In alloc_surplus_hugetlb_folio(), we increase nr_huge_pages and
surplus_huge_pages separately. In the middle window, if we set
nr_hugepages to smaller and satisfy count < persistent_huge_pages(h), the
surplus_huge_pages will be increased by adjust_pool_surplus().
After adding delay in the middle window, we can reproduce the problem
easily by following step:
1. echo 3 > /proc/sys/vm/nr_overcommit_hugepages
2. mmap two hugepages. When nr_huge_pages=2 and surplus_huge_pages=1,
goto step 3.
3. echo 0 > /proc/sys/vm/nr_huge_pages
Finally, nr_huge_pages is less than surplus_huge_pages.
To fix the problem, call only_alloc_fresh_hugetlb_folio() instead and
move down __prep_account_new_huge_page() into the hugetlb_lock.
Link: https://lkml.kernel.org/r/20250305035409.2391344-1-liushixin2@huawei.com
Fixes: 0c397daea1 ("mm, hugetlb: further simplify hugetlb allocation API")
Signed-off-by: Liu Shixin <liushixin2@huawei.com>
Acked-by: Peter Xu <peterx@redhat.com>
Acked-by: Oscar Salvador <osalvador@suse.de>
Cc: David Hildenbrand <david@redhat.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Liu Shixin <liushixin2@huawei.com>
Cc: Muchun Song <muchun.song@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "Add a command line option that enables control of how many
threads should be used to allocate huge pages", v2.
Allocating huge pages can take a very long time on servers with terabytes
of memory even when they are allocated at boot time where the allocation
happens in parallel.
Before this series, the kernel used a hard coded value of 2 threads per
NUMA node for these allocations. This value might have been good enough
in the past but it is not sufficient to fully utilize newer systems.
This series changes the default so the kernel uses 25% of the available
hardware threads for these allocations. In addition, we allow the user
that wish to micro-optimize the allocation time to override this value via
a new kernel parameter.
We tested this on 2 generations of Xeon CPUs and the results show a big
improvement of the overall allocation time.
+-----------------------+-------+-------+-------+-------+-------+
| threads | 8 | 16 | 32 | 64 | 128 |
+-----------------------+-------+-------+-------+-------+-------+
| skylake 144 cpus | 44s | 22s | 16s | 19s | 20s |
| cascade lake 192 cpus | 39s | 20s | 11s | 10s | 9s |
+-----------------------+-------+-------+-------+-------+-------+
On skylake, we see an improvment of 2.75x when using 32 threads, on
cascade lake we can get even better at 4.3x when we use 128 threads.
This speedup is quite significant and users of large machines like these
should have the option to make the machines boot as fast as possible.
This patch (of 3):
Before this patch, the kernel currently used a hard coded value of 2
threads per NUMA node for these allocations.
This patch changes this policy and the kernel now uses 25% of the
available hardware threads for the allocations.
Link: https://lkml.kernel.org/r/20250227-hugepage-parameter-v2-0-7db8c6dc0453@cyberus-technology.de
Link: https://lkml.kernel.org/r/20250227-hugepage-parameter-v2-1-7db8c6dc0453@cyberus-technology.de
Signed-off-by: Thomas Prescher <thomas.prescher@cyberus-technology.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Muchun Song <muchun.song@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
For large systems, the overhead of vmemmap pages for hugetlb is
substantial. It's about 1.5% of memory, which is about 45G for a 3T
system. If you want to configure most of that system for hugetlb (e.g.
to use as backing memory for VMs), there is a chance of running out of
memory on boot, even though you know that the 45G will become available
later.
To avoid this scenario, and since it's a waste to first allocate and then
free that 45G during boot, do pre-HVO for hugetlb bootmem allocated pages
('gigantic' pages).
pre-HVO is done by adding functions that are called from
sparse_init_nid_early and sparse_init_nid_late. The first is called
before memmap allocation, so it takes care of allocating memmap HVO-style.
The second verifies that all bootmem pages look good, specifically it
checks that they do not intersect with multiple zones. This can only be
done from sparse_init_nid_late path, when zones have been initialized.
The hugetlb page size must be aligned to the section size, and aligned to
the size of memory described by the number of page structures contained in
one PMD (since pre-HVO is not prepared to split PMDs). This should be
true for most 'gigantic' pages, it is for 1G pages on x86, where both of
these alignment requirements are 128M.
This will only have an effect if hugetlb_bootmem_alloc was called early in
boot. If not, it won't do anything, and HVO for bootmem hugetlb pages
works as before.
Link: https://lkml.kernel.org/r/20250228182928.2645936-20-fvdl@google.com
Signed-off-by: Frank van der Linden <fvdl@google.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Dan Carpenter <dan.carpenter@linaro.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Joao Martins <joao.m.martins@oracle.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roman Gushchin (Cruise) <roman.gushchin@linux.dev>
Cc: Usama Arif <usamaarif642@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Bootmem hugetlb pages are allocated using memblock, which isn't (and
mostly can't be) aware of zones.
So, they may end up crossing zone boundaries. This would create
confusion, a hugetlb page that is part of multiple zones is bad. Worse,
HVO might then end up stealthily re-assigning pages to a different zone
when a hugetlb page is freed, since the tail page structures beyond the
first vmemmap page would inherit the zone of the first page structures.
While the chance of this happening is low, you can definitely create a
configuration where this happens (especially using ZONE_MOVABLE).
To avoid this issue, check if bootmem hugetlb pages intersect with
multiple zones during the gather phase, and discard them, handing them to
the page allocator, if they do. Record the number of invalid bootmem
pages per node and subtract them from the number of available pages at the
end, making it easier to do these checks in multiple places later on.
Link: https://lkml.kernel.org/r/20250228182928.2645936-14-fvdl@google.com
Signed-off-by: Frank van der Linden <fvdl@google.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Dan Carpenter <dan.carpenter@linaro.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Joao Martins <joao.m.martins@oracle.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Roman Gushchin (Cruise) <roman.gushchin@linux.dev>
Cc: Usama Arif <usamaarif642@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
In dissolve_free_huge_page(), free huge pages are dissolved without
adjusting surplus count. However, free huge pages may be accounted as
surplus pages, and will lead to wrong surplus count.
I reproduce this issue on qemu. The steps are:
1) Node1 is memory-less at first. Hot-add memory to node1 by executing
the two commands in qemu monitor:
object_add memory-backend-ram,id=mem1,size=1G
device_add pc-dimm,id=dimm1,memdev=mem1,node=1
2) online one memory block of Node1 with：
echo online_movable > /sys/devices/system/node/node1/memoryX/state
3) create 64 huge pages for node1
4) run a program to reserve (don't consume) all the huge pages
5) echo 0 > nr_huge_pages for node1. After this step, free huge pages in
Node1 are surplus.
6) create 80 huge pages for node0
7) offline memory of node1, The memory range to offline contains the free
surplus huge pages created in step3) ~ step5)
echo offline > /sys/devices/system/node/node1/memoryX/state
8) kill the program in step 4)
The result:
Node0 Node1
total 80 0
free 80 0
surplus 0 61
To fix it, adjust surplus when destroying huge pages if the node has
surplus pages in dissolve_free_hugetlb_folio().
The result with this patch:
Node0 Node1
total 80 0
free 80 0
surplus 0 0
Link: https://lkml.kernel.org/r/20250304132106.2872754-1-tujinjiang@huawei.com
Fixes: c8721bbbdd ("mm: memory-hotplug: enable memory hotplug to handle hugepage")
Signed-off-by: Jinjiang Tu <tujinjiang@huawei.com>
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Oscar Salvador <osalvador@suse.de>
Cc: Jinjiang Tu <tujinjiang@huawei.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nanyong Sun <sunnanyong@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Pull misc fixes from Andrew Morton:
"33 hotfixes. 24 are cc:stable and the remainder address post-6.13
issues or aren't considered necessary for -stable kernels.
26 are for MM and 7 are for non-MM.
- "mm: memory_failure: unmap poisoned folio during migrate properly"
from Ma Wupeng fixes a couple of two year old bugs involving the
migration of hwpoisoned folios.
- "selftests/damon: three fixes for false results" from SeongJae Park
fixes three one year old bugs in the SAMON selftest code.
The remainder are singletons and doubletons. Please see the individual
changelogs for details"
* tag 'mm-hotfixes-stable-2025-03-08-16-27' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm: (33 commits)
mm/page_alloc: fix uninitialized variable
rapidio: add check for rio_add_net() in rio_scan_alloc_net()
rapidio: fix an API misues when rio_add_net() fails
MAINTAINERS: .mailmap: update Sumit Garg's email address
Revert "mm/page_alloc.c: don't show protection in zone's ->lowmem_reserve[] for empty zone"
mm: fix finish_fault() handling for large folios
mm: don't skip arch_sync_kernel_mappings() in error paths
mm: shmem: remove unnecessary warning in shmem_writepage()
userfaultfd: fix PTE unmapping stack-allocated PTE copies
userfaultfd: do not block on locking a large folio with raised refcount
mm: zswap: use ATOMIC_LONG_INIT to initialize zswap_stored_pages
mm: shmem: fix potential data corruption during shmem swapin
mm: fix kernel BUG when userfaultfd_move encounters swapcache
selftests/damon/damon_nr_regions: sort collected regiosn before checking with min/max boundaries
selftests/damon/damon_nr_regions: set ops update for merge results check to 100ms
selftests/damon/damos_quota: make real expectation of quota exceeds
include/linux/log2.h: mark is_power_of_2() with __always_inline
NFS: fix nfs_release_folio() to not deadlock via kcompactd writeback
mm, swap: avoid BUG_ON in relocate_cluster()
mm: swap: use correct step in loop to wait all clusters in wait_for_allocation()
...
Since the introduction of commit c77c0a8ac4 ("mm/hugetlb: defer freeing
of huge pages if in non-task context"), which supports deferring the
freeing of hugetlb pages, the allocation of contiguous memory through
cma_alloc() may fail probabilistically.
In the CMA allocation process, if it is found that the CMA area is
occupied by in-use hugetlb folios, these in-use hugetlb folios need to be
migrated to another location. When there are no available hugetlb folios
in the free hugetlb pool during the migration of in-use hugetlb folios,
new folios are allocated from the buddy system. A temporary state is set
on the newly allocated folio. Upon completion of the hugetlb folio
migration, the temporary state is transferred from the new folios to the
old folios. Normally, when the old folios with the temporary state are
freed, it is directly released back to the buddy system. However, due to
the deferred freeing of hugetlb pages, the PageBuddy() check fails,
ultimately leading to the failure of cma_alloc().
Here is a simplified call trace illustrating the process:
cma_alloc()
->__alloc_contig_migrate_range() // Migrate in-use hugetlb folios
->unmap_and_move_huge_page()
->folio_putback_hugetlb() // Free old folios
->test_pages_isolated()
->__test_page_isolated_in_pageblock()
->PageBuddy(page) // Check if the page is in buddy
To resolve this issue, we have implemented a function named
wait_for_freed_hugetlb_folios(). This function ensures that the hugetlb
folios are properly released back to the buddy system after their
migration is completed. By invoking wait_for_freed_hugetlb_folios()
before calling PageBuddy(), we ensure that PageBuddy() will succeed.
Link: https://lkml.kernel.org/r/1739936804-18199-1-git-send-email-yangge1116@126.com
Fixes: c77c0a8ac4 ("mm/hugetlb: defer freeing of huge pages if in non-task context")
Signed-off-by: Ge Yang <yangge1116@126.com>
Reviewed-by: Muchun Song <muchun.song@linux.dev>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <21cnbao@gmail.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Pull arm64 fixes from Will Deacon:
"Ryan's been hard at work finding and fixing mm bugs in the arm64 code,
so here's a small crop of fixes for -rc5.
The main changes are to fix our zapping of non-present PTEs for
hugetlb entries created using the contiguous bit in the page-table
rather than a block entry at the level above. Prior to these fixes, we
were pulling the contiguous bit back out of the PTE in order to
determine the size of the hugetlb page but this is clearly bogus if
the thing isn't present and consequently both the clearing of the
PTE(s) and the TLB invalidation were unreliable.
Although the problem was found by code inspection, we really don't
want this sitting around waiting to trigger and the changes are CC'd
to stable accordingly.
Note that the diffstat looks a lot worse than it really is;
huge_ptep_get_and_clear() now takes a size argument from the core code
and so all the arch implementations of that have been updated in a
pretty mechanical fashion.
- Fix a sporadic boot failure due to incorrect randomization of the
linear map on systems that support it
- Fix the zapping (both clearing the entries *and* invalidating the
TLB) of hugetlb PTEs constructed using the contiguous bit"
* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
arm64: hugetlb: Fix flush_hugetlb_tlb_range() invalidation level
arm64: hugetlb: Fix huge_ptep_get_and_clear() for non-present ptes
mm: hugetlb: Add huge page size param to huge_ptep_get_and_clear()
arm64/mm: Fix Boot panic on Ampere Altra
In order to fix a bug, arm64 needs to be told the size of the huge page
for which the huge_pte is being cleared in huge_ptep_get_and_clear().
Provide for this by adding an `unsigned long sz` parameter to the
function. This follows the same pattern as huge_pte_clear() and
set_huge_pte_at().
This commit makes the required interface modifications to the core mm as
well as all arches that implement this function (arm64, loongarch, mips,
parisc, powerpc, riscv, s390, sparc). The actual arm64 bug will be fixed
in a separate commit.
Cc: stable@vger.kernel.org
Fixes: 66b3923a1a ("arm64: hugetlb: add support for PTE contiguous bit")
Acked-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Alexandre Ghiti <alexghiti@rivosinc.com> # riscv
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Acked-by: Alexander Gordeev <agordeev@linux.ibm.com> # s390
Link: https://lore.kernel.org/r/20250226120656.2400136-2-ryan.roberts@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
When using the HugeTLB kernel command-line to allocate 1G pages from a
specific node, such as:
default_hugepagesz=1G hugepages=1:1
If node 1 happens to not have enough memory for the requested number of 1G
pages, the allocation falls back to other nodes. A quick way to reproduce
this is by creating a KVM guest with a memory-less node and trying to
allocate 1 1G page from it. Instead of failing, the allocation will
fallback to other nodes.
This defeats the purpose of node specific allocation. Also, specific node
allocation for 2M pages don't have this behavior: the allocation will just
fail for the pages it can't satisfy.
This issue happens because HugeTLB calls memblock_alloc_try_nid_raw() for
1G boot-time allocation as this function falls back to other nodes if the
allocation can't be satisfied. Use memblock_alloc_exact_nid_raw()
instead, which ensures that the allocation will only be satisfied from the
specified node.
Link: https://lkml.kernel.org/r/20250211034856.629371-1-luizcap@redhat.com
Fixes: b5389086ad ("hugetlbfs: extend the definition of hugepages parameter to support node allocation")
Signed-off-by: Luiz Capitulino <luizcap@redhat.com>
Acked-by: Oscar Salvador <osalvador@suse.de>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: "Mike Rapoport (IBM)" <rppt@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Zhenguo Yao <yaozhenguo1@gmail.com>
Cc: Frank van der Linden <fvdl@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
gather_bootmem_prealloc() assumes the start nid as 0 and size as
num_node_state(N_MEMORY). That means in case if memory attached numa
nodes are interleaved, then gather_bootmem_prealloc_parallel() will fail
to scan few of these nodes.
Since memory attached numa nodes can be interleaved in any fashion, hence
ensure that the current code checks for all numa node ids
(.size = nr_node_ids). Let's still keep max_threads as N_MEMORY, so that
it can distributes all nr_node_ids among the these many no. threads.
e.g. qemu cmdline
========================
numa_cmd="-numa node,nodeid=1,memdev=mem1,cpus=2-3 -numa node,nodeid=0,cpus=0-1 -numa dist,src=0,dst=1,val=20"
mem_cmd="-object memory-backend-ram,id=mem1,size=16G"
w/o this patch for cmdline (default_hugepagesz=1GB hugepagesz=1GB hugepages=2):
==========================
~ # cat /proc/meminfo |grep -i huge
AnonHugePages: 0 kB
ShmemHugePages: 0 kB
FileHugePages: 0 kB
HugePages_Total: 0
HugePages_Free: 0
HugePages_Rsvd: 0
HugePages_Surp: 0
Hugepagesize: 1048576 kB
Hugetlb: 0 kB
with this patch for cmdline (default_hugepagesz=1GB hugepagesz=1GB hugepages=2):
===========================
~ # cat /proc/meminfo |grep -i huge
AnonHugePages: 0 kB
ShmemHugePages: 0 kB
FileHugePages: 0 kB
HugePages_Total: 2
HugePages_Free: 2
HugePages_Rsvd: 0
HugePages_Surp: 0
Hugepagesize: 1048576 kB
Hugetlb: 2097152 kB
Link: https://lkml.kernel.org/r/f8d8dad3a5471d284f54185f65d575a6aaab692b.1736592534.git.ritesh.list@gmail.com
Fixes: b78b27d029 ("hugetlb: parallelize 1G hugetlb initialization")
Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
Reported-by: Pavithra Prakash <pavrampu@linux.ibm.com>
Suggested-by: Muchun Song <muchun.song@linux.dev>
Tested-by: Sourabh Jain <sourabhjain@linux.ibm.com>
Reviewed-by: Luiz Capitulino <luizcap@redhat.com>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Donet Tom <donettom@linux.ibm.com>
Cc: Gang Li <gang.li@linux.dev>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Add the const qualifier to all the ctl_tables in the tree except for
watchdog_hardlockup_sysctl, memory_allocation_profiling_sysctls,
loadpin_sysctl_table and the ones calling register_net_sysctl (./net,
drivers/inifiniband dirs). These are special cases as they use a
registration function with a non-const qualified ctl_table argument or
modify the arrays before passing them on to the registration function.
Constifying ctl_table structs will prevent the modification of
proc_handler function pointers as the arrays would reside in .rodata.
This is made possible after commit 78eb4ea25c ("sysctl: treewide:
constify the ctl_table argument of proc_handlers") constified all the
proc_handlers.
Created this by running an spatch followed by a sed command:
Spatch:
virtual patch
@
depends on !(file in "net")
disable optional_qualifier
@
identifier table_name != {
watchdog_hardlockup_sysctl,
iwcm_ctl_table,
ucma_ctl_table,
memory_allocation_profiling_sysctls,
loadpin_sysctl_table
};
@@
+ const
struct ctl_table table_name [] = { ... };
sed:
sed --in-place \
-e "s/struct ctl_table .table = &uts_kern/const struct ctl_table *table = \&uts_kern/" \
kernel/utsname_sysctl.c
Reviewed-by: Song Liu <song@kernel.org>
Acked-by: Steven Rostedt (Google) <rostedt@goodmis.org> # for kernel/trace/
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> # SCSI
Reviewed-by: Darrick J. Wong <djwong@kernel.org> # xfs
Acked-by: Jani Nikula <jani.nikula@intel.com>
Acked-by: Corey Minyard <cminyard@mvista.com>
Acked-by: Wei Liu <wei.liu@kernel.org>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Bill O'Donnell <bodonnel@redhat.com>
Acked-by: Baoquan He <bhe@redhat.com>
Acked-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Acked-by: Anna Schumaker <anna.schumaker@oracle.com>
Signed-off-by: Joel Granados <joel.granados@kernel.org>
We replaced a simple put_page() by a putback_active_hugepage() call in
commit 3aaa76e125 ("mm: migrate: hugetlb: putback destination hugepage
to active list"), to set the "active" flag on the dst hugetlb folio.
Nowadays, we decoupled the "active" list from the flag, by calling the
flag "migratable".
Calling "putback" on something that wasn't allocated is weird and not
future proof, especially if we might reach that path when migration failed
and we just want to free the freshly allocated hugetlb folio.
Let's simply handle the migratable flag and the active list flag in
move_hugetlb_state(), where we know that allocation succeeded and already
handle the temporary flag; use a simple folio_put() to return our
reference.
Link: https://lkml.kernel.org/r/20250113131611.2554758-4-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
vma_has_reserves() is a helper "trying" to know whether the vma should
consume one reservation when allocating the hugetlb folio.
However it's not clear on why we need such complexity, as such information
is already represented in the "chg" variable.
From alloc_hugetlb_folio() context, "chg" (or in the function's context,
"gbl_chg") is defined as:
- If gbl_chg=1, the allocation cannot reuse an existing reservation
- If gbl_chg=0, the allocation should reuse an existing reservation
Firstly, map_chg is defined as following, to cover all cases of hugetlb
reservation scenarios (mostly, via vma_needs_reservation(), but
cow_from_owner is an outlier):
CONDITION HAS RESERVATION?
========= ================
- SHARED: always check against per-inode resv_map
(ignore NONRESERVE)
- If resv exists ==> YES [1]
- If not ==> NO [2]
- PRIVATE: complicated...
- Request came from a CoW from owner resv map ==> NO [3]
(when cow_from_owner==true)
- If does not own a resv_map at all.. ==> NO [4]
(examples: VM_NORESERVE, private fork())
- If owns a resv_map, but resv donsn't exists ==> NO [5]
- If owns a resv_map, and resv exists ==> YES [6]
Further on, gbl_chg considered spool setup, so that is a decision based on
all the context.
If we look at vma_has_reserves(), it almost does check that has already
been processed by map_chg accounting (I marked each return value to the
case above):
static bool vma_has_reserves(struct vm_area_struct *vma, long chg)
{
if (vma->vm_flags & VM_NORESERVE) {
if (vma->vm_flags & VM_MAYSHARE && chg == 0)
return true; ==> [1]
else
return false; ==> [2] or [4]
}
if (vma->vm_flags & VM_MAYSHARE) {
if (chg)
return false; ==> [2]
else
return true; ==> [1]
}
if (is_vma_resv_set(vma, HPAGE_RESV_OWNER)) {
if (chg)
return false; ==> [5]
else
return true; ==> [6]
}
return false; ==> [4]
}
It didn't check [3], but [3] case was actually already covered now by the
"chg" / "gbl_chg" / "map_chg" calculations.
In short, vma_has_reserves() doesn't provide anything more than return
"!chg".. so just simplify all the things.
There're a lot of comments describing truncation races, IIUC there should
have no race as long as map_chg is properly done.
Link: https://lkml.kernel.org/r/20250107204002.2683356-6-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
Cc: Ackerley Tng <ackerleytng@google.com>
Cc: Breno Leitao <leitao@debian.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
alloc_hugetlb_folio() isn't a function easy to read, especially on
reservation accountings for either VMA or globally (majorly, spool only).
The 1st complexity lies in the special private CoW path, aka,
cow_from_owner=true case.
The 2nd complexity may be the confusing updates of gbl_chg after it's set
once, which looks like they can change anytime on the fly.
Logically, cow_from_user is only about vma reservation. We could already
decouple the flag and consolidate it into map charge flag very early.
Then we don't need to keep checking the CoW special flag every time.
This patch does it by making map_chg a tri-state flag. Tri-state needed
is unfortunate, and it's because currently vma_needs_reservation() has a
side effect internally, that it must be followed by either a end() or
commit().
We keep the same semantic as before on one thing: "if (map_chg)" means we
need a separate per-vma resv count. It keeps most of the old code like
before untouched with the new enum.
After this patch, we take these steps to decide these variables, hopefully
slightly easier to follow:
- First, decide map_chg. This will take cow_from_owner into account,
once and for all. It's about whether we could take a resv count from
the vma, no matter it's shared, private, etc.
- Then, decide gbl_chg. The only diff here is spool, comparing to
map_chg.
Now only update each flag once and for all, instead of keep any of them
flipping which can be very hard to follow.
With cow_from_owner merged into map_chg, we could remove quite a few such
checks all over. Side benefit of such is that we can get rid of one more
confusing flag, which is deferred_reserve.
Cleanup the comments a bit too. E.g., MAP_NORESERVE may not need to check
against spool limit, AFAIU, if it's on a shared mapping, and if the page
cache folio has its inode's resv map available (in which case map_chg
would have been set zero, hence the code should be correct, not the
comment).
There's one trivial detail that needs attention that this patch touched,
which is this check right after vma_commit_reservation():
if (map_chg > map_commit)
It changes to:
if (unlikely(map_chg == MAP_CHG_NEEDED && retval == 0))
It should behave the same like before, because previously the only way to
make "map_chg > map_commit" happen is map_chg=1 && map_commit=0. That's
exactly the rewritten line. Meanwhile, either commit() or end() will need
to be skipped if ENFORCE, to keep the old behavior.
Even though it looks a lot changed, but no functional change expected.
Link: https://lkml.kernel.org/r/20250107204002.2683356-5-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
Cc: Ackerley Tng <ackerleytng@google.com>
Cc: Breno Leitao <leitao@debian.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The old name "avoid_reserve" can be too generic and can be used wrongly in
the new call sites that want to allocate a hugetlb folio.
It's confusing on two things: (1) whether one can opt-in to avoid global
reservation, and (2) whether it should take more than one count.
In reality, this flag is only used in an extremely hacky path, in an
extremely hacky way in hugetlb CoW path only, and always use with 1 saying
"skip global reservation". Rename the flag to avoid future abuse of this
flag, making it a boolean so as to reflect its true representation that
it's not a counter. To make it even harder to abuse, add a comment above
the function to explain it.
Link: https://lkml.kernel.org/r/20250107204002.2683356-4-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Ackerley Tng <ackerleytng@google.com>
Cc: Breno Leitao <leitao@debian.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
When fork() and stumble on top of a dma-pinned hugetlb private page, CoW
must happen during fork() to guarantee dma coherency.
In this specific path, hugetlb pages need to be allocated for the child
process. Stop using avoid_reserve=1 flag here: it's not required to be
used here, as dest_vma (which is destined to be a MAP_PRIVATE hugetlb vma)
will have no private vma resv map, and that will make sure it won't be
able to use a vma reservation later.
No functional change intended with this change. Said that, it's still
wanted to do this, so as to reduce the usage of avoid_reserve to the only
one user, which is also why this flag was introduced initially in commit
04f2cbe356 ("hugetlb: guarantee that COW faults for a process that
called mmap(MAP_PRIVATE) on hugetlbfs will succeed"). I don't see whoever
else should set it at all.
Further patch will clean up resv accounting based on this.
Link: https://lkml.kernel.org/r/20250107204002.2683356-3-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Ackerley Tng <ackerleytng@google.com>
Cc: Breno Leitao <leitao@debian.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "mm/hugetlb: Refactor hugetlb allocation resv accounting",
v2.
This is a follow up on Ackerley's series here as replacement:
https://lore.kernel.org/r/cover.1728684491.git.ackerleytng@google.com
The goal of this series is to cleanup hugetlb resv accounting, especially
during folio allocation, to decouple a few things:
- Hugetlb folios v.s. Hugetlbfs: IOW, the hope is in the future hugetlb
folios can be allocated completely without hugetlbfs.
- Decouple VMA v.s. hugetlb folio allocations: allocating a hugetlb folio
should not always require a hugetlbfs VMA. For example, either it got
allocated from the inode level (see hugetlbfs_fallocate() where it used
a pesudo VMA for allocation), or it can be allocated by other kernel
subsystems.
It paves way for other users to allocate hugetlb folios out of either
system reservations, or subpools (instead of hugetlbfs, as a file system).
For longer term, this prepares hugetlb as a separate concept versus
hugetlbfs, so that hugetlb folios can be allocated by not only hugetlbfs
and other things.
Tests I've done:
- I had a reproducer in patch 1 for the bug I found, this will start to
work after patch 1 or the whole set applied.
- Hugetlb regression tests (on x86_64 2MBs), includes:
- All vmtests on hugetlbfs
- libhugetlbfs test suite (which may fail some tests, but no new failures
will be introduced by this series, so all such failures happen before
this series so shouldn't be relevant).
This patch (of 7):
Since commit 04f2cbe356 ("hugetlb: guarantee that COW faults for a
process that called mmap(MAP_PRIVATE) on hugetlbfs will succeed"),
avoid_reserve was introduced for a special case of CoW on hugetlb private
mappings, and only if the owner VMA is trying to allocate yet another
hugetlb folio that is not reserved within the private vma reserved map.
Later on, in commit d85f69b0b5 ("mm/hugetlb: alloc_huge_page handle
areas hole punched by fallocate"), alloc_huge_page() enforced to not
consume any global reservation as long as avoid_reserve=true. This
operation doesn't look correct, because even if it will enforce the
allocation to not use global reservation at all, it will still try to take
one reservation from the spool (if the subpool existed). Then since the
spool reserved pages take from global reservation, it'll also take one
reservation globally.
Logically it can cause global reservation to go wrong.
I wrote a reproducer below, trigger this special path, and every run of
such program will cause global reservation count to increment by one, until
it hits the number of free pages:
#define _GNU_SOURCE /* See feature_test_macros(7) */
#include <stdio.h>
#include <fcntl.h>
#include <errno.h>
#include <unistd.h>
#include <stdlib.h>
#include <sys/mman.h>
#define MSIZE (2UL << 20)
int main(int argc, char *argv[])
{
const char *path;
int *buf;
int fd, ret;
pid_t child;
if (argc < 2) {
printf("usage: %s <hugetlb_file>\n", argv[0]);
return -1;
}
path = argv[1];
fd = open(path, O_RDWR | O_CREAT, 0666);
if (fd < 0) {
perror("open failed");
return -1;
}
ret = fallocate(fd, 0, 0, MSIZE);
if (ret != 0) {
perror("fallocate");
return -1;
}
buf = mmap(NULL, MSIZE, PROT_READ|PROT_WRITE,
MAP_PRIVATE, fd, 0);
if (buf == MAP_FAILED) {
perror("mmap() failed");
return -1;
}
/* Allocate a page */
*buf = 1;
child = fork();
if (child == 0) {
/* child doesn't need to do anything */
exit(0);
}
/* Trigger CoW from owner */
*buf = 2;
munmap(buf, MSIZE);
close(fd);
unlink(path);
return 0;
}
It can only reproduce with a sub-mount when there're reserved pages on the
spool, like:
# sysctl vm.nr_hugepages=128
# mkdir ./hugetlb-pool
# mount -t hugetlbfs -o min_size=8M,pagesize=2M none ./hugetlb-pool
Then run the reproducer on the mountpoint:
# ./reproducer ./hugetlb-pool/test
Fix it by taking the reservation from spool if available. In general,
avoid_reserve is IMHO more about "avoid vma resv map", not spool's.
I copied stable, however I have no intention for backporting if it's not a
clean cherry-pick, because private hugetlb mapping, and then fork() on top
is too rare to hit.
Link: https://lkml.kernel.org/r/20250107204002.2683356-1-peterx@redhat.com
Link: https://lkml.kernel.org/r/20250107204002.2683356-2-peterx@redhat.com
Fixes: d85f69b0b5 ("mm/hugetlb: alloc_huge_page handle areas hole punched by fallocate")
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Ackerley Tng <ackerleytng@google.com>
Tested-by: Ackerley Tng <ackerleytng@google.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Breno Leitao <leitao@debian.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
My machine has 4 NUMA nodes, each equipped with 32GB of memory. I have
configured each NUMA node with 16GB of CMA and 16GB of in-use hugetlb
pages. The allocation of contiguous memory via cma_alloc() can fail
probabilistically.
When there are free hugetlb folios in the hugetlb pool, during the
migration of in-use hugetlb folios, new folios are allocated from the free
hugetlb pool. After the migration is completed, the old folios are
released back to the free hugetlb pool instead of being returned to the
buddy system. This can cause test_pages_isolated() check to fail,
ultimately leading to the failure of cma_alloc().
Call trace:
cma_alloc()
__alloc_contig_migrate_range() // migrate in-use hugepage
test_pages_isolated()
__test_page_isolated_in_pageblock()
PageBuddy(page) // check if the page is in buddy
To address this issue, we introduce a function named
replace_free_hugepage_folios(). This function will replace the hugepage
in the free hugepage pool with a new one and release the old one to the
buddy system. After the migration of in-use hugetlb pages is completed,
we will invoke replace_free_hugepage_folios() to ensure that these
hugepages are properly released to the buddy system. Following this step,
when test_pages_isolated() is executed for inspection, it will
successfully pass.
Additionally, when alloc_contig_range() is used to migrate multiple in-use
hugetlb pages, it can result in some in-use hugetlb pages being released
back to the free hugetlb pool and subsequently being reallocated and used
again. For example:
[huge 0] [huge 1]
To migrate huge 0, we obtain huge x from the pool. After the migration is
completed, we return the now-freed huge 0 back to the pool. When it's
time to migrate huge 1, we can simply reuse the now-freed huge 0 from the
pool. As a result, when replace_free_hugepage_folios() is executed, it
cannot release huge 0 back to the buddy system. To address this issue, we
should prevent the reuse of isolated free hugepages during the migration
process.
Link: https://lkml.kernel.org/r/1734503588-16254-1-git-send-email-yangge1116@126.com
Link: https://lkml.kernel.org/r/1736582300-11364-1-git-send-email-yangge1116@126.com
Signed-off-by: yangge <yangge1116@126.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <21cnbao@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: SeongJae Park <sj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Eric reported that PTRACE_POKETEXT fails when applications use hugetlb for
mapping text using huge pages. Before commit 1d8d14641f ("mm/hugetlb:
support write-faults in shared mappings"), PTRACE_POKETEXT worked by
accident, but it was buggy and silently ended up mapping pages writable
into the page tables even though VM_WRITE was not set.
In general, FOLL_FORCE|FOLL_WRITE does currently not work with hugetlb.
Let's implement FOLL_FORCE|FOLL_WRITE properly for hugetlb, such that what
used to work in the past by accident now properly works, allowing
applications using hugetlb for text etc. to get properly debugged.
This change might also be required to implement uprobes support for
hugetlb [1].
[1] https://lore.kernel.org/lkml/ZiK50qob9yl5e0Xz@bender.morinfr.org/
Link: https://lkml.kernel.org/r/Z1NshNfWuzUCPebA@bender.morinfr.org
Signed-off-by: Guillaume Morin <guillaume@morinfr.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Eric Hagberg <ehagberg@janestreet.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
If we have to trigger a hugetlb folio copy during fork() because the anon
folio might be pinned, we currently unconditionally create a writable PTE.
However, the VMA might not have write permissions (VM_WRITE) at that
point.
Fix it by checking the VMA for VM_WRITE. Make the code less error prone
by moving checking for VM_WRITE into make_huge_pte(), and letting callers
only specify whether we should try making it writable.
A simple reproducer that longterm-pins the folios using liburing to then
mprotect(PROT_READ) the folios befor fork() [1] results in:
Before:
[FAIL] access should not have worked
After:
[PASS] access did not work as expected
[1] https://gitlab.com/davidhildenbrand/scratchspace/-/raw/main/reproducers/hugetlb-mkwrite-fork.c
This is rather a corner case, so stable might not be warranted.
Link: https://lkml.kernel.org/r/20241204153100.1967364-1-david@redhat.com
Fixes: 4eae4efa2c ("hugetlb: do early cow when page pinned on src mm")
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Peter Xu <peterx@redhat.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Guillaume Morin <guillaume@morinfr.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Previously, surplus allocations triggered by mmap were typically made from
the node where the process was running. On a page fault, the area was
reliably dequeued from the hugepage_freelists for that node. However,
since commit 003af997c8 ("hugetlb: force allocating surplus hugepages on
mempolicy allowed nodes"), dequeue_hugetlb_folio_vma() may fall back to
other nodes unnecessarily even if there is no MPOL_BIND policy, causing
folios to be dequeued from nodes other than the current one.
Also, allocating from the node where the current process is running is
likely to result in a performance win, as mmap-ing processes often touch
the area not so long after allocation. This change minimizes surprises
for users relying on the previous behavior while maintaining the benefit
introduced by the commit.
So, prioritize the node the current process is running on when possible.
Link: https://lkml.kernel.org/r/20241204165503.628784-1-koichiro.den@canonical.com
Signed-off-by: Koichiro Den <koichiro.den@canonical.com>
Acked-by: Aristeu Rozanski <aris@ruivo.org>
Cc: Aristeu Rozanski <aris@redhat.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The folio refcount may be increased unexpectly through try_get_folio() by
caller such as split_huge_pages. In huge_pmd_unshare(), we use refcount
to check whether a pmd page table is shared. The check is incorrect if
the refcount is increased by the above caller, and this can cause the page
table leaked:
BUG: Bad page state in process sh pfn:109324
page: refcount:0 mapcount:0 mapping:0000000000000000 index:0x66 pfn:0x109324
flags: 0x17ffff800000000(node=0|zone=2|lastcpupid=0xfffff)
page_type: f2(table)
raw: 017ffff800000000 0000000000000000 0000000000000000 0000000000000000
raw: 0000000000000066 0000000000000000 00000000f2000000 0000000000000000
page dumped because: nonzero mapcount
...
CPU: 31 UID: 0 PID: 7515 Comm: sh Kdump: loaded Tainted: G B 6.13.0-rc2master+ #7
Tainted: [B]=BAD_PAGE
Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015
Call trace:
show_stack+0x20/0x38 (C)
dump_stack_lvl+0x80/0xf8
dump_stack+0x18/0x28
bad_page+0x8c/0x130
free_page_is_bad_report+0xa4/0xb0
free_unref_page+0x3cc/0x620
__folio_put+0xf4/0x158
split_huge_pages_all+0x1e0/0x3e8
split_huge_pages_write+0x25c/0x2d8
full_proxy_write+0x64/0xd8
vfs_write+0xcc/0x280
ksys_write+0x70/0x110
__arm64_sys_write+0x24/0x38
invoke_syscall+0x50/0x120
el0_svc_common.constprop.0+0xc8/0xf0
do_el0_svc+0x24/0x38
el0_svc+0x34/0x128
el0t_64_sync_handler+0xc8/0xd0
el0t_64_sync+0x190/0x198
The issue may be triggered by damon, offline_page, page_idle, etc, which
will increase the refcount of page table.
1. The page table itself will be discarded after reporting the
"nonzero mapcount".
2. The HugeTLB page mapped by the page table miss freeing since we
treat the page table as shared and a shared page table will not be
unmapped.
Fix it by introducing independent PMD page table shared count. As
described by comment, pt_index/pt_mm/pt_frag_refcount are used for s390
gmap, x86 pgds and powerpc, pt_share_count is used for x86/arm64/riscv
pmds, so we can reuse the field as pt_share_count.
Link: https://lkml.kernel.org/r/20241216071147.3984217-1-liushixin2@huawei.com
Fixes: 39dde65c99 ("[PATCH] shared page table for hugetlb page")
Signed-off-by: Liu Shixin <liushixin2@huawei.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Ken Chen <kenneth.w.chen@intel.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nanyong Sun <sunnanyong@huawei.com>
Cc: Jane Chu <jane.chu@oracle.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
In current kernel, hugetlb_wp() calls copy_user_large_folio() with the
fault address. Where the fault address may be not aligned with the huge
page size. Then, copy_user_large_folio() may call
copy_user_gigantic_page() with the address, while
copy_user_gigantic_page() requires the address to be huge page size
aligned. So, this may cause memory corruption or information leak,
addtional, use more obvious naming 'addr_hint' instead of 'addr' for
copy_user_gigantic_page().
Link: https://lkml.kernel.org/r/20241028145656.932941-2-wangkefeng.wang@huawei.com
Fixes: 530dd9926d ("mm: memory: improve copy_user_large_folio()")
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
This patch introduces a new counter to memory.stat that tracks hugeTLB
usage, only if hugeTLB accounting is done to memory.current. This feature
is enabled the same way hugeTLB accounting is enabled, via the
memory_hugetlb_accounting mount flag for cgroupsv2.
1. Why is this patch necessary?
Currently, memcg hugeTLB accounting is an opt-in feature [1] that adds
hugeTLB usage to memory.current. However, the metric is not reported in
memory.stat. Given that users often interpret memory.stat as a breakdown
of the value reported in memory.current, the disparity between the two
reports can be confusing. This patch solves this problem by including the
metric in memory.stat as well, but only if it is also reported in
memory.current (it would also be confusing if the value was reported in
memory.stat, but not in memory.current)
Aside from the consistency between the two files, we also see benefits in
observability. Userspace might be interested in the hugeTLB footprint of
cgroups for many reasons. For instance, system admins might want to
verify that hugeTLB usage is distributed as expected across tasks: i.e.
memory-intensive tasks are using more hugeTLB pages than tasks that don't
consume a lot of memory, or are seen to fault frequently. Note that this
is separate from wanting to inspect the distribution for limiting purposes
(in which case, hugeTLB controller makes more sense).
2. We already have a hugeTLB controller. Why not use that?
It is true that hugeTLB tracks the exact value that we want. In fact, by
enabling the hugeTLB controller, we get all of the observability benefits
that I mentioned above, and users can check the total hugeTLB usage,
verify if it is distributed as expected, etc.
With this said, there are 2 problems:
(a) They are still not reported in memory.stat, which means the
disparity between the memcg reports are still there.
(b) We cannot reasonably expect users to enable the hugeTLB controller
just for the sake of hugeTLB usage reporting, especially since
they don't have any use for hugeTLB usage enforcing [2].
3. Implementation Details:
In the alloc / free hugetlb functions, we call lruvec_stat_mod_folio
regardless of whether memcg accounts hugetlb. mem_cgroup_commit_charge
which is called from alloc_hugetlb_folio will set memcg for the folio only
if the CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING cgroup mount option is used, so
lruvec_stat_mod_folio accounts per-memcg hugetlb counters only if the
feature is enabled. Regardless of whether memcg accounts for hugetlb, the
newly added global counter is updated and shown in /proc/vmstat.
The global counter is added because vmstats is the preferred framework for
cgroup stats. It makes stat items consistent between global and cgroups.
It also provides a per-node breakdown, which is useful. Because it does
not use cgroup-specific hooks, we also keep generic MM code separate from
memcg code.
[1] https://lore.kernel.org/all/20231006184629.155543-1-nphamcs@gmail.com/
[2] Of course, we can't make a new patch for every feature that can be
duplicated. However, since the existing solution of enabling the
hugeTLB controller is an imperfect solution that still leaves a
discrepancy between memory.stat and memory.curent, I think that it
is reasonable to isolate the feature in this case.
Link: https://lkml.kernel.org/r/20241101204402.1885383-1-joshua.hahnjy@gmail.com
Signed-off-by: Joshua Hahn <joshua.hahnjy@gmail.com>
Suggested-by: Nhat Pham <nphamcs@gmail.com>
Suggested-by: Shakeel Butt <shakeel.butt@linux.dev>
Suggested-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Shakeel Butt <shakeel.butt@linux.dev>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Chris Down <chris@chrisdown.name>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Roman Gushchin <roman.gushchin@linux.dev>
Reviewed-by: Nhat Pham <nphamcs@gmail.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Michal Koutný <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
