Patch series "Free some vmemmap pages of HugeTLB page", v23.
This patch series will free some vmemmap pages(struct page structures)
associated with each HugeTLB page when preallocated to save memory.
In order to reduce the difficulty of the first version of code review. In
this version, we disable PMD/huge page mapping of vmemmap if this feature
was enabled. This acutely eliminates a bunch of the complex code doing
page table manipulation. When this patch series is solid, we cam add the
code of vmemmap page table manipulation in the future.
The struct page structures (page structs) are used to describe a physical
page frame. By default, there is an one-to-one mapping from a page frame
to it's corresponding page struct.
The HugeTLB pages consist of multiple base page size pages and is
supported by many architectures. See hugetlbpage.rst in the Documentation
directory for more details. On the x86 architecture, HugeTLB pages of
size 2MB and 1GB are currently supported. Since the base page size on x86
is 4KB, a 2MB HugeTLB page consists of 512 base pages and a 1GB HugeTLB
page consists of 4096 base pages. For each base page, there is a
corresponding page struct.
Within the HugeTLB subsystem, only the first 4 page structs are used to
contain unique information about a HugeTLB page. HUGETLB_CGROUP_MIN_ORDER
provides this upper limit. The only 'useful' information in the remaining
page structs is the compound_head field, and this field is the same for
all tail pages.
By removing redundant page structs for HugeTLB pages, memory can returned
to the buddy allocator for other uses.
When the system boot up, every 2M HugeTLB has 512 struct page structs which
size is 8 pages(sizeof(struct page) * 512 / PAGE_SIZE).
HugeTLB struct pages(8 pages) page frame(8 pages)
+-----------+ ---virt_to_page---> +-----------+ mapping to +-----------+
| | | 0 | -------------> | 0 |
| | +-----------+ +-----------+
| | | 1 | -------------> | 1 |
| | +-----------+ +-----------+
| | | 2 | -------------> | 2 |
| | +-----------+ +-----------+
| | | 3 | -------------> | 3 |
| | +-----------+ +-----------+
| | | 4 | -------------> | 4 |
| 2MB | +-----------+ +-----------+
| | | 5 | -------------> | 5 |
| | +-----------+ +-----------+
| | | 6 | -------------> | 6 |
| | +-----------+ +-----------+
| | | 7 | -------------> | 7 |
| | +-----------+ +-----------+
| |
| |
| |
+-----------+
The value of page->compound_head is the same for all tail pages. The
first page of page structs (page 0) associated with the HugeTLB page
contains the 4 page structs necessary to describe the HugeTLB. The only
use of the remaining pages of page structs (page 1 to page 7) is to point
to page->compound_head. Therefore, we can remap pages 2 to 7 to page 1.
Only 2 pages of page structs will be used for each HugeTLB page. This
will allow us to free the remaining 6 pages to the buddy allocator.
Here is how things look after remapping.
HugeTLB struct pages(8 pages) page frame(8 pages)
+-----------+ ---virt_to_page---> +-----------+ mapping to +-----------+
| | | 0 | -------------> | 0 |
| | +-----------+ +-----------+
| | | 1 | -------------> | 1 |
| | +-----------+ +-----------+
| | | 2 | ----------------^ ^ ^ ^ ^ ^
| | +-----------+ | | | | |
| | | 3 | ------------------+ | | | |
| | +-----------+ | | | |
| | | 4 | --------------------+ | | |
| 2MB | +-----------+ | | |
| | | 5 | ----------------------+ | |
| | +-----------+ | |
| | | 6 | ------------------------+ |
| | +-----------+ |
| | | 7 | --------------------------+
| | +-----------+
| |
| |
| |
+-----------+
When a HugeTLB is freed to the buddy system, we should allocate 6 pages
for vmemmap pages and restore the previous mapping relationship.
Apart from 2MB HugeTLB page, we also have 1GB HugeTLB page. It is similar
to the 2MB HugeTLB page. We also can use this approach to free the
vmemmap pages.
In this case, for the 1GB HugeTLB page, we can save 4094 pages. This is a
very substantial gain. On our server, run some SPDK/QEMU applications
which will use 1024GB HugeTLB page. With this feature enabled, we can
save ~16GB (1G hugepage)/~12GB (2MB hugepage) memory.
Because there are vmemmap page tables reconstruction on the
freeing/allocating path, it increases some overhead. Here are some
overhead analysis.
1) Allocating 10240 2MB HugeTLB pages.
a) With this patch series applied:
# time echo 10240 > /proc/sys/vm/nr_hugepages
real 0m0.166s
user 0m0.000s
sys 0m0.166s
# bpftrace -e 'kprobe:alloc_fresh_huge_page { @start[tid] = nsecs; }
kretprobe:alloc_fresh_huge_page /@start[tid]/ { @latency = hist(nsecs -
@start[tid]); delete(@start[tid]); }'
Attaching 2 probes...
@latency:
[8K, 16K) 5476 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
[16K, 32K) 4760 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ |
[32K, 64K) 4 | |
b) Without this patch series:
# time echo 10240 > /proc/sys/vm/nr_hugepages
real 0m0.067s
user 0m0.000s
sys 0m0.067s
# bpftrace -e 'kprobe:alloc_fresh_huge_page { @start[tid] = nsecs; }
kretprobe:alloc_fresh_huge_page /@start[tid]/ { @latency = hist(nsecs -
@start[tid]); delete(@start[tid]); }'
Attaching 2 probes...
@latency:
[4K, 8K) 10147 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
[8K, 16K) 93 | |
Summarize: this feature is about ~2x slower than before.
2) Freeing 10240 2MB HugeTLB pages.
a) With this patch series applied:
# time echo 0 > /proc/sys/vm/nr_hugepages
real 0m0.213s
user 0m0.000s
sys 0m0.213s
# bpftrace -e 'kprobe:free_pool_huge_page { @start[tid] = nsecs; }
kretprobe:free_pool_huge_page /@start[tid]/ { @latency = hist(nsecs -
@start[tid]); delete(@start[tid]); }'
Attaching 2 probes...
@latency:
[8K, 16K) 6 | |
[16K, 32K) 10227 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
[32K, 64K) 7 | |
b) Without this patch series:
# time echo 0 > /proc/sys/vm/nr_hugepages
real 0m0.081s
user 0m0.000s
sys 0m0.081s
# bpftrace -e 'kprobe:free_pool_huge_page { @start[tid] = nsecs; }
kretprobe:free_pool_huge_page /@start[tid]/ { @latency = hist(nsecs -
@start[tid]); delete(@start[tid]); }'
Attaching 2 probes...
@latency:
[4K, 8K) 6805 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
[8K, 16K) 3427 |@@@@@@@@@@@@@@@@@@@@@@@@@@ |
[16K, 32K) 8 | |
Summary: The overhead of __free_hugepage is about ~2-3x slower than before.
Although the overhead has increased, the overhead is not significant.
Like Mike said, "However, remember that the majority of use cases create
HugeTLB pages at or shortly after boot time and add them to the pool. So,
additional overhead is at pool creation time. There is no change to
'normal run time' operations of getting a page from or returning a page to
the pool (think page fault/unmap)".
Despite the overhead and in addition to the memory gains from this series.
The following data is obtained by Joao Martins. Very thanks to his
effort.
There's an additional benefit which is page (un)pinners will see an improvement
and Joao presumes because there are fewer memmap pages and thus the tail/head
pages are staying in cache more often.
Out of the box Joao saw (when comparing linux-next against linux-next +
this series) with gup_test and pinning a 16G HugeTLB file (with 1G pages):
get_user_pages(): ~32k -> ~9k
unpin_user_pages(): ~75k -> ~70k
Usually any tight loop fetching compound_head(), or reading tail pages
data (e.g. compound_head) benefit a lot. There's some unpinning
inefficiencies Joao was fixing[2], but with that in added it shows even
more:
unpin_user_pages(): ~27k -> ~3.8k
[1] https://lore.kernel.org/linux-mm/20210409205254.242291-1-mike.kravetz@oracle.com/
[2] https://lore.kernel.org/linux-mm/20210204202500.26474-1-joao.m.martins@oracle.com/
This patch (of 9):
Move bootmem info registration common API to individual bootmem_info.c.
And we will use {get,put}_page_bootmem() to initialize the page for the
vmemmap pages or free the vmemmap pages to buddy in the later patch. So
move them out of CONFIG_MEMORY_HOTPLUG_SPARSE. This is just code movement
without any functional change.
Link: https://lkml.kernel.org/r/20210510030027.56044-1-songmuchun@bytedance.com
Link: https://lkml.kernel.org/r/20210510030027.56044-2-songmuchun@bytedance.com
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Acked-by: Mike Kravetz <mike.kravetz@oracle.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Miaohe Lin <linmiaohe@huawei.com>
Tested-by: Chen Huang <chenhuang5@huawei.com>
Tested-by: Bodeddula Balasubramaniam <bodeddub@amazon.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: x86@kernel.org
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Paul E. McKenney <paulmck@kernel.org>
Cc: Pawan Gupta <pawan.kumar.gupta@linux.intel.com>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Oliver Neukum <oneukum@suse.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Mina Almasry <almasrymina@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Barry Song <song.bao.hua@hisilicon.com>
Cc: HORIGUCHI NAOYA <naoya.horiguchi@nec.com>
Cc: Joao Martins <joao.m.martins@oracle.com>
Cc: Xiongchun Duan <duanxiongchun@bytedance.com>
Cc: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There maybe an overflow in memblock_overlaps_region() if it is called with
base and size such that
base + size > PHYS_ADDR_MAX
Make sure that memblock_overlaps_region() caps the size to prevent such
overflow and remove now duplicated call to memblock_cap_size() from
memblock_is_region_reserved().
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Tested-by: Tony Lindgren <tony@atomide.com>
When CONFIG_SPARSEMEM=y the ranges of the memory map that are freed are not
aligned to the pageblock boundaries which breaks assumptions about
homogeneity of the memory map throughout core mm code.
Make sure that the freed memory map is always aligned on pageblock
boundaries regardless of the memory model selection.
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Tested-by: Tony Lindgren <tony@atomide.com>
The code that frees unused memory map uses rounds start and end of the
holes that are freed to MAX_ORDER_NR_PAGES to preserve continuity of the
memory map for MAX_ORDER regions.
Lots of core memory management functionality relies on homogeneity of the
memory map within each pageblock which size may differ from MAX_ORDER in
certain configurations.
Although currently, for the architectures that use free_unused_memmap(),
pageblock_order and MAX_ORDER are equivalent, it is cleaner to have common
notation thought mm code.
Replace MAX_ORDER_NR_PAGES with pageblock_nr_pages and update the comments
to make it more clear why the alignment to pageblock boundaries is
required.
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Tested-by: Tony Lindgren <tony@atomide.com>
Merge misc updates from Andrew Morton:
"191 patches.
Subsystems affected by this patch series: kthread, ia64, scripts,
ntfs, squashfs, ocfs2, kernel/watchdog, and mm (gup, pagealloc, slab,
slub, kmemleak, dax, debug, pagecache, gup, swap, memcg, pagemap,
mprotect, bootmem, dma, tracing, vmalloc, kasan, initialization,
pagealloc, and memory-failure)"
* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (191 commits)
mm,hwpoison: make get_hwpoison_page() call get_any_page()
mm,hwpoison: send SIGBUS with error virutal address
mm/page_alloc: split pcp->high across all online CPUs for cpuless nodes
mm/page_alloc: allow high-order pages to be stored on the per-cpu lists
mm: replace CONFIG_FLAT_NODE_MEM_MAP with CONFIG_FLATMEM
mm: replace CONFIG_NEED_MULTIPLE_NODES with CONFIG_NUMA
docs: remove description of DISCONTIGMEM
arch, mm: remove stale mentions of DISCONIGMEM
mm: remove CONFIG_DISCONTIGMEM
m68k: remove support for DISCONTIGMEM
arc: remove support for DISCONTIGMEM
arc: update comment about HIGHMEM implementation
alpha: remove DISCONTIGMEM and NUMA
mm/page_alloc: move free_the_page
mm/page_alloc: fix counting of managed_pages
mm/page_alloc: improve memmap_pages dbg msg
mm: drop SECTION_SHIFT in code comments
mm/page_alloc: introduce vm.percpu_pagelist_high_fraction
mm/page_alloc: limit the number of pages on PCP lists when reclaim is active
mm/page_alloc: scale the number of pages that are batch freed
...
__get_hwpoison_page() could fail to grab refcount by some race condition,
so it's helpful if we can handle it by retrying. We already have retry
logic, so make get_hwpoison_page() call get_any_page() when called from
memory_failure().
As a result, get_hwpoison_page() can return negative values (i.e. error
code), so some callers are also changed to handle error cases.
soft_offline_page() does nothing for -EBUSY because that's enough and
users in userspace can easily handle it. unpoison_memory() is also
unchanged because it's broken and need thorough fixes (will be done
later).
Link: https://lkml.kernel.org/r/20210603233632.2964832-3-nao.horiguchi@gmail.com
Signed-off-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Tony Luck <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Dave Hansen reported the following about Feng Tang's tests on a machine
with persistent memory onlined as a DRAM-like device.
Feng Tang tossed these on a "Cascade Lake" system with 96 threads and
~512G of persistent memory and 128G of DRAM. The PMEM is in "volatile
use" mode and being managed via the buddy just like the normal RAM.
The PMEM zones are big ones:
present 65011712 = 248 G
high 134595 = 525 M
The PMEM nodes, of course, don't have any CPUs in them.
With your series, the pcp->high value per-cpu is 69584 pages or about
270MB per CPU. Scaled up by the 96 CPU threads, that's ~26GB of
worst-case memory in the pcps per zone, or roughly 10% of the size of
the zone.
This should not cause a problem as such although it could trigger reclaim
due to pages being stored on per-cpu lists for CPUs remote to a node. It
is not possible to treat cpuless nodes exactly the same as normal nodes
but the worst-case scenario can be mitigated by splitting pcp->high across
all online CPUs for cpuless memory nodes.
Link: https://lkml.kernel.org/r/20210616110743.GK30378@techsingularity.net
Suggested-by: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Dave Hansen <dave.hansen@intel.com>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: "Tang, Feng" <feng.tang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The per-cpu page allocator (PCP) only stores order-0 pages. This means
that all THP and "cheap" high-order allocations including SLUB contends on
the zone->lock. This patch extends the PCP allocator to store THP and
"cheap" high-order pages. Note that struct per_cpu_pages increases in
size to 256 bytes (4 cache lines) on x86-64.
Note that this is not necessarily a universal performance win because of
how it is implemented. High-order pages can cause pcp->high to be
exceeded prematurely for lower-orders so for example, a large number of
THP pages being freed could release order-0 pages from the PCP lists.
Hence, much depends on the allocation/free pattern as observed by a single
CPU to determine if caching helps or hurts a particular workload.
That said, basic performance testing passed. The following is a netperf
UDP_STREAM test which hits the relevant patches as some of the network
allocations are high-order.
netperf-udp
5.13.0-rc2 5.13.0-rc2
mm-pcpburst-v3r4 mm-pcphighorder-v1r7
Hmean send-64 261.46 ( 0.00%) 266.30 * 1.85%*
Hmean send-128 516.35 ( 0.00%) 536.78 * 3.96%*
Hmean send-256 1014.13 ( 0.00%) 1034.63 * 2.02%*
Hmean send-1024 3907.65 ( 0.00%) 4046.11 * 3.54%*
Hmean send-2048 7492.93 ( 0.00%) 7754.85 * 3.50%*
Hmean send-3312 11410.04 ( 0.00%) 11772.32 * 3.18%*
Hmean send-4096 13521.95 ( 0.00%) 13912.34 * 2.89%*
Hmean send-8192 21660.50 ( 0.00%) 22730.72 * 4.94%*
Hmean send-16384 31902.32 ( 0.00%) 32637.50 * 2.30%*
Functionally, a patch like this is necessary to make bulk allocation of
high-order pages work with similar performance to order-0 bulk
allocations. The bulk allocator is not updated in this series as it would
have to be determined by bulk allocation users how they want to track the
order of pages allocated with the bulk allocator.
Link: https://lkml.kernel.org/r/20210611135753.GC30378@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This introduces a new sysctl vm.percpu_pagelist_high_fraction. It is
similar to the old vm.percpu_pagelist_fraction. The old sysctl increased
both pcp->batch and pcp->high with the higher pcp->high potentially
reducing zone->lock contention. However, the higher pcp->batch value also
potentially increased allocation latency while the PCP was refilled. This
sysctl only adjusts pcp->high so that zone->lock contention is potentially
reduced but allocation latency during a PCP refill remains the same.
# grep -E "high:|batch" /proc/zoneinfo | tail -2
high: 649
batch: 63
# sysctl vm.percpu_pagelist_high_fraction=8
# grep -E "high:|batch" /proc/zoneinfo | tail -2
high: 35071
batch: 63
# sysctl vm.percpu_pagelist_high_fraction=64
high: 4383
batch: 63
# sysctl vm.percpu_pagelist_high_fraction=0
high: 649
batch: 63
[mgorman@techsingularity.net: fix documentation]
Link: https://lkml.kernel.org/r/20210528151010.GQ30378@techsingularity.net
Link: https://lkml.kernel.org/r/20210525080119.5455-7-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Dave Hansen <dave.hansen@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When a task is freeing a large number of order-0 pages, it may acquire the
zone->lock multiple times freeing pages in batches. This may
unnecessarily contend on the zone lock when freeing very large number of
pages. This patch adapts the size of the batch based on the recent
pattern to scale the batch size for subsequent frees.
As the machines I used were not large enough to test this are not large
enough to illustrate a problem, a debugging patch shows patterns like the
following (slightly editted for clarity)
Baseline vanilla kernel
time-unmap-14426 [...] free_pcppages_bulk: free 63 count 378 high 378
time-unmap-14426 [...] free_pcppages_bulk: free 63 count 378 high 378
time-unmap-14426 [...] free_pcppages_bulk: free 63 count 378 high 378
time-unmap-14426 [...] free_pcppages_bulk: free 63 count 378 high 378
time-unmap-14426 [...] free_pcppages_bulk: free 63 count 378 high 378
With patches
time-unmap-7724 [...] free_pcppages_bulk: free 126 count 814 high 814
time-unmap-7724 [...] free_pcppages_bulk: free 252 count 814 high 814
time-unmap-7724 [...] free_pcppages_bulk: free 504 count 814 high 814
time-unmap-7724 [...] free_pcppages_bulk: free 751 count 814 high 814
time-unmap-7724 [...] free_pcppages_bulk: free 751 count 814 high 814
Link: https://lkml.kernel.org/r/20210525080119.5455-5-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Dave Hansen <dave.hansen@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The PCP high watermark is based on the number of online CPUs so the
watermarks must be adjusted during CPU hotplug. At the time of
hot-remove, the number of online CPUs is already adjusted but during
hot-add, a delta needs to be applied to update PCP to the correct value.
After this patch is applied, the high watermarks are adjusted correctly.
# grep high: /proc/zoneinfo | tail -1
high: 649
# echo 0 > /sys/devices/system/cpu/cpu4/online
# grep high: /proc/zoneinfo | tail -1
high: 664
# echo 1 > /sys/devices/system/cpu/cpu4/online
# grep high: /proc/zoneinfo | tail -1
high: 649
Link: https://lkml.kernel.org/r/20210525080119.5455-4-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The pcp high watermark is based on the batch size but there is no
relationship between them other than it is convenient to use early in
boot.
This patch takes the first step and bases pcp->high on the zone low
watermark split across the number of CPUs local to a zone while the batch
size remains the same to avoid increasing allocation latencies. The
intent behind the default pcp->high is "set the number of PCP pages such
that if they are all full that background reclaim is not started
prematurely".
Note that in this patch the pcp->high values are adjusted after memory
hotplug events, min_free_kbytes adjustments and watermark scale factor
adjustments but not CPU hotplug events which is handled later in the
series.
On a test KVM instance;
Before grep -E "high:|batch" /proc/zoneinfo | tail -2
high: 378
batch: 63
After grep -E "high:|batch" /proc/zoneinfo | tail -2
high: 649
batch: 63
[mgorman@techsingularity.net: fix __setup_per_zone_wmarks for parallel memory
hotplug]
Link: https://lkml.kernel.org/r/20210528105925.GN30378@techsingularity.net
Link: https://lkml.kernel.org/r/20210525080119.5455-3-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "Calculate pcp->high based on zone sizes and active CPUs", v2.
The per-cpu page allocator (PCP) is meant to reduce contention on the zone
lock but the sizing of batch and high is archaic and neither takes the
zone size into account or the number of CPUs local to a zone. With larger
zones and more CPUs per node, the contention is getting worse.
Furthermore, the fact that vm.percpu_pagelist_fraction adjusts both batch
and high values means that the sysctl can reduce zone lock contention but
also increase allocation latencies.
This series disassociates pcp->high from pcp->batch and then scales
pcp->high based on the size of the local zone with limited impact to
reclaim and accounting for active CPUs but leaves pcp->batch static. It
also adapts the number of pages that can be on the pcp list based on
recent freeing patterns.
The motivation is partially to adjust to larger memory sizes but is also
driven by the fact that large batches of page freeing via release_pages()
often shows zone contention as a major part of the problem. Another is a
bug report based on an older kernel where a multi-terabyte process can
takes several minutes to exit. A workaround was to use
vm.percpu_pagelist_fraction to increase the pcp->high value but testing
indicated that a production workload could not use the same values because
of an increase in allocation latencies. Unfortunately, I cannot reproduce
this test case myself as the multi-terabyte machines are in active use but
it should alleviate the problem.
The series aims to address both and partially acts as a pre-requisite.
pcp only works with order-0 which is useless for SLUB (when using high
orders) and THP (unconditionally). To store high-order pages on PCP, the
pcp->high values need to be increased first.
This patch (of 6):
The vm.percpu_pagelist_fraction is used to increase the batch and high
limits for the per-cpu page allocator (PCP). The intent behind the sysctl
is to reduce zone lock acquisition when allocating/freeing pages but it
has a problem. While it can decrease contention, it can also increase
latency on the allocation side due to unreasonably large batch sizes.
This leads to games where an administrator adjusts
percpu_pagelist_fraction on the fly to work around contention and
allocation latency problems.
This series aims to alleviate the problems with zone lock contention while
avoiding the allocation-side latency problems. For the purposes of
review, it's easier to remove this sysctl now and reintroduce a similar
sysctl later in the series that deals only with pcp->high.
Link: https://lkml.kernel.org/r/20210525080119.5455-1-mgorman@techsingularity.net
Link: https://lkml.kernel.org/r/20210525080119.5455-2-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Dave Hansen <dave.hansen@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
alloc_contig_dump_pages() aims for helping debugging page migration
failure by elevated page refcount compared to expected_count. (for the
detail, please look at migrate_page_move_mapping)
However, -ENOMEM is just the case that system is under memory pressure
state, not relevant with page refcount at all. Thus, the dumping page
list is not helpful for the debugging point of view.
Link: https://lkml.kernel.org/r/YKa2Wyo9xqIErpfa@google.com
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: John Dias <joaodias@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Historically when freeing pages, free_one_page() assumed that callers had
IRQs disabled and the zone->lock could be acquired with spin_lock(). This
confuses the scope of what local_lock_irq is protecting and what
zone->lock is protecting in free_unref_page_list in particular.
This patch uses spin_lock_irqsave() for the zone->lock in free_one_page()
instead of relying on callers to have disabled IRQs.
free_unref_page_commit() is changed to only deal with PCP pages protected
by the local lock. free_unref_page_list() then first frees isolated pages
to the buddy lists with free_one_page() and frees the rest of the pages to
the PCP via free_unref_page_commit(). The end result is that
free_one_page() is no longer depending on side-effects of local_lock to be
correct.
Note that this may incur a performance penalty while memory hot-remove is
running but that is not a common operation.
[lkp@intel.com: Ensure CMA pages get addded to correct pcp list]
Link: https://lkml.kernel.org/r/20210512095458.30632-9-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
NUMA statistics are maintained on the zone level for hits, misses, foreign
etc but nothing relies on them being perfectly accurate for functional
correctness. The counters are used by userspace to get a general overview
of a workloads NUMA behaviour but the page allocator incurs a high cost to
maintain perfect accuracy similar to what is required for a vmstat like
NR_FREE_PAGES. There even is a sysctl vm.numa_stat to allow userspace to
turn off the collection of NUMA statistics like NUMA_HIT.
This patch converts NUMA_HIT and friends to be NUMA events with similar
accuracy to VM events. There is a possibility that slight errors will be
introduced but the overall trend as seen by userspace will be similar.
The counters are no longer updated from vmstat_refresh context as it is
unnecessary overhead for counters that may never be read by userspace.
Note that counters could be maintained at the node level to save space but
it would have a user-visible impact due to /proc/zoneinfo.
[lkp@intel.com: Fix misplaced closing brace for !CONFIG_NUMA]
Link: https://lkml.kernel.org/r/20210512095458.30632-4-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The PCP (per-cpu page allocator in page_alloc.c) shares locking
requirements with vmstat and the zone lock which is inconvenient and
causes some issues. For example, the PCP list and vmstat share the same
per-cpu space meaning that it's possible that vmstat updates dirty cache
lines holding per-cpu lists across CPUs unless padding is used. Second,
PREEMPT_RT does not want to disable IRQs for too long in the page
allocator.
This series splits the locking requirements and uses locks types more
suitable for PREEMPT_RT, reduces the time when special locking is required
for stats and reduces the time when IRQs need to be disabled on
!PREEMPT_RT kernels.
Why local_lock? PREEMPT_RT considers the following sequence to be unsafe
as documented in Documentation/locking/locktypes.rst
local_irq_disable();
spin_lock(&lock);
The pcp allocator has this sequence for rmqueue_pcplist (local_irq_save)
-> __rmqueue_pcplist -> rmqueue_bulk (spin_lock). While it's possible to
separate this out, it generally means there are points where we enable
IRQs and reenable them again immediately. To prevent a migration and the
per-cpu pointer going stale, migrate_disable is also needed. That is a
custom lock that is similar, but worse, than local_lock. Furthermore, on
PREEMPT_RT, it's undesirable to leave IRQs disabled for too long. By
converting to local_lock which disables migration on PREEMPT_RT, the
locking requirements can be separated and start moving the protections for
PCP, stats and the zone lock to PREEMPT_RT-safe equivalent locking. As a
bonus, local_lock also means that PROVE_LOCKING does something useful.
After that, it's obvious that zone_statistics incurs too much overhead and
leaves IRQs disabled for longer than necessary on !PREEMPT_RT kernels.
zone_statistics uses perfectly accurate counters requiring IRQs be
disabled for parallel RMW sequences when inaccurate ones like vm_events
would do. The series makes the NUMA statistics (NUMA_HIT and friends)
inaccurate counters that then require no special protection on
!PREEMPT_RT.
The bulk page allocator can then do stat updates in bulk with IRQs enabled
which should improve the efficiency. Technically, this could have been
done without the local_lock and vmstat conversion work and the order
simply reflects the timing of when different series were implemented.
Finally, there are places where we conflate IRQs being disabled for the
PCP with the IRQ-safe zone spinlock. The remainder of the series reduces
the scope of what is protected by disabled IRQs on !PREEMPT_RT kernels.
By the end of the series, page_alloc.c does not call local_irq_save so the
locking scope is a bit clearer. The one exception is that modifying
NR_FREE_PAGES still happens in places where it's known the IRQs are
disabled as it's harmless for PREEMPT_RT and would be expensive to split
the locking there.
No performance data is included because despite the overhead of the stats,
it's within the noise for most workloads on !PREEMPT_RT. However, Jesper
Dangaard Brouer ran a page allocation microbenchmark on a E5-1650 v4 @
3.60GHz CPU on the first version of this series. Focusing on the array
variant of the bulk page allocator reveals the following.
(CPU: Intel(R) Xeon(R) CPU E5-1650 v4 @ 3.60GHz)
ARRAY variant: time_bulk_page_alloc_free_array: step=bulk size
Baseline Patched
1 56.383 54.225 (+3.83%)
2 40.047 35.492 (+11.38%)
3 37.339 32.643 (+12.58%)
4 35.578 30.992 (+12.89%)
8 33.592 29.606 (+11.87%)
16 32.362 28.532 (+11.85%)
32 31.476 27.728 (+11.91%)
64 30.633 27.252 (+11.04%)
128 30.596 27.090 (+11.46%)
While this is a positive outcome, the series is more likely to be
interesting to the RT people in terms of getting parts of the PREEMPT_RT
tree into mainline.
This patch (of 9):
The per-cpu page allocator lists and the per-cpu vmstat deltas are stored
in the same struct per_cpu_pages even though vmstats have no direct impact
on the per-cpu page lists. This is inconsistent because the vmstats for a
node are stored on a dedicated structure. The bigger issue is that the
per_cpu_pages structure is not cache-aligned and stat updates either cache
conflict with adjacent per-cpu lists incurring a runtime cost or padding
is required incurring a memory cost.
This patch splits the per-cpu pagelists and the vmstat deltas into
separate structures. It's mostly a mechanical conversion but some
variable renaming is done to clearly distinguish the per-cpu pages
structure (pcp) from the vmstats (pzstats).
Superficially, this appears to increase the size of the per_cpu_pages
structure but the movement of expire fills a structure hole so there is no
impact overall.
[mgorman@techsingularity.net: make it W=1 cleaner]
Link: https://lkml.kernel.org/r/20210514144622.GA3735@techsingularity.net
[mgorman@techsingularity.net: make it W=1 even cleaner]
Link: https://lkml.kernel.org/r/20210516140705.GB3735@techsingularity.net
[lkp@intel.com: check struct per_cpu_zonestat has a non-zero size]
[vbabka@suse.cz: Init zone->per_cpu_zonestats properly]
Link: https://lkml.kernel.org/r/20210512095458.30632-1-mgorman@techsingularity.net
Link: https://lkml.kernel.org/r/20210512095458.30632-2-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A customer experienced a low-memory situation and decided to issue a
SIGKILL (i.e. a fatal signal). Instead of promptly terminating as one
would expect, the aforementioned task remained unresponsive.
Further investigation indicated that the task was "stuck" in the
reclaim/compaction retry loop. Now, it does not make sense to retry
compaction when a fatal signal is pending.
In the context of try_to_compact_pages(), indeed COMPACT_SKIPPED can be
returned; albeit, not every zone, on the zone list, would be considered in
the case a fatal signal is found to be pending. Yet, in
should_compact_retry(), given the last known compaction result, each zone,
on the zone list, can be considered/or checked (see
compaction_zonelist_suitable()). For example, if a zone was found to
succeed, then reclaim/compaction would be tried again (notwithstanding the
above).
This patch ensures that compaction is not needlessly retried irrespective
of the last known compaction result e.g. if it was skipped, in the
unlikely case a fatal signal is found pending. So, OOM is at least
attempted.
Link: https://lkml.kernel.org/r/20210520142901.3371299-1-atomlin@redhat.com
Signed-off-by: Aaron Tomlin <atomlin@redhat.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "Constify struct page arguments".
While working on various solutions to the 32-bit struct page size
regression, one of the problems I found was the networking stack expects
to be able to pass const struct page pointers around, and the mm doesn't
provide a lot of const-friendly functions to call. The root tangle of
problems is that a lot of functions call VM_BUG_ON_PAGE(), which calls
dump_page(), which calls a lot of functions which don't take a const
struct page (but could be const).
This patch (of 6):
The only caller of __dump_page() now opencodes dump_page(), so remove it
as an externally visible symbol.
Link: https://lkml.kernel.org/r/20210416231531.2521383-1-willy@infradead.org
Link: https://lkml.kernel.org/r/20210416231531.2521383-2-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Recently there has been introduced a page bulk allocator for users which
need to get number of pages per one call request.
For order-0 pages switch to an alloc_pages_bulk_array_node() instead of
alloc_pages_node(), the reason is the former is not capable of allocating
set of pages, thus a one call is per one page.
Second, according to my tests the bulk allocator uses less cycles even for
scenarios when only one page is requested. Running the "perf" on same
test case shows below difference:
<default>
- 45.18% __vmalloc_node
- __vmalloc_node_range
- 35.60% __alloc_pages
- get_page_from_freelist
3.36% __list_del_entry_valid
3.00% check_preemption_disabled
1.42% prep_new_page
<default>
<patch>
- 31.00% __vmalloc_node
- __vmalloc_node_range
- 14.48% __alloc_pages_bulk
3.22% __list_del_entry_valid
- 0.83% __alloc_pages
get_page_from_freelist
<patch>
The "test_vmalloc.sh" also shows performance improvements:
fix_size_alloc_test_4MB loops: 1000000 avg: 89105095 usec
fix_size_alloc_test loops: 1000000 avg: 513672 usec
full_fit_alloc_test loops: 1000000 avg: 748900 usec
long_busy_list_alloc_test loops: 1000000 avg: 8043038 usec
random_size_alloc_test loops: 1000000 avg: 4028582 usec
fix_align_alloc_test loops: 1000000 avg: 1457671 usec
fix_size_alloc_test_4MB loops: 1000000 avg: 62083711 usec
fix_size_alloc_test loops: 1000000 avg: 449207 usec
full_fit_alloc_test loops: 1000000 avg: 735985 usec
long_busy_list_alloc_test loops: 1000000 avg: 5176052 usec
random_size_alloc_test loops: 1000000 avg: 2589252 usec
fix_align_alloc_test loops: 1000000 avg: 1365009 usec
For example 4MB allocations illustrates ~30% gain, all the
rest is also better.
Link: https://lkml.kernel.org/r/20210516202056.2120-3-urezki@gmail.com
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Oleksiy Avramchenko <oleksiy.avramchenko@sonymobile.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
