It's possible for free pages to become stranded on per-cpu pagesets
(pcps) that, if drained, could be merged with buddy pages on the zone's
free area to form large order pages, including up to MAX_ORDER.
Consider a verbose example using the tools/vm/page-types tool at the
beginning of a ZONE_NORMAL ('B' indicates a buddy page and 'S' indicates
a slab page). Pages on pcps do not have any page flags set.
109954 1 _______S________________________________________________________
109955 2 __________B_____________________________________________________
109957 1 ________________________________________________________________
109958 1 __________B_____________________________________________________
109959 7 ________________________________________________________________
109960 1 __________B_____________________________________________________
109961 9 ________________________________________________________________
10996a 1 __________B_____________________________________________________
10996b 3 ________________________________________________________________
10996e 1 __________B_____________________________________________________
10996f 1 ________________________________________________________________
...
109f8c 1 __________B_____________________________________________________
109f8d 2 ________________________________________________________________
109f8f 2 __________B_____________________________________________________
109f91 f ________________________________________________________________
109fa0 1 __________B_____________________________________________________
109fa1 7 ________________________________________________________________
109fa8 1 __________B_____________________________________________________
109fa9 1 ________________________________________________________________
109faa 1 __________B_____________________________________________________
109fab 1 _______S________________________________________________________
The compaction migration scanner is attempting to defragment this memory
since it is at the beginning of the zone. It has done so quite well,
all movable pages have been migrated. From pfn [0x109955, 0x109fab),
there are only buddy pages and pages without flags set.
These pages may be stranded on pcps that could otherwise allow this
memory to be coalesced if freed back to the zone free area. It is
possible that some of these pages may not be on pcps and that something
has called alloc_pages() and used the memory directly, but we rely on
the absence of __GFP_MOVABLE in these cases to allocate from
MIGATE_UNMOVABLE pageblocks to try to keep these MIGRATE_MOVABLE
pageblocks as free as possible.
These buddy and pcp pages, spanning 1,621 pages, could be coalesced and
allow for three transparent hugepages to be dynamically allocated.
Running the numbers for all such spans on the system, it was found that
there were over 400 such spans of only buddy pages and pages without
flags set at the time this /proc/kpageflags sample was collected.
Without this support, there were _no_ order-9 or order-10 pages free.
When kcompactd fails to defragment memory such that a cc.order page can
be allocated, drain all pcps for the zone back to the buddy allocator so
this stranding cannot occur. Compaction for that order will
subsequently be deferred, which acts as a ratelimit on this drain.
Link: http://lkml.kernel.org/r/alpine.DEB.2.20.1803010340100.88270@chino.kir.corp.google.com
Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
should_failslab() is a convenient function to hook into for directed
error injection into kmalloc(). However, it is only available if a
config flag is set.
The following BCC script, for example, fails kmalloc() calls after a
btrfs umount:
from bcc import BPF
prog = r"""
BPF_HASH(flag);
#include <linux/mm.h>
int kprobe__btrfs_close_devices(void *ctx) {
u64 key = 1;
flag.update(&key, &key);
return 0;
}
int kprobe__should_failslab(struct pt_regs *ctx) {
u64 key = 1;
u64 *res;
res = flag.lookup(&key);
if (res != 0) {
bpf_override_return(ctx, -ENOMEM);
}
return 0;
}
"""
b = BPF(text=prog)
while 1:
b.kprobe_poll()
This patch refactors the should_failslab implementation so that the
function is always available for error injection, independent of flags.
This change would be similar in nature to commit f5490d3ec921 ("block:
Add should_fail_bio() for bpf error injection").
Link: http://lkml.kernel.org/r/20180222020320.6944-1-hmclauchlan@fb.com
Signed-off-by: Howard McLauchlan <hmclauchlan@fb.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Johannes Weiner <jweiner@fb.com>
Cc: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
THP split makes non-atomic change of tail page flags. This is almost ok
because tail pages are locked and isolated but this breaks recent
changes in page locking: non-atomic operation could clear bit
PG_waiters.
As a result concurrent sequence get_page_unless_zero() -> lock_page()
might block forever. Especially if this page was truncated later.
Fix is trivial: clone flags before unfreezing page reference counter.
This race exists since commit 6290602709 ("mm: add PageWaiters
indicating tasks are waiting for a page bit") while unsave unfreeze
itself was added in commit 8df651c705 ("thp: cleanup
split_huge_page()").
clear_compound_head() also must be called before unfreezing page
reference because after successful get_page_unless_zero() might follow
put_page() which needs correct compound_head().
And replace page_ref_inc()/page_ref_add() with page_ref_unfreeze() which
is made especially for that and has semantic of smp_store_release().
Link: http://lkml.kernel.org/r/151844393341.210639.13162088407980624477.stgit@buzz
Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When page_mapping() is called and the mapping is dereferenced in
page_evicatable() through shrink_active_list(), it is possible for the
inode to be truncated and the embedded address space to be freed at the
same time. This may lead to the following race.
CPU1 CPU2
truncate(inode) shrink_active_list()
... page_evictable(page)
truncate_inode_page(mapping, page);
delete_from_page_cache(page)
spin_lock_irqsave(&mapping->tree_lock, flags);
__delete_from_page_cache(page, NULL)
page_cache_tree_delete(..)
... mapping = page_mapping(page);
page->mapping = NULL;
...
spin_unlock_irqrestore(&mapping->tree_lock, flags);
page_cache_free_page(mapping, page)
put_page(page)
if (put_page_testzero(page)) -> false
- inode now has no pages and can be freed including embedded address_space
mapping_unevictable(mapping)
test_bit(AS_UNEVICTABLE, &mapping->flags);
- we've dereferenced mapping which is potentially already free.
Similar race exists between swap cache freeing and page_evicatable()
too.
The address_space in inode and swap cache will be freed after a RCU
grace period. So the races are fixed via enclosing the page_mapping()
and address_space usage in rcu_read_lock/unlock(). Some comments are
added in code to make it clear what is protected by the RCU read lock.
Link: http://lkml.kernel.org/r/20180212081227.1940-1-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Both kernelcore= and movablecore= can be used to define the amount of
ZONE_NORMAL and ZONE_MOVABLE on a system, respectively. This requires
the system memory capacity to be known when specifying the command line,
however.
This introduces the ability to define both kernelcore= and movablecore=
as a percentage of total system memory. This is convenient for systems
software that wants to define the amount of ZONE_MOVABLE, for example,
as a proportion of a system's memory rather than a hardcoded byte value.
To define the percentage, the final character of the parameter should be
a '%'.
mhocko: "why is anyone using these options nowadays?"
rientjes:
:
: Fragmentation of non-__GFP_MOVABLE pages due to low on memory
: situations can pollute most pageblocks on the system, as much as 1GB of
: slab being fragmented over 128GB of memory, for example. When the
: amount of kernel memory is well bounded for certain systems, it is
: better to aggressively reclaim from existing MIGRATE_UNMOVABLE
: pageblocks rather than eagerly fallback to others.
:
: We have additional patches that help with this fragmentation if you're
: interested, specifically kcompactd compaction of MIGRATE_UNMOVABLE
: pageblocks triggered by fallback of non-__GFP_MOVABLE allocations and
: draining of pcp lists back to the zone free area to prevent stranding.
[rientjes@google.com: updates]
Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1802131700160.71590@chino.kir.corp.google.com
Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1802121622470.179479@chino.kir.corp.google.com
Signed-off-by: David Rientjes <rientjes@google.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Recently the following BUG was reported:
Injecting memory failure for pfn 0x3c0000 at process virtual address 0x7fe300000000
Memory failure: 0x3c0000: recovery action for huge page: Recovered
BUG: unable to handle kernel paging request at ffff8dfcc0003000
IP: gup_pgd_range+0x1f0/0xc20
PGD 17ae72067 P4D 17ae72067 PUD 0
Oops: 0000 [#1] SMP PTI
...
CPU: 3 PID: 5467 Comm: hugetlb_1gb Not tainted 4.15.0-rc8-mm1-abc+ #3
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.9.3-1.fc25 04/01/2014
You can easily reproduce this by calling madvise(MADV_HWPOISON) twice on
a 1GB hugepage. This happens because get_user_pages_fast() is not aware
of a migration entry on pud that was created in the 1st madvise() event.
I think that conversion to pud-aligned migration entry is working, but
other MM code walking over page table isn't prepared for it. We need
some time and effort to make all this work properly, so this patch
avoids the reported bug by just disabling error handling for 1GB
hugepage.
[n-horiguchi@ah.jp.nec.com: v2]
Link: http://lkml.kernel.org/r/1517284444-18149-1-git-send-email-n-horiguchi@ah.jp.nec.com
Link: http://lkml.kernel.org/r/1517207283-15769-1-git-send-email-n-horiguchi@ah.jp.nec.com
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Punit Agrawal <punit.agrawal@arm.com>
Tested-by: Michael Ellerman <mpe@ellerman.id.au>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
During memory hotplugging we traverse struct pages three times:
1. memset(0) in sparse_add_one_section()
2. loop in __add_section() to set do: set_page_node(page, nid); and
SetPageReserved(page);
3. loop in memmap_init_zone() to call __init_single_pfn()
This patch removes the first two loops, and leaves only loop 3. All
struct pages are initialized in one place, the same as it is done during
boot.
The benefits:
- We improve memory hotplug performance because we are not evicting the
cache several times and also reduce loop branching overhead.
- Remove condition from hotpath in __init_single_pfn(), that was added
in order to fix the problem that was reported by Bharata in the above
email thread, thus also improve performance during normal boot.
- Make memory hotplug more similar to the boot memory initialization
path because we zero and initialize struct pages only in one
function.
- Simplifies memory hotplug struct page initialization code, and thus
enables future improvements, such as multi-threading the
initialization of struct pages in order to improve hotplug
performance even further on larger machines.
[pasha.tatashin@oracle.com: v5]
Link: http://lkml.kernel.org/r/20180228030308.1116-7-pasha.tatashin@oracle.com
Link: http://lkml.kernel.org/r/20180215165920.8570-7-pasha.tatashin@oracle.com
Signed-off-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Bharata B Rao <bharata@linux.vnet.ibm.com>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "optimize memory hotplug", v3.
This patchset:
- Improves hotplug performance by eliminating a number of struct page
traverses during memory hotplug.
- Fixes some issues with hotplugging, where boundaries were not
properly checked. And on x86 block size was not properly aligned with
end of memory
- Also, potentially improves boot performance by eliminating condition
from __init_single_page().
- Adds robustness by verifying that that struct pages are correctly
poisoned when flags are accessed.
The following experiments were performed on Xeon(R) CPU E7-8895 v3 @
2.60GHz with 1T RAM:
booting in qemu with 960G of memory, time to initialize struct pages:
no-kvm:
TRY1 TRY2
BEFORE: 39.433668 39.39705
AFTER: 36.903781 36.989329
with-kvm:
BEFORE: 10.977447 11.103164
AFTER: 10.929072 10.751885
Hotplug 896G memory:
no-kvm:
TRY1 TRY2
BEFORE: 848.740000 846.910000
AFTER: 783.070000 786.560000
with-kvm:
TRY1 TRY2
BEFORE: 34.410000 33.57
AFTER: 29.810000 29.580000
This patch (of 6):
Start qemu with the following arguments:
-m 64G,slots=2,maxmem=66G -object memory-backend-ram,id=mem1,size=2G
Which: boots machine with 64G, and adds a device mem1 with 2G which can
be hotplugged later.
Also make sure that config has the following turned on:
CONFIG_MEMORY_HOTPLUG
CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE
CONFIG_ACPI_HOTPLUG_MEMORY
Using the qemu monitor hotplug the memory (make sure config has (qemu)
device_add pc-dimm,id=dimm1,memdev=mem1
The operation will fail with the following trace:
WARNING: CPU: 0 PID: 91 at drivers/base/memory.c:205
pages_correctly_reserved+0xe6/0x110
Modules linked in:
CPU: 0 PID: 91 Comm: systemd-udevd Not tainted 4.16.0-rc1_pt_master #29
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
BIOS rel-1.11.0-0-g63451fca13-prebuilt.qemu-project.org 04/01/2014
RIP: 0010:pages_correctly_reserved+0xe6/0x110
Call Trace:
memory_subsys_online+0x44/0xa0
device_online+0x51/0x80
store_mem_state+0x5e/0xe0
kernfs_fop_write+0xfa/0x170
__vfs_write+0x2e/0x150
vfs_write+0xa8/0x1a0
SyS_write+0x4d/0xb0
do_syscall_64+0x5d/0x110
entry_SYSCALL_64_after_hwframe+0x21/0x86
---[ end trace 6203bc4f1a5d30e8 ]---
The problem is detected in: drivers/base/memory.c
static bool pages_correctly_reserved(unsigned long start_pfn)
205 if (WARN_ON_ONCE(!pfn_valid(pfn)))
This function loops through every section in the newly added memory
block and verifies that the first pfn is valid, meaning section exists,
has mapping (struct page array), and is online.
The block size on x86 is usually 128M, but when machine is booted with
more than 64G of memory, the block size is changed to 2G: $ cat
/sys/devices/system/memory/block_size_bytes 80000000
or
$ dmesg | grep "block size"
[ 0.086469] x86/mm: Memory block size: 2048MB
During memory hotplug, and hotremove we verify that the range is section
size aligned, but we actually must verify that it is block size aligned,
because that is the proper unit for hotplug operations. See:
Documentation/memory-hotplug.txt
So, when the start_pfn of newly added memory is not block size aligned,
we can get a memory block that has only part of it with properly
populated sections.
In our case the start_pfn starts from the last_pfn (end of physical
memory).
$ dmesg | grep last_pfn
[ 0.000000] e820: last_pfn = 0x1040000 max_arch_pfn = 0x400000000
0x1040000 == 65G, and so is not 2G aligned!
The fix is to enforce that memory that is hotplugged and hotremoved is
block size aligned.
With this fix, running the above sequence yield to the following result:
(qemu) device_add pc-dimm,id=dimm1,memdev=mem1
Block size [0x80000000] unaligned hotplug range: start 0x1040000000,
size 0x80000000
acpi PNP0C80:00: add_memory failed
acpi PNP0C80:00: acpi_memory_enable_device() error
acpi PNP0C80:00: Enumeration failure
Link: http://lkml.kernel.org/r/20180213193159.14606-2-pasha.tatashin@oracle.com
Signed-off-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Bharata B Rao <bharata@linux.vnet.ibm.com>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The kasan quarantine is designed to delay freeing slab objects to catch
use-after-free. The quarantine can be large (several percent of machine
memory size). When kmem_caches are deleted related objects are flushed
from the quarantine but this requires scanning the entire quarantine
which can be very slow. We have seen the kernel busily working on this
while holding slab_mutex and badly affecting cache_reaper, slabinfo
readers and memcg kmem cache creations.
It can easily reproduced by following script:
yes . | head -1000000 | xargs stat > /dev/null
for i in `seq 1 10`; do
seq 500 | (cd /cg/memory && xargs mkdir)
seq 500 | xargs -I{} sh -c 'echo $BASHPID > \
/cg/memory/{}/tasks && exec stat .' > /dev/null
seq 500 | (cd /cg/memory && xargs rmdir)
done
The busy stack:
kasan_cache_shutdown
shutdown_cache
memcg_destroy_kmem_caches
mem_cgroup_css_free
css_free_rwork_fn
process_one_work
worker_thread
kthread
ret_from_fork
This patch is based on the observation that if the kmem_cache to be
destroyed is empty then there should not be any objects of this cache in
the quarantine.
Without the patch the script got stuck for couple of hours. With the
patch the script completed within a second.
Link: http://lkml.kernel.org/r/20180327230603.54721-1-shakeelb@google.com
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I have noticed on debug kernel with SLAB, the size of some non-root
slabs were larger than their corresponding root slabs.
e.g. for radix_tree_node:
$cat /proc/slabinfo | grep radix
name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab> ...
radix_tree_node 15052 15075 4096 1 1 ...
$cat /cgroup/memory/temp/memory.kmem.slabinfo | grep radix
name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab> ...
radix_tree_node 1581 158 4120 1 2 ...
However for SLUB in debug kernel, the sizes were same. On further
inspection it is found that SLUB always use kmem_cache.object_size to
measure the kmem_cache.size while SLAB use the given kmem_cache.size.
In the debug kernel the slab's size can be larger than its object_size.
Thus in the creation of non-root slab, the SLAB uses the root's size as
base to calculate the non-root slab's size and thus non-root slab's size
can be larger than the root slab's size. For SLUB, the non-root slab's
size is measured based on the root's object_size and thus the size will
remain same for root and non-root slab.
This patch makes slab's object_size the default base to measure the
slab's size.
Link: http://lkml.kernel.org/r/20180313165428.58699-1-shakeelb@google.com
Fixes: 794b1248be ("memcg, slab: separate memcg vs root cache creation paths")
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
