Prior to "percpu: implement partial chunk depopulation",
pcpu_depopulate_chunk() was called only on the destruction path. This
meant the virtual address range was on its way back to vmalloc which
will handle flushing the tlbs for us.
However, with pcpu_reclaim_populated(), we are now calling
pcpu_depopulate_chunk() during the active lifecycle of a chunk.
Therefore, we need to flush the tlb as well otherwise we can end up
accessing the wrong page through an invalid tlb mapping as reported in
[1].
[1] https://lore.kernel.org/lkml/20210702191140.GA3166599@roeck-us.net/
Fixes: f183324133 ("percpu: implement partial chunk depopulation")
Reported-and-tested-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
The current implementation of the memcg accounting of the percpu
memory is based on the idea of having two separate sets of chunks for
accounted and non-accounted memory. This approach has an advantage
of not wasting any extra memory for memcg data for non-accounted
chunks, however it complicates the code and leads to a higher chunks
number due to a lower chunk utilization.
Instead of having two chunk types it's possible to declare all* chunks
memcg-aware unless the kernel memory accounting is disabled globally
by a boot option. The size of objcg_array is usually small in
comparison to chunks themselves (it obviously depends on the number of
CPUs), so even if some chunk will have no accounted allocations, the
memory waste isn't significant and will likely be compensated by
a higher chunk utilization. Also, with time more and more percpu
allocations will likely become accounted.
* The first chunk is initialized before the memory cgroup subsystem,
so we don't know for sure whether we need to allocate obj_cgroups.
Because it's small, let's make it free for use. Then we don't need
to allocate obj_cgroups for it.
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
From Roman ("percpu: partial chunk depopulation"):
In our [Facebook] production experience the percpu memory allocator is
sometimes struggling with returning the memory to the system. A typical
example is a creation of several thousands memory cgroups (each has
several chunks of the percpu data used for vmstats, vmevents,
ref counters etc). Deletion and complete releasing of these cgroups
doesn't always lead to a shrinkage of the percpu memory, so that
sometimes there are several GB's of memory wasted.
The underlying problem is the fragmentation: to release an underlying
chunk all percpu allocations should be released first. The percpu
allocator tends to top up chunks to improve the utilization. It means
new small-ish allocations (e.g. percpu ref counters) are placed onto
almost filled old-ish chunks, effectively pinning them in memory.
This patchset solves this problem by implementing a partial depopulation
of percpu chunks: chunks with many empty pages are being asynchronously
depopulated and the pages are returned to the system.
To illustrate the problem the following script can be used:
--
cd /sys/fs/cgroup
mkdir percpu_test
echo "+memory" > percpu_test/cgroup.subtree_control
cat /proc/meminfo | grep Percpu
for i in `seq 1 1000`; do
mkdir percpu_test/cg_"${i}"
for j in `seq 1 10`; do
mkdir percpu_test/cg_"${i}"_"${j}"
done
done
cat /proc/meminfo | grep Percpu
for i in `seq 1 1000`; do
for j in `seq 1 10`; do
rmdir percpu_test/cg_"${i}"_"${j}"
done
done
sleep 10
cat /proc/meminfo | grep Percpu
for i in `seq 1 1000`; do
rmdir percpu_test/cg_"${i}"
done
rmdir percpu_test
--
It creates 11000 memory cgroups and removes every 10 out of 11.
It prints the initial size of the percpu memory, the size after
creating all cgroups and the size after deleting most of them.
Results:
vanilla:
./percpu_test.sh
Percpu: 7488 kB
Percpu: 481152 kB
Percpu: 481152 kB
with this patchset applied:
./percpu_test.sh
Percpu: 7488 kB
Percpu: 481408 kB
Percpu: 135552 kB
The total size of the percpu memory was reduced by more than 3.5 times.
This patch:
This patch implements partial depopulation of percpu chunks.
As of now, a chunk can be depopulated only as a part of the final
destruction, if there are no more outstanding allocations. However
to minimize a memory waste it might be useful to depopulate a
partially filed chunk, if a small number of outstanding allocations
prevents the chunk from being fully reclaimed.
This patch implements the following depopulation process: it scans
over the chunk pages, looks for a range of empty and populated pages
and performs the depopulation. To avoid races with new allocations,
the chunk is previously isolated. After the depopulation the chunk is
sidelined to a special list or freed. New allocations prefer using
active chunks to sidelined chunks. If a sidelined chunk is used, it is
reintegrated to the active lists.
The depopulation is scheduled on the free path if the chunk is all of
the following:
1) has more than 1/4 of total pages free and populated
2) the system has enough free percpu pages aside of this chunk
3) isn't the reserved chunk
4) isn't the first chunk
If it's already depopulated but got free populated pages, it's a good
target too. The chunk is moved to a special slot,
pcpu_to_depopulate_slot, chunk->isolated is set, and the balance work
item is scheduled. On isolation, these pages are removed from the
pcpu_nr_empty_pop_pages. It is constantly replaced to the
to_depopulate_slot when it meets these qualifications.
pcpu_reclaim_populated() iterates over the to_depopulate_slot until it
becomes empty. The depopulation is performed in the reverse direction to
keep populated pages close to the beginning. Depopulated chunks are
sidelined to preferentially avoid them for new allocations. When no
active chunk can suffice a new allocation, sidelined chunks are first
checked before creating a new chunk.
Signed-off-by: Roman Gushchin <guro@fb.com>
Co-developed-by: Dennis Zhou <dennis@kernel.org>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Tested-by: Pratik Sampat <psampat@linux.ibm.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Percpu memory is becoming more and more widely used by various subsystems,
and the total amount of memory controlled by the percpu allocator can make
a good part of the total memory.
As an example, bpf maps can consume a lot of percpu memory, and they are
created by a user. Also, some cgroup internals (e.g. memory controller
statistics) can be quite large. On a machine with many CPUs and big
number of cgroups they can consume hundreds of megabytes.
So the lack of memcg accounting is creating a breach in the memory
isolation. Similar to the slab memory, percpu memory should be accounted
by default.
To implement the perpcu accounting it's possible to take the slab memory
accounting as a model to follow. Let's introduce two types of percpu
chunks: root and memcg. What makes memcg chunks different is an
additional space allocated to store memcg membership information. If
__GFP_ACCOUNT is passed on allocation, a memcg chunk should be be used.
If it's possible to charge the corresponding size to the target memory
cgroup, allocation is performed, and the memcg ownership data is recorded.
System-wide allocations are performed using root chunks, so there is no
additional memory overhead.
To implement a fast reparenting of percpu memory on memcg removal, we
don't store mem_cgroup pointers directly: instead we use obj_cgroup API,
introduced for slab accounting.
[akpm@linux-foundation.org: fix CONFIG_MEMCG_KMEM=n build errors and warning]
[akpm@linux-foundation.org: move unreachable code, per Roman]
[cuibixuan@huawei.com: mm/percpu: fix 'defined but not used' warning]
Link: http://lkml.kernel.org/r/6d41b939-a741-b521-a7a2-e7296ec16219@huawei.com
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Bixuan Cui <cuibixuan@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Dennis Zhou <dennis@kernel.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Tobin C. Harding <tobin@kernel.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Waiman Long <longman@redhat.com>
Cc: Bixuan Cui <cuibixuan@huawei.com>
Cc: Michal Koutný <mkoutny@suse.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Link: http://lkml.kernel.org/r/20200623184515.4132564-3-guro@fb.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Based on 1 normalized pattern(s):
this file is released under the gplv2
extracted by the scancode license scanner the SPDX license identifier
GPL-2.0-only
has been chosen to replace the boilerplate/reference in 68 file(s).
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Armijn Hemel <armijn@tjaldur.nl>
Reviewed-by: Allison Randal <allison@lohutok.net>
Cc: linux-spdx@vger.kernel.org
Link: https://lkml.kernel.org/r/20190531190114.292346262@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Previously, block size was flexible based on the constraint that the
GCD(PCPU_BITMAP_BLOCK_SIZE, PAGE_SIZE) > 1. However, this carried the
overhead that keeping a floating number of populated free pages required
scanning over the free regions of a chunk.
Setting the block size to be fixed at PAGE_SIZE lets us know when an
empty page becomes used as we will break a full contig_hint of a block.
This means we no longer have to scan the whole chunk upon breaking a
contig_hint which empty page management piggybacked off. A later patch
takes advantage of this to optimize the allocation path by only scanning
forward using the scan_hint introduced later too.
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
percpu-km is used on UP systems which only has one group,
so the group offset will be always 0, there is no need
to subtract pcpu_group_offsets[0] when assigning chunk->base_addr
Signed-off-by: Peng Fan <peng.fan@nxp.com>
Acked-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
From Michael Cree:
"Bisection lead to commit b38d08f318 ("percpu: restructure
locking") as being the cause of lockups at initial boot on
the kernel built for generic Alpha.
On a suggestion by Tejun Heo that:
So, the only thing I can think of is that it's calling
spin_unlock_irq() while irq handling isn't set up yet.
Can you please try the followings?
1. Convert all spin_[un]lock_irq() to
spin_lock_irqsave/unlock_irqrestore()."
Fixes: b38d08f318 ("percpu: restructure locking")
Reported-and-tested-by: Michael Cree <mcree@orcon.net.nz>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
The prior patch added support for passing gfp flags through to the
underlying allocators. This patch allows users to pass along gfp flags
(currently only __GFP_NORETRY and __GFP_NOWARN) to the underlying
allocators. This should allow users to decide if they are ok with
failing allocations recovering in a more graceful way.
Additionally, gfp passing was done as additional flags in the previous
patch. Instead, change this to caller passed semantics. GFP_KERNEL is
also removed as the default flag. It continues to be used for internally
caused underlying percpu allocations.
V2:
Removed gfp_percpu_mask in favor of doing it inline.
Removed GFP_KERNEL as a default flag for __alloc_percpu_gfp.
Signed-off-by: Dennis Zhou <dennisszhou@gmail.com>
Suggested-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Percpu memory using the vmalloc area based chunk allocator lazily
populates chunks by first requesting the full virtual address space
required for the chunk and subsequently adding pages as allocations come
through. To ensure atomic allocations can succeed, a workqueue item is
used to maintain a minimum number of empty pages. In certain scenarios,
such as reported in [1], it is possible that physical memory becomes
quite scarce which can result in either a rather long time spent trying
to find free pages or worse, a kernel panic.
This patch adds support for __GFP_NORETRY and __GFP_NOWARN passing them
through to the underlying allocators. This should prevent any
unnecessary panics potentially caused by the workqueue item. The passing
of gfp around is as additional flags rather than a full set of flags.
The next patch will change these to caller passed semantics.
V2:
Added const modifier to gfp flags in the balance path.
Removed an extra whitespace.
[1] https://lkml.org/lkml/2018/2/12/551
Signed-off-by: Dennis Zhou <dennisszhou@gmail.com>
Suggested-by: Daniel Borkmann <daniel@iogearbox.net>
Reported-by: syzbot+adb03f3f0bb57ce3acda@syzkaller.appspotmail.com
Acked-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
The percpu memory allocator is experiencing scalability issues when
allocating and freeing large numbers of counters as in BPF.
Additionally, there is a corner case where iteration is triggered over
all chunks if the contig_hint is the right size, but wrong alignment.
This patch replaces the area map allocator with a basic bitmap allocator
implementation. Each subsequent patch will introduce new features and
replace full scanning functions with faster non-scanning options when
possible.
Implementation:
This patchset removes the area map allocator in favor of a bitmap
allocator backed by metadata blocks. The primary goal is to provide
consistency in performance and memory footprint with a focus on small
allocations (< 64 bytes). The bitmap removes the heavy memmove from the
freeing critical path and provides a consistent memory footprint. The
metadata blocks provide a bound on the amount of scanning required by
maintaining a set of hints.
In an effort to make freeing fast, the metadata is updated on the free
path if the new free area makes a page free, a block free, or spans
across blocks. This causes the chunk's contig hint to potentially be
smaller than what it could allocate by up to the smaller of a page or a
block. If the chunk's contig hint is contained within a block, a check
occurs and the hint is kept accurate. Metadata is always kept accurate
on allocation, so there will not be a situation where a chunk has a
later contig hint than available.
Evaluation:
I have primarily done testing against a simple workload of allocation of
1 million objects (2^20) of varying size. Deallocation was done by in
order, alternating, and in reverse. These numbers were collected after
rebasing ontop of a80099a152. I present the worst-case numbers here:
Area Map Allocator:
Object Size | Alloc Time (ms) | Free Time (ms)
----------------------------------------------
4B | 310 | 4770
16B | 557 | 1325
64B | 436 | 273
256B | 776 | 131
1024B | 3280 | 122
Bitmap Allocator:
Object Size | Alloc Time (ms) | Free Time (ms)
----------------------------------------------
4B | 490 | 70
16B | 515 | 75
64B | 610 | 80
256B | 950 | 100
1024B | 3520 | 200
This data demonstrates the inability for the area map allocator to
handle less than ideal situations. In the best case of reverse
deallocation, the area map allocator was able to perform within range
of the bitmap allocator. In the worst case situation, freeing took
nearly 5 seconds for 1 million 4-byte objects. The bitmap allocator
dramatically improves the consistency of the free path. The small
allocations performed nearly identical regardless of the freeing
pattern.
While it does add to the allocation latency, the allocation scenario
here is optimal for the area map allocator. The area map allocator runs
into trouble when it is allocating in chunks where the latter half is
full. It is difficult to replicate this, so I present a variant where
the pages are second half filled. Freeing was done sequentially. Below
are the numbers for this scenario:
Area Map Allocator:
Object Size | Alloc Time (ms) | Free Time (ms)
----------------------------------------------
4B | 4118 | 4892
16B | 1651 | 1163
64B | 598 | 285
256B | 771 | 158
1024B | 3034 | 160
Bitmap Allocator:
Object Size | Alloc Time (ms) | Free Time (ms)
----------------------------------------------
4B | 481 | 67
16B | 506 | 69
64B | 636 | 75
256B | 892 | 90
1024B | 3262 | 147
The data shows a parabolic curve of performance for the area map
allocator. This is due to the memmove operation being the dominant cost
with the lower object sizes as more objects are packed in a chunk and at
higher object sizes, the traversal of the chunk slots is the dominating
cost. The bitmap allocator suffers this problem as well. The above data
shows the inability to scale for the allocation path with the area map
allocator and that the bitmap allocator demonstrates consistent
performance in general.
The second problem of additional scanning can result in the area map
allocator completing in 52 minutes when trying to allocate 1 million
4-byte objects with 8-byte alignment. The same workload takes
approximately 16 seconds to complete for the bitmap allocator.
V2:
Fixed a bug in pcpu_alloc_first_chunk end_offset was setting the bitmap
using bytes instead of bits.
Added a comment to pcpu_cnt_pop_pages to explain bitmap_weight.
Signed-off-by: Dennis Zhou <dennisszhou@gmail.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Add support for tracepoints to the following events: chunk allocation,
chunk free, area allocation, area free, and area allocation failure.
This should let us replay percpu memory requests and evaluate
corresponding decisions.
Signed-off-by: Dennis Zhou <dennisz@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
There is limited visibility into the use of percpu memory leaving us
unable to reason about correctness of parameters and overall use of
percpu memory. These counters and statistics aim to help understand
basic statistics about percpu memory such as number of allocations over
the lifetime, allocation sizes, and fragmentation.
New Config: PERCPU_STATS
Signed-off-by: Dennis Zhou <dennisz@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Use the normal mechanism to make the logging output consistently
"percpu:" instead of a mix of "PERCPU:" and "percpu:"
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Most of the mm subsystem uses pr_<level> so make it consistent.
Miscellanea:
- Realign arguments
- Add missing newline to format
- kmemleak-test.c has a "kmemleak: " prefix added to the
"Kmemleak testing" logging message via pr_fmt
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Tejun Heo <tj@kernel.org> [percpu]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
pcpu_nr_empty_pop_pages counts the number of empty populated pages
across all chunks and chunk->nr_populated counts the number of
populated pages in a chunk. Both will be used to implement pre/async
population for atomic allocations.
pcpu_chunk_[de]populated() are added to update chunk->populated,
chunk->nr_populated and pcpu_nr_empty_pop_pages together. All
successful chunk [de]populations should be followed by the
corresponding pcpu_chunk_[de]populated() calls.
Signed-off-by: Tejun Heo <tj@kernel.org>
At first, the percpu allocator required a sleepable context for both
alloc and free paths and used pcpu_alloc_mutex to protect everything.
Later, pcpu_lock was introduced to protect the index data structure so
that the free path can be invoked from atomic contexts. The
conversion only updated what's necessary and left most of the
allocation path under pcpu_alloc_mutex.
The percpu allocator is planned to add support for atomic allocation
and this patch restructures locking so that the coverage of
pcpu_alloc_mutex is further reduced.
* pcpu_alloc() now grab pcpu_alloc_mutex only while creating a new
chunk and populating the allocated area. Everything else is now
protected soley by pcpu_lock.
After this change, multiple instances of pcpu_extend_area_map() may
race but the function already implements sufficient synchronization
using pcpu_lock.
This also allows multiple allocators to arrive at new chunk
creation. To avoid creating multiple empty chunks back-to-back, a
new chunk is created iff there is no other empty chunk after
grabbing pcpu_alloc_mutex.
* pcpu_lock is now held while modifying chunk->populated bitmap.
After this, all data structures are protected by pcpu_lock.
Signed-off-by: Tejun Heo <tj@kernel.org>
percpu-km instantiates the whole chunk on creation and doesn't make
use of chunk->populated bitmap and leaves it as zero. While this
currently doesn't cause any problem, the inconsistency makes it
difficult to build further logic on top of chunk->populated. This
patch makes percpu-km fill chunk->populated on creation so that the
bitmap is always consistent.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Christoph Lameter <cl@linux.com>
Previously, pcpu_[de]populate_chunk() were called with the range which
may contain multiple target regions in it and
pcpu_[de]populate_chunk() iterated over the regions. This has the
benefit of batching up cache flushes for all the regions; however,
we're planning to add more bookkeeping logic around [de]population to
support atomic allocations and this delegation of iterations gets in
the way.
This patch moves the region iterations out of
pcpu_[de]populate_chunk() into its callers - pcpu_alloc() and
pcpu_reclaim() - so that we can later add logic to track more states
around them. This change may make cache and tlb flushes more frequent
but multi-region [de]populations are rare anyway and if this actually
becomes a problem, it's not difficult to factor out cache flushes as
separate callbacks which are directly invoked from percpu.c.
Signed-off-by: Tejun Heo <tj@kernel.org>
percpu-vm and percpu-km implement separate versions of
pcpu_[de]populate_chunk() and some part which is or should be common
are currently in the specific implementations. Make the following
changes.
* Allocate area clearing is moved from the pcpu_populate_chunk()
implementations to pcpu_alloc(). This makes percpu-km's version
noop.
* Quick exit tests in pcpu_[de]populate_chunk() of percpu-vm are moved
to their respective callers so that they are applied to percpu-km
too. This doesn't make any meaningful difference as both functions
are noop for percpu-km; however, this is more consistent and will
help implementing atomic allocation support.
Signed-off-by: Tejun Heo <tj@kernel.org>
Percpu allocator should clear memory before returning it but the km
allocator forgot to do it. Fix it.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Peter Zijlstra <peterz@infradead.org>
On UP, percpu allocations were redirected to kmalloc. This has the
following problems.
* For certain amount of allocations (determined by
PERCPU_DYNAMIC_EARLY_SLOTS and PERCPU_DYNAMIC_EARLY_SIZE), percpu
allocator can be used before the usual kernel memory allocator is
brought online. On SMP, this is used to initialize the kernel
memory allocator.
* percpu allocator honors alignment upto PAGE_SIZE but kmalloc()
doesn't. For example, workqueue makes use of larger alignments for
cpu_workqueues.
Currently, users of percpu allocators need to handle UP differently,
which is somewhat fragile and ugly. Other than small amount of
memory, there isn't much to lose by enabling percpu allocator on UP.
It can simply use kernel memory based chunk allocation which was added
for SMP archs w/o MMUs.
This patch removes mm/percpu_up.c, builds mm/percpu.c on UP too and
makes UP build use percpu-km. As percpu addresses and kernel
addresses are always identity mapped and static percpu variables don't
need any special treatment, nothing is arch dependent and mm/percpu.c
implements generic setup_per_cpu_areas() for UP.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Pekka Enberg <penberg@cs.helsinki.fi>
Implement an alternate percpu chunk management based on kernel memeory
for nommu SMP architectures. Instead of mapping into vmalloc area,
chunks are allocated as a contiguous kernel memory using
alloc_pages(). As such, percpu allocator on nommu will have the
following restrictions.
* It can't fill chunks on-demand page-by-page. It has to allocate
each chunk fully upfront.
* It can't support sparse chunk for NUMA configurations. SMP w/o mmu
is crazy enough. Let's hope no one does NUMA w/o mmu. :-P
* If chunk size isn't power-of-two multiple of PAGE_SIZE, the
unaligned amount will be wasted on each chunk. So, archs which use
this better align chunk size.
For instructions on how to use this, read the comment on top of
mm/percpu-km.c.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: David Howells <dhowells@redhat.com>
Cc: Graff Yang <graff.yang@gmail.com>
Cc: Sonic Zhang <sonic.adi@gmail.com>
Dennis Zhou