... such that we can avoid the tree walks to get the node with the
smallest key. Semantically the same, as the previously used rb_first(),
but O(1). The main overhead is the extra footprint for the cached rb_node
pointer, which should not matter for epoll.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Davidlohr Bueso <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Cc: Jan Kara <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
... such that we can avoid the tree walks to get the node with the
smallest key. Semantically the same, as the previously used rb_first(),
but O(1). The main overhead is the extra footprint for the cached rb_node
pointer, which should not matter for procfs.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Davidlohr Bueso <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Allow interval trees to quickly check for overlaps to avoid unnecesary
tree lookups in interval_tree_iter_first().
As of this patch, all interval tree flavors will require using a
'rb_root_cached' such that we can have the leftmost node easily
available. While most users will make use of this feature, those with
special functions (in addition to the generic insert, delete, search
calls) will avoid using the cached option as they can do funky things
with insertions -- for example, vma_interval_tree_insert_after().
[[email protected]: fix deadlock from typo vm_lock_anon_vma()]
Link: http://lkml.kernel.org/r/[email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Davidlohr Bueso <[email protected]>
Signed-off-by: Jérôme Glisse <[email protected]>
Acked-by: Christian König <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Acked-by: Doug Ledford <[email protected]>
Acked-by: Michael S. Tsirkin <[email protected]>
Cc: David Airlie <[email protected]>
Cc: Jason Wang <[email protected]>
Cc: Christian Benvenuti <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
We can work with a single rb_root_cached root to test both cached and
non-cached rbtrees. In addition, also add a test to measure latencies
between rb_first and its fast counterpart.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Davidlohr Bueso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
While overall the code is very nicely commented, it might not be
immediately obvious from the diagrams what is going on. Add a very
brief summary of each case. Opposite cases where the node is the left
child are left untouched.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Davidlohr Bueso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The only times the nil-parent (root node) condition is true is when the
node is the first in the tree, or after fixing rbtree rule #4 and the
case 1 rebalancing made the node the root. Such conditions do not apply
most of the time:
(i) The common case in an rbtree is to have more than a single node,
so this is only true for the first rb_insert().
(ii) While there is a chance only one first rotation is needed, cases
where the node's uncle is black (cases 2,3) are more common as we can
have the following scenarios during the rotation looping:
case1 only, case1+1, case2+3, case1+2+3, case3 only, etc.
This patch, therefore, adds an unlikely() optimization to this
conditional. When profiling with CONFIG_PROFILE_ANNOTATED_BRANCHES, a
kernel build shows that the incorrect rate is less than 15%, and for
workloads that involve insert mostly trees overtime tend to have less
than 2% incorrect rate.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Davidlohr Bueso <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Patch series "rbtree: Cache leftmost node internally", v4.
A series to extending rbtrees to internally cache the leftmost node such
that we can have fast overlap check optimization for all interval tree
users[1]. The benefits of this series are that:
(i) Unify users that do internal leftmost node caching.
(ii) Optimize all interval tree users.
(iii) Convert at least two new users (epoll and procfs) to the new interface.
This patch (of 16):
Red-black tree semantics imply that nodes with smaller or greater (or
equal for duplicates) keys always be to the left and right,
respectively. For the kernel this is extremely evident when considering
our rb_first() semantics. Enabling lookups for the smallest node in the
tree in O(1) can save a good chunk of cycles in not having to walk down
the tree each time. To this end there are a few core users that
explicitly do this, such as the scheduler and rtmutexes. There is also
the desire for interval trees to have this optimization allowing faster
overlap checking.
This patch introduces a new 'struct rb_root_cached' which is just the
root with a cached pointer to the leftmost node. The reason why the
regular rb_root was not extended instead of adding a new structure was
that this allows the user to have the choice between memory footprint
and actual tree performance. The new wrappers on top of the regular
rb_root calls are:
- rb_first_cached(cached_root) -- which is a fast replacement
for rb_first.
- rb_insert_color_cached(node, cached_root, new)
- rb_erase_cached(node, cached_root)
In addition, augmented cached interfaces are also added for basic
insertion and deletion operations; which becomes important for the
interval tree changes.
With the exception of the inserts, which adds a bool for updating the
new leftmost, the interfaces are kept the same. To this end, porting rb
users to the cached version becomes really trivial, and keeping current
rbtree semantics for users that don't care about the optimization
requires zero overhead.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Davidlohr Bueso <[email protected]>
Reviewed-by: Jan Kara <[email protected]>
Acked-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
GENMASK(_ULL) performs a left-shift of ~0UL(L), which technically
results in an integer overflow. clang raises a warning if the overflow
occurs in a preprocessor expression. Clear the low-order bits through a
substraction instead of the left-shift to avoid the overflow.
(akpm: no change in .text size in my testing)
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Matthias Kaehlcke <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Patch series "Define CPU_BIG_ENDIAN or warn for inconsistencies", v3.
While working on enabling queued rwlock on SPARC, found this following
code in include/asm-generic/qrwlock.h which uses CONFIG_CPU_BIG_ENDIAN to
clear a byte.
static inline u8 *__qrwlock_write_byte(struct qrwlock *lock)
{
return (u8 *)lock + 3 * IS_BUILTIN(CONFIG_CPU_BIG_ENDIAN);
}
Problem is many of the fixed big endian architectures don't define
CPU_BIG_ENDIAN and clears the wrong byte.
Define CPU_BIG_ENDIAN for all the fixed big endian architecture to fix it.
Also found few more references of this config parameter in
drivers/of/base.c
drivers/of/fdt.c
drivers/tty/serial/earlycon.c
drivers/tty/serial/serial_core.c
Be aware that this may cause regressions if someone has worked-around
problems in the above code already. Remove the work-around.
Here is our original discussion
https://lkml.org/lkml/2017/5/24/620
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Babu Moger <[email protected]>
Suggested-by: Arnd Bergmann <[email protected]>
Acked-by: Geert Uytterhoeven <[email protected]>
Acked-by: David S. Miller <[email protected]>
Acked-by: Stafford Horne <[email protected]>
Cc: Yoshinori Sato <[email protected]>
Cc: Jonas Bonn <[email protected]>
Cc: Stefan Kristiansson <[email protected]>
Cc: "James E.J. Bottomley" <[email protected]>
Cc: Helge Deller <[email protected]>
Cc: Alexander Viro <[email protected]>
Cc: Michal Simek <[email protected]>
Cc: Michael Ellerman <[email protected]> (powerpc)
Cc: Peter Zijlstra <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Max Filippov <[email protected]>
Cc: Greg KH <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
First, number of CPUs can't be negative number.
Second, different signnnedness leads to suboptimal code in the following
cases:
1)
kmalloc(nr_cpu_ids * sizeof(X));
"int" has to be sign extended to size_t.
2)
while (loff_t *pos < nr_cpu_ids)
MOVSXD is 1 byte longed than the same MOV.
Other cases exist as well. Basically compiler is told that nr_cpu_ids
can't be negative which can't be deduced if it is "int".
Code savings on allyesconfig kernel: -3KB
add/remove: 0/0 grow/shrink: 25/264 up/down: 261/-3631 (-3370)
function old new delta
coretemp_cpu_online 450 512 +62
rcu_init_one 1234 1272 +38
pci_device_probe 374 399 +25
...
pgdat_reclaimable_pages 628 556 -72
select_fallback_rq 446 369 -77
task_numa_find_cpu 1923 1807 -116
Link: http://lkml.kernel.org/r/20170819114959.GA30580@avx2
Signed-off-by: Alexey Dobriyan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Patch series "Multibyte memset variations", v4.
A relatively common idiom we're missing is a function to fill an area of
memory with a pattern which is larger than a single byte. I first
noticed this with a zram patch which wanted to fill a page with an
'unsigned long' value. There turn out to be quite a few places in the
kernel which can benefit from using an optimised function rather than a
loop; sometimes text size, sometimes speed, and sometimes both. The
optimised PowerPC version (not included here) improves performance by
about 30% on POWER8 on just the raw memset_l().
Most of the extra lines of code come from the three testcases I added.
This patch (of 8):
memset16(), memset32() and memset64() are like memset(), but allow the
caller to fill the destination with a value larger than a single byte.
memset_l() and memset_p() allow the caller to use unsigned long and
pointer values respectively.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Matthew Wilcox <[email protected]>
Cc: "H. Peter Anvin" <[email protected]>
Cc: "James E.J. Bottomley" <[email protected]>
Cc: "Martin K. Petersen" <[email protected]>
Cc: David Miller <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Ivan Kokshaysky <[email protected]>
Cc: Matt Turner <[email protected]>
Cc: Michael Ellerman <[email protected]>
Cc: Minchan Kim <[email protected]>
Cc: Ralf Baechle <[email protected]>
Cc: Richard Henderson <[email protected]>
Cc: Russell King <[email protected]>
Cc: Sam Ravnborg <[email protected]>
Cc: Sergey Senozhatsky <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
If there are large numbers of hugepages to iterate while reading
/proc/pid/smaps, the page walk never does cond_resched(). On archs
without split pmd locks, there can be significant and observable
contention on mm->page_table_lock which cause lengthy delays without
rescheduling.
Always reschedule in smaps_pte_range() if necessary since the pagewalk
iteration can be expensive.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: David Rientjes <[email protected]>
Cc: Minchan Kim <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: "Kirill A. Shutemov" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
If initializing a small swap file fails because the swap file has a
problem (holes, etc.) then we need to free the cluster info as part of
cleanup. Unfortunately a previous patch changed the code to use kvzalloc
but did not change all the vfree calls to use kvfree.
Found by running generic/357 from xfstests.
Link: http://lkml.kernel.org/r/20170831233515.GR3775@magnolia
Fixes: 54f180d3c1 ("mm, swap: use kvzalloc to allocate some swap data structures")
Signed-off-by: Darrick J. Wong <[email protected]>
Reviewed-by: "Huang, Ying" <[email protected]>
Acked-by: David Rientjes <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: <[email protected]> [4.12+]
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
We are by error initializing alloc_flags before gfp_allowed_mask is
applied. This could cause problems after pm_restrict_gfp_mask() is called
during suspend operation. Apply gfp_allowed_mask before initializing
alloc_flags so that the first allocation attempt uses correct flags.
Link: http://lkml.kernel.org/r/[email protected]
Fixes: 83d4ca8148 ("mm, page_alloc: move __GFP_HARDWALL modifications out of the fastpath")
Signed-off-by: Tetsuo Handa <[email protected]>
Acked-by: Michal Hocko <[email protected]>
Acked-by: Vlastimil Babka <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: Jesper Dangaard Brouer <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
online_mem_sections() accidentally marks online only the first section
in the given range. This is a typo which hasn't been noticed because I
haven't tested large 2GB blocks previously. All users of
pfn_to_online_page would get confused on the the rest of the pfn range
in the block.
All we need to fix this is to use iterator (pfn) rather than start_pfn.
Link: http://lkml.kernel.org/r/[email protected]
Fixes: 2d070eab2e ("mm: consider zone which is not fully populated to have holes")
Signed-off-by: Michal Hocko <[email protected]>
Acked-by: Vlastimil Babka <[email protected]>
Cc: Anshuman Khandual <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Seen while reading the code, in handle_mm_fault(), in the case
arch_vma_access_permitted() is failing the call to
mem_cgroup_oom_disable() is not made.
To fix that, move the call to mem_cgroup_oom_enable() after calling
arch_vma_access_permitted() as it should not have entered the memcg OOM.
Link: http://lkml.kernel.org/r/[email protected]
Fixes: bae473a423 ("mm: introduce fault_env")
Signed-off-by: Laurent Dufour <[email protected]>
Acked-by: Kirill A. Shutemov <[email protected]>
Acked-by: Michal Hocko <[email protected]>
Cc: <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The fadvise() manpage is silent on fadvise()'s effect on memory-based
filesystems (shmem, hugetlbfs & ramfs) and pseudo file systems (procfs,
sysfs, kernfs). The current implementaion of fadvise is mostly a noop
for such filesystems except for FADV_DONTNEED which will trigger
expensive remote LRU cache draining. This patch makes the noop of
fadvise() on such file systems very explicit.
However this change has two side effects for ramfs and one for tmpfs.
First fadvise(FADV_DONTNEED) could remove the unmapped clean zero'ed
pages of ramfs (allocated through read, readahead & read fault) and
tmpfs (allocated through read fault). Also fadvise(FADV_WILLNEED) could
create such clean zero'ed pages for ramfs. This change removes those
possibilities.
One of our generic libraries does fadvise(FADV_DONTNEED). Recently we
observed high latency in fadvise() and noticed that the users have
started using tmpfs files and the latency was due to expensive remote
LRU cache draining. For normal tmpfs files (have data written on them),
fadvise(FADV_DONTNEED) will always trigger the unneeded remote cache
draining.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Shakeel Butt <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: Johannes Weiner <[email protected]>
Cc: Hillf Danton <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Cc: Hugh Dickins <[email protected]>
Cc: Greg Thelen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
page_zone_id() is a specialized function to compare the zone for the pages
that are within the section range. If the section of the pages are
different, page_zone_id() can be different even if their zone is the same.
This wrong usage doesn't cause any actual problem since
__munlock_pagevec_fill() would be called again with failed index.
However, it's better to use more appropriate function here.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Joonsoo Kim <[email protected]>
Acked-by: Vlastimil Babka <[email protected]>
Cc: Minchan Kim <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
There is significant overhead in cache bouncing caused by zone counters
(NUMA associated counters) update in parallel in multi-threaded page
allocation (suggested by Dave Hansen).
This patch updates NUMA counter threshold to a fixed size of MAX_U16 - 2,
as a small threshold greatly increases the update frequency of the global
counter from local per cpu counter(suggested by Ying Huang).
The rationality is that these statistics counters don't affect the
kernel's decision, unlike other VM counters, so it's not a problem to use
a large threshold.
With this patchset, we see 31.3% drop of CPU cycles(537-->369) for per
single page allocation and reclaim on Jesper's page_bench03 benchmark.
Benchmark provided by Jesper D Brouer(increase loop times to 10000000):
https://github.com/netoptimizer/prototype-kernel/tree/master/kernel/mm/
bench
Threshold CPU cycles Throughput(88 threads)
32 799 241760478
64 640 301628829
125 537 358906028 <==> system by default (base)
256 468 412397590
512 428 450550704
4096 399 482520943
20000 394 489009617
30000 395 488017817
65533 369(-31.3%) 521661345(+45.3%) <==> with this patchset
N/A 342(-36.3%) 562900157(+56.8%) <==> disable zone_statistics
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Kemi Wang <[email protected]>
Reported-by: Jesper Dangaard Brouer <[email protected]>
Suggested-by: Dave Hansen <[email protected]>
Suggested-by: Ying Huang <[email protected]>
Acked-by: Mel Gorman <[email protected]>
Cc: Aaron Lu <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Christopher Lameter <[email protected]>
Cc: Johannes Weiner <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: Tim Chen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Patch series "Separate NUMA statistics from zone statistics", v2.
Each page allocation updates a set of per-zone statistics with a call to
zone_statistics(). As discussed in 2017 MM summit, these are a
substantial source of overhead in the page allocator and are very rarely
consumed. This significant overhead in cache bouncing caused by zone
counters (NUMA associated counters) update in parallel in multi-threaded
page allocation (pointed out by Dave Hansen).
A link to the MM summit slides:
http://people.netfilter.org/hawk/presentations/MM-summit2017/MM-summit2017-JesperBrouer.pdf
To mitigate this overhead, this patchset separates NUMA statistics from
zone statistics framework, and update NUMA counter threshold to a fixed
size of MAX_U16 - 2, as a small threshold greatly increases the update
frequency of the global counter from local per cpu counter (suggested by
Ying Huang). The rationality is that these statistics counters don't
need to be read often, unlike other VM counters, so it's not a problem
to use a large threshold and make readers more expensive.
With this patchset, we see 31.3% drop of CPU cycles(537-->369, see
below) for per single page allocation and reclaim on Jesper's
page_bench03 benchmark. Meanwhile, this patchset keeps the same style
of virtual memory statistics with little end-user-visible effects (only
move the numa stats to show behind zone page stats, see the first patch
for details).
I did an experiment of single page allocation and reclaim concurrently
using Jesper's page_bench03 benchmark on a 2-Socket Broadwell-based
server (88 processors with 126G memory) with different size of threshold
of pcp counter.
Benchmark provided by Jesper D Brouer(increase loop times to 10000000):
https://github.com/netoptimizer/prototype-kernel/tree/master/kernel/mm/bench
Threshold CPU cycles Throughput(88 threads)
32 799 241760478
64 640 301628829
125 537 358906028 <==> system by default
256 468 412397590
512 428 450550704
4096 399 482520943
20000 394 489009617
30000 395 488017817
65533 369(-31.3%) 521661345(+45.3%) <==> with this patchset
N/A 342(-36.3%) 562900157(+56.8%) <==> disable zone_statistics
This patch (of 3):
In this patch, NUMA statistics is separated from zone statistics
framework, all the call sites of NUMA stats are changed to use
numa-stats-specific functions, it does not have any functionality change
except that the number of NUMA stats is shown behind zone page stats
when users *read* the zone info.
E.g. cat /proc/zoneinfo
***Base*** ***With this patch***
nr_free_pages 3976 nr_free_pages 3976
nr_zone_inactive_anon 0 nr_zone_inactive_anon 0
nr_zone_active_anon 0 nr_zone_active_anon 0
nr_zone_inactive_file 0 nr_zone_inactive_file 0
nr_zone_active_file 0 nr_zone_active_file 0
nr_zone_unevictable 0 nr_zone_unevictable 0
nr_zone_write_pending 0 nr_zone_write_pending 0
nr_mlock 0 nr_mlock 0
nr_page_table_pages 0 nr_page_table_pages 0
nr_kernel_stack 0 nr_kernel_stack 0
nr_bounce 0 nr_bounce 0
nr_zspages 0 nr_zspages 0
numa_hit 0 *nr_free_cma 0*
numa_miss 0 numa_hit 0
numa_foreign 0 numa_miss 0
numa_interleave 0 numa_foreign 0
numa_local 0 numa_interleave 0
numa_other 0 numa_local 0
*nr_free_cma 0* numa_other 0
... ...
vm stats threshold: 10 vm stats threshold: 10
... ...
The next patch updates the numa stats counter size and threshold.
[[email protected]: coding-style fixes]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Kemi Wang <[email protected]>
Reported-by: Jesper Dangaard Brouer <[email protected]>
Acked-by: Mel Gorman <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: Johannes Weiner <[email protected]>
Cc: Christopher Lameter <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Ying Huang <[email protected]>
Cc: Aaron Lu <[email protected]>
Cc: Tim Chen <[email protected]>
Cc: Dave Hansen <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This is an enhancement to avoid a non cooperative userfaultfd manager
having to unregister all regions before it can close the uffd after all
userfaultfd activity completed.
The UFFDIO_UNREGISTER would serialize against the handle_userfault by
taking the mmap_sem for writing, but we can simply repeat the page fault
if we detect the uffd was closed and so the regular page fault paths
should takeover.
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Andrea Arcangeli <[email protected]>
Acked-by: Mike Rapoport <[email protected]>
Cc: Mike Kravetz <[email protected]>
Cc: Pavel Emelyanov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
