dl_rq->dl_nr_migratory is increased whenever a DL entity is enqueued and it has
nr_cpus_allowed > 1. Unlike the pushable_dl_tasks tree, dl_rq->dl_nr_migratory
includes a dl_rq's current task. This means a dl_rq can have a migratable
current, N non-migratable queued tasks, and be flagged as overloaded and have
its CPU set in the dlo_mask, despite having an empty pushable_tasks tree.
Make an dl_rq's overload logic be driven by {enqueue,dequeue}_pushable_dl_task(),
in other words make DL RQs only be flagged as overloaded if they have at
least one runnable-but-not-current migratable task.
o push_dl_task() is unaffected, as it is a no-op if there are no pushable
tasks.
o pull_dl_task() now no longer scans runqueues whose sole migratable task is
their current one, which it can't do anything about anyway.
It may also now pull tasks to a DL RQ with dl_nr_running > 1 if only its
current task is migratable.
Since dl_rq->dl_nr_migratory becomes unused, remove it.
RT had the exact same mechanism (rt_rq->rt_nr_migratory) which was dropped
in favour of relying on rt_rq->pushable_tasks, see:
612f769edd ("sched/rt: Make rt_rq->pushable_tasks updates drive rto_mask")
Signed-off-by: Valentin Schneider <vschneid@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Juri Lelli <juri.lelli@redhat.com>
Link: https://lore.kernel.org/r/20230928150251.463109-1-vschneid@redhat.com
Sebastian noted that the rto_push_work IRQ work can be queued for a CPU
that has an empty pushable_tasks list, which means nothing useful will be
done in the IPI other than queue the work for the next CPU on the rto_mask.
rto_push_irq_work_func() only operates on tasks in the pushable_tasks list,
but the conditions for that irq_work to be queued (and for a CPU to be
added to the rto_mask) rely on rq_rt->nr_migratory instead.
nr_migratory is increased whenever an RT task entity is enqueued and it has
nr_cpus_allowed > 1. Unlike the pushable_tasks list, nr_migratory includes a
rt_rq's current task. This means a rt_rq can have a migratible current, N
non-migratible queued tasks, and be flagged as overloaded / have its CPU
set in the rto_mask, despite having an empty pushable_tasks list.
Make an rt_rq's overload logic be driven by {enqueue,dequeue}_pushable_task().
Since rt_rq->{rt_nr_migratory,rt_nr_total} become unused, remove them.
Note that the case where the current task is pushed away to make way for a
migration-disabled task remains unchanged: the migration-disabled task has
to be in the pushable_tasks list in the first place, which means it has
nr_cpus_allowed > 1.
Reported-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Valentin Schneider <vschneid@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Link: https://lore.kernel.org/r/20230811112044.3302588-1-vschneid@redhat.com
Where CFS is currently a WFQ based scheduler with only a single knob,
the weight. The addition of a second, latency oriented parameter,
makes something like WF2Q or EEVDF based a much better fit.
Specifically, EEVDF does EDF like scheduling in the left half of the
tree -- those entities that are owed service. Except because this is a
virtual time scheduler, the deadlines are in virtual time as well,
which is what allows over-subscription.
EEVDF has two parameters:
- weight, or time-slope: which is mapped to nice just as before
- request size, or slice length: which is used to compute
the virtual deadline as: vd_i = ve_i + r_i/w_i
Basically, by setting a smaller slice, the deadline will be earlier
and the task will be more eligible and ran earlier.
Tick driven preemption is driven by request/slice completion; while
wakeup preemption is driven by the deadline.
Because the tree is now effectively an interval tree, and the
selection is no longer 'leftmost', over-scheduling is less of a
problem.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20230531124603.931005524@infradead.org
In order to move to an eligibility based scheduling policy, we need
to have a better approximation of the ideal scheduler.
Specifically, for a virtual time weighted fair queueing based
scheduler the ideal scheduler will be the weighted average of the
individual virtual runtimes (math in the comment).
As such, compute the weighted average to approximate the ideal
scheduler -- note that the approximation is in the individual task
behaviour, which isn't strictly conformant.
Specifically consider adding a task with a vruntime left of center, in
this case the average will move backwards in time -- something the
ideal scheduler would of course never do.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20230531124603.654144274@infradead.org
Commit e6fe3f422b ("sched: Make multiple runqueue task counters
32-bit") changed the type for rq->nr_uninterruptible from "unsigned
long" to "unsigned int", but left wrong cast print to
/sys/kernel/debug/sched/debug and to the console.
For example, nr_uninterruptible's value is fffffff7 with type
"unsigned int", (long)nr_uninterruptible shows 4294967287 while
(int)nr_uninterruptible prints -9. So using int cast fixes wrong
printing.
Signed-off-by: Yan Yan <yanyan.yan@antgroup.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20230506074253.44526-1-yanyan.yan@antgroup.com
Pull MM updates from Andrew Morton:
- Yu Zhao's Multi-Gen LRU patches are here. They've been under test in
linux-next for a couple of months without, to my knowledge, any
negative reports (or any positive ones, come to that).
- Also the Maple Tree from Liam Howlett. An overlapping range-based
tree for vmas. It it apparently slightly more efficient in its own
right, but is mainly targeted at enabling work to reduce mmap_lock
contention.
Liam has identified a number of other tree users in the kernel which
could be beneficially onverted to mapletrees.
Yu Zhao has identified a hard-to-hit but "easy to fix" lockdep splat
at [1]. This has yet to be addressed due to Liam's unfortunately
timed vacation. He is now back and we'll get this fixed up.
- Dmitry Vyukov introduces KMSAN: the Kernel Memory Sanitizer. It uses
clang-generated instrumentation to detect used-unintialized bugs down
to the single bit level.
KMSAN keeps finding bugs. New ones, as well as the legacy ones.
- Yang Shi adds a userspace mechanism (madvise) to induce a collapse of
memory into THPs.
- Zach O'Keefe has expanded Yang Shi's madvise(MADV_COLLAPSE) to
support file/shmem-backed pages.
- userfaultfd updates from Axel Rasmussen
- zsmalloc cleanups from Alexey Romanov
- cleanups from Miaohe Lin: vmscan, hugetlb_cgroup, hugetlb and
memory-failure
- Huang Ying adds enhancements to NUMA balancing memory tiering mode's
page promotion, with a new way of detecting hot pages.
- memcg updates from Shakeel Butt: charging optimizations and reduced
memory consumption.
- memcg cleanups from Kairui Song.
- memcg fixes and cleanups from Johannes Weiner.
- Vishal Moola provides more folio conversions
- Zhang Yi removed ll_rw_block() :(
- migration enhancements from Peter Xu
- migration error-path bugfixes from Huang Ying
- Aneesh Kumar added ability for a device driver to alter the memory
tiering promotion paths. For optimizations by PMEM drivers, DRM
drivers, etc.
- vma merging improvements from Jakub Matěn.
- NUMA hinting cleanups from David Hildenbrand.
- xu xin added aditional userspace visibility into KSM merging
activity.
- THP & KSM code consolidation from Qi Zheng.
- more folio work from Matthew Wilcox.
- KASAN updates from Andrey Konovalov.
- DAMON cleanups from Kaixu Xia.
- DAMON work from SeongJae Park: fixes, cleanups.
- hugetlb sysfs cleanups from Muchun Song.
- Mike Kravetz fixes locking issues in hugetlbfs and in hugetlb core.
Link: https://lkml.kernel.org/r/CAOUHufZabH85CeUN-MEMgL8gJGzJEWUrkiM58JkTbBhh-jew0Q@mail.gmail.com [1]
* tag 'mm-stable-2022-10-08' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm: (555 commits)
hugetlb: allocate vma lock for all sharable vmas
hugetlb: take hugetlb vma_lock when clearing vma_lock->vma pointer
hugetlb: fix vma lock handling during split vma and range unmapping
mglru: mm/vmscan.c: fix imprecise comments
mm/mglru: don't sync disk for each aging cycle
mm: memcontrol: drop dead CONFIG_MEMCG_SWAP config symbol
mm: memcontrol: use do_memsw_account() in a few more places
mm: memcontrol: deprecate swapaccounting=0 mode
mm: memcontrol: don't allocate cgroup swap arrays when memcg is disabled
mm/secretmem: remove reduntant return value
mm/hugetlb: add available_huge_pages() func
mm: remove unused inline functions from include/linux/mm_inline.h
selftests/vm: add selftest for MADV_COLLAPSE of uffd-minor memory
selftests/vm: add file/shmem MADV_COLLAPSE selftest for cleared pmd
selftests/vm: add thp collapse shmem testing
selftests/vm: add thp collapse file and tmpfs testing
selftests/vm: modularize thp collapse memory operations
selftests/vm: dedup THP helpers
mm/khugepaged: add tracepoint to hpage_collapse_scan_file()
mm/madvise: add file and shmem support to MADV_COLLAPSE
...
Patch series "memory tiering: hot page selection", v4.
To optimize page placement in a memory tiering system with NUMA balancing,
the hot pages in the slow memory nodes need to be identified.
Essentially, the original NUMA balancing implementation selects the mostly
recently accessed (MRU) pages to promote. But this isn't a perfect
algorithm to identify the hot pages. Because the pages with quite low
access frequency may be accessed eventually given the NUMA balancing page
table scanning period could be quite long (e.g. 60 seconds). So in this
patchset, we implement a new hot page identification algorithm based on
the latency between NUMA balancing page table scanning and hint page
fault. Which is a kind of mostly frequently accessed (MFU) algorithm.
In NUMA balancing memory tiering mode, if there are hot pages in slow
memory node and cold pages in fast memory node, we need to promote/demote
hot/cold pages between the fast and cold memory nodes.
A choice is to promote/demote as fast as possible. But the CPU cycles and
memory bandwidth consumed by the high promoting/demoting throughput will
hurt the latency of some workload because of accessing inflating and slow
memory bandwidth contention.
A way to resolve this issue is to restrict the max promoting/demoting
throughput. It will take longer to finish the promoting/demoting. But
the workload latency will be better. This is implemented in this patchset
as the page promotion rate limit mechanism.
The promotion hot threshold is workload and system configuration
dependent. So in this patchset, a method to adjust the hot threshold
automatically is implemented. The basic idea is to control the number of
the candidate promotion pages to match the promotion rate limit.
We used the pmbench memory accessing benchmark tested the patchset on a
2-socket server system with DRAM and PMEM installed. The test results are
as follows,
pmbench score promote rate
(accesses/s) MB/s
------------- ------------
base 146887704.1 725.6
hot selection 165695601.2 544.0
rate limit 162814569.8 165.2
auto adjustment 170495294.0 136.9
From the results above,
With hot page selection patch [1/3], the pmbench score increases about
12.8%, and promote rate (overhead) decreases about 25.0%, compared with
base kernel.
With rate limit patch [2/3], pmbench score decreases about 1.7%, and
promote rate decreases about 69.6%, compared with hot page selection
patch.
With threshold auto adjustment patch [3/3], pmbench score increases about
4.7%, and promote rate decrease about 17.1%, compared with rate limit
patch.
Baolin helped to test the patchset with MySQL on a machine which contains
1 DRAM node (30G) and 1 PMEM node (126G).
sysbench /usr/share/sysbench/oltp_read_write.lua \
......
--tables=200 \
--table-size=1000000 \
--report-interval=10 \
--threads=16 \
--time=120
The tps can be improved about 5%.
This patch (of 3):
To optimize page placement in a memory tiering system with NUMA balancing,
the hot pages in the slow memory node need to be identified. Essentially,
the original NUMA balancing implementation selects the mostly recently
accessed (MRU) pages to promote. But this isn't a perfect algorithm to
identify the hot pages. Because the pages with quite low access frequency
may be accessed eventually given the NUMA balancing page table scanning
period could be quite long (e.g. 60 seconds). The most frequently
accessed (MFU) algorithm is better.
So, in this patch we implemented a better hot page selection algorithm.
Which is based on NUMA balancing page table scanning and hint page fault
as follows,
- When the page tables of the processes are scanned to change PTE/PMD
to be PROT_NONE, the current time is recorded in struct page as scan
time.
- When the page is accessed, hint page fault will occur. The scan
time is gotten from the struct page. And The hint page fault
latency is defined as
hint page fault time - scan time
The shorter the hint page fault latency of a page is, the higher the
probability of their access frequency to be higher. So the hint page
fault latency is a better estimation of the page hot/cold.
It's hard to find some extra space in struct page to hold the scan time.
Fortunately, we can reuse some bits used by the original NUMA balancing.
NUMA balancing uses some bits in struct page to store the page accessing
CPU and PID (referring to page_cpupid_xchg_last()). Which is used by the
multi-stage node selection algorithm to avoid to migrate pages shared
accessed by the NUMA nodes back and forth. But for pages in the slow
memory node, even if they are shared accessed by multiple NUMA nodes, as
long as the pages are hot, they need to be promoted to the fast memory
node. So the accessing CPU and PID information are unnecessary for the
slow memory pages. We can reuse these bits in struct page to record the
scan time. For the fast memory pages, these bits are used as before.
For the hot threshold, the default value is 1 second, which works well in
our performance test. All pages with hint page fault latency < hot
threshold will be considered hot.
It's hard for users to determine the hot threshold. So we don't provide a
kernel ABI to set it, just provide a debugfs interface for advanced users
to experiment. We will continue to work on a hot threshold automatic
adjustment mechanism.
The downside of the above method is that the response time to the workload
hot spot changing may be much longer. For example,
- A previous cold memory area becomes hot
- The hint page fault will be triggered. But the hint page fault
latency isn't shorter than the hot threshold. So the pages will
not be promoted.
- When the memory area is scanned again, maybe after a scan period,
the hint page fault latency measured will be shorter than the hot
threshold and the pages will be promoted.
To mitigate this, if there are enough free space in the fast memory node,
the hot threshold will not be used, all pages will be promoted upon the
hint page fault for fast response.
Thanks Zhong Jiang reported and tested the fix for a bug when disabling
memory tiering mode dynamically.
Link: https://lkml.kernel.org/r/20220713083954.34196-1-ying.huang@intel.com
Link: https://lkml.kernel.org/r/20220713083954.34196-2-ying.huang@intel.com
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Tested-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Wei Xu <weixugc@google.com>
Cc: osalvador <osalvador@suse.de>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Zhong Jiang <zhongjiang-ali@linux.alibaba.com>
Cc: Oscar Salvador <osalvador@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Collect all utility functionality source code files into a single kernel/sched/build_utility.c file,
via #include-ing the .c files:
kernel/sched/clock.c
kernel/sched/completion.c
kernel/sched/loadavg.c
kernel/sched/swait.c
kernel/sched/wait_bit.c
kernel/sched/wait.c
CONFIG_CPU_FREQ:
kernel/sched/cpufreq.c
CONFIG_CPU_FREQ_GOV_SCHEDUTIL:
kernel/sched/cpufreq_schedutil.c
CONFIG_CGROUP_CPUACCT:
kernel/sched/cpuacct.c
CONFIG_SCHED_DEBUG:
kernel/sched/debug.c
CONFIG_SCHEDSTATS:
kernel/sched/stats.c
CONFIG_SMP:
kernel/sched/cpupri.c
kernel/sched/stop_task.c
kernel/sched/topology.c
CONFIG_SCHED_CORE:
kernel/sched/core_sched.c
CONFIG_PSI:
kernel/sched/psi.c
CONFIG_MEMBARRIER:
kernel/sched/membarrier.c
CONFIG_CPU_ISOLATION:
kernel/sched/isolation.c
CONFIG_SCHED_AUTOGROUP:
kernel/sched/autogroup.c
The goal is to amortize the 60+ KLOC header bloat from over a dozen build units into
a single build unit.
The build time of build_utility.c also roughly matches the build time of core.c and
fair.c - allowing better load-balancing of scheduler-only rebuilds.
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Peter Zijlstra <peterz@infradead.org>
The older format of /proc/pid/sched printed home node info which
required the mempolicy and task lock around mpol_get(). However
the format has changed since then and there is no need for
sched_show_numa() any more to have mempolicy argument,
asssociated mpol_get/put and task_lock/unlock. Remove them.
Fixes: 397f2378f1 ("sched/numa: Fix numa balancing stats in /proc/pid/sched")
Signed-off-by: Bharata B Rao <bharata@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Link: https://lore.kernel.org/r/20220118050515.2973-1-bharata@amd.com
Adds accounting for "forced idle" time, which is time where a cookie'd
task forces its SMT sibling to idle, despite the presence of runnable
tasks.
Forced idle time is one means to measure the cost of enabling core
scheduling (ie. the capacity lost due to the need to force idle).
Forced idle time is attributed to the thread responsible for causing
the forced idle.
A few details:
- Forced idle time is displayed via /proc/PID/sched. It also requires
that schedstats is enabled.
- Forced idle is only accounted when a sibling hyperthread is held
idle despite the presence of runnable tasks. No time is charged if
a sibling is idle but has no runnable tasks.
- Tasks with 0 cookie are never charged forced idle.
- For SMT > 2, we scale the amount of forced idle charged based on the
number of forced idle siblings. Additionally, we split the time up and
evenly charge it to all running tasks, as each is equally responsible
for the forced idle.
Signed-off-by: Josh Don <joshdon@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20211018203428.2025792-1-joshdon@google.com
When !SCHEDSTATS schedstat_enabled() is an unconditional 0 and the
whole block doesn't exist, however GCC figures the scoped variable
'stats' is unused and complains about it.
Upgrade the warning from -Wunused-variable to -Wunused-but-set-variable
by writing it in two statements. This fixes the build because the new
warning is in W=1.
Given that whole if(0) {} thing, I don't feel motivated to change
things overly much and quite strongly feel this is the compiler being
daft.
Fixes: cb3e971c435d ("sched: Make struct sched_statistics independent of fair sched class")
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Currently in schedstats we have sum_sleep_runtime and iowait_sum, but
there's no metric to show how long the task is in D state. Once a task in
D state, it means the task is blocked in the kernel, for example the
task may be waiting for a mutex. The D state is more frequent than
iowait, and it is more critital than S state. So it is worth to add a
metric to measure it.
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20210905143547.4668-5-laoar.shao@gmail.com
If we want to use the schedstats facility to trace other sched classes, we
should make it independent of fair sched class. The struct sched_statistics
is the schedular statistics of a task_struct or a task_group. So we can
move it into struct task_struct and struct task_group to achieve the goal.
After the patch, schestats are orgnized as follows,
struct task_struct {
...
struct sched_entity se;
struct sched_rt_entity rt;
struct sched_dl_entity dl;
...
struct sched_statistics stats;
...
};
Regarding the task group, schedstats is only supported for fair group
sched, and a new struct sched_entity_stats is introduced, suggested by
Peter -
struct sched_entity_stats {
struct sched_entity se;
struct sched_statistics stats;
} __no_randomize_layout;
Then with the se in a task_group, we can easily get the stats.
The sched_statistics members may be frequently modified when schedstats is
enabled, in order to avoid impacting on random data which may in the same
cacheline with them, the struct sched_statistics is defined as cacheline
aligned.
As this patch changes the core struct of scheduler, so I verified the
performance it may impact on the scheduler with 'perf bench sched
pipe', suggested by Mel. Below is the result, in which all the values
are in usecs/op.
Before After
kernel.sched_schedstats=0 5.2~5.4 5.2~5.4
kernel.sched_schedstats=1 5.3~5.5 5.3~5.5
[These data is a little difference with the earlier version, that is
because my old test machine is destroyed so I have to use a new
different test machine.]
Almost no impact on the sched performance.
No functional change.
[lkp@intel.com: reported build failure in earlier version]
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mel Gorman <mgorman@suse.de>
Link: https://lore.kernel.org/r/20210905143547.4668-3-laoar.shao@gmail.com
Use a small, non-scaled min granularity for SCHED_IDLE entities, when
competing with normal entities. This reduces the latency of getting
a normal entity back on cpu, at the expense of increased context
switch frequency of SCHED_IDLE entities.
The benefit of this change is to reduce the round-robin latency for
normal entities when competing with a SCHED_IDLE entity.
Example: on a machine with HZ=1000, spawned two threads, one of which is
SCHED_IDLE, and affined to one cpu. Without this patch, the SCHED_IDLE
thread runs for 4ms then waits for 1.4s. With this patch, it runs for
1ms and waits 340ms (as it round-robins with the other thread).
Signed-off-by: Josh Don <joshdon@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org>
Link: https://lore.kernel.org/r/20210820010403.946838-4-joshdon@google.com
