mirrors/linux-stable
1
0
Fork 0
mirror of https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git synced 2024-09-28 21:33:52 +00:00
Code Issues Releases Wiki Activity
linux-stable/kernel
History
Raghavendra K T 84db47ca71 sched/numa: Fix mm numa_scan_seq based unconditional scan 
Since commit fc137c0dda ("sched/numa: enhance vma scanning logic")

NUMA Balancing allows updating PTEs to trap NUMA hinting faults if the
task had previously accessed VMA. However unconditional scan of VMAs are
allowed during initial phase of VMA creation until process's
mm numa_scan_seq reaches 2 even though current task had not accessed VMA.

Rationale:
 - Without initial scan subsequent PTE update may never happen.
 - Give fair opportunity to all the VMAs to be scanned and subsequently
understand the access pattern of all the VMAs.

But it has a corner case where, if a VMA is created after some time,
process's mm numa_scan_seq could be already greater than 2.

For e.g., values of mm numa_scan_seq when VMAs are created by running
mmtest autonuma benchmark briefly looks like:
start_seq=0 : 459
start_seq=2 : 138
start_seq=3 : 144
start_seq=4 : 8
start_seq=8 : 1
start_seq=9 : 1
This results in no unconditional PTE updates for those VMAs created after
some time.

Fix:
 - Note down the initial value of mm numa_scan_seq in per VMA start_seq.
 - Allow unconditional scan till start_seq + 2.

Result:
SUT: AMD EPYC Milan with 2 NUMA nodes 256 cpus.
base kernel: upstream 6.6-rc6 with Mels patches [1] applied.

kernbench
==========		base                  patched %gain
Amean    elsp-128      165.09 ( 0.00%)      164.78 *   0.19%*

Duration User       41404.28    41375.08
Duration System      9862.22     9768.48
Duration Elapsed      519.87      518.72

Ops NUMA PTE updates           1041416.00      831536.00
Ops NUMA hint faults            263296.00      220966.00
Ops NUMA pages migrated         258021.00      212769.00
Ops AutoNUMA cost                 1328.67        1114.69

autonumabench

NUMA01_THREADLOCAL
==================
Amean  elsp-NUMA01_THREADLOCAL   81.79 (0.00%)  67.74 *  17.18%*

Duration User       54832.73    47379.67
Duration System        75.00      185.75
Duration Elapsed      576.72      476.09

Ops NUMA PTE updates                  394429.00    11121044.00
Ops NUMA hint faults                    1001.00     8906404.00
Ops NUMA pages migrated                  288.00     2998694.00
Ops AutoNUMA cost                          7.77       44666.84

Signed-off-by: Raghavendra K T <raghavendra.kt@amd.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Mel Gorman <mgorman@suse.de>
Link: https://lore.kernel.org/r/2ea7cbce80ac7c62e90cbfb9653a7972f902439f.1697816692.git.raghavendra.kt@amd.com
2023-11-15 09:57:46 +01:00
..
bpf bpf: Let verifier consider {task,cgroup} is trusted in bpf_iter_reg 2023-11-07 15:24:25 -08:00
cgroup sched: psi: fix unprivileged polling against cgroups 2023-11-14 22:27:00 +01:00
configs hardening: Provide Kconfig fragments for basic options 2023-09-22 09:50:55 -07:00
debug kdb: Corrects comment for kdballocenv 2023-11-06 17:13:55 +00:00
dma swiotlb: fix out-of-bounds TLB allocations with CONFIG_SWIOTLB_DYNAMIC 2023-11-08 16:27:05 +01:00
entry entry: Remove empty addr_limit_user_check() 2023-08-23 10:32:39 +02:00
events Many singleton patches against the MM code. The patch series which are 2023-11-02 19:38:47 -10:00
futex io_uring-futex-2023-10-30 2023-11-01 11:25:08 -10:00
gcov gcov: annotate struct gcov_iterator with __counted_by 2023-10-18 14:43:22 -07:00
irq As usual, lots of singleton and doubleton patches all over the tree and 2023-11-02 20:53:31 -10:00
kcsan mm: delete checks for xor_unlock_is_negative_byte() 2023-10-18 14:34:17 -07:00
livepatch livepatch: Fix missing newline character in klp_resolve_symbols() 2023-09-20 11:24:18 +02:00
locking RCU pull request for v6.7 2023-10-30 18:01:41 -10:00
module This update includes the following changes: 2023-11-02 16:15:30 -10:00
power Power management updates for 6.7-rc1 2023-10-31 15:38:12 -10:00
printk TTY/Serial changes for 6.7-rc1 2023-11-03 15:44:25 -10:00
rcu RCU fixes for v6.7 2023-11-08 09:47:52 -08:00
sched sched/numa: Fix mm numa_scan_seq based unconditional scan 2023-11-15 09:57:46 +01:00
time Networking changes for 6.7. 2023-10-31 05:10:11 -10:00
trace tracing/kprobes: Fix the order of argument descriptions 2023-11-11 08:00:43 +09:00
.gitignore
acct.c fs: rename __mnt_{want,drop}_write*() helpers 2023-09-11 15:05:50 +02:00
async.c
audit.c audit: move trailing statements to next line 2023-08-15 18:16:14 -04:00
audit.h audit: correct audit_filter_inodes() definition 2023-07-21 12:17:25 -04:00
audit_fsnotify.c
audit_tree.c As usual, lots of singleton and doubleton patches all over the tree and 2023-11-02 20:53:31 -10:00
audit_watch.c audit: don't take task_lock() in audit_exe_compare() code path 2023-10-25 22:29:47 -04:00
auditfilter.c audit: move trailing statements to next line 2023-08-15 18:16:14 -04:00
auditsc.c audit,io_uring: io_uring openat triggers audit reference count underflow 2023-10-13 18:34:46 +02:00
backtracetest.c
bounds.c
capability.c lsm: constify the 'target' parameter in security_capget() 2023-08-08 16:48:47 -04:00
cfi.c
compat.c sched_getaffinity: don't assume 'cpumask_size()' is fully initialized 2023-03-14 19:32:38 -07:00
configs.c
context_tracking.c locking/atomic: treewide: use raw_atomic*_<op>() 2023-06-05 09:57:20 +02:00
cpu.c asm-generic updates for v6.7 2023-11-01 15:28:33 -10:00
cpu_pm.c
crash_core.c crash_core.c: remove unneeded functions 2023-10-04 10:41:58 -07:00
crash_dump.c
cred.c lsm/stable-6.7 PR 20231030 2023-10-30 20:13:17 -10:00
delayacct.c delayacct: track delays from IRQ/SOFTIRQ 2023-04-18 16:39:34 -07:00
dma.c
exec_domain.c
exit.c As usual, lots of singleton and doubleton patches all over the tree and 2023-11-02 20:53:31 -10:00
exit.h exit: add internal include file with helpers 2023-09-21 12:03:50 -06:00
extable.c
fail_function.c
fork.c As usual, lots of singleton and doubleton patches all over the tree and 2023-11-02 20:53:31 -10:00
freezer.c freezer,sched: Use saved_state to reduce some spurious wakeups 2023-09-18 08:14:36 +02:00
gen_kheaders.sh Revert "kheaders: substituting --sort in archive creation" 2023-05-28 16:20:21 +09:00
groups.c groups: Convert group_info.usage to refcount_t 2023-09-29 11:28:39 -07:00
hung_task.c kernel/hung_task.c: set some hung_task.c variables storage-class-specifier to static 2023-04-08 13:45:37 -07:00
iomem.c kernel/iomem.c: remove __weak ioremap_cache helper 2023-08-21 13:37:28 -07:00
irq_work.c trace: Add trace_ipi_send_cpu() 2023-03-24 11:01:29 +01:00
jump_label.c
kallsyms.c kallsyms: Change func signature for cleanup_symbol_name() 2023-08-25 15:00:36 -07:00
kallsyms_internal.h
kallsyms_selftest.c Modules changes for v6.6-rc1 2023-08-29 17:32:32 -07:00
kallsyms_selftest.h
kcmp.c file: convert to SLAB_TYPESAFE_BY_RCU 2023-10-19 11:02:48 +02:00
Kconfig.freezer
Kconfig.hz
Kconfig.kexec docs: fix link s390/zfcpdump.rst 2023-10-04 10:41:56 -07:00
Kconfig.locks
Kconfig.preempt
kcov.c kcov: add prototypes for helper functions 2023-06-09 17:44:17 -07:00
kexec.c kernel: kexec: copy user-array safely 2023-10-09 16:59:47 +10:00
kexec_core.c crash_core: move crashk_*res definition into crash_core.c 2023-10-04 10:41:58 -07:00
kexec_elf.c
kexec_file.c integrity-v6.6 2023-08-30 09:16:56 -07:00
kexec_internal.h
kheaders.c kheaders: Use array declaration instead of char 2023-03-24 20:10:59 -07:00
kprobes.c kprobes: kretprobe scalability improvement 2023-10-18 23:59:54 +09:00
ksyms_common.c kallsyms: make kallsyms_show_value() as generic function 2023-06-08 12:27:20 -07:00
ksysfs.c crash: hotplug support for kexec_load() 2023-08-24 16:25:14 -07:00
kthread.c As usual, lots of singleton and doubleton patches all over the tree and 2023-11-02 20:53:31 -10:00
latencytop.c
Makefile v6.5-rc1-modules-next 2023-06-28 15:51:08 -07:00
module_signature.c
notifier.c notifiers: add tracepoints to the notifiers infrastructure 2023-04-08 13:45:38 -07:00
nsproxy.c nsproxy: Convert nsproxy.count to refcount_t 2023-08-21 11:29:12 -07:00
padata.c padata: Fix refcnt handling in padata_free_shell() 2023-10-27 18:04:24 +08:00
panic.c panic: use atomic_try_cmpxchg in panic() and nmi_panic() 2023-10-04 10:41:56 -07:00
params.c kernel: params: Remove unnecessary ‘0’ values from err 2023-07-10 12:47:01 -07:00
pid.c pidfd: prevent a kernel-doc warning 2023-09-19 13:21:33 -07:00
pid_namespace.c pid: pid_ns_ctl_handler: remove useless comment 2023-10-04 10:41:57 -07:00
pid_sysctl.h memfd: replace ratcheting feature from vm.memfd_noexec with hierarchy 2023-08-21 13:37:59 -07:00
profile.c
ptrace.c mm: make __access_remote_vm() static 2023-10-18 14:34:15 -07:00
range.c
reboot.c kernel/reboot: Add device to sys_off_handler 2023-07-28 11:33:09 +01:00
regset.c
relay.c kernel: relay: remove unnecessary NULL values from relay_open_buf 2023-08-18 10:18:55 -07:00
resource.c resource: Unify next_resource() and next_resource_skip_children() 2023-10-05 13:58:07 +02:00
resource_kunit.c
rseq.c
scftorture.c scftorture: Pause testing after memory-allocation failure 2023-07-14 15:02:57 -07:00
scs.c
seccomp.c seccomp: Add missing kerndoc notations 2023-08-17 12:32:15 -07:00
signal.c As usual, lots of singleton and doubleton patches all over the tree and 2023-11-02 20:53:31 -10:00
smp.c CSD lock commits for v6.7 2023-10-30 17:56:53 -10:00
smpboot.c kthread: add kthread_stop_put 2023-10-04 10:41:57 -07:00
smpboot.h
softirq.c sched/core: introduce sched_core_idle_cpu() 2023-07-13 15:21:50 +02:00
stackleak.c stackleak: allow to specify arch specific stackleak poison function 2023-04-20 11:36:35 +02:00
stacktrace.c stacktrace: Export stack_trace_save_tsk 2023-09-11 23:59:47 -04:00
static_call.c
static_call_inline.c
stop_machine.c
sys.c As usual, lots of singleton and doubleton patches all over the tree and 2023-11-02 20:53:31 -10:00
sys_ni.c asm-generic updates for v6.7 2023-11-01 15:28:33 -10:00
sysctl-test.c
sysctl.c asm-generic updates for v6.7 2023-11-01 15:28:33 -10:00
task_work.c task_work: add kerneldoc annotation for 'data' argument 2023-09-19 13:21:32 -07:00
taskstats.c taskstats: fill_stats_for_tgid: use for_each_thread() 2023-10-04 10:41:57 -07:00
torture.c torture: Print out torture module parameters 2023-09-24 17:24:01 +02:00
tracepoint.c
tsacct.c
ucount.c sysctl: Add size to register_sysctl 2023-08-15 15:26:17 -07:00
uid16.c
uid16.h
umh.c sysctl: fix unused proc_cap_handler() function warning 2023-06-29 15:19:43 -07:00
up.c smp: Change function signatures to use call_single_data_t 2023-09-13 14:59:24 +02:00
user-return-notifier.c
user.c binfmt_misc: enable sandboxed mounts 2023-10-11 08:46:01 -07:00
user_namespace.c As usual, lots of singleton and doubleton patches all over the tree and 2023-11-02 20:53:31 -10:00
usermode_driver.c
utsname.c
utsname_sysctl.c utsname: simplify one-level sysctl registration for uts_kern_table 2023-04-13 11:49:35 -07:00
vhost_task.c vhost: Fix worker hangs due to missed wake up calls 2023-06-08 15:43:09 -04:00
watch_queue.c kernel: watch_queue: copy user-array safely 2023-10-09 16:59:48 +10:00
watchdog.c watchdog: move softlockup_panic back to early_param 2023-11-01 12:10:02 -07:00
watchdog_buddy.c watchdog/hardlockup: move SMP barriers from common code to buddy code 2023-06-19 16:25:28 -07:00
watchdog_perf.c watchdog/perf: add a weak function for an arch to detect if perf can use NMIs 2023-06-09 17:44:21 -07:00
workqueue.c As usual, lots of singleton and doubleton patches all over the tree and 2023-11-02 20:53:31 -10:00
workqueue_internal.h workqueue: Drop the special locking rule for worker->flags and worker_pool->flags 2023-08-07 15:57:22 -10:00