Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge branch 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Pull scheduler fix from Thomas Gleixner:
"Revert the new NUMA aware placement approach which turned out to
create more problems than it solved"
* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
Revert "sched/numa: Delay retrying placement for automatic NUMA balance after wake_affine()"
+1
-56
+1
-56
kernel/sched/fair.c
+1
-56
kernel/sched/fair.c
···
1854
1854
static void numa_migrate_preferred(struct task_struct *p)
1855
1855
{
1856
1856
unsigned long interval = HZ;
1857
-
unsigned long numa_migrate_retry;
1858
1857
1859
1858
/* This task has no NUMA fault statistics yet */
1860
1859
if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults))
···
1861
1862
1862
1863
/* Periodically retry migrating the task to the preferred node */
1863
1864
interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
1864
-
numa_migrate_retry = jiffies + interval;
1865
-
1866
-
/*
1867
-
* Check that the new retry threshold is after the current one. If
1868
-
* the retry is in the future, it implies that wake_affine has
1869
-
* temporarily asked NUMA balancing to backoff from placement.
1870
-
*/
1871
-
if (numa_migrate_retry > p->numa_migrate_retry)
1872
-
return;
1873
-
1874
-
/* Safe to try placing the task on the preferred node */
1875
-
p->numa_migrate_retry = numa_migrate_retry;
1865
+
p->numa_migrate_retry = jiffies + interval;
1876
1866
1877
1867
/* Success if task is already running on preferred CPU */
1878
1868
if (task_node(p) == p->numa_preferred_nid)
···
5910
5922
return this_eff_load < prev_eff_load ? this_cpu : nr_cpumask_bits;
5911
5923
}
5912
5924
5913
-
#ifdef CONFIG_NUMA_BALANCING
5914
-
static void
5915
-
update_wa_numa_placement(struct task_struct *p, int prev_cpu, int target)
5916
-
{
5917
-
unsigned long interval;
5918
-
5919
-
if (!static_branch_likely(&sched_numa_balancing))
5920
-
return;
5921
-
5922
-
/* If balancing has no preference then continue gathering data */
5923
-
if (p->numa_preferred_nid == -1)
5924
-
return;
5925
-
5926
-
/*
5927
-
* If the wakeup is not affecting locality then it is neutral from
5928
-
* the perspective of NUMA balacing so continue gathering data.
5929
-
*/
5930
-
if (cpu_to_node(prev_cpu) == cpu_to_node(target))
5931
-
return;
5932
-
5933
-
/*
5934
-
* Temporarily prevent NUMA balancing trying to place waker/wakee after
5935
-
* wakee has been moved by wake_affine. This will potentially allow
5936
-
* related tasks to converge and update their data placement. The
5937
-
* 4 * numa_scan_period is to allow the two-pass filter to migrate
5938
-
* hot data to the wakers node.
5939
-
*/
5940
-
interval = max(sysctl_numa_balancing_scan_delay,
5941
-
p->numa_scan_period << 2);
5942
-
p->numa_migrate_retry = jiffies + msecs_to_jiffies(interval);
5943
-
5944
-
interval = max(sysctl_numa_balancing_scan_delay,
5945
-
current->numa_scan_period << 2);
5946
-
current->numa_migrate_retry = jiffies + msecs_to_jiffies(interval);
5947
-
}
5948
-
#else
5949
-
static void
5950
-
update_wa_numa_placement(struct task_struct *p, int prev_cpu, int target)
5951
-
{
5952
-
}
5953
-
#endif
5954
-
5955
5925
static int wake_affine(struct sched_domain *sd, struct task_struct *p,
5956
5926
int this_cpu, int prev_cpu, int sync)
5957
5927
{
···
5925
5979
if (target == nr_cpumask_bits)
5926
5980
return prev_cpu;
5927
5981
5928
-
update_wa_numa_placement(p, prev_cpu, target);
5929
5982
schedstat_inc(sd->ttwu_move_affine);
5930
5983
schedstat_inc(p->se.statistics.nr_wakeups_affine);
5931
5984
return target;