Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'sched_ext-for-6.16-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/sched_ext
Pull sched_ext fixes from Tejun Heo:
"Two fixes in the built-in idle selection helpers:
- Fix prev_cpu handling to guarantee that idle selection never
returns a CPU that's not allowed
- Skip cross-node search when !NUMA which could lead to infinite
looping due to a bug in NUMA iterator"
* tag 'sched_ext-for-6.16-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/sched_ext:
sched_ext: idle: Skip cross-node search with !CONFIG_NUMA
sched_ext: idle: Properly handle invalid prev_cpu during idle selection
+19
-18
+19
-18
kernel/sched/ext_idle.c
+19
-18
kernel/sched/ext_idle.c
···
138
138
goto retry;
139
139
}
140
140
141
+
#ifdef CONFIG_NUMA
141
142
/*
142
143
* Tracks nodes that have not yet been visited when searching for an idle
143
144
* CPU across all available nodes.
···
187
186
188
187
return cpu;
189
188
}
189
+
#else
190
+
static inline s32
191
+
pick_idle_cpu_from_online_nodes(const struct cpumask *cpus_allowed, int node, u64 flags)
192
+
{
193
+
return -EBUSY;
194
+
}
195
+
#endif
190
196
191
197
/*
192
198
* Find an idle CPU in the system, starting from @node.
···
455
447
const struct cpumask *llc_cpus = NULL, *numa_cpus = NULL;
456
448
const struct cpumask *allowed = cpus_allowed ?: p->cpus_ptr;
457
449
int node = scx_cpu_node_if_enabled(prev_cpu);
450
+
bool is_prev_allowed;
458
451
s32 cpu;
459
452
460
453
preempt_disable();
454
+
455
+
/*
456
+
* Check whether @prev_cpu is still within the allowed set. If not,
457
+
* we can still try selecting a nearby CPU.
458
+
*/
459
+
is_prev_allowed = cpumask_test_cpu(prev_cpu, allowed);
461
460
462
461
/*
463
462
* Determine the subset of CPUs usable by @p within @cpus_allowed.
···
478
463
allowed = local_cpus;
479
464
} else {
480
465
cpu = -EBUSY;
481
-
goto out_enable;
482
-
}
483
-
484
-
/*
485
-
* If @prev_cpu is not in the allowed CPUs, skip topology
486
-
* optimizations and try to pick any idle CPU usable by the
487
-
* task.
488
-
*
489
-
* If %SCX_OPS_BUILTIN_IDLE_PER_NODE is enabled, prioritize
490
-
* the current node, as it may optimize some waker->wakee
491
-
* workloads.
492
-
*/
493
-
if (!cpumask_test_cpu(prev_cpu, allowed)) {
494
-
node = scx_cpu_node_if_enabled(smp_processor_id());
495
-
cpu = scx_pick_idle_cpu(allowed, node, flags);
496
466
goto out_enable;
497
467
}
498
468
}
···
525
525
* then avoid a migration.
526
526
*/
527
527
cpu = smp_processor_id();
528
-
if (cpus_share_cache(cpu, prev_cpu) &&
528
+
if (is_prev_allowed && cpus_share_cache(cpu, prev_cpu) &&
529
529
scx_idle_test_and_clear_cpu(prev_cpu)) {
530
530
cpu = prev_cpu;
531
531
goto out_unlock;
···
562
562
/*
563
563
* Keep using @prev_cpu if it's part of a fully idle core.
564
564
*/
565
-
if (cpumask_test_cpu(prev_cpu, idle_cpumask(node)->smt) &&
565
+
if (is_prev_allowed &&
566
+
cpumask_test_cpu(prev_cpu, idle_cpumask(node)->smt) &&
566
567
scx_idle_test_and_clear_cpu(prev_cpu)) {
567
568
cpu = prev_cpu;
568
569
goto out_unlock;
···
612
611
/*
613
612
* Use @prev_cpu if it's idle.
614
613
*/
615
-
if (scx_idle_test_and_clear_cpu(prev_cpu)) {
614
+
if (is_prev_allowed && scx_idle_test_and_clear_cpu(prev_cpu)) {
616
615
cpu = prev_cpu;
617
616
goto out_unlock;
618
617
}