diff options
| -rw-r--r-- | kernel/sched/ext.c | 29 |
1 files changed, 21 insertions, 8 deletions
diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c index e01144340d67..54edd0e2132a 100644 --- a/kernel/sched/ext.c +++ b/kernel/sched/ext.c @@ -2344,6 +2344,25 @@ static bool task_can_run_on_remote_rq(struct task_struct *p, struct rq *rq, SCHED_WARN_ON(task_cpu(p) == cpu); /* + * If @p has migration disabled, @p->cpus_ptr is updated to contain only + * the pinned CPU in migrate_disable_switch() while @p is being switched + * out. However, put_prev_task_scx() is called before @p->cpus_ptr is + * updated and thus another CPU may see @p on a DSQ inbetween leading to + * @p passing the below task_allowed_on_cpu() check while migration is + * disabled. + * + * Test the migration disabled state first as the race window is narrow + * and the BPF scheduler failing to check migration disabled state can + * easily be masked if task_allowed_on_cpu() is done first. + */ + if (unlikely(is_migration_disabled(p))) { + if (trigger_error) + scx_ops_error("SCX_DSQ_LOCAL[_ON] cannot move migration disabled %s[%d] from CPU %d to %d", + p->comm, p->pid, task_cpu(p), cpu); + return false; + } + + /* * We don't require the BPF scheduler to avoid dispatching to offline * CPUs mostly for convenience but also because CPUs can go offline * between scx_bpf_dsq_insert() calls and here. Trigger error iff the @@ -2351,17 +2370,11 @@ static bool task_can_run_on_remote_rq(struct task_struct *p, struct rq *rq, */ if (!task_allowed_on_cpu(p, cpu)) { if (trigger_error) - scx_ops_error("SCX_DSQ_LOCAL[_ON] verdict target cpu %d not allowed for %s[%d]", - cpu_of(rq), p->comm, p->pid); + scx_ops_error("SCX_DSQ_LOCAL[_ON] target CPU %d not allowed for %s[%d]", + cpu, p->comm, p->pid); return false; } - /* - * If @p has migration disabled, @p->cpus_ptr only contains its current - * CPU and the above task_allowed_on_cpu() test should have failed. - */ - SCHED_WARN_ON(is_migration_disabled(p)); - if (!scx_rq_online(rq)) return false; |