]> git.ipfire.org Git - thirdparty/kernel/linux.git/commit
sched_ext: Enforce scheduling authority in dispatch and select_cpu operations
authorTejun Heo <tj@kernel.org>
Fri, 6 Mar 2026 17:58:03 +0000 (07:58 -1000)
committerTejun Heo <tj@kernel.org>
Fri, 6 Mar 2026 17:58:03 +0000 (07:58 -1000)
commita5fa0708cbfda4d3c2c6a447de7c4b0b23595527
treeb10473d69b54dbb752fe994f99462e824782564d
parent105dcd005be2ac1d5541921db8feb1d0f98d59d5
sched_ext: Enforce scheduling authority in dispatch and select_cpu operations

Add checks to enforce scheduling authority boundaries when multiple
schedulers are present:

1. In scx_dsq_insert_preamble() and the dispatch retry path, ignore attempts
   to insert tasks that the scheduler doesn't own, counting them via
   SCX_EV_INSERT_NOT_OWNED. As BPF schedulers are allowed to ignore
   dequeues, such attempts can occur legitimately during sub-scheduler
   enabling when tasks move between schedulers. The counter helps distinguish
   normal cases from scheduler bugs.

2. For scx_bpf_dsq_insert_vtime() and scx_bpf_select_cpu_and(), error out
   when sub-schedulers are attached. These functions lack the aux__prog
   parameter needed to identify the calling scheduler, so they cannot be used
   safely with multiple schedulers. BPF programs should use the arg-wrapped
   versions (__scx_bpf_dsq_insert_vtime() and __scx_bpf_select_cpu_and())
   instead.

These checks ensure that with multiple concurrent schedulers, scheduler
identity can be properly determined and unauthorized task operations are
prevented or tracked.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
kernel/sched/ext.c
kernel/sched/ext_idle.c
kernel/sched/ext_internal.h