]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
sched_ext: idle: Recheck prev_cpu after narrowing allowed mask
authorDavid Carlier <devnexen@gmail.com>
Thu, 30 Apr 2026 09:27:47 +0000 (10:27 +0100)
committerTejun Heo <tj@kernel.org>
Mon, 4 May 2026 21:01:04 +0000 (11:01 -1000)
scx_select_cpu_dfl() narrows @allowed to @cpus_allowed & @p->cpus_ptr
when the BPF caller supplies a @cpus_allowed that differs from
@p->cpus_ptr and @p doesn't have full affinity. However,
@is_prev_allowed was computed against the original (wider)
@cpus_allowed, so the prev_cpu fast paths could pick a @prev_cpu that
is in @cpus_allowed but not in @p->cpus_ptr, violating the intended
invariant that the returned CPU is always usable by @p. The kernel
masks this via the SCX_EV_SELECT_CPU_FALLBACK fallback, but the
behavior contradicts the documented contract.

Move the @is_prev_allowed evaluation past the narrowing block so it
tests against the final @allowed mask.

Fixes: ee9a4e92799d ("sched_ext: idle: Properly handle invalid prev_cpu during idle selection")
Cc: stable@vger.kernel.org # v6.16+
Assisted-by: Claude <noreply@anthropic.com>
Signed-off-by: David Carlier <devnexen@gmail.com>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
kernel/sched/ext_idle.c

index 7468560a6d80414504b6e1266814c2216706d0cd..6e1980763270dffdf29cab6d587c921c56ba3ec5 100644 (file)
@@ -465,12 +465,6 @@ s32 scx_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags,
 
        preempt_disable();
 
-       /*
-        * Check whether @prev_cpu is still within the allowed set. If not,
-        * we can still try selecting a nearby CPU.
-        */
-       is_prev_allowed = cpumask_test_cpu(prev_cpu, allowed);
-
        /*
         * Determine the subset of CPUs usable by @p within @cpus_allowed.
         */
@@ -487,6 +481,12 @@ s32 scx_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags,
                }
        }
 
+       /*
+        * Check whether @prev_cpu is still within the allowed set. If not,
+        * we can still try selecting a nearby CPU.
+        */
+       is_prev_allowed = cpumask_test_cpu(prev_cpu, allowed);
+
        /*
         * This is necessary to protect llc_cpus.
         */