Re: [PATCH v4 sched_ext/for-7.3 39/40] tools/sched_ext: scx_qmap - Expand hierarchical sub-scheduling

On Wed, 08 Jul 2026 22:03:52 +0000, [email protected] wrote:

1. Not reachable in qmap contexts - the cid table is populated for every
   possible cpu before a scheduler can run, so scx_bpf_task_cid() never
   returns negative once the callbacks fire. The cmask helpers also take
   u32 and bounds-check, so even a negative would fail the range test
   rather than access memory. No change.

2. Real policy nit - fixed in v5 by masking the highpri target pick with
   self_cids. Kernel-side enforcement is copy-down by design (an ancestor
   keeps its caps on delegated cids), so exclusivity is the demo's own
   policy to enforce.

3. Benign - a stale or zeroed id just makes scx_bpf_sub_dispatch() miss
   the hash lookup and return false, a wasted no-op call at worst. v5
   reads the id once into a local anyway.

4. Same as the v3 round - self-healing, the fields are re-derived by the
   following redistribute() and the clean lock can't wrap the cmask ops
   as they call kfuncs:

     https://lore.kernel.org/r/[email protected]

5. Known tolerated nit - the weight is advisory and self-corrects on the
   next pass. v5 adds a comment noting it.

Thanks.

-- 
tejun