On Tue, Jan 20, 2026 at 10:00 PM Paul Moses [off-list ref] wrote:

quoted

Also, the AI review [2] pointed out a real issue.
It's easy to reproduce by running something like:

tc action add action gate base-time 200000000000ns \
  sched-entry close 0ns index 10

This was never allowed. A zero interval has always been invalid for a gate schedule entry

The issue is not whether this is valid or not (as you said, it isn't).
The issue is that, with your patch, this will result in a null-ptr deref.
You can avoid this by initialising the timer before calling tcf_idr_release.

[...]
I will prepare and test v3 with your first 8 suggestions and await further input on best practices for avoiding a monolithic patch and on appropriate levels of validation in this specific case.

Ok, the main suggestion is to logically break down (as much as possible),
the introduction of RCU and the other additional cleanups/checks.

cheers,
Victor