On 8/11/26 9:54 AM, Tejun Heo wrote:
Hello,
On Mon, Aug 10, 2026 at 05:17:26PM -0700, JP Kobryn wrote:
...
The get/put kfuncs above use KF_ACQUIRE/RELEASE so the verifier can
reject memcg access after the put. Replacing them with a generic API
would call for tracking a new lifetime relationship between the css and
memcg (or other derived) pointer.
So, if you need css -> memcg casting, I think that can just be RCU
protected. In fact, even for the iteration and access, RCU protection
probably makes more sense then KF_ACQUIRE/RELEASE. It's simpler and less
overhead especially given that most bpf progs that people care about are
non-sleepable and already in RCU critical section anyway.
It's just that if flushing is involved, the cond_resched() within
css_rstat_flush() would preclude RCU.
