Project management for areas/mem
Table of Contents
- Important medium-term tasks @medium
- TODO page_pool handling in-flight frames during shutdown
- DONE Test if __xdp_return() can hit no page_pool id issue
- DONE xdp_rxq_info_unreg_mem_model take over calling page_pool_destroy
- TODO page_pool inflight: howto delay/postpone work in kernel
- NEXT page_pool inflight: Code solution#3
- TODO Create tracepoints for page_pool
- TODO Create bpftrace scripts monitoring page_pool stats (via tracepoints)
- Site navigation
This document contains org-mode tasks and TODOs for files under: areas/mem/.
It is recommended to use emacs when viewing and editing these .org files, as
the github rendering view removes the TODO and DONE marking on the tasks.
Top-level headings indicate category, as explained in ../xdp-project.html.
Important medium-term tasks @medium
TODO page_pool handling in-flight frames during shutdown
DONE Test if __xdp_return() can hit no page_pool id issue
Confirmed this can happen.
DONE xdp_rxq_info_unreg_mem_model take over calling page_pool_destroy
When delaying freeing the registered XDP memory model, then we cannot have drivers calling page_pool_destroy() directly. Figure out a clean code way to do this.
TODO page_pool inflight: howto delay/postpone work in kernel
What is the best way to delay taking down page_pool object when driver calls xdp_rxq_info_unreg_mem_model(), in the case page_pool detect/have in-flight frames?
Should we create a workqueue?
#include <linux/workqueue.h> static struct workqueue_struct *wq; wq = alloc_workqueue("xdp_mem_dismantle_wq", 0, 0); struct work_struct free_work; // embed in struct xdp_mem_allocator? INIT_WORK(&free_work, __unreg_mem_model_schedule); queue_work(wq, &free_work); queue_delayed_work(wq, struct delayed_work *dwork, delay);
It might be overkill to create a new workqueue, what about instead using the default worker thread 'events/n'.
The function calls used instead of queue_work and queue_delayed_work.
schedule_work(wq); schedule_delayed_work(wq, delay);
There is an example in function amba_deferred_retry_func() how to reschedule
the work-queue function from itself via schedule_delayed_work.