On 2026-02-11 10:54:44 [-0800], Jakub Kicinski wrote: > Are you concerned about the latency of delivering the TS to the user > space app / socket? Or purely reading the TS out of the HW fifo to make > space for another packet to be timestamped?
The problem with the global workqueue is the latency. If the system is idle it works as-is but if it gets busy the workqueue can be significantly delayed. You can't give this workqueue a higher priority because the workqueue is anonymous and changes. The next best thing here is the aux_worker because everything here is executed always under the same process/ PID so you can apply a priority and ensure that is handled before other "less important" tasks. However. There is actually no need for the workqueue so you could do it directly (which is what some other driver do). But then I stumbled upon the locking issue… Sebastian
