Re: [PATCH] net: af_packet: add af_packet hrtimer mode

On Wed, 2025-08-08 at 14:01 +0800, Eric wrote:

I have some doubts why AF_PACKET timer would be involved in the performance
of an application.

Are you sure about your requirements ?

I do not know what application you are running, but I wonder if using
TX timestamps,
and EDT model to deliver your packets in a more accurate way would be better.

Some qdiscs already use hrtimer and get sub 50 usec latency out of the box,
or whatever time a cpu sleeping in deep state takes to wake up.

Dear Eric,

Let me explain our requirements. Initially, we used regular UDP reception
function to process LiDAR data packets, and at that time, the packet latency
was acceptable. However, due to insufficient CPU resources on our system, we
needed to optimize the CPU consumption of various tasks, including the task
of processing LiDAR data packets.

Therefore, our original requirement was to reduce the CPU consumption and
context switching overhead of the program that processes LiDAR data. To 
achieve this, we employed the AF_PACKET + MMAP programming model and found
that it indeed resulted in significant improvements. However, we noticed a
side effect: even with a 2-millisecond retire timeout set, the processing
latency for received packets still fluctuated to over 8 milliseconds.

Nevertheless, we still wanted to continue using the AF_PACKET + MMAP
programming model to improve CPU utilization. Therefore, we modified the
implementation of AF_PACKET and switched to using hrtimer, which resulted
in keeping the processing latency within 2 milliseconds. In this way, we
can continue to use the AF_PACKET + MMAP programming model to optimize
the LiDAR processing tasks.


Thanks
Xin Zhao