Yes, I won't be using a DKMS driver and will test with the in-tree driver. I rolled a PC back to the Debian 6.1 RT kernel and the default in-tree R8169 driver yesterday with isolcpus=2,3. I had a shot at it last night with the napi thread but was confused by the trace results so will try again. I could not see our PacketErrorTest script appearing in the trace but it stopped the trace when it should. Also our RT process did not appear to be on an isolated thread which seemed odd. I did not get as far as setting the affinity for napi. I've also installed Bookworm RC3 on another PC with the default RT kernel on a Gigabyte Brix N3160 with an R8111/R8168 NIC and the in-tree R8169 driver. This PC historically had max latency of about 130000 ns back with kernel 4.x and now its reporting 23000 ns with the 6.1 kernel over 24 hours testing with the linuxcnc tools. Quite amazing what you guys have achieved! It will be a few days before I can lug my hardware to work to test on it as well for another data point. I really appreciate your willingness to help. This remains a frustrating issue for many Linuxcnc users who are using my compiled 6.3 kernel to get out of jail. Bear with me as I bumble my way through! Rod Webster Rod Webster VMN® www.vmn.com.au Ph: 1300 896 832 Mob: +61 435 765 611 On Mon, 22 May 2023 at 19:32, Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> wrote: > > On 2023-05-19 21:41:30 [+1000], Rod Webster wrote: > > I will do some testing tomorrow now I know what's going on but I > > wonder how relevant it will be given this turnaround? > > I'm not going to look at the dkms/OOT driver. If there is an actual > problem or something odd with mainline and RT and I will look into it. > I managed to find an older box which is using the r8169 and I'm going to > boot and see if there the driver is doing something odd what I didn't > see earlier… > > Please make sure that you are using the in-tree driver. If you > experience high latencies then I'm willing to help investigate. > > > Rod Webster > > Sebastian
