Mesa hm2/hm2_7i96s.0: error finishing read

JT wrote:

I might try that tomorrow on my debian 13 test SSD but I'm not trying it on my dev PC LOL.

JT

Let us know how you get on. Just check the package name I used against the experimental link I shared in case they've rolled forward to kernel 6.1 or something silly!

So I followed PCW's advice and added his flags to GRUB commandline
Some improvement in latency but got a following error within a few seconds
So I decided to upgrade the kernel

Followed these instructions wiki.debian.org/DebianExperimental

added this to /etc/sources.listthen ran these commands
Rebooted into the 6.0 PREEMPT_RT Kernel
Latency seemed improved but got latency errors and the 7i76e TMAX did not seeme to improve. I might remove PCW's settings and see what happens.

JT wrote:

I just installed the 5.10.0-18-rt-amd64_5.10.140-1_amd64.deb on my Debian 11 test machine and have my 7i96 running for an hour now with no errors.

JT

Me too!
Reverted to a Gigabyte N3160 USFF PC rthat already had Debian 11 installed (non-rt) and installed 5.10.0-18-rt-amd64 kernel with Synaptic. Compiled master as RIP
Latency was awesome after adding isolcpus=2,3 but I was still getting error finishing read.
Coaxed Linux to tell me that the NIC hardware was a Realtek RT8168.  I installed the Debian r8168-dkms driver and the problem seems to be solved.
hm2_7i76e.0.read.tmax has been sitting stable at 717140  for an hour or so. Before changing the driver it was up over 2,000,000!

So I think its fair to say the issue is Linux is installing the wrong NIC driver by default or has poor driver support for some Realtek NIC's from kernel 5.10 and on.

 

My little favourite PC has been re-released as the Odroid H3+ I'm tempted to grab one but it uses 2 x Realtek RTL8125 NIC's. Much better option thna the PI with An Intel Celeron.. if the NIC works... USD $165 so not a bad price for something that fits in the palm of your hand... I have a couple of the Odroid H2+ but th eone I use for LCNC is running Ethercat

Gents,
          I am seeing something similar with an 7i95, The machine was stable on 2.8 but we have made our own GUI with qtpyvcp and python 3 and to run this we need 2.9 and running 2.9 we are getting spurious error finishing read! iter= error, I have been trying to pin this down as to what is going but am struggling. As can be seen from the HAL scope attached the latency is good most of the time then every so often (every coupe of seconds) the hm2_7i95.0.read.time jumps by factor of 10x then sure enough the read timeout error occurs. I have played around with kernel settings this has virtualy no effect, using the r8168-dkms driver has massive effect on bringing down the general read.time but the the spurious spikes are still there.
5.10.0-18-rt-amd64
PRETTY_NAME="Debian GNU/Linux 11 (bullseye)"
model name      : Intel(R) Celeron(R) CPU  J1900  @ 1.99GHz