Mesa hm2/hm2_7i96s.0: error finishing read

GRUB_CMDLINE_LINUX_DEFAULT="quiet lapic=notscdeadline hpet=disable i915.i915_enable_rc6=0 i915.powersave=0 intel_idle.max_cstate=1 processor.max_cstate=1 isolcpus=1 idle=poll"

That little lot significantly improves things it makes much more stable, it reduces the frequency and magnitude of the big delay spikes.

You might also try running Intels powertop utility and setting all power saving options to "bad"

So extending the servo period does not seem to overcome the issues as the read.tmas also extends when the timing spikes occur, i left it on soak test overnight and sure enough i got an "error finishing read! iter=" my read hm2_7i95.0.read.tmax hit a record 3376032 and servo .tmax hit 3597792 both with servo period set to SERVO_PERIOD = 2000000 I will try the exact hw with my older 2.8 build and also some different hw with this setup and post the results. Is it possible to make the error finishing read non fatal?

The issue follows the OS / Kernel or LinuxCNC version, not the hardware (hardware has small effect). Will post the stats later.

"Error finishing read" means that too many successive timeouts have occurred

Default settings are: timeout means >80% of the servo period,
and 5 timeouts in a row result in a fatal I/O error: "error finishing read"

You can tune the controlling parameters:

hm2_7i96s.0.packet-read-timeout (default = 80%)
hm2_7i96s.0.packet-error-increment (default = 2)
hm2_7i96s.0.packet-error-decrement (default = 1)
hm2_7i96s.0.packet-error-limit (default=10)

many thanks, is the packet-error exposed to hal at all i was struggling to see it.

Yes:
hm2_7i96s.0.packet-error
hm2_7i96s.0.packet-error-exceeded
hm2_7i96s.0.packet-error-level

Note that packet-error can be useful to prevent bogus velocity corrections
in the case of lost/dropped packets:

# position command and feedback
net emcmot.00.pos-cmd joint.0.motor-pos-cmd => pid.0.command mux2.0.in1
net stepgen0fb hm2_[HOSTMOT2](BOARD).0.stepgen.00.position-fb mux2.0.in0
net motor.00.pos-fb <= mux2.0.out joint.0.motor-pos-fb pid.0.feedback
net motor.00.command pid.0.output hm2_[HOSTMOT2](BOARD).0.stepgen.00.velocity-cmd
net ioerror hm2_[HOSTMOT2](BOARD).0.packet-error mux2.0.sel

This hal section uses the packet error pin to select the commanded position
rather that the [stale] feedback position for loop feedback in the event of packet
loss
 

Ok so next steps? There are still to many variables to figure out if its kernel or linuxcnc its self.
Has anyone got 2.9 running a mesa Ethernet card 7i95 or 96 etc stably?

I looked at building 2.8 for deb-11 to see if its in CNC or the OS but i looks like this is not really possible.

I looked at getting the 4.19 series kernel in deb-11 and that is not easy either, would have to be a source config and build.

Yes 4.19 is a better choice on most hardware but
5.10 and 6.0 work for me on my test CPU (HP Elite 800).

I dont think the LinuxCNC version is related as there have been no
significant upper level driver changes to hm2_eth for quite a while

( Though this reminds me that I should add and accumulated error count )