Solved Problems with net/realtek-re-kmod and FreeBSD 14.0

I tried disabling LRO and TSO by adding -tso -lro -vlanhwtso to the ifconfig_xxx in /etc/rc.conf and it didn't help. Those options are not listed in the options line of the ifconfig output for the interface either.
 
eofster said:
I tried disabling LRO and TSO by adding -tso -lro -vlanhwtso to the ifconfig_xxx in /etc/rc.conf and it didn't help. Those options are not listed in the options line of the ifconfig output for the interface either.
Click to expand...
I think so, yes. You can see the options from sysctl.
 
eofster said:
I tried disabling LRO and TSO by adding -tso -lro -vlanhwtso to the ifconfig_xxx in /etc/rc.conf and it didn't help. Those options are not listed in the options line of the ifconfig output for the interface either.
Click to expand...
One thing we can try is to change this in the source:

Code: 
        /* Not enable LRO for OS version lower than 11.0 */
#if OS_VER < VERSION(11,0)
        ifp->if_capenable &= ~IFCAP_LRO;
#endif

I can delete the conditional code and thus disable LRO, then rebuild the driver.

[LATER]

Tried to switch off options with ifconfig and got it working. But there was no logic in it. Seems that the problem is elsewhere. Not just some option not working, but deeper in the driver logic. Some FreeBSD driver expert should look at it.
 
Argentum said:
One thing we can try is to change this in the source:

Code: 
        /* Not enable LRO for OS version lower than 11.0 */
#if OS_VER < VERSION(11,0)
        ifp->if_capenable &= ~IFCAP_LRO;
#endif

I can delete the conditional code and thus disable LRO, then rebuild the driver.
Click to expand...

I tried that, but it did not help :(
 
elgrande said:
I tried that, but it did not help :(
Click to expand...
Tried that also yesterday with the same results. The bug is not that simple.

The other thing I tried yesterday was switching off, one by one, the checksum options manually with ifconfig. At some point it started working, but the situation was nor repeatable, meaning the second time I tried the same sequence, it did not work.

But the clue seems to be in my previous post - it is not about how the driver communicates with HW, it is somewhere higher in the stack. It, sure, needs an experienced driver expert to look at...
 
Already some time ago, the solution was to throw out the RTL8125 from the machine and replace it with Intel based card. The legacy work-around driver actually worked, but the speed in 1Gbit bode was less than 500Mbit/s.
 
The .198 version works sort of with 14.2. However, the newer ones do not -- in some cases, not at all. As in can't get an ARP -> no traffic.

The kernel-included version runs ok EXCEPT that some other hosts will cause it to take watchdog timeouts and reset. It comes back up but obviously that plays hell with speed when it decides to go offline for a couple of seconds at a crack.

Performance with .198 is meh. It gets nearly the full gigabit at least, but it won't negotiate on a 2.5g port reliably and what's maybe considerably worse is that if you run iperf3 against it you see retries all over the place; not consistent at some level, but wildly scattershot which is going to play devil with congestion control algorithms and impact real-world transmission speeds.

And it does.

What I don't understand is that I have a 2.5G RealTek on my PC Mobo running Win11 and it chooches without a problem. Full speed to my server which has a 10G interconnect from that switch, and it blasts right at wire speed. No problem at all So obviously this is a driver issue.... but its pretty bad.
 
You must log in or register to reply here.
Back
Top