Re: rge(4) RTL8126 simple testing
- In reply to: Adrian Chadd : "Re: rge(4) RTL8126 simple testing"
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Sat, 28 Mar 2026 18:59:17 UTC
On Wed, 25 Mar 2026, Adrian Chadd wrote: > sounds like maybe ASPM isn't setup completely correctly on RC/EP? As far as I know about ASPM, admittedly not a lot, it is disabled on my motherboard. The BIOS says it is disabled, and a BIOS would never lie. Right? ;) Sean -- scf@FreeBSD.org > On Wed, 25 Mar 2026 at 13:37, Sean C. Farley <scf@freebsd.org> wrote: >> >> On Tue, 24 Mar 2026, Sean C. Farley wrote: >> >>> On Fri, 6 Mar 2026, Adrian Chadd wrote: >>> >>>> me and when i poke the foundation about it :) >>> >>> Another point of interest with these NICs. The one that is connected >>> to the Nokia ONT XS-010X-Q stops functioning at random times, usually >>> days apart. The interface shows it as up and the ONT looks like it is >>> happy and even blinks showing data is flowing (in at least one >>> direction), but the only packets I see on the NIC are ARP requests for >>> the upstream router. >>> >>> The other two NICs have not shown this issue. >>> >>> The resolution is to either bring the NIC down then up or merely >>> changing the options on the NIC. This makes me suspect it is the NIC >>> at fault but unsure if it is the driver or this particular NIC. I >>> guess I can swap the NICs around to see if the problem resolves itself >>> or follows the NIC. >>> >>> One other thing to point out is that this ONT is new, so it could >>> still be the faulty bit. >> >> I have not tested this very long, but there is a possibility that hangs >> were caused by an issue with the bus the one NIC was on. The comment I >> just added to the issue to update the port >> https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=294029: >> >> If anyone becomes curious about the "ClockPM disabled" symptom I >> had, it was due to something going wrong with the PCIe bus >> associated with the CPU. Removing other cards from slots associated >> with the CPU did not help. Moving the NIC to a PCH-associated slot >> did. This should stop the hangs. It already removed all retries >> seen using iperf against a public server. >> >> I discovered it after bouncing things off of AI and discovering >> about the "-e" option to pciconf. "Hmm. What shows when I provide >> this -e option, to report errors, from the manpage? Oh!" >> >> ... >> ecap 0001[100] = AER 2 0 fatal 0 non-fatal 2 corrected >> ecap 0002[148] = VC 1 max VC0 >> ecap 0003[170] = Serial 1 01000000684ce000 >> ecap 0019[180] = PCIe Sec 1 lane errors 0 >> ecap 0017[190] = TPH Requester 1 >> ecap 0018[21c] = LTR 1 >> ecap 001e[224] = L1 PM Substates 1 >> ecap 000b[234] = Vendor [1] ID 0002 Rev 4 Length 256 >> PCI-e errors = Correctable Error Detected >> Unsupported Request Detected >> Corrected = Replay Timer Timeout >> Advisory Non-Fatal Error >> >> Sean >> -- >> scf@FreeBSD.org >> >> >>>> On Thu, 5 Mar 2026 at 09:05, Sean C. Farley <scf@freebsd.org> wrote: >>>>> >>>>> On Wed, 4 Mar 2026, Adrian Chadd wrote: >>>>> >>>>>> That's all good to know; if the vendor driver / linux driver can't >>>>>> negotiate 2.5g then we can try poking realtek directly for help. >>>>>> They're more likely to help if reproducible in these drivers. >>>>>> >>>>>> -a >>>>> >>>>> Just to make sure this is not lost: Who does the poking and prodding? >>>>> While I can do it, I have no idea where. :) >>>>> >>>>> Sean >>>>> -- >>>>> scf@FreeBSD.org >>>>> >>>>> >>>>>> On Wed, 4 Mar 2026 at 13:02, Sean C. Farley <scf@freebsd.org> wrote: >>>>>>> >>>>>>> On Sat, 28 Feb 2026, Sean C. Farley wrote: >>>>>>> >>>>>>>> I recently upgraded my home network using a few BrosTrend 5Gb cards >>>>>>>> (RTL8126) with 14-STABLE (1eb2beb3686c) and the latest commit >>>>>>>> (3170b35) to the GitHub repo for if_rge. It was either these cards or >>>>>>>> a couple of Intel X550-T2 cards, but less power and less heat were the >>>>>>>> deciding factors. >>>>>>>> >>>>>>>> For the most part, they seem to work very well, except that they >>>>>>>> cannot negotiate a 2500Base-T connection. They can negotiate >>>>>>>> 1000baseT and 5000Base-T. This was with two different cards >>>>>>>> connecting to two different devices. If manually set, they will try >>>>>>>> for awhile than fall-back to 1000baseT. >>>>>>>> >>>>>>>> One other oddity is that rge0 has "ClockPM disabled", but I have not >>>>>>>> looked into that. >>>>>>>> >>>>>>>> The networking hardware: >>>>>>>> - 3 BrosTrend 5Gb NIC's[1] (5Gb) >>>>>>>> - Zyxel NWA130BE[2] (2.5Gb port) >>>>>>>> - TRENDnet TEG-S591[3] (connected the NIC to its 10Gb port) >>>>>>>> - Nokia ONT XS-010X-Q[4] (10Gb port) >>> >>> *snip* >>> >>>>>>>> 1. https://www.brostrend.com/products/p2 >>>>>>>> 2. >>>>>>>> https://www.zyxel.com/us/en-us/products/wireless/be11000-6-stream-wifi-7-triple-radio-nebulaflex-access-point-nwa130be >>>>>>>> 3. >>>>>>>> https://www.trendnet.com/products/2_5g-10g-switches/9-port-multi-gig-switch-TEG-S591-v1 >>>>>>>> 4. https://www.nokia.com/asset/210704/