You also run 1G over 10g phy? Something in dmesg except the link-down/up? Do you have direct connection or with ppe involved?
I just tried connecting it up to my normal 2.5ge lan as a subtended router for a test. Connected the combo wan 2.5/10 rj45 port to my existing lan on a 2.5ge switch and a single laptop to the port 1. Same essential result, no kernel messages or errors. Though there are plenty of weird things in the boot log. Attaching boot log here for you and a sample speedtest attempt that fails. Maybe I was sold a bad board? Not sure what I can do, also I don’t understand what you are referring to “ppe”. Forgive me, I’m quite new to a lot of this.
dmesg_log.txt (216.1 KB) speedtest_log.txt (1.9 KB)
I’ve updated my thread referenced above - in my case routing is not involved as traffic originates / terminates in BPI, iperf3 is run on it. maybe you could try similar test to rule out routing
This is already the second case, and I’m starting to worry that I ordered it right at the start of sales without waiting for user feedback 
I think you are right.
- But maybe you’re lucky and it only needs a mature software/firmware stack.
Aside from that:
- There are current comments about the BE14 (Just to get a feel for the environment you’re in):
Thanks, friend. I’ve seen those issues, so I gave up on WiFi boards. I have routers in access point mode, and I’ll use those, even though it’s a bit inconvenient that the WiFi setup won’t be in one web interface.
That was exactly my plan, just want a powerful router. But unable to work out of the box in the most basic routing job possible is a deal breaker. I should have waited longer, I expect this will be solved via software in 2 to 6 months.
1 Like
@shadowmite as I’ve discovered, in my case
ethtool -K eth1 tso off
works around the problem. (obviously this is not a real solution, CPU usage+ interrupt storm)
very interested in your results 
Hi everyone. I’ve run into exactly the same issue.
Connected both via ONU stick and current router (LAN→WAN scheme). Speed is heavily throttled — upload only, limited to 100 Mbps instead of 1 Gbps.
А не может это быть случайно настройка адаптера самого устройства с которого ты проверяешь выход !? Имею в виду если это ноутбук или комп, то есть физическая настройка адаптера ?
Edit (moderated): Could it be a setting in the adapter of the device you’re testing the output from? I mean, if it’s a laptop or computer, is the adapter physically configured?
Привет, не, там проблема в драйвере скорее всего, автонег криво работает, я уже разобрался как победить, но как решить проблему пока не знаю. Но отписался в этой теме
Edit (moderated):
Hey, no, it’s most likely a driver issue. Autoneg isn’t working properly. I’ve already figured out how to fix it, but I don’t know how to solve it yet. I’ve posted in this thread, though.
Здравствуй. Понимаю о чём речь (косвенно), просто если работает на один поток 2 канала, то есть вход выход то скорость режется на 2, то есть если поток 1000, то должно получится = 900+100 или 500+500. Ведь в документации не описывается эта как “тонкость” потому что это всем известно. так вот я про что ; если так, то причина может быть именно в этом, что то блочит (ограничивает) отдачу потому что на вход смелые 900. А ты не пытался на сайте openWrt задавать вопрос ? Может это просто в прошивке. По простецки в винде это решается настройками самого адаптера, где есть такая возможность тыками решать проблему 100 или 1000, вот о чём я…
Edit (moderated):
Hello. I understand what you’re talking about (indirectly). It’s just that if two channels are working on a single stream, meaning input and output, the speed is halved. So, if the stream is 1000, it should be 900+100 or 500+500. The documentation doesn’t describe this as a “detail” because everyone knows it. So, that’s what I’m getting at. If so, that could be the problem: something is blocking (limiting) the output because the input is a whopping 900. Have you tried asking the question on the OpenWrt website? It might just be a firmware issue. In Windows, this can be easily resolved by adjusting the adapter settings, which allow you to tweak the settings to change between 100 and 1000, that’s what I’m getting at…
Мне кажется, что Вы не читали мои сообщение в ветке по ссылке
Ethernet работает в full-duplex нет коллизий и классического разделения одной общей среды между направлениями, а speedtest измеряет download и upload последовательно, а не одновременно.
Если отключить auto-neg и зафиксировать режим, соответствующий тому, что реально видно в Link partner advertised link modes, проблема исчезает. При auto-neg линк поднимается как 2.5G, но advertised modes остаются до 1G, из-за этого ломается TX. Это баг драйвера, а не какие-то два канала.
Edit (moderated):
I don’t think you read my post in the linked thread
Ethernet operates in full-duplex mode, there are no collisions or typical sharing of a single shared medium between directions, and speedtest measures download and upload sequentially, not simultaneously.
If you disable auto-neg and lock the mode to match what’s actually shown in Link Partner advertised link modes, the problem disappears. With auto-neg, the link rises to 2.5G, but advertised modes remain at 1G, which breaks the TX. This is a driver bug, not a specific issue with the two channels.
Please write in english only
2 Likes
i believe this is the same issue i was talking about in the other thread. when the aeonsemi phy is attached to a non 10gbit port the upload speeds are very bad.
start reading from here : [BPI-R4-PRO] Mainline OpenWRT - #16 by pameruoso
the issue is not there on the official bpi sinovoip image of the 29th december afaik.
Good to know,but had not find time yet to find difference to the new driver…they added a nearly complete new driver in sdk which has many additional files (mostly for debugging it seems). It was also mentioned in commit message that there are some adjustments regarding speed issues.
Afair it was this commit/patch:
or maybe this.
"Added support for 1000BASE-T and 2.5GBASE-T for 10G RJ45 WAN/LAN”. which still we don’t know what it corresponds to.
Hi. I tested it on this exact firmware, and the issue reproduces. I shared a link to a related thread a bit earlier and described the flow of how it happens for me there. Hopefully, they’ll fix it.
Where is this comment? In sdk or the image? But @abs1ck tells that issue is not fixed…maybe there are 2 different (auto neg and throughput) issues?
now that I recall i have 1 gbit in upload but i never tested the 10g aeonsemi with a 1gbit partner… so if abs1ck says it shows with 1gbit he could be right because i tested that only with 2,5gbit interfaces.