Hope this helps:
The RB450G has 5 ethernet ports, numbered from left to right.
Port 1 is identified as eth0 and member of br-lan.
Ping to the subnet configured on eth1 works on all ports from 2 to 5.

This seems a bit reversed from the rs pro.
I've tested with both ping and tcpdump so I confirm the port numbering.

Thank you.

First of all, thank you very much for testing the patch on the rb 450g.

Could you try this patch? It is almost the same as the previous, except it prints port status changes into the kernel log (using the switch internal port numbers).

This should tell you which external port label corresponds to which internal port.

Edit: Btw, it would be really nice if you could test a "cold boot" from flash, since there were reports that no ethernet works when directly booting from flash.

KM

(Last edited by KanjiMonster on 12 Mar 2010, 00:21)

Here is the result for my RB450G:
==>Eth1
ar8316: port 5 up (1000Mbps/Full duplex)
eth0: link up (1000Mbps/Full duplex)
br-lan: port 1(eth0) entering forwarding state
ar8316: port 5 down

==>Eth2
eth0: link down
br-lan: port 1(eth0) entering disabled state
ar8316: port 1 up (1000Mbps/Full duplex)
ar8316: port 1 down

==>Eth3
ar8316: port 4 up (1000Mbps/Full duplex)
ar8316: port 4 down

==>Eth4
ar8316: port 3 up (1000Mbps/Full duplex)
ar8316: port 3 down

==>Eth5
ar8316: port 2 up (1000Mbps/Full duplex)
ar8316: port 2 down

Same for me.
I'm browsing the other threads to understand how to flash the image and boot it so after the weekend I should have an answer regarding the cold boot situation.

Thank you

---Later that day
when flash booted:
ar8316: port 5 up (100Mbps/Full duplex)
ar8316: port 5 down
ar8316: port 1 up (100Mbps/Full duplex)
ar8316: port 1 down
ar8316: port 4 up (100Mbps/Full duplex)
ar8316: port 4 down
ar8316: port 3 up (100Mbps/Full duplex)
ar8316: port 3 down
ar8316: port 2 up (100Mbps/Full duplex)
ar8316: port 2 down
ar8316: port 2 down

ping works from all ports

---

Thank you KM for the switch and rabinnh for the partition size patch.

(Last edited by flaxrt on 12 Mar 2010, 19:48)

Thank you for taking the time for testing.

Could you paste me the following lines from cold boot:

ar8316: updating reg 0x00008 from 0x81461bea to 0x81461bea
ar8316: updating reg 0x000b0 from 0xcc37cc37 to 0xcc37cc37
ar8316: updating reg 0x000b4 from 0xca37ca37 to 0xcc37cc37
ar8316: updating reg 0x000b8 from 0xca37ca37 to 0xca37ca37
ar8316: updating reg 0x000bc from 0x00000000 to 0x00000000
ar8316: updating phy4 dbg reg 0x12 from 0x4c0c to 0x480c
ar8316: updating phy4 dbg reg 0x00 from 0x824e to 0x824e
ar8316: updating phy4 dbg reg 0x05 from 0x3d47 to 0x3d47

or say if there were the same (I'm interested in the original values).

---

Minor Update

I updated the second patch that adds the switch configuration.

Changes:
* RB-450G: Now the wan is again eth0 and eth1 is the lan again. I switched this previously because I didn't see that its already switched in the initialization routine of the routerboard.

KM

(Last edited by KanjiMonster on 12 Mar 2010, 20:07)

KanjiMonster, pleace read private message. I can give You remote access to PC connected to rb450g by console and two eth.

ar8316: updating reg 0x00008 from 0x01061b60 to 0x81461bea
ar8316: updating reg 0x000b0 from 0xcc00cc00 to 0xcc37cc37
ar8316: updating reg 0x000b4 from 0xca35ca35 to 0xcc37cc37
ar8316: updating reg 0x000b8 from 0xcf35cf35 to 0xca37ca37
ar8316: updating reg 0x000bc from 0x00000000 to 0x00000000
ar8316: updating phy4 dbg reg 0x12 from 0x4c0c to 0x480c
ar8316: updating phy4 dbg reg 0x00 from 0x82ee to 0x824e
ar8316: updating phy4 dbg reg 0x05 from 0x3c46 to 0x3d47

Great. Using this information, here are updated patches:

March 13 Version
Changelog:
- The ar8316 driver
  * The port status update is part of the regular version for now
  * Don't modify the LED registers, these seem to be always setup by the rb-450g
  * Don't do a hw init on the rs pro, this should speed up the boot there
  * Made the multicast fix ar8316 specific for now
- Network/profile configuration updates
  * Include the driver also in 2.6.33 and 2.6.34

KM

eth0 is now wan , eth1 is in the lan bridge.
The port order is the same ( 5,1,4,3,2 ).

Thank you.

PS: What do you usually use to test throughput ? I've used iperf but not much option there for parallel connections and simulating multiple nat-ed ip's ...

(Last edited by flaxrt on 13 Mar 2010, 13:51)

Thanks for the feedback. Unfortunately I can't think of an easy way to "fix" the port numbering.

The options I can think of are:
* Making the switch a platform device, adding port mappings (external number -> internal number), and supplying these as platform_data. Requires quite a bit of changes in the driver
* Modifying switch.sh to allow an e.g. a switch.int_port_numbers = '0 5 1 4 3 2' to do the translation before sending them to swconfig
* Creating a DSA driver (which currently has several drawback, like no wirespeed switching support), which allows the ports to appear as their own ethernet interfaces.

I currently would prefer option 2, which should be the easiest to implement (and can be developed directly on the device).

I only used iperf until now. Using windows as my main OS really does limit my options there ;-).

KM

(Last edited by KanjiMonster on 13 Mar 2010, 14:34)

Great work, just was I was waiting for...

I'll give a run on my RouterStation Pro this weekend and see what I can break

Thanks for all your work.

Regarding the switch "issue" I still have to read the docs as I am new to openwrt so I will not bug you for a while.

As for performance I'll wait until I can test it on a server with Ge adapters to see how it goes
In 100 MBit it tested for 92.6Mb with Masquerade and 46% cpu load ( 0.10 system load ), using iperf with -P 200 -t 5M

Sounds about right; I got ~200 Mbit with iperf and using an gbit connection between the rs pro and my pc.

KM

Sunday afternoon edition

Just a minor update for the configuration patch:

* The switch section is now named with the interface name; having no name seemed to have confused luci (it didn't bother swconfig). That means that configuring the switch with luci should work in theory (didn't test this).

KM

Edit:
No need to rebuild for this to work on device, just change in /etc/config/network

config switch
        option name     eth1
        option reset    1
        option enable_vlan 1

to

config switch eth1
        option name     eth1
        option reset    1
        option enable_vlan 1

(Last edited by KanjiMonster on 14 Mar 2010, 16:15)

I praise you KanjiMonster. I hope I could repay you at least by inviting you for some beer sometime
Finally it made my RouterStation PRO worth something..
Bit sad that guys from community must develop these drivers, that should have been done by Atheros or Ubiquiti..

I've tested it lot this weekend and it works nicely.
I've tested tagged VLAN setup with your switch patch (each port in different VLAN), eth0 VLANs and bridging between VLANs.
Only strange thing noticed so far is capturing packets with tcpdump. It captures few packets than pauses and this repeats.
On wifi interface it works fine. But well it may be unrelated to your patch and is related to Atheros network driver.
Will check on vanilla OpenWRT trunk later.

Thanks again,
Vlado

I did some tests about routing performance. rb450g with openwrt can route about 23 kpps RX counter and 23 kpps TX (four "ping -f" in my PCs through rb450g). In Cisco kind of measurement it is 46 kpps total performance or so.

root@rb450g:~# lsmod
Module                  Size  Used by    Not tainted
sch_htb                12384  2
sch_sfq                 4416  2
cls_route               5040  4
act_police              3328  2
iptable_filter           768  0
ip_tables               8464  1 iptable_filter
x_tables                9312  1 ip_tables
leds_gpio               1456  0
button_hotplug          2576  0
gpio_buttons            1968  0
input_polldev           1360  1 gpio_buttons
input_core             16896  4 button_hotplug,gpio_buttons,input_polldev
root@rb450g:~#

I unloaded all conntrack and nat modules. "top" was
CPU:   0% usr   0% sys   0% nic  13% idle   0% io   0% irq  86% sirq

http://mikc.ru/files/docs/tests2010.pdf -- here is no exactly rb450 but rb400 and rb750G. So they at 3 times more (70-90kpps). They use Agilent N2X , I don't know how it works but they results in 2-3 times faster.

I'll try to use linux packet kernel generator but need some free time for this tests ...

My strange iperf results are:

1. [  3]  0.0-60.0 sec  2.37 GBytes    339 Mbits/sec <- client is on eth0 (Eth1/PoE)
CPU was about 75% sirq

2. [  3]  0.0-120.0 sec  8.62 GBytes    617 Mbits/sec <- client in on eth1.10 (Eth2)
CPU was about 95% sirq

Why such difference ? :-).

I have questions about vlan setup:

1. At this moment we can use 1-32 vlan tags and no other numbers, isn't it ?

2. The default generated config for vlan contains all ports, is it right ?
I think it shoud to not contain port 5 (eth0)

config switch_vlan
        option device   eth1
        option vlan     1
        option ports    "0 1 2 3 4 5" <- 5 is eth0

Also I tried this test vlan 10

config switch_vlan
        option device   eth1
        option vlan     10
        option ports    "0t 1t 2t 3t 4t 5t"

and eth0 does not see tag 10 packets by tcpdump, i.e. works right.

So there is 6 ports (0 1 2 3 4 5) , one of them internal (to eth1) , one is eth0. Configuration of some vlan should have five not six port numbers (0 1 2 3 4). Where am I misstaken ?

Has anyone right /etc/config/network example where ich port is separate l3 interface ?

(Last edited by warm on 15 Mar 2010, 10:22)

Thanks for your testing. Do you see the same performance on the single eth0 port?

It might have to do that I guess routeros is a specialized os, and they have access to data sheets, so they know hot to optimize their code.

Yeah, but I'm hoping to fix that in the next days. For now, you can also just edit the ar8216.c and change the #define at the top ('though you will get problems if you make it bigger than ~650).

As youself have noticed, the fifth port is separated; at least in default setup. But it looks like it can be made part of the switch again as the RouterOS offers this. Therefore I left it in there, also because it doesn't seem to cause harm.
Perhaps I'll be able to figure out one day how its doing that (anybody with JTAG and a RB450G caring to read out some switch registers while RouterOS is loaded? ;-).

The following should work:

config interface loopback
        option ifname   lo
        option proto    static
        option ipaddr   127.0.0.1
        option netmask  255.0.0.0

config interface port1
        option ifname   "eth1.1"
        option proto    static
        option ipaddr   192.168.1.1
        option netmask  255.255.255.0

config interface port2
        option ifname   "eth1.2"
        option proto    static
        option ipaddr   192.168.2.1
        option netmask  255.255.255.0

config interface port3
        option ifname   "eth1.3"
        option proto    static
        option ipaddr   192.168.3.1
        option netmask  255.255.255.0

config interface port4
        option ifname   "eth1.4"
        option proto    static
        option ipaddr   192.168.4.1
        option netmask  255.255.255.0

config interface wan
        option ifname   eth0
        option proto    dhcp

config switch eth1
        option name     eth1
        option reset    1
        option enable_vlan 1

config switch_vlan
        option device   eth1
        option vlan     1
        option ports    "0t 1"

config switch_vlan
        option device   eth1
        option vlan     2
        option ports    "0t 2"

config switch_vlan
        option device   eth1
        option vlan     3
        option ports    "0t 3"

config switch_vlan
        option device   eth1
        option vlan     4
        option ports    "0t 4"

KM

P.S: Crititque (especially if I'm writing stupid code ;) is always welcome :)

(Last edited by KanjiMonster on 15 Mar 2010, 11:45)

What do You mean on "single eth0 port" ? To create eth0.2 and eth0.3 and to do tests from eth0.2 to eth0.3 and vice versa ?

I mean the wan port ("eth1"?, the switch port 5). It would be interesting to know whether there is a difference in packets/s between the single wan port and going through the switch.

Finally figured out how to use the whole VID range

Using the example for the separate interfaces for the ports:

config interface loopback
        option ifname   lo
        option proto    static
        option ipaddr   127.0.0.1
        option netmask  255.0.0.0

config interface port1
        option ifname   "eth1.100"
        option proto    static
        option ipaddr   192.168.1.1
        option netmask  255.255.255.0

config interface port2
        option ifname   "eth1.200"
        option proto    static
        option ipaddr   192.168.2.1
        option netmask  255.255.255.0

config interface port3
        option ifname   "eth1.300"
        option proto    static
        option ipaddr   192.168.3.1
        option netmask  255.255.255.0

config interface port4
        option ifname   "eth1.400"
        option proto    static
        option ipaddr   192.168.4.1
        option netmask  255.255.255.0

config interface wan
        option ifname   eth0
        option proto    dhcp

config switch eth1
        option name     eth1
        option reset    1
        option enable_vlan 1

config switch_vlan
        option device   eth1
        option vlan     1
        option pvid     100
        option ports    "0t 1"

config switch_vlan
        option device   eth1
        option vlan     2
        option pvid     200
        option ports    "0t 2"

config switch_vlan
        option device   eth1
        option vlan     3
        option pvid     300
        option ports    "0t 3"

config switch_vlan
        option device   eth1
        option vlan     4
        option pvid     400
        option ports    "0t 4"

This was too obvious.

So now the MAX_VLANS really *is* the maximum number of entries on the vlan table, not the highest VID.

KM

exactly this config does not work on my rb450g but it works without one zerro on pvid (10-40):-). Interface with wlan 40 works but with 100,200...400 no.

I find that I have different PCs on my tests so that is why results are different by direction. Now I always used one machine like iperf client (send traffic) and other like receiver. Resalts was the same regardless direction. I.e. eth1.10->eth1.40 = eth1.40->eth1.10 = 500Mbit.

So I have bad hardware to test gigabit :-).

(Last edited by warm on 15 Mar 2010, 13:52)

Mh, strange, its working fine on my routerstation (with one interface less of course).

I noticed that it doesn't always clean up correctly (sometimes there are still some bridges to remove with brctl, or there are still vlans configured) after changing /etc/config/network, and network stop && network start isn't the same as network restart (the switch gets only reconfigured when restarting).

KM

Yes, now it works *after reboot* but I rebooted router in a first time too ... may be I did mistake somewhere in configs when tried first time. Also I noticed that output of "swconfig dev eth1 show" does not change after network stop and network start, it changes only after reboot.

All of this not so important, the main point is that router works :-). Is it time to include Your rare patches in trunk ?

You were probably bitten by the problem that /etc/init.d/network start doesn't call setup_switch. This was fixed by nbd today.

My plan is to do a sort-of release candidate (with cleaned up source) tomorrow here, for final testing. I'll be away till sunday (leipzig book fair), and if I come back and there are no further complains (or only minor ;-), I'll post it to openwrt-devel (in hope for some ar8216 testing). Then it (hopefully) shouldn't take long until the patches gets applied to trunk.

KM

Kanji,

Thanks for all the hard work.

Which version of the kernel do I apply your patch to?  I had a build tree with 2.6.30 and some of them didn't get applied.

Thanks,
rabinnh

BTW, when they do go in the trunk, maybe we could actually have a Mikrotik/RB450G target ;-)

No kernel version specific except the config-* modifications, which only add the line

CONFIG_AR8216_PHY=y

and thats all.

Everything else shouldn't depend on the kernel version.

KM