Poor Ethernet performance with AM1705 and LAN8710a, RMII connection

Dan Edmondson

Other Parts Discussed in Thread: AM1705, AM1707, OMAPL138

Hi all,

We have a custom board based on the evmam1707, but using the am1705. I am using the mach-davinci/board-da830-evm.c with modifications as the board file. Same with the GEL. We are using PSP-SDK-03.22.00.06.

The Lan8710a PHY is strapped in 100FD, autoneg enabled and is found by the mdio probe in U-boot and the Kernel, and the phy does show 100FD, autoneg complete. We are using the internal 50MHz clock for RMII.

The smsc driver in linux is not being used at this time. Mostly because I don't know what needs to be in the board file for this device. The datasheet that has the performance numbers below mentions the da850, and a chart shows that board using the LAN 8710. Perhaps I can follow the code in board-da850-evm.c?

Here is dmesg of mdio probe:

Fixed MDIO Bus: probed
davinci_mdio davinci_mdio.0: davinci mdio revision 1.5
davinci_mdio davinci_mdio.0: detected phy mask ffffff7f
davinci_mdio.0: probed
davinci_mdio davinci_mdio.0: phy[7]: device davinci_mdio-0:07, driver unknown

Here is dmesg output when Ethernet is started:

[ ok ] Starting enhanced syslogd: rsyslogd.
[ ok ] Starting periodic command scheduler: cron.
net eth0: no phy, defaulting to 100/full
Starting DAS NTPDATE init...

The Ethernet connection is working, we can ping, run RPC, rsyslog, etc. However, the performance is not there at all.

I found these benchmarks in the SDK-03.20.00.08 datasheet for the DA850.

DA830/OMAP-L137

**Ethernet 100Mbps Mode Performance**
TCP Window Size(in KBytes)	Bandwidth (in Mbits/sec)	Transfer size (in MBytes)	Interval (in Seconds)
16	55.7	398	60
32	63.4	454	60
64	65	465	60
128	65	464	60

The performance numbers were captured using the iperf tool. Usage details are mentioned below:

Server side command switch : "-s"
Client side command : "-c <server ip> -w <window size> -d -t60"
Iperf tool is run on the DUT1 in server mode and on DUT2 in client mode. Version 1.7.0 is used on both sides.
Data captured here is for "iperf" in client mode.
Cross cable is used to measure performance.
Speed is set to 100Mbps
Power Management disabled for measurement

Using the same client options, these are the numbers I'm seeing:

We also see many, many TX overruns on the connection:

root@spm1:/etc/init.d# ifconfig
eth0      Link encap:Ethernet HWaddr 70:b3:d5:3c:21:33
          inet addr:192.168.101.1 Bcast:192.168.101.255 Mask:255.255.255.0
          UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
          RX packets:145857 errors:0 dropped:0 overruns:0 frame:0
          TX packets:141531 errors:0 dropped:0 overruns:5954 carrier:0
          collisions:0 txqueuelen:1000
          RX bytes:132752375 (126.6 MiB) TX bytes:128138572 (122.2 MiB)
          Interrupt:33

The receiving connection shows that it dropped 7, Rx packets.

Please advise as to where to focus. Do I need to get that SMSC driver running?

Thanks,

Dan

over 10 years ago

0 Titusrathinaraj Stalin over 10 years ago

TI__Guru** 116100 points

Hi Dan,

You have to enabe SMSC PHY driver for EMAC support.
You may not modify anything in board file regarding this, except pinmux, EMAC & MDIO init etc.,

The log should be like this,

[ 1.589454] davinci_emac davinci_emac.1: using random MAC addr: d6:36:33:99:43:ab
[ 1.612820] davinci_mdio davinci_mdio.0: resetting idled controller
[ 1.620142] net eth0: attached PHY driver [SMSC LAN8710/LAN8720] (mii_bus:phy_addr=davinci_mdio-0:07, id=7c0f1)
[ 4.613594] PHY: davinci_mdio-0:07 - Link is Up - 100/Full
[ 4.643391] Sending DHCP requests ., OK
[ 5.163834] IP-Config: Got DHCP answer from 0.0.0.0, my address is 10.100.1.10
[ 5.173663] IP-Config: Complete:

0 Dan Edmondson over 10 years ago in reply to Titusrathinaraj Stalin

Intellectual 860 points

Hi Titus,

We did have CONFIG_NET_VENDOR_SMSC, but not CONFIG_SMSC_PHY enabled. Note that these are listed as MII in the .config file. We are running RMII, if that makes any difference in the config.

In dmesg, mdio probe finds the phy and knows what driver to use:

When the connection is brought up, it doesn't seem find the phy:

[ ok ] Starting enhanced syslogd: rsyslogd.
[ ok ] Starting periodic command scheduler: cron.
net eth0: no phy, defaulting to 100/full
Starting DAS NTPDATE init...

The Ethernet connection is working, but still very slow.

Using the old mii-tool it seems the kernel still doesn't know there is a phy. Again, we are running rmii if that makes a difference here.

root@spm1:~# mii-tool eth0
SIOCGMIIPHY on 'eth0' failed: No such device

thanks,

Dan

0 Titusrathinaraj Stalin over 10 years ago in reply to Dan Edmondson

TI__Guru** 116100 points

Hi Dan,

Sorry for the delayed responses on this.

Do you have any option to configure the RMII with external clock ?
Actually, In DA830 SDI EVM board (OMAPL137 spectrum digital) board uses RMII with external clock mode.

We have a custom board based on the evmam1707, but using the am1705. I am using the mach-davinci/board-da830-evm.c with modifications as the board file. Same with the GEL. We are using PSP-SDK-03.22.00.06.

Have you checked the AM1707 EVM for RMII clock ie internal or external ?

Did you probe the EMAC clock ?

We've seen the same similar problem on AM335x.

e2e.ti.com/.../292923

0 Dan Edmondson over 10 years ago in reply to Titusrathinaraj Stalin

Intellectual 860 points

Hi Titus,

I have also been kept from getting back to this. Today I have been able to track down some information on the RMII support on the OMAP L137 AM1705. Since we are basing our BSP on the da830 from SDK 03.22.00.06 for a L137 I assume this note in the Ethernet section is applicable to this situation.

Performance and Benchmarks

**Ethernet 100Mbps Mode Performance**
TCP Window Size(in KBytes)	Interval (in Seconds)	Transfer Rate MII PHY (in Mbps)						Transfer Rate RMII PHY (in Mbps)
		DA850 Spectrum Digital EVM		DA830 Spectrum Digital EVM		DA830-WiFi Spectrum Digital EVM		DA850 Spectrum Digital EVM		DA830 Spectrum Digital EVM		DA830-WiFi Spectrum Digital EVM
		300 MHz	456 MHz	300 MHz	456 MHz	300 MHz	456 MHz	300 MHz	456 MHz	300 MHz	456 MHz	300 MHz	456 MHz
16	60	41.3	47.4	39.2	45.4	40.1	44.2	43.4	51.7	NA	NA	NA	NA
32	60	46.9	52.3	44.8	51.2	43.4	51.4	48.3	55.4	NA	NA	NA	NA
64	60	44.2	54.7	41.2	52.7	40.2	51.7	48.1	55.8	NA	NA	NA	NA
128	60	45.4	53.6	44.3	51.2	44.2	50.2	48.9	55.9	NA	NA	NA	NA

NOTE
RMII PHY support is available only on DA850/OMAPL138/AM18x SDI.
CPU load during the performance test is 90%

Titus, is that the trick for supporting RMII on this processor? An external clock? If so, I can pass that along to the HW guys while they also investigate changing layout for MII traces. I know they checked clock last Friday and said what they could see of the 50 MHz clocking looked good. But if this is an internal issue may be they can't see the problem. We are using the internal 50 MHz RMII clock.

Thanks,

Dan

0 Titusrathinaraj Stalin over 10 years ago in reply to Dan Edmondson

TI__Guru** 116100 points

You can check out the AM1705 TRM (spruh93b)

Refer to the AM1705 TRM chapter 6.3.4 (Page no 104 )

The SYSCLK7 output clock does not meet the RMII reference clock specification of 50 MHz +/-50 ppm.

Please refer to the following thread too.

e2e.ti.com/.../395955

0 Dan Edmondson over 10 years ago in reply to Titusrathinaraj Stalin

Intellectual 860 points

Excellent, Titus! I will pass this on to the hw group.

Also, I will check on getting the SMSC driver loaded properly.

Dan

0 Dan Edmondson over 10 years ago in reply to Dan Edmondson

Intellectual 860 points

It turns out the hw already has an external 50 MHz clock so I just need to try it.

I have built a new UBL ROM image and changed the pin mux to exclude the RMII clock, as can bee seen in reg 9.

I have also re-run menuconfig for the kernel, selected SMSC vendor and LAN8710 under the MII drivers (it is the only one I could find, again we use RMII).

The results are exactly the same as from my first post.

and later:

[ ok ] Starting periodic command scheduler: cron.
net eth0: no phy, defaulting to 100/full
Starting DAS NTPDATE init...

The best performance numbers I get with iperf are 10 - 12 Mbps

0 Titusrathinaraj Stalin over 10 years ago in reply to Dan Edmondson

TI__Guru** 116100 points

Have you tested this RMII behavior in AM1707 EVM (P9 port) ?

Have you selected the RMII mode in CFGCHIP3 register (8th bit ) ?

Could you share your board file or send to me (x0213399@ti.com)

0 Dan Edmondson over 10 years ago in reply to Titusrathinaraj Stalin

Intellectual 860 points

Hi Titus,

I used the AM1707 to figure out things for our custom boards and the evm was left in a bad state software-wise. It could be resurrected but would take time I don't have at the moment. One of the hardware guys is going over the schematics for that board to compare our connections.

I am confused by your comments on the CFGCHIP3 register. Those fields are marked reserved in the 1705 ref guide: sprugu3, March 2010

Table 10-47. Chip Configuration 3 Register (CFGCHIP3) Field Descriptions
Bit Field Value Description
31-16 Reserved 0 Reserved
15-8 Reserved FFh Reserved. Write the default value when modifying this register.
7-3 Reserved 0 Reserved. Write the default value to all bits when modifying this register

I will send you the current board file today.

0 Titusrathinaraj Stalin over 10 years ago in reply to Dan Edmondson

TI__Guru** 116100 points

Apologize, you are right. Actually I've looked into OMAPL138 which has both RGMII and MII.

Processors

Processors forum

Poor Ethernet performance with AM1705 and LAN8710a, RMII connection

DA830/OMAP-L137

Performance and Benchmarks