Forums » Software Development »
Ethernet packet loss
Added by Stéphane Peter almost 12 years ago
Hello,
We are using the OMAP-L138 with SysBios on both ARM and DSP. In a test where we are sending nonsense data from the ARM to a PC, we reach a bandwidth from over 4MB/s with 8K UDP packets at 300MHz. When we now increase the CPU frequency to 456MHz, we observe packet losses within the UDP frames:
No. Time Source Destination Protocol Length Info 299100 129.230709 10.128.30.66 10.128.20.74 IPv4 1514 Fragmented IP protocol (proto=UDP 0x11, off=0, ID=be9f) 299101 129.230830 10.128.30.66 10.128.20.74 IPv4 1514 Fragmented IP protocol (proto=UDP 0x11, off=1480, ID=be9f) 299102 129.230956 10.128.30.66 10.128.20.74 IPv4 1514 Fragmented IP protocol (proto=UDP 0x11, off=2960, ID=be9f) ----------------------------------------------------------------- missing ----------------------------------------------------------------- 299103 129.231204 10.128.30.66 10.128.20.74 IPv4 1514 Fragmented IP protocol (proto=UDP 0x11, off=5920, ID=be9f) 299104 129.231264 10.128.30.66 10.128.20.74 IPv4 622 Fragmented IP protocol (proto=UDP 0x11, off=7400, ID=be9f)
Can anybody setup a similar test with Linux to verify if this could be a HW issue (like http://e2e.ti.com/support/dsp/omap_applications_processors/f/42/t/212811.aspx)?
Thanks in advance
Stéphane
Replies (5)
RE: Ethernet packet loss - Added by Stéphane Peter almost 12 years ago
I also made a post in the E2E: http://e2e.ti.com/support/embedded/bios/f/355/t/244852.aspx
Stéphane
RE: Ethernet packet loss - Added by Michael Williamson almost 12 years ago
Hello,
I looked at your E2E post, and I think I may understand the problem, but I am not sure.
By changing the frequency in the bootloader and not in the kernel, I'm not sure if the kernel is initializing with the right idea of what the clock tree / configuration is. It may be assuming a 300 MHz configuration when setting up the clock tree for all the peripherals (specifically the EMAC controller).
Can you please, instead of altering the clock configuration u-Boot, change the frequency using the user controlled governor within the kernel and see if you have the same issue? This will automatically adjust the CORE voltage and ensures all the peripherals are advised of the clock change.
There are instructions for doing this here with our standard build.
If the EMAC performs better with this approach, then we know it's not a hardware issue and a problem in software. I would suggest compiling the kernel with PM_DEBUG enabled and DEBUGFS enable so that you can dump out the clock tree (find /mnt/debug/clock) and see how the kernel thinks the clocks are set up. There may be some initialization assumptions about the clock rates in the initialization code that requires attention with the u-Boot direct approach.
We have not observed these issues with 456 operation, but we always modify the speed using the kernel PM framework.
-Mike
RE: Ethernet packet loss - Added by Stéphane Peter almost 12 years ago
Hi Mike,
It’s not a u-Boot/Linux problem.
We are running SysBios on the ARM after leaving ROM code…
Best regards
Stéphane
| boot sequence.jpg (144 KB) boot sequence.jpg |
RE: Ethernet packet loss - Added by Michael Williamson almost 12 years ago
Are you using a custom board? Are you using the MII or the RMII port? If RMII, what device is sourcing the reference clock?
-Mike
RE: Ethernet packet loss - Added by Stéphane Peter almost 12 years ago
Hi Mike,
We are using your L138-FG-225-RC in combination with a propriatary base (=industrial IO) board --> MII is connected and used.
Stéphane
