On Sunday, July 7, 2013 1:00:28 PM UTC-6, eskimobob wrote: > > I too have seen the problem both with Angstrom and Ubuntu (13.04). > RobertCNelson suggests: "I believe it's a problem in the cpsw ethernet > driver." (see above). Presumably both Angstrom and Ubuntu pull in the same > low level driver. > > It is not clear (to me at least) whether that means it is a problem which > the community can/should fix or whether it is something that TI need to be > made aware of so that they can fix it. > Can anyone clarify? >
CPSW refers to the ethernet subsystem in the SoC, which includes a VLAN-aware 3-port switch, two MACs, a timestamping engine for IEEE 1588, and a DMA-based host interface along with various glue. It's all documented in the TRM, so theoretically anyone could fix it, but it's a little less straightforward than your typical ethernet interface. Really nifty hardware though, if you can get it working right. I've written and debugged a full driver suite for CPSW in the context of another TI SoC and another OS, so I could probably track this problem down if I can get some time to familiarize myself with the Linux drivers. It's pretty easy to get it into a state like this where the host port DMA engine is wedged in one direction, as there's a little ownership handshaking dance you have to do with the DMA descriptors and the documentation on it is a little vague in some cases, and it will kill the engine if you get things in an inconsistent state. Luckily it stores the reason for wedging when this happens, so it *might* be relatively easy to diagnose. There are also various opportunities for wedging things in other places, though, so we'll have to see what's really happening. Anyway, I'll try to check it out in the next couple of days and I'll report if I make any progress. -- For more options, visit http://beagleboard.org/discuss --- You received this message because you are subscribed to the Google Groups "BeagleBoard" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/groups/opt_out.
