Am 13.10.2010 10:50, schrieb Wolfgang Grandegger: > On 10/13/2010 10:20 AM, Michal Sojka wrote: >> On Tue, 12 Oct 2010, Wolfgang Grandegger wrote: >>> Yes, packet losses can also happen, when the message buffers are not >>> read out fast enough. But then you should see overrun errors, which you >>> should see with "candump any,0:0,#FFFFFFFF". See also: >>> >>> http://lxr.linux.no/#linux+v2.6.35.7/drivers/net/can/mscan/mscan.c#L353 >> >> Right. > > It's still unclear to me what errors you get? RX overrun errors in > hardware or software? candump is able to detect and report both. > >>> Also "# ip -d -s link show can0" should list them. >>> >>>> We were able to avoid losing messages by reading them in hard-irq >>>> context (with lincan driver), but I'm not sure whether this will help if >>>> you use CF (we do not use it). >>> >>> Well, if the CF card driver is causing huge latencies, it should be >>> fixed. It should not be too difficult to spot the latency evil with ftrace. >> >> I've heard that one company that experienced these problems tried to >> solve it with Thomas Gleixner and he said that it is impossible to fix. >> But that's only what I've heard, I do not know any details. > > Hey, it's software! Everything is possible ;-). Feel free to add Thomas > to CC to get more information.
Well, that's the point. According to the errata and some information we found deep in the internet, there is a hardware bug in the FIFO of the ATA-controller. The ATA-FIFO has to be polled to avoid data loss and this is the reason for the high latencies. (source: Freescale MPC5200B Errata) Greetings, René _______________________________________________ Socketcan-users mailing list [email protected] https://lists.berlios.de/mailman/listinfo/socketcan-users
