2010/11/4 Felix Fietkau <[email protected]>: > On 2010-11-04 9:59 PM, Björn Smedman wrote: >> Hi all, >> >> I'm trying to stabilize ath9k on an Atheros AR913x based platform. >> >> One type of problem is that sometimes, under high ath9k tx/rx load, >> the system locks up with lots of partial oopses like below spewing out >> on the serial. I'm not really qualified to read these but ath9k has a >> history of memory corruption and DMA problems. That in combination >> with the rater weird message, that __bzero and print_modules contain >> reserved instructions, makes me think that the problem may be that >> pages containing executable kernel code are accidentally DMA:ed over. >> :P Is that even possible? If so, can it be made impossible? :) >> >> Other than that, what could it be? >> >> Reserved instruction in kernel code[#1]: >> Cpu 0 [...] >> >> And so on until serial breaks down... Then all leds except two come on >> and it freezes in that position. Not even the hardware watchdog can >> get to it it seems. :( > If not even the watchdog can recover it, then it doesn't really sound > like a DMA issue. Last time I saw something like that on one of my > devices, the power supply was too weak. Are you sure that it's not > caused by a power or heat issue - or some other kind of hardware problem?
Could be heat I guess... But I've taken the cover off and I'm in Sweden, so doesn't feel very likely. :P It doesn't happen very often (maybe every week or so in normal operation as a home ap) but it happens right away in some ath9k related situations, e.g. if ath_alloc_rx_buf() fails in the RX DMA logic. That's under the assumption that "all leds except two come on and hardware watchdog is a kitten" is always this same problem though, which I guess it doesn't necessarily have to be. Just out of curiosity, is it possible to DMA over executable code? /Björn _______________________________________________ openwrt-devel mailing list [email protected] https://lists.openwrt.org/mailman/listinfo/openwrt-devel
