On Fri, 2009-05-01 at 11:47 -0700, David Brownell wrote: > On Friday 01 May 2009, Kevin Hilman wrote: > > In debugging this, I've noticed that only using the banked IRQ along > > with some strategically placed printk's also gets things working > > without the direct IRQ. This suggests there's a timing problem or > > race someplace that needs flushing out. > > Right, that's much the same as noting that LOCKDEP alone > makes it behave fine (for me). > > There's also that patch (from Sneha?) that changed the dm9000 > driver TX path a bit.
In my spare time, I'm looking to use DMA for dm9000 to off load the dm355. So far, I have the transmit DMA working under light traffic load. When I run performance bench mark, the dm9000 stops working. I'm using the MV 2.6.18 kernel, so I'm not sure if this is related to what Kevin is seeing. > > I looked at the AEMIF timings, and they didn't look wrong. > One phase seemed maybe longer than necessary, but not wrong. > > Another experiment: does setting the dm9000 to trigger on > both edges (vs just one) help? The code was originally written this way. During DM355 chip bring up, it was discovered that the dm9000 can only handle rising edge. Having trigger from both edges made the performance much worse. Regards, Steve > > - Dave > > > _______________________________________________ > Davinci-linux-open-source mailing list > Davinci-linux-open-source@linux.davincidsp.com > http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source _______________________________________________ Davinci-linux-open-source mailing list Davinci-linux-open-source@linux.davincidsp.com http://linux.davincidsp.com/mailman/listinfo/davinci-linux-open-source
