From: john r. <jo...@cf...> - 2006-09-25 18:22:02
|
Hi All , I've had a little time today to continue looking at the ethernet driver for the smc91C11x chip. It does use DMA but is sets up the DMA controller, kicks off the transfer and then BUSY waits for the transfer to complete by calling cpu_relax. On arm this is simply an optimization barrier (nop) and the cpu doesn't reschedule. Does anyone have any opinions about whether I can do better than 1Meg/sec with a gumstix? If someone with more experience gives me some encouragement I'll try to optimize the network driver with real async DMA. John john roll wrote: > Dave, > > I've looked over the scm91x driver and it looks well written, uses PXA > DMA and shows no evidence of busy waiting. Specifically it uses an interrupt > when data cannot be send right away. > > I guess my real question is : Does anyone know if I should expect better > performance from the PXA255 or the smc91x controller? If better io is > expected I can keep digging. > > John > > > Dave Hylands wrote: > >>Hi John, >> > > ... > >>There are typically two large consumers of cpu cycles: >> >>1 - Copying memory around. This can often be replaced by using DMA. I >>don't know if the existing driver uses DMA or not. >> >>2 - Busy waiting for stuff. This typically requires a redesign. >> >>< (Can you also tell me how to push on a string?) >> >>You push on a string by soaking it in a cyanoacrylate and letting it dry :) >> > > > > ------------------------------------------------------------------------- > Using Tomcat but need to do more? Need to support web services, security? > Get stuff done quickly with pre-integrated technology to make your job easier > Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo > http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 > _______________________________________________ > gumstix-users mailing list > gum...@li... > https://lists.sourceforge.net/lists/listinfo/gumstix-users |