On Mon, Jan 10, 2011 at 22:41, Arnd Bergmann wrote: > On Tuesday 11 January 2011, Mike Frysinger wrote: >> throughput seems to be a bit lower. Âthe Blackfin JTAG hardware has >> basically a 4 byte fifo that is filled/consumed in a single shot. Âand >> the Blackfin can produce much faster than the host can consume. Âso >> atm, i have it send out 4 bytes, and the hvc layers take care of >> calling back into me at some point. Âif i add a busy loop to run a few >> hundred milliseconds (like HZ/4), it runs much nicer. Âbut obviously >> doesnt give anyone else time to run. Âi cant schedule or anything as >> the write layers are called with a spin_lock_irqsave. Âany tips for >> how to speed this up a bit ? Âor is it a wash with hvc ? > > You could try lowering the hvc MIN_TIMEOUT value. If you want to > change it for blackfin, you could probably make it a configuration > option. However, if it really takes a few milliseconds that you need > to wait between two accesses, MIN_TIMEOUT should already be small > enough. > > You should really not need to sleep in your output function, as > the khvcd() timeout logic tries handling this in the best way. > There may be a bug in that logic though (wouldn't be the first one > there), so try to see what the timeouts are when you get into > this problem. np, i'll poke around a bit. i'm ok with merging this driver now since it seems to be functionality equivalent to the old code. -mike -- To unsubscribe from this list: send the line "unsubscribe linux-serial" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
