Re: bus_dmamap_sync() for USB

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Tue, Jun 24, 2008 at 03:50:25PM -0400, Michael Lorenz wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Hello,
> 
> On Jun 24, 2008, at 14:34, Manuel Bouyer wrote:
> 
> >The fwohci0 is the same as your, but the USB part is newer than your,
> >if I trust the revision numbers. It'd be nice to see what changed
> >between the 2 revisions, but I don't know where to find this info.
> 
> The uhci/ehci works perfectly fine on sparc64. I noticed a few things  
> though:
> - - throughput is much higher. On the O2 bonnie++ maxed out at about  
> 2.5MB/s, in my U60 it goes past 5MB/s
> - - the O2 spends a lot of time in interrupt code ( >20% at times )  
> during bonnie++ runs, on the U60 it's less than 2%
> So, maybe there's something fishy in the O2-specific ( or maybe MIPS- 
> specific ) interrupt code.

I guess some of it is because the descriptors are mapped uncached on the
mips. This includes some parts that are never accessed by hardware.
This could be improved by splitting memory allocation in 2 parts
(one for descriptors used by hardware, and one for the host-only part
of the descriptors).

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--