On Sun, 2010-07-25 at 18:34 -0400, John Martin wrote:
> On 07/25/10 02:02 PM, Garrett D'Amore wrote:
>
> > There may *also* be a limit on the number of IOMMU entries. I know some
> > platforms had such limitations. Perhaps you can try reducing from 1000
> > frames prealloc'd to something smaller -- say 512? Is this a 10GbE
> > driver? If its only 1GbE, then 512 ought to be enough.
>
> Since the original mail shows a SG list length of 16, is there
> some clue that can be given to the OPL nexus that DVMA/IOMMU
> space allocation is not needed and the transfers can be done
> using the real page addresses (or has VM forced a level of
> abstraction)?
There's a FORCE_PHYSICAL flag which *could* be used. Whether it will
actually have the desired effect or not, I'm uncertain.
Note that this should not be used on SPARC with SCSA drivers. SCSA
makes some assumptions on SPARC, and does not work well with drivers
that do not use virtual DMA. (This is #ifdef'd code in SCSA for SPARC,
notably x86 does not have this issue.)
- Garrett
>
> _______________________________________________
> driver-discuss mailing list
> [email protected]
> http://mail.opensolaris.org/mailman/listinfo/driver-discuss
_______________________________________________
driver-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/driver-discuss
