Ok.. The PPC team are finally going whole hog on the VIO stuff and
going after net front/back.
I'm looking at linux: network_alloc_rx_buffers() which allocates a
bunch of skbuf pages and eventually calls XENMEM_decrease_reservation
to remove them from the domain.
We use large pages, and in fact we are forced to a special area to
handle real mode exceptions called an RMA.
Consider it a permanent large page (64M) that is used for real mode,
this is our minimum memory.
In order to decrease a reservation that mean that memory must be
greater than this minimum memory by at least as much as we want to
decrease. We also want to make sure that pages that are part of the
decrease do not come from the RMA.
Any thoughts on this would be appreciated.
Also, I know others are thinking about super-pages, any progress on
that with regard to grant transfers?
Xen-ppc-devel mailing list