Re: [dpdk-dev] [RFC] ethdev: introduce DMA memory mapping for external memory

[Shahaf Shuler <shahafs@mellanox.com>]

Hi Anatoly,

Any last inputs on this one? Would like to prepare a patch for 19.05 and it would much effect the design. 

Thursday, November 22, 2018 1:32 PM, Shahaf Shuler:
> Subject: RE: [dpdk-dev] [RFC] ethdev: introduce DMA memory mapping for
> external memory
> 


[...]


> > > , and we will have to rewrite current DMA mapping code to do
> > >> mappings explicitly, rather than relying on callbacks mechanism.
> > >> So, instead of e.g. VFIO mapping being performed via callback, we
> > >> would instead iterate over each rte_device and call its respective
> > >> DMA mapping function explicitly - for internal memory, but not for
> > >> external. This
> > >> *may* work, but i need to think it through a bit more :)
> > >>
> > >> This also still leaves a question of external unregistered memory -
> > >> this is a wholly separate issue.
> > >>
> > >> TL;DR i'm tentatively OK with the proposal, but we need to work
> > >> certain things out before we go too far along the path and find out
> > >> that we've reached a dead end.
> > >
> > > Fully agree, this is why the RFC is for.
> > > I think our biggest question is whether or not VFIO will behave OK
> > > with
> > double mapping or we need some tracking mechanism in DPDK.
> >
> > Yes, i will check this for VFIO and get back to you.
> 
> Great thanks.

?

> 
> >
> > >
> > >>
> > >> --
> > >> Thanks,
> > >> Anatoly
> >
> >
> > --
> > Thanks,
> > Anatoly