On Fri, Jan 07, 2022 at 11:18:07AM -0400, Jason Gunthorpe wrote: > On Thu, Jan 06, 2022 at 10:06:42PM -0500, Daniel Jordan wrote: > > > > At least it seems like it is not an insurmountable problem if it makes > > > an appreciable difference.. > > > > Ok, I can think more about this. > > Unfortunately iommufd is not quite ready yet, otherwise I might > suggest just focus on that not type 1 optimizations. Depends on your > timeframe I suppose. Ok, I see. Well, sooner the better I guess, we've been carrying changes for this a while. > > > After seeing Daniels's patches I've been wondering if the pin step in > > > iommufd's draft could be parallized on a per-map basis without too > > > much trouble. It might give Daniel a way to do a quick approach > > > comparison.. > > > > Sorry, comparison between what? I can take a look at iommufd tomorrow > > though and see if your comment makes more sense. > > I think it might be easier to change the iommufd locking than the > type1 locking to allow kernel-side parallel map ioctls. It is already > almost properly locked for this right now, just the iopt lock covers a > little bit too much. > > It could give some idea what kind of performance user managed > concurrency gives vs kernel auto threading. Aha, I see, thanks!
