On Thu, Sep 29, 2022 at 12:14:10PM +0200, Jan Beulich wrote: > On 28.09.2022 16:11, Roger Pau Monne wrote: > > The current calls to memory_type_changed() are wider than strictly > > necessary. Move them inside the iocap handlers and also limit to only > > issue them when required. > > > > I would really like to get some feedback on the Arm change, since this > > is now a prereq for the actual fix. > > > > Thanks, Roger. > > > > Roger Pau Monne (2): > > arm/vgic: drop const attribute from gic_iomem_deny_access() > > x86/ept: limit calls to memory_type_changed() > > Are there intentions for having these on 4.17?
I wasn't sure. From XenServer PoV it's certainly a bug fix, otherwise some workloads related to GPU passthrough are simply too slow to be usable. I would certainly be fine with it making it's way into 4.17, let me add Henry: Cons: - Changes the number of issued memory_type_changed(), so there's a risk I misplaced some of the conditions and we end up with wrong cache types in the guest p2m due to missing memory_type_changed() calls. That however won't affect Xen itself, just the guest. Pros: - Removes unneeded memory_type_changed(), thus making some operations faster. It's effect it's greatly dependent on using a set of hypercalls against a domain, which doesn't seem common in upstream. It's possible other products based on Xen apart from XenServer will also see an speedup as a result. Thanks, Roger.
