[vfio-users] AMD Ryzen KVM/NPT/IOMMU issue

2017-06-28 Thread Graham Neville
Although not related to graphics card performance, there is definitely another issue with regards to running KVM nested L2 guests when npt=1. Thought I'd mention this in case it helps with identifying performance issues with NPT. I'm unable to start any L2 guests with KVM acceleration (--enable-k

Re: [vfio-users] AMD Ryzen Nested Page Table Performance Oddities

2017-04-23 Thread Graham Neville
at 11:49 PM, Nick Sarnie wrote: > On Thu, Apr 20, 2017 at 12:57 PM, Graham Neville > wrote: > > So turns out a Linux guest with Q35 and OVMF can work with npt on or off. > > However a Linux guest with i440fx only works with npt disabled. > > > > I'll f

Re: [vfio-users] AMD Ryzen Nested Page Table Performance Oddities

2017-04-20 Thread Graham Neville
> On Tue, Apr 18, 2017 at 10:09 AM, Graham Neville > wrote: > > Has there been any feedback on this at all? > > > > I'm still struggling to get a Windows and Linux VM working at the same > time > > with npt=1. Annoying as I'm having to reboot in to

Re: [vfio-users] AMD Ryzen Nested Page Table Performance Oddities

2017-04-18 Thread Graham Neville
ure out what is going on. Also, Alex has reproduced this himself on both Nvidia and AMD GPUs. On Fri, Apr 7, 2017 at 5:49 PM, Graham Neville wrote: > Sarnex, > > I have similar oddities with npt with Ryzen. As you say with kvm-amd.npt=0 > the GPU performance is so much better (In m

Re: [vfio-users] AMD Ryzen Nested Page Table Performance Oddities

2017-04-07 Thread Graham Neville
Sarnex, I have similar oddities with npt with Ryzen. As you say with kvm-amd.npt=0 the GPU performance is so much better (In my Witcher3 tests I'm going from 40fps to 75fps!). However with npt disabled the Windows10 VM slows down an awful lot in general tasks. I've also noticed that I can only ru

Re: [vfio-users] Ryzen Primary GPU passthrough success and woes

2017-04-03 Thread Graham Neville
the GPU > is in use in pass-through? > > watch -d cat /proc/interrupts > > Look at the bottom for PIN (Posted-interrupt notification event) and PIW > (Posted-interrupt wakeup event). > > Thanks! > > On 30 Mar 2017, at 15:46, Graham Neville wrote: > > Finally go

Re: [vfio-users] Ryzen Primary GPU passthrough success and woes

2017-03-30 Thread Graham Neville
On Thu, Mar 30, 2017 at 6:04 AM, Graham Neville wrote: > I am using a ROM file which I dumped myself using GPU-Z, this was the only > was I could get a display as well. With Q35 I do

Re: [vfio-users] Ryzen Primary GPU passthrough success and woes

2017-03-29 Thread Graham Neville
On Wed, Mar 29, 2017 at 5:44 PM, Graham Neville wrote: > Thanks for the link. I've tried a number of things but still no further down > the line. > > I've tried the following one-by-one > > kvm_amd.avic=1 > Properly isolate

Re: [vfio-users] Ryzen Primary GPU passthrough success and woes

2017-03-29 Thread Graham Neville
ter for you without AVIC than they > did for me, but perhaps the extra improvement in IRQ latency would fix > the hangs you get during intensive graphics operations. > > > On Tue, Mar 28, 2017 at 6:02 PM, Graham Neville > wrote: > > I've managed to get pci-e passthough work

[vfio-users] Ryzen Primary GPU passthrough success and woes

2017-03-28 Thread Graham Neville
I've managed to get pci-e passthough working on a gigabyte gaming 3 matx MB and Ryzen 1700, no ACS patch, using only 1 GPU - AMD r9 290. However I'm facing a problem with the whole KVM setup and not sure what it's related to. For the Windows10 guest with the GPU passed through it crashes (guest onl