On Thursday, November 15, 2018 at 2:56:39 PM UTC+1, John Mitchell wrote: > On Thursday, November 15, 2018 at 1:49:04 PM UTC+1, Radosław Szkodziński > wrote: > > On Tue, Nov 13, 2018 at 8:23 PM John Mitchell <sonwon.1@<snip>> wrote: > > > > > > On Tuesday, November 13, 2018 at 5:21:22 PM UTC+1, Radosław Szkodziński > > > wrote: > > > > On Tue, Nov 13, 2018 at 12:02 PM John Mitchell <snip> wrote: > > > > > > > > > > On Tuesday, November 13, 2018 at 5:01:06 AM UTC+1, Radosław > > > > > Szkodziński wrote: > > > > > > On Thu, Nov 8, 2018 at 11:48 AM Marek Marczykowski-Górecki > > > > > > <marmarek@<snip>> wrote: > > > > > > > > > > > > > > -----BEGIN PGP SIGNED MESSAGE----- > > > > > > > Hash: SHA256 > > > > > > > > > > > > > > On Thu, Nov 08, 2018 at 01:52:37AM -0800, John Mitchell wrote: > > > > > > > > On Thursday, November 8, 2018 at 10:24:13 AM UTC+1, Laszlo > > > > > > > > Zrubecz wrote: > > > > > > > > <snip> > > > > > > > > > > Thank you for the informative reply. > > > > > > > > > > It appears the Vega reset bug is related to the VM presenting the > > > > > card as a legacy PCI device rather than a PCIe device. > > > > > > > > AMD has said that Vega needs PSP reset issued before disabling IOMMU > > > > otherwise its internal bus will be all sorts of messed up and > > > > unavailable. > > > > Alternatively you can quirk this device to not be reset and put in D3 > > > > at all, then it will be reset by the driver on next guest boot, but > > > > that has some security implications. > > > > > > I've discovered a way to reset the card inside of Windows during shutdown > > > and startup. So I think I am good there. I admit I haven't been able to > > > test this for myself. > > > > If you happen to have the information on how to make Windows 10 1803 > > or preferably 1809 and AMD drivers 18.xx to reset the GPU, please > > share, since that'd be faster for me than patching vfio-pci with PSP > > reset code. > > > > I've tried various approaches including ejecting the device and they > > have all failed. (The device is clearly enabled - the power lights > > show it.) The dmesg shows IOVA errors as the GPU is trying to access > > old invalid mapping and IOMMU prevents this. > > With 17.xx drivers everything just works as the driver resets the GPU > > on next boot with the flashing power light... > > > > -- > > Radosław Szkodziński > > Check this thread, > > Linux Host, Windows Guest, GPU passthrough reinitialization fix > > My understanding this only works with Windows Pro versions. What isn't > working with this method are forced reboots from Windows Update. Easy to > control with Windows 7 however Windows 10 will be a pain because of forced > updates. It would be nice if this was incorporated into the VM startup and > shutdown sequence, not sure if this is feasible since it requires Windows > Device Console (Devcon.exe). > > Also, I saw a thread somewhere where the graphic card shutdown sequence could > be initiated via a different method inside of Windows that should work with > Windows Updates. I didn't save that thread and it was untested when I read > the thread. > > I've also read that Linux kernel 4.16 has a fix for the graphic card reset > problem. So maybe the problem is gone with Linux OSes using a new kernel? I > am not familiar with the Qubes kernel? I assume it is a derivative of the > Linux kernel. Is that correct? If it is which version is the current Qubes > OS derived from? > > I hope that helps. > > Sincerely, > > John Mitchell
https://forum.level1techs.com/t/linux-host-windows-guest-gpu-passthrough-reinitialization-fix/121097 -- You received this message because you are subscribed to the Google Groups "qubes-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to qubes-devel+unsubscr...@googlegroups.com. To post to this group, send email to qubes-devel@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/qubes-devel/076b4ec0-cc8a-4714-84dc-657e8c6bee3d%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.