Re: [edk2-devel] [PATCH v3 1/4] OvmfPkg/Sec: Setup MTRR early in the boot process.

["Sun, Yi Y" <yi.y.sun@intel.com>]

> -----Original Message-----
> From: Gerd Hoffmann <kraxel@redhat.com>
> Sent: Thursday, February 1, 2024 5:43 PM
> To: Sun, Yi Y <yi.y.sun@intel.com>
> Cc: Xu, Min M <min.m.xu@intel.com>; devel@edk2.groups.io;
> lersek@redhat.com; Michael Roth <michael.roth@amd.com>; Oliver Steffen
> <osteffen@redhat.com>; Yao, Jiewen <jiewen.yao@intel.com>; Tom
> Lendacky <thomas.lendacky@amd.com>; Ard Biesheuvel
> <ardb+tianocore@kernel.org>; Aktas, Erdem <erdemaktas@google.com>;
> Huang, Jiaqing <jiaqing.huang@intel.com>
> Subject: Re: RE: [edk2-devel] [PATCH v3 1/4] OvmfPkg/Sec: Setup MTRR early
> in the boot process.
> 
> On Thu, Feb 01, 2024 at 06:10:26AM +0000, Sun, Yi Y wrote:
> > Hi, all,
> >
> > Per our test, this issue only happens when the mdev/vdev is assigned
> > to VM. But PF(physical function)/VF(virtual function) assignment is
> > good. Per Jiaqing's investigation, it is because that the mdev/vdev
> > passthrough flow is different with PF/VF. For vdev/mdev passthrough to
> > VM, the enforce_cache_coherency was enabled after device bound to
> > iommufd. But it doesn't update the kvm coherency state after that.
> > This makes the memory type be UNCACHABLE which causes the reading is
> > very slow.
> >
> > So, what is your opinion to fix it? Which side to fix it is better,
> > ovmf or vfio? Thanks!
> 
> Both?  For OVMF it surely makes sense to properly setup MTRRs.  And if the
> cache coherency behavior of vfio is not correct that should be fixed too of
> course.
> 
Sounds good. Thanks!

BRs,
Yi Sun

> take care,
>   Gerd



-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#114979): https://edk2.groups.io/g/devel/message/114979
Mute This Topic: https://groups.io/mt/104052591/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-