RE: git: 6171e026be11 - main - bhyve: add support for MTRR

Hi,

this commit shouldn't add real MTRR emulation. The nvidia GPU driver
requires MTRR to work properly. If MTRR is missing, the driver
will fail on startup. Adding a dummy MTRR implementation is
sufficient to solve this issue.


Best regards
Corvin

Beckhoff Automation GmbH & Co. KG | Managing Director: Dipl. Phys. Hans Beckhoff
Registered office: Verl, Germany | Register court: Guetersloh HRA 7075


-----Original Message-----
From: Konstantin Belousov <kostikbel@gmail.com>
Sent: Tuesday, January 18, 2022 5:32 PM
To: John Baldwin <jhb@freebsd.org>
Cc: Emmanuel Vadot <manu@freebsd.org>; src-committers@freebsd.org; dev-commits-src-all@freebsd.org; dev-commits-src-main@freebsd.org
Subject: Re: git: 6171e026be11 - main - bhyve: add support for MTRR

CAUTION: External Email!!
On Tue, Jan 18, 2022 at 07:19:15AM -0800, John Baldwin wrote:
> On 1/14/22 3:42 AM, Emmanuel Vadot wrote:
> > The branch main has been updated by manu:
> >
> > URL: https://nospamproxywebp.beckhoff.com/enqsig/link?id=BCAAAABGd7m3lfMOLOIhx3uNR_1Qav4Ffx8-jQB0torD82ts_KYAAABj7YMrJMLJ7cXPyMvbFX9WJKJ3O_9eSEsrjC4dOUoxa4zlBD0VGYmBMUrUqe0J6nLTsxbGabU9mr5yyGtEAvpJO3CAuBBgfol1iNi8mky-rPfp7A4-KzOkZzVVDoTvyV5onuiY7akaRkSHD6_6whenu6bm6k3330UqkbIJnhHHTAH_aCuNdUpRPB21mCqUtWVbmJR0l1ZlmblFbs21-oJr5_5aqmOQ0
> >
> > commit 6171e026be11824495cebe8baf559af673a8e533
> > Author:     Corvin Köhne <CorvinK@beckhoff.com>
> > AuthorDate: 2022-01-14 09:58:48 +0000
> > Commit:     Emmanuel Vadot <manu@FreeBSD.org>
> > CommitDate: 2022-01-14 11:41:44 +0000
> >
> >      bhyve: add support for MTRR
> >      Some guests or driver might depend on MTRR to work properly. E.g. the
> >      nvidia gpu driver won't work without MTRR.
> >      Reviewed by:    markj
> >      MFC after:      2 weeks
> >      Sponsored by:   Beckhoff Automation GmbH & Co. KG
> >      Differential Revision:  https://nospamproxywebp.beckhoff.com/enqsig/link?id=BCAAAABGd7m3lfMOLOIhx3uNR_1Qav4Ffx8-jQB0torD82ts_HgAAADXKw6wUJtKFi6JF3K0iUvqHd4jzUMtqPnMIo16BSb3OZ4KOWvJONCUys3_dW62_stIRwAB1O3CWRQezI1ZSyN_mqt7E2_iAJ_f4pJPER5NYObzKaqbtVmN0KRAdilVPRQ_5JGBJM3VaTlUYexq-jay78fOBZtyFBs1
>
> So this doesn't look like actual emulation, but more like a dummy version?
> That is, the values aren't initialized to the "normal" state (WB for guest RAM,
> UC- for everything else), and the state isn't enforced in any way (I think the
> only sane way to actually implement these would be to use MTRR to set PAT values
> in EPT entries if EPT even supports PAT).  Perhaps this mostly doesn't matter
> since we don't directly map BARs in EPT anyway?  (Though perhaps for passthrough
> we might in which case you'd want to honor these settings in those?)
>
I do not believe that it is safe to enable actual application of the cache
modes set by guest, to the EPT pages. At very least, there are a lot of
erratas claiming broken situations like inconsistent mappings modes causing
machine checks or reboots on Intel processors.

Or, it should be only enabled for specific trusted vms.  For passthough,
I think we only need to set proper mode for BARs in EPT in advance.