Re: [PATCH v5 06/14] vpci/header: implement guest BAR register handlers

[Jan Beulich <jbeulich@xxxxxxxx>]

On 03.02.2022 13:45, Oleksandr Andrushchenko wrote:
>>> Also memory decoding needs to be initially disabled when used by
>>> guests, in order to prevent the BAR being placed on top of a RAM
>>> region. The guest physmap will be different from the host one, so it's
>>> possible for BARs to end up placed on top of RAM regions initially
>>> until the firmware or OS places them at a suitable address.
>> Agree, memory decoding must be disabled
> Isn't it already achieved by the toolstack resetting the PCI device
> while assigning  it to a guest?

Iirc the tool stack would reset a device only after getting it back from
a DomU. When coming straight from Dom0 or DomIO, no reset would be
performed. Furthermore, (again iirc) there are cases where there's no
known (standard) way to reset a device. Assigning such to a guest when
it previously was owned by another one is risky (and hence needs an
admin knowing what they're doing), but may be acceptable in particular
when e.g. simply rebooting a guest.

IOW - I don't think you can rely on the bit being in a particular state.

Jan