Re: [Xen-devel] [PATCH v3 09/35] OvmfPkg/OvmfXen: use a TimerLib instance that depends only on the CPU

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Mon, Jul 22, 2019 at 09:28:20PM +0200, Laszlo Ersek wrote:
> On 07/22/19 15:49, Anthony PERARD wrote:
> > On Mon, Jul 15, 2019 at 04:22:19PM +0200, Roger Pau Monné wrote:
> >> On Thu, Jul 04, 2019 at 03:42:07PM +0100, Anthony PERARD wrote:
> >>> ACPI Timer does not work in a PVH guest, but local APIC works on both
> >>
> >> This is not accurate. It's not that the ACPI timer doesn't work, it's
> >> just that it's not present. The PM_TMR_BLK should be set to 0 to
> >> indicate the lack of PM timer support, or else there's something
> >> broken.
> > 
> > I'll reword that first sentence.
> > 
> > OVMF doesn't look at the PM_TMR_BLK value when initializing that timer,
> > it only looks at the PCI host bridge device ID because OVMF is built
> > with QEMU in mind and there are only two possibles choices, QEMU is
> > running with a piix or q35 machine type, I think.
> 
> We should split this statement in two. :)
> 
> OVMF doesn't look at ACPI payload because it is a design goal to keep
> the guest firmware un-enlightened about such ACPI contents that arrive
> from the hypervisor. Parsing ACPI in firmware always looks attractive
> until someone actually writes the code, and then it always ends in
> misery -- at the latest when people realize they have to parse AML.
> Parsing ACPI is only feasible when you have a full-blown ACPICA (or
> similar) subsystem, and edk2 doesn't. Therefore, OVMF looks at either
> hardware, or specialized paravirt information channels such as fw_cfg
> files, that are easy to parse by design.

IMO passing information using such side-channels always looks
attractive at first sight, until you realize at some point later that
you just have ended up with a completely custom interface that
duplicates ACPI. Having that said, Xen manages to get most of what it
needs from static ACPI tables, but I'm not sure if OVMF could manage
to do so also.

Xen has quite a lot of baggage here, like using xenstore/xenbus
instead of PCI, or custom 'start info pages' instead of ACPI, which we
are currently trying to partially move away from when possible.

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel