[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH V13 3/4] pc: add a Virtual Machine Generation ID
Michael S. Tsirkin
Re: [Qemu-devel] [PATCH V13 3/4] pc: add a Virtual Machine Generation ID device
Thu, 5 Mar 2015 09:35:28 +0100
On Thu, Mar 05, 2015 at 01:51:26PM +1100, David Gibson wrote:
> On Tue, 3 Mar 2015 12:06:05 +0100
> Igor Mammedov <address@hidden> wrote:
> > On Tue, 3 Mar 2015 12:03:26 +1100
> > David Gibson <address@hidden> wrote:
> > > On Mon, 2 Mar 2015 22:06:22 +0100
> > > "Michael S. Tsirkin" <address@hidden> wrote:
> > >
> > > > On Mon, Mar 02, 2015 at 06:05:43PM +0100, Igor Mammedov wrote:
> > > > > On Sun, 1 Mar 2015 16:09:33 +0100
> > > > > "Michael S. Tsirkin" <address@hidden> wrote:
> > > > >
> > > > > > On Wed, Feb 25, 2015 at 05:08:52PM +0000, Igor Mammedov wrote:
> > > > > > > Based on Microsoft's sepecifications (paper can be dowloaded from
> > > > > > > http://go.microsoft.com/fwlink/?LinkId=260709), add a device
> > > > > > > description to the SSDT ACPI table and its implementation.
> > > > > > >
> > > > > > > The GUID is set using "vmgenid.uuid" property.
> > > > > > >
> > > > > > > Example of using vmgenid device:
> > > > > > > -device
> > > > > > > vmgenid,id=FOO,uuid="324e6eaf-d1d1-4bf6-bf41-b9bb6c91fb87"
> > > > > >
> > > > > > If you do this, doesn't windows then prompt for a driver?
> > > > > it doesn't since PCI_CLASS_MEMORY_RAM is displayed as driver less
> > > > > "PCI standard RAM Controller" binding in device manager.
> > > > >
> > > > > There was an issue with
> > > > > virtio balloon device + pseries firmware + kernel bug
> > > > > http://lists.gnu.org/archive/html/qemu-devel/2012-03/msg04704.html
> > > > > but it shouldn't be an issue for x86 targets with which device is
> > > > > supposed to be used.
> > > > > CCing David and Laszlo in case UEFI might do some crazy stuff like
> > > > > pseries firmware.
> > > >
> > > > I have to say, if it's not RAM, using PCI_CLASS_MEMORY_RAM seems
> > > > wrong. Can't we tag it in ACPI in some way?
> > I'm not sure that it's not a RAM though, (according to MS spec
> > it could be RAM, ROM or something else). This device is some RAM
> > controller and though it provides some RAM as its resource
> > it doesn't mean that it provides RAM that could be used freely as
> > as a generic RAM reported to host via E802 table.
> I think describing it as RAM is.. tremendously counter-intuitive at
> very best.
> > > I agree. PCI_CLASS_MEMORY_RAM means something quite specific, and this
> > > device isn't it. AFAICT, this would break pseries guests exactly like
> > > the balloon device did before we removed the bogus class code.
> > Could you send me a reference to source that explains what
> > PCI_CLASS_MEMORY_RAM means?
> > From what I've goggled out so far it's not RAM by itself but
> > some RAM controller.
> I was mistaken. Had a chat to someone with access to the PCI spec, and
> in fact PCI_CLASS_MEMORY_RAM specifies hardly anything.
> So, yeah, I think the fact that RAM class devices get treated as
> "normal" RAM by the pseries firmware is a bug in that firmware. I'll
> look into getting that fixed.
Thanks, pls keep us updated.
> > I've noticed that for example NVidia cards could expose
> > several of such devices (my guess it's VRAM) so BAR exposed by
> > such device shouldn't be treated as generic RAM blindly (like it seems
> > pseries does) without vendor specific driver for example like ivshmem.
> > I've also checked linux kernel, it only has unused define for this
> > class id.
> So, I still don't think PCI_CLASS_MEMORY_RAM is the best description.
> But, I wonder, is there any point to the argument. I wasn't clear if
> this was based on a Hyper-V specification that specifies the class
> code. If so, we're pretty much stuck with it, no matter what other
> irritations it causes.
No. AFAIK, there's no spec that requires this class value.
This is simply a work around found by Igor to avoid getting driver
prompt from windows. Apparently, if you set this class value, windows
does not prompt for driver even if there's no matching driver.
However, this behaviour seems to be undocumented.
Why this is the case, and whether this will keep being the case
in the future, I think it's safe to say, we don't know.
> > > I'm lacking context to see what the purpose of this device is, and
> > > whether we could ever want something similar on ppc64.
> > That is to provide guest with some UUID and possibly change it
> > at runtime and notify guest via ACPI about change as specified
> > by Microsoft spec.
> Ok. Can the device be used without extra ACPI stuff?
Absolutely: all the ACPI stuff does is return back BAR(0) value.
You can get at that without ACPI just as well.
> If not, it's
> clearly useless on pseries. If so, it's probably not the most natural
> way of presenting such information to a pseries guest, but it might be
> useful as a method that already exists.
> David Gibson <address@hidden>
> Senior Software Engineer, Virtualization, Red Hat