RE: [PATCH RESEND v2 3/3] PCI: hv: New paravirtual PCI front-end for Hyper-V VMs

[Jake Oshins]

> -----Original Message-----
> From: Bjorn Helgaas [mailto:helgaas@xxxxxxxxxx]
> Sent: Tuesday, February 16, 2016 2:44 PM
> To: Jake Oshins <jakeo@xxxxxxxxxxxxx>
> Cc: bhelgaas@xxxxxxxxxx; linux-pci@xxxxxxxxxxxxxxx;
> gregkh@xxxxxxxxxxxxxxxxxxx; KY Srinivasan <kys@xxxxxxxxxxxxx>; linux-
> kernel@xxxxxxxxxxxxxxx; devel@xxxxxxxxxxxxxxxxxxxxxx; olaf@xxxxxxxxx;
> apw@xxxxxxxxxxxxx; vkuznets@xxxxxxxxxx; tglx@xxxxxxxxxxxxx; Haiyang
> Zhang <haiyangz@xxxxxxxxxxxxx>; marc.zyngier@xxxxxxx; Hadden
> Hoppert <haddenh@xxxxxxxxxxxxx>
> Subject: Re: [PATCH RESEND v2 3/3] PCI: hv: New paravirtual PCI front-end
> for Hyper-V VMs
> 
> On Tue, Feb 16, 2016 at 09:46:55PM +0000, Jake Oshins wrote:
> > > -----Original Message-----
> > > From: Bjorn Helgaas [mailto:helgaas@xxxxxxxxxx]
> > > Sent: Tuesday, February 16, 2016 8:46 AM
> > > To: Jake Oshins <jakeo@xxxxxxxxxxxxx>
> > > Cc: bhelgaas@xxxxxxxxxx; linux-pci@xxxxxxxxxxxxxxx;
> > > gregkh@xxxxxxxxxxxxxxxxxxx; KY Srinivasan <kys@xxxxxxxxxxxxx>;
> > > linux- kernel@xxxxxxxxxxxxxxx; devel@xxxxxxxxxxxxxxxxxxxxxx;
> > > olaf@xxxxxxxxx; apw@xxxxxxxxxxxxx; vkuznets@xxxxxxxxxx;
> > > tglx@xxxxxxxxxxxxx; Haiyang Zhang <haiyangz@xxxxxxxxxxxxx>;
> > > marc.zyngier@xxxxxxx; Hadden Hoppert <haddenh@xxxxxxxxxxxxx>
> > > Subject: Re: [PATCH RESEND v2 3/3] PCI: hv: New paravirtual PCI
> > > front-end for Hyper-V VMs
> > >
> > > Hi Jake,
> > >
> > > Looks good to me overall; I marked a few nits below.
> > >
> > > The only real question I have is about domain number allocation.
> > > See the note below.
> > >
> > [snip]
> > > > +
> > > > +	/*
> > > > +	 * The PCI bus "domain" is what is called "segment" in
> > > > +	 * ACPI and other specs.  Pull it from the instance ID,
> > > > +	 * to get something unique.  Bytes 8 and 9 are what is used
> > > > +	 * in Windows guests, so do the same thing for consistency.
> > > > +	 */
> > > > +
> > > > +	hbus->sysdata.domain = hdev->dev_instance.b[9] |
> > > > +			       hdev->dev_instance.b[8] << 8;
> > >
> > > How do we know this is unique?  We don't have any idea what the
> > > platform will put in _SEG, so I think there's a potential conflict
> > > here.  The Intel VMD driver (arch/x86/pci/vmd.c) has a similar
> > > problem, and it looks for unused domain numbers starting at 0x10000
> > > (see vmd_find_free_domain()).
> > >
> >
> > Bjorn, thank you for your very thorough reviews.  I deeply appreciate it.  I
> checked the Hyper-V code and it currently does guarantee that these values
> are unique.  When I resend the series, I'll add a comment to that effect.  I'll
> also add a comment to Hyper-V that says that it has to stay that way.
> 
> I'm not familiar with how Hyper-V works, but I guess you're saying that
> Hyper-V controls what the guest sees via _SEG as well as what it sees via the
> instance ID.
> 
> Bjorn

Yes, exactly.  To the point, the Hyper-V Generation 1 VM (which is an emulation of a 440BX/PIIX4 platform) has one root PCI bus with no _SEG object at all, which implies a value of zero.  The Hyper-V Generation 2 VM (which doesn't emulate any existing machine) doesn't have any root PCI bus at all until a device is passed through and this driver is loaded.

-- Jake