Re: [PATCH v2 0/8] VMD add second rootbus support

[Bjorn Helgaas <helgaas@kernel.org>]

On Wed, Nov 20, 2024 at 10:10:21AM +0100, Mariusz Tkaczyk wrote:
> On Fri, 15 Nov 2024 11:48:55 -0600
> Bjorn Helgaas <helgaas@kernel.org> wrote:
> 
> > The "BUS0" nomenclature seems heavily embedded in the vmd driver but
> > is really a misnomer.  Maybe that reflects similar terminology in an
> > internal spec?  Any hard-wiring of bus numbers reflects a property of
> > the way a *Root Port* works, so using the right name will make this
> > easier to understand, especially since there are other Root Ports with
> > the same hard wiring.
> 
> It is named as BUS_0 and BUS_1 because it refers to kernel's
> pci_create_root_bus()`, we simplified it for better code wrapping.
> 
> Looks like it confuses you, do you like ROOT_BUS0 and ROOT_BUS1 better?
> 
> I don't see better fit than "root bus" because we are creating new root buses
> accordingly.

I was trying to use standard PCIe spec terminology where possible, but
I misspoke above, sorry.  I think we should pretend I never wrote that
paragraph, and also ignore my suggestion about replacing "rootbus"
with "Root Port" in the 7/8 commit log (sorry to everybody who never
got the patches I was responding to).

A "root bus" is the *primary* bus of a Root Port, so the root bus
number is determined by the host bridge, which I guess would be
considered part of a PCIe Root Complex.

> Internally, it is called as "VMD domain" and each VMD Domain is presented as
> separate pci root bus.
> 
> The original naming is presented in the sysfs links:
> - "domain" -> root bus 0
> = "domain1" -> root bus 1
> 
> i.e:
> domain -> pci10000:e0/pci_bus/10000:e0
> domain1 -> pci10000:e1/pci_bus/10000:e1

That's fine, I'm not suggesting any sysfs change to this.

Bjorn