Re: [Qemu-devel] [qemu-s390x] [RFC PATCH 1/3] hw/Kconfig: PCI bus implies PCI_DEVICES

[Cornelia Huck <cohuck@redhat.com>]

On Mon, 15 Jul 2019 18:08:45 +0200
Markus Armbruster <armbru@redhat.com> wrote:

> Philippe Mathieu-Daudé <philmd@redhat.com> writes:
> 
> > On 7/15/19 3:19 PM, Thomas Huth wrote:  
> >> On 15/07/2019 13.09, Cornelia Huck wrote:  
> >>> On Mon, 15 Jul 2019 13:04:28 +0200
> >>> Philippe Mathieu-Daudé <philmd@redhat.com> wrote:
> >>>  
> >>>> On 7/15/19 12:56 PM, Cornelia Huck wrote:  
> >>>>> On Mon, 15 Jul 2019 12:48:55 +0200
> >>>>> Thomas Huth <thuth@redhat.com> wrote:
> >>>>>     
> >>>>>> On 15/07/2019 12.19, Peter Maydell wrote:    
> >>>>>>> On Mon, 15 Jul 2019 at 11:15, Thomas Huth <thuth@redhat.com> wrote:      
> >>>>>>>>
> >>>>>>>> On 15/07/2019 11.55, Philippe Mathieu-Daudé wrote:      
> >>>>>>>>> If a controller device provides a PCI bus, we can plug any PCI
> >>>>>>>>> daughter card on it.
> >>>>>>>>>
> >>>>>>>>> Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
> >>>>>>>>> ---      
> >>>>>>>       
> >>>>>>>>> diff --git a/hw/pci/Kconfig b/hw/pci/Kconfig
> >>>>>>>>> index 77f8b005ff..0f7267db35 100644
> >>>>>>>>> --- a/hw/pci/Kconfig
> >>>>>>>>> +++ b/hw/pci/Kconfig
> >>>>>>>>> @@ -1,5 +1,6 @@
> >>>>>>>>>  config PCI
> >>>>>>>>>      bool
> >>>>>>>>> +    imply PCI_DEVICES      
> >>>>>>>>
> >>>>>>>> No, please don't change this. This was done on purpose, since almost all
> >>>>>>>> PCI_DEVICES do not work on s390x (so s390x does *not* imply PCI_DEVICES).      
> >>>>>>>
> >>>>>>> But that means that every board that provides PCI has to have an
> >>>>>>> "imply PCI_DEVICES" line, which is pretty clunky just to work
> >>>>>>> around an s390x limitation.
> >>>>>>>
> >>>>>>> Is there some way in the Kconfig syntax for s390x to say
> >>>>>>> "no PCI_DEVICES" so we can have the corner-case be handled
> >>>>>>> by the s390x Kconfig in one place rather than in 20 places
> >>>>>>> affecting everywhere except s390x?      
> >>>>>>
> >>>>>> IIRC the problem on s390x are the legacy IRQs. s390x has only MSIs. So I
> >>>>>> guess the correct way to fix this would be to introduce some
> >>>>>> PCI_LEGACY_IRQ switch and let all old devices that do not work with MSI
> >>>>>> depend on it.    
> >>>>>
> >>>>> s/MSI/MSI-X/, IIRC. Not sure how far 'legacy' would stretch.    
> >>>>
> >>>> Maybe we can have something like PCI_LEGACY_DEVICES and PCI_MSI_DEVICES?
> >>>>
> >>>> So if s390x only selects PCI_LEGACY (not PCI_MSI) bus, then it only get
> >>>> legacy devices?  
> >>>
> >>> Wrong way around? We need MSI-X for s390x, not plain MSI or
> >>> 'legacy' (whatever that is).  
> >> 
> >> With "legacy" I meant the old level-triggered interrupts from the early
> >> PCI (non-express) days. Sorry for being imprecise here.
> >> 
> >> So maybe we need two new switches, PCI_CLASSIC (or so) and PCI_MSIX, and
> >> then the PCI devices should be marked with "default y if PCI_CLASSIC" if
> >> they do not have MSIX support, and with "default y if PCI_MSIX" if they
> >> have MSI-X support?  
> >
> > Something like that :)
> >
> > Per Wikipedia:
> >
> >   Conventional PCI and PCI-X are sometimes called Parallel PCI
> >   in order to distinguish them technologically from their more
> >   recent successor PCI Express, which adopted a serial,
> >   lane-based architecture.
> >
> >   The PCI-SIG introduced the serial PCI Express in c. 2004. At
> >   the same time, they renamed PCI as Conventional PCI.
> >
> >   PCI Express does not have physical interrupt lines at all.
> >   It uses message-signaled interrupts exclusively.
> >
> > What about PCI_CONVENTIONAL then?  
> 
> What kinds of PCI devices are we trying to name?
> 
> Is it INTx vs. MSI vs. MSI-X?

I think for s390x we need (INTx || MSI) vs MSI-X...

> 
> Is it Conventional PCI vs. PCI Express?

...while this is probably more INTx vs (MSI || MSI-X)?