Re: [RFC PATCH v1 2/4] xen/arm: Discovering PCI devices and add the PCI devices in XEN.

[Julien Grall <julien@xxxxxxx>]

Hi,

On 24/07/2020 08:14, Oleksandr Andrushchenko wrote:
> On 7/23/20 11:44 PM, Stefano Stabellini wrote:
> > On Thu, 23 Jul 2020, Rahul Singh wrote:
> > > Hardware domain is in charge of doing the PCI enumeration and will
> > > discover the PCI devices and then will communicate to XEN via hyper
> > > call PHYSDEVOP_pci_device_add to add the PCI devices in XEN.
> > >
> > > Change-Id: Ie87e19741689503b4b62da911c8dc2ee318584ac
> > Same question about Change-Id
> >
> >
> > > Signed-off-by: Rahul Singh <rahul.singh@xxxxxxx>
> > > ---
> > >    xen/arch/arm/physdev.c | 42 +++++++++++++++++++++++++++++++++++++++---
> > >    1 file changed, 39 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/xen/arch/arm/physdev.c b/xen/arch/arm/physdev.c
> > > index e91355fe22..274720f98a 100644
> > > --- a/xen/arch/arm/physdev.c
> > > +++ b/xen/arch/arm/physdev.c
> > > @@ -9,12 +9,48 @@
> > >    #include <xen/errno.h>
> > >    #include <xen/sched.h>
> > >    #include <asm/hypercall.h>
> > > -
> > > +#include <xen/guest_access.h>
> > > +#include <xsm/xsm.h>
> > >    
> > >    int do_physdev_op(int cmd, XEN_GUEST_HANDLE_PARAM(void) arg)
> > >    {
> > > -    gdprintk(XENLOG_DEBUG, "PHYSDEVOP cmd=%d: not implemented\n", cmd);
> > > -    return -ENOSYS;
> > > +    int ret = 0;
> > > +
> > > +    switch ( cmd )
> > > +    {
> > > +#ifdef CONFIG_HAS_PCI
> In the cover letter you were saying "we are not enabling the HAS_PCI and HAS_VPCI 
> flags for ARM".
>
> Is this still valid?
>
> > > +        case PHYSDEVOP_pci_device_add:
> > > +            {
> > > +                struct physdev_pci_device_add add;
> > > +                struct pci_dev_info pdev_info;
> > > +                nodeid_t node = NUMA_NO_NODE;
> > > +
> > > +                ret = -EFAULT;
> > > +                if ( copy_from_guest(&add, arg, 1) != 0 )
> > > +                    break;
> > > +
> > > +                pdev_info.is_extfn = !!(add.flags & XEN_PCI_DEV_EXTFN);
> > > +                if ( add.flags & XEN_PCI_DEV_VIRTFN )
> > > +                {
> > > +                    pdev_info.is_virtfn = 1;
> > > +                    pdev_info.physfn.bus = add.physfn.bus;
> > > +                    pdev_info.physfn.devfn = add.physfn.devfn;
> > > +                }
> > > +                else
> > > +                    pdev_info.is_virtfn = 0;
> > > +
> > > +                ret = pci_add_device(add.seg, add.bus, add.devfn,
> > > +                                &pdev_info, node);
> > > +
> > > +                break;
> > > +            }
> > > +#endif
> > > +        default:
> > > +            gdprintk(XENLOG_DEBUG, "PHYSDEVOP cmd=%d: not implemented\n", cmd);
> > > +            ret = -ENOSYS;
> > > +    }
> > I think we should make the implementation common between arm and x86 by
> > creating xen/common/physdev.c:do_physdev_op as a shared entry point for
> > PHYSDEVOP hypercalls implementations. See for instance:
> >
> > xen/common/sysctl.c:do_sysctl
> >
> > and
> >
> > xen/arch/arm/sysctl.c:arch_do_sysctl
> > xen/arch/x86/sysctl.c:arch_do_sysctl
> >
> >
> > Jan, Andrew, Roger, any opinions?
> I think we can also have a look at [1] by Julien. That implementation,
>
> IMO, had some thoughts on making Arm/x86 code common where possible
There are some ideas on how I would like to see the split, although they 
need some cleanup. :)
In particular, I was expecting some preparatory work to get the existing 
PCI code non-x86 specific.
Regarding the hypercall Stefano mentionned, I think it should be 
possible to make it common if we abstract the NUMA node lookup part.
Cheers,

> [1] 
> https://xenbits.xen.org/gitweb/?p=people/julieng/xen-unstable.git;a=shortlog;h=refs/heads/dev-pci
--
Julien Grall