Re: [RFC 1/2] arm64: PCI: Allow use arch-specific pci sysdata

[Marc Zyngier <maz@kernel.org>]

Thanks Bjorn for looping me in.

On Fri, 19 Mar 2021 21:12:46 +0000,
Bjorn Helgaas <helgaas@kernel.org> wrote:
> 
> [+cc Arnd (author of 37d6a0a6f470 ("PCI: Add
> pci_register_host_bridge() interface"), which I think would make my
> idea below possible), Marc (IRQ domains maintainer)]
> 
> On Sat, Mar 20, 2021 at 12:19:55AM +0800, Boqun Feng wrote:
> > Currently, if an architecture selects CONFIG_PCI_DOMAINS_GENERIC, the
> > ->sysdata in bus and bridge will be treated as struct pci_config_window,
> > which is created by generic ECAM using the data from acpi.
> 
> It might be a mistake that we put the struct pci_config_window
> pointer, which is really arch-independent, in the ->sysdata element,
> which normally contains a pointer to arch- or host bridge-dependent 
> data.
> 
> > However, for a virtualized PCI bus, there might be no enough data in of
> > or acpi table to create a pci_config_window. This is similar to the case
> > where CONFIG_PCI_DOMAINS_GENERIC=n, IOW, architectures use their own
> > structure for sysdata, so no apci table lookup is required.
> > 
> > In order to enable Hyper-V's virtual PCI (which doesn't have acpi table
> > entry for PCI) on ARM64 (which selects CONFIG_PCI_DOMAINS_GENERIC), we
> > introduce arch-specific pci sysdata (similar to the one for x86) for
> > ARM64, and allow the core PCI code to detect the type of sysdata at the
> > runtime. The latter is achieved by adding a pci_ops::use_arch_sysdata
> > field.
> > 
> > Originally-by: Sunil Muthuswamy <sunilmut@microsoft.com>
> > Signed-off-by: Boqun Feng (Microsoft) <boqun.feng@gmail.com>
> > ---
> >  arch/arm64/include/asm/pci.h | 29 +++++++++++++++++++++++++++++
> >  arch/arm64/kernel/pci.c      | 15 ++++++++++++---
> >  include/linux/pci.h          |  3 +++
> >  3 files changed, 44 insertions(+), 3 deletions(-)
> > 
> > diff --git a/arch/arm64/include/asm/pci.h b/arch/arm64/include/asm/pci.h
> > index b33ca260e3c9..dade061a0658 100644
> > --- a/arch/arm64/include/asm/pci.h
> > +++ b/arch/arm64/include/asm/pci.h
> > @@ -22,6 +22,16 @@
> >  
> >  extern int isa_dma_bridge_buggy;
> >  
> > +struct pci_sysdata {
> > +	int domain;	/* PCI domain */
> > +	int node;	/* NUMA Node */
> > +#ifdef CONFIG_ACPI
> > +	struct acpi_device *companion;	/* ACPI companion device */
> > +#endif
> > +#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
> > +	void *fwnode;			/* IRQ domain for MSI assignment */

Why isn't this more strongly typed? pci_host_bridge_msi_domain()
definitely expects this to be the real thing. And the comment is
wrong.

[...]

> > +#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
> > +static inline void *_pci_root_bus_fwnode(struct pci_bus *bus)
> > +{
> > +	struct pci_sysdata *sd = bus->sysdata;
> > +
> > +	if (bus->ops->use_arch_sysdata)
> > +		return sd->fwnode;
> > +
> > +	/*
> > +	 * bus->sysdata is not struct pci_sysdata, fwnode should be able to
> > +	 * be queried from of/acpi.
> > +	 */
> > +	return NULL;
> > +}
> > +#define pci_root_bus_fwnode	_pci_root_bus_fwnode
> 
> Ugh.  pci_root_bus_fwnode() is another callback to find the
> irq_domain.  Only one call, from pci_host_bridge_msi_domain(), which
> itself is only called from pci_set_bus_msi_domain().  This feels like
> another case where we could simplify things by having the host bridge
> driver figure out the irq_domain explicitly when it creates the
> pci_host_bridge.  It seems like that's where we have the most
> information about how to find the irq_domain.

Urgh. This is a perfect copy paste of the x86 horror, warts and all.
I can't say I'm thrilled (another way to say "Gawd, Noes! Never!").

One thing I am sure of is that I do not want to add more custom
indirection to build the MSI topology. We barely got rid of the
msi_controller structure, and this is the same thing by another
name. Probably worse, actually.

In this case, I don't see the point in going via a fwnode indirection
given that there is no firmware tables the first place.

As for finding the irq domain from the host bridge, that's not doable
in most cases on arm64, as it is pretty likely that the host bridge
knows nothing about MSIs when they are implemented in the GIC (see my
recent msi_controller removal series that has a few patches about
that).

Having an optional callback to host bridges to obtain the MSI domain
may be possible in some cases though (there might be a chicken/egg
problem for some drivers though...).

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.