Re: [PATCH v6 06/30] PCI: Combine PCI domain and bus number in u32 arg

From: Bjorn Helgaas
Date: Thu Mar 12 2015 - 15:49:39 EST


On Thu, Mar 12, 2015 at 08:14:40PM +0800, Yijing Wang wrote:
> On 2015/3/12 9:29, Bjorn Helgaas wrote:
> > On Mon, Mar 09, 2015 at 10:34:03AM +0800, Yijing Wang wrote:
> >> Currently, we use int type for bus number in
> >> pci_create_root_bus(), pci_scan_root_bus() and
> >> pci_scan_bus_legacy. Because PCI bus number
> >> always <= 255, so we could change the bus number
> >> argument type to u32, and combine PCI domain and
> >> bus number in one.
> >
> > This makes no sense. Or rather, it only states the obvious: an 8-bit value
> > and a 16-bit value will both fit in a 32-bit value. But it doesn't say
> > *why* you think it's a good idea to pass a single value that contains both
> > domain and bus numbers. The obvious thing is to pass two separate values,
> > and you don't say why passing a single combined value is better.
>
> Sorry for my poor description for this patch, I combined the domain and bus, because
> I think now we have too many args at pci_scan_root_bus() or other scan functions,
>
> struct pci_bus *pci_scan_root_bus(struct device *parent, int bus,
> struct pci_ops *ops, void *sysdata, struct list_head *resources)
>
> Now we have five args yet, plus the new introduced domain and pci_host_bridge_ops,
> it will become 7.
>
> I thought introduced a new structure which contain the necessary info to scan root bus/ host bridge,
>
> E.g
>
> struct pci_scan_info {
> int bus;
> struct device *parent;
> struct pci_ops *ops;
> void *sysdata;
> struct list_head *resource;
> int domain;
> struct pci_host_bridge_ops;
> }
>
> Do you like this one or keep it like now ?
>
> pci_scan_root_bus(struct device *parent, int domain, int bus,
> struct pci_ops *ops, void *sysdata, struct list_head *resources, struct pci_host_bridge_ops *ops)

I don't think reducing the number of arguments is a good argument for
squashing some of them together.

I don't really want to add a structure like that because it adds management
complexity for all the callers because it contains per-bridge things (bus,
parent, domain, resource, sysdata). Things like struct pci_ops and struct
pci_host_bridge_ops are much simpler because drivers can statically
allocate a single copy and use it for multiple devices.

I think it might make sense to put the struct pci_ops pointer inside struct
pci_host_bridge_ops. That would get rid of one of the arguments.

You might also be able to get rid of the "bus" argument, since the caller
should be passing an IORESOURCE_BUS resource in the resource list, and
"bus" should be the same as res->start.

Bjorn
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/