On Tue, May 1, 2012 at 12:27 AM, Yinghai Lu <yinghai@xxxxxxxxxx> wrote:
> On Mon, Apr 30, 2012 at 9:02 PM, Bjorn Helgaas <bhelgaas@xxxxxxxxxx> wrote:
>> On Mon, Apr 30, 2012 at 6:32 PM, Yinghai Lu <yinghai@xxxxxxxxxx> wrote:
>>> On Mon, Apr 30, 2012 at 4:25 PM, Bjorn Helgaas <bhelgaas@xxxxxxxxxx> wrote:
>>>>> --- a/include/linux/pci.h
>>>>> +++ b/include/linux/pci.h
>>>>> @@ -419,6 +419,7 @@ struct pci_bus {
>>>>> struct list_head slots; /* list of slots on this bus */
>>>>> struct resource *resource[PCI_BRIDGE_RESOURCE_NUM];
>>>>> struct list_head resources; /* address space routed to this bus */
>>>>> + struct resource busn_res; /* track registered bus num range */
>>>>>
>>>>> struct pci_ops *ops; /* configuration access functions */
>>>>> void *sysdata; /* hook for sys-specific extension */
>>>>
>>>> struct pci_bus already includes "secondary" and "subordinate". This
>>>> new "busn_res" looks like it will contain the same information. Why
>>>> do we need both?
>>>
>>> In some case the could be different.
>>> for root bus from _CRS, busn_res could bigger than subordinate,
>>> because scan_childbus will update subordinate.
>>
>> For a bus below a P2P bridge, I think it's always the case that the
>> bridge's Subordinate Bus Number in config space == bus->subordinate ==
>> bus->busn_res.end (correct me if I'm wrong). I don't like the
>> redundancy in this case.
>
> there are about 70 bus->subordinate reference and 40 bus->secondary reference.
>
> could try to update them in following patch set.
If you're proposing this:
1. add bus->busn_res
2. remove bus->subordinate and bus->secondary
I fully support that, and I'd like to merge both pieces at the same
time (different patches is fine; I just want to make sure both pieces
actually happen).
>> For a root bus where you set bus->busn_res from _CRS and
>> bus->subordinate = pci_scan_child_bus(), bus->busn_res.end will
>> generally be different from bus->subordinate, but there's no point in
>> keeping track of bus->subordinate.
>>
>> The reason we care about secondary and subordinate is so we can
>> allocate bus numbers when enumerating devices behind a bridge. The
>> only thing we need for that is the aperture of the upstream bridge and
>> the apertures of any peer bridges on the same bus. Let's say we have
>> this:
>>
>> pci 00:00.0 bridge to [bus a-b]
>> pci a:01.0 bridge to [bus c-d] (already enumerated)
>> pci a:02.0 bridge to [bus e-f] (already enumerated)
>> pci a:03.0 bridge to [bus x-y] (enumerating now)
>>
>> We know [c-d] is contained in [a-b]; [e-f] is contained in [a-b]; a <
>> c; and a < e. To enumerate behind a:03.0, we need to assign x & y
>> such that a < x; [x-y] is contained in [a-b]; and [x-y] does not
>> overlap [c-d] or [e-f]. The value from pci_scan_child_bus() is
>> probably useful for setting y, but we don't have to save it in the
>> struct pci_bus for that.
>
> busn alloc will try to solve x-y may need big range than [a,b], it
> will extend top of b and parents of bus a.
> instead of just b+1 blindly.
>
> and will have more strict checking to avoid overlapping.
Obviously the completely general problem of allocating bus numbers may
require traversing up the tree. My point is that I don't think it's
necessary to keep both busn_res.end and subordinate to do that.
>>> and also we have one resource to insert it into the resource tree, so
>>> later could probe/allocate bus num range.
>>
>> Sorry, I didn't understand this.
>
> Using busn_res to track and allocate busn range, by put them in the
> resource tree could reuse resource allocating code.
Yes, I agree that replacing secondary & subordinate with a struct
resource is a good idea. That will allow a resource tree of bus
numbers, as well as other useful things like the ability to "%pR".
I just don't want *both* busn_resource and secondary & subordinate.
Bjorn
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
