Re: [PATCH 10/11] PCI: Lower bound bridge window alignment

["Ilpo Järvinen" <ilpo.jarvinen@linux.intel.com>]

[-- Attachment #1: Type: text/plain, Size: 1737 bytes --]

On Wed, 29 Apr 2026, Ilpo Järvinen wrote:

> pci_resource_alignment() does not consider bridge windows special,
> yet their alignment is subject to different requirements from BAR
> alignment.
> 
> Add lower bound to bridge window alignment to help callers out to
> always have large enough alignment.
> 
> Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>

Hi Bjorn,

Could you please pull this change and the subsequent one, the commits in 
pci/resource:

ae09d28ecbbc ("PCI: Lower bound bridge window alignment")
cf996b886675 ("PCI: Return valid alignment for assigned resources")

The rest of the changes in this series seem okay and can proceed if you're 
okay with keeping a partial series.

The reason for this request is the dev->bus vs dev->subordinate issue 
sashiko mentioned in its review.

I've tried to come up with a solution to that but it has become so ugly
I have not been very happy with it. Maybe there's no other way but the 
problem boils down with pci_min_window_alignment() having to be capable of 
dealing two cases, each lacking one of the key pointers (bus or bridge 
dev):

1) root bus without bus->self
2) bridge without a subordinate bus struct (if subordinate bus' alloc 
   failed)

Only way to solve that I could think of is passing both the bus and the 
bridge device to pci_min_window_alignment() and the related arch side 
function pcibios_window_alignment(). The ugliest parts then involve 
getting a bus pointer compatible with both of those cases, like this:

	struct pci_controller *phb = pci_bus_to_host(bus ? bus : bridge->bus);

But it's too late in the cycle now to try even that, IMO, so better to 
wait to the next cycle.

-- 
 i.