Re: [PATCH] ARM: dts: BCM5301X: drop extra AXI bus ranges that break PCIe

On Wed, Jul 29, 2026, at 22:55, Rosen Penev wrote:
> On Wed, Jul 29, 2026 at 1:43 PM Arnd Bergmann <[email protected]> wrote:
>> On Wed, Jul 29, 2026, at 22:26, Rosen Penev wrote:
>> > On Wed, Jul 29, 2026 at 12:31 PM Arnd Bergmann <[email protected]> wrote:
>>
>> Sure, but my point is that it fixes a regression introduced by
>> 767012397976 ("ARM: dts: BCM5301X: Describe PCIe controllers fully"),
>> which tried to address a warning about missing ranges in dts.
>>
>> As far as I can tell, the patch worked correctly on the
>> platforms that had the right windows set (presumably bcm47094/ac56u)
>> but failed when the information was wrong.
> Not quite. Full information here: https://github.com/openwrt/openwrt/pull/24415
>
> TL;DR: fails on both platforms for different reasons.

Right, but the pci host bridge driver patch does address the case
where the dts file describes the correct window, by just using the
the window from ROM as before, but it now silently ignores incorrect
DT data, which is probably worse than failing the boot, since that
makes it harder to debug when it goes wrong the next time.

I can see multiple options for addressing this:

- add a warning for any incorrect dtb and fix up the dts source
  data by board

- add a warning for any incorrect dtb and fix up the windows
  in the boot loader on the live dtb that gets passed to the
  kernel.
  
- assume the dts is correct and reprogram the outbound windows
  (assuming there is a register level interface for that), ignoring
  the ROM data

- revert the 767012397976 ("ARM: dts: BCM5301X: Describe PCIe
  controllers fully") patch and add an exception to the pci
  host bridge binding so that this controller does not need
  to describe the windows

     Arnd