[hack] help U-Boot find the correct PCIe inbound offset for NVMe on Raspberry Pi 5

[Matthias Brugger]

On 04/03/2026 20:43, Pedro Falcato wrote:
> Hi Torsten,
> 
> On Mon, Mar 02, 2026 at 04:42:56PM +0100, Torsten Duwe wrote:
>> On Mon, 2 Mar 2026 13:59:40 +0100
>> Torsten Duwe <duwe at lst.de> wrote:
>>
>>> This proposed patch uses the already existent dev_phys_to_bus(),
>>> which can dig up the correct offset from associated DT nodes and
>>> subtract it.
>>
>> Just to make it clear, that patch is a necessary but not a sufficient
>> condition to boot the RPi5 from NVMe. The PCIe bus node above
>> the NVMe is generated dynamically has no DT node with dma-ranges and so
>> dev_phys_to_bus() still returns zero. In order to test NVMe on the
>> RPi5 you can either:
>>
> 
> Say I wanted to test NVMe booting on the pi5 (which I actually do).
> I would need:
> 1) this patch
> 2) one of the two BUS_ADDR() related diffs
> 3) the fixes posted in https://lore.kernel.org/u-boot/20251105163553.15F2A227AAC@verein.lst.de/
> 4) anything else?
> 

Just as a heads-up. If you interested to test this on openSUSE, you can install 
uboot-rpiarm64 and raspberrypi-firmware-dt package from this open build service 
project:
https://build.opensuse.org/project/show/home:mbrugger:branches:RPi5

That has both series on top of v2026.01 (plus other openSUSE stuff) and the DT 
patch.

Regards,
Matthias

> Thanks,
> Pedro
> 
>> hard code the 64GiB value when building your private U-Boot binary:
>>
>> -#define BUS_ADDR(a)         dev_phys_to_bus(dev->udev, (a))
>> +#define BUS_ADDR(a)            ((a)+0x1000000000LL)
>>
>> OR use this tricky DT change Andrea has prepared:
>>
>> --- a/dts/upstream/src/arm64/broadcom/bcm2712.dtsi
>> +++ b/dts/upstream/src/arm64/broadcom/bcm2712.dtsi
>> @@ -571,6 +571,15 @@
>>   				<0x03000000 0xff 0xfffff000 0x10 0x00131000 0x00 0x00001000>;
>>   			status = "disabled";
>> +
>> +			pci at 0,0 {
>> +				reg = <0x00 0x00 0x00 0x00 0x00>;
>> +				device_type = "pci";
>> +				#address-cells = <0x03>;
>> +				#size-cells = <0x02>;
>> +				ranges;
>> +				dma-ranges;
>> +			};
>>   		};
>>   
>>   		pcie2: pcie at 1000120000 {
>>
>> This pci at 0,0 node needs to be below the pcie at 1000110000 RC node so it
>> can "bridge" dev_phys_to_bus() to the RC node when it climbs up the
>> device tree and so helps it find the correct value.
>>
>> AFAICS the proper fix should be to add some dma-ranges property when
>> dynamic PCI devices are created during enumeration/scan or make
>> dev_phys_to_bus() skip those nodes for the lookup.
>>
>> 	Torsten