Re: [PATCH] ARM: bcm2835_defconfig: add minimal support for Raspberry Pi4
From: Florian Fainelli
Date: Mon Feb 17 2020 - 16:44:19 EST
On 2/17/2020 12:18 PM, Nicolas Saenz Julienne wrote:
> [ Adding Florian to the coversation ]
> On Mon, 2020-02-17 at 13:22 +0100, Marek Szyprowski wrote:
>> Hi Nicolas,
>> On 14.02.2020 16:14, Nicolas Saenz Julienne wrote:
>>> IMO bcm2711_defconfig if the last resort solution. I don't think you can
>>> do bcm2835_lpae_defconfig as RPi and RPi2 SoCs don't support LPAE.
>> Okay, if you want I can send a patch adding bcm2711_defconfig.
>>> An intemediate solution is being discussed here:
>> Right, I also agree that multi_v7_lpae_defconfig is needed. Best would
>> be to have both (bcm2711 for quick tests of board-dedicated kernel and
>> multi for distributions).
> So I understand you'd be creating a new bcm2711_defconfig based on
> bcm2835_defconfig plus whatever is needed. Sounds OK to me. It'd be nice to
> have a small kernel config to do bisects with.
> Any comments Florian, Stefan?
If we can make bcm2711_defconfig a fragment that applies to
bcm2835_defconfig then we are not maintaining a completely new
configuration file and we take advantage of all existing coverage from
bcm2835_defconfig. A completely new bcm2711_defconfig would be hard to
justify IMHO when multi_v7_lpae_defconfig is sort of what we would prefer.
BTW, if you register the PCI outbound window as part of 2711's machine
descriptor map_io callback, you should have it trickled down from
iotable_init() -> create_mapping() -> __create_mapping() ->
create_36bit_mapping which should allow the creation of such a mapping
into the 32-bit virtual address space of the kernel. You would not quite
be able to use the entire 4GB of DRAM in such a configuration because
your virtual address space already needs ~41MB of register space + 64MB
of PCIe outbound space but at least bcm2835_defconfig would keep working.
NB: this only works AFAICT if you do this at map_io() time, not sure if
ioremap() will accept a >= 4GB physical address.