Re: [PATCH v3 10/10] riscv: dts: eswin: add HiFive Premier P550 board device tree

From: Sjoerd Simons
Date: Tue Apr 15 2025 - 03:40:52 EST


Hey,

On Mon, 2025-04-14 at 11:00 -0500, Samuel Holland wrote:
> Hi Ariel,
>
> On 2025-04-14 7:55 AM, Ariel D'Alessandro wrote:
> > Hi Pinkesh,
> >
> > On 4/10/25 12:25 PM, Pinkesh Vaghela wrote:
> > > From: Min Lin <linmin@xxxxxxxxxxxxxxxxxx>
> >
<snip>

> > Although commit log says that this includes DRAM configuration, looks like
> > it's
> > missing? In order to test this patchset, had to add this following memory
> > definition (picked from vendor kernel repository):
> >
> >     L50: memory@80000000 {
> >             compatible = "sifive,axi4-mem-port", "sifive,axi4-port",
> > "sifive,mem-port";
> >             device_type = "memory";
> >             reg = <0x0 0x80000000 0x7f 0x80000000>;
> >             sifive,port-width-bytes = <32>;
> >     };
>
> That is a misstatement in the commit message. The memory node is not included
> in
> the static devicetree because the amount of RAM installed on the board is
> variable. It is the responsibility of firmware to provide the memory map,
> either
> through EFI or by patching the memory node into the DT at runtime. I believe
> the
> current BSP U-Boot does the former but not the latter.

Amount of RAM being variable is pretty common on devices using FDT these days;
Typically the dts still gets a memory node that's a reasonable default, with the
expectation that e.g. u-boot will fix it up. If you look at other risc-v
devicetrees in upstream they (almost?) all come with a pre-defined memory node.
For the P550 board a default memory node for 16G ram seems reasonable (as that
seems the minimal SKU?)

That all being said. Indeed the sifive BSP u-boot doesn't seem to call the
relevant `fdt_fixup_memory` to fixup the memory node, hence us having issues
booting with u-boot directly (without going through EFI). Honestly this was a
bit of a surprise to me as only most other architectures that's just done by
common code, but that doesn't seem to be the case for risc-v (either upstream or
downstream)

--
Sjoerd Simons
Collabora