On Mon, Sep 15, 2025 at 05:25:23PM -0700, E Shattow wrote: > Hi Tom, Add CC: Heinrich > > On 9/15/25 07:32, Tom Rini wrote: > > On Mon, Sep 15, 2025 at 07:45:49AM +0800, Vivian Wang wrote: > > > >> Setting fdt_high to all ones is discouraged and does not appear to be > >> useful for RISC-V QEMU. Moreover, it causes a boot failure when the FDT > >> generated internally by QEMU is used while booting. Remove it to allow > >> U-Boot to pick a suitable address and relocate the FDT. > >> > >> Closes: > >> https://lore.kernel.org/u-boot/8397369a-9b0b-4798-9c30-3a8116565...@iscas.ac.cn > >> Signed-off-by: Vivian Wang <wangruik...@iscas.ac.cn> > > > > > I'm very glad to see this finally and would like to see the rest of the > > RISC-V platforms follow and do the same now. > > > > Reviewed-by: Tom Rini <tr...@konsulko.com> > > > > When you say you would like the rest of the RISC-V platforms (to) follow > and do the same; What do you mean? Heinrich and I were trying to > figure this out off of something you said on IRC awhile back. I was > taking a guess it was the qemu config (and this patch makes sense to me > so I think that's part of it), but what else? Is that a reference to > sifive-unleashed config fragment, are there others? > > If you give us a detailed hint what we should be looking at it would go > a long way and we can test / figure out the rest. Is initrd_high > all-ones on your radar too?
The simple answer is that no platforms should ever set fdt_high=0xff... nor initrd_high=0xff... by default. This tells U-Boot to never try and relocate in memory these parts of the boot even when U-Boot knows that they will overlap or are not correctly aligned. This in turn leads to harder to debug failures at run time. Unfortunately RISC-V platforms were introduced with these set and whatever checkpatch.pl check I had missed them at the time (aside, most recently checkpatch.pl was missing this in .env files, I fixed that today). -- Tom
signature.asc
Description: PGP signature
