On Wed Aug 5, 2026 at 4:13 PM CDT, Mark Kettenis wrote:
> Randolph, Ilias,
>
> After commit 623f6c5b6ab7fa2 ("boot: image-fdt: free old dtb reservations")
> U-Boot now produces some nasty bogus error messages on Apple Silicon:
>
>
> U-Boot 2026.10-rc1-00115-gbaa64b2f8928 (Aug 05 2026 - 22:33:45 +0200)
>
> Model: Apple Mac Mini (M2 Pro, 2023)
> DRAM:  15.3 GiB
> ERROR: reserving fdt memory region failed (addr=103d7f4c000 size=854000 
> flags=2): -22
> ERROR: reserving fdt memory region failed (addr=103d6c4c000 size=1000000 
> flags=2): -22
> ERROR: reserving fdt memory region failed (addr=103d7c4c000 size=300000 
> flags=2): -22
> ERROR: reserving fdt memory region failed (addr=10000218000 size=604000 
> flags=2): -22
> ERROR: reserving fdt memory region failed (addr=103fff74000 size=4000 
> flags=2): -22
> ERROR: reserving fdt memory region failed (addr=103fff78000 size=40000 
> flags=2): -22
> ERROR: reserving fdt memory region failed (addr=103fffb8000 size=4000 
> flags=2): -22
>
>
> The issue here is that these memory regions have been reserved already
> by the previous boot stage and have already been excluded from the
> memory listed under the /memory node:
>
>     Node 0x22568
>         name: 'memory'
>         device_type: 'memory'
>         reg: 
> 00000100.011e0000.00000003.d2a34000.00000103.d3c18000.00000000.00004000
>
> These regions are listed under /reserved-memory since they are
> referenced by other nodes in the device tree.
>
> I suspect there are scenarios on other hardware platforms where this
> will happen as well.  Can this be fixed?

It seems that these messages being valid errors or not are still up for debate,
and varies per platform. Rockchip has a similar issue, where the reservations
are explicitly outside of the memory node. In these cases it's my opinion that
LMB should never be asked to reserve these regions.

LMB should only know about regions that overlap with the main memory node.
Masking these errors is still potentially a problem. Simon pointed out that in
the event of a partial overlap it generates the same -EINVAL error.

Reply via email to