On fre, jan 24, 2025 at 13:39, Ilias Apalodimas ilias.apalodimas@linaro.org wrote:
Heinrich, Tobias
There's a slight problem that I forgot when commenting v2.
Heinrich's idea of plugging this into blkmap is eventually the right thing to do.
However, when I started coding this I only added the pmem memory as 'reserved' in the DT hoping that would work. Unfortunately, this depends on a kernel config option. I've managed to track down the problem here[0], but I haven't found time to test it properly and send it upstream. So for this feature to work reliably we *need* to remove the memory map we hand over to the OS.
Since using EFI memmap function into the blkmap code makes no sense, can we perhaps merge v2 (or a variant of it), which only targets EFI, with an explanation of *why* while I try to sort out the kernel issue?
I was not a part of the first two iterations of this series, but my view is basically this:
Adding some flag to memory backed slices of block maps, that the fdt-fixup code can use to know whether a pmem node should be injected or not, is completely fine by me.
What I am opposed to is adding restrictions on how block maps can be composed, i.e., limiting a block map to only contain either linear or memory mappings.