[PATCH v3 2/5] arm64: use fixmap region for permanent FDT mapping

[Mark Rutland]

> >> +NOTE: versions prior to v4.1 require, in addition to the requirements
> >> +listed above, that the dtb be placed above the kernel Image inside the
> >> +same naturally aligned 512 MB region.
> >
> > Minor nit: This would read a little better if we just said "versions
> > prior to v4.1 also require that ...", dropping "in addition to the
> > requirements listed above".
> >
> 
> OK
> 
> Actually, I realized that this is incorrect anyway: it is not the same
> naturally aligned 512 MB region, it is actually 'within 512 MB of the
> start of the kernel Image' since that itself is naturally aligned to
> 512 MB in the virtual address space.

Surely starting at TEXT_OFFSET bytes below the kernel image? ;)

> >> +static void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
> >> +{
> >> +     const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
> >> +     phys_addr_t dt_phys_base = dt_phys & ~(FIX_FDT_SIZE - 1);
> >> +
> >> +     /*
> >> +      * Make sure that the FDT region can be mapped without the need to
> >> +      * allocate additional translation table pages, so that it is safe
> >> +      * to call create_pgd_mapping() this early.
> >> +      * On 4k pages, we'll use section mappings for the region so we
> >> +      * only have to be in the same PUD as the rest of the fixmap.
> >> +      * On 64k pages, we need to be in the same PMD as well, as the region
> >> +      * will be mapped using PTEs.
> >> +      */
> >> +     BUILD_BUG_ON(dt_virt_base & (SZ_2M - 1));
> >
> > s/SZ_2M/FIX_FDT_SIZE/
> >
> > Perhaps we should also have:
> >
> > BUILD_BUG_ON_NOT_POWER_OF_2(FIX_FDT_SIZE);
> >
> > ... given it's necessary for the address masking we do here.
> >
> > [...]
> >
> 
> I was going to suggest to use SZ_2M for the masking, but map a 4 MB
> window. That way, we can relax the requirement from "should not cross
> a 2 MB boundary' to 'should not exceed 2 MB in size', which is
> arguable easier to adhere to, since the latter is a property of the
> DTB itself, whereas the former is a property of whatever your malloc()
> gives you. That means the mask would remain SZ_2M, and the size would
> become SZ_4M

That sounds good so long as it doesn't get in the way of increasing
FIX_FDT_SIZE in future.

> > I've tested this on my Juno (atop of arm64 for-next/core), and all seems
> > well, so with those points addressed:
> >
> > Reviewed-by: Mark Rutland <mark.rutland at arm.com>
> > Tested-by: Mark Rutland <mark.rutland at arm.com>
> 
> Thanks, but this code will likely look quite different if we are going
> to cater for moving the kernel out of the linear range, so by then I
> probably won't be able to retain these tags.

Ok. I guess you'll merge the two series together?

Thanks,
Mark.