Re: [PATCH 18/28] of: create default early_init_dt_add_memory_arch

[Catalin Marinas <catalin.marinas@arm.com>]

On Tue, Sep 17, 2013 at 02:01:36PM +0100, Rob Herring wrote:
> On Tue, Sep 17, 2013 at 3:46 AM, Catalin Marinas
> <catalin.marinas@arm.com> wrote:
> > On 17 Sep 2013, at 00:09, Rob Herring <robherring2@gmail.com> wrote:
> >> --- a/arch/arm64/kernel/setup.c
> >> +++ b/arch/arm64/kernel/setup.c
> >> @@ -147,24 +147,6 @@ static void __init setup_machine_fdt(phys_addr_t dt_phys)
> >>       pr_info("Machine: %s\n", machine_name);
> >> }
> >>
> >> -void __init early_init_dt_add_memory_arch(u64 base, u64 size)
> >> -{
> >> -     base &= PAGE_MASK;
> >> -     size &= PAGE_MASK;
> >> -     if (base + size < PHYS_OFFSET) {
> >> -             pr_warning("Ignoring memory block 0x%llx - 0x%llx\n",
> >> -                        base, base + size);
> >> -             return;
> >> -     }
> >> -     if (base < PHYS_OFFSET) {
> >> -             pr_warning("Ignoring memory range 0x%llx - 0x%llx\n",
> >> -                        base, PHYS_OFFSET);
> >> -             size -= PHYS_OFFSET - base;
> >> -             base = PHYS_OFFSET;
> >> -     }
> >> -     memblock_add(base, size);
> >> -}
> >> -
> >> /*
> >>  * Limit the memory size that was specified via FDT.
> >>  */
> >
> > ...
> >
> >> --- a/drivers/of/fdt.c
> >> +++ b/drivers/of/fdt.c
> >> @@ -688,6 +688,17 @@ u64 __init dt_mem_next_cell(int s, __be32 **cellp)
> >>       return of_read_number(p, s);
> >> }
> >>
> >> +void __init __weak early_init_dt_add_memory_arch(u64 base, u64 size)
> >> +{
> >> +#ifdef CONFIG_HAVE_MEMBLOCK
> >> +     base &= PAGE_MASK;
> >> +     size &= PAGE_MASK;
> >> +     memblock_add(base, size);
> >> +#else
> >> +     pr_err("%s: ignoring memory (%llx, %llx)\n", __func__, base, size);
> >> +#endif
> >> +}
> >
> > Are the arm64 changes equivalent here?  There are some safety checks to
> > cope with the kernel being loaded at a higher offset than the
> > recommended one (PHYS_OFFSET calculated automatically).
> 
> I tried to keep that, but PHYS_OFFSET is not universally defined. My
> reasoning is this range checking is hardly specific to an
> architecture. Perhaps if memory always starts at 0 you don't need it.
> If arm64 really needs these checks, then all architectures do.
> 
> Perhaps "__virt_to_phys(PAGE_OFFSET)" instead of PHYS_OFFSET would work for all?

I think virt_to_phys() or __pa() should work, the __virt_to_phys() is
only defined by a few architectures.

-- 
Catalin