Re: [PATCH 1/3] memblock: define functions to set the usable memory range

[Frank van der Linden <fllinden@amazon.com>]

On Thu, Jan 13, 2022 at 07:33:11PM +0200, Mike Rapoport wrote:
> On Tue, Jan 11, 2022 at 08:44:41PM +0000, Frank van der Linden wrote:
> > On Tue, Jan 11, 2022 at 12:31:58PM +0200, Mike Rapoport wrote:
> > > > --- a/include/linux/memblock.h
> > > > +++ b/include/linux/memblock.h
> > > > @@ -481,6 +481,8 @@ phys_addr_t memblock_reserved_size(void);
> > > >  phys_addr_t memblock_start_of_DRAM(void);
> > > >  phys_addr_t memblock_end_of_DRAM(void);
> > > >  void memblock_enforce_memory_limit(phys_addr_t memory_limit);
> > > > +void memblock_set_usable_range(phys_addr_t base, phys_addr_t size);
> > > > +void memblock_enforce_usable_range(void);
> > > >  void memblock_cap_memory_range(phys_addr_t base, phys_addr_t size);
> > > >  void memblock_mem_limit_remove_map(phys_addr_t limit);
> > >
> > > We already have 3 very similar interfaces that deal with memory capping.
> > > Now you suggest to add fourth that will "generically" solve a single use
> > > case of DT, EFI and kdump interaction on arm64.
> > >
> > > Looks like a workaround for a fundamental issue of incompatibility between
> > > DT and EFI wrt memory registration.
> >
> > Yep, I figured this would be the main argument against this - arm64
> > already added several other more-or-less special cased interfaces over
> > time.
> >
> > I'm more than happy to solve this in a different way.
> >
> > What would you suggest:
> >
> > 1) Try to merge the similar interfaces in to one.
> > 2) Just deal with it at a lower (arm64) level?
> > 3) Some other way?
> 
> We've discussed this with Ard on IRC, and our conclusion was that on arm64
> kdump kernel should have memblock.memory exactly the same as the normal
> kernel. Then, the memory outside usable-memory-range should be reserved so
> that kdump kernel won't step over it.
> 
> With that, simple (untested) patch below could be what we need:
> 
> diff --git a/drivers/of/fdt.c b/drivers/of/fdt.c
> index bdca35284ceb..371418dffaf1 100644
> --- a/drivers/of/fdt.c
> +++ b/drivers/of/fdt.c
> @@ -1275,7 +1275,8 @@ void __init early_init_dt_scan_nodes(void)
>         of_scan_flat_dt(early_init_dt_scan_memory, NULL);
> 
>         /* Handle linux,usable-memory-range property */
> -       memblock_cap_memory_range(cap_mem_addr, cap_mem_size);
> +       memblock_reserve(0, cap_mem_addr);
> +       memblock_reserve(cap_mem_addr + cap_mem_size, PHYS_ADDR_MAX);
>  }
> 
>  bool __init early_init_dt_scan(void *params)

Thanks for discussing this further! I tried the above change, although
I wrapped it in an if (cap_mem_size != 0), so that a normal kernel
doesn't get its entire memory marked as reserved.

The crash kernel hung without producing output - I haven't chased that
down. This particular kernel was a bit older (5.15), so I'll try again
with 5.17-rc to make sure.

- Frank