[PATCH v24 9/9] Documentation: dt: chosen properties for arm64 kdump

[AKASHI Takahiro]

On Wed, Aug 31, 2016 at 08:45:46AM +0900, AKASHI Takahiro wrote:
> Rob,
> 
> On Tue, Aug 30, 2016 at 11:34:10AM -0500, Rob Herring wrote:
> > On Sun, Aug 21, 2016 at 11:28 PM, AKASHI Takahiro
> > <takahiro.akashi at linaro.org> wrote:
> > > Rob,
> > > [Cc: Mark]
> > >
> > > On Fri, Aug 19, 2016 at 08:26:41AM -0500, Rob Herring wrote:
> > >> On Tue, Aug 09, 2016 at 10:57:47AM +0900, AKASHI Takahiro wrote:
> > >> > From: James Morse <james.morse at arm.com>
> > >> >
> > >> > Add documentation for
> > >> >     linux,crashkernel-base and crashkernel-size,
> > >> >     linux,usable-memory-range, and
> > >> >     linux,elfcorehdr
> > >> > used by arm64 kexec/kdump to decribe the kdump reserved area, and
> > >> > the elfcorehdr's location within it.
> > >> >
> > >> > Signed-off-by: James Morse <james.morse at arm.com>
> > >> > [takahiro.akashi at linaro.org:
> > >> >     renamed "usable-memory" to "usable-memory-range",
> > >> >     added "linux,crashkernel-base" and "-size" ]
> > >> > Signed-off-by: AKASHI Takahiro <takahiro.akashi at linaro.org>
> > >> > ---
> > >> >  Documentation/devicetree/bindings/chosen.txt | 50 ++++++++++++++++++++++++++++
> > >> >  1 file changed, 50 insertions(+)
> > >> >
> > >> > diff --git a/Documentation/devicetree/bindings/chosen.txt b/Documentation/devicetree/bindings/chosen.txt
> > >> > index 6ae9d82..236188a 100644
> > >> > --- a/Documentation/devicetree/bindings/chosen.txt
> > >> > +++ b/Documentation/devicetree/bindings/chosen.txt
> > >> > @@ -52,3 +52,53 @@ This property is set (currently only on PowerPC, and only needed on
> > >> >  book3e) by some versions of kexec-tools to tell the new kernel that it
> > >> >  is being booted by kexec, as the booting environment may differ (e.g.
> > >> >  a different secondary CPU release mechanism)
> > >> > +
> > >> > +linux,crashkernel-base
> > >> > +linux,crashkernel-size
> > >> > +----------------------
> > >> > +These properties (currently used on PowerPC and arm64) indicates
> > >> > +the base address and the size, respectively, of the reserved memory
> > >> > +range for crash dump kernel.
> > >> > +e.g.
> > >> > +
> > >> > +/ {
> > >> > +   chosen {
> > >> > +           linux,crashkernel-base = <0x9 0xf0000000>;
> > >> > +           linux,crashkernel-size = <0x0 0x10000000>;
> > >> > +   };
> > >> > +};
> > >> > +
> > >> > +linux,usable-memory-range
> > >> > +-------------------------
> > >> > +
> > >> > +This property (currently used only on arm64) holds the memory range,
> > >> > +the base address and the size, which can be used as system ram on
> > >> > +the *current* kernel. Note that, if this property is present, any memory
> > >> > +regions under "memory" nodes in DT blob or ones marked as "conventional
> > >> > +memory" in EFI memory map should be ignored.
> > >> > +e.g.
> > >> > +
> > >> > +/ {
> > >> > +   chosen {
> > >> > +           linux,usable-memory-range = <0x9 0xf0000000 0x0 0x10000000>;
> > >> > +   };
> > >> > +};
> > >>
> > >> I've read the discussion on this. I think you should use the existing
> > >> linux,usable-memory property in the memory nodes. If UEFI systems don't
> > >> have memory nodes, then you can find an UEFI way to describe this. DT is
> > >> not the dumping ground for what doesn't fit in UEFI. How do x86 systems
> > >> work?
> > >
> > > Kdump for x86 passes the range of usable memory to crash dump kernel
> > > either via:
> > > (a) "memmap=nn at ss" command line parameter, or
> > > (b) faked BIOS e820 map (a sort of memory map table)
> > >
> > > (a) was rejected by Mark [1], and (b) is x86-specific.
> > >
> > > I believe that my approach is better because it works in the same way
> > > either on UEFI systems or DT-based systems.
> > 
> > So we have a new way for both UEFI and DT. If UEFI folks can't come up
> > with a standard way to do things in the UEFI standard, then we just
> > dump them into DT?
> 
> No, I don't think so.
> According to "Documentation/devicetree/bindings/chosen.txt,
>   "The chosen node does not represent a real device, but serves as a place
>    for passing data between firmware and the operating system, like boot
>    arguments."
> 
> Since kexec/dump is some sort of boot loader, it all makes sense to
> add a new property as an interface from the old kernel(kexec/dump)
> to the new kernel. So my approach doesn't break any DT rules.
> 
> Please note that, even on UEFI/APCI systems, there are already a few
> properties used under /chosen, like "linux,uefi-system-table" and 
> "linux,mmap-*."
> (Those properties are currently *not* defined in this document, though.)
> 
> > I'd rather see alignment with existing DT systems
> > (PPC) and in particular the tools.
> 
> I'm not sure what you mean here, but I guess:
> 1) On DT-only systems, we should follow the way that PPC does.
>    (That is, adding "usable-memory" property to each "memory" node.)
> 2) On UEFI/ACPI system, we must invent a new own way for our purpose
>    because, as I said before, there is no standard, but x86-specific way. 
> 
> Is this what you want to do?
> 
> If so,
> do you agree to my approach of adding "linux,usable-memory-range" property
> to /chosen *for UEFI/ACPI use*?

I've got another idea here: using "/memreserve/" field.
All the memory regions used by the crashed kernel would be added
as "/memreserve/" and be memblock_reserve'd later in crash dump kernel.
Consequently, they will be excluded from the usable memory.

Obviously,
- this approach will work both on UEFI/ACPI and DT-only systems, and
- patch#2 ("memblock: add memblock_cap_memory_range()"), as well as
  "linux,usable-memory-range" property, is no longer necessary.
  (So nothing new, except for "linux.elfcorehdr," will be invented.)

But on the other hand,
- arm64 and ppc do the different ways even though both are DT systems,
- those reserved regions are still *in* linear mapping and appear as
  "System RAM" in /proc/iomem, and
- this may imply that we cannot do kexec on crash dump kernel.

-Takahiro AKASHI


> Thanks,
> -Takahiro AKASHI
> 
> > Rob