Re: [PATCH v21 8/8] Documentation: dt: usable-memory and elfcorehdr nodes for arm64 kexec

Am Donnerstag, 07 Juli 2016, 11:00:25 schrieb AKASHI Takahiro:

On Wed, Jul 06, 2016 at 04:29:18PM -0300, Thiago Jung Bauermann wrote:

quoted

Hi,

Am Mittwoch, 06 Juli 2016, 16:52:26 schrieb AKASHI Takahiro:

quoted

+linux,usable-memory
+-------------------
+
+This property is set on PowerPC and arm64 by kexec-tools during kdump
+to tell the crash kernel the base address of its reserved area of
memory, and +the size. e.g.
+
+/ {
+	chosen {
+		linux,usable-memory = <0x9 0xf0000000 0x0 0x10000000>;
+	};
+};

Again, this description is wrong for PowerPC. See messages from myself
and Michael Ellerman:

https://lists.infradead.org/pipermail/kexec/2016-June/016250.html

https://lists.infradead.org/pipermail/kexec/2016-June/016253.html

Oops, I must have missed your previous comments. Apologies.

No problem.

Yes, I know that, and I used to implement the same functionality before.
It did work for dtb-based systems, but not for UEFI(ACPI)-based systems
because UEFI doesn't export memory regions information via a device tree,
but rather via ACPI table. So "/memory" node won't appear.
So I went back with "mem=" command line approach, and later this
"/chosen/" approach.

Ah, I didn't realize there could be dtb and UEFI systems.

quoted

IMHO, it would be simpler if ARM used linux,usable-memory in the same
way
that PowerPC does, for consistency.

Well, this property won't conflict with per-"/memory" ones
if we take it that the former, if present, supersedes the latter.
Sophistic?
What about changing the name to usable-memory-limit?
(I know that you have another one, "memory-limit" though.)

Again, I would like to defer to arm64 maintainers.

My personal opinion is that having a property with a different name would be 
less confusing, but it's not a strong opinion.

I would suggest calling it usable-memory-range, but I'm fine with whatever 
is decided by the maintainers.

-- 
[]'s
Thiago Jung Bauermann
IBM Linux Technology Center