[PATCH 18/19] arm64: kdump: update a kernel doc

[Mark Rutland]

On Fri, Jan 22, 2016 at 03:23:14PM +0900, AKASHI Takahiro wrote:
> On 01/21/2016 09:02 PM, Mark Rutland wrote:
> >On Thu, Jan 21, 2016 at 03:53:42PM +0900, AKASHI Takahiro wrote:
> >>On 01/20/2016 08:49 PM, Mark Rutland wrote:
> >>>On Wed, Jan 20, 2016 at 03:07:53PM +0900, AKASHI Takahiro wrote:
> >>>>On 01/20/2016 11:49 AM, Dave Young wrote:
> >>>>>Firmware do not know kernel endianniess, kernel should respect firmware
> >>>>>maps and adapt to it, it sounds like a generic issue not specfic to kexec.
> >>>>
> >>>>On arm64, a kernel image header has a bit field to specify the image's endianness.
> >>>>Anyway, our current implementation replies on a user-supplied dtb to start BE kernel.
> >>>
> >>>The firmware should _never_ care about the kernel's endianness. The
> >>>bootlaoder or first kernel shouldn't care about the next kernel's
> >>>endianness apart from in exceptional circumstances. The DTB for a LE
> >>>kernel should look identical to that passed to a BE kernel.
> >>
> >>Please note that I didn't say anything different from your last two statements.
> >>The current arm64 kexec implementation doesn't do anything specific to BE,
> >>but as far as BE kernel doesn't support UEFI, users are responsible for
> >>providing a proper dtb.
> >
> >I'm just confused as to what you mean by a "proper dtb" in that case.
> >
> >If you just mean one with memory nodes hacked in, then that would
> >currently be a way to make that work, yes.
> 
> One of useful cases that I have in my mind is kdump.
> We may want to use a small sub-set of dtb, especially devices, to
> make the reboot more reliable. Device drivers are likely to be vulnerable
> at crash.

I don't think that we can reliably have userspace carve out devices from
the DTB or from ACPI tables in order to achieve that. That's going to
end up complex and/or incomplete. We also can't do this in the
kexec_load_file / Secure Boot case.

That's not to say we cannot try, as it's possible when using kexec_load.
However, it's only going to be possible on a subset of systems, and it
would probably make sense to reserve this approach to those cases we
cannot work around by other means (e.g. whitelisting "safe" devices in
the kdump kernel, forcing explicit resets, etc).

> >It seems like the better option is to fix the BE kernel to support a
> >UEFI memory map, as that solves other issues.
> 
> Why did Ard throw away his patch?

In the absence of kexec it wasn't necessary, it only supported a subset
of the runtime services (and no other features like DMI IIRC), and it
looked like it would be painful to debug (if something went wrong while
a CPU was in LE mode, we couldn't even panic()).

Given BE kernels on UEFI were never supported until that point, there
wasn't a compelling reason to support that case.

Even if we support the UEFI memory map, I don't think it's worth the
effort to support runtime services, ACPI, and related code that's only
ever been tested on LE. So realistically this would only work on systems
using UEFI && DT rather than UEFI && ACPI.

> So, are you now suggesting that we put both "elfcorehdr=" and
> "usable-memory=" under /chosen in dtb?

Yes.

> That's fair enough.  (as far as nobody cares about incompatibility
> with other archs.)

Glad to hear! :)

Thanks,
Mark.