[PATCH 0/2] arm64: kexec_file_load vs memory reservations

[Marc Zyngier]

[- Bhupesh, as his RH address now bounces]

On Thu, 13 May 2021 04:17:38 +0100,
Dave Young <dyoung at redhat.com> wrote:
> 
> Hi Marc,
> On 05/12/21 at 07:04pm, Marc Zyngier wrote:
> > + Dave Young, which I accidentally missed in my initial post
> > 
> > On Thu, 29 Apr 2021 14:35:31 +0100,
> > Marc Zyngier <maz at kernel.org> wrote:
> > > 
> > > It recently became apparent that using kexec with kexec_file_load() on
> > > arm64 is pretty similar to playing Russian roulette.
> > > 
> > > Depending on the amount of memory, the HW supported and the firmware
> > > interface used, your secondary kernel may overwrite critical memory
> > > regions without which the secondary kernel cannot boot (the GICv3 LPI
> > > tables being a prime example of such reserved regions).
> > > 
> > > It turns out that there is at least two ways for reserved memory
> > > regions to be described to kexec: /proc/iomem for the userspace
> > > implementation, and memblock.reserved for kexec_file. And of course,
> > > our LPI tables are only reserved using the resource tree, leading to
> > > the aforementioned stamping. Similar things could happen with ACPI
> > > tables as well.
> > > 
> > > On my 24xA53 system artificially limited to 256MB of RAM (yes, it
> > > boots with that little memory), trying to kexec a secondary kernel
> > > failed every times. I can only presume that this was mostly tested
> > > using kdump, which preserves the entire kernel memory range.
> > > 
> > > This small series aims at triggering a discussion on what are the
> > > expectations for kexec_file, and whether we should unify the two
> > > reservation mechanisms.
> > > 
> > > And in the meantime, it gets things going...
> > > 
> > > Marc Zyngier (2):
> > >   firmware/efi: Tell memblock about EFI reservations
> > >   ACPI: arm64: Reserve the ACPI tables in memblock
> > > 
> > >  arch/arm64/kernel/acpi.c   |  1 +
> > >  drivers/firmware/efi/efi.c | 23 ++++++++++++++++++++++-
> > >  2 files changed, 23 insertions(+), 1 deletion(-)
> > 
> > Any comment on this?
> > 
> > I've separately started working on using the resource tree to slice
> > and dice the memblocks that are candidate for kexec_file_load(), but
> > I'd like some consensus on whether this is the right way to address
> > the issue.
> > 
> > Without something like this, kexec_file_load() is not usable on arm64,
> > so I'm pretty eager to see the back of this bug.
> 
> The arm64 memory reservation is tricky, I do not think I understand it
> correctly.  Previously there were a lot discussion, Ard and AKASHI
> should know more about it, see if they can provide comments.

I'll let them chime in. It looks like most of the discussions were
around kdump, which doesn't suffer from this issue (the memory is
reserved upfront).

> About the problem you see, another way is to just add an arch weak
> function like powerpc: arch_kexec_locate_mem_hole, and walking resource
> tree for kexec_file_load as well.  But I might be wrong since I did not
> follow up the arm64 specific history.

Right, this would avoid messing with the core code. However, the
problem remains in the sense that there is no clear definition of what
"reserved memory" is in general, and where it is described. For
example, x86 places the ACPI tables in reserved memblocks, while arm64
doesn't (unless we use my second patch).

To reliably use the resource tree, we need to ensure that it contains
all the reservations that appeared in memblock too. And if we can't
have a single reference, then we have to consider the union of the two
trees.

Thoughts?

	M.

-- 
Without deviation from the norm, progress is not possible.