[PATCH 1/1] kernel/crash_core.c - Add crashkernel=auto for x86 and ARM

[Guilherme Piccoli]

Hi Saeed, thanks for your patch/idea! Comments inline, below.

On Wed, Nov 18, 2020 at 8:29 PM Saeed Mirzamohammadi
<saeed.mirzamohammadi at oracle.com> wrote:
>
> This adds crashkernel=auto feature to configure reserved memory for
> vmcore creation to both x86 and ARM platforms based on the total memory
> size.
>
> Cc: stable at vger.kernel.org
> Signed-off-by: John Donnelly <john.p.donnelly at oracle.com>
> Signed-off-by: Saeed Mirzamohammadi <saeed.mirzamohammadi at oracle.com>
> ---
>  Documentation/admin-guide/kdump/kdump.rst |  5 +++++
>  arch/arm64/Kconfig                        | 26 ++++++++++++++++++++++-
>  arch/arm64/configs/defconfig              |  1 +
>  arch/x86/Kconfig                          | 26 ++++++++++++++++++++++-
>  arch/x86/configs/x86_64_defconfig         |  1 +
>  kernel/crash_core.c                       | 20 +++++++++++++++--
>  6 files changed, 75 insertions(+), 4 deletions(-)
>
> diff --git a/Documentation/admin-guide/kdump/kdump.rst b/Documentation/admin-guide/kdump/kdump.rst
> index 75a9dd98e76e..f95a2af64f59 100644
> --- a/Documentation/admin-guide/kdump/kdump.rst
> +++ b/Documentation/admin-guide/kdump/kdump.rst
> @@ -285,7 +285,12 @@ This would mean:
>      2) if the RAM size is between 512M and 2G (exclusive), then reserve 64M
>      3) if the RAM size is larger than 2G, then reserve 128M
>
> +Or you can use crashkernel=auto if you have enough memory. The threshold
> +is 1G on x86_64 and arm64. If your system memory is less than the threshold,
> +crashkernel=auto will not reserve memory. The size changes according to
> +the system memory size like below:
>
> +    x86_64/arm64: 1G-64G:128M,64G-1T:256M,1T-:512M

As mentioned in the thread, this was tried before and never got merged
- I'm not sure the all the reasons, but I speculate that a stronger
reason is that it'd likely fail in many cases. I've seen cases of 256G
servers that require crashkernel=600M (or more), due to the amount of
devices. Also, the minimum nowadays would likely be 96M or more - I'm
looping Cascardo and Dann (Debian/Ubuntu maintainers of kdump stuff)
so they maybe can jump in with even more examples/considerations.

What we've been trying to do in Ubuntu/Debian is using an estimator
approach [0] - this is purely userspace and tries to infer the amount
of necessary memory a kdump minimal[1] kernel would take. I'm not
-1'ing your approach totally, but I think a bit more consideration is
needed in the ranges, at least accounting the number of devices of the
machine or something like that.

Cheers,


Guilherme

[0] https://salsa.debian.org/debian/makedumpfile/-/merge_requests/7
[1] Minimal as having a reduced initrd + "shrinking" parameters (like
nr_cpus=1).