Re: crashkernel only allowing smaller allocations as kernel version increases

From: Robert LeBlanc
Date: Thu Dec 08 2016 - 14:23:56 EST

OK, after looking through some code, it looks like the behavior was
change (back?) to only allocating under 896M with commit
55a20ee7804ab64ac90bcdd4e2868a42829e2784. In order to allocate more
RAM than that, you have to specify crashkernel=1024M,high which
reserves some RAM in low memory and the requested amount in high
memory. This is not explained in Documentation/kdump/kdump.txt. I'll
work on a pull request for Documentation.
Robert LeBlanc
PGP Fingerprint 79A2 9CA4 6CC4 45DD A904 C70E E654 3BB2 FA62 B9F1

On Thu, Dec 8, 2016 at 11:36 AM, Robert LeBlanc <robert@xxxxxxxxxxxxx> wrote:
> I've been trying to capture a core to help in debuging a problem, but
> it seems that as I go up in kernel versions, I can only allocate less
> and less memory to crashkernel and with 4.9-rc8 I can't allocate
> enough to prevent OOM (specifically on CentOS with iSCSI root). I'm
> seeing the same reduction of allocatable memory on both Debian and
> CentOS on the same hardware
> As an example:
> Debian
> 3.16 - 1 GB is OK
> 4.9-rc7 - ~512MB is OK, but gets "kernel version is not supported".
> CentOS
> 3.10 - 1 GB is OK
> 4.8 - ~800MB is OK
> 4.9-rc8 - ~512MB is OK, but OOMs
> CentOS 4.9 kernel was built with default config, Debian 4.9 was built
> by coping the distro 3.16.0-4 config and then running make menuconfig.
> Trying to specify memory amount above this causes kdump to report that
> no memory has been reserved. The hardware has 512GB of RAM. Am I
> missing some config option that limiting how much crashkernel can
> reserve?
> I also verified on my test desktop with 8GB of RAM that setting
> crashkernel=1024M also fails to allocate memory with 4.9-rc8, but
> works fine with 128M for instance.
> Thanks,
> ----------------
> Robert LeBlanc
> PGP Fingerprint 79A2 9CA4 6CC4 45DD A904 C70E E654 3BB2 FA62 B9F1