On 08/27/24 at 08:37am, John Paul Adrian Glaubitz wrote:

On Tue, 2024-08-27 at 14:22 +0800, Baoquan He wrote:

quoted

About why it's enabled by default, as Michael has explained in another
thread, distros usualy needs to enable it by default because vmcore
dumping is a very important feature on servers, even guest instances. 
Even though kdump codes are enabled to built in, not providing
crashkernel= value won't make vmcore dumping take effect, it won't cost
system resources in that case.

OK, thanks for the explanation. But as we have found out in the mean time,
the assumption was wrong to enable it by default for all architectures as
some architectures cannot boot a crash dump kernel with their default bootloader
but only through kexec.

Can we have a follow-up patch to disable crash dump kernels where they're
not needed? I mean, not every platform supported by Linux is obviously a
x86-based or POWER-based server.

Yes, while isn't Dave's patch a good one to fix it? In Dave's patch, the
default enabling of CRASH_DUMP has been taken off, change to rely on
ARCH_DEFAULT_CRASH_DUMP provided by each arch.

 config CRASH_DUMP
        bool "kernel crash dumps"
-       default y
+       default ARCH_DEFAULT_CRASH_DUMP