Comment # 5 on bug 947825 from Takashi Iwai

(In reply to Petr Tesa����k from comment #4)
> Takashi, first, the "low" and "high" suffixes for crashkernel= parameter
> refer to DMA32 zone ("low", i.e. <=4G) and Normal zone ("high", i.e. >4G).

Thanks, I figured out by myself after the report, but I wasn't sure about the
terminology here, as this confused me.

> Second, the allocation is indeed much lower than in previous versions, but
> there is some logic behind the suggested number. If the logic misses some
> important aspects, I'll gladly assist in resolution.
> 
> Let me start with a general observation I made a couple of years ago, namely
> that memory required to save a dump does not depend on the total RAM.
> 
> I'm aware of some changes in systemd-udevd that cause OOM in SLE12 SP1. It's
> quite likely that they also affect Leap. Can you try adding
> "rd.udev.children-max=2" to KDUMP_COMMANDLINE_APPEND in /etc/sysconfig/kdump
> and try with the suggested value?

OK, if this tight size is intentional, I'm fine with this.  It's hard to tell
whether 111MB really works because kdump fails in anyway in a different point
on Leap (it didn't seem to mount the local fs).  More investigation needed.