Re: Booting the kernel on very large NUMA systems
On 12/29/2013 08:41 PM, Ben Hutchings wrote:
>> "/sgiroot splash=silent showopts stop_machine.lazy=1 add_efi_memmap
>> nortsched processor.max_cstate=1 nobau log_buf_len=8M kdb=on
>> cgroup_disable=memory earlyprintk=ttyS0,115200n8 pcie_aspm=on nohz=off
>> crashkernel=512M intel_iommu=off init=/sbin/bootcpuset
>
> This is an interesting init command. Does it set CPU affinity and then
> invoke the 'real' init?
Yes, according to the documentation it seems to do exactly that [1].
>> Any idea?
> [...]
>
> I know that stop_machine used to be very slow indeed on large systems,
> and was used in too many places, and both of these problems have been
> addressed over the last few years. Have you tested with a more recent
> kernel version from unstable?
Last kernel I tested was 3.11. I am testing with 3.12-1-amd64 from
unstable now which I could install it by temporarily disabling NUMAlink.
So far, the machine seems stuck again, unfortunately. There also
is no output on the serial console whatsoever which puzzles me since
last time I could at least see the message about the kernel being
decompressed.
I could maybe try a vanilla 3.13 kernel and one with the SuSE
stop_machine patch applied.
Adrian
> [1]
http://techpubs.sgi.com/library/tpl/cgi-bin/getdoc.cgi?coll=linux&db=man&fname=/usr/share/catman/man8/bootcpuset.8.html
--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer - glaubitz@debian.org
`. `' Freie Universitaet Berlin - glaubitz@physik.fu-berlin.de
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913
Reply to: