Hi Wenjian, On 08/01/16 at 08:23am, Zhou Wenjian wrote: > v1->v2: change nr_cpus to maxcpus > > SMP dump-capture kernel is useful to improve the performance of kdump in > some cases. So add the description of bringing up SMP dump-capture kernel. > > Signed-off-by: Zhou Wenjian <zhouwj-fnst at cn.fujitsu.com> Discussed with people, it could be better to adjust the description about nr_cpus and maxcpus part. I think you can still describe nr_cpus/maxcpus in patch 1/2, and keep parallel dumping part in 2/2. Originally maxcpus=1 is used for all ARCHes. Later people found nr_cpus=1 is better since nr_cpus decides the number of possible cpu while maxcpus decides the max working cpu after system boot. So nr_cpus can save memory because percpu will pre-allocate memory for each possible cpu for hotplug. So on x86 nr_cpus is used because much memory can be saved if possible cpu number is very large. So you can mention that both maxcpus and nr_cpus can be used but nr_cpus has advantage if it has been implemented in some ARCHes like x86_64. And I guess you mush have tested parallel dumping feature with nr_cpus specified, it makes sense to tell people with the real situation. Thanks Baoquan > --- > Documentation/kdump/kdump.txt | 6 ++++++ > 1 file changed, 6 insertions(+) > > diff --git a/Documentation/kdump/kdump.txt b/Documentation/kdump/kdump.txt > index 88ff63d..c5762b7 100644 > --- a/Documentation/kdump/kdump.txt > +++ b/Documentation/kdump/kdump.txt > @@ -394,6 +394,12 @@ Notes on loading the dump-capture kernel: > dump. Hence generally it is useful either to build a UP dump-capture > kernel or specify maxcpus=1 option while loading dump-capture kernel. > > +* We should bring up a SMP dump-capture kernel if we intend to use multi-threads > + programs with it, such as parallel dump feature of makedumpfile. Otherwise, > + the multi-threads program may have a great performance degradation. > + To bring up a SMP dump-capture kernel, we should specify maxcpus=[X] and > + disable_cpu_apic=[Y] options while loading it. > + > * For s390x there are two kdump modes: If a ELF header is specified with > the elfcorehdr= kernel parameter, it is used by the kdump kernel as it > is done on all other architectures. If no elfcorehdr= kernel parameter is > -- > 1.8.3.1 > > >
