Re: [PATCH] softlockup: keep current softlockup period before touching timestamp

[Ingo Molnar <mingo@elte.hu>]

* Jia Zhang <jia.zhang2008@gmail.com> wrote:

> From: Jia Zhang <jia.zhang2008@gmail.com>
> 
> Imaging such a scenario, softlockup deadline is 60s and a softirq runs 
> at most 65s. Assume that currently that softirq runs 50s. At this 
> moment, issue a "echo 40 > /proc/sys/kernel/softlockup_thresh" and 
> thus lead to recounting the timeout. The remaining time for that 
> softirq is 15s less than new threshold 40s, which means the warning 
> will be depressed after 15s. It is not what we expected because 
> softlockup actually feeds 100s(old+new) for that softirq. To fix this 
> problem, softlockup should keep current softlockup period before 
> touching timestamp.
> 
> Signed-off-by: Jia Zhang <jia.zhang2008@gmail.com>
> ---
>  include/linux/sched.h |    5 +----
>  kernel/softlockup.c   |   26 ++++++++++++++------------
>  kernel/sysctl.c       |    2 +-
>  3 files changed, 16 insertions(+), 17 deletions(-)

your patch does not apply to the latest kernel - mind redoing it against 
latest -tip:

  http://people.redhat.com/mingo/tip.git/README

also please re-post the other patch of yours as well, the "allow 
checking for remote CPUs" - that too looks like a worthwile improvement.

Thanks,

	Ingo