Re: offlining cpus breakage

Hi,

On Wednesday 07 January 2015 03:07 PM, Alexey Kardashevskiy wrote:

Hi!

"ppc64_cpu --smt=off" produces multiple error on the latest upstream kernel
(sha1 bdec419):

NMI watchdog: BUG: soft lockup - CPU#20 stuck for 23s! [swapper/20:0]

or

INFO: rcu_sched detected stalls on CPUs/tasks: { 2 7 8 9 10 11 12 13 14 15
16 17 18 19 20 21 22 23 2
4 25 26 27 28 29 30 31} (detected by 6, t=2102 jiffies, g=1617, c=1616,
q=1441)

and many others, all about lockups

I did bisecting and found out that reverting these helps:

77b54e9f213f76a23736940cf94bcd765fc00f40 powernv/powerpc: Add winkle
support for offline cpus
7cba160ad789a3ad7e68b92bf20eaad6ed171f80 powernv/cpuidle: Redesign idle
states management
8eb8ac89a364305d05ad16be983b7890eb462cc3 powerpc/powernv: Enable Offline
CPUs to enter deep idle states

btw reverting just two of them produces a compile error.

It is pseries_le_defconfig, POWER8 machine:
timebase        : 512000000
platform        : PowerNV
model           : palmetto
machine         : PowerNV palmetto
firmware        : OPAL v3


Please help to fix it. Thanks.

Upon investigation, we figured that the cpu is stuck in cpu_idle_poll
loop in kernel/sched/idle.c leading us to believe the bug is in timer
offload framework which fastsleep uses. Preeti and I are working on a
fix. We'll post it out as soon as possible.

Thanks,
Shreyas