[PATCH 0/2] Fix CPU Online handling for unbounded worker threads

[Gautham R. Shenoy]

Hi,

This patchset fixes a couple of issues in the CPU_ONLINE notification
handling for the workqueues with respect to unbounded worker threads.

Patch 1 ensures that the affinity of a unbound worker thread
associated with a node whose very first CPU has come online is set
correctly. In the existing code path we will never call
set_cpus_allowed_ptr() for unbound worker threads that have been
created on a CPU Online operation after boot.

Patch 2 fixes the following WARN_ON() reported by Abdul when
set_cpus_allowed_ptr() for an unbound worker thread is invoked when
only one of the CPUs in its cpumask is online but not yet active.

 ------------[ cut here ]------------
 WARNING: CPU: 40 PID: 248 at kernel/sched/core.c:1166 __set_cpus_allowed_ptr+0x21c/0x290
 Modules linked in:
CPU: 40 PID: 248 Comm: cpuhp/40 Not tainted 4.6.0-autotest #1
task: c000000f27284200 ti: c000000f273fc000 task.ti: c000000f273fc000
NIP: c00000000010488c LR: c000000000104874 CTR: 0000000000000000
REGS: c000000f273ff7d0 TRAP: 0700   Not tainted  (4.6.0-autotest)
MSR: 9000000100029033 <SF,HV,EE,ME,IR,DR,RI,LE,TM[E]>  CR: 28002804  XER: 20000000
CFAR: c0000000005b0888 SOFTE: 0
GPR00: c00000000010478c c000000f273ffa50 c0000000013ce400 0000000000000000
GPR04: c00000000140ed98 0000000000000800 c0000007f64d9408 0000000000000000
GPR08: 0000000000000000 0000000000000028 c00000000140ee90 0000000000000020
GPR12: 0000000000002200 c00000000fb96800 c0000000000f44a8 c0000007fa158480
GPR16: c0000007fc621a70 c000000f2721f800 0000000000000000 0000000000000001
GPR20: c000000001571ef0 0000000000000000 c00000000134879f c0000000012bc510
GPR24: 0000010000000000 0000000000000000 c00000000140ea98 c0000007f64d9408
GPR28: c0000007fbc21c00 ffffffffffffffea 0000000000000000 c000000f27280000
NIP [c00000000010488c] __set_cpus_allowed_ptr+0x21c/0x290
LR [c000000000104874] __set_cpus_allowed_ptr+0x204/0x290
Call Trace:
[c000000f273ffa50] [c00000000010478c] __set_cpus_allowed_ptr+0x11c/0x290 (unreliable)
[c000000f273ffac0] [c0000000000ed4b0] workqueue_cpu_up_callback+0x2c0/0x470
[c000000f273ffb70] [c0000000000f5c58] notifier_call_chain+0x98/0x100
[c000000f273ffbc0] [c0000000000c5ed0] __cpu_notify+0x70/0xe0
[c000000f273ffc00] [c0000000000c6028] notify_online+0x38/0x50
[c000000f273ffc30] [c0000000000c5214] cpuhp_invoke_callback+0x84/0x250
[c000000f273ffc90] [c0000000000c562c] cpuhp_up_callbacks+0x5c/0x120
[c000000f273ffce0] [c0000000000c64d4] cpuhp_thread_fun+0x184/0x1c0
[c000000f273ffd20] [c0000000000fa050] smpboot_thread_fn+0x290/0x2a0
[c000000f273ffd80] [c0000000000f45b0] kthread+0x110/0x130
[c000000f273ffe30] [c000000000009570] ret_from_kernel_thread+0x5c/0x6c
Instruction dump:
419eff3c 3d420004 38a00800 388a0998 7f63db78 484abfa1 60000000 2fa30000
409eff1c 813f0378 2f890001 419eff10 <0fe00000> 4bffff08 60000000 60000000
 ---[ end trace cbc1c5cfbc9591d0 ]---

The patches are based on 4.7-rc2. I have tested the patches on a
multi-node x86_64 and a ppc64

Gautham R. Shenoy (2):
  workqueue: Move wq_update_unbound_numa() to the beginning of
    CPU_ONLINE
  workqueue:Fix affinity of an unbound worker of a node with 1 online
    CPU

 kernel/workqueue.c | 27 +++++++++++++++++++--------
 1 file changed, 19 insertions(+), 8 deletions(-)

-- 
1.9.3