Hello. On 05/17/2012 08:13 PM, David Daney wrote:
From: Yong Zhang<yong.zhang@xxxxxxxxxxxxx>
Since commit 5fbd036b [sched: Cleanup cpu_active madness] and commit 2baab4e9 [sched: Fix select_fallback_rq() vs cpu_active/cpu_online], it's more safe to put notify_cpu_starting() and set_cpu_online() with irq disabled, otherwise we will have a typical race condition which above two commits try to resolve:
CPU1 CPU2 __cpu_up(); mp_ops->boot_secondary(); start_secondary(); ->init_secondary(); local_irq_enable(); <IRQ> do something; wake up softirqd; try_to_wake_up(); select_fallback_rq(); /* select wrong cpu */ set_cpu_online();
This patchset fix the above issue as well as set_cpu_online is called on the caller cpu.
BTW, I'm only running it on Cavium board because of limited source, so if anyone is interested to test it on other board, that's great :)
Signed-off-by: Yong Zhang<yong.zhang0@xxxxxxxxx>
Yong Zhang (8): MIPS: Octeon: delay enable irq to ->smp_finish() MIPS: BMIPS: delay irq enable to ->smp_finish() MIPS: SMTC: delay irq enable to ->smp_finish() MIPS: Yosemite: delay irq enable to ->smp_finish() MIPS: call ->smp_finish() a little late MIPS: call set_cpu_online() on the uping cpu with irq disabled MIPS: smp: Warn on too early irq enable MIPS: sync-r4k: remove redundant irq operation
This entire patch set (modulo the change log grammar items noted by Sergei):
I noted not only change log grammar, also comment grammar. And a missing summary in commit reference. :-)
Acked-by: David Daney <david.daney@xxxxxxxxxx>
WBR, Sergei
