Re: [RFC PATCH v5 1/2] Use lock_device_hotplug() in cpu_probe_store() and cpu_release_store()

[Li Zhong <zhong@linux.vnet.ibm.com>]

On Thu, 2014-04-24 at 10:32 -0400, Tejun Heo wrote:
> Hello,
> 
> On Thu, Apr 24, 2014 at 04:37:23PM +0800, Li Zhong wrote:
> > On Wed, 2014-04-23 at 10:39 -0400, Tejun Heo wrote:
> > After thinking it harder, I still couldn't see ABBA here ... 
> > 
> > the active protection taken here is for "probe/release" which will not
> > be waited for removing something like "online" under cpu#? Or my
> > assumption that s_active for different files are different locks are
> > completely wrong? Or I missed something else? 
> 
> I'm probably confused about the locking.  I was thinking a scenario
> like the following.
> 
> A. CPU on/offline
> 
>    grabs s_active protection of online node
>    grabs cpu subsys mutex
>    perform on/offline
>    releases cpu subsys mutex
>    releases s_active protection of online node

so the chain here is s_active(online) -> cpu_subsys_mutex

> 
> B. CPU release
> 
>    grabs s_active protection of release node
>    grabs cpu subsys mutex
>    performs removal of the CPU
>    removes the online node
>    releases cpu subsys mutex
>    releases s_active protection of release node

and the chain here is s_active(release) -> cpu_subsys_mutex ->
s_active(online)

> 
> A nests cpu subsys mutex under s_active of the online node.  B nests
> s_active of the online node under the cpu subsys mutex.  What am I
> missing?

>From the above two chain, I think the problem is cpu_subsys_mutex and
s_active(online), which is the deadlock we are trying to solve in patch
#2. I seems to me s_active(release) here doesn't have lock issues? 

Thanks, Zhong

> 
> Thanks.
>