[Top] [All Lists]

[xfs-masters] Re: Linux 2.6.17-rc2 - notifier chain problem?

To: Andrew Morton <akpm@xxxxxxxx>, torvalds@xxxxxxxx
Subject: [xfs-masters] Re: Linux 2.6.17-rc2 - notifier chain problem?
From: Chandra Seetharaman <sekharan@xxxxxxxxxx>
Date: Fri, 28 Apr 2006 16:12:00 -0700
Cc: Ashok Raj <ashok.raj@xxxxxxxxx>, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx>, herbert@xxxxxxxxxxxx, linux-kernel@xxxxxxxxxxxxxxx, linux-xfs@xxxxxxxxxxx, xfs-masters@xxxxxxxxxxx
In-reply-to: <20060426132644.A31761@xxxxxxxxxxxxxxxxxxxx>
Organization: IBM
References: <Pine.LNX.4.44L0.0604261144010.6376-100000@xxxxxxxxxxxxxxxxxxxx> <1146075534.24650.11.camel@linuxchandra> <20060426114348.51e8e978.akpm@xxxxxxxx> <20060426122926.A31482@xxxxxxxxxxxxxxxxxxxx> <1146082893.24650.27.camel@linuxchandra> <20060426132644.A31761@xxxxxxxxxxxxxxxxxxxx>
Reply-to: xfs-masters@xxxxxxxxxxx
Sender: xfs-masters-bounce@xxxxxxxxxxx
On Wed, 2006-04-26 at 13:26 -0700, Ashok Raj wrote:

Hi All,

Looks like the patches I provided is a step backward from where Ashok &
Andrew were taking the register_cpu_notifier stuff to.

After some discussions with Ashok we both think the following would be
the right direction:
        1 revert the changes i pushed recently
        2 make all usages of register_cpu_notifier to be _init and 
          __initdata (if hotplug cpu is defined these are removed)
        3 export the symbols register_cpu_notifier and
          unregister_cpu_notifier only in CONFIG_HOTPLUG_CPU is defined
        4 move the hot plug cpu based usages of register_cpu_notifier
          inside #ifdef CONFIG_HOTPLUF_CPU(like xfs's usage).

I have few questions:
 - any problems with the above direction (mainly 3) ?
 - Should we proceed in this direction ?
 - is it too late for 2.6.17 ? if not late how much time do we have ?
Many thanks to Alan for bringing up the issue.



> On Wed, Apr 26, 2006 at 01:21:33PM -0700, Chandra Seetharaman wrote:
> > > 
> > > The problem we ran into was some of the startup code depends on the 
> > > notifier
> > > call chain for smp bringup, hence we couldn't nuke it similar to 
> > > hotcpu_notifier().
> > 
> > I do not understand the problem. If everybody that uses
> > register_cpu_notifier() starts using __cpuinit and __cpuinitdata (or the
> > devinit siblings), then the notifier mechanism will not be any different
> > than what they are now, right ? (both in hotplug cpu and non-hotplug cpu
> > case) Or am i missing something ?
> Well, register_cpu_notifier() is an exported function. There are several 
> modules that use this today like cpufreq etc which disqualifies it to be
> a init style function.
> either that function should be devinit and be present premanently, or
> should be mapped to null macro for correctness.
> Otherwise module loaders will start to oops when they call into 
> register.

    Chandra Seetharaman               | Be careful what you choose....
              - sekharan@xxxxxxxxxx   |      .......you may get it.

<Prev in Thread] Current Thread [Next in Thread>