Re: [PATCH] cpu: rid cpu_hotplug_disabled check for cpu_down()

[li guang]

å 2013-04-29äç 10:29 +0530ïSrivatsa S. Bhatåéï
> On 04/29/2013 10:12 AM, li guang wrote:
> > å 2013-04-29äç 10:00 +0530ïSrivatsa S. Bhatåéï
> >> On 04/29/2013 08:19 AM, liguang wrote:
> >>> in cpu_down(), _cpu_down() will do
> >>> "
> >>>         if (num_online_cpus() == 1)
> >>>                  return -EBUSY;
> >>> "
> >>> when cpu_hotplug_disabled was set, num_online_cpus
> >>> will return 1 for there's only 1 boot cpu.
> >>> so, it's unnecessary to check cpu_hotplug_disabled
> >>> here.
> >>>
> >>
> >> The 2 checks serve very different purposes; they are not the same!
> > 
> > purposes are different, but I think effects are same for this case,
> > the statement 'if ((num_online_cpus() == 1)' seems
> > have same effect with cpu_hotplug_disabled here,
> > because when cpu_hotplug_disabled, only boot cpu is online
> > 
> 
> Why do you keep saying that? They are *two* *different* checks!
> Whether they happen to have the same effect or not completely depends
> on the particular *usecase*. And for example, in the suspend/resume case,
> when the flag 'cpu_hotplug_disabled' is set, *all* CPUs are online!
> Only much later do we actually call disable_nonboot_cpus() to offline
> the non-boot CPUs. Take a look at the following functions, then you'll
> see what scenario I'm referring to:
> cpu_hotplug_disable_before_freeze(), cpu_hotplug_disable_after_thaw()
> 
> Also, recently, there was another usecase for the cpu_hotplug_disabled
> flag, in the reboot code. Here is the link to that discussion:
> http://thread.gmane.org/gmane.linux.kernel/1480380
> 

OK, Thanks!

> 
> >>
> >> The num_online_cpus() check is to ensure that the user doesn't do
> >> something insane like trying to offline the last online CPU in the
> >> system.
> >>
> >> Whereas, the flag 'cpu_hotplug_disabled' is used to prevent user-
> >> triggered CPU hotplug (such as those initiated through sysfs).
> >> This is useful in cases where the system itself wants to initiate CPU
> >> hotplug and it doesn't want annoying races with CPU hotplug going
> >> on in parallel due to other reasons. One such case is suspend/resume.
> >> That's why, if you have noticed, the suspend/resume code invokes the
> >> _cpu_down() version, in order to bypass the flag and get its job done.
> >>
> >> So, no, I think the check needs to stay.
> >>
> >> Regards,
> >> Srivatsa S. Bhat
> >>
> >>> Signed-off-by: liguang <lig.fnst@xxxxxxxxxxxxxx>
> >>> ---
> >>>  kernel/cpu.c |    6 ------
> >>>  1 files changed, 0 insertions(+), 6 deletions(-)
> >>>
> >>> diff --git a/kernel/cpu.c b/kernel/cpu.c
> >>> index b5e4ab2..cd166d3 100644
> >>> --- a/kernel/cpu.c
> >>> +++ b/kernel/cpu.c
> >>> @@ -330,14 +330,8 @@ int __ref cpu_down(unsigned int cpu)
> >>>
> >>>  	cpu_maps_update_begin();
> >>>
> >>> -	if (cpu_hotplug_disabled) {
> >>> -		err = -EBUSY;
> >>> -		goto out;
> >>> -	}
> >>> -
> >>>  	err = _cpu_down(cpu, 0);
> >>>
> >>> -out:
> >>>  	cpu_maps_update_done();
> >>>  	return err;
> >>>  }
> >>>
> >>
> > 
> > 
> 


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/