Re: [PATCH] markers: comment usage of marker_synchronize_unregister()

[Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>]

* Wu Fengguang (fengguang.wu@intel.com) wrote:
> Add more comments to marker_synchronize_unregister() in order to
> reduce the chance of misusing.
> 
> Based on comments from Lai Jiangshan <laijs@cn.fujitsu.com>.
> 
> Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
> Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
> Signed-off-by: Wu Fengguang <wfg@linux.intel.com>
> ---
> 
> I'm still not sure about the last sentence. Can anyone clarify on
> this? Thanks!
> 
> diff --git a/include/linux/marker.h b/include/linux/marker.h
> index 889196c..89ce1b8 100644
> --- a/include/linux/marker.h
> +++ b/include/linux/marker.h
> @@ -164,6 +164,12 @@ extern void *marker_get_private_data(const char *name, marker_probe_func *probe,
>   * marker_synchronize_unregister must be called between the last marker probe
>   * unregistration and the end of module exit to make sure there is no caller
>   * executing a probe when it is freed.
> + *
> + * It must be called _also_ between unregistration and destruction the data
> + * that unregistration-ed probes need to make sure there is no caller executing
> + * a probe when it's data is destroyed.

it's -> its

And the way it's written, this last sentence is a bit misleading. One
might think that the synchronize_unregister has to be called two, when
in fact it just has to be called once, but it must be called at a moment
in time between unregister and free of any resource used by the probes,
including the code which is removed by module unload.

> + *
> + * It works reliably only when all probe routines do not sleep and reschedule.

Per definition, preemption is disabled around marker probe execution, so
I don't see why we should add this last sentence ?

Mathieu

>   */
>  #define marker_synchronize_unregister() synchronize_sched()
>  

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68