[ofa-general] Re: [PATCH] opensm: serialize osm_state_mgr_process()

[Sasha Khapyorsky]

On 14:34 Tue 28 Aug     , Sasha Khapyorsky wrote:
> 

[snip...]

> diff --git a/opensm/opensm/osm_sm.c b/opensm/opensm/osm_sm.c
> index 6fc9496..c052456 100644
> --- a/opensm/opensm/osm_sm.c
> +++ b/opensm/opensm/osm_sm.c
> @@ -69,50 +69,51 @@
>  
>  /**********************************************************************
>   **********************************************************************/
> -void __osm_sm_sweeper(IN void *p_ptr)
> +static void __osm_sm_sweeper(IN void *p_ptr)
>  {
>  	ib_api_status_t status;
>  	osm_sm_t *const p_sm = (osm_sm_t *) p_ptr;
> +	unsigned signals, i;
>  
>  	OSM_LOG_ENTER(p_sm->p_log, __osm_sm_sweeper);
>  
> -	if (p_sm->thread_state == OSM_THREAD_STATE_INIT) {
> -		p_sm->thread_state = OSM_THREAD_STATE_RUN;
> -	}
> -
> -	/* If the sweep interval was updated before - then run only if
> -	 * it is not zero. */
> -	while (p_sm->thread_state == OSM_THREAD_STATE_RUN &&
> -	       p_sm->p_subn->opt.sweep_interval != 0) {
> -		/*  do the sweep only if we are in MASTER state */
> -		if (p_sm->p_subn->sm_state == IB_SMINFO_STATE_MASTER ||
> -		    p_sm->p_subn->sm_state == IB_SMINFO_STATE_DISCOVERING)
> -			osm_state_mgr_process(&p_sm->state_mgr,
> -					      OSM_SIGNAL_SWEEP);
> -
> +	while (p_sm->thread_state == OSM_THREAD_STATE_RUN) {
>  		/*
>  		 * Wait on the event with a timeout.
>  		 * Sweeps may be initiated "off schedule" by simply
>  		 * signaling the event.
>  		 */
> -		status = cl_event_wait_on(&p_sm->signal,
> +		status = cl_event_wait_on(&p_sm->signal_event,
> +					  p_sm->p_subn->opt.sweep_interval ?
>  					  p_sm->p_subn->opt.sweep_interval *
> -					  1000000, TRUE);
> -
> -		if (status == CL_SUCCESS) {
> -			if (osm_log_is_active(p_sm->p_log, OSM_LOG_DEBUG)) {
> -				osm_log(p_sm->p_log, OSM_LOG_DEBUG,
> -					"__osm_sm_sweeper: "
> -					"Off schedule sweep signalled\n");
> -			}
> -		} else {
> -			if (status != CL_TIMEOUT) {
> -				osm_log(p_sm->p_log, OSM_LOG_ERROR,
> -					"__osm_sm_sweeper: ERR 2E01: "
> -					"Event wait failed (%s)\n",
> -					CL_STATUS_MSG(status));
> -			}
> +					  1000000 : EVENT_NO_TIMEOUT, TRUE);
> +
> +		if (status == CL_SUCCESS)
> +			osm_log(p_sm->p_log, OSM_LOG_DEBUG,
> +				"__osm_sm_sweeper: "
> +				"Off schedule sweep signalled\n");
> +		else if (status != CL_TIMEOUT) {
> +			osm_log(p_sm->p_log, OSM_LOG_ERROR,
> +				"__osm_sm_sweeper: ERR 2E01: "
> +				"Event wait failed (%s)\n",
> +				CL_STATUS_MSG(status));
> +			continue;
>  		}
> +
> +		cl_spinlock_acquire(&p_sm->signal_lock);
> +		signals = p_sm->signal_mask;
> +		p_sm->signal_mask = 0;
> +		cl_spinlock_release(&p_sm->signal_lock);
> +
> +		/*  do the sweep only if we are in MASTER state */
> +		if (status == CL_TIMEOUT &&
> +		    (p_sm->p_subn->sm_state == IB_SMINFO_STATE_MASTER ||
> +		     p_sm->p_subn->sm_state == IB_SMINFO_STATE_DISCOVERING))
> +			p_sm->signal_mask |= 1 << OSM_SIGNAL_SWEEP;

Actually here should be just:

			signals != 1 << OSM_SIGNAL_SWEEP;

> +
> +		for (i = 0 ; signals ; signals >>= 1 , i++)
> +			if (signals&1)
> +				osm_state_mgr_process(&p_sm->state_mgr, i);
>  	}
>  
>  	OSM_LOG_EXIT(p_sm->p_log);