[dpdk-dev] [PATCH v3 3/3] examples/l3fwd: Handle SIGINT and SIGTERM in l3fwd

[Wang, Zhihong]

> -----Original Message-----
> From: Ananyev, Konstantin
> Sent: Wednesday, December 30, 2015 7:30 PM
> To: Wang, Zhihong <zhihong.wang at intel.com>; dev at dpdk.org
> Cc: stephen at networkplumber.org; Qiu, Michael <michael.qiu at intel.com>
> Subject: RE: [PATCH v3 3/3] examples/l3fwd: Handle SIGINT and SIGTERM in
> l3fwd
> 
> 
> 
> > -----Original Message-----
> > From: Wang, Zhihong
> > Sent: Wednesday, December 30, 2015 3:15 AM
> > To: Ananyev, Konstantin; dev at dpdk.org
> > Cc: stephen at networkplumber.org; Qiu, Michael
> > Subject: RE: [PATCH v3 3/3] examples/l3fwd: Handle SIGINT and SIGTERM
> > in l3fwd
> >
> > > > +static uint8_t
> > > > +start_ports(void)
> > > > +{
> > > > +	unsigned portid, nb_ports, avail_ports;
> > > > +	int ret;
> > > > +
> > > > +	nb_ports = rte_eth_dev_count();
> > > > +	avail_ports = 0;
> > > > +	for (portid = 0; portid < nb_ports; portid++) {
> > > > +		if ((enabled_port_mask & (1 << portid)) == 0)
> > > > +			continue;
> > > > +		avail_ports++;
> > > > +		port_started = true;
> > >
> > > Why do you need it at each iteration?
> >
> > Only become true when the first enabled port about to started. In case there's
> no port enabled at all.
> > In my opinion no need to optimize since it's not performance sensitive
> > and the logic is correct :)
> >
> >
> > >
> > > > +		printf("Starting port %d...", portid);
> > > > +		ret = rte_eth_dev_start(portid);
> > > > +		if (ret < 0)
> > > > +			rte_exit(EXIT_FAILURE,
> > > > +					"rte_eth_dev_start: err=%d, port=%d\n",
> > > > +					ret, portid);
> > > > +		/*
> > > > +		 * If enabled, put device in promiscuous mode.
> > > > +		 * This allows IO forwarding mode to forward packets
> > > > +		 * to itself through 2 cross-connected  ports of the
> > > > +		 * target machine.
> > > > +		 */
> > > > +		if (promiscuous_on)
> > > > +			rte_eth_promiscuous_enable(portid);
> > > > +		printf(" Done\n");
> > > > +	}
> > > > +
> > > > +	return avail_ports;
> > > > +}
> >
> > [...]
> >
> > > > +static void
> > > > +signal_handler(int signum)
> > > > +{
> > > > +	if (signum == SIGINT || signum == SIGTERM) {
> > > > +		printf("\nSignal %d received, preparing to exit...\n",
> > > > +				signum);
> > > > +		if (port_started) {
> > > > +			printf("Ports started already...\n");
> > > > +			signo_quit = signum;
> > > > +			force_quit = true;
> > > > +		} else {
> > >
> > >
> > > Hmm, and what if signal_handler() would be executed not in the
> > > context of master lcore?
> > > Then there could be a raise condition, and you could end up here,
> > > while master lcore would be in the middle of
> start_ports()->rte_eth_dev_start().
> >
> > Good point! Then we need rte_atomic16_cmpset() to avoid the race condition.
> >
> >
> > > Probably not a big deal, but why do you need this  if (port_started)
> > > {...} else {...} at all?
> > > Why not just:
> >
> > If no port has been started, then just kill itself.
> > This is for cases like when you just started it and then want to shut
> > it down, it'll wait a long time for initialization (memory, etc.) before the
> force_quit signal take effect.
> 
> Do you mean rte_eal_init()?
> Then why not to install non-default signal handlers after rte_eal_init()?
> Konstantin

Yes that does sounds better :)



> 
> >
> >
> > >
> > > signal_handler(int signum)
> > > {
> > > 	signo_quit = signum;
> > > 	force_quit = true;
> > > }
> > > ?
> > >
> > > Konstantin
> > >
> > > > +			printf("Ports not started yet...\n");
> > > > +			printf("Bye...\n");
> > > > +			/* exit with the expected status */
> > > > +			signal(signum, SIG_DFL);
> > > > +			kill(getpid(), signum);
> > > > +		}
> > > > +	}
> > > > +}
> > > > +