Re: [dpdk-dev] [PATCH] eal_interrupts.c: properly init struct epoll_event (valgrind)

[Stephen Hemminger <stephen@networkplumber.org>]

On Thu, 17 Mar 2016 15:18:15 +0100
Thomas Monjalon <thomas.monjalon@6wind.com> wrote:

> Hi Stephen,
> 
> Please, could you turn it into a real patch with your sign-off?
> Thanks
> 
> 2016-02-14 12:22, Stephen Hemminger:
> > A better patch would be to move the data structure into the
> > code block used, and get rid of the useless else (rte_panic never returns);
> > and fix the indentation, and use C99 initialization which should make valgrind
> > happier.
> > 
> > The moral is don't just slap memsets around
> > 
> > diff --git a/lib/librte_eal/linuxapp/eal/eal_interrupts.c b/lib/librte_eal/linuxapp/eal/eal_interrupts.c
> > index 06b26a9..d53826e 100644
> > --- a/lib/librte_eal/linuxapp/eal/eal_interrupts.c
> > +++ b/lib/librte_eal/linuxapp/eal/eal_interrupts.c
> > @@ -799,8 +799,6 @@ eal_intr_handle_interrupts(int pfd, unsigned totalfds)
> >  static __attribute__((noreturn)) void *
> >  eal_intr_thread_main(__rte_unused void *arg)
> >  {
> > -	struct epoll_event ev;
> > -
> >  	/* host thread, never break out */
> >  	for (;;) {
> >  		/* build up the epoll fd with all descriptors we are to
> > @@ -834,20 +832,22 @@ eal_intr_thread_main(__rte_unused void *arg)
> >  		TAILQ_FOREACH(src, &intr_sources, next) {
> >  			if (src->callbacks.tqh_first == NULL)
> >  				continue; /* skip those with no callbacks */
> > -			ev.events = EPOLLIN | EPOLLPRI;
> > -			ev.data.fd = src->intr_handle.fd;
> > +
> > +			struct epoll_event ev = {
> > +				.events = EPOLLIN | EPOLLPRI,
> > +				.data.fd = src->intr_handle.fd,
> > +			};
> >  
> >  			/**
> >  			 * add all the uio device file descriptor
> >  			 * into wait list.
> >  			 */
> >  			if (epoll_ctl(pfd, EPOLL_CTL_ADD,
> > -					src->intr_handle.fd, &ev) < 0){
> > +					src->intr_handle.fd, &ev) < 0)
> >  				rte_panic("Error adding fd %d epoll_ctl, %s\n",
> >  					src->intr_handle.fd, strerror(errno));
> > -			}
> > -			else
> > -				numfds++;
> > +
> > +			numfds++;
> >  		}
> >  		rte_spinlock_unlock(&intr_lock);
> >  		/* serve the interrupt */
> 
> 

Sure I thought Matthew would since he reported the issue and had the ability
to test it.