Re: [dpdk-dev] [PATCH v6] eal: add function to check if primary proc alive

["Van Haaren, Harry" <harry.van.haaren@intel.com>]

Hi David,

> From: David Marchand [mailto:david.marchand@6wind.com]
> Subject: Re: [PATCH v6] eal: add function to check if primary proc alive

> When I look at this new api, I am under the impression that you are
> supposed to check for primary liveliness once dpdk init has finished
> (from your secondary process point of view), not before and not while
> it is initialising.

The issue is that if a secondary process is initialized, it holds a read
lock on  /var/run/.rte_config  and this prevents a primary from starting.

So we *must* detect a primary process being ready to attach to, *without*
having called  rte_eal_init()  in the secondary process.


> Why do you need to move this ?

Issues arise when a primary and secondary process both scan the PCI devices
at the same time. Moving  rte_eal_mcfg_complete()  solves this race-cond
because the secondary process will wait until the primary is finished.


> > +       if (config_file_path)
> > +               config_fd = open(config_file_path, O_RDONLY);
> > +       else {
> > +               char default_path[PATH_MAX+1];
> > +               snprintf(default_path, PATH_MAX, RUNTIME_CONFIG_FMT,
> > +                        default_config_dir, "rte");
> > +               config_fd = open(default_path, O_RDONLY);
> 
> Can't you reuse eal_runtime_config_path() here ?

No, as rte_eal_init() has not been called, for the same reason as above.
As rte_eal_init() has not been called, the shared config that is read by
eal_runtime_config_path() has not been initialized.


-Harry