2017-01-06 17:30, Shreyansh Jain:
> On Friday 06 January 2017 04:08 PM, Shreyansh Jain wrote:
> > On Wednesday 04 January 2017 03:16 AM, Thomas Monjalon wrote:
> >> 2016-12-26 18:53, Shreyansh Jain:
> >>> --- a/lib/librte_eal/linuxapp/eal/eal.c
> >>> +++ b/lib/librte_eal/linuxapp/eal/eal.c
> >>> @@ -844,6 +845,9 @@ rte_eal_init(int argc, char **argv)
> >>> if (rte_eal_intr_init() < 0)
> >>> rte_panic("Cannot init interrupt-handling thread\n");
> >>>
> >>> + if (rte_eal_bus_scan())
> >>> + rte_panic("Cannot scan the buses for devices\n");
> >>
> >> Yes, definitely. Just one scan functions which scan registered bus.
> >>
> >>> @@ -884,6 +888,9 @@ rte_eal_init(int argc, char **argv)
> >>> if (rte_eal_pci_probe())
> >>> rte_panic("Cannot probe PCI\n");
> >>>
> >>> + if (rte_eal_bus_probe())
> >>> + rte_panic("Cannot probe devices\n");
> >>> +
> >>> if (rte_eal_dev_init() < 0)
> >>> rte_panic("Cannot init pmd devices\n");
> >>
> >> What is the benefit of initializing (probe) a device outside of the scan?
> >> Currently, it is done in two steps, so you are keeping the same
> >> behaviour.
> >
> > Yes, only for compatibility to existing model of two-step process.
> > Ideally, only a single step is enough (init->probe).
> >
> > During the discussion in [1] also this point was raised - at that time
> > for VDEV and applicability to PCI.
> >
> > [1] http://dpdk.org/ml/archives/dev/2016-December/051306.html
> >
> > If you want, I can merge these two. I postponed it because 1) it is an
> > independent change and should really impact bus and 2) I was not sure
> > of dependency of init *before* pthread_create for all workers.
>
> I forgot _not_ in above - rephrasing:
>
> If you want, I can merge these two. I postponed it because 1) it is an
> independent change and should _not_ really impact bus and 2) I was not
> sure of dependency of init *before* pthread_create for all workers.
I'm OK with your approach.
> >> I imagine a model where the scan function decide to initialize the
> >> device and can require some help from a callback to make this decision.
> >> So the whitelist/blacklist policy can be implemented with callbacks at
> >> the scan level and possibly the responsibility of the application.
> >> Note that the callback model would be a change for a next release.
> >
> > Agree. But, that is not really part of Bus patches - isn't it? Or, you
> > want to change that with this series?
No it is not the scope of this series.
Please could you add it in the cover letter as a next step?
Thanks
