On Tue, 20 Dec 2016 12:46:17 +0530
Shreyansh Jain <shreyansh.j...@nxp.com> wrote:
> On Tuesday 20 December 2016 03:29 AM, Stephen Hemminger wrote:
> > Since multiple buses and device types need to be supported.
> > Provide type field in driver.
> > ---
> > lib/librte_eal/common/include/rte_dev.h | 15 ++++++++++++---
> > lib/librte_eal/common/include/rte_pci.h | 1 +
> > lib/librte_eal/common/include/rte_vdev.h | 1 +
> > 3 files changed, 14 insertions(+), 3 deletions(-)
> >
> > diff --git a/lib/librte_eal/common/include/rte_dev.h
> > b/lib/librte_eal/common/include/rte_dev.h
> > index e5471a22..3f4e26e6 100644
> > --- a/lib/librte_eal/common/include/rte_dev.h
> > +++ b/lib/librte_eal/common/include/rte_dev.h
> > @@ -144,12 +144,21 @@ void rte_eal_device_insert(struct rte_device *dev);
> > void rte_eal_device_remove(struct rte_device *dev);
> >
> > /**
> > + * Type of device driver
> > + */
> > +enum rte_driver_type {
> > + PMD_VIRTUAL,
> > + PMD_PCI,
> > +};
>
> So the expectation is that if a new device type is introduced (anything
> other than virtual or PCI), this enum is expanded?
>
> Broadly, we have two possible ways being discussed on ML for device
> relationship with PMD/APIs:
>
> 1. the way you have done: each device belongs to a particular type (enum
> rte_driver_type) and based on its type, operations would be performed on
> it (using conditional operators, for example). We continue to have a
> common device list containing all type of devices.
It is more about providing feedback to applications in info_get which is
the only place they should care.
>
> 2. disassociating the device (rte_device) completely from its type and
> basing it on a Bus type. So, we have separate buses for each device type
> and hence no need for separation of logic based on device type.
>
> I think implementation similar to (1) existed before it was removed in
> 6751f6de.
>
> I have an (obvious) inclination towards (2) because that helps plugging
> in more drivers/device types without expecting the contributor to change
> the EAL - however trivial the change is.
>
The difference is that virtual devices don't actually live on bus.
So it is really a property of the device not a bus per say.
