Re: [dpdk-dev] [PATCH 2/2] hyperv: VMBUS support infrastucture

[Stephen Hemminger <stephen@networkplumber.org>]

On Thu, 15 Dec 2016 12:19:44 +0530
Shreyansh Jain <shreyansh.jain@nxp.com> wrote:

> > @@ -1866,7 +1871,11 @@ typedef int (*eth_dev_uninit_t)(struct rte_eth_dev *eth_dev);
> >   * - The size of the private data to allocate for each matching device.
> >   */
> >  struct eth_driver {
> > -	struct rte_pci_driver pci_drv;    /**< The PMD is also a PCI driver. */
> > +	union {
> > +		struct rte_pci_driver pci_drv;    /**< The PMD PCI driver. */
> > +		struct rte_vmbus_driver vmbus_drv;/**< The PMD VMBUS drv. */
> > +	};
> > +
> >  	eth_dev_init_t eth_dev_init;      /**< Device init function. */
> >  	eth_dev_uninit_t eth_dev_uninit;  /**< Device uninit function. */
> >  	unsigned int dev_private_size;    /**< Size of device private data. */  
> 
> It is not a scale-able model where we have to change eth_driver/eth_dev 
> for every new device type, other than PCI. Maybe VMBus is _very_ close 
> to PCI so no changes are required in PCI layer (common, linuxapp, 
> bsdapp) - but, for others it won't stop there.
> 
> At the least, rte_pci_driver/rte_pci_device should be removed from 
> eth_driver & rte_eth_dev, respectively - relying on rte_driver and 
> rte_device.
> 
> This is the primary reason work on the SoC patchset and now the new Bus 
> model is being done.

Agreed. the better long term model is to use C style inheritance where
rte_pci_driver has eth_driver inside. 
The other alternative is to make the second element an opaque pointer.

But that was too big a change, and not necessary to get VMBUS to work.
Longer term refactoring will take more effort. Go ahead and address it
with a better bus model, but that probably isn't going to be ready for
a couple of releases.