[dpdk-dev] [PATCH v2 1/6] ethdev: fix port data mismatched in multiple process model

[Yuanhan Liu]

On Wed, Jan 04, 2017 at 06:34:40PM +0100, Thomas Monjalon wrote:
> +Cc Sergio (maintainer of the secondary process thing)
> 
> 2016-12-28 19:02, Yuanhan Liu:
> > --- a/lib/librte_ether/rte_ethdev.c
> > +++ b/lib/librte_ether/rte_ethdev.c
> > @@ -201,9 +201,6 @@ rte_eth_dev_allocate(const char *name)
> >  		return NULL;
> >  	}
> >  
> > -	if (rte_eth_dev_data == NULL)
> > -		rte_eth_dev_data_alloc();
> > -
> 
> It is dangerous to move this to rte_eth_dev_pci_probe.
> Please keep it here and duplicate it in eth_dev_attach.

Oh, right, I missed the fact that it could be invoked from other places.

> [...]
> > +/*
> > + * Attach to a port already registered by the primary process, which
> > + * makes sure that the same device would both have the same port id
> > + * in the primary and secondary process.
> > + */
> > +static struct rte_eth_dev *
> > +eth_dev_attach(const char *name)
> 
> Maybe that the word "secondary" could help to differentiate of
> the function rte_eth_dev_attach().

Yes, it's better. How about "_attach_secondary", or "_attach_to_primary"?

> > +{
> > +	uint8_t i;
> > +	struct rte_eth_dev *eth_dev;
> > +
> > +	for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
> > +		if (strcmp(rte_eth_dev_data[i].name, name) == 0)
> > +			break;
> > +	}
> > +	if (i == RTE_MAX_ETHPORTS) {
> > +		RTE_PMD_DEBUG_TRACE(
> > +			"device %s is not driven by the primary process\n",
> > +			name);
> > +		return NULL;
> > +	}
> > +
> > +	RTE_ASSERT(eth_dev->data->port_id == i);
> > +
> > +	eth_dev = &rte_eth_devices[i];
> > +	eth_dev->data = &rte_eth_dev_data[i];
> > +	eth_dev->attached = DEV_ATTACHED;
> > +	nb_ports++;
> 
> I am a bit nervous when I see these lines duplicated from rte_eth_dev_allocate.
> Not sure whether it deserves a common function or not.

I don't think so, they do share some common assignments, but the assignments
are actually not the same. The primary one has few more: notably, they are:

- eth_dev->data
- eth_dev->data->name
- eth_dev->data->port_id

> 
> [...]
> > @@ -246,9 +275,26 @@ rte_eth_dev_pci_probe(struct rte_pci_driver *pci_drv,
> > -	eth_dev = rte_eth_dev_allocate(ethdev_name);
> > -	if (eth_dev == NULL)
> > -		return -ENOMEM;
> > +	if (rte_eth_dev_data == NULL)
> > +		rte_eth_dev_data_alloc();
> > +
> > +	if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
> > +		eth_dev = rte_eth_dev_allocate(ethdev_name);
> > +		if (eth_dev == NULL)
> > +			return -ENOMEM;
> > +	} else {
> > +		/*
> > +		 * if we failed to attach a device, it means that
> > +		 * device is skipped, due to some errors. Take
> > +		 * virtio-net device as example, it could be the
> > +		 * device is managed by virtio-net kernel driver.
> > +		 * For such case, we return a positive value, to
> > +		 * let EAL skip it as well.
> > +		 */
> 
> This comment (a bit too long) should be placed between "if" and "return".

Okay.

	--yliu