Re: [dpdk-stable] [PATCH v2] drivers/net: fix segfault in secondary process

[Thomas Monjalon <thomas@monjalon.net>]

20/07/2018 14:44, Ferruh Yigit:
> --- a/drivers/net/tap/rte_eth_tap.c
> +++ b/drivers/net/tap/rte_eth_tap.c
>  	if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
>  	    strlen(params) == 0) {
> -		eth_dev = rte_eth_dev_attach_secondary(name);
> +		eth_dev = rte_eth_vdev_secondary_probe(dev, name, &ops);
>  		if (!eth_dev) {
>  			TAP_LOG(ERR, "Failed to probe %s", name);
>  			return -1;
>  		}
> -		/* TODO: request info from primary to set up Rx and Tx */
> -		eth_dev->dev_ops = &ops;
> -		rte_eth_dev_probing_finish(eth_dev);
>  		return 0;
>  	}
>  
> --- a/lib/librte_ethdev/rte_ethdev_vdev.h
> +++ b/lib/librte_ethdev/rte_ethdev_vdev.h
> +static inline struct rte_eth_dev *
> +rte_eth_vdev_secondary_probe(struct rte_vdev_device *vdev,
> +		const char *name, const struct eth_dev_ops *ops)
> +{
> +	struct rte_eth_dev *eth_dev;
> +
> +	eth_dev = rte_eth_dev_attach_secondary(name);
> +	if (!eth_dev)
> +		return NULL;
> +
> +	/* TODO: request info from primary to set up Rx and Tx */
> +	eth_dev->dev_ops = ops;
> +	eth_dev->device = &vdev->device;
> +	rte_eth_dev_probing_finish(eth_dev);
> +	return eth_dev;
> +}

As you can see in the following patch,

	net/tap: add queues when attaching from secondary process
	https://patches.dpdk.org/patch/43229/

The synchronization with primary process can be specific to the PMD.
It will use some private data and can need IPC call to translate some
file descriptors.
Moreover, we will need to think about a more complex sync in order
to update queues when a process re-configure queues number,
and also attach in primary a device created in secondary.
We need also to handle hotplug as Qi already proposed.

My suggestion: just fix the PMDs without creating an helper.
We could design a helper later when we will have thought to the global
issue we want to solve.