[PATCH] vdpa/ifc: fix update_datapath error handling

[Maxime Coquelin]

Hi,

On 10/18/22 09:22, Taekyung Kim wrote:
> Stop and return the error code if update_datapath fails.
> update_datapath prepares resources for the vdpa device.
> The driver should not perform any further actions
> if update_datapath returns an error.

You need to add Fixes tag and Cc stable at dpdk.org, so that it is
backported to relevant LTS branches.

> 
> Signed-off-by: Taekyung Kim <kim.tae.kyung at navercorp.com>
> ---
>   drivers/vdpa/ifc/ifcvf_vdpa.c | 20 ++++++++++++++++----
>   1 file changed, 16 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/vdpa/ifc/ifcvf_vdpa.c b/drivers/vdpa/ifc/ifcvf_vdpa.c
> index d5ac583589..795967e998 100644
> --- a/drivers/vdpa/ifc/ifcvf_vdpa.c
> +++ b/drivers/vdpa/ifc/ifcvf_vdpa.c
> @@ -1063,7 +1063,10 @@ ifcvf_dev_config(int vid)
>   	internal = list->internal;
>   	internal->vid = vid;
>   	rte_atomic32_set(&internal->dev_attached, 1);
> -	update_datapath(internal);
> +	if (update_datapath(internal) < 0) {
> +		DRV_LOG(ERR, "failed to update datapath: %p", vdev);
> +		return -1;
> +	}
>   
>   	if (rte_vhost_host_notifier_ctrl(vid, RTE_VHOST_QUEUE_ALL, true) != 0)
>   		DRV_LOG(NOTICE, "vDPA (%s): software relay is used.",
> @@ -1105,7 +1108,10 @@ ifcvf_dev_close(int vid)
>   		internal->sw_fallback_running = false;
>   	} else {
>   		rte_atomic32_set(&internal->dev_attached, 0);
> -		update_datapath(internal);
> +		if (update_datapath(internal) < 0) {
> +			DRV_LOG(ERR, "failed to update datapath: %p", vdev);
> +			return -1;
> +		}
>   	}
>   
>   	internal->configured = 0;
> @@ -1632,7 +1638,10 @@ ifcvf_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
>   	pthread_mutex_unlock(&internal_list_lock);
>   
>   	rte_atomic32_set(&internal->started, 1);
> -	update_datapath(internal);
> +	if (update_datapath(internal) < 0) {
> +		DRV_LOG(ERR, "failed to update datapath: %s", pci_dev->name);
> +		return -1;
> +	}

This is not enough, you need to free resources allocated before,
otherwise you introduce memory leaks.

Basically, you should goto error instead o directly returning -1.

>   
>   	rte_kvargs_free(kvlist);
>   	return 0;
> @@ -1661,7 +1670,10 @@ ifcvf_pci_remove(struct rte_pci_device *pci_dev)
>   
>   	internal = list->internal;
>   	rte_atomic32_set(&internal->started, 0);
> -	update_datapath(internal);
> +	if (update_datapath(internal) < 0) {
> +		DRV_LOG(ERR, "failed to update datapath: %s", pci_dev->name);
> +		return -1;
> +	}

I think we should not return early here, because we still want to clean
as much as possible on remove. maybe just keep the error message.

>   	rte_pci_unmap_device(internal->pdev);
>   	rte_vfio_container_destroy(internal->vfio_container_fd);

Thanks,
Maxime