Re: [dpdk-dev] [PATCH] net/failsafe: fix Rx interrupt reinstallation

["Gaëtan Rivet" <gaetan.rivet@6wind.com>]

Hi Matan,

On Tue, Feb 13, 2018 at 10:59:32PM +0000, Matan Azrad wrote:
> Fail-safe dev_start() operation can be called by both the application
> and the hot-plug alarm mechanism.
> 
> The installation of Rx interrupt are triggered from dev_start() in any
> time it is called while actually the Rx interrupt should be installed
> only by the application calls.
> 
> So, each plug-in event causes reinstallation which causes memory leak.
> 
> Trigger the Rx interrupt installation only for application calls.
> 
> Fixes: 9e0360aebf23 ("net/failsafe: register as Rx interrupt mode")
> 
> Signed-off-by: Matan Azrad <matan@mellanox.com>
> ---
>  drivers/net/failsafe/failsafe_ops.c | 10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/net/failsafe/failsafe_ops.c b/drivers/net/failsafe/failsafe_ops.c
> index 057e435..bbbd335 100644
> --- a/drivers/net/failsafe/failsafe_ops.c
> +++ b/drivers/net/failsafe/failsafe_ops.c
> @@ -181,10 +181,12 @@
>  	int ret;
>  
>  	fs_lock(dev, 0);
> -	ret = failsafe_rx_intr_install(dev);
> -	if (ret) {
> -		fs_unlock(dev, 0);
> -		return ret;
> +	if (PRIV(dev)->alarm_lock == 0) {

I dislike having to rely on unrelated context of execution to decide a
code-path.

I'd prefer to make interrupt installation dependent on the interrupt
state instead.

I think it should be possible to forbid reinstallation within
failsafe_rx_intr_install directly, e.g.

diff --git a/drivers/net/failsafe/failsafe_intr.c b/drivers/net/failsafe/failsafe_intr.c
index f6ff04dc8..46c3aa5f2 100644
--- a/drivers/net/failsafe/failsafe_intr.c
+++ b/drivers/net/failsafe/failsafe_intr.c
@@ -523,7 +523,8 @@ failsafe_rx_intr_install(struct rte_eth_dev *dev)
        const struct rte_intr_conf *const intr_conf =
                        &priv->dev->data->dev_conf.intr_conf;

-       if (intr_conf->rxq == 0)
+       if (intr_conf->rxq == 0 ||
+           dev->intr_handle != NULL)
                return 0;
        if (fs_rx_intr_vec_install(priv) < 0)
                return -rte_errno;

This way the logic is self-dependent and the check limited to this
component.

There might be better way to do this, it's only an example to explain my
point.

> +		ret = failsafe_rx_intr_install(dev);
> +		if (ret) {
> +			fs_unlock(dev, 0);
> +			return ret;
> +		}
>  	}
>  	FOREACH_SUBDEV(sdev, i, dev) {
>  		if (sdev->state != DEV_ACTIVE)
> -- 
> 1.9.5
> 

-- 
Gaëtan Rivet
6WIND