Re: [dpdk-dev] [PATCH v2 1/2] eventdev: add device stop flush callback

[Jerin Jacob <jerin.jacob@caviumnetworks.com>]

-----Original Message-----
> When an event device is stopped, it drains all event queues. These events
> may contain pointers, so to prevent memory leaks eventdev now supports a
> user-provided flush callback that is called during the queue drain process.
> This callback is stored in process memory, so the callback must be
> registered by any process that may call rte_event_dev_stop().
> 
> This commit also clarifies the behavior of rte_event_dev_stop().
> 
> This follows this mailing list discussion:
> http://dpdk.org/ml/archives/dev/2018-January/087484.html
> 
> Signed-off-by: Gage Eads <gage.eads@intel.com>
> ---
> v2: allow a NULL callback pointer to unregister the callback
> 
>  lib/librte_eventdev/rte_eventdev.c           | 17 +++++++++
>  lib/librte_eventdev/rte_eventdev.h           | 55 +++++++++++++++++++++++++++-
>  lib/librte_eventdev/rte_eventdev_version.map |  6 +++
>  3 files changed, 76 insertions(+), 2 deletions(-)
> 
> diff --git a/lib/librte_eventdev/rte_eventdev.c b/lib/librte_eventdev/rte_eventdev.c
> index 851a119..1aacb7b 100644
> --- a/lib/librte_eventdev/rte_eventdev.c
> +++ b/lib/librte_eventdev/rte_eventdev.c
> @@ -1123,6 +1123,23 @@ rte_event_dev_start(uint8_t dev_id)
>  	return 0;
>  }
>  
> +typedef void (*eventdev_stop_flush_t)(uint8_t dev_id, struct rte_event event,
> +		void *arg);
> +/**< Callback function called during rte_event_dev_stop(), invoked once per
> + * flushed event.
> + */
> +
>  #define RTE_EVENTDEV_NAME_MAX_LEN	(64)
>  /**< @internal Max length of name of event PMD */
>  
> @@ -1176,6 +1194,11 @@ struct rte_eventdev {
>  	event_dequeue_burst_t dequeue_burst;
>  	/**< Pointer to PMD dequeue burst function. */
>  
> +	eventdev_stop_flush_t dev_stop_flush;
> +	/**< Optional, user-provided event flush function */
> +	void *dev_stop_flush_arg;
> +	/**< User-provided argument for event flush function */
> +

I think, we can move this additions to the internal rte_eventdev_data structure. Reasons are
1) Changes to "struct rte_eventdev" would call for ABI change
2) We can keep "struct rte_eventdev" only for fast path functions,
slow path functions can have additional redirection.

>  	struct rte_eventdev_data *data;
>  	/**< Pointer to device data */
>  	const struct rte_eventdev_ops *dev_ops;
> @@ -1822,6 +1845,34 @@ rte_event_dev_xstats_reset(uint8_t dev_id,
>   */
>  int rte_event_dev_selftest(uint8_t dev_id);
>  
> +/**
> + * Registers a callback function to be invoked during rte_event_dev_stop() for
> + * each flushed event. This function can be used to properly dispose of queued
> + * events, for example events containing memory pointers.
> + *
> + * The callback function is only registered for the calling process. The
> + * callback function must be registered in every process that can call
> + * rte_event_dev_stop().
> + *
> + * To unregister a callback, call this function with a NULL callback pointer.
> + *
> + * @param dev_id
> + *   The identifier of the device.
> + * @param callback
> + *   Callback function invoked once per flushed event.
> + * @param userdata
> + *   Argument supplied to callback.
> + *
> + * @return
> + *  - 0 on success.
> + *  - -EINVAL if *dev_id* is invalid
> + *
> + * @see rte_event_dev_stop()
> + */
> +int
> +rte_event_dev_stop_flush_callback_register(uint8_t dev_id,
> +		eventdev_stop_flush_t callback, void *userdata);
> +
IMO, It would be better if we place this function near to rte_event_dev_stop().

Other than above minor changes, It looks good to me.