Re: [dpdk-dev] [v4 1/3] cryptodev: support enqueue callback functions

["Ananyev, Konstantin" <konstantin.ananyev@intel.com>]

> 
> This patch adds APIs to add/remove callback functions. The callback
> function will be called for each burst of crypto ops received on a
> given crypto device queue pair.
> 
> Signed-off-by: Abhinandan Gujjar <abhinandan.gujjar@intel.com>
> ---
>  config/rte_config.h                            |   1 +
>  lib/librte_cryptodev/meson.build               |   2 +-
>  lib/librte_cryptodev/rte_cryptodev.c           | 230 +++++++++++++++++++++++++
>  lib/librte_cryptodev/rte_cryptodev.h           | 158 ++++++++++++++++-
>  lib/librte_cryptodev/rte_cryptodev_version.map |   2 +
>  5 files changed, 391 insertions(+), 2 deletions(-)
> 
> diff --git a/config/rte_config.h b/config/rte_config.h
> index 03d90d7..e999d93 100644
> --- a/config/rte_config.h
> +++ b/config/rte_config.h
> @@ -61,6 +61,7 @@
>  /* cryptodev defines */
>  #define RTE_CRYPTO_MAX_DEVS 64
>  #define RTE_CRYPTODEV_NAME_LEN 64
> +#define RTE_CRYPTO_CALLBACKS 1
> 
>  /* compressdev defines */
>  #define RTE_COMPRESS_MAX_DEVS 64
> diff --git a/lib/librte_cryptodev/meson.build b/lib/librte_cryptodev/meson.build
> index c4c6b3b..8c5493f 100644
> --- a/lib/librte_cryptodev/meson.build
> +++ b/lib/librte_cryptodev/meson.build
> @@ -9,4 +9,4 @@ headers = files('rte_cryptodev.h',
>  	'rte_crypto.h',
>  	'rte_crypto_sym.h',
>  	'rte_crypto_asym.h')
> -deps += ['kvargs', 'mbuf']
> +deps += ['kvargs', 'mbuf', 'rcu']
> diff --git a/lib/librte_cryptodev/rte_cryptodev.c b/lib/librte_cryptodev/rte_cryptodev.c
> index 3d95ac6..0880d9b 100644
> --- a/lib/librte_cryptodev/rte_cryptodev.c
> +++ b/lib/librte_cryptodev/rte_cryptodev.c
> @@ -448,6 +448,91 @@ struct rte_cryptodev_sym_session_pool_private_data {
>  	return 0;
>  }
> 
> +#ifdef RTE_CRYPTO_CALLBACKS
> +/* spinlock for crypto device enq callbacks */
> +static rte_spinlock_t rte_cryptodev_callback_lock = RTE_SPINLOCK_INITIALIZER;
> +
> +static void
> +cryptodev_cb_cleanup(struct rte_cryptodev *dev)
> +{
> +	struct rte_cryptodev_cb **prev_cb, *curr_cb;
> +	struct rte_cryptodev_enq_cb_rcu *list;
> +	uint16_t qp_id;
> +
> +	if (dev->enq_cbs == NULL)
> +		return;
> +
> +	for (qp_id = 0; qp_id < dev->data->nb_queue_pairs; qp_id++) {
> +		list = &dev->enq_cbs[qp_id];
> +		prev_cb = &list->next;
> +
> +		while (*prev_cb != NULL) {
> +			curr_cb = *prev_cb;
> +			/* Remove the user cb from the callback list. */
> +			__atomic_store_n(prev_cb, curr_cb->next,
> +				__ATOMIC_RELAXED);
> +			rte_rcu_qsbr_synchronize(list->qsbr,
> +				RTE_QSBR_THRID_INVALID);

You call this function (cb_cleanup) only at dev_confiture().
At that moment DP threads can't do enqueue/dequeue anyway.
So you can safely skip all this synchronization code here and just do:

cb = list->next;
while (cb != NULL) {
	next = cb->next;
	rte_free(cb);
	cb = next;
}


> +			rte_free(curr_cb);

One thing that makes it sort of grey area:
we do free() for cb itself, but user provided data will be sort of 'lost'.
As it is not referenced from our cb struct anymore...
I see two options here - first just document explicitly that callbacks wouldn't
survive cryptodev_configure() and it is user responsibility to remove all
installed callbacks before doing dev_configure() to avoid possible memory leakage.
Another option - add user provided cleanup() function pointer into struct rte_cryptodev_cb
and call it here if provided:
struct rte_cryptodev_cb {
	struct rte_cryptodev_cb *next;
	/** < Pointer to next callback */
	rte_cryptodev_callback_fn fn;
	/** < Pointer to callback function */
	void *arg;
	/** < Pointer to argument */
	void (*cleanup)(void *);	
};

And here:
	If (curr_cb->cleanup != NULL)
		curr_cb->cleanup(curr_cb->arg);

 	rte_free(curr_cb);

Rest of the code - LGTM.
So with that addressed:
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
 
> +		}
> +
> +		rte_free(list->qsbr);
> +	}
> +
> +	rte_free(dev->enq_cbs);
> +	dev->enq_cbs = NULL;
> +}
> +
> +static int
> +cryptodev_cb_init(struct rte_cryptodev *dev)
> +{
> +	struct rte_cryptodev_enq_cb_rcu *list;
> +	struct rte_rcu_qsbr *qsbr;
> +	uint16_t qp_id;
> +	size_t size;
> +
> +	/* Max thread set to 1, as one DP thread accessing a queue-pair */
> +	const uint32_t max_threads = 1;
> +
> +	dev->enq_cbs = rte_zmalloc(NULL,
> +				   sizeof(struct rte_cryptodev_enq_cb_rcu) *
> +				   dev->data->nb_queue_pairs, 0);
> +	if (dev->enq_cbs == NULL) {
> +		CDEV_LOG_ERR("Failed to allocate memory for callbacks");
> +		rte_errno = ENOMEM;
> +		return -1;
> +	}
> +
> +	/* Create RCU QSBR variable */
> +	size = rte_rcu_qsbr_get_memsize(max_threads);
> +
> +	for (qp_id = 0; qp_id < dev->data->nb_queue_pairs; qp_id++) {
> +		list = &dev->enq_cbs[qp_id];
> +		qsbr = rte_zmalloc(NULL, size, RTE_CACHE_LINE_SIZE);
> +		if (qsbr == NULL) {
> +			CDEV_LOG_ERR("Failed to allocate memory for RCU on "
> +				"queue_pair_id=%d", qp_id);
> +			goto cb_init_err;
> +		}
> +
> +		if (rte_rcu_qsbr_init(qsbr, max_threads)) {
> +			CDEV_LOG_ERR("Failed to initialize for RCU on "
> +				"queue_pair_id=%d", qp_id);
> +			goto cb_init_err;
> +		}
> +
> +		list->qsbr = qsbr;
> +	}
> +
> +	return 0;
> +
> +cb_init_err:
> +	rte_errno = ENOMEM;
> +	cryptodev_cb_cleanup(dev);
> +	return -1;
> +
> +}
> +#endif
> 
>  const char *
>  rte_cryptodev_get_feature_name(uint64_t flag)
> @@ -927,6 +1012,11 @@ struct rte_cryptodev *
> 
>  	RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->dev_configure, -ENOTSUP);
> 
> +#ifdef RTE_CRYPTO_CALLBACKS
> +	rte_spinlock_lock(&rte_cryptodev_callback_lock);
> +	cryptodev_cb_cleanup(dev);
> +	rte_spinlock_unlock(&rte_cryptodev_callback_lock);
> +#endif
>  	/* Setup new number of queue pairs and reconfigure device. */
>  	diag = rte_cryptodev_queue_pairs_config(dev, config->nb_queue_pairs,
>  			config->socket_id);
> @@ -936,6 +1026,15 @@ struct rte_cryptodev *
>  		return diag;
>  	}
> 
> +#ifdef RTE_CRYPTO_CALLBACKS
> +	rte_spinlock_lock(&rte_cryptodev_callback_lock);
> +	diag = cryptodev_cb_init(dev);
> +	rte_spinlock_unlock(&rte_cryptodev_callback_lock);
> +	if (diag) {
> +		CDEV_LOG_ERR("Callback init failed for dev_id=%d", dev_id);
> +		return -ENOMEM;
> +	}
> +#endif
>  	rte_cryptodev_trace_configure(dev_id, config);
>  	return (*dev->dev_ops->dev_configure)(dev, config);
>  }
> @@ -1136,6 +1235,137 @@ struct rte_cryptodev *
>  			socket_id);
>  }
> 
> +#ifdef RTE_CRYPTO_CALLBACKS
> +struct rte_cryptodev_cb *
> +rte_cryptodev_add_enq_callback(uint8_t dev_id,
> +			       uint16_t qp_id,
> +			       rte_cryptodev_callback_fn cb_fn,
> +			       void *cb_arg)
> +{
> +	struct rte_cryptodev *dev;
> +	struct rte_cryptodev_enq_cb_rcu *list;
> +	struct rte_cryptodev_cb *cb, *tail;
> +
> +	if (!cb_fn)
> +		return NULL;
> +
> +	if (!rte_cryptodev_pmd_is_valid_dev(dev_id)) {
> +		CDEV_LOG_ERR("Invalid dev_id=%d", dev_id);
> +		return NULL;
> +	}
> +
> +	dev = &rte_crypto_devices[dev_id];
> +	if (qp_id >= dev->data->nb_queue_pairs) {
> +		CDEV_LOG_ERR("Invalid queue_pair_id=%d", qp_id);
> +		return NULL;
> +	}
> +
> +	cb = rte_zmalloc(NULL, sizeof(*cb), 0);
> +	if (cb == NULL) {
> +		CDEV_LOG_ERR("Failed to allocate memory for callback on "
> +			     "dev=%d, queue_pair_id=%d", dev_id, qp_id);
> +		rte_errno = ENOMEM;
> +		return NULL;
> +	}
> +
> +	rte_spinlock_lock(&rte_cryptodev_callback_lock);
> +
> +	cb->fn = cb_fn;
> +	cb->arg = cb_arg;
> +
> +	/* Add the callbacks in fifo order. */
> +	list = &dev->enq_cbs[qp_id];
> +	tail = list->next;
> +
> +	if (tail) {
> +		while (tail->next)
> +			tail = tail->next;
> +		/* Stores to cb->fn and cb->param should complete before
> +		 * cb is visible to data plane.
> +		 */
> +		__atomic_store_n(&tail->next, cb, __ATOMIC_RELEASE);
> +	} else {
> +		/* Stores to cb->fn and cb->param should complete before
> +		 * cb is visible to data plane.
> +		 */
> +		__atomic_store_n(&list->next, cb, __ATOMIC_RELEASE);
> +	}
> +
> +	rte_spinlock_unlock(&rte_cryptodev_callback_lock);
> +
> +	return cb;
> +}
> +
> +int
> +rte_cryptodev_remove_enq_callback(uint8_t dev_id,
> +				  uint16_t qp_id,
> +				  struct rte_cryptodev_cb *cb)
> +{
> +	struct rte_cryptodev *dev;
> +	struct rte_cryptodev_cb **prev_cb, *curr_cb;
> +	struct rte_cryptodev_enq_cb_rcu *list;
> +	int ret;
> +
> +	ret = -EINVAL;
> +
> +	if (!cb) {
> +		CDEV_LOG_ERR("cb is NULL");
> +		return ret;
> +	}
> +
> +	if (!rte_cryptodev_pmd_is_valid_dev(dev_id)) {
> +		CDEV_LOG_ERR("Invalid dev_id=%d", dev_id);
> +		return ret;
> +	}
> +
> +	dev = &rte_crypto_devices[dev_id];
> +	if (qp_id >= dev->data->nb_queue_pairs) {
> +		CDEV_LOG_ERR("Invalid queue_pair_id=%d", qp_id);
> +		return ret;
> +	}
> +
> +	rte_spinlock_lock(&rte_cryptodev_callback_lock);
> +	if (dev->enq_cbs == NULL) {
> +		CDEV_LOG_ERR("Callback not initialized");
> +		goto cb_err;
> +	}
> +
> +	list = &dev->enq_cbs[qp_id];
> +	if (list == NULL) {
> +		CDEV_LOG_ERR("Callback list is NULL");
> +		goto cb_err;
> +	}
> +
> +	if (list->qsbr == NULL) {
> +		CDEV_LOG_ERR("Rcu qsbr is NULL");
> +		goto cb_err;
> +	}
> +
> +	prev_cb = &list->next;
> +	for (; *prev_cb != NULL; prev_cb = &curr_cb->next) {
> +		curr_cb = *prev_cb;
> +		if (curr_cb == cb) {
> +			/* Remove the user cb from the callback list. */
> +			__atomic_store_n(prev_cb, curr_cb->next,
> +				__ATOMIC_RELAXED);
> +			ret = 0;
> +			break;
> +		}
> +	}
> +
> +	if (!ret) {
> +		/* Call sync with invalid thread id as this is part of
> +		 * control plane API
> +		 */
> +		rte_rcu_qsbr_synchronize(list->qsbr, RTE_QSBR_THRID_INVALID);
> +		rte_free(cb);
> +	}
> +
> +cb_err:
> +	rte_spinlock_unlock(&rte_cryptodev_callback_lock);
> +	return ret;
> +}
> +#endif
> 
>  int
>  rte_cryptodev_stats_get(uint8_t dev_id, struct rte_cryptodev_stats *stats)
> diff --git a/lib/librte_cryptodev/rte_cryptodev.h b/lib/librte_cryptodev/rte_cryptodev.h
> index 0935fd5..1b7d7ef 100644
> --- a/lib/librte_cryptodev/rte_cryptodev.h
> +++ b/lib/librte_cryptodev/rte_cryptodev.h
> @@ -23,6 +23,7 @@
>  #include "rte_dev.h"
>  #include <rte_common.h>
>  #include <rte_config.h>
> +#include <rte_rcu_qsbr.h>
> 
>  #include "rte_cryptodev_trace_fp.h"
> 
> @@ -522,6 +523,34 @@ struct rte_cryptodev_qp_conf {
>  	/**< The mempool for creating sess private data in sessionless mode */
>  };
> 
> +#ifdef RTE_CRYPTO_CALLBACKS
> +/**
> + * Function type used for pre processing crypto ops when enqueue burst is
> + * called.
> + *
> + * The callback function is called on enqueue burst immediately
> + * before the crypto ops are put onto the hardware queue for processing.
> + *
> + * @param	dev_id		The identifier of the device.
> + * @param	qp_id		The index of the queue pair in which ops are
> + *				to be enqueued for processing. The value
> + *				must be in the range [0, nb_queue_pairs - 1]
> + *				previously supplied to
> + *				*rte_cryptodev_configure*.
> + * @param	ops		The address of an array of *nb_ops* pointers
> + *				to *rte_crypto_op* structures which contain
> + *				the crypto operations to be processed.
> + * @param	nb_ops		The number of operations to process.
> + * @param	user_param	The arbitrary user parameter passed in by the
> + *				application when the callback was originally
> + *				registered.
> + * @return			The number of ops to be enqueued to the
> + *				crypto device.
> + */
> +typedef uint16_t (*rte_cryptodev_callback_fn)(uint16_t dev_id, uint16_t qp_id,
> +		struct rte_crypto_op **ops, uint16_t nb_ops, void *user_param);
> +#endif
> +
>  /**
>   * Typedef for application callback function to be registered by application
>   * software for notification of device events
> @@ -822,7 +851,6 @@ struct rte_cryptodev_config {
>  		enum rte_cryptodev_event_type event,
>  		rte_cryptodev_cb_fn cb_fn, void *cb_arg);
> 
> -
>  typedef uint16_t (*dequeue_pkt_burst_t)(void *qp,
>  		struct rte_crypto_op **ops,	uint16_t nb_ops);
>  /**< Dequeue processed packets from queue pair of a device. */
> @@ -839,6 +867,33 @@ typedef uint16_t (*enqueue_pkt_burst_t)(void *qp,
>  /** Structure to keep track of registered callbacks */
>  TAILQ_HEAD(rte_cryptodev_cb_list, rte_cryptodev_callback);
> 
> +#ifdef RTE_CRYPTO_CALLBACKS
> +/**
> + * @internal
> + * Structure used to hold information about the callbacks to be called for a
> + * queue pair on enqueue.
> + */
> +struct rte_cryptodev_cb {
> +	struct rte_cryptodev_cb *next;
> +	/** < Pointer to next callback */
> +	rte_cryptodev_callback_fn fn;
> +	/** < Pointer to callback function */
> +	void *arg;
> +	/** < Pointer to argument */
> +};
> +
> +/**
> + * @internal
> + * Structure used to hold information about the RCU for a queue pair.
> + */
> +struct rte_cryptodev_enq_cb_rcu {
> +	struct rte_cryptodev_cb *next;
> +	/** < Pointer to next callback */
> +	struct rte_rcu_qsbr *qsbr;
> +	/** < RCU QSBR variable per queue pair */
> +};
> +#endif
> +
>  /** The data structure associated with each crypto device. */
>  struct rte_cryptodev {
>  	dequeue_pkt_burst_t dequeue_burst;
> @@ -867,6 +922,10 @@ struct rte_cryptodev {
>  	__extension__
>  	uint8_t attached : 1;
>  	/**< Flag indicating the device is attached */
> +
> +	struct rte_cryptodev_enq_cb_rcu *enq_cbs;
> +	/**< User application callback for pre enqueue processing */
> +
>  } __rte_cache_aligned;
> 
>  void *
> @@ -989,6 +1048,31 @@ struct rte_cryptodev_data {
>  {
>  	struct rte_cryptodev *dev = &rte_cryptodevs[dev_id];
> 
> +#ifdef RTE_CRYPTO_CALLBACKS
> +	if (unlikely(dev->enq_cbs != NULL)) {
> +		struct rte_cryptodev_enq_cb_rcu *list;
> +		struct rte_cryptodev_cb *cb;
> +
> +		/* __ATOMIC_RELEASE memory order was used when the
> +		* call back was inserted into the list.
> +		* Since there is a clear dependency between loading
> +		* cb and cb->fn/cb->next, __ATOMIC_ACQUIRE memory order is
> +		* not required.
> +		*/
> +		list = &dev->enq_cbs[qp_id];
> +		rte_rcu_qsbr_thread_online(list->qsbr, 0);
> +		cb = __atomic_load_n(&list->next, __ATOMIC_RELAXED);
> +
> +		while (cb != NULL) {
> +			nb_ops = cb->fn(dev_id, qp_id, ops, nb_ops,
> +					cb->arg);
> +			cb = cb->next;
> +		};
> +
> +		rte_rcu_qsbr_thread_offline(list->qsbr, 0);
> +	}
> +#endif
> +
>  	rte_cryptodev_trace_enqueue_burst(dev_id, qp_id, (void **)ops, nb_ops);
>  	return (*dev->enqueue_burst)(
>  			dev->data->queue_pairs[qp_id], ops, nb_ops);
> @@ -1730,6 +1814,78 @@ struct rte_crypto_raw_dp_ctx {
>  rte_cryptodev_raw_dequeue_done(struct rte_crypto_raw_dp_ctx *ctx,
>  		uint32_t n);
> 
> +#ifdef RTE_CRYPTO_CALLBACKS
> +/**
> + * @warning
> + * @b EXPERIMENTAL: this API may change without prior notice
> + *
> + * Add a user callback for a given crypto device and queue pair which will be
> + * called on crypto ops enqueue.
> + *
> + * This API configures a function to be called for each burst of crypto ops
> + * received on a given crypto device queue pair. The return value is a pointer
> + * that can be used later to remove the callback using
> + * rte_cryptodev_remove_enq_callback().
> + *
> + * Multiple functions are called in the order that they are added.
> + *
> + * @param	dev_id		The identifier of the device.
> + * @param	qp_id		The index of the queue pair in which ops are
> + *				to be enqueued for processing. The value
> + *				must be in the range [0, nb_queue_pairs - 1]
> + *				previously supplied to
> + *				*rte_cryptodev_configure*.
> + * @param	cb_fn		The callback function
> + * @param	cb_arg		A generic pointer parameter which will be passed
> + *				to each invocation of the callback function on
> + *				this crypto device and queue pair.
> + *
> + * @return
> + *   NULL on error.
> + *   On success, a pointer value which can later be used to remove the callback.
> + */
> +
> +__rte_experimental
> +struct rte_cryptodev_cb *
> +rte_cryptodev_add_enq_callback(uint8_t dev_id,
> +			       uint16_t qp_id,
> +			       rte_cryptodev_callback_fn cb_fn,
> +			       void *cb_arg);
> +
> +
> +/**
> + * @warning
> + * @b EXPERIMENTAL: this API may change without prior notice
> + *
> + * Remove a user callback function for given crypto device and queue pair.
> + *
> + * This function is used to removed callbacks that were added to a crypto
> + * device queue pair using rte_cryptodev_add_enq_callback().
> + *
> + *
> + *
> + * @param	dev_id		The identifier of the device.
> + * @param	qp_id		The index of the queue pair in which ops are
> + *				to be enqueued for processing. The value
> + *				must be in the range [0, nb_queue_pairs - 1]
> + *				previously supplied to
> + *				*rte_cryptodev_configure*.
> + * @param	cb		Pointer to user supplied callback created via
> + *				rte_cryptodev_add_enq_callback().
> + *
> + * @return
> + *   - 0: Success. Callback was removed.
> + *   - -EINVAL:  The dev_id or the qp_id is out of range, or the callback
> + *               is NULL or not found for the crypto device queue pair.
> + */
> +
> +__rte_experimental
> +int rte_cryptodev_remove_enq_callback(uint8_t dev_id,
> +				      uint16_t qp_id,
> +				      struct rte_cryptodev_cb *cb);
> +
> +#endif
> +
>  #ifdef __cplusplus
>  }
>  #endif
> diff --git a/lib/librte_cryptodev/rte_cryptodev_version.map b/lib/librte_cryptodev/rte_cryptodev_version.map
> index 7e4360f..5d8d6b0 100644
> --- a/lib/librte_cryptodev/rte_cryptodev_version.map
> +++ b/lib/librte_cryptodev/rte_cryptodev_version.map
> @@ -101,6 +101,7 @@ EXPERIMENTAL {
>  	rte_cryptodev_get_qp_status;
> 
>  	# added in 20.11
> +	rte_cryptodev_add_enq_callback;
>  	rte_cryptodev_configure_raw_dp_ctx;
>  	rte_cryptodev_get_raw_dp_ctx_size;
>  	rte_cryptodev_raw_dequeue;
> @@ -109,4 +110,5 @@ EXPERIMENTAL {
>  	rte_cryptodev_raw_enqueue;
>  	rte_cryptodev_raw_enqueue_burst;
>  	rte_cryptodev_raw_enqueue_done;
> +	rte_cryptodev_remove_enq_callback;
>  };
> --
> 1.9.1