Re: [dpdk-dev] [PATCH 1/4] vhost: inroduce operation to get vDPA queue stats

[Maxime Coquelin <maxime.coquelin@redhat.com>]

Hi Matan,

On 4/2/20 1:26 PM, Matan Azrad wrote:
> The vDPA device offloads all the datapath of the vhost device to the HW
> device.
> 
> In order to expose to the user traffic information this patch introduce
> new API to get traffic statistics per virtio queue.
> 
> The statistics are taken directly from the vDPA driver managing the HW
> device.
> 
> Signed-off-by: Matan Azrad <matan@mellanox.com>
> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
> ---
>  doc/guides/rel_notes/release_20_05.rst    |  4 +++
>  doc/guides/vdpadevs/features/default.ini  |  1 +
>  doc/guides/vdpadevs/features_overview.rst |  3 +++
>  lib/librte_vhost/rte_vdpa.h               | 45 ++++++++++++++++++++++++++++++-
>  lib/librte_vhost/rte_vhost_version.map    |  1 +
>  lib/librte_vhost/vdpa.c                   | 14 ++++++++++
>  6 files changed, 67 insertions(+), 1 deletion(-)

...

> diff --git a/lib/librte_vhost/rte_vdpa.h b/lib/librte_vhost/rte_vdpa.h
> index 9a3deb3..d6cbf48 100644
> --- a/lib/librte_vhost/rte_vdpa.h
> +++ b/lib/librte_vhost/rte_vdpa.h
> @@ -37,6 +37,27 @@ struct rte_vdpa_dev_addr {
>  	};
>  };
>  
> +struct rte_vdpa_queue_stats {
> +	/** Number of descriptors received by device */
> +	uint64_t received_desc;
> +	/** Number of descriptors completed by the device */
> +	uint64_t completed_desc;
> +	/** Number of bad descriptors received by device */
> +	uint32_t bad_desc;
> +	/**
> +	 * Number of chained descriptors received that exceed the max allowed
> +	 * chain by device
> +	 */
> +	uint32_t exceed_max_chain;
> +	/**
> +	 * Number of times device tried to read or write buffer that is not
> +	 * registered to the device
> +	 */
> +	uint32_t invalid_buffer;
> +	/** Number of errors detected by the device */
> +	uint32_t errors;
> +};
> +

I think doing it like that, we risk to keep the rte_vdpa_get_stats API
always experimental, as every vendor will want to add their own counters
and so break the ABI.

How about implementing something similar to rte_eth_xstat?
As these stats are for debugging purpose, it would give you much more
flexibility in adding new counters as HW or firmwares evolves.

What do you think?

Thanks,
Maxime

>  /**
>   * vdpa device operations
>   */
> @@ -73,8 +94,11 @@ struct rte_vdpa_dev_ops {
>  	int (*get_notify_area)(int vid, int qid,
>  			uint64_t *offset, uint64_t *size);
>  
> +	/** Get statistics of the queue */
> +	int (*get_stats)(int did, int qid, struct rte_vdpa_queue_stats *stats);
> +
>  	/** Reserved for future extension */
> -	void *reserved[5];
> +	void *reserved[4];
>  };
>  
>  /**
> @@ -200,4 +224,23 @@ struct rte_vdpa_device *
>  __rte_experimental
>  int
>  rte_vdpa_relay_vring_used(int vid, uint16_t qid, void *vring_m);
> +
> +/**
> + * @warning
> + * @b EXPERIMENTAL: this API may change without prior notice
> + *
> + * Get vDPA device queue statistics.
> + *
> + * @param did
> + *  device id
> + * @param qid
> + *  queue id
> + * @param stats
> + *  queue statistics pointer.
> + * @return
> + *  0 on success, non-zero on failure.
> + */
> +__rte_experimental
> +int
> +rte_vdpa_get_stats(int did, uint16_t qid, struct rte_vdpa_queue_stats *stats);
>  #endif /* _RTE_VDPA_H_ */
> diff --git a/lib/librte_vhost/rte_vhost_version.map b/lib/librte_vhost/rte_vhost_version.map
> index 051d08c..c9dcff4 100644
> --- a/lib/librte_vhost/rte_vhost_version.map
> +++ b/lib/librte_vhost/rte_vhost_version.map
> @@ -38,6 +38,7 @@ EXPERIMENTAL {
>  	rte_vdpa_find_device_id;
>  	rte_vdpa_get_device;
>  	rte_vdpa_get_device_num;
> +	rte_vdpa_get_stats;
>  	rte_vhost_driver_attach_vdpa_device;
>  	rte_vhost_driver_detach_vdpa_device;
>  	rte_vhost_driver_get_vdpa_device_id;
> diff --git a/lib/librte_vhost/vdpa.c b/lib/librte_vhost/vdpa.c
> index 2b86708..57900fc 100644
> --- a/lib/librte_vhost/vdpa.c
> +++ b/lib/librte_vhost/vdpa.c
> @@ -227,3 +227,17 @@ struct rte_vdpa_device *
>  		free_ind_table(idesc);
>  	return -1;
>  }
> +
> +int
> +rte_vdpa_get_stats(int did, uint16_t qid, struct rte_vdpa_queue_stats *stats)
> +{
> +	struct rte_vdpa_device *vdpa_dev;
> +
> +	vdpa_dev = rte_vdpa_get_device(did);
> +	if (!vdpa_dev)
> +		return -ENODEV;
> +
> +	RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_stats, -ENOTSUP);
> +
> +	return vdpa_dev->ops->get_stats(did, qid, stats);
> +}
>