dpdk: Inquiry about vring cleanup during packets transmission

dpdk: Inquiry about vring cleanup during packets transmission

[wangzengyuan]

[-- Attachment #1: Type: text/plain, Size: 1006 bytes --]

Hi,
         I am writing to inquire about the vring cleanup process during packets transmission.
In the virtio_xmit_pkts function, there is the following code:
         nb_used = virtqueue_nused(vq);

         if (likely(nb_used > vq->vq_nentries - vq->vq_free_thresh))
                   virtio_xmit_cleanup(vq, nb_used);
In other words, cleaning is performed when the number of items used in the vring exceeds (vq->vq_nentries - vq->vq_free_thresh). In the case of an vring size of 4096, at least (4096-32) items need to be cleaned at once, which will take a considerable amount of time.
I'm curious why not clean up fewer items each time to avoid taking up too much CPU time in one transmission. Because during the debugging process, I found that cleaning up thousands of items at once takes up a considerable amount of time.
As I am not familiar with this process, I would appreciate it if you could provide me with some information on what its purpose is.

Best regards,

Zengyuan Wang

[-- Attachment #2: Type: text/html, Size: 4241 bytes --]

Re: dpdk: Inquiry about vring cleanup during packets transmission

[Maxime Coquelin]

Hello,

On 5/27/23 09:08, wangzengyuan wrote:
> Hi,
> 
>           I am writing to inquire about the vring cleanup process during 
> packets transmission.
> 
> In the virtio_xmit_pkts function, there is the following code:
> 
>           nb_used = virtqueue_nused(vq);
> 
>           if (likely(nb_used > vq->vq_nentries - vq->vq_free_thresh))
> 
>                     virtio_xmit_cleanup(vq, nb_used);
> 
> In other words, cleaning is performed when the number of items used in 
> the vring exceeds (vq->vq_nentries - vq->vq_free_thresh). In the case of 
> an vring size of 4096, at least (4096-32) items need to be cleaned at 
> once, which will take a considerable amount of time.
> 
> I'm curious why not clean up fewer items each time to avoid taking up 
> too much CPU time in one transmission. Because during the debugging 
> process, I found that cleaning up thousands of items at once takes up a 
> considerable amount of time.
> 
> As I am not familiar with this process, I would appreciate it if you 
> could provide me with some information on what its purpose is.

Both the Tx and Rx queues free threshold are configurable via ethdev
APIs:

int rte_eth_tx_queue_setup(uint16_t port_id, uint16_t tx_queue_id,
		uint16_t nb_tx_desc, unsigned int socket_id,
		const struct rte_eth_txconf *tx_conf);

int rte_eth_rx_queue_setup(uint16_t port_id, uint16_t rx_queue_id,
		uint16_t nb_rx_desc, unsigned int socket_id,
		const struct rte_eth_rxconf *rx_conf,
		struct rte_mempool *mb_pool);

As you are using large rings, your application may use above APIs to set 
more appropriate values.

Regards,
Maxime

> Best regards,
> 
> Zengyuan Wang
>