[dpdk-users] [Crypto-QAT] queue pairs per QAT device

[dpdk-users] [Crypto-QAT] queue pairs per QAT device

[Meng Wang]

Hi,

I'm trying to understand how queue pairs get setup on QAT. I have 
created one VF for each lcore. From qat code it seems support up to 2 
qps per device/VF, one for each direction. Then I found that each queue 
pair has its own qat_queue rx/tx.

/* Maximum number of qps on a device for any service type */
#define ADF_MAX_QPS_ON_ANY_SERVICE    2
#define ADF_RING_DIR_TX            0
#define ADF_RING_DIR_RX            1

How does the direction of qp correlate to direction of qat_queue? Is one 
queue pair sufficient to transmit packet between QAT and application per 
core?

Thanks,
Meng

Re: [dpdk-users] [Crypto-QAT] queue pairs per QAT device

[Trahe, Fiona]

Hi Meng,

> -----Original Message-----
> From: users [mailto:users-bounces@dpdk.org] On Behalf Of Meng Wang
> Sent: Monday, May 13, 2019 7:47 PM
> To: users@dpdk.org
> Subject: [dpdk-users] [Crypto-QAT] queue pairs per QAT device
> 
> Hi,
> 
> I'm trying to understand how queue pairs get setup on QAT. I have
> created one VF for each lcore. From qat code it seems support up to 2
> qps per device/VF, one for each direction. Then I found that each queue
> pair has its own qat_queue rx/tx.
> 
> /* Maximum number of qps on a device for any service type */
> #define ADF_MAX_QPS_ON_ANY_SERVICE    2
> #define ADF_RING_DIR_TX            0
> #define ADF_RING_DIR_RX            1
> 
> How does the direction of qp correlate to direction of qat_queue? Is one
> queue pair sufficient to transmit packet between QAT and application per
> core?
> 
> Thanks,
> Meng
[Fiona] There is no direction associated with a qp - as you noted each queue pair
has both a transmit and a receive queue. (or 2 rings, if you're familiar with other QAT drivers)
So yes, one qp is sufficient and you should enqueue and dequeue to/from the same qp.