RE: RFC: Using and renaming 8-bit reserved field of rte_crypto_op for implementation specific

[Akhil Goyal <gakhil@marvell.com>]

Hi Ganapati,

>> Is it not possible to use rte_event_crypto_adapter_enqueue
>> if you want to send the event context to cryptodev?
> [Ganapati] No, event crypto adapter sends only ev::event_ptr as rte_crypto_op to cryptodev and not event context.

>> While using rte_cryptodev_enqueue() all previous stage event context is meant to be lost and 
>> It would send a new crypto request to cryptodev and is not supposed to be aware of event context.
> [Ganapati] Yes, proposal is for sending implementation specific value from eventdev to crypodev and vice versa

As discussed in a separate mail thread, 
impl_opaque in rte_crypto_op is not generic and is specific to your use case.
The crypto dev wont be able to make difference whether to alter it (for any other usecase) or not(for event case).
As per definition of impl_opaque is meant to be consumed by driver. Hence this will contradict the usage.
Since impl_opaque in rte_crypto_op is exposed to driver. Drivers are free to use it.

You may consider using mbuf dynamic fields for setting some userdata for your use case for each packet.
See rte_security_dynfield and add a fastpath cryptodev API to set pkt userdata which is opaque to driver
(similar to rte_security_set_pkt_metadata).
Or else add a similar schema to introduce dynamic fields in rte_crypto_op.

Regards,
Akhil