Hi DPDK,
Any comments on this proposal?
Thanks,
Ganapati
From: Kundapura, Ganapati <ganapati.kundapura@intel.com>
Sent: Wednesday, March 6, 2024 10:27 AM
To: Akhil Goyal <gakhil@marvell.com>; dpdk-dev <dev@dpdk.org>; fanzhang.oss@gmail.com; Ji, Kai <kai.ji@intel.com>; Power, Ciara <ciara.power@intel.com>; Kusztal, ArkadiuszX <arkadiuszx.kusztal@intel.com>; Gujjar, Abhinandan S <abhinandan.gujjar@intel.com>;
Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Jerin Jacob <jerinjacobk@gmail.com>
Subject: RE: RFC: Using and renaming 8-bit reserved field of rte_crypto_op for implementation specific
Hi Akhil,
No changes in sequence of API’s by adding ‘uint8_t impl_opaque’ to ‘struct rte_crypto_op’.
It’s required in case application/event dispatcher passes some implementation specific value in rte_event::impl_opaque, to restore the value
back on to rte_event::impl_opaque after enqueue to and dequeue from cryptodev.
Here is the pseudocode for one of the use case
Application/event dispatcher passes implementation specific value in rte_event::impl_opaque.
struct rte_event ev;
rte_event_dequeue_burst(…, &ev, …)
struct rte_crypto_op *crypto_op = ev.event_ptr; // ev.impl_opaque some implementation specific value
rte_cryptodev_enqueue_burst(…, crypto_op, …) ; // ev.impl_opaque is not passed to crypto_op
With rte_crypto_op::impl_opaque field which is unchanged in library/driver
crypto_op->impl_opaque = ev.impl_opaque;
rte_cryptodev_enqueue_burst(…, crypto_op, …) ;
…
rte_crypto_dequeue_burst(…, crypto_op, …)
ev.event_ptr = crypto_op;
…
rte_event_enqueue_burst(…, &ev, …); // ev::impl_opaque value is lost
with rte_crypto_op::impl_opaque field
ev.event_ptr = crypto_op;
ev.impl_opaque = crypto_op->impl_opaque; // implementation specific value in rte_event::impl_opaque restored back
rte_event_enqueue_burst(…, &ev, …);
Thanks,
Ganapati
From: Akhil Goyal <gakhil@marvell.com>
Sent: Tuesday, March 5, 2024 10:18 PM
To: Kundapura, Ganapati <ganapati.kundapura@intel.com>; dpdk-dev <dev@dpdk.org>;
fanzhang.oss@gmail.com; Ji, Kai <kai.ji@intel.com>; Power, Ciara <ciara.power@intel.com>; Kusztal, ArkadiuszX <arkadiuszx.kusztal@intel.com>;
Gujjar, Abhinandan S <abhinandan.gujjar@intel.com>; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Jerin Jacob <jerinjacobk@gmail.com>
Subject: RE: RFC: Using and renaming 8-bit reserved field of rte_crypto_op for implementation specific
Hi Ganapati,
Can you please explain the flow with a sequence of APIs to be used.
Regards,
Akhil
From: Kundapura, Ganapati <ganapati.kundapura@intel.com>
Sent: Tuesday, March 5, 2024 12:44 PM
To: dpdk-dev <dev@dpdk.org>; Akhil Goyal <gakhil@marvell.com>;
fanzhang.oss@gmail.com; Ji, Kai <kai.ji@intel.com>; Power, Ciara <ciara.power@intel.com>; Kusztal, ArkadiuszX <arkadiuszx.kusztal@intel.com>;
Gujjar, Abhinandan S <abhinandan.gujjar@intel.com>; Jayatheerthan, Jay <jay.jayatheerthan@intel.com>; Jerin Jacob <jerinjacobk@gmail.com>
Subject: [EXTERNAL] RFC: Using and renaming 8-bit reserved field of rte_crypto_op for implementation specific
Prioritize security for external emails: Confirm sender and content safety before clicking links or opening attachments
Hi dpdk-dev,
Can 'uint8_t reserved[1]' of 'struct rte_crypto_op' be renamed
to 'uint8_t impl_opaque' for implementation specific?
An implementation may use this field to hold implementation specific
value to share value between dequeue and enqueue operation and crypto library/driver
can also use this field to share implementation specfic value to event crypto adapter/application.
'struct rte_event' has 'uint8_t impl_opaque' member
struct rte_event {
...
uint8_t impl_opaque;
/**< Implementation specific opaque value.
* An implementation may use this field to hold
* implementation specific value to share between
* dequeue and enqueue operation.
* The application should not modify this field.
*/
...
};
Event crypto adapter, on dequeuing the event, enqueues rte_event::event_ptr
to cryptodev as rte_crypto_op and converts the dequeued crypto op to rte_event
without restoring the implementation specific opaque value.
By having the 'uint8_t impl_opaque' member in 'struct rte_crypto_op' as
diff --git a/lib/cryptodev/rte_crypto.h b/lib/cryptodev/rte_crypto.h
index dbc2700..af46ec9 100644
--- a/lib/cryptodev/rte_crypto.h
+++ b/lib/cryptodev/rte_crypto.h
@@ -146,10 +146,13 @@ struct rte_crypto_op {
/**< TLS record */
} param1;
/**< Additional per operation parameter 1. */
- uint8_t reserved[1];
- /**< Reserved bytes to fill 64 bits for
- * future additions
+ uint8_t impl_opaque;
+ /**< Implementation specific opaque value.
+ * An implementation may use this field to hold
+ * implementation specific value to share between
+ * dequeue and enqueue operation.
*/
+
which is untouched in library/driver and rte_event::impl_opaque field can be restored
while enqueuing the event back to eventdev.
Also crypto library/driver can use rte_crypto_op::impl_opaque field to
share implementation specific opaque value to the event crypto adapter/application.
I look forward to feedback on this proposal. Patch will be submitted
for review once the initial feedback is received.
Thank you,
Ganapati