DPDK patches and discussions
 help / color / mirror / Atom feed
* cryptodev API question with out of order support
@ 2023-03-20  9:28 Suanming Mou
  2023-03-20 11:58 ` Doherty, Declan
  0 siblings, 1 reply; 3+ messages in thread
From: Suanming Mou @ 2023-03-20  9:28 UTC (permalink / raw)
  To: declan.doherty; +Cc: dev

[-- Attachment #1: Type: text/plain, Size: 361 bytes --]

Hi Declan,

While reading the cryptodev API and define docs, I don't see some places mention the out of order support.
Does current crypto enqueue and dequeue function support out of order mode? Or should we add a hint capability flag for that?

Not sure if such topics have been discussed before, can you please help to clarify?

Thanks,
Suanming Mou

[-- Attachment #2: Type: text/html, Size: 2260 bytes --]

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: cryptodev API question with out of order support
  2023-03-20  9:28 cryptodev API question with out of order support Suanming Mou
@ 2023-03-20 11:58 ` Doherty, Declan
  2023-03-20 12:05   ` Suanming Mou
  0 siblings, 1 reply; 3+ messages in thread
From: Doherty, Declan @ 2023-03-20 11:58 UTC (permalink / raw)
  To: Suanming Mou; +Cc: dev

Hey Suanming,

I haven't been actively involved in the development of DPDK in a number 
of years but as far as I'm aware (and I don't believe this has changed 
recently), out of order processing from the cryptodev API user 
perspective is not supported, and for most scenarios I'm aware of (for 
symmetric crypto processing) it would not be desired, as data path 
protocols like IPsec or TLS expect packet order to be maintained.

I know the crypto scheduler PMD allows out of order processing of 
operations on worker cores, when it is working in a load balancing mode, 
but it's default behaviour is to guarantee that packets/operations are 
returned in order to the user after processing.

If there are use cases for out of order processing (maybe for asymmetric 
crypto) then I expected that at a minimum a per queue pair setup option 
would be required and possibly a completion queue mechanism might be 
required.


Regards
Declan


On 20/03/2023 09:28, Suanming Mou wrote:
> Hi Declan,
> 
> While reading the cryptodev API and define docs, I don’t see some places 
> mention the out of order support.
> 
> Does current crypto enqueue and dequeue function support out of order 
> mode? Or should we add a hint capability flag for that?
> 



> Not sure if such topics have been discussed before, can you please help 
> to clarify?
> 
> Thanks,
> 
> Suanming Mou
> 

^ permalink raw reply	[flat|nested] 3+ messages in thread

* RE: cryptodev API question with out of order support
  2023-03-20 11:58 ` Doherty, Declan
@ 2023-03-20 12:05   ` Suanming Mou
  0 siblings, 0 replies; 3+ messages in thread
From: Suanming Mou @ 2023-03-20 12:05 UTC (permalink / raw)
  To: Doherty, Declan; +Cc: dev

Hi Declan,

I see, thank you very much for the clear clarification.

BR,
Suanming

> -----Original Message-----
> From: Doherty, Declan <declan.doherty@intel.com>
> Sent: Monday, March 20, 2023 7:58 PM
> To: Suanming Mou <suanmingm@nvidia.com>
> Cc: dev@dpdk.org
> Subject: Re: cryptodev API question with out of order support
> 
> Hey Suanming,
> 
> I haven't been actively involved in the development of DPDK in a number of
> years but as far as I'm aware (and I don't believe this has changed recently), out
> of order processing from the cryptodev API user perspective is not supported,
> and for most scenarios I'm aware of (for symmetric crypto processing) it would
> not be desired, as data path protocols like IPsec or TLS expect packet order to
> be maintained.
> 
> I know the crypto scheduler PMD allows out of order processing of operations
> on worker cores, when it is working in a load balancing mode, but it's default
> behaviour is to guarantee that packets/operations are returned in order to the
> user after processing.
> 
> If there are use cases for out of order processing (maybe for asymmetric
> crypto) then I expected that at a minimum a per queue pair setup option would
> be required and possibly a completion queue mechanism might be required.
> 
> 
> Regards
> Declan
> 
> 
> On 20/03/2023 09:28, Suanming Mou wrote:
> > Hi Declan,
> >
> > While reading the cryptodev API and define docs, I don’t see some
> > places mention the out of order support.
> >
> > Does current crypto enqueue and dequeue function support out of order
> > mode? Or should we add a hint capability flag for that?
> >
> 
> 
> 
> > Not sure if such topics have been discussed before, can you please
> > help to clarify?
> >
> > Thanks,
> >
> > Suanming Mou
> >

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2023-03-20 12:05 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-03-20  9:28 cryptodev API question with out of order support Suanming Mou
2023-03-20 11:58 ` Doherty, Declan
2023-03-20 12:05   ` Suanming Mou

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).