Yes, atomicity and ordering are scheduler features. Polled queues don’t have 
those features. Also scheduled queues cannot be polled and we’ll have to keep 
it that way to ensure HW support for scheduling.

-Petri


From: [email protected] 
[mailto:[email protected]] On Behalf Of ext Alexandru Badicioiu
Sent: Thursday, September 04, 2014 3:38 PM
To: Robbie King (robking)
Cc: [email protected]
Subject: Re: [lng-odp] [PATCHv3 0/4] IPsec example application

Hi Robbie,
I think polled queues can't be ATOMIC or ORDERED. When a queue is created these 
attributes are part of the scheduling parameters:
qparam.sched.sync  = ODP_SCHED_SYNC_ORDERED;

Thanks,
Alex

On 4 September 2014 15:34, Robbie King (robking) 
<[email protected]<mailto:[email protected]>> wrote:
This is a good segue to a question I have.  When polling queues created
as ORDERED or ATOMIC, are multiple cores able to poll these queues
simultaneously and order/atomicity will be preserved?  Or with polled
queues is the burden on the application.

If ordering and atomicity is guaranteed by the implementation when only
polled queues are supported, then I think it's fair to say that applications
can choose polled versus scheduled.  If that's not true and implementations
that only support polling have no ordering/atomicity guarantees, then that
seems to be more like DPDK versus ODP.

My apologies if these details are already spelled out somewhere.

Thanks,
Robbie

-----Original Message-----
From: Anders Roxell 
[mailto:[email protected]<mailto:[email protected]>]
Sent: Thursday, September 04, 2014 8:25 AM
To: Taras Kondratiuk
Cc: Robbie King (robking); 
[email protected]<mailto:[email protected]>; Mike Holmes (Google 
Drive) ([email protected]<mailto:[email protected]>)
Subject: Re: [lng-odp] [PATCHv3 0/4] IPsec example application
On 2014-09-04 09:23, Taras Kondratiuk wrote:
> On 09/03/2014 09:14 PM, Robbie King (robking) wrote:
> > Taras, do you have a bug ID for the "IPSEC_POLL_QUEUES" workaround?
> > Anders has requested that be included in my IPsec application patch.
>
> Anders, why do you think it need to be tracked as a bug? It is just a
> different way of using API. Not all applications must use scheduler
> API. We have the same in all our packet examples.

I just had a problem with this sentence:
"Some platforms require a temporary hack to get around using
odp_schedule"

"temporary hack" in my ears that should be a bug, maybe we can rename it
and say that this isn't a hack?

and I asked if this should be a bug.

Cheers,
Anders
_______________________________________________
lng-odp mailing list
[email protected]<mailto:[email protected]>
http://lists.linaro.org/mailman/listinfo/lng-odp

_______________________________________________
lng-odp mailing list
[email protected]
http://lists.linaro.org/mailman/listinfo/lng-odp

Reply via email to