All of lore.kernel.org
 help / color / mirror / Atom feed
* maximum QP size
@ 2020-02-07  5:18 Dimitris Dimitropoulos
  2020-02-09 14:57 ` Leon Romanovsky
  0 siblings, 1 reply; 3+ messages in thread
From: Dimitris Dimitropoulos @ 2020-02-07  5:18 UTC (permalink / raw)
  To: linux-rdma

Hi,

I'm running the ibv_rc_pingpong example and I notice as as rx-depth
increases the QP create fails, eg at 50k. My understanding is this
variable controls the number of RECV WR that will be posted in advance
before SEND WRs are posted.

Is there a limitation on the queue size (the size is unit32_t) and if
so why ? Also is there a way around it ?

Regards
Dimitris

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

* Re: maximum QP size
  2020-02-07  5:18 maximum QP size Dimitris Dimitropoulos
@ 2020-02-09 14:57 ` Leon Romanovsky
  2020-02-10 17:38   ` Dimitris Dimitropoulos
  0 siblings, 1 reply; 3+ messages in thread
From: Leon Romanovsky @ 2020-02-09 14:57 UTC (permalink / raw)
  To: Dimitris Dimitropoulos; +Cc: linux-rdma

On Thu, Feb 06, 2020 at 09:18:00PM -0800, Dimitris Dimitropoulos wrote:
> Hi,
>
> I'm running the ibv_rc_pingpong example and I notice as as rx-depth
> increases the QP create fails, eg at 50k. My understanding is this
> variable controls the number of RECV WR that will be posted in advance
> before SEND WRs are posted.
>
> Is there a limitation on the queue size (the size is unit32_t) and if
> so why ? Also is there a way around it ?

The rx-depth is translated to the size of completion queue buffer.
That buffer used by hardware to post completions - writing completion
queue elements (CQEs) and it is allocated when creating the CQ.

Maximum number of CQEs can be retrieved by the ibv_devinfo -v, see max_cqe field.

Thanks

>
> Regards
> Dimitris

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

* Re: maximum QP size
  2020-02-09 14:57 ` Leon Romanovsky
@ 2020-02-10 17:38   ` Dimitris Dimitropoulos
  0 siblings, 0 replies; 3+ messages in thread
From: Dimitris Dimitropoulos @ 2020-02-10 17:38 UTC (permalink / raw)
  To: Leon Romanovsky; +Cc: linux-rdma

Thank you, there is a wealth of information with the "-v" switch. For
the Mellanox Connect-5 that I'm using I see:

max_cq:                         16777216
max_cqe:                        4194303

I'm encountering a smaller limit on the maximum entries, so I presume
this is the max total number of cqe for all queues in the system.

Regards
Dimitris

> > Hi,
> >
> > I'm running the ibv_rc_pingpong example and I notice as as rx-depth
> > increases the QP create fails, eg at 50k. My understanding is this
> > variable controls the number of RECV WR that will be posted in advance
> > before SEND WRs are posted.
> >
> > Is there a limitation on the queue size (the size is unit32_t) and if
> > so why ? Also is there a way around it ?
>
> The rx-depth is translated to the size of completion queue buffer.
> That buffer used by hardware to post completions - writing completion
> queue elements (CQEs) and it is allocated when creating the CQ.
>
> Maximum number of CQEs can be retrieved by the ibv_devinfo -v, see max_cqe field.

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

end of thread, other threads:[~2020-02-10 17:38 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-02-07  5:18 maximum QP size Dimitris Dimitropoulos
2020-02-09 14:57 ` Leon Romanovsky
2020-02-10 17:38   ` Dimitris Dimitropoulos

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.