[PATCH / RFC] sched: Correct subport calcuation

[PATCH / RFC] sched: Correct subport calcuation

[Simon Kagstrom]

Signed-off-by: Simon Kagstrom <simon.kagstrom@netinsight.net>
---
I'm a total newbie to the rte_sched design and implementation, so I've
added the RFC.

We get crashes (at other places in the scheduler) without this code.

 lib/librte_sched/rte_sched.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/lib/librte_sched/rte_sched.c b/lib/librte_sched/rte_sched.c
index 1609ea8..b46ecfb 100644
--- a/lib/librte_sched/rte_sched.c
+++ b/lib/librte_sched/rte_sched.c
@@ -1869,7 +1869,7 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
 
 	/* Install new pipe in the grinder */
 	grinder->pindex = pipe_qindex >> 4;
-	grinder->subport = port->subport + (grinder->pindex / port->n_pipes_per_subport);
+	grinder->subport = port->subport + (grinder->pindex / port->n_subports_per_port);
 	grinder->pipe = port->pipe + grinder->pindex;
 	grinder->pipe_params = NULL; /* to be set after the pipe structure is prefetched */
 	grinder->productive = 0;
-- 
1.9.1

Re: [PATCH / RFC] sched: Correct subport calcuation

[Simon Kågström]

Hi again!

Any news about this patch? I'm off for parental leave starting next week
(until january), so any comments (or simply dropping it!) would be good
to have before that :-)

// Simon

On 2016-06-10 08:29, Simon Kagstrom wrote:
> Signed-off-by: Simon Kagstrom <simon.kagstrom@netinsight.net>
> ---
> I'm a total newbie to the rte_sched design and implementation, so I've
> added the RFC.
> 
> We get crashes (at other places in the scheduler) without this code.
> 
>  lib/librte_sched/rte_sched.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/lib/librte_sched/rte_sched.c b/lib/librte_sched/rte_sched.c
> index 1609ea8..b46ecfb 100644
> --- a/lib/librte_sched/rte_sched.c
> +++ b/lib/librte_sched/rte_sched.c
> @@ -1869,7 +1869,7 @@ grinder_next_pipe(struct rte_sched_port *port, uint32_t pos)
>  
>  	/* Install new pipe in the grinder */
>  	grinder->pindex = pipe_qindex >> 4;
> -	grinder->subport = port->subport + (grinder->pindex / port->n_pipes_per_subport);
> +	grinder->subport = port->subport + (grinder->pindex / port->n_subports_per_port);
>  	grinder->pipe = port->pipe + grinder->pindex;
>  	grinder->pipe_params = NULL; /* to be set after the pipe structure is prefetched */
>  	grinder->productive = 0;
> 

Re: [PATCH / RFC] sched: Correct subport calcuation

[Dumitrescu, Cristian]

Hi Simon,

I am going to take a look at it this week and come back to you.

Thanks,
Cristian

> -----Original Message-----
> From: Simon Kågström [mailto:simon.kagstrom@netinsight.net]
> Sent: Tuesday, June 21, 2016 7:41 AM
> To: Dumitrescu, Cristian <cristian.dumitrescu@intel.com>;
> stephen@networkplumber.org; dev@dpdk.org;
> thomas.monjalon@6wind.com
> Subject: Re: [dpdk-dev] [PATCH / RFC] sched: Correct subport calcuation
> 
> Hi again!
> 
> Any news about this patch? I'm off for parental leave starting next week
> (until january), so any comments (or simply dropping it!) would be good
> to have before that :-)
> 
> // Simon
> 
> On 2016-06-10 08:29, Simon Kagstrom wrote:
> > Signed-off-by: Simon Kagstrom <simon.kagstrom@netinsight.net>
> > ---
> > I'm a total newbie to the rte_sched design and implementation, so I've
> > added the RFC.
> >
> > We get crashes (at other places in the scheduler) without this code.
> >
> >  lib/librte_sched/rte_sched.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/lib/librte_sched/rte_sched.c b/lib/librte_sched/rte_sched.c
> > index 1609ea8..b46ecfb 100644
> > --- a/lib/librte_sched/rte_sched.c
> > +++ b/lib/librte_sched/rte_sched.c
> > @@ -1869,7 +1869,7 @@ grinder_next_pipe(struct rte_sched_port *port,
> uint32_t pos)
> >
> >  	/* Install new pipe in the grinder */
> >  	grinder->pindex = pipe_qindex >> 4;
> > -	grinder->subport = port->subport + (grinder->pindex / port-
> >n_pipes_per_subport);
> > +	grinder->subport = port->subport + (grinder->pindex / port-
> >n_subports_per_port);
> >  	grinder->pipe = port->pipe + grinder->pindex;
> >  	grinder->pipe_params = NULL; /* to be set after the pipe structure is
> prefetched */
> >  	grinder->productive = 0;
> >

Re: [PATCH / RFC] sched: Correct subport calcuation

[Dumitrescu, Cristian]

> -----Original Message-----
> From: Simon Kagstrom [mailto:simon.kagstrom@netinsight.net]
> Sent: Friday, June 10, 2016 7:29 AM
> To: Dumitrescu, Cristian <cristian.dumitrescu@intel.com>;
> stephen@networkplumber.org; dev@dpdk.org;
> thomas.monjalon@6wind.com
> Subject: [PATCH / RFC] sched: Correct subport calcuation
> 
> Signed-off-by: Simon Kagstrom <simon.kagstrom@netinsight.net>
> ---
> I'm a total newbie to the rte_sched design and implementation, so I've
> added the RFC.
> 
> We get crashes (at other places in the scheduler) without this code.
> 
>  lib/librte_sched/rte_sched.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/lib/librte_sched/rte_sched.c b/lib/librte_sched/rte_sched.c
> index 1609ea8..b46ecfb 100644
> --- a/lib/librte_sched/rte_sched.c
> +++ b/lib/librte_sched/rte_sched.c
> @@ -1869,7 +1869,7 @@ grinder_next_pipe(struct rte_sched_port *port,
> uint32_t pos)
> 
>  	/* Install new pipe in the grinder */
>  	grinder->pindex = pipe_qindex >> 4;
> -	grinder->subport = port->subport + (grinder->pindex / port-
> >n_pipes_per_subport);
> +	grinder->subport = port->subport + (grinder->pindex / port-
> >n_subports_per_port);
>  	grinder->pipe = port->pipe + grinder->pindex;
>  	grinder->pipe_params = NULL; /* to be set after the pipe structure is
> prefetched */
>  	grinder->productive = 0;
> --
> 1.9.1

Hi Simon,

NACK.

Each port has an array of queues (size is port->n_queues_per_port), which are organized into equal size groups associated with pipes and subports:
- Each pipe is assigned the next group of RTE_SCHED_QUEUES_PER_PIPE (i.e. 16) queues in ascending order;
- Each subport is assigned the next group of port->n_pipes_per_subport pipes (congurable parameter).
The following is true:
	n_queues_per_port = RTE_SCHED_QUEUES_PER_PIPE * n_pipes_per_subport * n_subports_per_port

Given a queue index (pipe_qindex), the following are true:
- Pipe index: pindex = pipe_qindex >> 4;
- Subport index (let's call it sindex): sindex = pindex / n_pipes_per_subport, right?

I don't know why you get crashes in your application, but what I do know is that your proposed method to compute sindex is not correct :)

Regards,
Cristian