Thanks Liron, I've checked the code and that seems to be another issue. The
intent is that when classifier_enable is set the ODP classifier creates
queues as needed.

+cc Bala for additional commentary.

On Tue, Oct 3, 2017 at 8:48 AM, Liron Himi <lir...@marvell.com> wrote:

> Hi Bill,
>
>
>
> I opened ‘https://bugs.linaro.org/show_bug.cgi?id=3289’ for this issue.
>
>
>
> I think there is another issue related to this topic in case PKTIO was
> configure in schedule mode.
>
> The ‘start’ function assumes ‘num_in_queue’ is bigger than ‘0’ but in the
> classifier_enable case it is ‘0’ so pktio is register with no real indexes,
> actually in ‘index’ array there is garbage. (see code below)
>
>
>
> If I want to have a local fix, what do you think should be the number of
> queues? ‘max_input_queues’ in capabilities ?
>
>
>
> Regards,
>
> Liron
>
>
>
>
>
> int odp_pktio_start(odp_pktio_t hdl)
>
> {
>
> …
>
>                 mode = entry->s.param.in_mode;
>
>
>
>                 if (mode == ODP_PKTIN_MODE_SCHED) {
>
>                                 unsigned i;
>
>                                 unsigned num = entry->s.num_in_queue;
>
>                                 int index[num];
>
>
>
>                                 for (i = 0; i < num; i++) {
>
>                                                 index[i] = i;
>
>
>
>                                                 if
> (entry->s.in_queue[i].queue == ODP_QUEUE_INVALID) {
>
>
> ODP_ERR("No input queue\n");
>
>                                                                 return -1;
>
>                                                 }
>
>                                 }
>
>
>
>                                 sched_fn->pktio_start(pktio_to_id(hdl),
> num, index);
>
>                 }
>
>
>
>                 return res;
>
> }
>
> *From:* Bill Fischofer [mailto:bill.fischo...@linaro.org]
> *Sent:* Monday, October 02, 2017 14:52
> *To:* Liron Himi <lir...@marvell.com>
> *Cc:* lng-odp@lists.linaro.org
> *Subject:* [EXT] Re: [lng-odp] odp_pktin_queue_param_t -
> classifier_enable & num_queues
>
>
>
> External Email
> ------------------------------
>
> Thanks Liron. This appears to be a bug in the implementation. It's only an
> error for num_queues to be zero if hash_enabled is set.
>
>
>
> On Mon, Oct 2, 2017 at 6:02 AM, Liron Himi <lir...@marvell.com> wrote:
>
> Hi,
>
> According to the API if 'classifier_enable' is enabled than 'num_queues'
> and 'queue_param' should be ignored.
> But looking at linux-generic implementation those parameters are always
> being checked regardless of 'classifier_enable' value.
> 'classifier_enable' is actually ignore at linux-generic implementation.
> The main problem is if 'num_queues' is '0' than 'odp_pktin_queue_config'
> return an error although it can be '0' if 'classifier_enable' is enabled.
>
> Regards,
> Liron
>
>
>

Reply via email to