On 24/06/2021 16:04, David Marchand wrote: > On Fri, Jun 4, 2021 at 11:19 PM Kevin Traynor <[email protected]> wrote: >> >> Previously, if pmd-rxq-affinity was used to pin an rxq to >> a core that was not in pmd-cpu-mask the rxq was not polled >> and the user received a warning. >> >> Now that pinned and non-pinned rxqs are assigned to PMDs in >> a common call to rxq scheduling, if an invalid core is >> selected in pmd-rxq-affinity the rxq can be assigned an >> available PMD (if any). >> >> A warning will still be logged as the requested core could >> not be used. >> >> Signed-off-by: Kevin Traynor <[email protected]> >> --- >> Documentation/topics/dpdk/pmd.rst | 6 +++--- >> lib/dpif-netdev.c | 30 ++++++++++++++++++++++++++++-- >> tests/pmd.at | 5 ++++- >> 3 files changed, 35 insertions(+), 6 deletions(-) >> >> diff --git a/Documentation/topics/dpdk/pmd.rst >> b/Documentation/topics/dpdk/pmd.rst >> index d1c45cdfb..29ba53954 100644 >> --- a/Documentation/topics/dpdk/pmd.rst >> +++ b/Documentation/topics/dpdk/pmd.rst >> @@ -108,7 +108,7 @@ means that this thread will only poll the *pinned* Rx >> queues. >> >> If there are no *non-isolated* PMD threads, *non-pinned* RX queues will >> not >> - be polled. Also, if the provided ``<core-id>`` is not available (e.g. the >> - ``<core-id>`` is not in ``pmd-cpu-mask``), the RX queue will not be >> polled >> - by any PMD thread. >> + be polled. If the provided ``<core-id>`` is not available (e.g. the >> + ``<core-id>`` is not in ``pmd-cpu-mask``), the RX queue will be assigned >> to >> + a *non-isolated* PMD, that will remain *non-isolated*. >> >> If ``pmd-rxq-affinity`` is not set for Rx queues, they will be assigned to >> PMDs >> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c >> index 61e0a516f..377573233 100644 >> --- a/lib/dpif-netdev.c >> +++ b/lib/dpif-netdev.c >> @@ -5027,4 +5027,25 @@ find_sched_pmd_by_pmd(struct sched_numa_list >> *numa_list, >> } >> >> +static struct sched_pmd * >> +find_sched_pmd_by_rxq(struct sched_numa_list *numa_list, >> + struct dp_netdev_rxq *rxq) >> +{ >> + struct sched_numa *numa; >> + >> + HMAP_FOR_EACH (numa, node, &numa_list->numas) { >> + for (unsigned i = 0; i < numa->n_pmds; i++) { >> + struct sched_pmd *sched_pmd; >> + >> + sched_pmd = &numa->pmds[i]; >> + for (int k = 0; k < sched_pmd->n_rxq; k++) { >> + if (sched_pmd->rxqs[k] == rxq) { >> + return sched_pmd; >> + } >> + } >> + } >> + } >> + return NULL; >> +} >> + >> static struct sched_numa * >> sched_numa_list_find_numa(struct sched_numa_list *numa_list, >> @@ -5408,5 +5429,6 @@ sched_numa_list_schedule(struct sched_numa_list >> *numa_list, >> "Core %2u cannot be pinned with " >> "port \'%s\' rx queue %d. Use pmd-cpu-mask to " >> - "enable a pmd on core %u.", >> + "enable a pmd on core %u. An alternative core " >> + "will be assigned.", >> rxq->core_id, >> netdev_rxq_get_name(rxq->rx), > > To continue with my suggestion on patch 1 on only pushing unpinned > queues to the rxqs array, you could add this rxq in rxqs[] ... > >> @@ -5453,5 +5475,9 @@ sched_numa_list_schedule(struct sched_numa_list >> *numa_list, >> >> if (rxq->core_id != OVS_CORE_UNSPEC) { >> - continue; >> + /* This rxq should have been pinned, check it was. */ >> + sched_pmd = find_sched_pmd_by_rxq(numa_list, rxq); >> + if (sched_pmd && sched_pmd->pmd->core_id == rxq->core_id) { >> + continue; >> + } > > ... and this check is unnneded. > > WDYT? >
ok, you've convinced me and as a bonus, i don't need find_sched_pmd_by_rxq() anymore, so it simplifies the code here and removes some other code. Thanks for the nice suggestion. >> } >> >> diff --git a/tests/pmd.at b/tests/pmd.at >> index 78105bf45..55977632a 100644 >> --- a/tests/pmd.at >> +++ b/tests/pmd.at >> @@ -552,7 +552,10 @@ AT_CHECK([ovs-vsctl set Open_vSwitch . >> other_config:pmd-cpu-mask=6]) >> >> dnl We removed the cores requested by some queues from pmd-cpu-mask. >> -dnl Those queues will not be polled. >> +dnl Those queues will be polled by remaining non-isolated pmds. >> AT_CHECK([ovs-appctl dpif-netdev/pmd-rxq-show | parse_pmd_rxq_show], [0], >> [dnl >> +p1 0 0 1 >> +p1 1 0 1 >> p1 2 0 2 >> +p1 3 0 1 >> ]) >> >> -- >> 2.31.1 >> > > > _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
