On 2/23/2023 2:41 PM, Shiyang He wrote:
> Under multi-process scenario, the secondary process gets queue state
> from the wrong location (the global variable 'ports'). Therefore, the
> secondary process can not forward since "stream_init" is not called.
>
> This commit fixes the issue by calling 'rte_eth_rx/tx_queue_info_get'
> to get queue state from shared memory.
>
> Fixes: 3c4426db54fc ("app/testpmd: do not poll stopped queues")
> Cc: sta...@dpdk.org
>
> Signed-off-by: Shiyang He <shiyangx...@intel.com>
> Acked-by: Yuying Zhang <yuying.zh...@intel.com>
>
> v3: Add return value description
> ---
> app/test-pmd/testpmd.c | 45 ++++++++++++++++++++++++++++++++++++++++--
> 1 file changed, 43 insertions(+), 2 deletions(-)
>
> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
> index 0c14325b8d..a050472aea 100644
> --- a/app/test-pmd/testpmd.c
> +++ b/app/test-pmd/testpmd.c
> @@ -2418,9 +2418,50 @@ start_packet_forwarding(int with_tx_first)
> if (!pkt_fwd_shared_rxq_check())
> return;
>
> - if (stream_init != NULL)
> - for (i = 0; i < cur_fwd_config.nb_fwd_streams; i++)
> + if (stream_init != NULL) {
> + for (i = 0; i < cur_fwd_config.nb_fwd_streams; i++) {
> + if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
> + struct fwd_stream *fs = fwd_streams[i];
> + struct rte_eth_rxq_info rx_qinfo;
> + struct rte_eth_txq_info tx_qinfo;
> + int32_t rc;
> + rc = rte_eth_rx_queue_info_get(fs->rx_port,
> + fs->rx_queue, &rx_qinfo);
> + if (rc == 0) {
> +
> ports[fs->rx_port].rxq[fs->rx_queue].state =
> + rx_qinfo.queue_state;
> + } else if (rc == -ENOTSUP) {
> + /* Set the rxq state to
> RTE_ETH_QUEUE_STATE_STARTED
> + * to ensure that the PMDs do not
> implement
> + * rte_eth_rx_queue_info_get can
> forward.
> + */
> +
> ports[fs->rx_port].rxq[fs->rx_queue].state =
> + RTE_ETH_QUEUE_STATE_STARTED;
> + } else {
> + TESTPMD_LOG(WARNING,
> + "Failed to get rx queue
> info\n");
> + }
> +
> + rc = rte_eth_tx_queue_info_get(fs->tx_port,
> + fs->tx_queue, &tx_qinfo);
> + if (rc == 0) {
> +
> ports[fs->tx_port].txq[fs->tx_queue].state =
> + tx_qinfo.queue_state;
> + } else if (rc == -ENOTSUP) {
> + /* Set the txq state to
> RTE_ETH_QUEUE_STATE_STARTED
> + * to ensure that the PMDs do not
> implement
> + * rte_eth_tx_queue_info_get can
> forward.
> + */
> +
> ports[fs->tx_port].txq[fs->tx_queue].state =
> + RTE_ETH_QUEUE_STATE_STARTED;
> + } else {
> + TESTPMD_LOG(WARNING,
> + "Failed to get tx queue
> info\n");
> + }
> + }
> stream_init(fwd_streams[i]);
> + }
> + }
>
Testpmd duplicates some dpdk/ethdev state/config in application level,
and this can bite in multiple cases, as it is happening here.
I am not sure if this was a design decision, but I think instead of
testpmd storing ethdev related state/config in application level, it
should store only application level state/config, and when ethdev
related state/config is required app should get it directly from ethdev.
It may be too late already for testpmd, there is a mixed usage, but I am
for preferring this approach when there is an opportunity.
For above issue, why queue state needs to be stored in application level
'port' variable?
Where is this queue state used?
Can it work to get queue state directly from ethdev where this state is
used, instead of storing it in the 'port' variable in advance?
And perhaps testpmd 'port' variable can be updated there, both for
primary and secondary, for backward compatibility (other existing users
of this queue state).
What do you think?
