On 9/8/2023 12:50 PM, David Marchand wrote:
> On Fri, Sep 8, 2023 at 1:32 PM Jie Hai <haij...@huawei.com> wrote:
>>
>> The DPDK framework reports the queue state, which is stored in
>> dev->data->tx_queue_state and dev->data->rx_queue_state. The
>> state is maintained by the driver. Users may determine whether
>> a queue participates in packet forwarding based on the state,
>> for example,
>
> The driver is maintaining this state in dev_start / dev_stop and per
> queue start/stop handlers.
>
>>
>> [1] 5028f207a4fa ("app/testpmd: fix secondary process packet forwarding"
>> [2] 141a520b35f7 ("app/testpmd: fix primary process not polling all queues")
>>
>> Therefore, the drivers need to modify the queue state in time
>> according to the actual situation, especially when dev_start
>> and dev_stop are called. see [3] for more information.
>>
>> [3] https://inbox.dpdk.org/dev/20230721160422.3848154-1-ferruh.yi...@amd.com/
>>
>> This patchset also resubmit the patch [2] and makes some fixes on the patch.
>
> I just had a quick look at some patches and I wonder if a better fix
> would be at the ethdev level, rather than fixing all drivers.
>
>
I came here to make the same comment,
Jie, I forgot if we discuss this already but,
does it work if we update queue state in 'rte_eth_dev_start()' &
'rte_eth_dev_stop()' when 'dev_start' & 'dev_stop' dev_ops succeeds?
This solves the case driver forgets to update the queue state.
