DPDK patches and discussions
 help / color / mirror / Atom feed
From: Ferruh Yigit <ferruh.yigit@amd.com>
To: "lihuisong (C)" <lihuisong@huawei.com>,
	"He, ShiyangX" <shiyangx.he@intel.com>,
	"dev@dpdk.org" <dev@dpdk.org>
Cc: "Zhou, YidingX" <yidingx.zhou@intel.com>,
	"stable@dpdk.org" <stable@dpdk.org>,
	"Zhang, Yuying" <yuying.zhang@intel.com>,
	"Singh, Aman Deep" <aman.deep.singh@intel.com>,
	"Burakov, Anatoly" <anatoly.burakov@intel.com>,
	Matan Azrad <matan@nvidia.com>,
	Dmitry Kozlyuk <dmitry.kozliuk@gmail.com>
Subject: Re: [PATCH v3] app/testpmd: fix secondary process not forwarding
Date: Wed, 8 Mar 2023 09:54:51 +0000	[thread overview]
Message-ID: <6acbe8d1-d030-3fd7-7c17-ed8958fec4bb@amd.com> (raw)
In-Reply-To: <59e12a4b-529a-96d5-616a-65899e52eee7@huawei.com>

On 3/8/2023 2:54 AM, lihuisong (C) wrote:
> 
> 在 2023/3/8 10:05, He, ShiyangX 写道:
>>
>>> -----Original Message-----
>>> From: Ferruh Yigit <ferruh.yigit@amd.com>
>>> Sent: Tuesday, March 7, 2023 7:41 PM
>>> To: He, ShiyangX <shiyangx.he@intel.com>; dev@dpdk.org
>>> Cc: Zhou, YidingX <yidingx.zhou@intel.com>; stable@dpdk.org; Zhang,
>>> Yuying
>>> <yuying.zhang@intel.com>; Singh, Aman Deep
>>> <aman.deep.singh@intel.com>; Burakov, Anatoly
>>> <anatoly.burakov@intel.com>; Matan Azrad <matan@nvidia.com>; Dmitry
>>> Kozlyuk <dmitry.kozliuk@gmail.com>
>>> Subject: Re: [PATCH v3] app/testpmd: fix secondary process not
>>> forwarding
>>>
>>> On 3/7/2023 3:25 AM, He, ShiyangX wrote:
>>>>
>>>>> -----Original Message-----
>>>>> From: Ferruh Yigit <ferruh.yigit@amd.com>
>>>>> Sent: Monday, March 6, 2023 11:06 PM
>>>>> To: He, ShiyangX <shiyangx.he@intel.com>; dev@dpdk.org
>>>>> Cc: Zhou, YidingX <yidingx.zhou@intel.com>; stable@dpdk.org; Zhang,
>>>>> Yuying <yuying.zhang@intel.com>; Singh, Aman Deep
>>>>> <aman.deep.singh@intel.com>; Burakov, Anatoly
>>>>> <anatoly.burakov@intel.com>; Matan Azrad <matan@nvidia.com>; Dmitry
>>>>> Kozlyuk <dmitry.kozliuk@gmail.com>
>>>>> Subject: Re: [PATCH v3] app/testpmd: fix secondary process not
>>>>> forwarding
>>>>>
>>>>> 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: stable@dpdk.org
>>>>>>
>>>>>> Signed-off-by: Shiyang He <shiyangx.he@intel.com>
>>>>>> Acked-by: Yuying Zhang <yuying.zhang@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?
>>>> Thanks for your comments!
>>>>
>>>> It is an effective method to get queue state directly from ethdev
>>>> where this
>>> state is used.
>>>> I also don't know the design meaning of the 'ports' variable. If
>>>> modification is needed, a higher level of design and more work are
>>>> required.
>>>>
>>>> As a bug fix, apart from extracting the code block into a function,
>>>> is the
>>> solution feasible?
>>>
>>> Hi Shiyang,
>>>
>>> As a bug fix, this issue (testpmd stored state not being up to date for
>>> secondary process) looks like have potential to occur many different
>>> flavors,
>>> that is why what about having a central update?
>>>
>>> I think 'start_port()' can be a good place for this kind of update:
>>>
>>> start_port() {
>>>
>>>     ...
>>>     if (secondary)
>>>         update_state()
>>> }
>>>
>>> update_state() {
>>>     update_queue_state()
>>> }
>>>
>>> update_queue_state() {
>>>     <your code goes here>
>>> }
>>>
>>>
>>> Having secondary checks and updates in multiple places can make code
>>> harder
>>> to understand.
>>>
>>> What do you think to update as above?
>>>
>>>
>>>
>> Thanks for your reply!
>> It is more reasonable to update the queue state in 'start_port()'.
>> I'll send a new patch asap.
> It's also necessary to update the queue state when start forwarding.
> Because the state may be changed after start port.


I was hoping updating on a single point can be sufficient, is this
needed because of testpmd commands?

> The state cannot be updated in real time(because of no notification),
> but it's helpful for secondary.


  reply	other threads:[~2023-03-08  9:55 UTC|newest]

Thread overview: 24+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-12-30  7:55 [PATCH] " Shiyang He
2022-12-30 17:23 ` Stephen Hemminger
2023-01-04  2:02   ` He, ShiyangX
2023-01-13  9:07     ` He, ShiyangX
2023-02-08  3:22       ` Zhang, Yuying
2023-02-08  6:38         ` He, ShiyangX
2023-02-20  5:39           ` Zhang, Yuying
2023-02-20 12:45 ` lihuisong (C)
2023-02-21  2:52   ` He, ShiyangX
2023-02-21  6:37     ` lihuisong (C)
2023-02-21  6:51       ` He, ShiyangX
2023-02-21 15:44 ` [PATCH v2] " Shiyang He
2023-02-22  6:20   ` lihuisong (C)
2023-02-23 14:41 ` [PATCH v3] " Shiyang He
2023-02-23  8:08   ` lihuisong (C)
2023-03-06 15:16     ` Ferruh Yigit
2023-03-06 15:05   ` Ferruh Yigit
2023-03-07  3:25     ` He, ShiyangX
2023-03-07 11:41       ` Ferruh Yigit
2023-03-08  2:05         ` He, ShiyangX
2023-03-08  2:54           ` lihuisong (C)
2023-03-08  9:54             ` Ferruh Yigit [this message]
2023-03-08 16:19   ` [PATCH v4] " Shiyang He
2023-03-08 10:20     ` Ferruh Yigit

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=6acbe8d1-d030-3fd7-7c17-ed8958fec4bb@amd.com \
    --to=ferruh.yigit@amd.com \
    --cc=aman.deep.singh@intel.com \
    --cc=anatoly.burakov@intel.com \
    --cc=dev@dpdk.org \
    --cc=dmitry.kozliuk@gmail.com \
    --cc=lihuisong@huawei.com \
    --cc=matan@nvidia.com \
    --cc=shiyangx.he@intel.com \
    --cc=stable@dpdk.org \
    --cc=yidingx.zhou@intel.com \
    --cc=yuying.zhang@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).