DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Li, Xiaoyun" <xiaoyun.li@intel.com>
To: "Yigit, Ferruh" <ferruh.yigit@intel.com>,
	"Wang, Jie1X" <jie1x.wang@intel.com>,
	"dev@dpdk.org" <dev@dpdk.org>
Cc: "andrew.rybchenko@oktetlabs.ru" <andrew.rybchenko@oktetlabs.ru>,
	"thomas@monjalon.net" <thomas@monjalon.net>,
	"jerinj@marvell.com" <jerinj@marvell.com>,
	"Ananyev, Konstantin" <konstantin.ananyev@intel.com>
Subject: Re: [dpdk-dev] [PATCH v8 2/2] app/testpmd: fix testpmd doesn't show RSS hash offload
Date: Sat, 18 Sep 2021 02:18:46 +0000	[thread overview]
Message-ID: <DM4PR11MB55343DD0A005F6BC6264F7F399DE9@DM4PR11MB5534.namprd11.prod.outlook.com> (raw)
In-Reply-To: <9d302c27-572b-2d03-4286-a19bc0b77779@intel.com>

Hi

> -----Original Message-----
> From: Yigit, Ferruh <ferruh.yigit@intel.com>
> Sent: Friday, September 17, 2021 18:20
> To: Li, Xiaoyun <xiaoyun.li@intel.com>; Wang, Jie1X <jie1x.wang@intel.com>;
> dev@dpdk.org
> Cc: andrew.rybchenko@oktetlabs.ru; thomas@monjalon.net;
> jerinj@marvell.com; Ananyev, Konstantin <konstantin.ananyev@intel.com>
> Subject: Re: [PATCH v8 2/2] app/testpmd: fix testpmd doesn't show RSS hash
> offload
> 
> On 9/9/2021 4:31 AM, Li, Xiaoyun wrote:
> > Hi
> >
> >> -----Original Message-----
> >> From: Yigit, Ferruh <ferruh.yigit@intel.com>
> >> Sent: Thursday, September 9, 2021 00:51
> >> To: Wang, Jie1X <jie1x.wang@intel.com>; dev@dpdk.org; Li, Xiaoyun
> >> <xiaoyun.li@intel.com>
> >> Cc: andrew.rybchenko@oktetlabs.ru; thomas@monjalon.net
> >> Subject: Re: [PATCH v8 2/2] app/testpmd: fix testpmd doesn't show RSS
> >> hash offload
> >>
> >> On 8/27/2021 9:17 AM, Jie Wang wrote:
> >>> The driver may change offloads info into dev->data->dev_conf in
> >>> dev_configure which may cause port->dev_conf and port->rx_conf
> >>> contain outdated values.
> >>>
> >>> This patch updates the offloads info if it changes to fix this issue.
> >>>
> >>> Fixes: ce8d561418d4 ("app/testpmd: add port configuration settings")
> >>>
> >>> Signed-off-by: Jie Wang <jie1x.wang@intel.com>
> >>> ---
> >>>  app/test-pmd/testpmd.c | 34 ++++++++++++++++++++++++++++++++++
> >>>  app/test-pmd/testpmd.h |  2 ++
> >>>  app/test-pmd/util.c    | 15 +++++++++++++++
> >>>  3 files changed, 51 insertions(+)
> >>>
> >>> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c index
> >>> 6cbe9ba3c8..bd67291160 100644
> >>> --- a/app/test-pmd/testpmd.c
> >>> +++ b/app/test-pmd/testpmd.c
> >>> @@ -2461,6 +2461,9 @@ start_port(portid_t pid)
> >>>  		}
> >>>
> >>>  		if (port->need_reconfig > 0) {
> >>> +			struct rte_eth_conf dev_conf_info;
> >>> +			int k;
> >>> +
> >>>  			port->need_reconfig = 0;
> >>>
> >>>  			if (flow_isolate_all) {
> >>> @@ -2498,6 +2501,37 @@ start_port(portid_t pid)
> >>>  				port->need_reconfig = 1;
> >>>  				return -1;
> >>>  			}
> >>> +			/* get rte_eth_conf info */
> >>> +			if (0 !=
> >>> +				eth_dev_conf_info_get_print_err(pi,
> >>> +							&dev_conf_info)) {
> >>> +				fprintf(stderr,
> >>> +					"port %d can not get device
> >> configuration info\n",
> >>> +					pi);
> >>> +				return -1;
> >>> +			}
> >>> +			/* Apply Rx offloads configuration */
> >>> +			if (dev_conf_info.rxmode.offloads !=
> >>> +				port->dev_conf.rxmode.offloads) {
> >>> +				port->dev_conf.rxmode.offloads =
> >>> +					dev_conf_info.rxmode.offloads;
> >>> +				for (k = 0;
> >>> +				     k < port->dev_info.max_rx_queues;
> >>> +				     k++)
> >>> +					port->rx_conf[k].offloads =
> >>> +
> >> 	dev_conf_info.rxmode.offloads;
> >>> +			}
> >>> +			/* Apply Tx offloads configuration */
> >>> +			if (dev_conf_info.txmode.offloads !=
> >>> +				port->dev_conf.txmode.offloads) {
> >>> +				port->dev_conf.txmode.offloads =
> >>> +					dev_conf_info.txmode.offloads;
> >>> +				for (k = 0;
> >>> +				     k < port->dev_info.max_tx_queues;
> >>> +				     k++)
> >>> +					port->tx_conf[k].offloads =
> >>> +
> >> 	dev_conf_info.txmode.offloads;
> >>> +			}
> >>>  		}
> >>
> >> Above implementation gets the configuration from device and applies
> >> it to the testpmd configuration.
> >>
> >> Instead, what about a long level target to get rid of testpmd
> >> specific copy of the configuration and rely and the config provided
> >> by devices. @Xiaoyun, what do you think, does this make sense?
> >
> > You mean remove port->dev_conf and rx/tx_conf completely in the future? Or
> keep it in initial stage?
> >
> > Now, port->dev_conf will take global tx/rx_mode, fdir_conf and change some
> based on dev_info capabilities. And then use dev_configure to apply them for
> device.
> > After this, actually, dev->data->dev_conf contains all device configuration.
> >
> > So It seems it's OK to remove port->dev_conf completely. Just testpmd needs
> to be refactored a lot and regression test in case of issues.
> > But from long term view, it's good to keep one source and avoid copy.
> >
> 
> Yes, this is the intention I have for long term. I expect that testpmd still will keep
> some configuration in application level but we can prevent some duplication.
> 
> And the main point is, by cleaning up testpmd we can recognize blockers and fix
> them in libraries to help user applications.
> 
> > As for rx/tx_conf, it takes device default tx/rx_conf in dev_info and some
> settings in testpmd parameters also offloads from dev_conf.
> > So keep port->rx/tx_conf? But then it still needs copy from dev_conf since this
> may change.
> >
> 
> I am not very clear what is suggested above, can you please elaborate?
> 
> And 'struct rte_port' seems has following structs that can be get from library:
> struct rte_eth_dev_info dev_info;
> struct rte_eth_conf     dev_conf;
> struct rte_eth_rxconf   rx_conf[]
> struct rte_eth_txconf   tx_conf[]
> 
> I don't think we can remove them, but perhaps reduce the usage of them, please
> see below.
> 
> >>
> >> So instead of above code, update where RSS hash offload information
> >> printed to use device retrieved config instead of testpmd config, will it work?
> >
> > It's OK for device offload configurations.
> > But queue offloads are a bit tricky since dev->data->dev_conf doesn't include
> queue conf.
> > And it's not fair to use device offload configurations for queue offloads since
> user can use cmdline to config queue offload and that info can only be saved in
> port->rx/tx_conf and configure the device in setup_queue.
> >
> 
> It is common in testpmd that, a command changes the application copy of the
> configs, and mark as device configuration is required (for port or for queue).
> So in later stage this changed configuration is applied to device.
> 
> This async approach has its benefits and I don't think we should change it.
> (Also has some disadvantages that we hit in the past, like detecting some
> configuration can't be applied in later stage when we try to apply the config, not
> when command is issued at first place.).
> 
> What we can do it, reduce the testpmd config usage for the case to gather user
> requests and apply them to device.
> But to display device configuration, or to decide based on device configuration
> we can user config values get by device by APIs.
> 
> What do you think, can above distinction makes sense, or does it work?
> 
> 
> And there is still a chance that application copy of config diverge from device
> config, and since we provide full config in our APIs (not changes), there is a
> chance to overwrite a device configuration.
> To prevent this it is possible to read device config and overwrite testpmd config
> with that, similar to what this patch does, but I am not sure where this sync can
> be done. What do you think about doing this just after device configured?

I'm not sure I fully understand.
So for showing cmd, just use API rte_eth_tx/rx_queue_info_get to get dev queue config and new added API rte_eth_dev_conf_info_get to get dev config.

And for the cases where port->dev_config is used as a right value, replace them with use getting API.
For example: "if (res->value == port->dev_conf.rxmode.max_rx_pkt_len)" will be changed like "if (res->value == rte_eth_dev_conf_info_get().rxmode.max_rx_pkt_len)"

But other things keep the same as what this patch does?

This makes sense to me if I understand it right.
> 
> Thanks,
> ferruh


  reply	other threads:[~2021-09-18  2:18 UTC|newest]

Thread overview: 80+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-07-09 15:57 [dpdk-dev] [PATCH] " Jie Wang
2021-07-09  9:27 ` Andrew Rybchenko
2021-07-12  3:12   ` Li, Xiaoyun
     [not found]     ` <DM8PR11MB5639B19DACFB1B4F4E70ACA4D1149@DM8PR11MB5639.namprd11.prod.outlook.com>
2021-07-13  3:30       ` Li, Xiaoyun
2021-07-16  9:09     ` [dpdk-dev] [dpdk-stable] " Ferruh Yigit
2021-07-13 17:04 ` [dpdk-dev] [PATCH v2] " Jie Wang
2021-07-15  2:29   ` Li, Xiaoyun
2021-07-15  2:40     ` Li, Xiaoyun
2021-07-15 11:33   ` [dpdk-dev] [PATCH v3] " Jie Wang
2021-07-15 11:57     ` [dpdk-dev] [PATCH v4] " Jie Wang
2021-07-15  4:53       ` Li, Xiaoyun
2021-07-16  8:30         ` Li, Xiaoyun
2021-07-16  8:52           ` [dpdk-dev] [dpdk-stable] " Ferruh Yigit
     [not found]             ` <DM8PR11MB5639C757A790F65CBFB647C2D1E19@DM8PR11MB5639.namprd11.prod.outlook.com>
2021-07-19 16:18               ` Ferruh Yigit
2021-07-22 11:03                 ` Andrew Rybchenko
2021-08-09  8:53                   ` Ferruh Yigit
2021-08-17 17:38       ` [dpdk-dev] [PATCH v5] " Jie Wang
2021-08-24 17:10         ` [dpdk-dev] [PATCH v6 0/2] testpmd shows incorrect rx_offload configuration Jie Wang
2021-08-24 17:10           ` [dpdk-dev] [PATCH v6 1/2] ethdev: add an API to get device configuration info Jie Wang
2021-08-24 17:10           ` [dpdk-dev] [PATCH v6 2/2] app/testpmd: fix testpmd doesn't show RSS hash offload Jie Wang
2021-08-26  7:09           ` [dpdk-dev] [PATCH v7 0/2] testpmd shows incorrect rx_offload configuration Jie Wang
2021-08-26  7:09             ` [dpdk-dev] [PATCH v7 1/2] ethdev: add an API to get device configuration info Jie Wang
2021-08-26  7:09             ` [dpdk-dev] [PATCH v7 2/2] app/testpmd: fix testpmd doesn't show RSS hash offload Jie Wang
2021-08-27  7:36               ` Li, Xiaoyun
2021-08-27  8:17             ` [dpdk-dev] [PATCH v8 0/2] testpmd shows incorrect rx_offload configuration Jie Wang
2021-08-27  8:17               ` [dpdk-dev] [PATCH v8 1/2] ethdev: add an API to get device configuration info Jie Wang
2021-09-08 16:46                 ` Ferruh Yigit
2021-08-27  8:17               ` [dpdk-dev] [PATCH v8 2/2] app/testpmd: fix testpmd doesn't show RSS hash offload Jie Wang
2021-08-30  5:57                 ` Li, Xiaoyun
2021-09-08 16:50                 ` Ferruh Yigit
2021-09-09  3:31                   ` Li, Xiaoyun
2021-09-17 10:20                     ` Ferruh Yigit
2021-09-18  2:18                       ` Li, Xiaoyun [this message]
2021-09-20  9:45                         ` Ferruh Yigit
2021-09-20  9:48                 ` Ferruh Yigit
2021-09-22  2:52                   ` Wang, Jie1X
2021-09-26  9:20               ` [dpdk-dev] [PATCH v9 0/3] testpmd shows incorrect rx_offload configuration Jie Wang
2021-09-26  9:20                 ` [dpdk-dev] [PATCH v9 1/3] ethdev: add an API to get device configuration info Jie Wang
2021-09-27  6:19                   ` Thomas Monjalon
2021-09-27  7:21                     ` Wang, Jie1X
2021-09-27  7:56                       ` Thomas Monjalon
2021-10-04 11:20                         ` Ferruh Yigit
2021-10-04 11:25                           ` Thomas Monjalon
2021-10-04 11:22                   ` Ferruh Yigit
2021-10-04 11:26                     ` Thomas Monjalon
2021-10-04 11:35                       ` Ferruh Yigit
2021-09-26  9:20                 ` [dpdk-dev] [PATCH v9 2/3] doc: update release notes for new API Jie Wang
2021-10-04 11:22                   ` Ferruh Yigit
2021-10-04 11:26                     ` Thomas Monjalon
2021-09-26  9:20                 ` [dpdk-dev] [PATCH v9 3/3] app/testpmd: fix testpmd doesn't show RSS hash offload Jie Wang
2021-10-08  3:41                 ` [dpdk-dev] [PATCH v10 0/2] testpmd shows incorrect rx_offload configuration Jie Wang
2021-10-08  3:41                   ` [dpdk-dev] [PATCH v10 1/2] ethdev: add an API to get device configuration info Jie Wang
2021-10-08 12:10                     ` Ferruh Yigit
2021-10-08  3:41                   ` [dpdk-dev] [PATCH v10 2/2] app/testpmd: fix testpmd doesn't show RSS hash offload Jie Wang
2021-10-08 12:12                     ` Ferruh Yigit
2021-10-11 18:01                   ` [dpdk-dev] [PATCH v11 0/2] testpmd shows incorrect rx_offload configuration Jie Wang
2021-10-11 18:01                     ` [dpdk-dev] [PATCH v11 1/2] ethdev: add an API to get device configuration Jie Wang
2021-10-11 10:08                       ` Somnath Kotur
2021-10-11 12:21                       ` Andrew Rybchenko
2021-10-11 18:01                     ` [dpdk-dev] [PATCH v11 2/2] app/testpmd: fix testpmd doesn't show RSS hash offload Jie Wang
2021-10-12  2:54                     ` [dpdk-dev] [PATCH v12 0/2] testpmd shows incorrect rx_offload configuration Jie Wang
2021-10-12  2:54                       ` [dpdk-dev] [PATCH v12 1/2] ethdev: add an API to get device configuration Jie Wang
2021-10-12  5:50                         ` Andrew Rybchenko
2021-10-12  2:54                       ` [dpdk-dev] [PATCH v12 2/2] app/testpmd: fix testpmd doesn't show RSS hash offload Jie Wang
2021-10-12 14:37                         ` Ferruh Yigit
2021-10-13  2:42                           ` Wang, Jie1X
2021-10-13  8:50                             ` Ferruh Yigit
2021-10-13 10:14                               ` Ferruh Yigit
2021-10-12 14:35                       ` [dpdk-dev] [PATCH v12 0/2] testpmd shows incorrect rx_offload configuration Ferruh Yigit
2021-10-14 10:31                       ` [dpdk-dev] [PATCH v13 " Jie Wang
2021-10-14 10:31                         ` [dpdk-dev] [PATCH v13 1/2] ethdev: add an API to get device configuration Jie Wang
2021-10-14 10:31                         ` [dpdk-dev] [PATCH v13 2/2] app/testpmd: fix testpmd doesn't show RSS hash offload Jie Wang
2021-10-14 12:56                         ` [dpdk-dev] [PATCH v13 0/2] testpmd shows incorrect rx_offload configuration Ferruh Yigit
2021-10-15 10:43                           ` Ferruh Yigit
2021-10-15 11:29                           ` Ferruh Yigit
2021-08-24 18:19         ` [dpdk-dev] [PATCH v6 " Jie Wang
2021-08-24 18:19           ` [dpdk-dev] [PATCH v6 1/2] ethdev: add an API to get device configuration info Jie Wang
2021-08-25 20:07             ` Ferruh Yigit
2021-08-26  6:00               ` Ajit Khaparde
2021-08-24 18:19           ` [dpdk-dev] [PATCH v6 2/2] app/testpmd: fix testpmd doesn't show RSS hash offload Jie Wang

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=DM4PR11MB55343DD0A005F6BC6264F7F399DE9@DM4PR11MB5534.namprd11.prod.outlook.com \
    --to=xiaoyun.li@intel.com \
    --cc=andrew.rybchenko@oktetlabs.ru \
    --cc=dev@dpdk.org \
    --cc=ferruh.yigit@intel.com \
    --cc=jerinj@marvell.com \
    --cc=jie1x.wang@intel.com \
    --cc=konstantin.ananyev@intel.com \
    --cc=thomas@monjalon.net \
    /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).