DPDK patches and discussions
 help / color / mirror / Atom feed
From: Ferruh Yigit <ferruh.yigit@amd.com>
To: Joshua Washington <joshwash@google.com>
Cc: Rushil Gupta <rushilg@google.com>,
	Stephen Hemminger <stephen@networkplumber.org>,
	Junfeng Guo <junfeng.guo@intel.com>,
	qi.z.zhang@intel.com, jingjing.wu@intel.com,
	beilei.xing@intel.com, dev@dpdk.org, jeroendb@google.com,
	jrkim@google.com, Xiaoyun Li <xiaoyun.li@intel.com>
Subject: Re: [PATCH] net/gve: add support for basic stats
Date: Tue, 31 Jan 2023 10:05:03 +0000	[thread overview]
Message-ID: <a440ab60-9624-f21e-396a-239bdf2aa1a1@amd.com> (raw)
In-Reply-To: <CALuQH+WaAR_dZ0ECdKLH8u4grv2Z3mfzogvDVnakGu5NQY80sA@mail.gmail.com>

On 1/31/2023 1:51 AM, Joshua Washington wrote:
> Hello,
> 
> I tested it out, and the updates to testpmd seem to work.
> 

Hi Joshua,

Thanks for testing, I will send a patch soon.

But this was testpmd issue, do you have any objection with the net/gve
patch, if not can you please record this with ack/review/tested-by tags,
so I can proceed with it.

> Before applying the second patch:
>   ---------------------- Forward statistics for port 0
>  ----------------------
>   RX-packets: 0              RX-dropped: 0             RX-total: 0
>   TX-packets: 43769238       TX-dropped: 62634         TX-total: 43831872
>  
> ----------------------------------------------------------------------------
> 
>   ---------------------- Forward statistics for port 1
>  ----------------------
>   RX-packets: 0              RX-dropped: 0             RX-total: 0
>   TX-packets: 43761119       TX-dropped: 70753         TX-total: 43831872
>  
> ----------------------------------------------------------------------------
> 
>   +++++++++++++++ Accumulated forward statistics for all
> ports+++++++++++++++
>   RX-packets: 0              RX-dropped: 0             RX-total: 0
>   TX-packets: 87530357       TX-dropped: 157302        TX-total: 87687659
>  
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> 
> 62634 + 70753 = 133387 != 157302
> 
> After applying the second patch:
>   ---------------------- Forward statistics for port 0
>  ----------------------
>   RX-packets: 0              RX-dropped: 0             RX-total: 0
>   TX-packets: 12590721       TX-dropped: 36638         TX-total: 12627359
>  
> ----------------------------------------------------------------------------
> 
>   ---------------------- Forward statistics for port 1
>  ----------------------
>   RX-packets: 0              RX-dropped: 0             RX-total: 0
>   TX-packets: 12596255       TX-dropped: 31746         TX-total: 12628001
>  
> ----------------------------------------------------------------------------
> 
>   +++++++++++++++ Accumulated forward statistics for all
> ports+++++++++++++++
>   RX-packets: 0              RX-dropped: 0             RX-total: 0
>   TX-packets: 25186976       TX-dropped: 68384         TX-total: 25255360
>  
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> 
> Thanks,
> Josh
> 
> On Wed, Jan 18, 2023 at 8:22 AM Ferruh Yigit <ferruh.yigit@amd.com
> <mailto:ferruh.yigit@amd.com>> wrote:
> 
>     On 12/19/2022 7:38 PM, Joshua Washington wrote:
>     > Hello,
>     >
>     > As it turns out, this error actually propagates to the "total"
>     stats as
>     > well, which I assume is just calculated by adding TX-packets and
>     > TX-dropped. Here are the full stats from the example that Rushil
>     mentioned:
>     >
>     >   ---------------------- Forward statistics for port 0
>     >  ----------------------
>     >   RX-packets: 2453802        RX-dropped: 0             RX-total:
>     2453802
>     >   TX-packets: 34266881       TX-dropped: 447034        TX-total:
>     34713915
>     >  
>     >
>     ----------------------------------------------------------------------------
>     >
>     >   ---------------------- Forward statistics for port 1
>     >  ----------------------
>     >   RX-packets: 34713915       RX-dropped: 0             RX-total:
>     34713915
>     >   TX-packets: 2453802        TX-dropped: 0             TX-total:
>     2453802
>     >  
>     >
>     ----------------------------------------------------------------------------
>     >
>     >   +++++++++++++++ Accumulated forward statistics for all
>     > ports+++++++++++++++
>     >   RX-packets: 37167717       RX-dropped: 0             RX-total:
>     37167717
>     >   TX-packets: 36720683       TX-dropped: 807630        TX-total:
>     37528313
>     >  
>     >
>     ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>     >
>     > It can be seen that the stats for the individual ports are consistent,
>     > but the TX-total and TX-dropped are not consistent with the stats for
>     > the individual ports, as I believe that the TX-total and RX-total
>     > accumulated stats should be equal.
>     >
> 
>     Hi Joshua, Rushil,
> 
>     As I checked for it, issue may be related to testpmd stats display,
> 
>     While displaying per port TX-dropped value, it only takes
>     'ports_stats[pt_id].tx_dropped' into account,
>     but for accumulated TX-dropped results it takes both
>     'ports_stats[pt_id].tx_dropped' & 'stats.oerrors' into account.
> 
>     If you can reproduce it easily, can you please test with following
>     change:
> 
>      diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
>      index 134d79a55547..49322d07d7d6 100644
>      --- a/app/test-pmd/testpmd.c
>      +++ b/app/test-pmd/testpmd.c
>      @@ -2056,6 +2056,8 @@ fwd_stats_display(void)
>                              fwd_cycles += fs->core_cycles;
>              }
>              for (i = 0; i < cur_fwd_config.nb_fwd_ports; i++) {
>      +               uint64_t tx_dropped = 0;
>      +
>                      pt_id = fwd_ports_ids[i];
>                      port = &ports[pt_id];
> 
>      @@ -2077,8 +2079,9 @@ fwd_stats_display(void)
>                      total_recv += stats.ipackets;
>                      total_xmit += stats.opackets;
>                      total_rx_dropped += stats.imissed;
>      -               total_tx_dropped += ports_stats[pt_id].tx_dropped;
>      -               total_tx_dropped += stats.oerrors;
>      +               tx_dropped += ports_stats[pt_id].tx_dropped;
>      +               tx_dropped += stats.oerrors;
>      +               total_tx_dropped += tx_dropped;
>                      total_rx_nombuf  += stats.rx_nombuf;
> 
>                      printf("\n  %s Forward statistics for port %-2d %s\n",
>      @@ -2105,8 +2108,8 @@ fwd_stats_display(void)
> 
>                      printf("  TX-packets: %-14"PRIu64" TX-dropped:
>     %-14"PRIu64
>                             "TX-total: %-"PRIu64"\n",
>      -                      stats.opackets, ports_stats[pt_id].tx_dropped,
>      -                      stats.opackets + ports_stats[pt_id].tx_dropped);
>      +                      stats.opackets, tx_dropped,
>      +                      stats.opackets + tx_dropped);
> 
>                      if (record_burst_stats) {
>                              if (ports_stats[pt_id].rx_stream)
> 
> 
>     >
>     > On Mon, Dec 19, 2022 at 11:17 AM Rushil Gupta <rushilg@google.com
>     <mailto:rushilg@google.com>
>     > <mailto:rushilg@google.com <mailto:rushilg@google.com>>> wrote:
>     >
>     >     Hi all
>     >     Josh just found out some inconsistencies in the Tx/Rx
>     statistics sum
>     >     for all ports. Not sure if we can screenshot here but it goes like
>     >     this:
>     >     Tx-dropped for port0: 447034
>     >     Tx-dropped for port1: 0
>     >     Accumulated forward statistics for all ports: 807630
>     >
>     >     Please note that this issue is only with Tx-dropped (not
>     >     Tx-packets/Tx-total).
>     >
>     >
>     >     On Wed, Dec 7, 2022 at 8:39 AM Stephen Hemminger
>     >     <stephen@networkplumber.org
>     <mailto:stephen@networkplumber.org>
>     <mailto:stephen@networkplumber.org
>     <mailto:stephen@networkplumber.org>>> wrote:
>     >     >
>     >     > On Wed, 7 Dec 2022 15:09:08 +0000
>     >     > Ferruh Yigit <ferruh.yigit@amd.com
>     <mailto:ferruh.yigit@amd.com> <mailto:ferruh.yigit@amd.com
>     <mailto:ferruh.yigit@amd.com>>>
>     >     wrote:
>     >     >
>     >     > > On 11/24/2022 7:33 AM, Junfeng Guo wrote:
>     >     > > > Add support for dev_ops of stats_get and stats_reset.
>     >     > > >
>     >     > > > Queue stats update will be moved into xstat [1], but the
>     basic
>     >     stats
>     >     > > > items may still be required. So we just keep the remaining
>     >     ones and
>     >     > > > will implement the queue stats via xstats in the coming
>     release.
>     >     > > >
>     >     > > > [1]
>     >     > > > https://elixir.bootlin.com/dpdk/v22.07/
>     <https://elixir.bootlin.com/dpdk/v22.07/>
>     >     <https://elixir.bootlin.com/dpdk/v22.07/
>     <https://elixir.bootlin.com/dpdk/v22.07/>> \
>     >     > > >     source/doc/guides/rel_notes/deprecation.rst#L118
>     >     > > >
>     >     > > > Signed-off-by: Xiaoyun Li <xiaoyun.li@intel.com
>     <mailto:xiaoyun.li@intel.com>
>     >     <mailto:xiaoyun.li@intel.com <mailto:xiaoyun.li@intel.com>>>
>     >     > > > Signed-off-by: Junfeng Guo <junfeng.guo@intel.com
>     <mailto:junfeng.guo@intel.com>
>     >     <mailto:junfeng.guo@intel.com <mailto:junfeng.guo@intel.com>>>
>     >     > >
>     >     > > <...>
>     >     > >
>     >     > > > +static int
>     >     > > > +gve_dev_stats_get(struct rte_eth_dev *dev, struct
>     >     rte_eth_stats *stats)
>     >     > > > +{
>     >     > > > +   uint16_t i;
>     >     > > > +
>     >     > > > +   for (i = 0; i < dev->data->nb_tx_queues; i++) {
>     >     > > > +           struct gve_tx_queue *txq =
>     dev->data->tx_queues[i];
>     >     > > > +           if (txq == NULL)
>     >     > > > +                   continue;
>     >     > > > +
>     >     > > > +           stats->opackets += txq->packets;
>     >     > > > +           stats->obytes += txq->bytes;
>     >     > > > +           stats->oerrors += txq->errors;
>     >     > >
>     >     > > Hi Junfeng, Qi, Jingjing, Beilei,
>     >     > >
>     >     > > Above logic looks wrong to me, did you test it?
>     >     > >
>     >     > > If the 'gve_dev_stats_get()' called multiple times (without
>     >     stats reset
>     >     > > in between), same values will be keep added to stats.
>     >     > > Some hw based implementations does this, because reading
>     from stats
>     >     > > registers automatically reset those registers but this
>     shouldn't
>     >     be case
>     >     > > for this driver.
>     >     > >
>     >     > > I expect it to be something like:
>     >     > >
>     >     > > local_stats = 0
>     >     > > foreach queue
>     >     > >       local_stats += queue->stats
>     >     > > stats = local_stats
>     >     >
>     >     > The zero of local_stats is unnecessary.
>     >     > The only caller of the PMD stats_get is rte_ethdev_stats_get
>     >     > and it zeros the stats structure before calling the PMD.
>     >     >
>     >     >
>     >     > int
>     >     > rte_eth_stats_get(uint16_t port_id, struct rte_eth_stats *stats)
>     >     > {
>     >     >         struct rte_eth_dev *dev;
>     >     >
>     >     >         RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -ENODEV);
>     >     >         dev = &rte_eth_devices[port_id];
>     >     >
>     >     >         memset(stats, 0, sizeof(*stats));
>     >     > ...
>     >     >         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
>     >     >         return eth_err(port_id, (*dev->dev_ops->stats_get)(dev,
>     >     stats));
>     >     >
>     >     > If any PMD has extra memset in their stats get that could be
>     removed.
>     >
>     >
>     >
>     > --
>     >
>     > Joshua Washington | Software Engineer | joshwash@google.com
>     <mailto:joshwash@google.com>
>     > <mailto:joshwash@google.com <mailto:joshwash@google.com>> | (414)
>     366-4423 <tel:(414)%20366-4423>
>     >  
> 
> 
> 
> -- 
> 
> Joshua Washington | Software Engineer | joshwash@google.com
> <mailto:joshwash@google.com> | (414) 366-4423
>  


  reply	other threads:[~2023-01-31 10:05 UTC|newest]

Thread overview: 17+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-11-24  7:33 Junfeng Guo
2022-11-24 16:59 ` Stephen Hemminger
2022-11-24 17:26   ` Ferruh Yigit
2022-11-26  3:16     ` Rushil Gupta
2022-11-26 17:34       ` Stephen Hemminger
2022-11-28 11:12         ` Ferruh Yigit
2022-11-28 17:24           ` Stephen Hemminger
2022-11-29  1:42             ` Guo, Junfeng
2022-12-07 15:09 ` Ferruh Yigit
2022-12-07 16:39   ` Stephen Hemminger
2022-12-19 19:17     ` Rushil Gupta
2022-12-19 19:38       ` Joshua Washington
2023-01-18 16:22         ` Ferruh Yigit
2023-01-31  1:51           ` Joshua Washington
2023-01-31 10:05             ` Ferruh Yigit [this message]
2023-02-02 23:00               ` Joshua Washington
2023-02-03 17:39 ` 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=a440ab60-9624-f21e-396a-239bdf2aa1a1@amd.com \
    --to=ferruh.yigit@amd.com \
    --cc=beilei.xing@intel.com \
    --cc=dev@dpdk.org \
    --cc=jeroendb@google.com \
    --cc=jingjing.wu@intel.com \
    --cc=joshwash@google.com \
    --cc=jrkim@google.com \
    --cc=junfeng.guo@intel.com \
    --cc=qi.z.zhang@intel.com \
    --cc=rushilg@google.com \
    --cc=stephen@networkplumber.org \
    --cc=xiaoyun.li@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).