patches for DPDK stable branches
 help / color / mirror / Atom feed
From: Stephen Hemminger <stephen@networkplumber.org>
To: Ferruh Yigit <ferruh.yigit@xilinx.com>
Cc: Long Li <longli@microsoft.com>,
	"longli@linuxonhyperv.com" <longli@linuxonhyperv.com>,
	"dev@dpdk.org" <dev@dpdk.org>,
	Stephen Hemminger <sthemmin@microsoft.com>,
	"stable@dpdk.org" <stable@dpdk.org>
Subject: Re: [Patch v2] net/netvsc: report correct stats values
Date: Thu, 5 May 2022 09:40:26 -0700	[thread overview]
Message-ID: <20220505094026.22e74f43@hermes.local> (raw)
In-Reply-To: <7809f41b-c21b-5ebe-d830-91015edb0cb8@xilinx.com>

On Thu, 5 May 2022 17:28:38 +0100
Ferruh Yigit <ferruh.yigit@xilinx.com> wrote:

> On 5/4/2022 7:38 PM, Long Li wrote:
> >> Subject: Re: [Patch v2] net/netvsc: report correct stats values
> >>
> >> On 5/3/2022 9:48 PM, Long Li wrote:  
> >>>> Subject: Re: [Patch v2] net/netvsc: report correct stats values
> >>>>
> >>>> On 5/3/2022 8:14 PM, Long Li wrote:  
> >>>>>> Subject: Re: [Patch v2] net/netvsc: report correct stats values
> >>>>>>
> >>>>>> On 5/3/2022 7:18 PM, Long Li wrote:  
> >>>>>>>> Subject: Re: [Patch v2] net/netvsc: report correct stats values
> >>>>>>>>
> >>>>>>>> On Tue, 26 Apr 2022 22:56:14 +0100 Ferruh Yigit
> >>>>>>>> <ferruh.yigit@xilinx.com> wrote:
> >>>>>>>>  
> >>>>>>>>>>       		if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
> >>>>>>>>>> -			stats->q_opackets[i] = txq->stats.packets;
> >>>>>>>>>> -			stats->q_obytes[i] = txq->stats.bytes;
> >>>>>>>>>> +			stats->q_opackets[i] += txq->stats.packets;
> >>>>>>>>>> +			stats->q_obytes[i] += txq->stats.bytes;  
> >>>>>>>>>
> >>>>>>>>> This is per queue stats, 'stats->q_opackets[i]', in next
> >>>>>>>>> iteration of the loop, 'i' will be increased and 'txq' will be
> >>>>>>>>> updated, so as far as I can see the above change has no affect.  
> >>>>>>>>
> >>>>>>>> Agree, that is why it was just assignment originally.  
> >>>>>>>
> >>>>>>> The condition here is a little different. NETVSC is a master
> >>>>>>> device with  
> >>>>>> another PMD running as a slave. When reporting stats values, it
> >>>>>> needs to add the values from the slave PMD. The original code just
> >>>>>> overwrites the values from its slave PMD.
> >>>>>>
> >>>>>> Where the initial values are coming from, 'hn_vf_stats_get()'?
> >>>>>>
> >>>>>> If 'hn_vf_stats_get()' fills the stats, what are the values kept in
> >>>>>> 'txq-  
> >>>>> stats.*'  
> >>>>>> in above updated loop?  
> >>>>>
> >>>>> Yes, hn_vf_stats_get() fills in the stats from the slave PMD.
> >>>>> txq->stats  
> >>>> values are from the master PMD. Those values are different and
> >>>> accounted separated from the values from the slave PMD.
> >>>>
> >>>> I see, since this is a little different than what most of the PMDs
> >>>> do, can you please put a little more info to the commit log? Or
> >>>> perhaps can add some comments to the code.  
> >>>
> >>> Ok, will do.
> >>>  
> >>>>
> >>>> And still 'stats->rx_nombuf' change is not required right? If so can
> >>>> you remove it in the next version?  
> >>>
> >>> It is still needed. NETVSC unconditionally calls the slave PMD to receive  
> >> packets, even if it can't allocate a mbuf to receive a synthetic packet itself. The
> >> accounting of rx_nombuf is valid because the synthetic packets (to NETVSC) and
> >> VF packets (to slave PMD) are routed separately from Hyper-V.
> >>
> >> I am not referring to the "+=" update, my comment was because 'stats-  
> >>> rx_nombuf' is overwritten in 'rte_eth_stats_get()' [1].  
> >> Is it still required?  
> > 
> > Yes, it is still needed. NETVSC calls the rte_eth_stats_get() on its slave PMD first, and stats->rx_nombuf is updated (overwritten) for its slave PMD. Afte that, it needs to add to its own dev->data->rx_mbuf_alloc_failed back to stats->rx_nombuf.
> >   
> 
> But its own stat also will be overwritten (not in PMD function, but in 
> ethdev layer).
> 'stats->rx_nombuf' assignment in the PMD seems has no effect and can be 
> removed.
> 
> I can't see how it is needed, can you please put a call stack to describe?

This here:


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];

	if (stats == NULL) {
		RTE_ETHDEV_LOG(ERR, "Cannot get ethdev port %u stats to NULL\n",
			port_id);
		return -EINVAL;
	}

	memset(stats, 0, sizeof(*stats));

	RTE_FUNC_PTR_OR_ERR_RET(*dev->dev_ops->stats_get, -ENOTSUP);
	stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
	return eth_err(port_id, (*dev->dev_ops->stats_get)(dev, stats));
}

Will fill in rx_nombuf from the current rx_mbuf_alloc_failed.
But it happens before the PMD specific stats function.

For the case of nested device like netvsc PMD that value needs to
be summed.

What happens with VF would be:

Call of rte_eth_stats_get on the netvsc PMD port.
    stats->rx_nombuf comes from rx_mbuf_alloc_failed of the vmbus part
Netvsc PMD stats get calls rte_eth_stats_get on the VF
Nested rte_eth_stats_get will set stats->rx_nombuf to the value from VF
   the previous value is overwritten.

At end of Netvsc PMD stats get it should be adding the value of any
allocation failures in the VMBUS part.

Summary:
rte_eth_stats_get:      stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;  // original
rte_eth_stats_get(vf):  stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;  // VF alloc failed
hn_dev_stats_get:       stats->rx_nombuf += dev->data->rx_mbuf_alloc_failed; // Vmbus alloc failed



  reply	other threads:[~2022-05-05 16:40 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-03-24 17:45 longli
2022-04-26 21:56 ` Ferruh Yigit
2022-04-26 22:45   ` Stephen Hemminger
2022-05-03 18:18     ` Long Li
2022-05-03 19:03       ` Ferruh Yigit
2022-05-03 19:14         ` Long Li
2022-05-03 19:55           ` Ferruh Yigit
2022-05-03 20:48             ` Long Li
2022-05-04 12:33               ` Ferruh Yigit
2022-05-04 18:38                 ` Long Li
2022-05-05 16:28                   ` Ferruh Yigit
2022-05-05 16:40                     ` Stephen Hemminger [this message]
2022-05-05 16:57                       ` Ferruh Yigit
2022-05-10  5:33                         ` Long Li
2022-05-10 11:29                           ` Ferruh Yigit
2022-05-10 18:03                             ` Long Li

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=20220505094026.22e74f43@hermes.local \
    --to=stephen@networkplumber.org \
    --cc=dev@dpdk.org \
    --cc=ferruh.yigit@xilinx.com \
    --cc=longli@linuxonhyperv.com \
    --cc=longli@microsoft.com \
    --cc=stable@dpdk.org \
    --cc=sthemmin@microsoft.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).