mlx5 VF packet lost between rx_port_unicast_packets and rx_good_packets
Francesco Montorsi
francesco.montorsi at infovista.com
Thu Nov 11 00:53:43 CET 2021
Hi all,
I hit the exact same problem reported by Yan.
I’m using:
* 2 Mellanox CX5 MT28800 installed on 2 different servers, connected together
* Device FW (as reported by DPDK): 16.31.1014
* DPDK 19.11.5 (from 6WindGate actually)
I sent roughly 360M packets from one server to the other using “testpmd” (in –forward-mode=txonly).
My DPDK application on the other server is reporting the following xstats counters (:
CounterName PORT0 PORT1 TOTAL
rx_good_packets: 76727920, 0, 76727920
tx_good_packets: 0, 0, 0
rx_good_bytes: 4910586880, 0, 4910586880
tx_good_bytes: 0, 0, 0
rx_missed_errors: 0, 0, 0
rx_errors: 0, 0, 0
tx_errors: 0, 0, 0
rx_mbuf_allocation_errors: 0, 0, 0
rx_q0packets: 0, 0, 0
rx_q0bytes: 0, 0, 0
rx_q0errors: 0, 0, 0
rx_q1packets: 0, 0, 0
rx_q1bytes: 0, 0, 0
rx_q1errors: 0, 0, 0
rx_q2packets: 0, 0, 0
rx_q2bytes: 0, 0, 0
rx_q2errors: 0, 0, 0
rx_q3packets: 0, 0, 0
rx_q3bytes: 0, 0, 0
rx_q3errors: 0, 0, 0
rx_q4packets: 0, 0, 0
rx_q4bytes: 0, 0, 0
rx_q4errors: 0, 0, 0
rx_q5packets: 76727920, 0, 76727920
rx_q5bytes: 4910586880, 0, 4910586880
rx_q5errors: 0, 0, 0
rx_q6packets: 0, 0, 0
rx_q6bytes: 0, 0, 0
rx_q6errors: 0, 0, 0
rx_q7packets: 0, 0, 0
rx_q7bytes: 0, 0, 0
rx_q7errors: 0, 0, 0
rx_q8packets: 0, 0, 0
rx_q8bytes: 0, 0, 0
rx_q8errors: 0, 0, 0
rx_q9packets: 0, 0, 0
rx_q9bytes: 0, 0, 0
rx_q9errors: 0, 0, 0
rx_q10packets: 0, 0, 0
rx_q10bytes: 0, 0, 0
rx_q10errors: 0, 0, 0
rx_q11packets: 0, 0, 0
rx_q11bytes: 0, 0, 0
rx_q11errors: 0, 0, 0
tx_q0packets: 0, 0, 0
tx_q0bytes: 0, 0, 0
rx_wqe_err: 0, 0, 0
rx_port_unicast_packets: 360316064, 0, 360316064
rx_port_unicast_bytes: 23060228096, 0, 23060228096
tx_port_unicast_packets: 0, 0, 0
tx_port_unicast_bytes: 0, 0, 0
rx_port_multicast_packets: 0, 0, 0
rx_port_multicast_bytes: 0, 0, 0
tx_port_multicast_packets: 0, 0, 0
tx_port_multicast_bytes: 0, 0, 0
rx_port_broadcast_packets: 0, 0, 0
rx_port_broadcast_bytes: 0, 0, 0
tx_port_broadcast_packets: 0, 0, 0
tx_port_broadcast_bytes: 0, 0, 0
tx_packets_phy: 0, 0, 0
rx_packets_phy: 0, 0, 0
rx_crc_errors_phy: 0, 0, 0
tx_bytes_phy: 0, 0, 0
rx_bytes_phy: 0, 0, 0
rx_in_range_len_errors_phy 0, 0, 0
rx_symbol_err_phy: 0, 0, 0
rx_discards_phy: 0, 0, 0
tx_discards_phy: 0, 0, 0
tx_errors_phy: 0, 0, 0
rx_out_of_buffer: 0, 0, 0
So rx_good_packets is roughly 76M pkts, while rx_port_unicast_packets has counted correctly all 360M pkts sent by testpmd.
Of course my application layer has been able to dequeue from the DPDK port only 76M pkts so the remaining (rx_port_unicast_packets- rx_good_packets) got lost but are not reported in the “imissed” counter or “ierrors” counter of rte_eth_stats…
It would be not so easy for me to test against latest DPDK… also from what Yan has reported, the issue is still there in DPDK stable 20.11.3…
@Mellanox maintainers: any update on this issue? Is there a workaround to get the dropped packet back into the rte_eth_stats counters?
Thanks
Francesco Montorsi
From: Yan, Xiaoping (NSB - CN/Hangzhou) <xiaoping.yan at nokia-sbell.com>
Sent: Friday, October 29, 2021 2:44 AM
To: Gerry Wan <gerryw at stanford.edu>; Asaf Penso <asafp at nvidia.com>; Slava Ovsiienko <viacheslavo at nvidia.com>; Matan Azrad <matan at nvidia.com>; Raslan Darawsheh <rasland at nvidia.com>
Cc: Martin Weiser <martin.weiser at allegro-packets.com>; David Marchand <david.marchand at redhat.com>; users at dpdk.org
Subject: RE: mlx5 VF packet lost between rx_port_unicast_packets and rx_good_packets
Hi,
Yes, it’s always zero…
It seems dpdk-stable-20.11-3 already include this patch.
[xiaopiya at fedora30 dpdk-stable-20.11.3]$ patch -p1 < ./4-4-net-mlx5-fix-imissed-statistics.diff
patching file drivers/net/mlx5/linux/mlx5_os.c
Reversed (or previously applied) patch detected! Assume -R? [n] n
Apply anyway? [n] ^C
[xiaopiya at fedora30 dpdk-stable-20.11.3]$ grep -r "mlx5_queue_counter_id_prepare" ./
./drivers/net/mlx5/linux/mlx5_os.c:mlx5_queue_counter_id_prepare(struct rte_eth_dev *dev)
./drivers/net/mlx5/linux/mlx5_os.c: mlx5_queue_counter_id_prepare(eth_dev);
./drivers/net/mlx5/linux/mlx5_os.c.rej:+mlx5_queue_counter_id_prepare(struct rte_eth_dev *dev)
./drivers/net/mlx5/linux/mlx5_os.c.rej:+ mlx5_queue_counter_id_prepare(eth_dev);
./4-4-net-mlx5-fix-imissed-statistics.diff:+mlx5_queue_counter_id_prepare(struct rte_eth_dev *dev)
./4-4-net-mlx5-fix-imissed-statistics.diff:+ mlx5_queue_counter_id_prepare(eth_dev);
Thank you.
Best regards
Yan Xiaoping
From: Gerry Wan <gerryw at stanford.edu<mailto:gerryw at stanford.edu>>
Sent: 2021年10月28日 9:58
To: Yan, Xiaoping (NSB - CN/Hangzhou) <xiaoping.yan at nokia-sbell.com<mailto:xiaoping.yan at nokia-sbell.com>>
Cc: Martin Weiser <martin.weiser at allegro-packets.com<mailto:martin.weiser at allegro-packets.com>>; David Marchand <david.marchand at redhat.com<mailto:david.marchand at redhat.com>>; Asaf Penso <asafp at nvidia.com<mailto:asafp at nvidia.com>>; Slava Ovsiienko <viacheslavo at nvidia.com<mailto:viacheslavo at nvidia.com>>; Matan Azrad <matan at nvidia.com<mailto:matan at nvidia.com>>; Raslan Darawsheh <rasland at nvidia.com<mailto:rasland at nvidia.com>>; users at dpdk.org<mailto:users at dpdk.org>
Subject: Re: mlx5 VF packet lost between rx_port_unicast_packets and rx_good_packets
Are the rx_missed_errors/rx_out_of_buffer counters always showing 0 no matter how fast you push your generator?
I had a similar issue with missing counters on DPDK 20.11 that was fixed on 21.05 by applying this patch:
http://patchwork.dpdk.org/project/dpdk/patch/1614249901-307665-5-git-send-email-matan@nvidia.com/
Potentially relevant thread:
https://inbox.dpdk.org/users/CAAcwi38rs2Vk9MKhRGS3kAK+=dYAnDdECT7f+Ts-f13cANYB+Q@mail.gmail.com/
On Wed, Oct 27, 2021 at 6:39 PM Yan, Xiaoping (NSB - CN/Hangzhou) <xiaoping.yan at nokia-sbell.com<mailto:xiaoping.yan at nokia-sbell.com>> wrote:
Hi,
I checked the counter from PF with ethtool -S, there is no counter named 'rx_prio0_buf_discard'
Anyway, I checked all counters from ethtool output, there is not any counter reflects the dropped packets.
Any suggestion from mlx maintainer? @Matan Azrad @Asaf Penso @Slava Ovsiienko @Raslan Darawsheh
Thank you.
Best regards
Yan Xiaoping
-----Original Message-----
From: Martin Weiser <martin.weiser at allegro-packets.com<mailto:martin.weiser at allegro-packets.com>>
Sent: 2021年10月27日 15:54
To: Yan, Xiaoping (NSB - CN/Hangzhou) <xiaoping.yan at nokia-sbell.com<mailto:xiaoping.yan at nokia-sbell.com>>; David Marchand <david.marchand at redhat.com<mailto:david.marchand at redhat.com>>
Cc: Asaf Penso <asafp at nvidia.com<mailto:asafp at nvidia.com>>; users at dpdk.org<mailto:users at dpdk.org>; Slava Ovsiienko <viacheslavo at nvidia.com<mailto:viacheslavo at nvidia.com>>; Matan Azrad <matan at nvidia.com<mailto:matan at nvidia.com>>; Raslan Darawsheh <rasland at nvidia.com<mailto:rasland at nvidia.com>>
Subject: Re: mlx5 VF packet lost between rx_port_unicast_packets and rx_good_packets
Hi,
you may want to check the counter 'rx_prio0_buf_discard' with ethtool (which is not available in DPDK xstats as it seems that this counter is global for the card and not available per port).
I opened a ticket a while ago regarding this issue:
https://bugs.dpdk.org/show_bug.cgi?id=749
Best regards,
Martin
Am 27.10.21 um 08:17 schrieb Yan, Xiaoping (NSB - CN/Hangzhou):
> Hi,
>
> I tried with dpdk 20.11-3 downloaded from
> https://fast.dpdk.org/rel/dpdk-20.11.3.tar.xz
> Problem still exist:
> 1. there is packet loss with 2mpps (small packet), 2. no counter for
> the dropped packet in NIC.
>
> traffic generator stats: sends 41990938, receives back 41986105, lost
> 4833 testpmd fwd stats: RX-packets: 41986110, TX-packets: 41986110
> port xstats: rx_unicast_packets: 41990938 (all packets reached to the NIC port), rx_good_packets: 41986111 (some is lost), but there is not any counter of the lost packet.
>
> Here is the log:
> [root at up-0 /]# dpdk-testpmd -l "2,3" --legacy-mem --socket-mem
> "5000,0" -a 0000:03:06.7 -- -i --nb-cores=1 --portmask=0x1 --rxd=512
> --txd=512
> EAL: Detected 28 lcore(s)
> EAL: Detected 2 NUMA nodes
> EAL: Detected static linkage of DPDK
> EAL: Multi-process socket /var/run/dpdk/rte/mp_socket
> EAL: Selected IOVA mode 'VA'
> EAL: No available hugepages reported in hugepages-2048kB
> EAL: Probing VFIO support...
> EAL: Probe PCI driver: mlx5_pci (15b3:1018) device: 0000:03:06.7
> (socket 0)
> mlx5_pci: cannot bind mlx5 socket: Read-only file system
> mlx5_pci: Cannot initialize socket: Read-only file system
> EAL: No legacy callbacks, legacy socket not created Interactive-mode
> selected
> testpmd: create a new mbuf pool <mb_pool_0>: n=155456, size=2176,
> socket=0
> testpmd: preferred mempool ops selected: ring_mp_mc
>
> Warning! port-topology=paired and odd forward ports number, the last port will pair with itself.
>
> Configuring Port 0 (socket 0)
> Port 0: 7A:9A:8A:A6:86:93
> Checking link statuses...
> Done
> testpmd> port stop 0
> Stopping ports...
> Checking link statuses...
> Done
> testpmd> vlan set filter on 0
> testpmd> rx_vlan add 767 0
> testpmd> port start 0
> Port 0: 7A:9A:8A:A6:86:93
> Checking link statuses...
> Done
> testpmd> set fwd 5tswap
> Set 5tswap packet forwarding mode
> testpmd> start
> 5tswap packet forwarding - ports=1 - cores=1 - streams=1 - NUMA
> support enabled, MP allocation mode: native Logical Core 3 (socket 0) forwards packets on 1 streams:
> RX P=0/Q=0 (socket 0) -> TX P=0/Q=0 (socket 0)
> peer=02:00:00:00:00:00
>
> 5tswap packet forwarding packets/burst=32
> nb forwarding cores=1 - nb forwarding ports=1
> port 0: RX queue number: 1 Tx queue number: 1
> Rx offloads=0x200 Tx offloads=0x0
> RX queue: 0
> RX desc=512 - RX free threshold=64
> RX threshold registers: pthresh=0 hthresh=0 wthresh=0
> RX Offloads=0x200
> TX queue: 0
> TX desc=512 - TX free threshold=0
> TX threshold registers: pthresh=0 hthresh=0 wthresh=0
> TX offloads=0x0 - TX RS bit threshold=0
>
> testpmd> show fwd stats all
>
> ---------------------- Forward statistics for port 0 ----------------------
> RX-packets: 41986110 RX-dropped: 0 RX-total: 41986110
> TX-packets: 41986110 TX-dropped: 0 TX-total: 41986110
>
> ----------------------------------------------------------------------
> ------
>
> testpmd> show port xstats 0
> ###### NIC extended statistics for port 0
> rx_good_packets: 41986111
> tx_good_packets: 41986111
> rx_good_bytes: 3106973594
> tx_good_bytes: 3106973594
> rx_missed_errors: 0
> rx_errors: 0
> tx_errors: 0
> rx_mbuf_allocation_errors: 0
> rx_q0_packets: 41986111
> rx_q0_bytes: 3106973594
> rx_q0_errors: 0
> tx_q0_packets: 41986111
> tx_q0_bytes: 3106973594
> rx_wqe_errors: 0
> rx_unicast_packets: 41990938
> rx_unicast_bytes: 3107329412
> tx_unicast_packets: 41986111
> tx_unicast_bytes: 3106973594
> rx_multicast_packets: 1
> rx_multicast_bytes: 114
> tx_multicast_packets: 0
> tx_multicast_bytes: 0
> rx_broadcast_packets: 5
> rx_broadcast_bytes: 1710
> tx_broadcast_packets: 0
> tx_broadcast_bytes: 0
> tx_phy_packets: 0
> rx_phy_packets: 0
> rx_phy_crc_errors: 0
> tx_phy_bytes: 0
> rx_phy_bytes: 0
> rx_phy_in_range_len_errors: 0
> rx_phy_symbol_errors: 0
> rx_phy_discard_packets: 0
> tx_phy_discard_packets: 0
> tx_phy_errors: 0
> rx_out_of_buffer: 0
> tx_pp_missed_interrupt_errors: 0
> tx_pp_rearm_queue_errors: 0
> tx_pp_clock_queue_errors: 0
> tx_pp_timestamp_past_errors: 0
> tx_pp_timestamp_future_errors: 0
> tx_pp_jitter: 0
> tx_pp_wander: 0
> tx_pp_sync_lost: 0
> testpmd> q
> Command not found
> testpmd> exit
> Command not found
> testpmd> quit
> Telling cores to stop...
> Waiting for lcores to finish...
>
> ---------------------- Forward statistics for port 0 ----------------------
> RX-packets: 41986112 RX-dropped: 0 RX-total: 41986112
> TX-packets: 41986112 TX-dropped: 0 TX-total: 41986112
>
> ----------------------------------------------------------------------
> ------
>
> Best regards
> Yan Xiaoping
>
> -----Original Message-----
> From: David Marchand <david.marchand at redhat.com<mailto:david.marchand at redhat.com>>
> Sent: 2021年10月18日 18:45
> To: Yan, Xiaoping (NSB - CN/Hangzhou) <xiaoping.yan at nokia-sbell.com<mailto:xiaoping.yan at nokia-sbell.com>>
> Cc: Asaf Penso <asafp at nvidia.com<mailto:asafp at nvidia.com>>; users at dpdk.org<mailto:users at dpdk.org>; Slava Ovsiienko
> <viacheslavo at nvidia.com<mailto:viacheslavo at nvidia.com>>; Matan Azrad <matan at nvidia.com<mailto:matan at nvidia.com>>; Raslan
> Darawsheh <rasland at nvidia.com<mailto:rasland at nvidia.com>>
> Subject: Re: mlx5 VF packet lost between rx_port_unicast_packets and
> rx_good_packets
>
> On Mon, Oct 18, 2021 at 11:28 AM Yan, Xiaoping (NSB - CN/Hangzhou) <xiaoping.yan at nokia-sbell.com<mailto:xiaoping.yan at nokia-sbell.com>> wrote:
>> I have cloned dpdk code from github
>>
>> [xiaopiya at fedora30 dpdk]$ git remote -v origin
>> https://github.com/DPDK/dpdk.git (fetch) origin
>> https://github.com/DPDK/dpdk.git (push)
>>
>> which tag should I use?
>>
>> Or do I have to download 20.11.3 from git.dpdk.org<http://git.dpdk.org>?
>>
>> Sorry, I don’t know the relation between https://github.com/DPDK and git.dpdk.org<http://git.dpdk.org>?
> Github DPDK/dpdk repo is a replication of the main repo hosted on dpdk.org<http://dpdk.org> servers.
>
> The official git repos and releases tarballs are on dpdk.org<http://dpdk.org> servers.
> The list of official releases tarballs is at:
> http://core.dpdk.org/download/ The main repo git is at:
> https://git.dpdk.org/dpdk/ The LTS/stable releases repo git is at:
> https://git.dpdk.org/dpdk-stable/
>
>
> --
> David Marchand
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mails.dpdk.org/archives/users/attachments/20211110/c3d3b48a/attachment-0001.htm>
More information about the users
mailing list