net/bnxt: wrong link status when lsc_intr is used

[Edwin Brossette]

Hello,

I am trying to operate a Broadcom BCM57414 2x10G nic using dpdk bnxt pmd. I
use DPDK 22.11.
However, doing so I stumbled over a number of different issues. Mainly,
using the dpdk rte_eth library, I don't seem to be able to correctly poll
the link status: I expect my nic has a problem using autoneg to set the
link speed/duplex, because these parameters remain unknown while autoneg is
on. However, after trying to set link up, instead of showing the link state
as down, I see the link being up, which is in truth not the case, as no
packets can transit on the line and the switch at the other end sees it
down.

When searching around and trying to debug the code, I found the function
bnxt_dev_info_get_op() sets my nic in interrupt mode:

> eth_dev->data->dev_conf.intr_conf.lsc = 1;

Which is a bit of a surprising thing to do for a function meant to get
info.
Thus my card ends up working in a mode I didn't configure it to, which may
be the cause of my issue: later when setting the link up in function
bnxt_dev_set_link_up_op():

>  if (!bp->link_info->link_up)
>     rc = bnxt_set_hwrm_link_config(bp, true);
>  if (!rc)
>     eth_dev->data->dev_link.link_status = 1;

So link_status in eth_dev gets set to 1 as long as the operation did not
return any error code. This is the case when setting my card's link up
(rc=0), although the link clearly can't get up, for whatever other bug is
present. Now this shouldn't be much of an issue given we will update the
link status at some point, mainly in rte_eth_link_get_nowait():

>  if (dev->data->dev_conf.intr_conf.lsc && dev->data->dev_started)
>     rte_eth_linkstatus_get(dev, eth_link);
>  else {
>     if (*dev->dev_ops->link_update == NULL)
>         return -ENOTSUP;
>     (*dev->dev_ops->link_update)(dev, 0);
>      *eth_link = dev->data->dev_link;

Here we can see in the else statement that the link status gets updated.
However because the pmd auto-configured the nic in interrupt mode when
calling the get_info function, we are not going through that else
statement. So when reading the value of the link_status, we read 1 instead
of 0. I suppose with interrupt mode enabled, the nic should be able to
update this variable on its own, but it is clearly not the case in my
setup: link status is never updated and incorrectly indicates the link is
UP.

I can suggest a testpmd reproduction setup using the --no-lsc-interrupt
option. With this option, dev_conf.intr_conf.lsc should be 0. In addition,
I added a log to the rte_eth library in rte_eth_dev_start() to display
dev_conf.intr_conf.lsc state when starting the port:

> + RTE_ETHDEV_LOG(ERR, "><><><><!!<devstart>:
dev->data->dev_conf.intr_conf=%d\n", dev->data->dev_conf.intr_conf.lsc);
>   diag = (*dev->dev_ops->dev_start)(dev);

Running testpmd, we can see the following outpout when starting port:

dpdk-testpmd --log-level=pmd.net.bnxt.driver:8 -a 0000:02:00.0 -a
0000:02:00.1 -- -i --rxq=2 --txq=2 --coremask=0x0c --total-num-mbufs=250000
--no-lsc-interrupt
[...]
Configuring Port 0 (socket 0)
bnxt_rx_queue_setup_op(): App supplied RXQ drop_en status : 1
bnxt_rx_queue_setup_op(): RX Buf MTU 1500
bnxt_rx_queue_setup_op(): RX Buf size is 9728
bnxt_rx_queue_setup_op(): App supplied RXQ drop_en status : 1
bnxt_rx_queue_setup_op(): RX Buf MTU 1500
bnxt_rx_queue_setup_op(): RX Buf size is 9728
><><><><!!<devstart>: dev->data->dev_conf.intr_conf=1
bnxt_mq_rx_configure(): pools = 1 nb_q_per_grp = 2
bnxt_mq_rx_configure(): rxq[0] = 0x105fb7ac0 vnic[0] = 0x100227080
bnxt_mq_rx_configure(): rxq[1] = 0x105fb0e40 vnic[0] = 0x100227080
bnxt_setup_one_vnic(): vnic[0] = 0x100227080 vnic->fw_grp_ids = 0x105fa7e00
bnxt_hwrm_vnic_alloc(): Alloc VNIC. Start 0, End 2
bnxt_hwrm_vnic_alloc(): VNIC ID 2
bnxt_setup_one_vnic(): rxq[0]->vnic=0x100227080 vnic->fw_grp_ids=0x105fa7e00
bnxt_setup_one_vnic(): rxq[1]->vnic=0x100227080 vnic->fw_grp_ids=0x105fa7e00
bnxt_setup_one_vnic(): vnic->rx_queue_cnt = 2
bnxt_hwrm_port_phy_qcfg(): Link Speed:0,Auto:4:64:140,Support:140,Force:0
bnxt_hwrm_port_phy_qcfg(): Link Signal:0,PAM::Auto:0,Support:0,Force:0
bnxt_hwrm_port_phy_qcfg(): Link Speed:0,Auto:4:64:140,Support:140,Force:0
bnxt_hwrm_port_phy_qcfg(): Link Signal:0,PAM::Auto:0,Support:0,Force:0
bnxt_ulp_port_init(): Skip ulp init for port: 0, TF is not enabled
bnxt_receive_function(): Using SSE vector mode receive for port 0
bnxt_transmit_function(): Using SSE vector mode transmit for port 0
Port 0: 00:0A:F7:B6:E3:D0
Configuring Port 1 (socket 0)
bnxt_rx_queue_setup_op(): App supplied RXQ drop_en status : 1
bnxt_rx_queue_setup_op(): RX Buf MTU 1500
bnxt_rx_queue_setup_op(): RX Buf size is 9728
bnxt_rx_queue_setup_op(): App supplied RXQ drop_en status : 1
bnxt_rx_queue_setup_op(): RX Buf MTU 1500
bnxt_rx_queue_setup_op(): RX Buf size is 9728
><><><><!!<devstart>: dev->data->dev_conf.intr_conf=1
bnxt_mq_rx_configure(): pools = 1 nb_q_per_grp = 2
bnxt_mq_rx_configure(): rxq[0] = 0x106200280 vnic[0] = 0x100200080
bnxt_mq_rx_configure(): rxq[1] = 0x105f10e40 vnic[0] = 0x100200080
bnxt_setup_one_vnic(): vnic[0] = 0x100200080 vnic->fw_grp_ids = 0x105f07e00
bnxt_hwrm_vnic_alloc(): Alloc VNIC. Start 0, End 2
bnxt_hwrm_vnic_alloc(): VNIC ID 3
bnxt_setup_one_vnic(): rxq[0]->vnic=0x100200080 vnic->fw_grp_ids=0x105f07e00
bnxt_setup_one_vnic(): rxq[1]->vnic=0x100200080 vnic->fw_grp_ids=0x105f07e00
bnxt_setup_one_vnic(): vnic->rx_queue_cnt = 2
bnxt_hwrm_port_phy_qcfg(): Link Speed:0,Auto:4:64:140,Support:140,Force:0
bnxt_hwrm_port_phy_qcfg(): Link Signal:0,PAM::Auto:0,Support:0,Force:0
bnxt_hwrm_port_phy_qcfg(): Link Speed:0,Auto:4:64:140,Support:140,Force:0
bnxt_hwrm_port_phy_qcfg(): Link Signal:0,PAM::Auto:0,Support:0,Force:0
bnxt_ulp_port_init(): Skip ulp init for port: 1, TF is not enabled
bnxt_receive_function(): Using SSE vector mode receive for port 1
bnxt_transmit_function(): Using SSE vector mode transmit for port 1
Port 1: 00:0A:F7:B6:E3:D1

Here, we can see that lsc interrupts are enabled even though we specified
not to enable them. Then given autoneg does not work on my nic, I can try
setting the link up and showing port info:

testpmd> set link-up port 0
bnxt_print_link_info(): Port 0 Link Up - speed 0 Mbps - half-duplex

testpmd> show port info 0

********************* Infos for port 0  *********************
MAC address: 00:0A:F7:B6:E3:D0
Device name: 0000:02:00.0
Driver name: net_bnxt
Firmware-version: 223.0.161.0
Devargs:
Connect to socket: 0
memory allocation on the socket: 0
Link status: up
Link speed: None
Link duplex: half-duplex
Autoneg status: Off
MTU: 1500
Promiscuous mode: enabled
Allmulticast mode: disabled
Maximum number of MAC addresses: 128
Maximum number of MAC addresses of hash filtering: 0
VLAN offload:
  strip off, filter off, extend off, qinq strip off
Hash key size in bytes: 40
Redirection table size: 128
Supported RSS offload flow types:
  ipv4  ipv4-tcp  ipv4-udp  ipv6  ipv6-tcp  ipv6-udp
  user-defined-50  user-defined-51
Minimum size of RX buffer: 1
Maximum configurable length of RX packet: 9600
Maximum configurable size of LRO aggregated packet: 0
Maximum number of VMDq pools: 64
Current number of RX queues: 2
Max possible RX queues: 117
Max possible number of RXDs per queue: 8192
Min possible number of RXDs per queue: 16
RXDs number alignment: 1
Current number of TX queues: 2
Max possible TX queues: 117
Max possible number of TXDs per queue: 4096
Min possible number of TXDs per queue: 16
TXDs number alignment: 1
Max segment number per packet: 65535
Max segment number per MTU/TSO: 65535
Device capabilities: 0x3( RUNTIME_RX_QUEUE_SETUP RUNTIME_TX_QUEUE_SETUP )
Switch name: 0000:02:00.0
Switch domain Id: 0
Switch Port Id: 32768
Device error handling mode: proactive

This shows link status is seen as up, even although link speed is None.

I was wondering if patching the code to move this line which sets lsc
interrupt on somewhere else might be reasonable, or if this could cause
further trouble. Maybe having a parameter to trigger it ON/OFF might be a
good addition.
May I have your opinion on this matter?

Sincerely,
Edwin Brossette
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mails.dpdk.org/archives/dev/attachments/20230119/6a4a8b61/attachment.htm>