Re: [PATCH] net/bonding: fix bond startup failure when NUMA is -1

["humin (Q)" <humin29@huawei.com>]

在 2023/6/16 14:08, Chaoyong He 写道:
>> 在 2023/6/16 11:20, Chaoyong He 写道:
>>> From: Zerun Fu <zerun.fu@corigine.com>
>>>
>>> After the mainline Linux kernel commit
>>> "fe205d984e7730f4d21f6f8ebc60f0698404ac31" (ACPI: Remove side
>> effect
>>> of partly creating a node in acpi_map_pxm_to_online_node) by Jonathan
>>> Cameron. When the system does not support NUMA architecture, the
>>> "socket_id" is expected to be -1. The valid "socket_id" in BOND PMD is
>>> greater than or equal to zero. So it will cause an error when DPDK
>>> checks the validity of the "socket_id" when starting the bond. This
>>> commit can fix this bug.
>>>
>>> Fixes: f294e04851fd ("net/bonding: fix socket ID check")
>>> Cc: stable@dpdk.org
>>>
>>> Signed-off-by: Zerun Fu <zerun.fu@corigine.com>
>>> Reviewed-by: Peng Zhang <peng.zhang@corigine.com>
>>> Reviewed-by: Chaoyong He <chaoyong.he@corigine.com>
>>> Reviewed-by: Long Wu <long.wu@corigine.com>
>>> ---
>>>    drivers/net/bonding/rte_eth_bond_args.c | 6 ++++++
>>>    drivers/net/bonding/rte_eth_bond_pmd.c  | 2 +-
>>>    2 files changed, 7 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/net/bonding/rte_eth_bond_args.c
>>> b/drivers/net/bonding/rte_eth_bond_args.c
>>> index 6553166f5c..c137efd55f 100644
>>> --- a/drivers/net/bonding/rte_eth_bond_args.c
>>> +++ b/drivers/net/bonding/rte_eth_bond_args.c
>>> @@ -212,6 +212,12 @@ bond_ethdev_parse_socket_id_kvarg(const char
>> *key __rte_unused,
>>>        if (*endptr != 0 || errno != 0)
>>>                return -1;
>>>
>>> +     /* SOCKET_ID_ANY also consider a valid socket id */
>>> +     if ((int8_t)socket_id == SOCKET_ID_ANY) {
>>> +             *(int *)extra_args = SOCKET_ID_ANY;
>>> +             return 0;
>>> +     }
>>> +
>>>        /* validate socket id value */
>>>        if (socket_id >= 0 && socket_id < RTE_MAX_NUMA_NODES) {
>>>                *(int *)extra_args = (int)socket_id; diff --git
>>> a/drivers/net/bonding/rte_eth_bond_pmd.c
>>> b/drivers/net/bonding/rte_eth_bond_pmd.c
>>> index f0c4f7d26b..390a5b4271 100644
>>> --- a/drivers/net/bonding/rte_eth_bond_pmd.c
>>> +++ b/drivers/net/bonding/rte_eth_bond_pmd.c
>>> @@ -3604,7 +3604,7 @@ static int
>>>    bond_alloc(struct rte_vdev_device *dev, uint8_t mode)
>>>    {
>>>        const char *name = rte_vdev_device_name(dev);
>>> -     uint8_t socket_id = dev->device.numa_node;
>>> +     int socket_id = dev->device.numa_node;
>> Well,  other point should be also modified, like :
>>
>> ***
>>
>> "socket %u.",    name, bonding_mode, socket_id);
>>
>> ***
>>
>> %u -- > %d.
> Okay, I will send a v2 patch fix this.
>
>> BTW,  I think  there is  no need  to add args like "socket_id=-1..." if we know
>> this server does not support NUMA.
>>
>> Default socket id is -1, so this is meaningless.
>>
> We found this bug when running 'dperf' app, and it is the 'dperf' app add this 'socket_id=-1' args.
> Maybe the 'dperf' app should change its logic?
> Please help correct me if I misunderstood, thanks.

I agree with your patch.

What I mentioned is just for further discussion.


>
>>>        struct bond_dev_private *internals = NULL;
>>>        struct rte_eth_dev *eth_dev = NULL;
>>>        uint32_t vlan_filter_bmp_size;