[PATCH v1] net/mlx5: fix probe optimization race condition

[rongwei liu]

On 2025/8/28 12:40, Stephen Hemminger wrote:
> On Thu, 28 Aug 2025 06:21:34 +0300
> Rongwei Liu <rongweil at nvidia.com> wrote:
> 
>> With dedicated RDMA link monitor, there are two threads
>> which can update the IB device port information.
>>
>> Add a new flag to avoid the race condition. Update should
>> go through RDMA link monitor once ready.
>>
>> Fixes: 51fb5c40c826 ("net/mlx5: optimize device probing")
>> Cc: rongweil at nvidia.com
>> Cc: stable at dpdk.org
>> Signed-off-by: Rongwei Liu <rongweil at nvidia.com>
>> Acked-by: Viacheslav Ovsiienko <viacheslavo at nvidia.com>
>> ---
> 
> If variable is modified (with out locking) on two threads it
> needs to atomic or volatile.
Exactly. Before this patch, it' user responsebility to seperate probe probing and sf manipualtion.
Obviously, customer didn't follow this very well.
Now logic change to:
1. Update all port information in probing thread.
2. Probe thread initiate the dedicated rdma monitor thread. Once ready, all port update will go to this thread.
3. Next port probing won't trigger PMD port information update.

No lock is required then.