[PATCH22.11 v1] net/mlx5: fix age checking crash
Luca Boccassi
luca.boccassi at gmail.com
Tue Oct 28 15:48:34 CET 2025
On Tue, 28 Oct 2025 at 13:10, Thomas Monjalon <thomas at monjalon.net> wrote:
>
> 28/10/2025 13:13, Luca Boccassi:
> > On Tue, 28 Oct 2025 at 08:17, Rongwei Liu <rongweil at nvidia.com> wrote:
> > >
> > > When aging is configured, there is a background thread
> > > which queries all the counters in the pool.
> > >
> > > Meantime, per queue flow insertion/deletion/update changes
> > > the counter pool too. It introduces a race condition between
> > > resetting counters's in_used and age_idx fields during flow deletion
> > > and reading them in the background thread.
> > >
> > > To resolve it, all key members of counter's struct
> > > are placed in a single uint32_t and they are accessed atomically.
> > >
> > > To avoid the occasional timestamp equalization with age_idx,
> > > query_gen_when_free is moved out of the union. The total memory
> > > size is kept the same.
> > >
> > > Fixes: 04a4de756e14 ("net/mlx5: support flow age action with HWS")
> > > Cc: michaelba at nvidia.com
> > > Cc: stable at dpdk.org
> > > Signed-off-by: Rongwei Liu <rongweil at nvidia.com>
> >
> > Hi,
> >
> > I cannot seem to find a reference to this patch on the main branch?
> > Was it merged with a different commit?
>
> It seems Rongwei didn't reuse the title updated while merging in main:
>
> https://git.dpdk.org/dpdk/commit/?id=820ca7361bb
>
> Note the new title is more precise, please use it:
> net/mlx5: fix flow aging race condition
Sounds good, thanks
More information about the stable
mailing list