[dpdk-dev] [PATCH 3/4] net/mlx5: rework PMD global data init
shahafs at mellanox.com
Tue Mar 19 07:54:13 CET 2019
Monday, March 18, 2019 11:21 PM, Yongseok Koh:
> Subject: Re: [PATCH 3/4] net/mlx5: rework PMD global data init
> On Thu, Mar 14, 2019 at 05:36:28AM -0700, Shahaf Shuler wrote:
> > Hi Koh,
> > Thursday, March 7, 2019 9:33 AM, Yongseok Koh:
> > > Subject: [PATCH 3/4] net/mlx5: rework PMD global data init
> > >
> > > There's more need to have PMD global data structure. It should be
> > > initialized once per a process regardless of how many PMD instances are
> > > mlx5_init_once() is called during probing and make sure all the init
> > > functions are called once per a process. The existing shared memory
> > > gets more extensively used for this purpose. As there could be
> > > multiple secondary processes, a static storage (local to process) is also
> > It is hard to understand from the commit log what was missing on the old
> Okay, will add more comments.
> > > As the reserved virtual address for UAR remap is a PMD global
> > > resource, this doesn't need to be stored in the device priv
> > > structure, but in the PMD global data.
> > I thought we agreed to drop those and have different VA for each process.
> > If so, is the extra work on the UAR here is needed?
> My plan was to do that in a separate patch for performance regression.
> Let me know if you want it to be done in this patchset.
I prefer to see how the UAR will look in the final design. If you can include such patch it wil be great.
Otherwise lets keep it as is.
> > > Signed-off-by: Yongseok Koh <yskoh at mellanox.com>
> > > ---
> > > drivers/net/mlx5/mlx5.c | 250
> > > ----------
> > > drivers/net/mlx5/mlx5.h | 19 +++-
> > > drivers/net/mlx5/mlx5_mp.c | 19 +++-
> > > drivers/net/mlx5/mlx5_txq.c | 7 +-
> > > 4 files changed, 217 insertions(+), 78 deletions(-)
> > >
> > > diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c index
> > > 6ed2418106..ea8fd55ee6 100644
> > > --- a/drivers/net/mlx5/mlx5.c
> > > +++ b/drivers/net/mlx5/mlx5.c
> > > @@ -128,16 +128,26 @@ struct mlx5_shared_data *mlx5_shared_data;
> > > /* Spinlock for mlx5_shared_data allocation. */ static
> > > rte_spinlock_t mlx5_shared_data_lock = RTE_SPINLOCK_INITIALIZER;
> > >
> > > +/* Process local data for secondary processes. */ static struct
> > > +mlx5_local_data mlx5_local_data;
> > Why not storing this context as part of ethdev-> process_private instead of
> declaring it static?
> Because it is not per-device data but per-PMD data.
> Will also have to rebase my patchsets when I send out v2.
More information about the dev