Re: [PATCH rdma-rc 1/3] RDMA/mlx5: Use xa_lock_irqsave when access to SRQ table

[Leon Romanovsky <leon@kernel.org>]

On Tue, Jul 07, 2020 at 08:43:03AM -0300, Jason Gunthorpe wrote:
> On Tue, Jul 07, 2020 at 02:06:10PM +0300, Leon Romanovsky wrote:
> > Fixes: b02a29eb8841 ("mlx5: Convert mlx5_srq_table to XArray")
>
> This didn't introduce the bug, when things were converted to xarray it
> already had the wrong spinlock type.
>
> I'm surprised this is only been found now since it has been wrong for
> years. Did something else change?

I checked internal bug tracker to see when this bug was discovered. It
looks like ennoblement of more debug kernel config options by default
helped to find it.

>
> > Signed-off-by: Maor Gottlieb <maorg@mellanox.com>
> > Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
> >  drivers/infiniband/hw/mlx5/srq_cmd.c | 10 ++++++----
> >  1 file changed, 6 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/infiniband/hw/mlx5/srq_cmd.c b/drivers/infiniband/hw/mlx5/srq_cmd.c
> > index 6f5eadc4d183..be0e5469dad0 100644
> > +++ b/drivers/infiniband/hw/mlx5/srq_cmd.c
> > @@ -82,12 +82,13 @@ struct mlx5_core_srq *mlx5_cmd_get_srq(struct mlx5_ib_dev *dev, u32 srqn)
> >  {
> >  	struct mlx5_srq_table *table = &dev->srq_table;
> >  	struct mlx5_core_srq *srq;
> > +	unsigned long flags;
> >
> > -	xa_lock(&table->array);
> > +	xa_lock_irqsave(&table->array, flags);
> >  	srq = xa_load(&table->array, srqn);
> >  	if (srq)
> >  		refcount_inc(&srq->common.refcount);
> > -	xa_unlock(&table->array);
> > +	xa_unlock_irqrestore(&table->array, flags);
>
> This and other places can just be xa_lock_irq as we are not in an atomic
> context here.

ok, let's change for the clarity.

Thanks