Re: [PATCH mlx5-next 1/5] RDMA/mlx5: Replace struct mlx5_core_mkey by u32 key

[Jason Gunthorpe]

On Tue, Jun 22, 2021 at 03:08:19PM +0300, Leon Romanovsky wrote:

> diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c
> index 03dc6c22843f..ae0472d92801 100644
> +++ b/drivers/infiniband/hw/mlx5/mr.c
> @@ -89,24 +89,39 @@ static void set_mkc_access_pd_addr_fields(void *mkc, int acc, u64 start_addr,
>  	MLX5_SET64(mkc, mkc, start_addr, start_addr);
>  }
>  
> -static void
> -assign_mkey_variant(struct mlx5_ib_dev *dev, struct mlx5_core_mkey *mkey,
> -		    u32 *in)
> +static void assign_mkey_variant(struct mlx5_ib_dev *dev, u32 *mkey, u32 *in)
>  {
>  	u8 key = atomic_inc_return(&dev->mkey_var);
>  	void *mkc;
>  
>  	mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
>  	MLX5_SET(mkc, mkc, mkey_7_0, key);
> -	mkey->key = key;
> +	*mkey = key;
> +}

Can this be tidied please? We set both mkey_7_0 and mkey then pass
them into mlx5_core_create_mkey which then does

	*mkey = (u32)mlx5_mkey_variant(*mkey) | mlx5_idx_to_mkey(mkey_index);

But isn't mlx5_mkey_variant(*mkey) just MLX5_GET(mkc, in, mkey_7_0)
and we can get rid of this confusing sequence?

> +
> +static void set_mkey_fields(void *mkc, struct mlx5_core_mkey *mkey)
> +{
> +	mkey->iova = MLX5_GET64(mkc, mkc, start_addr);
> +	mkey->size = MLX5_GET64(mkc, mkc, len);
> +	mkey->pd = MLX5_GET(mkc, mkc, pd);
> +	init_waitqueue_head(&mkey->wait);
>  }

Why isn't this called through the create_mkey_callback() path? I think
evey mkey should always have a valid waitqueue

Jason