netdev - Re: [PATCH mlx5-next 11/12] {net,IB}/mlx5: Move Page fault EQ and ODP logic to RDMA

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20181117201347.GA14593@mellanox.com>
Date:   Sat, 17 Nov 2018 20:13:53 +0000
From:   Jason Gunthorpe <jgg@...lanox.com>
To:     Saeed Mahameed <saeedm@...lanox.com>
CC:     Leon Romanovsky <leonro@...lanox.com>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        "linux-rdma@...r.kernel.org" <linux-rdma@...r.kernel.org>
Subject: Re: [PATCH mlx5-next 11/12] {net,IB}/mlx5: Move Page fault EQ and ODP
 logic to RDMA

On Fri, Nov 16, 2018 at 01:59:00PM -0800, Saeed Mahameed wrote:
> Use the new generic EQ API to move all ODP RDMA data structures and logic
> form mlx5 core driver into mlx5_ib driver.
> 
> Signed-off-by: Saeed Mahameed <saeedm@...lanox.com>
> Reviewed-by: Leon Romanovsky <leonro@...lanox.com>
> Reviewed-by: Tariq Toukan <tariqt@...lanox.com>
>  drivers/infiniband/hw/mlx5/main.c             |  10 +-
>  drivers/infiniband/hw/mlx5/mlx5_ib.h          |  15 +-
>  drivers/infiniband/hw/mlx5/odp.c              | 281 +++++++++++++++++-
>  drivers/net/ethernet/mellanox/mlx5/core/dev.c |  34 ---
>  drivers/net/ethernet/mellanox/mlx5/core/eq.c  | 252 ----------------
>  .../net/ethernet/mellanox/mlx5/core/lib/eq.h  |   8 -
>  .../net/ethernet/mellanox/mlx5/core/main.c    |  17 +-
>  .../ethernet/mellanox/mlx5/core/mlx5_core.h   |   2 -
>  include/linux/mlx5/driver.h                   |  49 ---
>  include/linux/mlx5/eq.h                       |  21 ++
>  10 files changed, 308 insertions(+), 381 deletions(-)
> 
> diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
> index 6fbc0cba1bac..fcf4a0328a90 100644
> +++ b/drivers/infiniband/hw/mlx5/main.c
> @@ -6040,6 +6040,11 @@ static int mlx5_ib_stage_odp_init(struct mlx5_ib_dev *dev)
>  	return mlx5_ib_odp_init_one(dev);
>  }
>  
> +void mlx5_ib_stage_odp_cleanup(struct mlx5_ib_dev *dev)
> +{
> +	mlx5_ib_odp_cleanup_one(dev);
> +}
> +
>  int mlx5_ib_stage_counters_init(struct mlx5_ib_dev *dev)
>  {
>  	if (MLX5_CAP_GEN(dev->mdev, max_qp_cnt)) {
> @@ -6225,7 +6230,7 @@ static const struct mlx5_ib_profile pf_profile = {
>  		     mlx5_ib_stage_dev_res_cleanup),
>  	STAGE_CREATE(MLX5_IB_STAGE_ODP,
>  		     mlx5_ib_stage_odp_init,
> -		     NULL),
> +		     mlx5_ib_stage_odp_cleanup),
>  	STAGE_CREATE(MLX5_IB_STAGE_COUNTERS,
>  		     mlx5_ib_stage_counters_init,
>  		     mlx5_ib_stage_counters_cleanup),
> @@ -6395,9 +6400,6 @@ static struct mlx5_interface mlx5_ib_interface = {
>  	.add            = mlx5_ib_add,
>  	.remove         = mlx5_ib_remove,
>  	.event          = mlx5_ib_event,
> -#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
> -	.pfault		= mlx5_ib_pfault,
> -#endif
>  	.protocol	= MLX5_INTERFACE_PROTOCOL_IB,
>  };
>  
> diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
> index b651a7a6fde9..d01af2d829b8 100644
> +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
> @@ -880,6 +880,15 @@ struct mlx5_ib_lb_state {
>  	bool			enabled;
>  };
>  
> +struct mlx5_ib_pf_eq {
> +	struct mlx5_ib_dev      *dev;
> +	struct mlx5_eq          *core;
> +	struct work_struct       work;
> +	spinlock_t               lock; /* Pagefaults spinlock */
> +	struct workqueue_struct  *wq;
> +	mempool_t                *pool;
> +};

I know this is being copied, but can we please not do this vertical
alignment madness in RDMA? It is such a nightmare to maintain this..

> +/* mempool_refill() was proposed but unfortunately wasn't accepted
> + * http://lkml.iu.edu/hypermail/linux/kernel/1512.1/05073.html
> + * Chip workaround.

'cheap workaround'

Jason