[PATCH] net/mlx5: change calculating inline room for Tx

[PATCH] net/mlx5: change calculating inline room for Tx

[Yongseok Koh]

Current implementation is error-prone if the max inline size
(txq->max_inilne) is decoupled from txq->inline_en and becomes zero. If it
becomes zero, HW can crash due to WQ overflow.

Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
---
 drivers/net/mlx5/mlx5_rxtx.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 2e208bf96..6254228a9 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -499,6 +499,8 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	unsigned int j = 0;
 	unsigned int k = 0;
 	unsigned int max;
+	unsigned int max_inline = txq->max_inline;
+	const unsigned int inline_en = !!max_inline && txq->inline_en;
 	uint16_t max_wqe;
 	unsigned int comp;
 	volatile struct mlx5_wqe_v *wqe = NULL;
@@ -685,14 +687,14 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 			}
 		}
 		/* Inline if enough room. */
-		if (txq->inline_en || tso) {
+		if (inline_en || tso) {
 			uintptr_t end = (uintptr_t)
 				(((uintptr_t)txq->wqes) +
 				 (1 << txq->wqe_n) * MLX5_WQE_SIZE);
-			unsigned int max_inline = txq->max_inline *
-						  RTE_CACHE_LINE_SIZE -
-						  (pkt_inline_sz - 2);
-			uintptr_t addr_end = (addr + max_inline) &
+			unsigned int inline_room = max_inline *
+						   RTE_CACHE_LINE_SIZE -
+						   (pkt_inline_sz - 2);
+			uintptr_t addr_end = (addr + inline_room) &
 					     ~(RTE_CACHE_LINE_SIZE - 1);
 			unsigned int copy_b = (addr_end > addr) ?
 				RTE_MIN((addr_end - addr), length) :
-- 
2.11.0

Re: [PATCH] net/mlx5: change calculating inline room for Tx

[Nélio Laranjeiro]

Hi Yongseok,

On Thu, Apr 13, 2017 at 10:46:51AM -0700, Yongseok Koh wrote:
> Current implementation is error-prone if the max inline size
> (txq->max_inilne) is decoupled from txq->inline_en and becomes zero. If it
> becomes zero, HW can crash due to WQ overflow.

By reading this log, it seems more to be a fix, no?

> Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
> Acked-by: Shahaf Shuler <shahafs@mellanox.com>
> ---
>  drivers/net/mlx5/mlx5_rxtx.c | 12 +++++++-----
>  1 file changed, 7 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
> index 2e208bf96..6254228a9 100644
> --- a/drivers/net/mlx5/mlx5_rxtx.c
> +++ b/drivers/net/mlx5/mlx5_rxtx.c
> @@ -499,6 +499,8 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
>  	unsigned int j = 0;
>  	unsigned int k = 0;
>  	unsigned int max;
> +	unsigned int max_inline = txq->max_inline;

Cannot it be declared as const too?  Seems it is only read in the whole
Tx burst function, as it a chance to be update in the middle?

> +	const unsigned int inline_en = !!max_inline && txq->inline_en;
>  	uint16_t max_wqe;
>  	unsigned int comp;
>  	volatile struct mlx5_wqe_v *wqe = NULL;
> @@ -685,14 +687,14 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
>  			}
>  		}
>  		/* Inline if enough room. */
> -		if (txq->inline_en || tso) {
> +		if (inline_en || tso) {
>  			uintptr_t end = (uintptr_t)
>  				(((uintptr_t)txq->wqes) +
>  				 (1 << txq->wqe_n) * MLX5_WQE_SIZE);
> -			unsigned int max_inline = txq->max_inline *
> -						  RTE_CACHE_LINE_SIZE -
> -						  (pkt_inline_sz - 2);
> -			uintptr_t addr_end = (addr + max_inline) &
> +			unsigned int inline_room = max_inline *
> +						   RTE_CACHE_LINE_SIZE -
> +						   (pkt_inline_sz - 2);
> +			uintptr_t addr_end = (addr + inline_room) &
>  					     ~(RTE_CACHE_LINE_SIZE - 1);
>  			unsigned int copy_b = (addr_end > addr) ?
>  				RTE_MIN((addr_end - addr), length) :
> -- 
> 2.11.0

Thanks,

-- 
Nélio Laranjeiro
6WIND

Re: [PATCH] net/mlx5: change calculating inline room for Tx

[Yongseok Koh]

On Fri, Apr 14, 2017 at 08:51:39AM +0200, Nélio Laranjeiro wrote:
> Hi Yongseok,
> 
> On Thu, Apr 13, 2017 at 10:46:51AM -0700, Yongseok Koh wrote:
> > Current implementation is error-prone if the max inline size
> > (txq->max_inilne) is decoupled from txq->inline_en and becomes zero. If it
> > becomes zero, HW can crash due to WQ overflow.
> 
> By reading this log, it seems more to be a fix, no?
It doesn't fix any existing bug in the code but to protect from potential
disaster by adding new features in the future. Also, there's a customer who
wants to change txq->max_inilne in runtime.

> 
> > Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
> > Acked-by: Shahaf Shuler <shahafs@mellanox.com>
> > ---
> >  drivers/net/mlx5/mlx5_rxtx.c | 12 +++++++-----
> >  1 file changed, 7 insertions(+), 5 deletions(-)
> > 
> > diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
> > index 2e208bf96..6254228a9 100644
> > --- a/drivers/net/mlx5/mlx5_rxtx.c
> > +++ b/drivers/net/mlx5/mlx5_rxtx.c
> > @@ -499,6 +499,8 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
> >  	unsigned int j = 0;
> >  	unsigned int k = 0;
> >  	unsigned int max;
> > +	unsigned int max_inline = txq->max_inline;
> 
> Cannot it be declared as const too?  Seems it is only read in the whole
> Tx burst function, as it a chance to be update in the middle?
This is used just for calculation not for conditional expression. So, I didn't
think this makes difference.

Thanks
Yongseok

Re: [PATCH] net/mlx5: change calculating inline room for Tx

[Nélio Laranjeiro]

On Thu, Apr 13, 2017 at 10:46:51AM -0700, Yongseok Koh wrote:
> Current implementation is error-prone if the max inline size
> (txq->max_inilne) is decoupled from txq->inline_en and becomes zero. If it
> becomes zero, HW can crash due to WQ overflow.
> 
> Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
> Acked-by: Shahaf Shuler <shahafs@mellanox.com>
> ---
>  drivers/net/mlx5/mlx5_rxtx.c | 12 +++++++-----
>  1 file changed, 7 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
> index 2e208bf96..6254228a9 100644
> --- a/drivers/net/mlx5/mlx5_rxtx.c
> +++ b/drivers/net/mlx5/mlx5_rxtx.c
> @@ -499,6 +499,8 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
>  	unsigned int j = 0;
>  	unsigned int k = 0;
>  	unsigned int max;
> +	unsigned int max_inline = txq->max_inline;
> +	const unsigned int inline_en = !!max_inline && txq->inline_en;
>  	uint16_t max_wqe;
>  	unsigned int comp;
>  	volatile struct mlx5_wqe_v *wqe = NULL;
> @@ -685,14 +687,14 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
>  			}
>  		}
>  		/* Inline if enough room. */
> -		if (txq->inline_en || tso) {
> +		if (inline_en || tso) {
>  			uintptr_t end = (uintptr_t)
>  				(((uintptr_t)txq->wqes) +
>  				 (1 << txq->wqe_n) * MLX5_WQE_SIZE);
> -			unsigned int max_inline = txq->max_inline *
> -						  RTE_CACHE_LINE_SIZE -
> -						  (pkt_inline_sz - 2);
> -			uintptr_t addr_end = (addr + max_inline) &
> +			unsigned int inline_room = max_inline *
> +						   RTE_CACHE_LINE_SIZE -
> +						   (pkt_inline_sz - 2);
> +			uintptr_t addr_end = (addr + inline_room) &
>  					     ~(RTE_CACHE_LINE_SIZE - 1);
>  			unsigned int copy_b = (addr_end > addr) ?
>  				RTE_MIN((addr_end - addr), length) :
> -- 
> 2.11.0
 

Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>

-- 
Nélio Laranjeiro
6WIND

Re: [PATCH] net/mlx5: change calculating inline room for Tx

[Ferruh Yigit]

On 4/18/2017 7:43 AM, Nélio Laranjeiro wrote:
> On Thu, Apr 13, 2017 at 10:46:51AM -0700, Yongseok Koh wrote:
>> Current implementation is error-prone if the max inline size
>> (txq->max_inilne) is decoupled from txq->inline_en and becomes zero. If it
>> becomes zero, HW can crash due to WQ overflow.
>>
>> Signed-off-by: Yongseok Koh <yskoh@mellanox.com>
>> Acked-by: Shahaf Shuler <shahafs@mellanox.com>

> Acked-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com>

Applied to dpdk-next-net/master, thanks.