[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <f1b53677-7432-4343-af63-b29bf5b17e0d@gmail.com>
Date: Sun, 3 Nov 2024 09:33:45 +0200
From: Tariq Toukan <ttoukan.linux@...il.com>
To: Daniel Machon <daniel.machon@...rochip.com>,
Tariq Toukan <tariqt@...dia.com>
Cc: "David S. Miller" <davem@...emloft.net>, Jakub Kicinski
<kuba@...nel.org>, Paolo Abeni <pabeni@...hat.com>,
Eric Dumazet <edumazet@...gle.com>, Andrew Lunn <andrew+netdev@...n.ch>,
netdev@...r.kernel.org, Saeed Mahameed <saeedm@...dia.com>,
Gal Pressman <gal@...dia.com>, Leon Romanovsky <leonro@...dia.com>,
William Tu <witu@...dia.com>, Parav Pandit <parav@...dia.com>
Subject: Re: [PATCH net-next 5/5] net/mlx5e: do not create xdp_redirect for
non-uplink rep
On 01/11/2024 15:22, Daniel Machon wrote:
> Hi Tariq, William
>
>> From: William Tu <witu@...dia.com>
>>
>> XDP and XDP socket require extra SQ/RQ/CQs. Most of these resources
>> are dynamically created: no XDP program loaded, no resources are
>> created. One exception is the SQ/CQ created for XDP_REDRIECT, used
>> for other netdev to forward packet to mlx5 for transmit. The patch
>> disables creation of SQ and CQ used for egress XDP_REDIRECT, by
>> checking whether ndo_xdp_xmit is set or not.
>>
>> For netdev without XDP support such as non-uplink representor, this
>> saves around 0.35MB of memory, per representor netdevice per channel.
>>
>> Signed-off-by: William Tu <witu@...dia.com>
>> Reviewed-by: Parav Pandit <parav@...dia.com>
>> Signed-off-by: Tariq Toukan <tariqt@...dia.com>
>> ---
>> .../net/ethernet/mellanox/mlx5/core/en_main.c | 17 +++++++++++------
>> 1 file changed, 11 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
>> index 2f609b92d29b..59d7a0e28f24 100644
>> --- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
>> +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
>> @@ -2514,6 +2514,7 @@ static int mlx5e_open_queues(struct mlx5e_channel *c,
>> struct mlx5e_params *params,
>> struct mlx5e_channel_param *cparam)
>> {
>> + const struct net_device_ops *netdev_ops = c->netdev->netdev_ops;
>> struct dim_cq_moder icocq_moder = {0, 0};
>> struct mlx5e_create_cq_param ccp;
>> int err;
>> @@ -2534,10 +2535,12 @@ static int mlx5e_open_queues(struct mlx5e_channel *c,
>> if (err)
>> goto err_close_icosq_cq;
>>
>> - c->xdpsq = mlx5e_open_xdpredirect_sq(c, params, cparam, &ccp);
>> - if (IS_ERR(c->xdpsq)) {
>> - err = PTR_ERR(c->xdpsq);
>> - goto err_close_tx_cqs;
>> + if (netdev_ops->ndo_xdp_xmit) {
>
> Is it possible to have ndo_xdp_xmit() set, but *not* have an XDP prog attached
> to the netdevice? I see that c->xdp = !!params->xdp_prog - could that be
> used instead?
>
> /Daniel
>
Hi Daniel,
Unlike drivers of other vendors, in mlx5e we have the
egress-xdp-redirect feature enabled without the need to load a dummy xdp
program in rx.
Powered by blists - more mailing lists