[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <24f5387bc2bec23627ea9cfaec4ed4e416c60510.camel@kernel.org>
Date: Mon, 19 Jul 2021 13:29:00 -0700
From: Saeed Mahameed <saeed@...nel.org>
To: David Ramirez <davramir@....edu>,
Leon Romanovsky <leonro@...dia.com>
Cc: "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
"linux-rdma@...r.kernel.org" <linux-rdma@...r.kernel.org>
Subject: Re: XDP applications running in driver mode on mlx5_core can't
access various helper functions
On Thu, 2021-07-15 at 13:59 +0000, David Ramirez wrote:
> Hey all,
>
> I am having issues with calling some bpf helper functions
> when running my XDP program in driver on the mlx5_core driver.
> Several of the helpers I've tried to use for ringbuf and maps always
> return 0.
> While this may seem to imply that for functions that merely return
> null instead of
> a pointer in case of an error are working as intended,
> some functions which return negative on failure and 0 on success are
> are also affected,
> as while they return 0, they do not result in the desired effect.
>
> Observed examples:
> - bpf_ringbuf_output always returns 0, but no data is pushed to the
> ringbuf
> - bpf_map_update_elem always returns 0, but the element is not
> updated
> - bpf_ringbuf_reserve always returns 0
> - bpf_map_lookup_elem always returns 0
>
> I'm uncertain if this is a driver specific issue or an ebpf issue.
this is purely ebpf issue or user prog issue.
1. the ebpf program/helpers are agnostic to run mode, driver/native
v.s. skb/generic modes.
2. the above helpers are pure bpf callbacks, the driver is not
involved.
> Testing with xdp in driver mode on veth devices works as expected,
> which suggests this is more likely a driver issue.
>
The driver or its implementation are not involved with the above
helpers, so further debug is required in the user prog or the helpers
logic themselves.
> Additional Details:
>
> Linux Distro: Ubuntu 21.04
> Linux Kernel version: 5.11.0-18-generic
> driver: mlx5_core
> version: 5.11.0-18-generic
>
> Thank you,
> David Ramirez
Powered by blists - more mailing lists