Re: [PATCH bpf-next 3/3] libbpf, xsk: select AF_XDP BPF program based on kernel version

[Jesper Dangaard Brouer <brouer@redhat.com>]

On Fri, 22 Jan 2021 11:53:51 +0100
Björn Töpel <bjorn.topel@gmail.com> wrote:

> From: Björn Töpel <bjorn.topel@intel.com>
> 
> Add detection for kernel version, and adapt the BPF program based on
> kernel support. This way, users will get the best possible performance
> from the BPF program.

You say "detection for kernel version", but doesn't the code detect the
feature rather than kernel version ?

If so, please update the description.

> Reviewed-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
> Acked-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
> Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
> Signed-off-by: Marek Majtyka  <alardam@gmail.com>
> ---
>  tools/lib/bpf/xsk.c | 82 +++++++++++++++++++++++++++++++++++++++++++--
>  1 file changed, 79 insertions(+), 3 deletions(-)
> 
> diff --git a/tools/lib/bpf/xsk.c b/tools/lib/bpf/xsk.c
> index e3e41ceeb1bc..1df8c133a5bc 100644
> --- a/tools/lib/bpf/xsk.c
> +++ b/tools/lib/bpf/xsk.c
> @@ -46,6 +46,11 @@
>   #define PF_XDP AF_XDP
>  #endif
>  
> +enum xsk_prog {
> +	XSK_PROG_FALLBACK,
> +	XSK_PROG_REDIRECT_FLAGS,
> +};
> +
>  struct xsk_umem {
>  	struct xsk_ring_prod *fill_save;
>  	struct xsk_ring_cons *comp_save;
> @@ -351,6 +356,55 @@ int xsk_umem__create_v0_0_2(struct xsk_umem **umem_ptr, void *umem_area,
>  COMPAT_VERSION(xsk_umem__create_v0_0_2, xsk_umem__create, LIBBPF_0.0.2)
>  DEFAULT_VERSION(xsk_umem__create_v0_0_4, xsk_umem__create, LIBBPF_0.0.4)
>  
> +
> +static enum xsk_prog get_xsk_prog(void)
> +{
> +	enum xsk_prog detected = XSK_PROG_FALLBACK;
> +	struct bpf_load_program_attr prog_attr;
> +	struct bpf_create_map_attr map_attr;
> +	__u32 size_out, retval, duration;
> +	char data_in = 0, data_out;
> +	struct bpf_insn insns[] = {
> +		BPF_LD_MAP_FD(BPF_REG_1, 0),
> +		BPF_MOV64_IMM(BPF_REG_2, 0),
> +		BPF_MOV64_IMM(BPF_REG_3, XDP_PASS),
> +		BPF_EMIT_CALL(BPF_FUNC_redirect_map),
> +		BPF_EXIT_INSN(),
> +	};
> +	int prog_fd, map_fd, ret;
> +
> +	memset(&map_attr, 0, sizeof(map_attr));
> +	map_attr.map_type = BPF_MAP_TYPE_XSKMAP;
> +	map_attr.key_size = sizeof(int);
> +	map_attr.value_size = sizeof(int);
> +	map_attr.max_entries = 1;
> +
> +	map_fd = bpf_create_map_xattr(&map_attr);
> +	if (map_fd < 0)
> +		return detected;
> +
> +	insns[0].imm = map_fd;
> +
> +	memset(&prog_attr, 0, sizeof(prog_attr));
> +	prog_attr.prog_type = BPF_PROG_TYPE_XDP;
> +	prog_attr.insns = insns;
> +	prog_attr.insns_cnt = ARRAY_SIZE(insns);
> +	prog_attr.license = "GPL";
> +
> +	prog_fd = bpf_load_program_xattr(&prog_attr, NULL, 0);
> +	if (prog_fd < 0) {
> +		close(map_fd);
> +		return detected;
> +	}
> +
> +	ret = bpf_prog_test_run(prog_fd, 0, &data_in, 1, &data_out, &size_out, &retval, &duration);
> +	if (!ret && retval == XDP_PASS)
> +		detected = XSK_PROG_REDIRECT_FLAGS;
> +	close(prog_fd);
> +	close(map_fd);
> +	return detected;
> +}
> +
>  static int xsk_load_xdp_prog(struct xsk_socket *xsk)
>  {
>  	static const int log_buf_size = 16 * 1024;
> @@ -358,7 +412,7 @@ static int xsk_load_xdp_prog(struct xsk_socket *xsk)
>  	char log_buf[log_buf_size];
>  	int err, prog_fd;
>  
> -	/* This is the C-program:
> +	/* This is the fallback C-program:
>  	 * SEC("xdp_sock") int xdp_sock_prog(struct xdp_md *ctx)
>  	 * {
>  	 *     int ret, index = ctx->rx_queue_index;
> @@ -414,9 +468,31 @@ static int xsk_load_xdp_prog(struct xsk_socket *xsk)
>  		/* The jumps are to this instruction */
>  		BPF_EXIT_INSN(),
>  	};
> -	size_t insns_cnt = sizeof(prog) / sizeof(struct bpf_insn);
>  
> -	prog_fd = bpf_load_program(BPF_PROG_TYPE_XDP, prog, insns_cnt,
> +	/* This is the post-5.3 kernel C-program:
> +	 * SEC("xdp_sock") int xdp_sock_prog(struct xdp_md *ctx)
> +	 * {
> +	 *     return bpf_redirect_map(&xsks_map, ctx->rx_queue_index, XDP_PASS);
> +	 * }
> +	 */
> +	struct bpf_insn prog_redirect_flags[] = {
> +		/* r2 = *(u32 *)(r1 + 16) */
> +		BPF_LDX_MEM(BPF_W, BPF_REG_2, BPF_REG_1, 16),
> +		/* r1 = xskmap[] */
> +		BPF_LD_MAP_FD(BPF_REG_1, ctx->xsks_map_fd),
> +		/* r3 = XDP_PASS */
> +		BPF_MOV64_IMM(BPF_REG_3, 2),
> +		/* call bpf_redirect_map */
> +		BPF_EMIT_CALL(BPF_FUNC_redirect_map),
> +		BPF_EXIT_INSN(),
> +	};
> +	size_t insns_cnt[] = {sizeof(prog) / sizeof(struct bpf_insn),
> +			      sizeof(prog_redirect_flags) / sizeof(struct bpf_insn),
> +	};
> +	struct bpf_insn *progs[] = {prog, prog_redirect_flags};
> +	enum xsk_prog option = get_xsk_prog();
> +
> +	prog_fd = bpf_load_program(BPF_PROG_TYPE_XDP, progs[option], insns_cnt[option],
>  				   "LGPL-2.1 or BSD-2-Clause", 0, log_buf,
>  				   log_buf_size);
>  	if (prog_fd < 0) {



-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Principal Kernel Engineer at Red Hat
  LinkedIn: http://www.linkedin.com/in/brouer