bpf.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH bpf-next] libbpf: remove explicit XSKMAP lookup from AF_XDP XDP program
@ 2019-10-20 17:07 Björn Töpel
  2019-10-20 19:53 ` Toke Høiland-Jørgensen
  0 siblings, 1 reply; 3+ messages in thread
From: Björn Töpel @ 2019-10-20 17:07 UTC (permalink / raw)
  To: netdev, ast, daniel
  Cc: Björn Töpel, bpf, magnus.karlsson, magnus.karlsson,
	toke, sridhar.samudrala

From: Björn Töpel <bjorn.topel@intel.com>

In commit 43e74c0267a3 ("bpf_xdp_redirect_map: Perform map lookup in
eBPF helper") the bpf_redirect_map() helper learned to do map lookup,
which means that the explicit lookup in the XDP program for AF_XDP is
not needed.

This commit removes the map lookup, which simplifies the BPF code and
improves the performance for the "rx_drop" [1] scenario with ~4%.

[1] # xdpsock -i eth0 -z -r

Suggested-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Björn Töpel <bjorn.topel@intel.com>
---
 tools/lib/bpf/xsk.c | 24 +++++-------------------
 1 file changed, 5 insertions(+), 19 deletions(-)

diff --git a/tools/lib/bpf/xsk.c b/tools/lib/bpf/xsk.c
index b0f532544c91..8e35de3cb443 100644
--- a/tools/lib/bpf/xsk.c
+++ b/tools/lib/bpf/xsk.c
@@ -278,29 +278,15 @@ static int xsk_load_xdp_prog(struct xsk_socket *xsk)
 	 *
 	 *     // A set entry here means that the correspnding queue_id
 	 *     // has an active AF_XDP socket bound to it.
-	 *     if (bpf_map_lookup_elem(&xsks_map, &index))
-	 *         return bpf_redirect_map(&xsks_map, index, 0);
-	 *
-	 *     return XDP_PASS;
+	 *     return bpf_redirect_map(&xsks_map, index, XDP_PASS);
 	 * }
 	 */
 	struct bpf_insn prog[] = {
-		/* r1 = *(u32 *)(r1 + 16) */
-		BPF_LDX_MEM(BPF_W, BPF_REG_1, BPF_REG_1, 16),
-		/* *(u32 *)(r10 - 4) = r1 */
-		BPF_STX_MEM(BPF_W, BPF_REG_10, BPF_REG_1, -4),
-		BPF_MOV64_REG(BPF_REG_2, BPF_REG_10),
-		BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4),
-		BPF_LD_MAP_FD(BPF_REG_1, xsk->xsks_map_fd),
-		BPF_EMIT_CALL(BPF_FUNC_map_lookup_elem),
-		BPF_MOV64_REG(BPF_REG_1, BPF_REG_0),
-		BPF_MOV32_IMM(BPF_REG_0, 2),
-		/* if r1 == 0 goto +5 */
-		BPF_JMP_IMM(BPF_JEQ, BPF_REG_1, 0, 5),
-		/* r2 = *(u32 *)(r10 - 4) */
+		/* r2 = *(u32 *)(r1 + 16) */
+		BPF_LDX_MEM(BPF_W, BPF_REG_2, BPF_REG_1, 16),
 		BPF_LD_MAP_FD(BPF_REG_1, xsk->xsks_map_fd),
-		BPF_LDX_MEM(BPF_W, BPF_REG_2, BPF_REG_10, -4),
-		BPF_MOV32_IMM(BPF_REG_3, 0),
+		/* r3 = XDP_PASS */
+		BPF_MOV32_IMM(BPF_REG_3, 2),
 		BPF_EMIT_CALL(BPF_FUNC_redirect_map),
 		/* The jumps are to this instruction */
 		BPF_EXIT_INSN(),
-- 
2.20.1


^ permalink raw reply related	[flat|nested] 3+ messages in thread

* Re: [PATCH bpf-next] libbpf: remove explicit XSKMAP lookup from AF_XDP XDP program
  2019-10-20 17:07 [PATCH bpf-next] libbpf: remove explicit XSKMAP lookup from AF_XDP XDP program Björn Töpel
@ 2019-10-20 19:53 ` Toke Høiland-Jørgensen
  2019-10-21  5:03   ` Björn Töpel
  0 siblings, 1 reply; 3+ messages in thread
From: Toke Høiland-Jørgensen @ 2019-10-20 19:53 UTC (permalink / raw)
  To: Björn Töpel, netdev, ast, daniel
  Cc: Björn Töpel, bpf, magnus.karlsson, magnus.karlsson,
	sridhar.samudrala

Björn Töpel <bjorn.topel@gmail.com> writes:

> From: Björn Töpel <bjorn.topel@intel.com>
>
> In commit 43e74c0267a3 ("bpf_xdp_redirect_map: Perform map lookup in
> eBPF helper") the bpf_redirect_map() helper learned to do map lookup,
> which means that the explicit lookup in the XDP program for AF_XDP is
> not needed.
>
> This commit removes the map lookup, which simplifies the BPF code and
> improves the performance for the "rx_drop" [1] scenario with ~4%.

Nice, 4% is pretty good!

I wonder if the program needs to be backwards-compatible (with pre-5.3
kernels), though?

You can do that by something like this:

ret = bpf_redirect_map(&xsks_map, index, XDP_PASS);
if (ret > 0)
  return ret;

if (bpf_map_lookup_elem(&xsks_map, &index))
   return bpf_redirect_map(&xsks_map, index, 0);
return XDP_PASS;


This works because bpf_redirect_map() prior to 43e74c0267a3 will return
XDP_ABORTED on a non-0 flags value.

-Toke


^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH bpf-next] libbpf: remove explicit XSKMAP lookup from AF_XDP XDP program
  2019-10-20 19:53 ` Toke Høiland-Jørgensen
@ 2019-10-21  5:03   ` Björn Töpel
  0 siblings, 0 replies; 3+ messages in thread
From: Björn Töpel @ 2019-10-21  5:03 UTC (permalink / raw)
  To: Toke Høiland-Jørgensen
  Cc: Netdev, Alexei Starovoitov, Daniel Borkmann,
	Björn Töpel, bpf, Magnus Karlsson, Karlsson, Magnus,
	Samudrala, Sridhar

On Sun, 20 Oct 2019 at 21:53, Toke Høiland-Jørgensen <toke@redhat.com> wrote:
>
> Björn Töpel <bjorn.topel@gmail.com> writes:
>
> > From: Björn Töpel <bjorn.topel@intel.com>
> >
> > In commit 43e74c0267a3 ("bpf_xdp_redirect_map: Perform map lookup in
> > eBPF helper") the bpf_redirect_map() helper learned to do map lookup,
> > which means that the explicit lookup in the XDP program for AF_XDP is
> > not needed.
> >
> > This commit removes the map lookup, which simplifies the BPF code and
> > improves the performance for the "rx_drop" [1] scenario with ~4%.
>
> Nice, 4% is pretty good!
>
> I wonder if the program needs to be backwards-compatible (with pre-5.3
> kernels), though?
>
> You can do that by something like this:
>
> ret = bpf_redirect_map(&xsks_map, index, XDP_PASS);
> if (ret > 0)
>   return ret;
>
> if (bpf_map_lookup_elem(&xsks_map, &index))
>    return bpf_redirect_map(&xsks_map, index, 0);
> return XDP_PASS;
>

Ah, yes. Thanks for pointing that out. I'll do a respin.


Thanks,
Björn

>
> This works because bpf_redirect_map() prior to 43e74c0267a3 will return
> XDP_ABORTED on a non-0 flags value.
>
> -Toke
>

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2019-10-21  5:03 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-10-20 17:07 [PATCH bpf-next] libbpf: remove explicit XSKMAP lookup from AF_XDP XDP program Björn Töpel
2019-10-20 19:53 ` Toke Høiland-Jørgensen
2019-10-21  5:03   ` Björn Töpel

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).