On Sat, Oct 19, 2019 at 11:48 AM Alexei Starovoitov
[off-list ref] wrote:

On Fri, Oct 18, 2019 at 11:33:40AM +0200, Magnus Karlsson wrote:

quoted

+
+   #include <linux/bpf.h>
+   #include "bpf_helpers.h"
+
+   #define MAX_SOCKS 16
+
+   struct {
+        __uint(type, BPF_MAP_TYPE_XSKMAP);
+        __uint(max_entries, MAX_SOCKS);
+        __uint(key_size, sizeof(int));
+        __uint(value_size, sizeof(int));
+   } xsks_map SEC(".maps");
+
+   struct {
+        __uint(type, BPF_MAP_TYPE_PERCPU_ARRAY);
+        __uint(max_entries, 1);
+        __type(key, int);
+        __type(value, unsigned int);
+   } rr_map SEC(".maps");

hmm. does xsks_map compile?

Yes. Actually, I wrote a new sample to demonstrate this feature and to
test the code above. I will send that patch set (contains some small
additions to libbpf also to be able to support this) to bpf-next.
Though, if I used the __type declarations of the rr_map PERCPU_ARRAY I
got this warning: "pr_warning("Error in
bpf_create_map_xattr(%s):%s(%d). Retrying without BTF.\n")", so I had
to change it to the type above that is also used for SOCKMAP. Some
enablement that is missing for XSKMAP? Have not dug into it.

quoted

+
+   SEC("xdp_sock") int xdp_sock_prog(struct xdp_md *ctx)
+   {
+     int key = 0, idx;
+     unsigned int *rr;
+
+     rr = bpf_map_lookup_elem(&rr_map, &key);
+     if (!rr)
+        return XDP_ABORTED;

could you please use global data and avoid lookup?
The run-time will be much faster.

Good idea. Will do and change the new sample app too.