Re: [PATCH net-next v3 2/2] netcons: Add udp send fail statistics to netconsole

On Thu, Sep 12, 2024 at 01:58:12PM -0400, Maksym Kutsevol wrote:

Hey Breno,
Thanks for looking into this.

On Thu, Sep 12, 2024 at 1:49 PM Breno Leitao [off-list ref] wrote:

quoted

Hello Maksym,

Thanks for the patch, it is looking good. A few nits:

On Thu, Sep 12, 2024 at 10:28:52AM -0700, Maksym Kutsevol wrote:

quoted

+/**
+ * netpoll_send_udp_count_errs - Wrapper for netpoll_send_udp that counts errors
+ * @nt: target to send message to
+ * @msg: message to send
+ * @len: length of message
+ *
+ * Calls netpoll_send_udp and classifies the return value. If an error
+ * occurred it increments statistics in nt->stats accordingly.
+ * Only calls netpoll_send_udp if CONFIG_NETCONSOLE_DYNAMIC is disabled.
+ */
+static void netpoll_send_udp_count_errs(struct netconsole_target *nt, const char *msg, int len)
+{
+     int result = netpoll_send_udp(&nt->np, msg, len);

Would you get a "variable defined but not used" type of eror if
CONFIG_NETCONSOLE_DYNAMIC is disabled?

Most probably yes, I'll check. If so, I'll add __maybe_unused in the
next iteration.

quoted

quoted

+
+     if (IS_ENABLED(CONFIG_NETCONSOLE_DYNAMIC)) {
+             if (result == NET_XMIT_DROP) {
+                     u64_stats_update_begin(&nt->stats.syncp);
+                     u64_stats_inc(&nt->stats.xmit_drop_count);
+                     u64_stats_update_end(&nt->stats.syncp);
+             } else if (result == -ENOMEM) {
+                     u64_stats_update_begin(&nt->stats.syncp);
+                     u64_stats_inc(&nt->stats.enomem_count);
+                     u64_stats_update_end(&nt->stats.syncp);
+             }
+     }

Would this look better?

        if (IS_ENABLED(CONFIG_NETCONSOLE_DYNAMIC)) {
                u64_stats_update_begin(&nt->stats.syncp);

                if (result == NET_XMIT_DROP)
                        u64_stats_inc(&nt->stats.xmit_drop_count);
                else if (result == -ENOMEM)
                        u64_stats_inc(&nt->stats.enomem_count);
                else
                        WARN_ONCE(true, "invalid result: %d\n", result)

                u64_stats_update_end(&nt->stats.syncp);
        }

1. It will warn on positive result
2. If the last `else` is removed, it attempts locking when the result
is positive, so I'd not do it this way.

Correct. We could replace the WARN_ONCE(true, ..) by
WARN_ONCE(result,..), but this might look worse.

Let's keep the way you proposed.

Other than that, the patch looks good.

Thanks