On 2023/10/8 14:45, Eric Dumazet wrote:

On Sat, Oct 7, 2023 at 8:34 AM Yajun Deng [off-list ref] wrote:

quoted

On 2023/10/7 13:29, Eric Dumazet wrote:

quoted

On Sat, Oct 7, 2023 at 7:06 AM Yajun Deng [off-list ref] wrote:

quoted

Although there is a kfree_skb_reason() helper function that can be used to
find the reason why this skb is dropped, but most callers didn't increase
one of rx_dropped, tx_dropped, rx_nohandler and rx_otherhost_dropped.

...

quoted

+
+void netdev_core_stats_inc(struct net_device *dev, u32 offset)
+{
+       /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
+       struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats);
+       unsigned long *field;
+
+       if (unlikely(!p))
+               p = netdev_core_stats_alloc(dev);
+
+       if (p) {
+               field = (unsigned long *)((void *)this_cpu_ptr(p) + offset);
+               WRITE_ONCE(*field, READ_ONCE(*field) + 1);

This is broken...

As I explained earlier, dev_core_stats_xxxx(dev) can be called from
many different contexts:

1) process contexts, where preemption and migration are allowed.
2) interrupt contexts.

Adding WRITE_ONCE()/READ_ONCE() is not solving potential races.

I _think_ I already gave you how to deal with this ?

Yes, I replied in v6.

https://lore.kernel.org/all/e25b5f3c-bd97-56f0-de86-b93a3172870d@linux.dev/ (local)

quoted

Please try instead:

+void netdev_core_stats_inc(struct net_device *dev, u32 offset)
+{
+       /* This READ_ONCE() pairs with the write in netdev_core_stats_alloc() */
+       struct net_device_core_stats __percpu *p = READ_ONCE(dev->core_stats);
+       unsigned long __percpu *field;
+
+       if (unlikely(!p)) {
+               p = netdev_core_stats_alloc(dev);
+               if (!p)
+                       return;
+       }
+       field = (__force unsigned long __percpu *)((__force void *)p + offset);
+       this_cpu_inc(*field);
+}

This wouldn't trace anything even the rx_dropped is in increasing. It
needs to add an extra operation, such as:

I honestly do not know what you are talking about.

Have you even tried to change your patch to use

field = (__force unsigned long __percpu *)((__force void *)p + offset);
this_cpu_inc(*field);

Yes, I tested this code. But the following couldn't show anything even 
if the rx_dropped is increasing.

'sudo python3 /usr/share/bcc/tools/trace netdev_core_stats_inc'

It needs to add anything else. The above command will show correctly.

Instead of the clearly buggy code you had instead :

     field = (unsigned long *)((void *)this_cpu_ptr(p) + offset);
      WRITE_ONCE(*field, READ_ONCE(*field) + 1);

If your v7 submission was ok for tracing what you wanted,
I fail to see why a v8 with 3 lines changed would not work.

Me too.

If I add a pr_info in your code, the kprobe will be ok.