Re: [PATCH net-next] net: skbuff: Skip early return in skb_unref when debugging

From: Paolo Abeni
Date: Tue Jul 30 2024 - 10:37:35 EST


On 7/30/24 16:10, Jakub Kicinski wrote:
On Tue, 30 Jul 2024 13:15:57 +0200 Paolo Abeni wrote:
I was under impression that entire reason for CONFIG_DEBUG_NET was
to enable more checks for fuzzers and the like, i.e. NOT for production
kernels.

I feel like I already had this discussion and I forgot the outcome, if
so I'm sorry. To me the "but is safe to select." part in the knob
description means this could be enabled in production, and AFAICS the
CONFIG_DEBUG_NET-enabled code so far respects that assumption.

I believe the previous discussion was page pool specific and there
wasn't as much of a conclusion as an acquiescence (read: we had more
important things on our minds than that argument ;)).

Should we set a bar for how much perf impact is okay?

I think that better specifying the general guidance/expectation should be enough. What about extending the knob description with something alike:
---
diff --git a/net/Kconfig.debug b/net/Kconfig.debug
index 5e3fffe707dd..058cf031913b 100644
--- a/net/Kconfig.debug
+++ b/net/Kconfig.debug
@@ -24,3 +24,5 @@ config DEBUG_NET
help
Enable extra sanity checks in networking.
This is mostly used by fuzzers, but is safe to select.
+ This could introduce some very minimal overhead and
+ is not suggested for production systems.