Re: [PATCH] Introduce the pkill_on_warn boot parameter

On Wed, Jul 27, 2022 at 9:17 AM Alexey Khoroshilov
<[email protected]> wrote:
>
> We see a number of cases where WARNING is used to inform userspace that
> it is doing something wrong, e.g.
> https://elixir.bootlin.com/linux/v5.19-rc8/source/net/can/j1939/socket.c#L181
> https://elixir.bootlin.com/linux/v5.19-rc8/source/drivers/video/fbdev/core/fbmem.c#L1023

That first case is entirely bogus.

WARN_ON() should only be used for "This cannot happen, but if it does,
I want to know how we got here".

But the second case is fine: Using "pr_warn()" is fine. A kernel
warning (without a backtrace) is a normal thing for something that is
deprecated or questionable, and you want to tell the user that "this
app is doing something wrong".

So if that j1939 thing is something that can be triggered by a user,
then the backtrace should be reported to the driver maintainer, and
then either

 (a) the WARN_ON_ONCE() should just be removed ("ok, this can happen,
we understand why it can happen, and it's fine")

 (b) the problem the WARN_ON_ONCE() reports about should be made
impossible some way

 (c) it might be downgraded to a pr_warn() if people really want to
tell user space that "guys, you're doing something wrong" and it's
considered a useful warning.

Honestly, for something like that j1939 can driver, I doubt (c) is
ever an option. The "return -EBUSY" is the only real information that
a user needs.

               Linus