Re: [BUG] printk/nbcon can use RCU illegally prior to CPU online

[Steven Rostedt <rostedt@goodmis.org>]

On Thu, 18 Jul 2024 14:34:41 -0500
Andrew Halaney <ahalaney@redhat.com> wrote:


> Heh, exactly why I made a "reproducer" for the printk/RCU side of the
> splat with linux-rt-devel before reporting.
> 
> So my takeaway is that printk is not considered safe to use in all
> contexts, including during SMP bringup (because RCU's at play and not
> yet ready to be used on those CPUs). So if we do printk for reasons we
> can't control (say some firmware problem causing an exception) and try
> to print you could get the RCU warning about printk/nbcon usage in
> addition to the actual error, and that's just something to accept.

Well, in most cases people don't run a debug kernel in production and
if this printk was triggered in production, that warning would not show
up (it's only there when lockdep is enabled).

> 
> Ideally though we don't printk() in those situations... so if its badly
> behaving code calling printk() that's a different story.

Right. printk() does get called in locations it shouldn't simply due to
a bug happening. But then, it's just a best effort to get information
to the user.

> 
> Let me know if you think that recap is bogus.

I think you summarized it well.

-- Steve