Re: [PATCH v2] printk_legacy_map: use LD_WAIT_CONFIG instead of LD_WAIT_SLEEP

[Petr Mladek <pmladek@suse.com>]

On Thu 2025-10-23 21:14:42, Sebastian Andrzej Siewior wrote:
> On 2025-10-23 17:46:58 [+0200], Oleg Nesterov wrote:
> > Again, quite possibly I am wrong. please see my reply in another thread,
> > https://lore.kernel.org/all/20251023152942.GC26461@redhat.com/
> > 
> > On 10/23, Sebastian Andrzej Siewior wrote:
> > >
> > > This does not matter. My point is that there no need for this ifdefery.
> > 
> > I disagree. Rightly or not. To me this ifdefery in printk.c make the intent
> > more clear.
> > 
> > I am still thinking about the possible cleanup of the current usage of
> > DEFINE_WAIT_OVERRIDE_MAP(), but I think that, whatever we do, this cleanup
> > should take CONFIG_RT into account.
> 
> Right. Please just do s/SLEEP/CONFIG as discussed. Please leave
> PROVE_RAW_LOCK_NESTING out of it while arguing why this change is
> correct.

It is clear that the commit message and the comment above the mapping
caused some confusion. I thought about better wording.

I wanted to be as clear as possible, But the problem is that everyone
has different background and might understand the same term
differently. Also I am not a native speaker.


/*
 * Some legacy console drivers might violate raw_spinlock/spinlock nesting
 * rules when printk() was called under a raw_spinlock and the driver used
 * a spinlock. It is not a real problem because the legacy drivers should
 * never be called directly from printk() in PREEMPT_RT.
 *
 * This map is used to pretend that printk() was called under a normal spinlock
 * to hide the above described locking violation. It still allows to catch
 * other problems, for example, possible ABBA deadlocks or sleeping locks.
 *
 * The mapping is not used in PREEMPT_RT which allows to catch bugs when
 * the legacy console driver would get called from an atomic context by mistake.
 */


And the commit message might be:

<commit_message>
printk_legacy_map: use LD_WAIT_CONFIG instead of LD_WAIT_SLEEP

printk_legacy_map is used to hide possible violations of
raw_spinlock/spinlock nesting when printk() calls legacy console
drivers directly. It is not a real problem in !PREEMPT_RT mode and
the problematic code path should never be called in PREEMPT_RT mode.

However, LD_WAIT_SLEEP is not exactly right. It fools lockdep as if it
is fine to acquire a sleeping lock.

Change DEFINE_WAIT_OVERRIDE_MAP(printk_legacy_map) to use LD_WAIT_CONFIG.

Also, update the comment to better describe the purpose of the mapping.
</commit_message>


Is this better and acceptable, please?
If not then please provide alternatives ;-)

Best Regards,
Petr