Re: [patch V2 26/45] genirq/chip: Rework irq_set_handler() variants

[Nathan Chancellor <nathan@kernel.org>]

On Sun, May 11, 2025 at 07:29:11PM +0200, Miguel Ojeda wrote:
> On Fri, 09 May 2025 14:22:11 +0100 Nathan Chancellor <nathan@kernel.org> wrote:
> >
> > I am investigating some cases where
> >
> > WARN(!irqs_disabled(), "Interrupts were enabled early\n");
> >
> > in start_kernel() in init/main.c is triggered in certain builds with
> > clang after patch 23 of this series (very bizarre since the conversion
> > seems to be correct) and I happened to notice that this conversion seems
> > to be incorrect? Should this be scoped_irqdesc_get_and_buslock() like
> > below?
> 
> Yeah, I am also seeing this in next-20250509 in a LLVM=1 arm64 defconfig + Rust
> build under QEMU.

I noticed that the warning was reproducible with just the first patch of
the series that adds the lock guards and patch 23 but also several other
individual patches within the series, as I could not just revert patch
23 on next-20250509 to fix it. I have no idea why yet because I have not
had the chance to actually sit down and dig into it but this diff fixes
every instance of the warning that I saw in my tests... :/ could be a
compiler bug or just some difference in behavior between compilers.

diff --git a/kernel/irq/internals.h b/kernel/irq/internals.h
index bd2db6ebb98e..94f463de8f26 100644
--- a/kernel/irq/internals.h
+++ b/kernel/irq/internals.h
@@ -176,10 +176,9 @@ __DEFINE_UNLOCK_GUARD(irqdesc_lock, struct irq_desc,
 static inline class_irqdesc_lock_t class_irqdesc_lock_constructor(unsigned int irq, bool bus,
 								  unsigned int check)
 {
-	class_irqdesc_lock_t _t = {
-		.bus	= bus,
-		.lock	= __irq_get_desc_lock(irq, &_t.flags, bus, check),
-	};
+	class_irqdesc_lock_t _t = {};
+	_t.bus	= bus;
+	_t.lock	= __irq_get_desc_lock(irq, &_t.flags, bus, check);
 	return _t;
 }