Re: [PATCH] irq: Make refcount_interrupt kunit test selectable

[Boqun Feng <boqun@kernel.org>]

On Mon, Aug 24, 2026 at 11:02:26AM -0400, Radu Rendec wrote:
> On Sun, 2026-08-23 at 22:23 -0700, Boqun Feng wrote:
> > On Sun, Aug 23, 2026 at 05:52:24PM -0400, Radu Rendec wrote:
> > > On Sun, 2026-08-23 at 14:08 -0700, Boqun Feng wrote:
> > > > On Sun, Aug 23, 2026 at 04:46:23PM -0400, Radu Rendec wrote:
> > > > > On Sun, 2026-08-23 at 17:17 +0000, Kuan-Wei Chiu wrote:
> > > > > > Currently, refcount_interrupt_test is built unconditionally when
> > > > > > CONFIG_KUNIT is enabled, causing it to run unexpectedly during boot.
> > > > > > 
> > > > > > Fix this by introducing CONFIG_REFCOUNT_INTERRUPT_KUNIT_TEST so the
> > > > > > test can be configured independently, following standard kunit
> > > > > > practices.
> > > > > > 
> > > > > > Fixes: 07a88e2bcd5b ("irq: Add KUnit test for refcounted interrupt enable/disable")
> > > > > > Signed-off-by: Kuan-Wei Chiu <visitorckw@gmail.com>
> > > > > > ---
> > > > > >  kernel/irq/Kconfig  | 12 ++++++++++++
> > > > > >  kernel/irq/Makefile |  2 +-
> > > > > >  2 files changed, 13 insertions(+), 1 deletion(-)
> > > > > 
> > > > > Can someone please explain me why this needs to be a separate unit test
> > > > > in the first place and can't be integrated into the (previously)
> > > > > existing unit test gated by IRQ_KUNIT_TEST?
> > > > > 
> > > > 
> > > > If you look at the irq_test.c, you will find that it primarily tests the
> > > > {request,disable,enable}_irq() API (i.e. allocating IRQ vectors and
> > > > enabling and disabling them), which is not the same scope as
> > > > local_interrupt_{disable,enable}() (i.e. operating the CPU irq disabling
> > > > status), so it makes sense to have a separate test case.
> > > 
> > > Thanks! If we want to keep the unit tests separated/grouped by scope,
> > > then yes, it makes sense for sure. What I don't like about this is:
> > >  * The naming of the two config options is inconsistent: IRQ_KUNIT_TEST
> > >    and REFCOUNT_INTERRUPT_KUNIT_TEST.
> > >  * The naming and description of the older test become misleading. The
> > >    way they are formulated, I would expect them to test "everything
> > >    related to IRQ" but now we have a separate option (and kunit test)
> > >    dedicated to the refcount stuff.
> > > 
> > 
> > These are valid points. However, I think the "inconsistency" comes from
> > refcount_interrupt_test.c being in kernel/irq/. IRQ management (code in
> > kernel/irq/) and IRQ disabling/enabling are very different things to me
> 
> I agree, they are different indeed, which is a good reason to keep the
> tests separated as you pointed out.
> 
> > (for example include/linux/irqflags.h is not listed in the "IRQ
> > SUBSYSTEM" entry).
> 
> It's not, and neither are a bunch of other IRQ related header files in
> include/linux/ (get_maintainer.pl returns just the generic linux-kernel
> mailing list). To me, that looks more like an omission, because it
> means they have no maintainer.
> 
> > IMO, refcount_interrupt_test.c should be moved into
> > lib/tests/.
> 
> I'm not sure about that. I looked, and and the tests in lib/tests/ are
> for the stuff that's implemented in lib/*.c. On the other hand, most of

Make sense.

> the IRQ refcount code is in include/linux/interrupt_rc.h, with some
> bits in kernel/softirq.c (e.g. the per-cpu state variable). If the IRQ
> refcount code hadn't consisted of mostly static inline functions, where
> would that code have been added? Still kernel/softirq.c? I'm thinking

Yeah, I think that's the place (for example,
local_interrupt_disable_state is defined here).

> the corresponding unit tests belong in the same place, whatever that
> is. Assuming kernel/softirq.c, I would have suggested to put the unit
> tests in kernel/tests/, but it doesn't exist :)

Right, I need to check whether we have other unit tests for kernel/ and
maybe start a discussion on where to put them for now and future.

> 
> > > If they must be separated/grouped by scope, I would find it much nicer
> > > and less confusing if they were called something like:
> > > IRQ_MANAGEMENT_KUNIT_TEST and
> > > IRQ_REFCOUNT_KUNIT_TEST,
> > > and also the corresponding .c files were somewhat consistent with the
> > > Kconfig option names. The description of the older test would have to
> > > be updated to make it clear that it's not just "everything IRQ".
> > > 
> > > In that scenario, new kunit test(s) related to different aspects of the
> > > IRQ subsystem would get their own Kconfig option and file, and would
> > > hopefully follow the same naming pattern.
> > > 
> > > To be clear, I'm not arguing that keeping the tests separate is wrong.
> > > I'm arguing that if we do that, we should also make some adjustments to
> > > keep everything clean and tidy.
> > > 
> > 
> > Understood, and thank you for bringing it up. 
> 
> No problem. And thank you for not taking it the wrong way :)
> 

Of course! Appreciate your input!

> > So how about we keep Kuan-Wei's fix as it is and move the
> > refcount_interrupt_test.c file into lib/tests/? It's more clean and tidy
> > to me this way. Thoughts?
> 
> I'm fine with taking Kuan-Wei's patch as it is. It fixes a real
> problem, and the test code is already there (in kernel/irq/), it's not
> like the patch is putting it there. Meanwhile, we can keep discussing
> to find a better home for the test code and move it in a future patch.
> I'll send a r-b tag separately.
> 

Thank you!

Regards,
Boqun

> > > > But yes, I agree Kuan-Wei's fix is needed.
> > > 
> > > Yes, I agree, those unit tests must not be gated by KUNIT alone, so
> > > something must be done to fix it.
> > > 
> > > > > Ironically, both the issue that this patch is trying to fix and my
> > > > > question were raised by sashiko here:
> > > > > https://lore.kernel.org/all/20260605055315.3E96E1F00893@smtp.kernel.org/
> > > > > 
> > > > > > diff --git a/kernel/irq/Kconfig b/kernel/irq/Kconfig
> > > > > > index 05cba4e16dad..6923f37eaab4 100644
> > > > > > --- a/kernel/irq/Kconfig
> > > > > > +++ b/kernel/irq/Kconfig
> > > > > > @@ -150,6 +150,18 @@ config IRQ_KUNIT_TEST
> > > > > >  
> > > > > >     If unsure, say N.
> > > > > >  
> > > > > > +config REFCOUNT_INTERRUPT_KUNIT_TEST
> > > > > > + tristate "Test refcounted interrupt enable/disable" if !KUNIT_ALL_TESTS
> > > > > > + depends on KUNIT
> > > > > > + default KUNIT_ALL_TESTS
> > > > > > + help
> > > > > > +   This builds the kunit tests for the refcounted interrupt
> > > > > > +   infrastructure. It verifies the correctness of single, nested,
> > > > > > +   and multiple interrupt enable/disable state changes and ensures
> > > > > > +   that the underlying reference counting mechanisms work as expected.
> > > > > > +
> > > > > > +   If unsure, say N.
> > > > > > +
> > > > > >  endmenu
> > > > > >  
> > > > > >  config GENERIC_IRQ_MULTI_HANDLER
> > > > > > diff --git a/kernel/irq/Makefile b/kernel/irq/Makefile
> > > > > > index 44c4d6fc502a..0e5df962a149 100644
> > > > > > --- a/kernel/irq/Makefile
> > > > > > +++ b/kernel/irq/Makefile
> > > > > > @@ -16,4 +16,4 @@ obj-$(CONFIG_SMP) += affinity.o
> > > > > >  obj-$(CONFIG_GENERIC_IRQ_DEBUGFS) += debugfs.o
> > > > > >  obj-$(CONFIG_GENERIC_IRQ_MATRIX_ALLOCATOR) += matrix.o
> > > > > >  obj-$(CONFIG_IRQ_KUNIT_TEST) += irq_test.o
> > > > > > -obj-$(CONFIG_KUNIT) += refcount_interrupt_test.o
> > > > > > +obj-$(CONFIG_REFCOUNT_INTERRUPT_KUNIT_TEST) += refcount_interrupt_test.o
> 
> > 
> -- 
> Best regards,
> Radu