[PATCH] genirq/test: Resolve irq lock inversion warnings

[Brian Norris <briannorris@chromium.org>]

irq_shutdown_and_deactivate() is normally called with the descriptor
lock held, and interrupts disabled. Nested a few levels down, it grabs
the global irq_resend_lock. Lockdep rightfully complains [1].

Grab the descriptor lock, and disable interrupts, to resolve the
complaint.

Tested with:

  tools/testing/kunit/kunit.py run 'irq_test_cases*' \
      --arch x86_64 --qemu_args '-smp 2' \
      --kconfig_add CONFIG_DEBUG_KERNEL=y \
      --kconfig_add CONFIG_PROVE_LOCKING=y \
      --raw_output=all

[1]
========================================================
WARNING: possible irq lock inversion dependency detected
6.16.0-11743-g6bcdbd62bd56 #2 Tainted: G                 N
--------------------------------------------------------
kunit_try_catch/40 just changed the state of lock:
ffffffff898b1538 (irq_resend_lock){+...}-{2:2}, at: clear_irq_resend+0x14/0x70
but this lock was taken by another, HARDIRQ-safe lock in the past:
 (&irq_desc_lock_class){-.-.}-{2:2}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
 Possible interrupt unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(irq_resend_lock);
                               local_irq_disable();
                               lock(&irq_desc_lock_class);
                               lock(irq_resend_lock);
  <Interrupt>
    lock(&irq_desc_lock_class);

[...]

 ... key      at: [<ffffffff898b1538>] irq_resend_lock+0x18/0x60
 ... acquired at:
   __lock_acquire+0x82b/0x2620
   lock_acquire+0xc7/0x2c0
   _raw_spin_lock+0x2b/0x40
   clear_irq_resend+0x14/0x70
   irq_shutdown_and_deactivate+0x29/0x80
   irq_shutdown_depth_test+0x1ce/0x600
   kunit_try_run_case+0x90/0x120
   kunit_generic_run_threadfn_adapter+0x1c/0x40
   kthread+0xf3/0x200
   ret_from_fork+0x140/0x1b0
   ret_from_fork_asm+0x1a/0x30

[    5.766715]     ok 2 irq_free_disabled_test
[    5.769030]
[    5.769106] ========================================================
[    5.769159] WARNING: possible irq lock inversion dependency detected
[    5.769355] 6.16.0-11743-g6bcdbd62bd56 #1 Tainted: G                 N
[    5.769413] --------------------------------------------------------
[    5.769465] kunit_try_catch/122 just changed the state of lock:
[    5.769532] ffffffffb81ace18 (irq_resend_lock){+...}-{2:2}, at: clear_irq_resend+0x14/0x70
[    5.769899] but this lock was taken by another, HARDIRQ-safe lock in the past:
[    5.769967]  (&irq_desc_lock_class){-.-.}-{2:2}
[    5.769989]
[    5.769989]
[    5.769989] and interrupts could create inverse lock ordering between them.
...
[    5.776956]  ret_from_fork_asm+0x1a/0x30
[    5.776983]  </TASK>
[    5.778916]     # irq_shutdown_depth_test: pass:1 fail:0 skip:0 total:1
[    5.778953]     ok 3 irq_shutdown_depth_test

Fixes: 66067c3c8a1e ("genirq: Add kunit tests for depth counts")
Reported-by: Guenter Roeck <linux@roeck-us.net>
Closes: https://lore.kernel.org/lkml/31a761e4-8f81-40cf-aaf5-d220ba11911c@roeck-us.net/
Signed-off-by: Brian Norris <briannorris@chromium.org>
---

On Tue, Aug 05, 2025 at 10:45:33AM -0700, Guenter Roeck wrote:
> Hi Brian,
> 
> On Thu, May 22, 2025 at 02:08:01PM -0700, Brian Norris wrote:
> > * irq_shutdown_depth_test: exercises similar behavior from
> >   irq_cpuhotplug_test, but directly using irq_*() APIs instead of going
> >   through CPU hotplug. This still requires CONFIG_SMP, because
> >   managed-affinity is stubbed out (and not all APIs are even present)
> >   without it.
> > 
> This test triggers warning tracebacks for me.
[...]
> Is this on purpose ?

No. I think it's an artifact of trying to imitate CPU hotplug, but doing
so insufficiently. I believe the surrounding patch fixes things.

Let me know if it helps. Thanks for the report.

 kernel/irq/irq_test.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/kernel/irq/irq_test.c b/kernel/irq/irq_test.c
index 5161b56a12f9..a75abebed7f2 100644
--- a/kernel/irq/irq_test.c
+++ b/kernel/irq/irq_test.c
@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: LGPL-2.1+
 
+#include <linux/cleanup.h>
 #include <linux/cpu.h>
 #include <linux/cpumask.h>
 #include <linux/interrupt.h>
@@ -134,7 +135,8 @@ static void irq_shutdown_depth_test(struct kunit *test)
 	disable_irq(virq);
 	KUNIT_EXPECT_EQ(test, desc->depth, 1);
 
-	irq_shutdown_and_deactivate(desc);
+	scoped_guard(raw_spinlock_irqsave, &desc->lock)
+		irq_shutdown_and_deactivate(desc);
 
 	KUNIT_EXPECT_FALSE(test, irqd_is_activated(data));
 	KUNIT_EXPECT_FALSE(test, irqd_is_started(data));
-- 
2.50.1.565.gc32cd1483b-goog