[PATCH] rtc-cmos: use spin_lock_irqsave in cmos_interrupt

[Mateusz Jończyk]

cmos_interrupt() can be called in a non-interrupt context, such as in
an ACPI event handler (which runs in an interrupt thread). Therefore,
usage of spin_lock(&rtc_lock) is insecure. Use spin_lock_irqsave() /
spin_unlock_irqrestore() instead.

Before a misguided
commit 6950d046eb6e ("rtc: cmos: Replace spin_lock_irqsave with spin_lock in hard IRQ")
the cmos_interrupt() function used spin_lock_irqsave(). That commit
changed it to spin_lock() and broke locking, which was partially fixed in
commit 13be2efc390a ("rtc: cmos: Disable irq around direct invocation of cmos_interrupt()")

That second commit did not take account of the ACPI fixed event handler
pathway, however. It introduced local_irq_disable() workarounds in
cmos_check_wkalrm(), which can cause problems on PREEMPT_RT kernels
and are now unnecessary.

Add an explicit comment so that this change will not be reverted by
mistake.

Cc: <stable@xxxxxxxxxxxxxxx>
Fixes: 6950d046eb6e ("rtc: cmos: Replace spin_lock_irqsave with spin_lock in hard IRQ")
Signed-off-by: Mateusz Jończyk <mat.jonczyk@xxxxx>
Reviewed-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
Tested-by: Chris Bainbridge <chris.bainbridge@xxxxxxxxx>
Reported-by: Chris Bainbridge <chris.bainbridge@xxxxxxxxx>
Closes: https://lore.kernel.org/all/aDtJ92foPUYmGheF@debian.local/

---

Changes after DRAFT version of the patch:
- rewrite commit message,
- test this locally (also on top of 5.10.238 for the stable backport),
- fix a grammar mistake in the comment.
---
 drivers/rtc/rtc-cmos.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/drivers/rtc/rtc-cmos.c b/drivers/rtc/rtc-cmos.c
index 8172869bd3d7..0743c6acd6e2 100644
--- a/drivers/rtc/rtc-cmos.c
+++ b/drivers/rtc/rtc-cmos.c
@@ -692,8 +692,12 @@ static irqreturn_t cmos_interrupt(int irq, void *p)
 {
 	u8		irqstat;
 	u8		rtc_control;
+	unsigned long	flags;
 
-	spin_lock(&rtc_lock);
+	/* We cannot use spin_lock() here, as cmos_interrupt() is also called
+	 * in a non-irq context.
+	 */
+	spin_lock_irqsave(&rtc_lock, flags);
 
 	/* When the HPET interrupt handler calls us, the interrupt
 	 * status is passed as arg1 instead of the irq number.  But
@@ -727,7 +731,7 @@ static irqreturn_t cmos_interrupt(int irq, void *p)
 			hpet_mask_rtc_irq_bit(RTC_AIE);
 		CMOS_READ(RTC_INTR_FLAGS);
 	}
-	spin_unlock(&rtc_lock);
+	spin_unlock_irqrestore(&rtc_lock, flags);
 
 	if (is_intr(irqstat)) {
 		rtc_update_irq(p, 1, irqstat);
@@ -1295,9 +1299,7 @@ static void cmos_check_wkalrm(struct device *dev)
 	 * ACK the rtc irq here
 	 */
 	if (t_now >= cmos->alarm_expires && cmos_use_acpi_alarm()) {
-		local_irq_disable();
 		cmos_interrupt(0, (void *)cmos->rtc);
-		local_irq_enable();
 		return;
 	}
 
-- 
2.25.1