rtc: remove BKL for ioctl()

Remove implicit use of BKL in ioctl() from the RTC framework.

Instead, the rtc->ops_lock is used.  That's the same lock that already
protects the RTC operations when they're issued through the exported
rtc_*() calls in drivers/rtc/interface.c ...  making this a bugfix, not
just a cleanup, since both ioctl calls and set_alarm() need to update IRQ
enable flags and that implies a common lock (which RTC drivers as a rule
do not provide on their own).

A new comment at the declaration of "struct rtc_class_ops" summarizes
current locking rules.  It's not clear to me that the exceptions listed
there should exist ...  if not, those are pre-existing problems which can
be fixed in a patch that doesn't relate to BKL removal.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Jonathan Corbet <corbet@lwn.net>
Acked-by: Alessandro Zummo <a.zummo@towertech.it>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 17 insertions, 0 deletions

diff --git a/include/linux/rtc.h b/include/linux/rtc.h
index f2d0d1527721..b01fe004cb5e 100644
--- a/include/linux/rtc.h
+++ b/include/linux/rtc.h
@@ -115,6 +115,23 @@ extern void rtc_time_to_tm(unsigned long time, struct rtc_time *tm);
 
 extern struct class *rtc_class;
 
+/*
+ * For these RTC methods the device parameter is the physical device
+ * on whatever bus holds the hardware (I2C, Platform, SPI, etc), which
+ * was passed to rtc_device_register().  Its driver_data normally holds
+ * device state, including the rtc_device pointer for the RTC.
+ *
+ * Most of these methods are called with rtc_device.ops_lock held,
+ * through the rtc_*(struct rtc_device *, ...) calls.
+ *
+ * The (current) exceptions are mostly filesystem hooks:
+ *   - the proc() hook for procfs
+ *   - non-ioctl() chardev hooks:  open(), release(), read_callback()
+ *   - periodic irq calls:  irq_set_state(), irq_set_freq()
+ *
+ * REVISIT those periodic irq calls *do* have ops_lock when they're
+ * issued through ioctl() ...
+ */
 struct rtc_class_ops {
         int (*open)(struct device *);
         void (*release)(struct device *);