doc: RCU callback locks need only _bh, not necessarily _irq
authorJiunn Chang <c0d1n61at3@gmail.com>
Thu, 27 Jun 2019 21:01:47 +0000 (16:01 -0500)
committerJonathan Corbet <corbet@lwn.net>
Fri, 28 Jun 2019 15:10:40 +0000 (09:10 -0600)
The UP.rst file calls for locks acquired within RCU callback functions
to use _irq variants (spin_lock_irqsave() or similar), which does work,
but can be overkill.  This commit therefore instead calls for _bh variants
(spin_lock_bh() or similar), while noting that _irq does work.

Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Signed-off-by: Jiunn Chang <c0d1n61at3@gmail.com>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Documentation/RCU/UP.rst

index 67715a4..e26dda2 100644 (file)
@@ -113,12 +113,13 @@ Answer to Quick Quiz #1:
 Answer to Quick Quiz #2:
        What locking restriction must RCU callbacks respect?
 
-       Any lock that is acquired within an RCU callback must be
-       acquired elsewhere using an _irq variant of the spinlock
-       primitive.  For example, if "mylock" is acquired by an
-       RCU callback, then a process-context acquisition of this
-       lock must use something like spin_lock_irqsave() to
-       acquire the lock.
+       Any lock that is acquired within an RCU callback must be acquired
+       elsewhere using an _bh variant of the spinlock primitive.
+       For example, if "mylock" is acquired by an RCU callback, then
+       a process-context acquisition of this lock must use something
+       like spin_lock_bh() to acquire the lock.  Please note that
+       it is also OK to use _irq variants of spinlocks, for example,
+       spin_lock_irqsave().
 
        If the process-context code were to simply use spin_lock(),
        then, since RCU callbacks can be invoked from softirq context,