percpu_ref: add PCPU_REF_DEAD

percpu_ref will be restructured so that percpu/atomic mode switching
and reference killing are dedoupled.  In preparation, add
PCPU_REF_DEAD and PCPU_REF_ATOMIC_DEAD which is OR of ATOMIC and DEAD.
For now, ATOMIC and DEAD are changed together and all PCPU_REF_ATOMIC
uses are converted to PCPU_REF_ATOMIC_DEAD without causing any
behavior changes.

percpu_ref_init() now specifies an explicit alignment when allocating
the percpu counters so that the pointer has enough unused low bits to
accomodate the flags.  Note that one flag was fine as min alignment
for percpu memory is 2 bytes but two flags are already too many for
the natural alignment of unsigned longs on archs like cris and m68k.

v2: The original patch had BUILD_BUG_ON() which triggers if unsigned
    long's alignment isn't enough to accomodate the flags, which
    triggered on cris and m64k.  percpu_ref_init() updated to specify
    the required alignment explicitly.  Reported by Fengguang.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Kent Overstreet <kmo@daterainc.com>
Cc: kbuild test robot <fengguang.wu@intel.com>

1 files changed, 11 insertions, 8 deletions

diff --git a/lib/percpu-refcount.c b/lib/percpu-refcount.c
index 7aef590c1ef..e2ff19f970c 100644
--- a/lib/percpu-refcount.c
+++ b/lib/percpu-refcount.c
@@ -34,7 +34,7 @@
 static unsigned long __percpu *percpu_count_ptr(struct percpu_ref *ref)
 {
 	return (unsigned long __percpu *)
-		(ref->percpu_count_ptr & ~__PERCPU_REF_ATOMIC);
+		(ref->percpu_count_ptr & ~__PERCPU_REF_ATOMIC_DEAD);
 }
 
 /**
@@ -52,10 +52,13 @@ static unsigned long __percpu *percpu_count_ptr(struct percpu_ref *ref)
 int percpu_ref_init(struct percpu_ref *ref, percpu_ref_func_t *release,
 		    gfp_t gfp)
 {
+	size_t align = max_t(size_t, 1 << __PERCPU_REF_FLAG_BITS,
+			     __alignof__(unsigned long));
+
 	atomic_long_set(&ref->count, 1 + PERCPU_COUNT_BIAS);
 
-	ref->percpu_count_ptr =
-		(unsigned long)alloc_percpu_gfp(unsigned long, gfp);
+	ref->percpu_count_ptr = (unsigned long)
+		__alloc_percpu_gfp(sizeof(unsigned long), align, gfp);
 	if (!ref->percpu_count_ptr)
 		return -ENOMEM;
 
@@ -80,7 +83,7 @@ void percpu_ref_exit(struct percpu_ref *ref)
 
 	if (percpu_count) {
 		free_percpu(percpu_count);
-		ref->percpu_count_ptr = __PERCPU_REF_ATOMIC;
+		ref->percpu_count_ptr = __PERCPU_REF_ATOMIC_DEAD;
 	}
 }
 EXPORT_SYMBOL_GPL(percpu_ref_exit);
@@ -145,10 +148,10 @@ static void percpu_ref_kill_rcu(struct rcu_head *rcu)
 void percpu_ref_kill_and_confirm(struct percpu_ref *ref,
 				 percpu_ref_func_t *confirm_kill)
 {
-	WARN_ONCE(ref->percpu_count_ptr & __PERCPU_REF_ATOMIC,
+	WARN_ONCE(ref->percpu_count_ptr & __PERCPU_REF_ATOMIC_DEAD,
 		  "%s called more than once on %pf!", __func__, ref->release);
 
-	ref->percpu_count_ptr |= __PERCPU_REF_ATOMIC;
+	ref->percpu_count_ptr |= __PERCPU_REF_ATOMIC_DEAD;
 	ref->confirm_switch = confirm_kill;
 
 	call_rcu_sched(&ref->rcu, percpu_ref_kill_rcu);
@@ -180,12 +183,12 @@ void percpu_ref_reinit(struct percpu_ref *ref)
 	 * Restore per-cpu operation.  smp_store_release() is paired with
 	 * smp_read_barrier_depends() in __ref_is_percpu() and guarantees
 	 * that the zeroing is visible to all percpu accesses which can see
-	 * the following __PERCPU_REF_ATOMIC clearing.
+	 * the following __PERCPU_REF_ATOMIC_DEAD clearing.
 	 */
 	for_each_possible_cpu(cpu)
 		*per_cpu_ptr(percpu_count, cpu) = 0;
 
 	smp_store_release(&ref->percpu_count_ptr,
-			  ref->percpu_count_ptr & ~__PERCPU_REF_ATOMIC);
+			  ref->percpu_count_ptr & ~__PERCPU_REF_ATOMIC_DEAD);
 }
 EXPORT_SYMBOL_GPL(percpu_ref_reinit);