Re: [PATCH net-next 03/12] ptr_ring: READ/WRITE_ONCE for __ptr_ring_empty

On Fri, Jan 26, 2018 at 10:37:58AM +0800, Jason Wang wrote:

On 2018年01月26日 07:36, Michael S. Tsirkin wrote:

quoted

Lockless __ptr_ring_empty requires that consumer head is read and
written at once, atomically. Annotate accordingly to make sure compiler
does it correctly.  Switch locked callers to __ptr_ring_peek which does
not support the lockless operation.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
---
  include/linux/ptr_ring.h | 11 ++++++++---
  1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/include/linux/ptr_ring.h b/include/linux/ptr_ring.h
index 8594c7b..9a72d8f 100644
--- a/include/linux/ptr_ring.h
+++ b/include/linux/ptr_ring.h

@@ -196,7 +196,9 @@ static inline void *__ptr_ring_peek(struct ptr_ring *r)
   */
  static inline bool __ptr_ring_empty(struct ptr_ring *r)
  {
-	return !__ptr_ring_peek(r);
+	if (likely(r->size))
+		return !r->queue[READ_ONCE(r->consumer_head)];
+	return true;
  }

So after patch 8, __ptr_ring_peek() did:

static inline void *__ptr_ring_peek(struct ptr_ring *r)
{
    if (likely(r->size))
        return READ_ONCE(r->queue[r->consumer_head]);
    return NULL;
}

Looks like a duplication.

Thanks

Nope - they are different.

The reason is that __ptr_ring_peek does not need to read the consumer_head once
since callers have a lock, and __ptr_ring_empty does not need to read
the queue once since it merely compares it to 0.

-- 
MST