mbuf: optimize refcnt update

[dpdk.git] / lib / librte_mbuf / rte_mbuf.h

diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
index 70b0987..6c9cfd6 100644 (file)
--- a/lib/librte_mbuf/rte_mbuf.h
+++ b/lib/librte_mbuf/rte_mbuf.h
@@ -217,6 +217,16 @@ const char *rte_get_rx_ol_flag_name(uint64_t mask);
  */
 const char *rte_get_tx_ol_flag_name(uint64_t mask);
 
+/**
+ * Some NICs need at least 2KB buffer to RX standard Ethernet frame without
+ * splitting it into multiple segments.
+ * So, for mbufs that planned to be involved into RX/TX, the recommended
+ * minimal buffer length is 2KB + RTE_PKTMBUF_HEADROOM.
+ */
+#define        RTE_MBUF_DEFAULT_DATAROOM       2048
+#define        RTE_MBUF_DEFAULT_BUF_SIZE       \
+       (RTE_MBUF_DEFAULT_DATAROOM + RTE_PKTMBUF_HEADROOM)
+
 /* define a set of marker types that can be used to refer to set points in the
  * mbuf */
 typedef void    *MARKER[0];   /**< generic marker for a point in a structure */
@@ -415,21 +425,6 @@ if (!(exp)) {                                                        \
 
 #ifdef RTE_MBUF_REFCNT_ATOMIC
 
-/**
- * Adds given value to an mbuf's refcnt and returns its new value.
- * @param m
- *   Mbuf to update
- * @param value
- *   Value to add/subtract
- * @return
- *   Updated value
- */
-static inline uint16_t
-rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
-{
-       return (uint16_t)(rte_atomic16_add_return(&m->refcnt_atomic, value));
-}
-
 /**
  * Reads the value of an mbuf's refcnt.
  * @param m
@@ -456,6 +451,33 @@ rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
        rte_atomic16_set(&m->refcnt_atomic, new_value);
 }
 
+/**
+ * Adds given value to an mbuf's refcnt and returns its new value.
+ * @param m
+ *   Mbuf to update
+ * @param value
+ *   Value to add/subtract
+ * @return
+ *   Updated value
+ */
+static inline uint16_t
+rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
+{
+       /*
+        * The atomic_add is an expensive operation, so we don't want to
+        * call it in the case where we know we are the uniq holder of
+        * this mbuf (i.e. ref_cnt == 1). Otherwise, an atomic
+        * operation has to be used because concurrent accesses on the
+        * reference counter can occur.
+        */
+       if (likely(rte_mbuf_refcnt_read(m) == 1)) {
+               rte_mbuf_refcnt_set(m, 1 + value);
+               return 1 + value;
+       }
+
+       return (uint16_t)(rte_atomic16_add_return(&m->refcnt_atomic, value));
+}
+
 #else /* ! RTE_MBUF_REFCNT_ATOMIC */
 
 /**
@@ -885,20 +907,7 @@ __rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
 {
        __rte_mbuf_sanity_check(m, 0);
 
-       /*
-        * Check to see if this is the last reference to the mbuf.
-        * Note: the double check here is deliberate. If the ref_cnt is "atomic"
-        * the call to "refcnt_update" is a very expensive operation, so we
-        * don't want to call it in the case where we know we are the holder
-        * of the last reference to this mbuf i.e. ref_cnt == 1.
-        * If however, ref_cnt != 1, it's still possible that we may still be
-        * the final decrementer of the count, so we need to check that
-        * result also, to make sure the mbuf is freed properly.
-        */
-       if (likely (rte_mbuf_refcnt_read(m) == 1) ||
-                       likely (rte_mbuf_refcnt_update(m, -1) == 0)) {
-
-               rte_mbuf_refcnt_set(m, 0);
+       if (likely(rte_mbuf_refcnt_update(m, -1) == 0)) {
 
                /* if this is an indirect mbuf, then
                 *  - detach mbuf