Increase default mbuf size and code simplification/cleanup

mbuf size was setup to achieve the best performance i.e.
using the smallest mbuf and not segmenting packets.
However this resulted in complex code, much dependent of the way
the pmd are working e.g. a change(fix) in recent dpdk i40e
implementation caused a 1782 (=1518+8+256) bytes mbuf to be too
small to hold a 1518 bytes packets.
Hence this change simplifies the mbuf size selection at the price
of a potential decreases in performance - as more memory is now used.
Except if jumbo frames are used, the mbuf size will now be the same
for all modes. The packets will not be segmented except if jumbo
frames are enabled.
If jumbo frames are enabled, packets are by default segmented, except
if the mbuf size is configured big enough in the config file.

Change-Id: I222fcac7a65c0d221d5d422f419deb9c0f864172
Signed-off-by: Xavier Simonart <xavier.simonart@intel.com>
Signed-off-by: Deepak S <deepak.s@linux.intel.com>

1 files changed, 18 insertions, 1 deletions

diff --git a/VNFs/DPPD-PROX/defaults.h b/VNFs/DPPD-PROX/defaults.h
index 4cf3cf7a..8f850d03 100644
--- a/VNFs/DPPD-PROX/defaults.h
+++ b/VNFs/DPPD-PROX/defaults.h
@@ -41,7 +41,24 @@ void set_port_defaults(void);
 #define MAX_RSS_QUEUE_BITS      9
 
 #define PROX_VLAN_TAG_SIZE	4
-#define MBUF_SIZE (ETHER_MAX_LEN + (unsigned)sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM +  2 * PROX_VLAN_TAG_SIZE)
+
+/* MBUF_SIZE can be configured based on the following:
+   - If only one segment is used ETH_TXQ_FLAGS_NOMULTSEGS can be used resulting
+     in vector mode used for transmission hence higher performance
+   - Only one segment is used by the rx function if the mbuf size is big enough
+   - Bigger mbufs result in more memory used, hence slighly lower performance (DTLB misses)
+   - Selecting the smaller mbuf is not obvious as pmds might behave slighly differently:
+     - on ixgbe a 1526 + 256 mbuf size will cause any packets bigger than 1024 bytes to be segmented
+     - on i40e a 1526 + 256 mbuf size will cause any packets bigger than 1408 bytes to be segmented
+     - other pmds might have additional requirements
+   As the performance decrease due to the usage of bigger mbuf is not very important, we prefer
+   here to use  the same, bigger, mbuf size for all pmds, making the code easier to support.
+   An mbuf size of 2048 + 128 + 128 + 8 can hold a 2048 packet, and only one segment will be used
+   except if jumbo frames are enabled. +8 (VLAN) is needed for i40e (and maybe other pmds).
+   TX_MBUF_SIZE is used for when transmitting only: in this case the mbuf size can be smaller.
+*/
+#define MBUF_SIZE (2048 + (unsigned)sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM + 2 * PROX_VLAN_TAG_SIZE)
+#define TX_MBUF_SIZE (ETHER_MAX_LEN + (unsigned)sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM +  2 * PROX_VLAN_TAG_SIZE)
 
 #define PROX_MTU   ETHER_MAX_LEN - ETHER_HDR_LEN - ETHER_CRC_LEN