: Remove unused fields in packet history structure

[CCID 3]: Remove unused fields in packet history structure

This removes two fields of the TX packet history structure which are not referenced by the CCID 3 code and are not needed:

* dccphtx_sent - is redundant since the fact that an entry is present in the TX history in itself is an indication that the packet has been sent (cf. dccp_write_xmit and ccid3_hc_tx_packet_sent);

* dccphtx_rtt - is nowhere referenced and is not even required:
the `Preventing Oscillations' mechanism in [RFC 3448, 4.5] uses a moving-average, but does not require to memorize past RTTs; the history field has no further use;

As a further benefit, the history entry size is reduced.

NB : Adding the newly created entry is now at the end of packet_sent - if it is
in the history before being fully filled in, list corruption is possible.

---
 net/dccp/ccids/ccid3.c              |    4 +---
 net/dccp/ccids/lib/packet_history.h |   12 ++----------
 2 files changed, 3 insertions(+), 13 deletions(-)

@@ -408,13 +408,11 @@ static void ccid3_hc_tx_packet_sent(stru
 		DCCP_CRIT("packet history - out of memory!");
 		return;
 	}
-	dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, packet);
 
 	do_gettimeofday(&now);
 	packet->dccphtx_tstamp = now;
 	packet->dccphtx_seqno  = dccp_sk(sk)->dccps_gss;
-	packet->dccphtx_rtt    = hctx->ccid3hctx_rtt;
-	packet->dccphtx_sent   = 1;
+	dccp_tx_hist_add_entry(&hctx->ccid3hctx_hist, packet);
 }
 
 static void ccid3_hc_tx_packet_recv(struct sock *sk, struct sk_buff *skb)

@@ -54,9 +54,7 @@
  */
 struct dccp_tx_hist_entry {
 	struct list_head dccphtx_node;
-	u64		 dccphtx_seqno:48,
-			 dccphtx_sent:1;
-	u32		 dccphtx_rtt;
+	u64		 dccphtx_seqno;
 	struct timeval	 dccphtx_tstamp;
 };
 

@@ -71,13 +69,7 @@ static inline struct dccp_tx_hist_entry 
 			dccp_tx_hist_entry_new(struct dccp_tx_hist *hist,
 					       const gfp_t prio)
 {
-	struct dccp_tx_hist_entry *entry = kmem_cache_alloc(hist->dccptxh_slab,
-							    prio);
-
-	if (entry != NULL)
-		entry->dccphtx_sent = 0;
-
-	return entry;
+	return kmem_cache_alloc(hist->dccptxh_slab, prio);
 }
 
 extern int dccp_tx_hist_get_send_time(struct dccp_tx_hist *, struct list_head *,

Indeed it reduced the size:

[acme@mica net-2.6.22]$ cat /tmp/before
/* <eb7c> /home/acme/git/net-2.6.22/net/dccp/ccids/lib/packet_history.h:55 */ struct dccp_tx_hist_entry {

        struct list_head           dccphtx_node;         /*     0    16 */
        u64                        dccphtx_seqno:48;     /*    16     8 */
        u64                        dccphtx_sent:1;       /*    16     8 */

/* XXX 15 bits hole, try to pack */

u32 dccphtx_rtt; /* 24 4 */

/* XXX 4 bytes hole, try to pack */

struct timeval dccphtx_tstamp; /* 32 16 */
}; /* size: 48, cachelines: 1 */
/* sum members: 44, holes: 1, sum holes: 4 */
/* bit holes: 1, sum bit holes: 15 bits */
/* last cacheline: 48 bytes */
[acme@mica net-2.6.22]$ pahole -e ../build/mica/net-2.6.22/net/dccp/ccids/ccid3.o dccp_tx_hist_entry /* <eb7c> /home/acme/git/net-2.6.22/net/dccp/ccids/lib/packet_history.h:55 */
struct dccp_tx_hist_entry {

        struct list_head           dccphtx_node;         /*     0    16 */
        u64                        dccphtx_seqno;        /*    16     8 */
        struct timeval             dccphtx_tstamp;       /*    24    16 */

}; /* size: 40, cachelines: 1 */
/* last cacheline: 40 bytes */
[acme@mica net-2.6.22]$

I wonder if we could make this a single linked list, lemme see.. I think so, and by doing that we get the struct down to 32 bytes, making it cacheline friendly, and also simplify the operations, i.e. that opencoded list operation would just go away, will leave this for later, in a TODO file.

- Arnaldo

Quoting Arnaldo Carvalho de Melo:

|  struct dccp_tx_hist_entry {
|          struct list_head           dccphtx_node;         /*     0    16 */
|          u64                        dccphtx_seqno;        /*    16     8 */
|          struct timeval             dccphtx_tstamp;       /*    24    16 */
|  }; /* size: 40, cachelines: 1 */
|     /* last cacheline: 40 bytes */
|  [acme@mica net-2.6.22]$
|  
|  I wonder if we could make this a single linked list, lemme see.. I
|  think so, and by doing that we get the struct down to 32 bytes, making
|  it cacheline friendly, and also simplify the operations, i.e. that
|  opencoded list operation would just go away, will leave this for
|  later, in a TODO file.

This is an excellent idea - I have just checked. It works, there is no sorting and just LIFO order.

Will you be working on this or would you like me to update this in terms of struct hlist_head?