A bunch of netmap fixes:

USERSPACE:
1. add support for devices with different number of rx and tx queues;

2. add better support for zero-copy operation, adding an extra field
   to the netmap ring to indicate how many buffers we have already processed
   but not yet released (with help from Eddie Kohler);

3. The two changes above unfortunately require an API change, so while
   at it add a version field and some spares to the ioctl() argument
   to help detect mismatches.

4. update the manual page for the two changes above;

5. update sample applications in tools/tools/netmap

KERNEL:

1. simplify the internal structures moving the global wait queues
   to the 'struct netmap_adapter';

2. simplify the functions that map kring<->nic ring indexes

3. normalize device-specific code, helps mainteinance;

4. start exploring the impact of micro-optimizations (prefetch etc.)
   in the ixgbe driver.
   Use 'legacy' descriptors on the tx ring and prefetch slots gives
   about 20% speedup at 900 MHz. Another 7-10% would come from removing
   the explict calls to bus_dmamap* in the core (they are effectively
   NOPs in this case, but it takes expensive load of the per-buffer
   dma maps to figure out that they are all NULL.

   Rx performance not investigated.

I am postponing the MFC so i can import a few more improvements
before merging.

3 files changed, 24 insertions, 13 deletions

diff --git a/tools/tools/netmap/bridge.c b/tools/tools/netmap/bridge.c
index 2b9122c..1b027d7 100644
--- a/tools/tools/netmap/bridge.c
+++ b/tools/tools/netmap/bridge.c
@@ -48,7 +48,7 @@ int verbose = 0;
 	} while (0)
 
 
-char *version = "$Id: bridge.c 9642 2011-11-07 21:39:47Z luigi $";
+char *version = "$Id: bridge.c 10637 2012-02-24 16:36:25Z luigi $";
 
 static int do_abort = 0;
 
@@ -136,6 +136,7 @@ netmap_open(struct my_ring *me, int ringid)
 	bzero(&req, sizeof(req));
 	strncpy(req.nr_name, me->ifname, sizeof(req.nr_name));
 	req.nr_ringid = ringid;
+	req.nr_version = NETMAP_API;
 	err = ioctl(fd, NIOCGINFO, &req);
 	if (err) {
 		D("cannot get info on %s", me->ifname);
@@ -162,17 +163,22 @@ netmap_open(struct my_ring *me, int ringid)
 	me->nifp = NETMAP_IF(me->mem, req.nr_offset);
 	me->queueid = ringid;
 	if (ringid & NETMAP_SW_RING) {
-		me->begin = req.nr_numrings;
+		me->begin = req.nr_rx_rings;
 		me->end = me->begin + 1;
+		me->tx = NETMAP_TXRING(me->nifp, req.nr_tx_rings);
+		me->rx = NETMAP_RXRING(me->nifp, req.nr_rx_rings);
 	} else if (ringid & NETMAP_HW_RING) {
+		D("XXX check multiple threads");
 		me->begin = ringid & NETMAP_RING_MASK;
 		me->end = me->begin + 1;
+		me->tx = NETMAP_TXRING(me->nifp, me->begin);
+		me->rx = NETMAP_RXRING(me->nifp, me->begin);
 	} else {
 		me->begin = 0;
-		me->end = req.nr_numrings;
+		me->end = req.nr_rx_rings; // XXX max of the two
+		me->tx = NETMAP_TXRING(me->nifp, 0);
+		me->rx = NETMAP_RXRING(me->nifp, 0);
 	}
-	me->tx = NETMAP_TXRING(me->nifp, me->begin);
-	me->rx = NETMAP_RXRING(me->nifp, me->begin);
 	return (0);
 error:
 	close(me->fd);
@@ -294,7 +300,7 @@ howmany(struct my_ring *me, int tx)
 	if (0 && verbose && tot && !tx)
 		D("ring %s %s %s has %d avail at %d",
 			me->ifname, tx ? "tx": "rx",
-			me->end > me->nifp->ni_num_queues ?
+			me->end > me->nifp->ni_rx_queues ?
 				"host":"net",
 			tot, NETMAP_TXRING(me->nifp, me->begin)->cur);
 	return tot;
@@ -392,8 +398,8 @@ main(int argc, char **argv)
 	D("Wait 2 secs for link to come up...");
 	sleep(2);
 	D("Ready to go, %s 0x%x/%d <-> %s 0x%x/%d.",
-		me[0].ifname, me[0].queueid, me[0].nifp->ni_num_queues,
-		me[1].ifname, me[1].queueid, me[1].nifp->ni_num_queues);
+		me[0].ifname, me[0].queueid, me[0].nifp->ni_rx_queues,
+		me[1].ifname, me[1].queueid, me[1].nifp->ni_rx_queues);
 
 	/* main loop */
 	signal(SIGINT, sigint_h);
diff --git a/tools/tools/netmap/pcap.c b/tools/tools/netmap/pcap.c
index f010b83..2a93e82 100644
--- a/tools/tools/netmap/pcap.c
+++ b/tools/tools/netmap/pcap.c
@@ -257,14 +257,14 @@ netmap_open(struct my_ring *me, int ringid)
 	me->nifp = NETMAP_IF(me->mem, req.nr_offset);
 	me->queueid = ringid;
 	if (ringid & NETMAP_SW_RING) {
-		me->begin = req.nr_numrings;
+		me->begin = req.nr_rx_rings;
 		me->end = me->begin + 1;
 	} else if (ringid & NETMAP_HW_RING) {
 		me->begin = ringid & NETMAP_RING_MASK;
 		me->end = me->begin + 1;
 	} else {
 		me->begin = 0;
-		me->end = req.nr_numrings;
+		me->end = req.nr_rx_rings;
 	}
 	/* request timestamps for packets */
 	for (i = me->begin; i < me->end; i++) {
diff --git a/tools/tools/netmap/pkt-gen.c b/tools/tools/netmap/pkt-gen.c
index 5627e9e..43eeda8 100644
--- a/tools/tools/netmap/pkt-gen.c
+++ b/tools/tools/netmap/pkt-gen.c
@@ -25,7 +25,7 @@
 
 /*
  * $FreeBSD$
- * $Id: pkt-gen.c 9827 2011-12-05 11:29:34Z luigi $
+ * $Id: pkt-gen.c 10637 2012-02-24 16:36:25Z luigi $
  *
  * Example program to show how to build a multithreaded packet
  * source/sink using the netmap device.
@@ -776,6 +776,7 @@ main(int arc, char **argv)
 	}
 
 	bzero(&nmr, sizeof(nmr));
+	nmr.nr_version = NETMAP_API;
 	/*
 	 * Open the netmap device to fetch the number of queues of our
 	 * interface.
@@ -796,11 +797,12 @@ main(int arc, char **argv)
 			D("map size is %d Kb", nmr.nr_memsize >> 10);
 		}
 		bzero(&nmr, sizeof(nmr));
+		nmr.nr_version = NETMAP_API;
 		strncpy(nmr.nr_name, ifname, sizeof(nmr.nr_name));
 		if ((ioctl(fd, NIOCGINFO, &nmr)) == -1) {
 			D("Unable to get if info for %s", ifname);
 		}
-		devqueues = nmr.nr_numrings;
+		devqueues = nmr.nr_rx_rings;
 	}
 
 	/* validate provided nthreads. */
@@ -841,6 +843,7 @@ main(int arc, char **argv)
 	 * We decide to put the first interface registration here to
 	 * give time to cards that take a long time to reset the PHY.
 	 */
+	nmr.nr_version = NETMAP_API;
 	if (ioctl(fd, NIOCREGIF, &nmr) == -1) {
 		D("Unable to register interface %s", ifname);
 		//continue, fail later
@@ -904,6 +907,7 @@ main(int arc, char **argv)
 
 		bzero(&tifreq, sizeof(tifreq));
 		strncpy(tifreq.nr_name, ifname, sizeof(tifreq.nr_name));
+		tifreq.nr_version = NETMAP_API;
 		tifreq.nr_ringid = (g.nthreads > 1) ? (i | NETMAP_HW_RING) : 0;
 
 		/*
@@ -930,7 +934,8 @@ main(int arc, char **argv)
 		targs[i].nmr = tifreq;
 		targs[i].nifp = tnifp;
 		targs[i].qfirst = (g.nthreads > 1) ? i : 0;
-		targs[i].qlast = (g.nthreads > 1) ? i+1 : tifreq.nr_numrings;
+		targs[i].qlast = (g.nthreads > 1) ? i+1 :
+			(td_body == receiver_body ? tifreq.nr_rx_rings : tifreq.nr_tx_rings);
 		targs[i].me = i;
 		targs[i].affinity = g.cpus ? i % g.cpus : -1;
 		if (td_body == sender_body) {