Merge branch 'xdp_xmit-bulking'

Jesper Dangaard Brouer says:

====================
This patchset change ndo_xdp_xmit API to take a bulk of xdp frames.

When kernel is compiled with CONFIG_RETPOLINE, every indirect function
pointer (branch) call hurts performance. For XDP this have a huge
negative performance impact.

This patchset reduce the needed (indirect) calls to ndo_xdp_xmit, but
also prepares for further optimizations.  The DMA APIs use of indirect
function pointer calls is the primary source the regression.  It is
left for a followup patchset, to use bulking calls towards the DMA API
(via the scatter-gatter calls).

The other advantage of this API change is that drivers can easier
amortize the cost of any sync/locking scheme, over the bulk of
packets.  The assumption of the current API is that the driver
implemementing the NDO will also allocate a dedicated XDP TX queue for
every CPU in the system.  Which is not always possible or practical to
configure. E.g. ixgbe cannot load an XDP program on a machine with
more than 96 CPUs, due to limited hardware TX queues.  E.g. virtio_net
is hard to configure as it requires manually increasing the
queues. E.g. tun driver chooses to use a per XDP frame producer lock
modulo smp_processor_id over avail queues.

I'm considered adding 'flags' to ndo_xdp_xmit, but it's not part of
this patchset.  This will be a followup patchset, once we know if this
will be needed (e.g. for non-map xdp_redirect flush-flag, and if
AF_XDP chooses to use ndo_xdp_xmit for TX).

---
V5: Fixed up issues spotted by Daniel and John

V4: Splitout the patches from 4 to 8 patches.  I cannot split the
driver changes from the NDO change, but I've tried to isolated the NDO
change together with the driver change as much as possible.
====================

Signed-off-by: Alexei Starovoitov <ast@kernel.org>

1 files changed, 24 insertions, 13 deletions

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 44d4f3d25350..d3dcfcb1c4b3 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -70,6 +70,7 @@
 #include <net/netns/generic.h>
 #include <net/rtnetlink.h>
 #include <net/sock.h>
+#include <net/xdp.h>
 #include <linux/seq_file.h>
 #include <linux/uio.h>
 #include <linux/skb_array.h>
@@ -1290,34 +1291,44 @@ static const struct net_device_ops tun_netdev_ops = {
 	.ndo_get_stats64	= tun_net_get_stats64,
 };
 
-static int tun_xdp_xmit(struct net_device *dev, struct xdp_frame *frame)
+static int tun_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **frames)
 {
 	struct tun_struct *tun = netdev_priv(dev);
 	struct tun_file *tfile;
 	u32 numqueues;
-	int ret = 0;
+	int drops = 0;
+	int cnt = n;
+	int i;
 
 	rcu_read_lock();
 
 	numqueues = READ_ONCE(tun->numqueues);
 	if (!numqueues) {
-		ret = -ENOSPC;
-		goto out;
+		rcu_read_unlock();
+		return -ENXIO; /* Caller will free/return all frames */
 	}
 
 	tfile = rcu_dereference(tun->tfiles[smp_processor_id() %
 					    numqueues]);
-	/* Encode the XDP flag into lowest bit for consumer to differ
-	 * XDP buffer from sk_buff.
-	 */
-	if (ptr_ring_produce(&tfile->tx_ring, tun_xdp_to_ptr(frame))) {
-		this_cpu_inc(tun->pcpu_stats->tx_dropped);
-		ret = -ENOSPC;
+
+	spin_lock(&tfile->tx_ring.producer_lock);
+	for (i = 0; i < n; i++) {
+		struct xdp_frame *xdp = frames[i];
+		/* Encode the XDP flag into lowest bit for consumer to differ
+		 * XDP buffer from sk_buff.
+		 */
+		void *frame = tun_xdp_to_ptr(xdp);
+
+		if (__ptr_ring_produce(&tfile->tx_ring, frame)) {
+			this_cpu_inc(tun->pcpu_stats->tx_dropped);
+			xdp_return_frame_rx_napi(xdp);
+			drops++;
+		}
 	}
+	spin_unlock(&tfile->tx_ring.producer_lock);
 
-out:
 	rcu_read_unlock();
-	return ret;
+	return cnt - drops;
 }
 
 static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
@@ -1327,7 +1338,7 @@ static int tun_xdp_tx(struct net_device *dev, struct xdp_buff *xdp)
 	if (unlikely(!frame))
 		return -EOVERFLOW;
 
-	return tun_xdp_xmit(dev, frame);
+	return tun_xdp_xmit(dev, 1, &frame);
 }
 
 static void tun_xdp_flush(struct net_device *dev)