xdp: base API for new XDP rx-queue info concept

This patch only introduce the core data structures and API functions.
All XDP enabled drivers must use the API before this info can used.

There is a need for XDP to know more about the RX-queue a given XDP
frames have arrived on.  For both the XDP bpf-prog and kernel side.

Instead of extending xdp_buff each time new info is needed, the patch
creates a separate read-mostly struct xdp_rxq_info, that contains this
info.  We stress this data/cache-line is for read-only info.  This is
NOT for dynamic per packet info, use the data_meta for such use-cases.

The performance advantage is this info can be setup at RX-ring init
time, instead of updating N-members in xdp_buff.  A possible (driver
level) micro optimization is that xdp_buff->rxq assignment could be
done once per XDP/NAPI loop.  The extra pointer deref only happens for
program needing access to this info (thus, no slowdown to existing
use-cases).

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>

4 files changed, 117 insertions, 1 deletions

diff --git a/include/linux/filter.h b/include/linux/filter.h
index 2b0df2703671..425056c7f96c 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -20,6 +20,7 @@
 #include <linux/set_memory.h>
 #include <linux/kallsyms.h>
 
+#include <net/xdp.h>
 #include <net/sch_generic.h>
 
 #include <uapi/linux/filter.h>
@@ -503,6 +504,7 @@ struct xdp_buff {
 	void *data_end;
 	void *data_meta;
 	void *data_hard_start;
+	struct xdp_rxq_info *rxq;
 };
 
 /* Compute the linear packet data range [data, data_end) which
diff --git a/include/net/xdp.h b/include/net/xdp.h
new file mode 100644
index 000000000000..86c41631a908
--- /dev/null
+++ b/include/net/xdp.h
@@ -0,0 +1,47 @@
+/* include/net/xdp.h
+ *
+ * Copyright (c) 2017 Jesper Dangaard Brouer, Red Hat Inc.
+ * Released under terms in GPL version 2.  See COPYING.
+ */
+#ifndef __LINUX_NET_XDP_H__
+#define __LINUX_NET_XDP_H__
+
+/**
+ * DOC: XDP RX-queue information
+ *
+ * The XDP RX-queue info (xdp_rxq_info) is associated with the driver
+ * level RX-ring queues.  It is information that is specific to how
+ * the driver have configured a given RX-ring queue.
+ *
+ * Each xdp_buff frame received in the driver carry a (pointer)
+ * reference to this xdp_rxq_info structure.  This provides the XDP
+ * data-path read-access to RX-info for both kernel and bpf-side
+ * (limited subset).
+ *
+ * For now, direct access is only safe while running in NAPI/softirq
+ * context.  Contents is read-mostly and must not be updated during
+ * driver NAPI/softirq poll.
+ *
+ * The driver usage API is a register and unregister API.
+ *
+ * The struct is not directly tied to the XDP prog.  A new XDP prog
+ * can be attached as long as it doesn't change the underlying
+ * RX-ring.  If the RX-ring does change significantly, the NIC driver
+ * naturally need to stop the RX-ring before purging and reallocating
+ * memory.  In that process the driver MUST call unregistor (which
+ * also apply for driver shutdown and unload).  The register API is
+ * also mandatory during RX-ring setup.
+ */
+
+struct xdp_rxq_info {
+	struct net_device *dev;
+	u32 queue_index;
+	u32 reg_state;
+} ____cacheline_aligned; /* perf critical, avoid false-sharing */
+
+int xdp_rxq_info_reg(struct xdp_rxq_info *xdp_rxq,
+		     struct net_device *dev, u32 queue_index);
+void xdp_rxq_info_unreg(struct xdp_rxq_info *xdp_rxq);
+void xdp_rxq_info_unused(struct xdp_rxq_info *xdp_rxq);
+
+#endif /* __LINUX_NET_XDP_H__ */
diff --git a/net/core/Makefile b/net/core/Makefile
index 1fd0a9c88b1b..6dbbba8c57ae 100644
--- a/net/core/Makefile
+++ b/net/core/Makefile
@@ -11,7 +11,7 @@ obj-$(CONFIG_SYSCTL) += sysctl_net_core.o
 obj-y		     += dev.o ethtool.o dev_addr_lists.o dst.o netevent.o \
 			neighbour.o rtnetlink.o utils.o link_watch.o filter.o \
 			sock_diag.o dev_ioctl.o tso.o sock_reuseport.o \
-			fib_notifier.o
+			fib_notifier.o xdp.o
 
 obj-y += net-sysfs.o
 obj-$(CONFIG_PROC_FS) += net-procfs.o
diff --git a/net/core/xdp.c b/net/core/xdp.c
new file mode 100644
index 000000000000..229bc5a0ee04
--- /dev/null
+++ b/net/core/xdp.c
@@ -0,0 +1,67 @@
+/* net/core/xdp.c
+ *
+ * Copyright (c) 2017 Jesper Dangaard Brouer, Red Hat Inc.
+ * Released under terms in GPL version 2.  See COPYING.
+ */
+#include <linux/types.h>
+#include <linux/mm.h>
+
+#include <net/xdp.h>
+
+#define REG_STATE_NEW		0x0
+#define REG_STATE_REGISTERED	0x1
+#define REG_STATE_UNREGISTERED	0x2
+#define REG_STATE_UNUSED	0x3
+
+void xdp_rxq_info_unreg(struct xdp_rxq_info *xdp_rxq)
+{
+	/* Simplify driver cleanup code paths, allow unreg "unused" */
+	if (xdp_rxq->reg_state == REG_STATE_UNUSED)
+		return;
+
+	WARN(!(xdp_rxq->reg_state == REG_STATE_REGISTERED), "Driver BUG");
+
+	xdp_rxq->reg_state = REG_STATE_UNREGISTERED;
+	xdp_rxq->dev = NULL;
+}
+EXPORT_SYMBOL_GPL(xdp_rxq_info_unreg);
+
+static void xdp_rxq_info_init(struct xdp_rxq_info *xdp_rxq)
+{
+	memset(xdp_rxq, 0, sizeof(*xdp_rxq));
+}
+
+/* Returns 0 on success, negative on failure */
+int xdp_rxq_info_reg(struct xdp_rxq_info *xdp_rxq,
+		     struct net_device *dev, u32 queue_index)
+{
+	if (xdp_rxq->reg_state == REG_STATE_UNUSED) {
+		WARN(1, "Driver promised not to register this");
+		return -EINVAL;
+	}
+
+	if (xdp_rxq->reg_state == REG_STATE_REGISTERED) {
+		WARN(1, "Missing unregister, handled but fix driver");
+		xdp_rxq_info_unreg(xdp_rxq);
+	}
+
+	if (!dev) {
+		WARN(1, "Missing net_device from driver");
+		return -ENODEV;
+	}
+
+	/* State either UNREGISTERED or NEW */
+	xdp_rxq_info_init(xdp_rxq);
+	xdp_rxq->dev = dev;
+	xdp_rxq->queue_index = queue_index;
+
+	xdp_rxq->reg_state = REG_STATE_REGISTERED;
+	return 0;
+}
+EXPORT_SYMBOL_GPL(xdp_rxq_info_reg);
+
+void xdp_rxq_info_unused(struct xdp_rxq_info *xdp_rxq)
+{
+	xdp_rxq->reg_state = REG_STATE_UNUSED;
+}
+EXPORT_SYMBOL_GPL(xdp_rxq_info_unused);