1 files changed, 143 insertions, 122 deletions

diff --git a/sys/net/netmap.h b/sys/net/netmap.h
index b5ab6d549084..0f2baebe15fc 100644
--- a/sys/net/netmap.h
+++ b/sys/net/netmap.h
@@ -38,6 +38,8 @@
  * Detailed info on netmap is available with "man netmap" or at
  * 
  *	http://info.iet.unipi.it/~luigi/netmap/
+ *
+ * This API is also used to communicate with the VALE software switch
  */
 
 #ifndef _NET_NETMAP_H_
@@ -46,106 +48,95 @@
 /*
  * --- Netmap data structures ---
  *
- * The data structures used by netmap are shown below. Those in
- * capital letters are in an mmapp()ed area shared with userspace,
- * while others are private to the kernel.
- * Shared structures do not contain pointers but only memory
- * offsets, so that addressing is portable between kernel and userspace.
+ * The userspace data structures used by netmap are shown below.
+ * They are allocated by the kernel and mmap()ed by userspace threads.
+ * Pointers are implemented as memory offsets or indexes,
+ * so that they can be easily dereferenced in kernel and userspace.
 
+   KERNEL (opaque, obviously)
 
- softc
-+----------------+
-| standard fields|
-| if_pspare[0] ----------+
-+----------------+       |
-                         |
-+----------------+<------+
-|(netmap_adapter)|
-|                |                             netmap_kring
-| tx_rings *--------------------------------->+---------------+
-|                |       netmap_kring         | ring    *---------.
-| rx_rings *--------->+---------------+       | nr_hwcur      |   |
-+----------------+    | ring    *--------.    | nr_hwavail    |   V
-                      | nr_hwcur      |  |    | selinfo       |   |
-                      | nr_hwavail    |  |    +---------------+   .
-                      | selinfo       |  |    |     ...       |   .
-                      +---------------+  |    |(ntx+1 entries)|
-                      |    ....       |  |    |               |
-                      |(nrx+1 entries)|  |    +---------------+
-                      |               |  |
-   KERNEL             +---------------+  |
-                                         |
   ====================================================================
                                          |
-   USERSPACE                             |      NETMAP_RING
-                                         +---->+-------------+
-                                             / | cur         |
-   NETMAP_IF  (nifp, one per file desc.)    /  | avail       |
-    +---------------+                      /   | buf_ofs     |
-    | ni_tx_rings   |                     /    +=============+
-    | ni_rx_rings   |                    /     | buf_idx     | slot[0]
-    |               |                   /      | len, flags  |
-    |               |                  /       +-------------+
-    +===============+                 /        | buf_idx     | slot[1]
-    | txring_ofs[0] | (rel.to nifp)--'         | len, flags  |
-    | txring_ofs[1] |                          +-------------+
-  (num_rings+1 entries)                     (nr_num_slots entries)
-    | txring_ofs[n] |                          | buf_idx     | slot[n-1]
-    +---------------+                          | len, flags  |
-    | rxring_ofs[0] |                          +-------------+
+   USERSPACE                             |      struct netmap_ring
+                                         +---->+--------------+
+                                             / | cur          |
+   struct netmap_if (nifp, 1 per fd)        /  | avail        |
+    +---------------+                      /   | buf_ofs      |
+    | ni_tx_rings   |                     /    +==============+
+    | ni_rx_rings   |                    /     | buf_idx, len | slot[0]
+    |               |                   /      | flags, ptr   |
+    |               |                  /       +--------------+
+    +===============+                 /        | buf_idx, len | slot[1]
+    | txring_ofs[0] | (rel.to nifp)--'         | flags, ptr   |
+    | txring_ofs[1] |                          +--------------+
+  (ni_tx_rings+1 entries)                     (num_slots entries)
+    | txring_ofs[t] |                          | buf_idx, len | slot[n-1]
+    +---------------+                          | flags, ptr   |
+    | rxring_ofs[0] |                          +--------------+
     | rxring_ofs[1] |
-  (num_rings+1 entries)
-    | txring_ofs[n] |
+  (ni_rx_rings+1 entries)
+    | rxring_ofs[r] |
     +---------------+
 
- * The private descriptor ('softc' or 'adapter') of each interface
- * is extended with a "struct netmap_adapter" containing netmap-related
- * info (see description in dev/netmap/netmap_kernel.h.
- * Among other things, tx_rings and rx_rings point to the arrays of
- * "struct netmap_kring" which in turn reache the various
- * "struct netmap_ring", shared with userspace.
-
- * The NETMAP_RING is the userspace-visible replica of the NIC ring.
- * Each slot has the index of a buffer, its length and some flags.
+ * For each "interface" (NIC, host stack, VALE switch port) attached to a
+ * file descriptor, the mmap()ed region contains a (logically readonly)
+ * struct netmap_if pointing to struct netmap_ring's.
+ * There is one netmap_ring per physical NIC ring, plus one tx/rx ring
+ * pair attached to the host stack (this pair is unused for VALE ports).
+ *
+ * All physical/host stack ports share the same memory region,
+ * so that zero-copy can be implemented between them.
+ * VALE switch ports instead have separate memory regions.
+ *
+ * The netmap_ring is the userspace-visible replica of the NIC ring.
+ * Each slot has the index of a buffer (MTU-sized and residing in the
+ * mmapped region), its length and some flags. An extra 64-bit pointer
+ * is provided for user-supplied buffers in the tx path.
+ *
  * In user space, the buffer address is computed as
  *	(char *)ring + buf_ofs + index*NETMAP_BUF_SIZE
- * In the kernel, buffers do not necessarily need to be contiguous,
- * and the virtual and physical addresses are derived through
- * a lookup table.
+ */
+
+/*
+ * struct netmap_slot is a buffer descriptor
  *
- * struct netmap_slot:
+ * buf_idx	the index of the buffer associated to the slot.
+ * len		the length of the payload
+ * flags	control operation on the slot, as defined below
  *
- * buf_idx	is the index of the buffer associated to the slot.
- * len		is the length of the payload
  * NS_BUF_CHANGED	must be set whenever userspace wants
  *		to change buf_idx (it might be necessary to
- *		reprogram the NIC slot)
+ *		reprogram the NIC)
+ *
  * NS_REPORT	must be set if we want the NIC to generate an interrupt
  *		when this slot is used. Leaving it to 0 improves
  *		performance.
+ *
  * NS_FORWARD	if set on a receive ring, and the device is in
  *		transparent mode, buffers released with the flag set
  *		will be forwarded to the 'other' side (host stack
  *		or NIC, respectively) on the next select() or ioctl()
  *
- *		The following will be supported from NETMAP_API = 5
  * NS_NO_LEARN	on a VALE switch, do not 'learn' the source port for
  *		this packet.
- * NS_INDIRECT	the netmap buffer contains a 64-bit pointer to
- *		the actual userspace buffer. This may be useful
- *		to reduce copies in a VM environment.
+ *
+ * NS_INDIRECT	(tx rings only) data is in a userspace buffer pointed
+ *		by the ptr field in the slot.
+ *
  * NS_MOREFRAG	Part of a multi-segment frame. The last (or only)
  *		segment must not have this flag.
+ *		Only supported on VALE ports.
+ *
  * NS_PORT_MASK	the high 8 bits of the flag, if not zero, indicate the
  *		destination port for the VALE switch, overriding
  *		the lookup table.
  */
 
 struct netmap_slot {
-	uint32_t buf_idx; /* buffer index */
-	uint16_t len;	/* packet length, to be copied to/from the hw ring */
-	uint16_t flags;	/* buf changed, etc. */
-#define	NS_BUF_CHANGED	0x0001	/* must resync the map, buffer changed */
+	uint32_t buf_idx;	/* buffer index */
+	uint16_t len;		/* packet length */
+	uint16_t flags;		/* buf changed, etc. */
+#define	NS_BUF_CHANGED	0x0001	/* buf_idx changed */
 #define	NS_REPORT	0x0002	/* ask the hardware to report results
 				 * e.g. by generating an interrupt
 				 */
@@ -157,62 +148,61 @@ struct netmap_slot {
 #define	NS_MOREFRAG	0x0020
 #define	NS_PORT_SHIFT	8
 #define	NS_PORT_MASK	(0xff << NS_PORT_SHIFT)
+				/*
+				 * in rx rings, the high 8 bits
+				 *  are the number of fragments.
+				 */
+#define	NS_RFRAGS(_slot)	( ((_slot)->flags >> 8) & 0xff)
+	uint64_t	ptr;	/* pointer for indirect buffers */
 };
 
 /*
+ * struct netmap_ring
+ *
  * Netmap representation of a TX or RX ring (also known as "queue").
  * This is a queue implemented as a fixed-size circular array.
  * At the software level, two fields are important: avail and cur.
  *
  * In TX rings:
- *	avail	indicates the number of slots available for transmission.
- *		It is updated by the kernel after every netmap system call.
- *		It MUST BE decremented by the application when it appends a
- *		packet.
+ *
+ *	avail	tells how many slots are available for transmission.
+ *		It is updated by the kernel in each netmap system call.
+ *		It MUST BE decremented by the user when it
+ *		adds a new packet to send.
+ *
  *	cur	indicates the slot to use for the next packet
  *		to send (i.e. the "tail" of the queue).
- *		It MUST BE incremented by the application before
+ *		It MUST BE incremented by the user before
  *		netmap system calls to reflect the number of newly
  *		sent packets.
  *		It is checked by the kernel on netmap system calls
  *		(normally unmodified by the kernel unless invalid).
  *
- *   The kernel side of netmap uses two additional fields in its own
- *   private ring structure, netmap_kring:
- *	nr_hwcur is a copy of nr_cur on an NIOCTXSYNC.
- *	nr_hwavail is the number of slots known as available by the
- *		hardware. It is updated on an INTR (inc by the
- *		number of packets sent) and on a NIOCTXSYNC
- *		(decrease by nr_cur - nr_hwcur)
- *		A special case, nr_hwavail is -1 if the transmit
- *		side is idle (no pending transmits).
- *
  * In RX rings:
+ *
  *	avail	is the number of packets available (possibly 0).
- *		It MUST BE decremented by the application when it consumes
- *		a packet, and it is updated to nr_hwavail on a NIOCRXSYNC
+ *		It is updated by the kernel in each netmap system call.
+ *		It MUST BE decremented by the user when it
+ *		consumes a packet.
+ *
  *	cur	indicates the first slot that contains a packet not
- *		processed yet (the "head" of the queue).
- *		It MUST BE incremented by the software when it consumes
+ *		yet processed (the "head" of the queue).
+ *		It MUST BE incremented by the user when it consumes
  *		a packet.
+ *
  *	reserved	indicates the number of buffers before 'cur'
- *		that the application has still in use. Normally 0,
- *		it MUST BE incremented by the application when it
+ *		that the user has not released yet. Normally 0,
+ *		it MUST BE incremented by the user when it
  *		does not return the buffer immediately, and decremented
  *		when the buffer is finally freed.
  *
- *   The kernel side of netmap uses two additional fields in the kring:
- *	nr_hwcur is a copy of nr_cur on an NIOCRXSYNC
- *	nr_hwavail is the number of packets available. It is updated
- *		on INTR (inc by the number of new packets arrived)
- *		and on NIOCRXSYNC (decreased by nr_cur - nr_hwcur).
  *
  * DATA OWNERSHIP/LOCKING:
- *	The netmap_ring is owned by the user program and it is only
- *	accessed or modified in the upper half of the kernel during
- *	a system call.
- *
- *	The netmap_kring is only modified by the upper half of the kernel.
+ *	The netmap_ring, all slots, and buffers in the range
+ *	[reserved-cur , cur+avail[ are owned by the user program,
+ *	and the kernel only touches them in the same thread context
+ *	during a system call.
+ *	Other buffers are reserved for use by the NIC's DMA engines.
  *
  * FLAGS
  *	NR_TIMESTAMP	updates the 'ts' field on each syscall. This is
@@ -228,7 +218,7 @@ struct netmap_slot {
  */
 struct netmap_ring {
 	/*
-	 * nr_buf_base_ofs is meant to be used through macros.
+	 * buf_ofs is meant to be used through macros.
 	 * It contains the offset of the buffer region from this
 	 * descriptor.
 	 */
@@ -253,23 +243,29 @@ struct netmap_ring {
 
 /*
  * Netmap representation of an interface and its queue(s).
+ * This is initialized by the kernel when binding a file
+ * descriptor to a port, and should be considered as readonly
+ * by user programs. The kernel never uses it.
+ *
  * There is one netmap_if for each file descriptor on which we want
- * to select/poll.  We assume that on each interface has the same number
- * of receive and transmit queues.
+ * to select/poll.
  * select/poll operates on one or all pairs depending on the value of
  * nmr_queueid passed on the ioctl.
  */
 struct netmap_if {
 	char		ni_name[IFNAMSIZ]; /* name of the interface. */
-	const u_int	ni_version;	/* API version, currently unused */
-	const u_int	ni_rx_rings;	/* number of rx rings */
-	const u_int	ni_tx_rings;	/* if zero, same as ni_rx_rings */
+	const uint32_t	ni_version;	/* API version, currently unused */
+	const uint32_t	ni_flags;	/* properties */
+#define	NI_PRIV_MEM	0x1		/* private memory region */
+
+	const uint32_t	ni_rx_rings;	/* number of rx rings */
+	const uint32_t	ni_tx_rings;	/* number of tx rings */
 	/*
 	 * The following array contains the offset of each netmap ring
-	 * from this structure. The first ni_tx_queues+1 entries refer
-	 * to the tx rings, the next ni_rx_queues+1 refer to the rx rings
+	 * from this structure. The first ni_tx_rings+1 entries refer
+	 * to the tx rings, the next ni_rx_rings+1 refer to the rx rings
 	 * (the last entry in each block refers to the host stack rings).
-	 * The area is filled up by the kernel on NIOCREG,
+	 * The area is filled up by the kernel on NIOCREGIF,
 	 * and then only read by userspace code.
 	 */
 	const ssize_t	ring_ofs[0];
@@ -282,23 +278,47 @@ struct netmap_if {
  * NIOCGINFO takes a struct ifreq, the interface name is the input,
  *	the outputs are number of queues and number of descriptor
  *	for each queue (useful to set number of threads etc.).
+ *	The info returned is only advisory and may change before
+ *	the interface is bound to a file descriptor.
  *
  * NIOCREGIF takes an interface name within a struct ifreq,
  *	and activates netmap mode on the interface (if possible).
  *
- *	For vale ports, starting with NETMAP_API = 5,
- *	nr_tx_rings and nr_rx_rings specify how many software rings
- *	are created (0 means 1).
+ *   nr_name	is the name of the interface
+ *
+ *   nr_tx_slots, nr_tx_slots, nr_tx_rings, nr_rx_rings
+ *	indicate the configuration of the port on return.
+ *
+ *	On input, non-zero values for nr_tx_rings, nr_tx_slots and the
+ *	rx counterparts may be used to reconfigure the port according
+ *	to the requested values, but this is not guaranteed.
+ *	The actual values are returned on completion of the ioctl().
+ *
+ *   nr_ringid
+ *	indicates how rings should be bound to the file descriptors.
+ *	The default (0) means all physical rings of a NIC are bound.
+ *	NETMAP_HW_RING plus a ring number lets you bind just
+ *	a single ring pair.
+ *	NETMAP_SW_RING binds only the host tx/rx rings
+ *	NETMAP_NO_TX_POLL prevents select()/poll() from pushing
+ *	out packets on the tx ring unless POLLOUT is specified.
+ *
+ *	NETMAP_PRIV_MEM is a return value used to indicate that
+ *	this ring is in a private memory region hence buffer
+ *	swapping cannot be used
+ *	
+ *   nr_cmd	is used to configure NICs attached to a VALE switch,
+ *	or to dump the configuration of a VALE switch.
+ *	
+ *	nr_cmd = NETMAP_BDG_ATTACH and nr_name = vale*:ifname
+ *	attaches the NIC to the switch, with nr_ringid specifying
+ *	which rings to use
  *
- *	NIOCREGIF is also used to attach a NIC to a VALE switch.
- *	In this case the name is vale*:ifname, and "nr_cmd"
- *	is set to 'NETMAP_BDG_ATTACH' or 'NETMAP_BDG_DETACH'.
- *	nr_ringid specifies which rings should be attached, 0 means all,
- *	NETMAP_HW_RING + n means only the n-th ring.
- *	The process can terminate after the interface has been attached.
+ *	nr_cmd = NETMAP_BDG_DETACH and nr_name = vale*:ifname
+ *	disconnects a previously attached NIC
  *
- * NIOCUNREGIF unregisters the interface associated to the fd.
- *	this is deprecated and will go away.
+ *	nr_cmd = NETMAP_BDG_LIST is used to list the configuration
+ *	of VALE switches, with additional arguments.
  *
  * NIOCTXSYNC, NIOCRXSYNC synchronize tx or rx queues,
  *	whose identity is set in NIOCREGIF through nr_ringid
@@ -312,7 +332,7 @@ struct netmap_if {
 struct nmreq {
 	char		nr_name[IFNAMSIZ];
 	uint32_t	nr_version;	/* API version */
-#define	NETMAP_API	4		/* current version */
+#define	NETMAP_API	5		/* current version */
 	uint32_t	nr_offset;	/* nifp offset in the shared region */
 	uint32_t	nr_memsize;	/* size of the shared region */
 	uint32_t	nr_tx_slots;	/* slots in tx rings */
@@ -320,6 +340,7 @@ struct nmreq {
 	uint16_t	nr_tx_rings;	/* number of tx rings */
 	uint16_t	nr_rx_rings;	/* number of rx rings */
 	uint16_t	nr_ringid;	/* ring(s) we care about */
+#define NETMAP_PRIV_MEM	0x8000		/* rings use private memory */
 #define NETMAP_HW_RING	0x4000		/* low bits indicate one hw ring */
 #define NETMAP_SW_RING	0x2000		/* process the sw ring */
 #define NETMAP_NO_TX_POLL	0x1000	/* no automatic txsync on poll */
@@ -343,7 +364,7 @@ struct nmreq {
  */
 #define NIOCGINFO	_IOWR('i', 145, struct nmreq) /* return IF info */
 #define NIOCREGIF	_IOWR('i', 146, struct nmreq) /* interface register */
-#define NIOCUNREGIF	_IO('i', 147) /* interface unregister */
+#define NIOCUNREGIF	_IO('i', 147) /* deprecated. Was interface unregister */
 #define NIOCTXSYNC	_IO('i', 148) /* sync tx queues */
 #define NIOCRXSYNC	_IO('i', 149) /* sync rx queues */
 #endif /* !NIOCREGIF */