2 * Copyright (C) 2011-2014 Matteo Landi, Luigi Rizzo
3 * Copyright (C) 2013-2016 Universita` di Pisa
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * The header contains the definitions of constants and function
32 * prototypes used only in kernelspace.
35 #ifndef _NET_NETMAP_KERN_H_
36 #define _NET_NETMAP_KERN_H_
40 #if defined(CONFIG_NETMAP_VALE)
43 #if defined(CONFIG_NETMAP_PIPE)
46 #if defined(CONFIG_NETMAP_MONITOR)
49 #if defined(CONFIG_NETMAP_GENERIC)
52 #if defined(CONFIG_NETMAP_PTNETMAP_GUEST)
53 #define WITH_PTNETMAP_GUEST
55 #if defined(CONFIG_NETMAP_PTNETMAP_HOST)
56 #define WITH_PTNETMAP_HOST
58 #if defined(CONFIG_NETMAP_SINK)
62 #elif defined (_WIN32)
63 #define WITH_VALE // comment out to disable VALE support
68 #else /* neither linux nor windows */
69 #define WITH_VALE // comment out to disable VALE support
73 #define WITH_PTNETMAP_HOST /* ptnetmap host support */
74 #define WITH_PTNETMAP_GUEST /* ptnetmap guest support */
78 #if defined(__FreeBSD__)
79 #include <sys/selinfo.h>
81 #define likely(x) __builtin_expect((long)!!(x), 1L)
82 #define unlikely(x) __builtin_expect((long)!!(x), 0L)
85 #define NM_LOCK_T struct mtx /* low level spinlock, used to protect queues */
87 #define NM_MTX_T struct sx /* OS-specific mutex (sleepable) */
88 #define NM_MTX_INIT(m) sx_init(&(m), #m)
89 #define NM_MTX_DESTROY(m) sx_destroy(&(m))
90 #define NM_MTX_LOCK(m) sx_xlock(&(m))
91 #define NM_MTX_UNLOCK(m) sx_xunlock(&(m))
92 #define NM_MTX_ASSERT(m) sx_assert(&(m), SA_XLOCKED)
94 #define NM_SELINFO_T struct nm_selinfo
95 #define NM_SELRECORD_T struct thread
96 #define MBUF_LEN(m) ((m)->m_pkthdr.len)
97 #define MBUF_TXQ(m) ((m)->m_pkthdr.flowid)
98 #define MBUF_TRANSMIT(na, ifp, m) ((na)->if_transmit(ifp, m))
99 #define GEN_TX_MBUF_IFP(m) ((m)->m_pkthdr.rcvif)
101 #define NM_ATOMIC_T volatile int // XXX ?
102 /* atomic operations */
103 #include <machine/atomic.h>
104 #define NM_ATOMIC_TEST_AND_SET(p) (!atomic_cmpset_acq_int((p), 0, 1))
105 #define NM_ATOMIC_CLEAR(p) atomic_store_rel_int((p), 0)
107 #if __FreeBSD_version >= 1100030
108 #define WNA(_ifp) (_ifp)->if_netmap
109 #else /* older FreeBSD */
110 #define WNA(_ifp) (_ifp)->if_pspare[0]
111 #endif /* older FreeBSD */
113 #if __FreeBSD_version >= 1100005
114 struct netmap_adapter *netmap_getna(if_t ifp);
117 #if __FreeBSD_version >= 1100027
118 #define MBUF_REFCNT(m) ((m)->m_ext.ext_count)
119 #define SET_MBUF_REFCNT(m, x) (m)->m_ext.ext_count = x
121 #define MBUF_REFCNT(m) ((m)->m_ext.ref_cnt ? *((m)->m_ext.ref_cnt) : -1)
122 #define SET_MBUF_REFCNT(m, x) *((m)->m_ext.ref_cnt) = x
125 #define MBUF_QUEUED(m) 1
133 // XXX linux struct, not used in FreeBSD
134 struct net_device_ops {
140 #define NM_BNS_GET(b)
141 #define NM_BNS_PUT(b)
143 #elif defined (linux)
145 #define NM_LOCK_T safe_spinlock_t // see bsd_glue.h
146 #define NM_SELINFO_T wait_queue_head_t
147 #define MBUF_LEN(m) ((m)->len)
148 #define MBUF_TRANSMIT(na, ifp, m) \
150 /* Avoid infinite recursion with generic. */ \
151 m->priority = NM_MAGIC_PRIORITY_TX; \
152 (((struct net_device_ops *)(na)->if_transmit)->ndo_start_xmit(m, ifp)); \
156 /* See explanation in nm_os_generic_xmit_frame. */
157 #define GEN_TX_MBUF_IFP(m) ((struct ifnet *)skb_shinfo(m)->destructor_arg)
159 #define NM_ATOMIC_T volatile long unsigned int
161 #define NM_MTX_T struct mutex /* OS-specific sleepable lock */
162 #define NM_MTX_INIT(m) mutex_init(&(m))
163 #define NM_MTX_DESTROY(m) do { (void)(m); } while (0)
164 #define NM_MTX_LOCK(m) mutex_lock(&(m))
165 #define NM_MTX_UNLOCK(m) mutex_unlock(&(m))
166 #define NM_MTX_ASSERT(m) mutex_is_locked(&(m))
170 #endif /* DEV_NETMAP */
172 #elif defined (__APPLE__)
174 #warning apple support is incomplete.
175 #define likely(x) __builtin_expect(!!(x), 1)
176 #define unlikely(x) __builtin_expect(!!(x), 0)
177 #define NM_LOCK_T IOLock *
178 #define NM_SELINFO_T struct selinfo
179 #define MBUF_LEN(m) ((m)->m_pkthdr.len)
181 #elif defined (_WIN32)
182 #include "../../../WINDOWS/win_glue.h"
184 #define NM_SELRECORD_T IO_STACK_LOCATION
185 #define NM_SELINFO_T win_SELINFO // see win_glue.h
186 #define NM_LOCK_T win_spinlock_t // see win_glue.h
187 #define NM_MTX_T KGUARDED_MUTEX /* OS-specific mutex (sleepable) */
189 #define NM_MTX_INIT(m) KeInitializeGuardedMutex(&m);
190 #define NM_MTX_DESTROY(m) do { (void)(m); } while (0)
191 #define NM_MTX_LOCK(m) KeAcquireGuardedMutex(&(m))
192 #define NM_MTX_UNLOCK(m) KeReleaseGuardedMutex(&(m))
193 #define NM_MTX_ASSERT(m) assert(&m.Count>0)
195 //These linknames are for the NDIS driver
196 #define NETMAP_NDIS_LINKNAME_STRING L"\\DosDevices\\NMAPNDIS"
197 #define NETMAP_NDIS_NTDEVICE_STRING L"\\Device\\NMAPNDIS"
199 //Definition of internal driver-to-driver ioctl codes
200 #define NETMAP_KERNEL_XCHANGE_POINTERS _IO('i', 180)
201 #define NETMAP_KERNEL_SEND_SHUTDOWN_SIGNAL _IO_direct('i', 195)
203 //Empty data structures are not permitted by MSVC compiler
204 //XXX_ale, try to solve this problem
205 struct net_device_ops{
208 typedef struct ethtool_ops{
211 typedef struct hrtimer{
217 /* MSVC does not have likely/unlikely support */
219 #define likely(x) (x)
220 #define unlikely(x) (x)
222 #define likely(x) __builtin_expect((long)!!(x), 1L)
223 #define unlikely(x) __builtin_expect((long)!!(x), 0L)
228 #error unsupported platform
230 #endif /* end - platform-specific code */
232 #ifndef _WIN32 /* support for emulated sysctl */
237 #define NM_ACCESS_ONCE(x) (*(volatile __typeof__(x) *)&(x))
239 #define NMG_LOCK_T NM_MTX_T
240 #define NMG_LOCK_INIT() NM_MTX_INIT(netmap_global_lock)
241 #define NMG_LOCK_DESTROY() NM_MTX_DESTROY(netmap_global_lock)
242 #define NMG_LOCK() NM_MTX_LOCK(netmap_global_lock)
243 #define NMG_UNLOCK() NM_MTX_UNLOCK(netmap_global_lock)
244 #define NMG_LOCK_ASSERT() NM_MTX_ASSERT(netmap_global_lock)
246 #if defined(__FreeBSD__)
247 #define nm_prerr printf
248 #define nm_prinf printf
249 #elif defined (_WIN32)
250 #define nm_prerr DbgPrint
251 #define nm_prinf DbgPrint
253 #define nm_prerr(fmt, arg...) printk(KERN_ERR fmt, ##arg)
254 #define nm_prinf(fmt, arg...) printk(KERN_INFO fmt, ##arg)
257 #define ND(format, ...)
258 #define D(format, ...) \
260 struct timeval __xxts; \
261 microtime(&__xxts); \
262 nm_prerr("%03d.%06d [%4d] %-25s " format "\n", \
263 (int)__xxts.tv_sec % 1000, (int)__xxts.tv_usec, \
264 __LINE__, __FUNCTION__, ##__VA_ARGS__); \
267 /* rate limited, lps indicates how many per second */
268 #define RD(lps, format, ...) \
270 static int t0, __cnt; \
271 if (t0 != time_second) { \
276 D(format, ##__VA_ARGS__); \
279 struct netmap_adapter;
282 struct netmap_priv_d;
284 /* os-specific NM_SELINFO_T initialzation/destruction functions */
285 void nm_os_selinfo_init(NM_SELINFO_T *);
286 void nm_os_selinfo_uninit(NM_SELINFO_T *);
288 const char *nm_dump_buf(char *p, int len, int lim, char *dst);
290 void nm_os_selwakeup(NM_SELINFO_T *si);
291 void nm_os_selrecord(NM_SELRECORD_T *sr, NM_SELINFO_T *si);
293 int nm_os_ifnet_init(void);
294 void nm_os_ifnet_fini(void);
295 void nm_os_ifnet_lock(void);
296 void nm_os_ifnet_unlock(void);
298 void nm_os_get_module(void);
299 void nm_os_put_module(void);
301 void netmap_make_zombie(struct ifnet *);
302 void netmap_undo_zombie(struct ifnet *);
304 /* os independent alloc/realloc/free */
305 void *nm_os_malloc(size_t);
306 void *nm_os_realloc(void *, size_t new_size, size_t old_size);
307 void nm_os_free(void *);
309 /* passes a packet up to the host stack.
310 * If the packet is sent (or dropped) immediately it returns NULL,
311 * otherwise it links the packet to prev and returns m.
312 * In this case, a final call with m=NULL and prev != NULL will send up
313 * the entire chain to the host stack.
315 void *nm_os_send_up(struct ifnet *, struct mbuf *m, struct mbuf *prev);
317 int nm_os_mbuf_has_offld(struct mbuf *m);
319 #include "netmap_mbq.h"
321 extern NMG_LOCK_T netmap_global_lock;
323 enum txrx { NR_RX = 0, NR_TX = 1, NR_TXRX };
325 static __inline const char*
326 nm_txrx2str(enum txrx t)
328 return (t== NR_RX ? "RX" : "TX");
331 static __inline enum txrx
332 nm_txrx_swap(enum txrx t)
334 return (t== NR_RX ? NR_TX : NR_RX);
337 #define for_rx_tx(t) for ((t) = 0; (t) < NR_TXRX; (t)++)
340 struct netmap_zmon_list {
341 struct netmap_kring *next;
342 struct netmap_kring *prev;
344 #endif /* WITH_MONITOR */
347 * private, kernel view of a ring. Keeps track of the status of
348 * a ring across system calls.
350 * nr_hwcur index of the next buffer to refill.
351 * It corresponds to ring->head
352 * at the time the system call returns.
354 * nr_hwtail index of the first buffer owned by the kernel.
355 * On RX, hwcur->hwtail are receive buffers
356 * not yet released. hwcur is advanced following
357 * ring->head, hwtail is advanced on incoming packets,
358 * and a wakeup is generated when hwtail passes ring->cur
359 * On TX, hwcur->rcur have been filled by the sender
360 * but not sent yet to the NIC; rcur->hwtail are available
361 * for new transmissions, and hwtail->hwcur-1 are pending
362 * transmissions not yet acknowledged.
364 * The indexes in the NIC and netmap rings are offset by nkr_hwofs slots.
365 * This is so that, on a reset, buffers owned by userspace are not
366 * modified by the kernel. In particular:
367 * RX rings: the next empty buffer (hwtail + hwofs) coincides with
368 * the next empty buffer as known by the hardware (next_to_check or so).
369 * TX rings: hwcur + hwofs coincides with next_to_send
371 * For received packets, slot->flags is set to nkr_slot_flags
372 * so we can provide a proper initial value (e.g. set NS_FORWARD
373 * when operating in 'transparent' mode).
375 * The following fields are used to implement lock-free copy of packets
376 * from input to output ports in VALE switch:
377 * nkr_hwlease buffer after the last one being copied.
378 * A writer in nm_bdg_flush reserves N buffers
379 * from nr_hwlease, advances it, then does the
380 * copy outside the lock.
381 * In RX rings (used for VALE ports),
382 * nkr_hwtail <= nkr_hwlease < nkr_hwcur+N-1
383 * In TX rings (used for NIC or host stack ports)
384 * nkr_hwcur <= nkr_hwlease < nkr_hwtail
385 * nkr_leases array of nkr_num_slots where writers can report
386 * completion of their block. NR_NOSLOT (~0) indicates
387 * that the writer has not finished yet
388 * nkr_lease_idx index of next free slot in nr_leases, to be assigned
390 * The kring is manipulated by txsync/rxsync and generic netmap function.
392 * Concurrent rxsync or txsync on the same ring are prevented through
393 * by nm_kr_(try)lock() which in turn uses nr_busy. This is all we need
394 * for NIC rings, and for TX rings attached to the host stack.
396 * RX rings attached to the host stack use an mbq (rx_queue) on both
397 * rxsync_from_host() and netmap_transmit(). The mbq is protected
398 * by its internal lock.
400 * RX rings attached to the VALE switch are accessed by both senders
401 * and receiver. They are protected through the q_lock on the RX ring.
403 struct netmap_kring {
404 struct netmap_ring *ring;
410 * Copies of values in user rings, so we do not need to look
411 * at the ring (which could be modified). These are set in the
412 * *sync_prologue()/finalize() routines.
418 uint32_t nr_kflags; /* private driver flags */
419 #define NKR_PENDINTR 0x1 // Pending interrupt.
420 #define NKR_EXCLUSIVE 0x2 /* exclusive binding */
421 #define NKR_FORWARD 0x4 /* (host ring only) there are
424 #define NKR_NEEDRING 0x8 /* ring needed even if users==0
425 * (used internally by pipes and
426 * by ptnetmap host ports)
430 uint32_t nr_pending_mode;
431 #define NKR_NETMAP_OFF 0x0
432 #define NKR_NETMAP_ON 0x1
434 uint32_t nkr_num_slots;
437 * On a NIC reset, the NIC ring indexes may be reset but the
438 * indexes in the netmap rings remain the same. nkr_hwofs
439 * keeps track of the offset between the two.
443 uint16_t nkr_slot_flags; /* initial value for flags */
445 /* last_reclaim is opaque marker to help reduce the frequency
446 * of operations such as reclaiming tx buffers. A possible use
447 * is set it to ticks and do the reclaim only once per tick.
449 uint64_t last_reclaim;
452 NM_SELINFO_T si; /* poll/select wait queue */
453 NM_LOCK_T q_lock; /* protects kring and ring. */
454 NM_ATOMIC_T nr_busy; /* prevent concurrent syscalls */
456 struct netmap_adapter *na;
458 /* The following fields are for VALE switch support */
459 struct nm_bdg_fwd *nkr_ft;
460 uint32_t *nkr_leases;
461 #define NR_NOSLOT ((uint32_t)~0) /* used in nkr_*lease* */
462 uint32_t nkr_hwlease;
463 uint32_t nkr_lease_idx;
465 /* while nkr_stopped is set, no new [tr]xsync operations can
466 * be started on this kring.
467 * This is used by netmap_disable_all_rings()
468 * to find a synchronization point where critical data
469 * structures pointed to by the kring can be added or removed
471 volatile int nkr_stopped;
473 /* Support for adapters without native netmap support.
474 * On tx rings we preallocate an array of tx buffers
475 * (same size as the netmap ring), on rx rings we
476 * store incoming mbufs in a queue that is drained by
479 struct mbuf **tx_pool;
480 struct mbuf *tx_event; /* TX event used as a notification */
481 NM_LOCK_T tx_event_lock; /* protects the tx_event mbuf */
482 struct mbq rx_queue; /* intercepted rx mbufs. */
484 uint32_t users; /* existing bindings for this ring */
486 uint32_t ring_id; /* kring identifier */
487 enum txrx tx; /* kind of ring (tx or rx) */
488 char name[64]; /* diagnostic */
490 /* [tx]sync callback for this kring.
491 * The default nm_kring_create callback (netmap_krings_create)
492 * sets the nm_sync callback of each hardware tx(rx) kring to
493 * the corresponding nm_txsync(nm_rxsync) taken from the
494 * netmap_adapter; moreover, it sets the sync callback
495 * of the host tx(rx) ring to netmap_txsync_to_host
496 * (netmap_rxsync_from_host).
498 * Overrides: the above configuration is not changed by
499 * any of the nm_krings_create callbacks.
501 int (*nm_sync)(struct netmap_kring *kring, int flags);
502 int (*nm_notify)(struct netmap_kring *kring, int flags);
505 struct netmap_kring *pipe; /* if this is a pipe ring,
506 * pointer to the other end
508 #endif /* WITH_PIPES */
511 int (*save_notify)(struct netmap_kring *kring, int flags);
515 /* array of krings that are monitoring this kring */
516 struct netmap_kring **monitors;
517 uint32_t max_monitors; /* current size of the monitors array */
518 uint32_t n_monitors; /* next unused entry in the monitor array */
519 uint32_t mon_pos[NR_TXRX]; /* index of this ring in the monitored ring array */
520 uint32_t mon_tail; /* last seen slot on rx */
522 /* circular list of zero-copy monitors */
523 struct netmap_zmon_list zmon_list[NR_TXRX];
526 * Monitors work by intercepting the sync and notify callbacks of the
527 * monitored krings. This is implemented by replacing the pointers
528 * above and saving the previous ones in mon_* pointers below
530 int (*mon_sync)(struct netmap_kring *kring, int flags);
531 int (*mon_notify)(struct netmap_kring *kring, int flags);
536 __declspec(align(64));
538 __attribute__((__aligned__(64)));
541 /* return 1 iff the kring needs to be turned on */
543 nm_kring_pending_on(struct netmap_kring *kring)
545 return kring->nr_pending_mode == NKR_NETMAP_ON &&
546 kring->nr_mode == NKR_NETMAP_OFF;
549 /* return 1 iff the kring needs to be turned off */
551 nm_kring_pending_off(struct netmap_kring *kring)
553 return kring->nr_pending_mode == NKR_NETMAP_OFF &&
554 kring->nr_mode == NKR_NETMAP_ON;
557 /* return the next index, with wraparound */
558 static inline uint32_t
559 nm_next(uint32_t i, uint32_t lim)
561 return unlikely (i == lim) ? 0 : i + 1;
565 /* return the previous index, with wraparound */
566 static inline uint32_t
567 nm_prev(uint32_t i, uint32_t lim)
569 return unlikely (i == 0) ? lim : i - 1;
575 * Here is the layout for the Rx and Tx rings.
579 +-----------------+ +-----------------+
581 |XXX free slot XXX| |XXX free slot XXX|
582 +-----------------+ +-----------------+
583 head->| owned by user |<-hwcur | not sent to nic |<-hwcur
585 +-----------------+ | |
586 cur->| available to | | |
587 | user, not read | +-----------------+
588 | yet | cur->| (being |
591 +-----------------+ + ------ +
592 tail->| |<-hwtail | |<-hwlease
593 | (being | ... | | ...
594 | prepared) | ... | | ...
595 +-----------------+ ... | | ...
596 | |<-hwlease +-----------------+
597 | | tail->| |<-hwtail
601 +-----------------+ +-----------------+
603 * The cur/tail (user view) and hwcur/hwtail (kernel view)
604 * are used in the normal operation of the card.
606 * When a ring is the output of a switch port (Rx ring for
607 * a VALE port, Tx ring for the host stack or NIC), slots
608 * are reserved in blocks through 'hwlease' which points
609 * to the next unused slot.
610 * On an Rx ring, hwlease is always after hwtail,
611 * and completions cause hwtail to advance.
612 * On a Tx ring, hwlease is always between cur and hwtail,
613 * and completions cause cur to advance.
615 * nm_kr_space() returns the maximum number of slots that
617 * nm_kr_lease() reserves the required number of buffers,
618 * advances nkr_hwlease and also returns an entry in
619 * a circular array where completions should be reported.
624 struct lut_entry *lut;
625 uint32_t objtotal; /* max buffer index */
626 uint32_t objsize; /* buffer size */
629 struct netmap_vp_adapter; // forward
632 * The "struct netmap_adapter" extends the "struct adapter"
633 * (or equivalent) device descriptor.
634 * It contains all base fields needed to support netmap operation.
635 * There are in fact different types of netmap adapters
636 * (native, generic, VALE switch...) so a netmap_adapter is
637 * just the first field in the derived type.
639 struct netmap_adapter {
641 * On linux we do not have a good way to tell if an interface
642 * is netmap-capable. So we always use the following trick:
643 * NA(ifp) points here, and the first entry (which hopefully
644 * always exists and is at least 32 bits) contains a magic
645 * value which we can use to detect that the interface is good.
648 uint32_t na_flags; /* enabled, and other flags */
649 #define NAF_SKIP_INTR 1 /* use the regular interrupt handler.
650 * useful during initialization
652 #define NAF_SW_ONLY 2 /* forward packets only to sw adapter */
653 #define NAF_BDG_MAYSLEEP 4 /* the bridge is allowed to sleep when
654 * forwarding packets coming from this
657 #define NAF_MEM_OWNER 8 /* the adapter uses its own memory area
658 * that cannot be changed
660 #define NAF_NATIVE 16 /* the adapter is native.
661 * Virtual ports (non persistent vale ports,
662 * pipes, monitors...) should never use
665 #define NAF_NETMAP_ON 32 /* netmap is active (either native or
666 * emulated). Where possible (e.g. FreeBSD)
667 * IFCAP_NETMAP also mirrors this flag.
669 #define NAF_HOST_RINGS 64 /* the adapter supports the host rings */
670 #define NAF_FORCE_NATIVE 128 /* the adapter is always NATIVE */
671 #define NAF_PTNETMAP_HOST 256 /* the adapter supports ptnetmap in the host */
672 #define NAF_ZOMBIE (1U<<30) /* the nic driver has been unloaded */
673 #define NAF_BUSY (1U<<31) /* the adapter is used internally and
674 * cannot be registered from userspace
676 int active_fds; /* number of user-space descriptors using this
677 interface, which is equal to the number of
678 struct netmap_if objs in the mapped region. */
680 u_int num_rx_rings; /* number of adapter receive rings */
681 u_int num_tx_rings; /* number of adapter transmit rings */
683 u_int num_tx_desc; /* number of descriptor in each queue */
686 /* tx_rings and rx_rings are private but allocated
687 * as a contiguous chunk of memory. Each array has
688 * N+1 entries, for the adapter queues and for the host queue.
690 struct netmap_kring *tx_rings; /* array of TX rings. */
691 struct netmap_kring *rx_rings; /* array of RX rings. */
693 void *tailroom; /* space below the rings array */
694 /* (used for leases) */
697 NM_SELINFO_T si[NR_TXRX]; /* global wait queues */
699 /* count users of the global wait queues */
700 int si_users[NR_TXRX];
702 void *pdev; /* used to store pci device */
704 /* copy of if_qflush and if_transmit pointers, to intercept
705 * packets from the network stack when netmap is active.
707 int (*if_transmit)(struct ifnet *, struct mbuf *);
709 /* copy of if_input for netmap_send_up() */
710 void (*if_input)(struct ifnet *, struct mbuf *);
712 /* references to the ifnet and device routines, used by
713 * the generic netmap functions.
715 struct ifnet *ifp; /* adapter is ifp->if_softc */
717 /*---- callbacks for this netmap adapter -----*/
719 * nm_dtor() is the cleanup routine called when destroying
721 * Called with NMG_LOCK held.
723 * nm_register() is called on NIOCREGIF and close() to enter
724 * or exit netmap mode on the NIC
725 * Called with NNG_LOCK held.
727 * nm_txsync() pushes packets to the underlying hw/switch
729 * nm_rxsync() collects packets from the underlying hw/switch
731 * nm_config() returns configuration information from the OS
732 * Called with NMG_LOCK held.
734 * nm_krings_create() create and init the tx_rings and
735 * rx_rings arrays of kring structures. In particular,
736 * set the nm_sync callbacks for each ring.
737 * There is no need to also allocate the corresponding
738 * netmap_rings, since netmap_mem_rings_create() will always
739 * be called to provide the missing ones.
740 * Called with NNG_LOCK held.
742 * nm_krings_delete() cleanup and delete the tx_rings and rx_rings
744 * Called with NMG_LOCK held.
746 * nm_notify() is used to act after data have become available
747 * (or the stopped state of the ring has changed)
748 * For hw devices this is typically a selwakeup(),
749 * but for NIC/host ports attached to a switch (or vice-versa)
750 * we also need to invoke the 'txsync' code downstream.
751 * This callback pointer is actually used only to initialize
753 * Return values are the same as for netmap_rx_irq().
755 void (*nm_dtor)(struct netmap_adapter *);
757 int (*nm_register)(struct netmap_adapter *, int onoff);
758 void (*nm_intr)(struct netmap_adapter *, int onoff);
760 int (*nm_txsync)(struct netmap_kring *kring, int flags);
761 int (*nm_rxsync)(struct netmap_kring *kring, int flags);
762 int (*nm_notify)(struct netmap_kring *kring, int flags);
763 #define NAF_FORCE_READ 1
764 #define NAF_FORCE_RECLAIM 2
765 #define NAF_CAN_FORWARD_DOWN 4
766 /* return configuration information */
767 int (*nm_config)(struct netmap_adapter *,
768 u_int *txr, u_int *txd, u_int *rxr, u_int *rxd);
769 int (*nm_krings_create)(struct netmap_adapter *);
770 void (*nm_krings_delete)(struct netmap_adapter *);
773 * nm_bdg_attach() initializes the na_vp field to point
774 * to an adapter that can be attached to a VALE switch. If the
775 * current adapter is already a VALE port, na_vp is simply a cast;
776 * otherwise, na_vp points to a netmap_bwrap_adapter.
777 * If applicable, this callback also initializes na_hostvp,
778 * that can be used to connect the adapter host rings to the
780 * Called with NMG_LOCK held.
782 * nm_bdg_ctl() is called on the actual attach/detach to/from
783 * to/from the switch, to perform adapter-specific
785 * Called with NMG_LOCK held.
787 int (*nm_bdg_attach)(const char *bdg_name, struct netmap_adapter *);
788 int (*nm_bdg_ctl)(struct netmap_adapter *, struct nmreq *, int);
790 /* adapter used to attach this adapter to a VALE switch (if any) */
791 struct netmap_vp_adapter *na_vp;
792 /* adapter used to attach the host rings of this adapter
793 * to a VALE switch (if any) */
794 struct netmap_vp_adapter *na_hostvp;
797 /* standard refcount to control the lifetime of the adapter
798 * (it should be equal to the lifetime of the corresponding ifp)
802 /* memory allocator (opaque)
803 * We also cache a pointer to the lut_entry for translating
804 * buffer addresses, the total number of buffers and the buffer size.
806 struct netmap_mem_d *nm_mem;
807 struct netmap_lut na_lut;
809 /* additional information attached to this adapter
810 * by other netmap subsystems. Currently used by
811 * bwrap, LINUX/v1000 and ptnetmap
815 /* array of pipes that have this adapter as a parent */
816 struct netmap_pipe_adapter **na_pipes;
817 int na_next_pipe; /* next free slot in the array */
818 int na_max_pipes; /* size of the array */
820 /* Offset of ethernet header for each packet. */
826 static __inline u_int
827 nma_get_ndesc(struct netmap_adapter *na, enum txrx t)
829 return (t == NR_TX ? na->num_tx_desc : na->num_rx_desc);
833 nma_set_ndesc(struct netmap_adapter *na, enum txrx t, u_int v)
841 static __inline u_int
842 nma_get_nrings(struct netmap_adapter *na, enum txrx t)
844 return (t == NR_TX ? na->num_tx_rings : na->num_rx_rings);
848 nma_set_nrings(struct netmap_adapter *na, enum txrx t, u_int v)
851 na->num_tx_rings = v;
853 na->num_rx_rings = v;
856 static __inline struct netmap_kring*
857 NMR(struct netmap_adapter *na, enum txrx t)
859 return (t == NR_TX ? na->tx_rings : na->rx_rings);
863 * If the NIC is owned by the kernel
864 * (i.e., bridge), neither another bridge nor user can use it;
865 * if the NIC is owned by a user, only users can share it.
866 * Evaluation must be done under NMG_LOCK().
868 #define NETMAP_OWNED_BY_KERN(na) ((na)->na_flags & NAF_BUSY)
869 #define NETMAP_OWNED_BY_ANY(na) \
870 (NETMAP_OWNED_BY_KERN(na) || ((na)->active_fds > 0))
873 * derived netmap adapters for various types of ports
875 struct netmap_vp_adapter { /* VALE software port */
876 struct netmap_adapter up;
881 * bdg_port is the port number used in the bridge;
882 * na_bdg points to the bridge this NA is attached to.
885 struct nm_bridge *na_bdg;
887 int autodelete; /* remove the ifp on last reference */
889 /* Maximum Frame Size, used in bdg_mismatch_datapath() */
891 /* Last source MAC on this port */
896 struct netmap_hw_adapter { /* physical device */
897 struct netmap_adapter up;
899 struct net_device_ops nm_ndo; // XXX linux only
900 struct ethtool_ops nm_eto; // XXX linux only
901 const struct ethtool_ops* save_ethtool;
903 int (*nm_hw_register)(struct netmap_adapter *, int onoff);
907 /* Mitigation support. */
908 struct nm_generic_mit {
909 struct hrtimer mit_timer;
911 int mit_ring_idx; /* index of the ring being mitigated */
912 struct netmap_adapter *mit_na; /* backpointer */
915 struct netmap_generic_adapter { /* emulated device */
916 struct netmap_hw_adapter up;
918 /* Pointer to a previously used netmap adapter. */
919 struct netmap_adapter *prev;
921 /* generic netmap adapters support:
922 * a net_device_ops struct overrides ndo_select_queue(),
923 * save_if_input saves the if_input hook (FreeBSD),
924 * mit implements rx interrupt mitigation,
926 struct net_device_ops generic_ndo;
927 void (*save_if_input)(struct ifnet *, struct mbuf *);
929 struct nm_generic_mit *mit;
931 netdev_tx_t (*save_start_xmit)(struct mbuf *, struct ifnet *);
933 /* Is the adapter able to use multiple RX slots to scatter
934 * each packet pushed up by the driver? */
937 /* Is the transmission path controlled by a netmap-aware
938 * device queue (i.e. qdisc on linux)? */
941 #endif /* WITH_GENERIC */
944 netmap_real_rings(struct netmap_adapter *na, enum txrx t)
946 return nma_get_nrings(na, t) + !!(na->na_flags & NAF_HOST_RINGS);
950 struct nm_bdg_polling_state;
952 * Bridge wrapper for non VALE ports attached to a VALE switch.
954 * The real device must already have its own netmap adapter (hwna).
955 * The bridge wrapper and the hwna adapter share the same set of
956 * netmap rings and buffers, but they have two separate sets of
957 * krings descriptors, with tx/rx meanings swapped:
960 * bwrap krings rings krings hwna
961 * +------+ +------+ +-----+ +------+ +------+
962 * |tx_rings->| |\ /| |----| |<-tx_rings|
963 * | | +------+ \ / +-----+ +------+ | |
966 * | | +------+/ \+-----+ +------+ | |
967 * |rx_rings->| | | |----| |<-rx_rings|
968 * | | +------+ +-----+ +------+ | |
971 * - packets coming from the bridge go to the brwap rx rings,
972 * which are also the hwna tx rings. The bwrap notify callback
973 * will then complete the hwna tx (see netmap_bwrap_notify).
975 * - packets coming from the outside go to the hwna rx rings,
976 * which are also the bwrap tx rings. The (overwritten) hwna
977 * notify method will then complete the bridge tx
978 * (see netmap_bwrap_intr_notify).
980 * The bridge wrapper may optionally connect the hwna 'host' rings
981 * to the bridge. This is done by using a second port in the
982 * bridge and connecting it to the 'host' netmap_vp_adapter
983 * contained in the netmap_bwrap_adapter. The brwap host adapter
984 * cross-links the hwna host rings in the same way as shown above.
986 * - packets coming from the bridge and directed to the host stack
987 * are handled by the bwrap host notify callback
988 * (see netmap_bwrap_host_notify)
990 * - packets coming from the host stack are still handled by the
991 * overwritten hwna notify callback (netmap_bwrap_intr_notify),
992 * but are diverted to the host adapter depending on the ring number.
995 struct netmap_bwrap_adapter {
996 struct netmap_vp_adapter up;
997 struct netmap_vp_adapter host; /* for host rings */
998 struct netmap_adapter *hwna; /* the underlying device */
1001 * When we attach a physical interface to the bridge, we
1002 * allow the controlling process to terminate, so we need
1003 * a place to store the n_detmap_priv_d data structure.
1004 * This is only done when physical interfaces
1005 * are attached to a bridge.
1007 struct netmap_priv_d *na_kpriv;
1008 struct nm_bdg_polling_state *na_polling_state;
1010 int netmap_bwrap_attach(const char *name, struct netmap_adapter *);
1011 int netmap_vi_create(struct nmreq *, int);
1013 #else /* !WITH_VALE */
1014 #define netmap_vi_create(nmr, a) (EOPNOTSUPP)
1015 #endif /* WITH_VALE */
1019 #define NM_MAXPIPES 64 /* max number of pipes per adapter */
1021 struct netmap_pipe_adapter {
1022 struct netmap_adapter up;
1024 u_int id; /* pipe identifier */
1025 int role; /* either NR_REG_PIPE_MASTER or NR_REG_PIPE_SLAVE */
1027 struct netmap_adapter *parent; /* adapter that owns the memory */
1028 struct netmap_pipe_adapter *peer; /* the other end of the pipe */
1029 int peer_ref; /* 1 iff we are holding a ref to the peer */
1030 struct ifnet *parent_ifp; /* maybe null */
1032 u_int parent_slot; /* index in the parent pipe array */
1035 #endif /* WITH_PIPES */
1038 /* return slots reserved to rx clients; used in drivers */
1039 static inline uint32_t
1040 nm_kr_rxspace(struct netmap_kring *k)
1042 int space = k->nr_hwtail - k->nr_hwcur;
1044 space += k->nkr_num_slots;
1045 ND("preserving %d rx slots %d -> %d", space, k->nr_hwcur, k->nr_hwtail);
1050 /* return slots reserved to tx clients */
1051 #define nm_kr_txspace(_k) nm_kr_rxspace(_k)
1054 /* True if no space in the tx ring, only valid after txsync_prologue */
1056 nm_kr_txempty(struct netmap_kring *kring)
1058 return kring->rcur == kring->nr_hwtail;
1061 /* True if no more completed slots in the rx ring, only valid after
1062 * rxsync_prologue */
1063 #define nm_kr_rxempty(_k) nm_kr_txempty(_k)
1066 * protect against multiple threads using the same ring.
1067 * also check that the ring has not been stopped or locked
1069 #define NM_KR_BUSY 1 /* some other thread is syncing the ring */
1070 #define NM_KR_STOPPED 2 /* unbounded stop (ifconfig down or driver unload) */
1071 #define NM_KR_LOCKED 3 /* bounded, brief stop for mutual exclusion */
1074 /* release the previously acquired right to use the *sync() methods of the ring */
1075 static __inline void nm_kr_put(struct netmap_kring *kr)
1077 NM_ATOMIC_CLEAR(&kr->nr_busy);
1081 /* true if the ifp that backed the adapter has disappeared (e.g., the
1082 * driver has been unloaded)
1084 static inline int nm_iszombie(struct netmap_adapter *na);
1086 /* try to obtain exclusive right to issue the *sync() operations on the ring.
1087 * The right is obtained and must be later relinquished via nm_kr_put() if and
1088 * only if nm_kr_tryget() returns 0.
1089 * If can_sleep is 1 there are only two other possible outcomes:
1090 * - the function returns NM_KR_BUSY
1091 * - the function returns NM_KR_STOPPED and sets the POLLERR bit in *perr
1093 * In both cases the caller will typically skip the ring, possibly collecting
1094 * errors along the way.
1095 * If the calling context does not allow sleeping, the caller must pass 0 in can_sleep.
1096 * In the latter case, the function may also return NM_KR_LOCKED and leave *perr
1097 * untouched: ideally, the caller should try again at a later time.
1099 static __inline int nm_kr_tryget(struct netmap_kring *kr, int can_sleep, int *perr)
1101 int busy = 1, stopped;
1102 /* check a first time without taking the lock
1103 * to avoid starvation for nm_kr_get()
1106 stopped = kr->nkr_stopped;
1107 if (unlikely(stopped)) {
1110 busy = NM_ATOMIC_TEST_AND_SET(&kr->nr_busy);
1111 /* we should not return NM_KR_BUSY if the ring was
1112 * actually stopped, so check another time after
1113 * the barrier provided by the atomic operation
1115 stopped = kr->nkr_stopped;
1116 if (unlikely(stopped)) {
1120 if (unlikely(nm_iszombie(kr->na))) {
1121 stopped = NM_KR_STOPPED;
1125 return unlikely(busy) ? NM_KR_BUSY : 0;
1130 if (stopped == NM_KR_STOPPED) {
1131 /* if POLLERR is defined we want to use it to simplify netmap_poll().
1132 * Otherwise, any non-zero value will do.
1135 #define NM_POLLERR POLLERR
1137 #define NM_POLLERR 1
1138 #endif /* POLLERR */
1140 *perr |= NM_POLLERR;
1142 } else if (can_sleep) {
1143 tsleep(kr, 0, "NM_KR_TRYGET", 4);
1149 /* put the ring in the 'stopped' state and wait for the current user (if any) to
1150 * notice. stopped must be either NM_KR_STOPPED or NM_KR_LOCKED
1152 static __inline void nm_kr_stop(struct netmap_kring *kr, int stopped)
1154 kr->nkr_stopped = stopped;
1155 while (NM_ATOMIC_TEST_AND_SET(&kr->nr_busy))
1156 tsleep(kr, 0, "NM_KR_GET", 4);
1159 /* restart a ring after a stop */
1160 static __inline void nm_kr_start(struct netmap_kring *kr)
1162 kr->nkr_stopped = 0;
1168 * The following functions are used by individual drivers to
1169 * support netmap operation.
1171 * netmap_attach() initializes a struct netmap_adapter, allocating the
1172 * struct netmap_ring's and the struct selinfo.
1174 * netmap_detach() frees the memory allocated by netmap_attach().
1176 * netmap_transmit() replaces the if_transmit routine of the interface,
1177 * and is used to intercept packets coming from the stack.
1179 * netmap_load_map/netmap_reload_map are helper routines to set/reset
1180 * the dmamap for a packet buffer
1182 * netmap_reset() is a helper routine to be called in the hw driver
1183 * when reinitializing a ring. It should not be called by
1184 * virtual ports (vale, pipes, monitor)
1186 int netmap_attach(struct netmap_adapter *);
1187 int netmap_attach_ext(struct netmap_adapter *, size_t size);
1188 void netmap_detach(struct ifnet *);
1189 int netmap_transmit(struct ifnet *, struct mbuf *);
1190 struct netmap_slot *netmap_reset(struct netmap_adapter *na,
1191 enum txrx tx, u_int n, u_int new_cur);
1192 int netmap_ring_reinit(struct netmap_kring *);
1194 /* Return codes for netmap_*x_irq. */
1196 /* Driver should do normal interrupt processing, e.g. because
1197 * the interface is not in netmap mode. */
1199 /* Port is in netmap mode, and the interrupt work has been
1200 * completed. The driver does not have to notify netmap
1201 * again before the next interrupt. */
1202 NM_IRQ_COMPLETED = -1,
1203 /* Port is in netmap mode, but the interrupt work has not been
1204 * completed. The driver has to make sure netmap will be
1205 * notified again soon, even if no more interrupts come (e.g.
1206 * on Linux the driver should not call napi_complete()). */
1207 NM_IRQ_RESCHED = -2,
1210 /* default functions to handle rx/tx interrupts */
1211 int netmap_rx_irq(struct ifnet *, u_int, u_int *);
1212 #define netmap_tx_irq(_n, _q) netmap_rx_irq(_n, _q, NULL)
1213 int netmap_common_irq(struct netmap_adapter *, u_int, u_int *work_done);
1217 /* functions used by external modules to interface with VALE */
1218 #define netmap_vp_to_ifp(_vp) ((_vp)->up.ifp)
1219 #define netmap_ifp_to_vp(_ifp) (NA(_ifp)->na_vp)
1220 #define netmap_ifp_to_host_vp(_ifp) (NA(_ifp)->na_hostvp)
1221 #define netmap_bdg_idx(_vp) ((_vp)->bdg_port)
1222 const char *netmap_bdg_name(struct netmap_vp_adapter *);
1223 #else /* !WITH_VALE */
1224 #define netmap_vp_to_ifp(_vp) NULL
1225 #define netmap_ifp_to_vp(_ifp) NULL
1226 #define netmap_ifp_to_host_vp(_ifp) NULL
1227 #define netmap_bdg_idx(_vp) -1
1228 #define netmap_bdg_name(_vp) NULL
1229 #endif /* WITH_VALE */
1232 nm_netmap_on(struct netmap_adapter *na)
1234 return na && na->na_flags & NAF_NETMAP_ON;
1238 nm_native_on(struct netmap_adapter *na)
1240 return nm_netmap_on(na) && (na->na_flags & NAF_NATIVE);
1244 nm_iszombie(struct netmap_adapter *na)
1246 return na == NULL || (na->na_flags & NAF_ZOMBIE);
1250 nm_update_hostrings_mode(struct netmap_adapter *na)
1252 /* Process nr_mode and nr_pending_mode for host rings. */
1253 na->tx_rings[na->num_tx_rings].nr_mode =
1254 na->tx_rings[na->num_tx_rings].nr_pending_mode;
1255 na->rx_rings[na->num_rx_rings].nr_mode =
1256 na->rx_rings[na->num_rx_rings].nr_pending_mode;
1259 /* set/clear native flags and if_transmit/netdev_ops */
1261 nm_set_native_flags(struct netmap_adapter *na)
1263 struct ifnet *ifp = na->ifp;
1265 /* We do the setup for intercepting packets only if we are the
1266 * first user of this adapapter. */
1267 if (na->active_fds > 0) {
1271 na->na_flags |= NAF_NETMAP_ON;
1272 #ifdef IFCAP_NETMAP /* or FreeBSD ? */
1273 ifp->if_capenable |= IFCAP_NETMAP;
1275 #if defined (__FreeBSD__)
1276 na->if_transmit = ifp->if_transmit;
1277 ifp->if_transmit = netmap_transmit;
1278 #elif defined (_WIN32)
1279 (void)ifp; /* prevent a warning */
1280 //XXX_ale can we just comment those?
1281 //na->if_transmit = ifp->if_transmit;
1282 //ifp->if_transmit = netmap_transmit;
1284 na->if_transmit = (void *)ifp->netdev_ops;
1285 ifp->netdev_ops = &((struct netmap_hw_adapter *)na)->nm_ndo;
1286 ((struct netmap_hw_adapter *)na)->save_ethtool = ifp->ethtool_ops;
1287 ifp->ethtool_ops = &((struct netmap_hw_adapter*)na)->nm_eto;
1289 nm_update_hostrings_mode(na);
1293 nm_clear_native_flags(struct netmap_adapter *na)
1295 struct ifnet *ifp = na->ifp;
1297 /* We undo the setup for intercepting packets only if we are the
1298 * last user of this adapapter. */
1299 if (na->active_fds > 0) {
1303 nm_update_hostrings_mode(na);
1305 #if defined(__FreeBSD__)
1306 ifp->if_transmit = na->if_transmit;
1307 #elif defined(_WIN32)
1308 (void)ifp; /* prevent a warning */
1309 //XXX_ale can we just comment those?
1310 //ifp->if_transmit = na->if_transmit;
1312 ifp->netdev_ops = (void *)na->if_transmit;
1313 ifp->ethtool_ops = ((struct netmap_hw_adapter*)na)->save_ethtool;
1315 na->na_flags &= ~NAF_NETMAP_ON;
1316 #ifdef IFCAP_NETMAP /* or FreeBSD ? */
1317 ifp->if_capenable &= ~IFCAP_NETMAP;
1322 * nm_*sync_prologue() functions are used in ioctl/poll and ptnetmap
1324 * We need netmap_ring* parameter, because in ptnetmap it is decoupled
1326 * The user-space ring pointers (head/cur/tail) are shared through
1327 * CSB between host and guest.
1331 * validates parameters in the ring/kring, returns a value for head
1332 * If any error, returns ring_size to force a reinit.
1334 uint32_t nm_txsync_prologue(struct netmap_kring *, struct netmap_ring *);
1338 * validates parameters in the ring/kring, returns a value for head
1339 * If any error, returns ring_size lim to force a reinit.
1341 uint32_t nm_rxsync_prologue(struct netmap_kring *, struct netmap_ring *);
1344 /* check/fix address and len in tx rings */
1345 #if 1 /* debug version */
1346 #define NM_CHECK_ADDR_LEN(_na, _a, _l) do { \
1347 if (_a == NETMAP_BUF_BASE(_na) || _l > NETMAP_BUF_SIZE(_na)) { \
1348 RD(5, "bad addr/len ring %d slot %d idx %d len %d", \
1349 kring->ring_id, nm_i, slot->buf_idx, len); \
1350 if (_l > NETMAP_BUF_SIZE(_na)) \
1351 _l = NETMAP_BUF_SIZE(_na); \
1353 #else /* no debug version */
1354 #define NM_CHECK_ADDR_LEN(_na, _a, _l) do { \
1355 if (_l > NETMAP_BUF_SIZE(_na)) \
1356 _l = NETMAP_BUF_SIZE(_na); \
1361 /*---------------------------------------------------------------*/
1363 * Support routines used by netmap subsystems
1364 * (native drivers, VALE, generic, pipes, monitors, ...)
1368 /* common routine for all functions that create a netmap adapter. It performs
1370 * - if the na points to an ifp, mark the ifp as netmap capable
1371 * using na as its native adapter;
1372 * - provide defaults for the setup callbacks and the memory allocator
1374 int netmap_attach_common(struct netmap_adapter *);
1375 /* common actions to be performed on netmap adapter destruction */
1376 void netmap_detach_common(struct netmap_adapter *);
1377 /* fill priv->np_[tr]xq{first,last} using the ringid and flags information
1378 * coming from a struct nmreq
1380 int netmap_interp_ringid(struct netmap_priv_d *priv, uint16_t ringid, uint32_t flags);
1381 /* update the ring parameters (number and size of tx and rx rings).
1382 * It calls the nm_config callback, if available.
1384 int netmap_update_config(struct netmap_adapter *na);
1385 /* create and initialize the common fields of the krings array.
1386 * using the information that must be already available in the na.
1387 * tailroom can be used to request the allocation of additional
1388 * tailroom bytes after the krings array. This is used by
1389 * netmap_vp_adapter's (i.e., VALE ports) to make room for
1390 * leasing-related data structures
1392 int netmap_krings_create(struct netmap_adapter *na, u_int tailroom);
1393 /* deletes the kring array of the adapter. The array must have
1394 * been created using netmap_krings_create
1396 void netmap_krings_delete(struct netmap_adapter *na);
1398 int netmap_hw_krings_create(struct netmap_adapter *na);
1399 void netmap_hw_krings_delete(struct netmap_adapter *na);
1401 /* set the stopped/enabled status of ring
1402 * When stopping, they also wait for all current activity on the ring to
1403 * terminate. The status change is then notified using the na nm_notify
1406 void netmap_set_ring(struct netmap_adapter *, u_int ring_id, enum txrx, int stopped);
1407 /* set the stopped/enabled status of all rings of the adapter. */
1408 void netmap_set_all_rings(struct netmap_adapter *, int stopped);
1409 /* convenience wrappers for netmap_set_all_rings */
1410 void netmap_disable_all_rings(struct ifnet *);
1411 void netmap_enable_all_rings(struct ifnet *);
1413 int netmap_do_regif(struct netmap_priv_d *priv, struct netmap_adapter *na,
1414 uint16_t ringid, uint32_t flags);
1415 void netmap_do_unregif(struct netmap_priv_d *priv);
1417 u_int nm_bound_var(u_int *v, u_int dflt, u_int lo, u_int hi, const char *msg);
1418 int netmap_get_na(struct nmreq *nmr, struct netmap_adapter **na,
1419 struct ifnet **ifp, struct netmap_mem_d *nmd, int create);
1420 void netmap_unget_na(struct netmap_adapter *na, struct ifnet *ifp);
1421 int netmap_get_hw_na(struct ifnet *ifp,
1422 struct netmap_mem_d *nmd, struct netmap_adapter **na);
1427 * The following bridge-related functions are used by other
1430 * VALE only supports unicast or broadcast. The lookup
1431 * function can return 0 .. NM_BDG_MAXPORTS-1 for regular ports,
1432 * NM_BDG_MAXPORTS for broadcast, NM_BDG_MAXPORTS+1 for unknown.
1433 * XXX in practice "unknown" might be handled same as broadcast.
1435 typedef u_int (*bdg_lookup_fn_t)(struct nm_bdg_fwd *ft, uint8_t *ring_nr,
1436 struct netmap_vp_adapter *);
1437 typedef int (*bdg_config_fn_t)(struct nm_ifreq *);
1438 typedef void (*bdg_dtor_fn_t)(const struct netmap_vp_adapter *);
1439 struct netmap_bdg_ops {
1440 bdg_lookup_fn_t lookup;
1441 bdg_config_fn_t config;
1445 u_int netmap_bdg_learning(struct nm_bdg_fwd *ft, uint8_t *dst_ring,
1446 struct netmap_vp_adapter *);
1448 #define NM_BRIDGES 8 /* number of bridges */
1449 #define NM_BDG_MAXPORTS 254 /* up to 254 */
1450 #define NM_BDG_BROADCAST NM_BDG_MAXPORTS
1451 #define NM_BDG_NOPORT (NM_BDG_MAXPORTS+1)
1453 /* these are redefined in case of no VALE support */
1454 int netmap_get_bdg_na(struct nmreq *nmr, struct netmap_adapter **na,
1455 struct netmap_mem_d *nmd, int create);
1456 struct nm_bridge *netmap_init_bridges2(u_int);
1457 void netmap_uninit_bridges2(struct nm_bridge *, u_int);
1458 int netmap_init_bridges(void);
1459 void netmap_uninit_bridges(void);
1460 int netmap_bdg_ctl(struct nmreq *nmr, struct netmap_bdg_ops *bdg_ops);
1461 int netmap_bdg_config(struct nmreq *nmr);
1463 #else /* !WITH_VALE */
1464 #define netmap_get_bdg_na(_1, _2, _3, _4) 0
1465 #define netmap_init_bridges(_1) 0
1466 #define netmap_uninit_bridges()
1467 #define netmap_bdg_ctl(_1, _2) EINVAL
1468 #endif /* !WITH_VALE */
1471 /* max number of pipes per device */
1472 #define NM_MAXPIPES 64 /* XXX how many? */
1473 void netmap_pipe_dealloc(struct netmap_adapter *);
1474 int netmap_get_pipe_na(struct nmreq *nmr, struct netmap_adapter **na,
1475 struct netmap_mem_d *nmd, int create);
1476 #else /* !WITH_PIPES */
1477 #define NM_MAXPIPES 0
1478 #define netmap_pipe_alloc(_1, _2) 0
1479 #define netmap_pipe_dealloc(_1)
1480 #define netmap_get_pipe_na(nmr, _2, _3, _4) \
1481 ({ int role__ = (nmr)->nr_flags & NR_REG_MASK; \
1482 (role__ == NR_REG_PIPE_MASTER || \
1483 role__ == NR_REG_PIPE_SLAVE) ? EOPNOTSUPP : 0; })
1487 int netmap_get_monitor_na(struct nmreq *nmr, struct netmap_adapter **na,
1488 struct netmap_mem_d *nmd, int create);
1489 void netmap_monitor_stop(struct netmap_adapter *na);
1491 #define netmap_get_monitor_na(nmr, _2, _3, _4) \
1492 ((nmr)->nr_flags & (NR_MONITOR_TX | NR_MONITOR_RX) ? EOPNOTSUPP : 0)
1495 #ifdef CONFIG_NET_NS
1496 struct net *netmap_bns_get(void);
1497 void netmap_bns_put(struct net *);
1498 void netmap_bns_getbridges(struct nm_bridge **, u_int *);
1500 #define netmap_bns_get()
1501 #define netmap_bns_put(_1)
1502 #define netmap_bns_getbridges(b, n) \
1503 do { *b = nm_bridges; *n = NM_BRIDGES; } while (0)
1506 /* Various prototypes */
1507 int netmap_poll(struct netmap_priv_d *, int events, NM_SELRECORD_T *td);
1508 int netmap_init(void);
1509 void netmap_fini(void);
1510 int netmap_get_memory(struct netmap_priv_d* p);
1511 void netmap_dtor(void *data);
1513 int netmap_ioctl(struct netmap_priv_d *priv, u_long cmd, caddr_t data, struct thread *);
1515 /* netmap_adapter creation/destruction */
1517 // #define NM_DEBUG_PUTGET 1
1519 #ifdef NM_DEBUG_PUTGET
1521 #define NM_DBG(f) __##f
1523 void __netmap_adapter_get(struct netmap_adapter *na);
1525 #define netmap_adapter_get(na) \
1527 struct netmap_adapter *__na = na; \
1528 D("getting %p:%s (%d)", __na, (__na)->name, (__na)->na_refcount); \
1529 __netmap_adapter_get(__na); \
1532 int __netmap_adapter_put(struct netmap_adapter *na);
1534 #define netmap_adapter_put(na) \
1536 struct netmap_adapter *__na = na; \
1537 D("putting %p:%s (%d)", __na, (__na)->name, (__na)->na_refcount); \
1538 __netmap_adapter_put(__na); \
1541 #else /* !NM_DEBUG_PUTGET */
1544 void netmap_adapter_get(struct netmap_adapter *na);
1545 int netmap_adapter_put(struct netmap_adapter *na);
1547 #endif /* !NM_DEBUG_PUTGET */
1553 #define NETMAP_BUF_BASE(_na) ((_na)->na_lut.lut[0].vaddr)
1554 #define NETMAP_BUF_SIZE(_na) ((_na)->na_lut.objsize)
1555 extern int netmap_no_pendintr;
1556 extern int netmap_mitigate;
1557 extern int netmap_verbose; /* for debugging */
1558 enum { /* verbose flags */
1559 NM_VERB_ON = 1, /* generic verbose */
1560 NM_VERB_HOST = 0x2, /* verbose host stack */
1561 NM_VERB_RXSYNC = 0x10, /* verbose on rxsync/txsync */
1562 NM_VERB_TXSYNC = 0x20,
1563 NM_VERB_RXINTR = 0x100, /* verbose on rx/tx intr (driver) */
1564 NM_VERB_TXINTR = 0x200,
1565 NM_VERB_NIC_RXSYNC = 0x1000, /* verbose on rx/tx intr (driver) */
1566 NM_VERB_NIC_TXSYNC = 0x2000,
1569 extern int netmap_txsync_retry;
1570 extern int netmap_flags;
1571 extern int netmap_generic_mit;
1572 extern int netmap_generic_ringsize;
1573 extern int netmap_generic_rings;
1574 extern int netmap_generic_txqdisc;
1575 extern int ptnetmap_tx_workers;
1578 * NA returns a pointer to the struct netmap adapter from the ifp,
1579 * WNA is used to write it.
1581 #define NA(_ifp) ((struct netmap_adapter *)WNA(_ifp))
1584 * On old versions of FreeBSD, NA(ifp) is a pspare. On linux we
1585 * overload another pointer in the netdev.
1587 * We check if NA(ifp) is set and its first element has a related
1588 * magic value. The capenable is within the struct netmap_adapter.
1590 #define NETMAP_MAGIC 0x52697a7a
1592 #define NM_NA_VALID(ifp) (NA(ifp) && \
1593 ((uint32_t)(uintptr_t)NA(ifp) ^ NA(ifp)->magic) == NETMAP_MAGIC )
1595 #define NM_ATTACH_NA(ifp, na) do { \
1599 ((uint32_t)(uintptr_t)NA(ifp)) ^ NETMAP_MAGIC; \
1602 #define NM_IS_NATIVE(ifp) (NM_NA_VALID(ifp) && NA(ifp)->nm_dtor == netmap_hw_dtor)
1604 #if defined(__FreeBSD__)
1606 /* Assigns the device IOMMU domain to an allocator.
1607 * Returns -ENOMEM in case the domain is different */
1608 #define nm_iommu_group_id(dev) (0)
1610 /* Callback invoked by the dma machinery after a successful dmamap_load */
1611 static void netmap_dmamap_cb(__unused void *arg,
1612 __unused bus_dma_segment_t * segs, __unused int nseg, __unused int error)
1616 /* bus_dmamap_load wrapper: call aforementioned function if map != NULL.
1617 * XXX can we do it without a callback ?
1620 netmap_load_map(struct netmap_adapter *na,
1621 bus_dma_tag_t tag, bus_dmamap_t map, void *buf)
1624 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE(na),
1625 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT);
1629 netmap_unload_map(struct netmap_adapter *na,
1630 bus_dma_tag_t tag, bus_dmamap_t map)
1633 bus_dmamap_unload(tag, map);
1636 /* update the map when a buffer changes. */
1638 netmap_reload_map(struct netmap_adapter *na,
1639 bus_dma_tag_t tag, bus_dmamap_t map, void *buf)
1642 bus_dmamap_unload(tag, map);
1643 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE(na),
1644 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT);
1648 #elif defined(_WIN32)
1652 int nm_iommu_group_id(bus_dma_tag_t dev);
1653 #include <linux/dma-mapping.h>
1656 netmap_load_map(struct netmap_adapter *na,
1657 bus_dma_tag_t tag, bus_dmamap_t map, void *buf)
1660 *map = dma_map_single(na->pdev, buf, NETMAP_BUF_SIZE(na),
1666 netmap_unload_map(struct netmap_adapter *na,
1667 bus_dma_tag_t tag, bus_dmamap_t map)
1669 u_int sz = NETMAP_BUF_SIZE(na);
1672 dma_unmap_single(na->pdev, *map, sz,
1678 netmap_reload_map(struct netmap_adapter *na,
1679 bus_dma_tag_t tag, bus_dmamap_t map, void *buf)
1681 u_int sz = NETMAP_BUF_SIZE(na);
1684 dma_unmap_single(na->pdev, *map, sz,
1688 *map = dma_map_single(na->pdev, buf, sz,
1693 * XXX How do we redefine these functions:
1696 * dma_map_single(&pdev->dev, virt_addr, len, direction)
1697 * dma_unmap_single(&adapter->pdev->dev, phys_addr, len, direction
1698 * The len can be implicit (on netmap it is NETMAP_BUF_SIZE)
1699 * unfortunately the direction is not, so we need to change
1700 * something to have a cross API
1704 struct e1000_buffer *buffer_info = &tx_ring->buffer_info[l];
1705 /* set time_stamp *before* dma to help avoid a possible race */
1706 buffer_info->time_stamp = jiffies;
1707 buffer_info->mapped_as_page = false;
1708 buffer_info->length = len;
1709 //buffer_info->next_to_watch = l;
1710 /* reload dma map */
1711 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1712 NETMAP_BUF_SIZE, DMA_TO_DEVICE);
1713 buffer_info->dma = dma_map_single(&adapter->pdev->dev,
1714 addr, NETMAP_BUF_SIZE, DMA_TO_DEVICE);
1716 if (dma_mapping_error(&adapter->pdev->dev, buffer_info->dma)) {
1717 D("dma mapping error");
1718 /* goto dma_error; See e1000_put_txbuf() */
1721 tx_desc->buffer_addr = htole64(buffer_info->dma); //XXX
1726 * The bus_dmamap_sync() can be one of wmb() or rmb() depending on direction.
1728 #define bus_dmamap_sync(_a, _b, _c)
1734 * functions to map NIC to KRING indexes (n2k) and vice versa (k2n)
1737 netmap_idx_n2k(struct netmap_kring *kr, int idx)
1739 int n = kr->nkr_num_slots;
1740 idx += kr->nkr_hwofs;
1751 netmap_idx_k2n(struct netmap_kring *kr, int idx)
1753 int n = kr->nkr_num_slots;
1754 idx -= kr->nkr_hwofs;
1764 /* Entries of the look-up table. */
1766 void *vaddr; /* virtual address. */
1767 vm_paddr_t paddr; /* physical address. */
1770 struct netmap_obj_pool;
1773 * NMB return the virtual address of a buffer (buffer 0 on bad index)
1774 * PNMB also fills the physical address
1776 static inline void *
1777 NMB(struct netmap_adapter *na, struct netmap_slot *slot)
1779 struct lut_entry *lut = na->na_lut.lut;
1780 uint32_t i = slot->buf_idx;
1781 return (unlikely(i >= na->na_lut.objtotal)) ?
1782 lut[0].vaddr : lut[i].vaddr;
1785 static inline void *
1786 PNMB(struct netmap_adapter *na, struct netmap_slot *slot, uint64_t *pp)
1788 uint32_t i = slot->buf_idx;
1789 struct lut_entry *lut = na->na_lut.lut;
1790 void *ret = (i >= na->na_lut.objtotal) ? lut[0].vaddr : lut[i].vaddr;
1793 *pp = (i >= na->na_lut.objtotal) ? lut[0].paddr : lut[i].paddr;
1795 *pp = (i >= na->na_lut.objtotal) ? (uint64_t)lut[0].paddr.QuadPart : (uint64_t)lut[i].paddr.QuadPart;
1802 * Structure associated to each netmap file descriptor.
1803 * It is created on open and left unbound (np_nifp == NULL).
1804 * A successful NIOCREGIF will set np_nifp and the first few fields;
1805 * this is protected by a global lock (NMG_LOCK) due to low contention.
1807 * np_refs counts the number of references to the structure: one for the fd,
1808 * plus (on FreeBSD) one for each active mmap which we track ourselves
1809 * (linux automatically tracks them, but FreeBSD does not).
1810 * np_refs is protected by NMG_LOCK.
1812 * Read access to the structure is lock free, because ni_nifp once set
1813 * can only go to 0 when nobody is using the entry anymore. Readers
1814 * must check that np_nifp != NULL before using the other fields.
1816 struct netmap_priv_d {
1817 struct netmap_if * volatile np_nifp; /* netmap if descriptor. */
1819 struct netmap_adapter *np_na;
1820 struct ifnet *np_ifp;
1821 uint32_t np_flags; /* from the ioctl */
1822 u_int np_qfirst[NR_TXRX],
1823 np_qlast[NR_TXRX]; /* range of tx/rx rings to scan */
1824 uint16_t np_txpoll; /* XXX and also np_rxpoll ? */
1825 int np_sync_flags; /* to be passed to nm_sync */
1827 int np_refs; /* use with NMG_LOCK held */
1829 /* pointers to the selinfo to be used for selrecord.
1830 * Either the local or the global one depending on the
1833 NM_SELINFO_T *np_si[NR_TXRX];
1834 struct thread *np_td; /* kqueue, just debugging */
1837 struct netmap_priv_d *netmap_priv_new(void);
1838 void netmap_priv_delete(struct netmap_priv_d *);
1840 static inline int nm_kring_pending(struct netmap_priv_d *np)
1842 struct netmap_adapter *na = np->np_na;
1847 for (i = np->np_qfirst[t]; i < np->np_qlast[t]; i++) {
1848 struct netmap_kring *kring = &NMR(na, t)[i];
1849 if (kring->nr_mode != kring->nr_pending_mode) {
1858 int netmap_pipe_txsync(struct netmap_kring *txkring, int flags);
1859 int netmap_pipe_rxsync(struct netmap_kring *rxkring, int flags);
1860 #endif /* WITH_PIPES */
1864 struct netmap_monitor_adapter {
1865 struct netmap_adapter up;
1867 struct netmap_priv_d priv;
1871 #endif /* WITH_MONITOR */
1876 * generic netmap emulation for devices that do not have
1877 * native netmap support.
1879 int generic_netmap_attach(struct ifnet *ifp);
1880 int generic_rx_handler(struct ifnet *ifp, struct mbuf *m);;
1882 int nm_os_catch_rx(struct netmap_generic_adapter *gna, int intercept);
1883 int nm_os_catch_tx(struct netmap_generic_adapter *gna, int intercept);
1885 int na_is_generic(struct netmap_adapter *na);
1888 * the generic transmit routine is passed a structure to optionally
1889 * build a queue of descriptors, in an OS-specific way.
1890 * The payload is at addr, if non-null, and the routine should send or queue
1891 * the packet, returning 0 if successful, 1 on failure.
1893 * At the end, if head is non-null, there will be an additional call
1894 * to the function with addr = NULL; this should tell the OS-specific
1895 * routine to send the queue and free any resources. Failure is ignored.
1897 struct nm_os_gen_arg {
1899 void *m; /* os-specific mbuf-like object */
1900 void *head, *tail; /* tailq, if the OS-specific routine needs to build one */
1901 void *addr; /* payload of current packet */
1902 u_int len; /* packet length */
1903 u_int ring_nr; /* packet length */
1904 u_int qevent; /* in txqdisc mode, place an event on this mbuf */
1907 int nm_os_generic_xmit_frame(struct nm_os_gen_arg *);
1908 int nm_os_generic_find_num_desc(struct ifnet *ifp, u_int *tx, u_int *rx);
1909 void nm_os_generic_find_num_queues(struct ifnet *ifp, u_int *txq, u_int *rxq);
1910 void nm_os_generic_set_features(struct netmap_generic_adapter *gna);
1912 static inline struct ifnet*
1913 netmap_generic_getifp(struct netmap_generic_adapter *gna)
1916 return gna->prev->ifp;
1918 return gna->up.up.ifp;
1921 void netmap_generic_irq(struct netmap_adapter *na, u_int q, u_int *work_done);
1923 //#define RATE_GENERIC /* Enables communication statistics for generic. */
1925 void generic_rate(int txp, int txs, int txi, int rxp, int rxs, int rxi);
1927 #define generic_rate(txp, txs, txi, rxp, rxs, rxi)
1931 * netmap_mitigation API. This is used by the generic adapter
1932 * to reduce the number of interrupt requests/selwakeup
1933 * to clients on incoming packets.
1935 void nm_os_mitigation_init(struct nm_generic_mit *mit, int idx,
1936 struct netmap_adapter *na);
1937 void nm_os_mitigation_start(struct nm_generic_mit *mit);
1938 void nm_os_mitigation_restart(struct nm_generic_mit *mit);
1939 int nm_os_mitigation_active(struct nm_generic_mit *mit);
1940 void nm_os_mitigation_cleanup(struct nm_generic_mit *mit);
1941 #else /* !WITH_GENERIC */
1942 #define generic_netmap_attach(ifp) (EOPNOTSUPP)
1943 #define na_is_generic(na) (0)
1944 #endif /* WITH_GENERIC */
1946 /* Shared declarations for the VALE switch. */
1949 * Each transmit queue accumulates a batch of packets into
1950 * a structure before forwarding. Packets to the same
1951 * destination are put in a list using ft_next as a link field.
1952 * ft_frags and ft_next are valid only on the first fragment.
1954 struct nm_bdg_fwd { /* forwarding entry for a bridge */
1955 void *ft_buf; /* netmap or indirect buffer */
1956 uint8_t ft_frags; /* how many fragments (only on 1st frag) */
1957 uint8_t _ft_port; /* dst port (unused) */
1958 uint16_t ft_flags; /* flags, e.g. indirect */
1959 uint16_t ft_len; /* src fragment len */
1960 uint16_t ft_next; /* next packet to same destination */
1963 /* struct 'virtio_net_hdr' from linux. */
1964 struct nm_vnet_hdr {
1965 #define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /* Use csum_start, csum_offset */
1966 #define VIRTIO_NET_HDR_F_DATA_VALID 2 /* Csum is valid */
1968 #define VIRTIO_NET_HDR_GSO_NONE 0 /* Not a GSO frame */
1969 #define VIRTIO_NET_HDR_GSO_TCPV4 1 /* GSO frame, IPv4 TCP (TSO) */
1970 #define VIRTIO_NET_HDR_GSO_UDP 3 /* GSO frame, IPv4 UDP (UFO) */
1971 #define VIRTIO_NET_HDR_GSO_TCPV6 4 /* GSO frame, IPv6 TCP */
1972 #define VIRTIO_NET_HDR_GSO_ECN 0x80 /* TCP has ECN set */
1976 uint16_t csum_start;
1977 uint16_t csum_offset;
1980 #define WORST_CASE_GSO_HEADER (14+40+60) /* IPv6 + TCP */
1982 /* Private definitions for IPv4, IPv6, UDP and TCP headers. */
1985 uint8_t version_ihl;
1995 /*The options start here. */
2003 uint8_t doff; /* Data offset + Reserved */
2018 uint8_t priority_version;
2019 uint8_t flow_lbl[3];
2021 uint16_t payload_len;
2029 /* Type used to store a checksum (in host byte order) that hasn't been
2032 #define rawsum_t uint32_t
2034 rawsum_t nm_os_csum_raw(uint8_t *data, size_t len, rawsum_t cur_sum);
2035 uint16_t nm_os_csum_ipv4(struct nm_iphdr *iph);
2036 void nm_os_csum_tcpudp_ipv4(struct nm_iphdr *iph, void *data,
2037 size_t datalen, uint16_t *check);
2038 void nm_os_csum_tcpudp_ipv6(struct nm_ipv6hdr *ip6h, void *data,
2039 size_t datalen, uint16_t *check);
2040 uint16_t nm_os_csum_fold(rawsum_t cur_sum);
2042 void bdg_mismatch_datapath(struct netmap_vp_adapter *na,
2043 struct netmap_vp_adapter *dst_na,
2044 const struct nm_bdg_fwd *ft_p,
2045 struct netmap_ring *dst_ring,
2046 u_int *j, u_int lim, u_int *howmany);
2048 /* persistent virtual port routines */
2049 int nm_os_vi_persist(const char *, struct ifnet **);
2050 void nm_os_vi_detach(struct ifnet *);
2051 void nm_os_vi_init_index(void);
2054 * kernel thread routines
2056 struct nm_kctx; /* OS-specific kernel context - opaque */
2057 typedef void (*nm_kctx_worker_fn_t)(void *data, int is_kthread);
2058 typedef void (*nm_kctx_notify_fn_t)(void *data);
2060 /* kthread configuration */
2061 struct nm_kctx_cfg {
2062 long type; /* kthread type/identifier */
2063 nm_kctx_worker_fn_t worker_fn; /* worker function */
2064 void *worker_private;/* worker parameter */
2065 nm_kctx_notify_fn_t notify_fn; /* notify function */
2066 int attach_user; /* attach kthread to user process */
2067 int use_kthread; /* use a kthread for the context */
2069 /* kthread configuration */
2070 struct nm_kctx *nm_os_kctx_create(struct nm_kctx_cfg *cfg,
2071 unsigned int cfgtype,
2073 int nm_os_kctx_worker_start(struct nm_kctx *);
2074 void nm_os_kctx_worker_stop(struct nm_kctx *);
2075 void nm_os_kctx_destroy(struct nm_kctx *);
2076 void nm_os_kctx_worker_wakeup(struct nm_kctx *nmk);
2077 void nm_os_kctx_send_irq(struct nm_kctx *);
2078 void nm_os_kctx_worker_setaff(struct nm_kctx *, int);
2079 u_int nm_os_ncpus(void);
2081 #ifdef WITH_PTNETMAP_HOST
2083 * netmap adapter for host ptnetmap ports
2085 struct netmap_pt_host_adapter {
2086 struct netmap_adapter up;
2088 /* the passed-through adapter */
2089 struct netmap_adapter *parent;
2090 /* parent->na_flags, saved at NETMAP_PT_HOST_CREATE time,
2091 * and restored at NETMAP_PT_HOST_DELETE time */
2092 uint32_t parent_na_flags;
2094 int (*parent_nm_notify)(struct netmap_kring *kring, int flags);
2097 /* ptnetmap HOST routines */
2098 int netmap_get_pt_host_na(struct nmreq *nmr, struct netmap_adapter **na,
2099 struct netmap_mem_d * nmd, int create);
2100 int ptnetmap_ctl(struct nmreq *nmr, struct netmap_adapter *na);
2102 nm_ptnetmap_host_on(struct netmap_adapter *na)
2104 return na && na->na_flags & NAF_PTNETMAP_HOST;
2106 #else /* !WITH_PTNETMAP_HOST */
2107 #define netmap_get_pt_host_na(nmr, _2, _3, _4) \
2108 ((nmr)->nr_flags & (NR_PTNETMAP_HOST) ? EOPNOTSUPP : 0)
2109 #define ptnetmap_ctl(_1, _2) EINVAL
2110 #define nm_ptnetmap_host_on(_1) EINVAL
2111 #endif /* !WITH_PTNETMAP_HOST */
2113 #ifdef WITH_PTNETMAP_GUEST
2114 /* ptnetmap GUEST routines */
2117 * netmap adapter for guest ptnetmap ports
2119 struct netmap_pt_guest_adapter {
2120 /* The netmap adapter to be used by netmap applications.
2121 * This field must be the first, to allow upcast. */
2122 struct netmap_hw_adapter hwup;
2124 /* The netmap adapter to be used by the driver. */
2125 struct netmap_hw_adapter dr;
2129 /* Reference counter to track users of backend netmap port: the
2130 * network stack and netmap clients.
2131 * Used to decide when we need (de)allocate krings/rings and
2132 * start (stop) ptnetmap kthreads. */
2137 int netmap_pt_guest_attach(struct netmap_adapter *na, void *csb,
2138 unsigned int nifp_offset, unsigned int memid);
2140 bool netmap_pt_guest_txsync(struct ptnet_ring *ptring, struct netmap_kring *kring,
2142 bool netmap_pt_guest_rxsync(struct ptnet_ring *ptring, struct netmap_kring *kring,
2144 int ptnet_nm_krings_create(struct netmap_adapter *na);
2145 void ptnet_nm_krings_delete(struct netmap_adapter *na);
2146 void ptnet_nm_dtor(struct netmap_adapter *na);
2147 #endif /* WITH_PTNETMAP_GUEST */
2149 #endif /* _NET_NETMAP_KERN_H_ */