1 /* 2 * Copyright (C) 2011-2014 Matteo Landi, Luigi Rizzo 3 * Copyright (C) 2013-2016 Universita` di Pisa 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 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. 14 * 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 25 * SUCH DAMAGE. 26 */ 27 28 /* 29 * $FreeBSD$ 30 * 31 * The header contains the definitions of constants and function 32 * prototypes used only in kernelspace. 33 */ 34 35 #ifndef _NET_NETMAP_KERN_H_ 36 #define _NET_NETMAP_KERN_H_ 37 38 #if defined(linux) 39 40 #if defined(CONFIG_NETMAP_VALE) 41 #define WITH_VALE 42 #endif 43 #if defined(CONFIG_NETMAP_PIPE) 44 #define WITH_PIPES 45 #endif 46 #if defined(CONFIG_NETMAP_MONITOR) 47 #define WITH_MONITOR 48 #endif 49 #if defined(CONFIG_NETMAP_GENERIC) 50 #define WITH_GENERIC 51 #endif 52 #if defined(CONFIG_NETMAP_PTNETMAP_GUEST) 53 #define WITH_PTNETMAP_GUEST 54 #endif 55 #if defined(CONFIG_NETMAP_PTNETMAP_HOST) 56 #define WITH_PTNETMAP_HOST 57 #endif 58 59 #elif defined (_WIN32) 60 #define WITH_VALE // comment out to disable VALE support 61 #define WITH_PIPES 62 #define WITH_MONITOR 63 #define WITH_GENERIC 64 65 #else /* neither linux nor windows */ 66 #define WITH_VALE // comment out to disable VALE support 67 #define WITH_PIPES 68 #define WITH_MONITOR 69 #define WITH_GENERIC 70 #define WITH_PTNETMAP_HOST /* ptnetmap host support */ 71 #define WITH_PTNETMAP_GUEST /* ptnetmap guest support */ 72 73 #endif 74 75 #if defined(__FreeBSD__) 76 #include <sys/selinfo.h> 77 78 #define likely(x) __builtin_expect((long)!!(x), 1L) 79 #define unlikely(x) __builtin_expect((long)!!(x), 0L) 80 #define __user 81 82 #define NM_LOCK_T struct mtx /* low level spinlock, used to protect queues */ 83 84 #define NM_MTX_T struct sx /* OS-specific mutex (sleepable) */ 85 #define NM_MTX_INIT(m) sx_init(&(m), #m) 86 #define NM_MTX_DESTROY(m) sx_destroy(&(m)) 87 #define NM_MTX_LOCK(m) sx_xlock(&(m)) 88 #define NM_MTX_UNLOCK(m) sx_xunlock(&(m)) 89 #define NM_MTX_ASSERT(m) sx_assert(&(m), SA_XLOCKED) 90 91 #define NM_SELINFO_T struct nm_selinfo 92 #define NM_SELRECORD_T struct thread 93 #define MBUF_LEN(m) ((m)->m_pkthdr.len) 94 #define MBUF_TXQ(m) ((m)->m_pkthdr.flowid) 95 #define MBUF_TRANSMIT(na, ifp, m) ((na)->if_transmit(ifp, m)) 96 #define GEN_TX_MBUF_IFP(m) ((m)->m_pkthdr.rcvif) 97 98 #define NM_ATOMIC_T volatile int // XXX ? 99 /* atomic operations */ 100 #include <machine/atomic.h> 101 #define NM_ATOMIC_TEST_AND_SET(p) (!atomic_cmpset_acq_int((p), 0, 1)) 102 #define NM_ATOMIC_CLEAR(p) atomic_store_rel_int((p), 0) 103 104 #if __FreeBSD_version >= 1100030 105 #define WNA(_ifp) (_ifp)->if_netmap 106 #else /* older FreeBSD */ 107 #define WNA(_ifp) (_ifp)->if_pspare[0] 108 #endif /* older FreeBSD */ 109 110 #if __FreeBSD_version >= 1100005 111 struct netmap_adapter *netmap_getna(if_t ifp); 112 #endif 113 114 #if __FreeBSD_version >= 1100027 115 #define MBUF_REFCNT(m) ((m)->m_ext.ext_count) 116 #define SET_MBUF_REFCNT(m, x) (m)->m_ext.ext_count = x 117 #else 118 #define MBUF_REFCNT(m) ((m)->m_ext.ref_cnt ? *((m)->m_ext.ref_cnt) : -1) 119 #define SET_MBUF_REFCNT(m, x) *((m)->m_ext.ref_cnt) = x 120 #endif 121 122 #define MBUF_QUEUED(m) 1 123 124 struct nm_selinfo { 125 struct selinfo si; 126 struct mtx m; 127 }; 128 129 130 // XXX linux struct, not used in FreeBSD 131 struct net_device_ops { 132 }; 133 struct ethtool_ops { 134 }; 135 struct hrtimer { 136 }; 137 #define NM_BNS_GET(b) 138 #define NM_BNS_PUT(b) 139 140 #elif defined (linux) 141 142 #define NM_LOCK_T safe_spinlock_t // see bsd_glue.h 143 #define NM_SELINFO_T wait_queue_head_t 144 #define MBUF_LEN(m) ((m)->len) 145 #define MBUF_TRANSMIT(na, ifp, m) \ 146 ({ \ 147 /* Avoid infinite recursion with generic. */ \ 148 m->priority = NM_MAGIC_PRIORITY_TX; \ 149 (((struct net_device_ops *)(na)->if_transmit)->ndo_start_xmit(m, ifp)); \ 150 0; \ 151 }) 152 153 /* See explanation in nm_os_generic_xmit_frame. */ 154 #define GEN_TX_MBUF_IFP(m) ((struct ifnet *)skb_shinfo(m)->destructor_arg) 155 156 #define NM_ATOMIC_T volatile long unsigned int 157 158 #define NM_MTX_T struct mutex /* OS-specific sleepable lock */ 159 #define NM_MTX_INIT(m) mutex_init(&(m)) 160 #define NM_MTX_DESTROY(m) do { (void)(m); } while (0) 161 #define NM_MTX_LOCK(m) mutex_lock(&(m)) 162 #define NM_MTX_UNLOCK(m) mutex_unlock(&(m)) 163 #define NM_MTX_ASSERT(m) mutex_is_locked(&(m)) 164 165 #ifndef DEV_NETMAP 166 #define DEV_NETMAP 167 #endif /* DEV_NETMAP */ 168 169 #elif defined (__APPLE__) 170 171 #warning apple support is incomplete. 172 #define likely(x) __builtin_expect(!!(x), 1) 173 #define unlikely(x) __builtin_expect(!!(x), 0) 174 #define NM_LOCK_T IOLock * 175 #define NM_SELINFO_T struct selinfo 176 #define MBUF_LEN(m) ((m)->m_pkthdr.len) 177 178 #elif defined (_WIN32) 179 #include "../../../WINDOWS/win_glue.h" 180 181 #define NM_SELRECORD_T IO_STACK_LOCATION 182 #define NM_SELINFO_T win_SELINFO // see win_glue.h 183 #define NM_LOCK_T win_spinlock_t // see win_glue.h 184 #define NM_MTX_T KGUARDED_MUTEX /* OS-specific mutex (sleepable) */ 185 186 #define NM_MTX_INIT(m) KeInitializeGuardedMutex(&m); 187 #define NM_MTX_DESTROY(m) do { (void)(m); } while (0) 188 #define NM_MTX_LOCK(m) KeAcquireGuardedMutex(&(m)) 189 #define NM_MTX_UNLOCK(m) KeReleaseGuardedMutex(&(m)) 190 #define NM_MTX_ASSERT(m) assert(&m.Count>0) 191 192 //These linknames are for the NDIS driver 193 #define NETMAP_NDIS_LINKNAME_STRING L"\\DosDevices\\NMAPNDIS" 194 #define NETMAP_NDIS_NTDEVICE_STRING L"\\Device\\NMAPNDIS" 195 196 //Definition of internal driver-to-driver ioctl codes 197 #define NETMAP_KERNEL_XCHANGE_POINTERS _IO('i', 180) 198 #define NETMAP_KERNEL_SEND_SHUTDOWN_SIGNAL _IO_direct('i', 195) 199 200 //Empty data structures are not permitted by MSVC compiler 201 //XXX_ale, try to solve this problem 202 struct net_device_ops{ 203 char data[1]; 204 }; 205 typedef struct ethtool_ops{ 206 char data[1]; 207 }; 208 typedef struct hrtimer{ 209 KTIMER timer; 210 BOOLEAN active; 211 KDPC deferred_proc; 212 }; 213 214 /* MSVC does not have likely/unlikely support */ 215 #ifdef _MSC_VER 216 #define likely(x) (x) 217 #define unlikely(x) (x) 218 #else 219 #define likely(x) __builtin_expect((long)!!(x), 1L) 220 #define unlikely(x) __builtin_expect((long)!!(x), 0L) 221 #endif //_MSC_VER 222 223 #else 224 225 #error unsupported platform 226 227 #endif /* end - platform-specific code */ 228 229 #ifndef _WIN32 /* support for emulated sysctl */ 230 #define SYSBEGIN(x) 231 #define SYSEND 232 #endif /* _WIN32 */ 233 234 #define NM_ACCESS_ONCE(x) (*(volatile __typeof__(x) *)&(x)) 235 236 #define NMG_LOCK_T NM_MTX_T 237 #define NMG_LOCK_INIT() NM_MTX_INIT(netmap_global_lock) 238 #define NMG_LOCK_DESTROY() NM_MTX_DESTROY(netmap_global_lock) 239 #define NMG_LOCK() NM_MTX_LOCK(netmap_global_lock) 240 #define NMG_UNLOCK() NM_MTX_UNLOCK(netmap_global_lock) 241 #define NMG_LOCK_ASSERT() NM_MTX_ASSERT(netmap_global_lock) 242 243 #define ND(format, ...) 244 #define D(format, ...) \ 245 do { \ 246 struct timeval __xxts; \ 247 microtime(&__xxts); \ 248 printf("%03d.%06d [%4d] %-25s " format "\n", \ 249 (int)__xxts.tv_sec % 1000, (int)__xxts.tv_usec, \ 250 __LINE__, __FUNCTION__, ##__VA_ARGS__); \ 251 } while (0) 252 253 /* rate limited, lps indicates how many per second */ 254 #define RD(lps, format, ...) \ 255 do { \ 256 static int t0, __cnt; \ 257 if (t0 != time_second) { \ 258 t0 = time_second; \ 259 __cnt = 0; \ 260 } \ 261 if (__cnt++ < lps) \ 262 D(format, ##__VA_ARGS__); \ 263 } while (0) 264 265 struct netmap_adapter; 266 struct nm_bdg_fwd; 267 struct nm_bridge; 268 struct netmap_priv_d; 269 270 /* os-specific NM_SELINFO_T initialzation/destruction functions */ 271 void nm_os_selinfo_init(NM_SELINFO_T *); 272 void nm_os_selinfo_uninit(NM_SELINFO_T *); 273 274 const char *nm_dump_buf(char *p, int len, int lim, char *dst); 275 276 void nm_os_selwakeup(NM_SELINFO_T *si); 277 void nm_os_selrecord(NM_SELRECORD_T *sr, NM_SELINFO_T *si); 278 279 int nm_os_ifnet_init(void); 280 void nm_os_ifnet_fini(void); 281 void nm_os_ifnet_lock(void); 282 void nm_os_ifnet_unlock(void); 283 284 void nm_os_get_module(void); 285 void nm_os_put_module(void); 286 287 void netmap_make_zombie(struct ifnet *); 288 void netmap_undo_zombie(struct ifnet *); 289 290 /* passes a packet up to the host stack. 291 * If the packet is sent (or dropped) immediately it returns NULL, 292 * otherwise it links the packet to prev and returns m. 293 * In this case, a final call with m=NULL and prev != NULL will send up 294 * the entire chain to the host stack. 295 */ 296 void *nm_os_send_up(struct ifnet *, struct mbuf *m, struct mbuf *prev); 297 298 int nm_os_mbuf_has_offld(struct mbuf *m); 299 300 #include "netmap_mbq.h" 301 302 extern NMG_LOCK_T netmap_global_lock; 303 304 enum txrx { NR_RX = 0, NR_TX = 1, NR_TXRX }; 305 306 static __inline const char* 307 nm_txrx2str(enum txrx t) 308 { 309 return (t== NR_RX ? "RX" : "TX"); 310 } 311 312 static __inline enum txrx 313 nm_txrx_swap(enum txrx t) 314 { 315 return (t== NR_RX ? NR_TX : NR_RX); 316 } 317 318 #define for_rx_tx(t) for ((t) = 0; (t) < NR_TXRX; (t)++) 319 320 321 /* 322 * private, kernel view of a ring. Keeps track of the status of 323 * a ring across system calls. 324 * 325 * nr_hwcur index of the next buffer to refill. 326 * It corresponds to ring->head 327 * at the time the system call returns. 328 * 329 * nr_hwtail index of the first buffer owned by the kernel. 330 * On RX, hwcur->hwtail are receive buffers 331 * not yet released. hwcur is advanced following 332 * ring->head, hwtail is advanced on incoming packets, 333 * and a wakeup is generated when hwtail passes ring->cur 334 * On TX, hwcur->rcur have been filled by the sender 335 * but not sent yet to the NIC; rcur->hwtail are available 336 * for new transmissions, and hwtail->hwcur-1 are pending 337 * transmissions not yet acknowledged. 338 * 339 * The indexes in the NIC and netmap rings are offset by nkr_hwofs slots. 340 * This is so that, on a reset, buffers owned by userspace are not 341 * modified by the kernel. In particular: 342 * RX rings: the next empty buffer (hwtail + hwofs) coincides with 343 * the next empty buffer as known by the hardware (next_to_check or so). 344 * TX rings: hwcur + hwofs coincides with next_to_send 345 * 346 * For received packets, slot->flags is set to nkr_slot_flags 347 * so we can provide a proper initial value (e.g. set NS_FORWARD 348 * when operating in 'transparent' mode). 349 * 350 * The following fields are used to implement lock-free copy of packets 351 * from input to output ports in VALE switch: 352 * nkr_hwlease buffer after the last one being copied. 353 * A writer in nm_bdg_flush reserves N buffers 354 * from nr_hwlease, advances it, then does the 355 * copy outside the lock. 356 * In RX rings (used for VALE ports), 357 * nkr_hwtail <= nkr_hwlease < nkr_hwcur+N-1 358 * In TX rings (used for NIC or host stack ports) 359 * nkr_hwcur <= nkr_hwlease < nkr_hwtail 360 * nkr_leases array of nkr_num_slots where writers can report 361 * completion of their block. NR_NOSLOT (~0) indicates 362 * that the writer has not finished yet 363 * nkr_lease_idx index of next free slot in nr_leases, to be assigned 364 * 365 * The kring is manipulated by txsync/rxsync and generic netmap function. 366 * 367 * Concurrent rxsync or txsync on the same ring are prevented through 368 * by nm_kr_(try)lock() which in turn uses nr_busy. This is all we need 369 * for NIC rings, and for TX rings attached to the host stack. 370 * 371 * RX rings attached to the host stack use an mbq (rx_queue) on both 372 * rxsync_from_host() and netmap_transmit(). The mbq is protected 373 * by its internal lock. 374 * 375 * RX rings attached to the VALE switch are accessed by both senders 376 * and receiver. They are protected through the q_lock on the RX ring. 377 */ 378 struct netmap_kring { 379 struct netmap_ring *ring; 380 381 uint32_t nr_hwcur; 382 uint32_t nr_hwtail; 383 384 /* 385 * Copies of values in user rings, so we do not need to look 386 * at the ring (which could be modified). These are set in the 387 * *sync_prologue()/finalize() routines. 388 */ 389 uint32_t rhead; 390 uint32_t rcur; 391 uint32_t rtail; 392 393 uint32_t nr_kflags; /* private driver flags */ 394 #define NKR_PENDINTR 0x1 // Pending interrupt. 395 #define NKR_EXCLUSIVE 0x2 /* exclusive binding */ 396 #define NKR_FORWARD 0x4 /* (host ring only) there are 397 packets to forward 398 */ 399 #define NKR_NEEDRING 0x8 /* ring needed even if users==0 400 * (used internally by pipes and 401 * by ptnetmap host ports) 402 */ 403 404 uint32_t nr_mode; 405 uint32_t nr_pending_mode; 406 #define NKR_NETMAP_OFF 0x0 407 #define NKR_NETMAP_ON 0x1 408 409 uint32_t nkr_num_slots; 410 411 /* 412 * On a NIC reset, the NIC ring indexes may be reset but the 413 * indexes in the netmap rings remain the same. nkr_hwofs 414 * keeps track of the offset between the two. 415 */ 416 int32_t nkr_hwofs; 417 418 uint16_t nkr_slot_flags; /* initial value for flags */ 419 420 /* last_reclaim is opaque marker to help reduce the frequency 421 * of operations such as reclaiming tx buffers. A possible use 422 * is set it to ticks and do the reclaim only once per tick. 423 */ 424 uint64_t last_reclaim; 425 426 427 NM_SELINFO_T si; /* poll/select wait queue */ 428 NM_LOCK_T q_lock; /* protects kring and ring. */ 429 NM_ATOMIC_T nr_busy; /* prevent concurrent syscalls */ 430 431 struct netmap_adapter *na; 432 433 /* The following fields are for VALE switch support */ 434 struct nm_bdg_fwd *nkr_ft; 435 uint32_t *nkr_leases; 436 #define NR_NOSLOT ((uint32_t)~0) /* used in nkr_*lease* */ 437 uint32_t nkr_hwlease; 438 uint32_t nkr_lease_idx; 439 440 /* while nkr_stopped is set, no new [tr]xsync operations can 441 * be started on this kring. 442 * This is used by netmap_disable_all_rings() 443 * to find a synchronization point where critical data 444 * structures pointed to by the kring can be added or removed 445 */ 446 volatile int nkr_stopped; 447 448 /* Support for adapters without native netmap support. 449 * On tx rings we preallocate an array of tx buffers 450 * (same size as the netmap ring), on rx rings we 451 * store incoming mbufs in a queue that is drained by 452 * a rxsync. 453 */ 454 struct mbuf **tx_pool; 455 struct mbuf *tx_event; /* TX event used as a notification */ 456 NM_LOCK_T tx_event_lock; /* protects the tx_event mbuf */ 457 struct mbq rx_queue; /* intercepted rx mbufs. */ 458 459 uint32_t users; /* existing bindings for this ring */ 460 461 uint32_t ring_id; /* kring identifier */ 462 enum txrx tx; /* kind of ring (tx or rx) */ 463 char name[64]; /* diagnostic */ 464 465 /* [tx]sync callback for this kring. 466 * The default nm_kring_create callback (netmap_krings_create) 467 * sets the nm_sync callback of each hardware tx(rx) kring to 468 * the corresponding nm_txsync(nm_rxsync) taken from the 469 * netmap_adapter; moreover, it sets the sync callback 470 * of the host tx(rx) ring to netmap_txsync_to_host 471 * (netmap_rxsync_from_host). 472 * 473 * Overrides: the above configuration is not changed by 474 * any of the nm_krings_create callbacks. 475 */ 476 int (*nm_sync)(struct netmap_kring *kring, int flags); 477 int (*nm_notify)(struct netmap_kring *kring, int flags); 478 479 #ifdef WITH_PIPES 480 struct netmap_kring *pipe; /* if this is a pipe ring, 481 * pointer to the other end 482 */ 483 #endif /* WITH_PIPES */ 484 485 #ifdef WITH_VALE 486 int (*save_notify)(struct netmap_kring *kring, int flags); 487 #endif 488 489 #ifdef WITH_MONITOR 490 /* array of krings that are monitoring this kring */ 491 struct netmap_kring **monitors; 492 uint32_t max_monitors; /* current size of the monitors array */ 493 uint32_t n_monitors; /* next unused entry in the monitor array */ 494 /* 495 * Monitors work by intercepting the sync and notify callbacks of the 496 * monitored krings. This is implemented by replacing the pointers 497 * above and saving the previous ones in mon_* pointers below 498 */ 499 int (*mon_sync)(struct netmap_kring *kring, int flags); 500 int (*mon_notify)(struct netmap_kring *kring, int flags); 501 502 uint32_t mon_tail; /* last seen slot on rx */ 503 uint32_t mon_pos; /* index of this ring in the monitored ring array */ 504 #endif 505 } 506 #ifdef _WIN32 507 __declspec(align(64)); 508 #else 509 __attribute__((__aligned__(64))); 510 #endif 511 512 /* return 1 iff the kring needs to be turned on */ 513 static inline int 514 nm_kring_pending_on(struct netmap_kring *kring) 515 { 516 return kring->nr_pending_mode == NKR_NETMAP_ON && 517 kring->nr_mode == NKR_NETMAP_OFF; 518 } 519 520 /* return 1 iff the kring needs to be turned off */ 521 static inline int 522 nm_kring_pending_off(struct netmap_kring *kring) 523 { 524 return kring->nr_pending_mode == NKR_NETMAP_OFF && 525 kring->nr_mode == NKR_NETMAP_ON; 526 } 527 528 /* return the next index, with wraparound */ 529 static inline uint32_t 530 nm_next(uint32_t i, uint32_t lim) 531 { 532 return unlikely (i == lim) ? 0 : i + 1; 533 } 534 535 536 /* return the previous index, with wraparound */ 537 static inline uint32_t 538 nm_prev(uint32_t i, uint32_t lim) 539 { 540 return unlikely (i == 0) ? lim : i - 1; 541 } 542 543 544 /* 545 * 546 * Here is the layout for the Rx and Tx rings. 547 548 RxRING TxRING 549 550 +-----------------+ +-----------------+ 551 | | | | 552 |XXX free slot XXX| |XXX free slot XXX| 553 +-----------------+ +-----------------+ 554 head->| owned by user |<-hwcur | not sent to nic |<-hwcur 555 | | | yet | 556 +-----------------+ | | 557 cur->| available to | | | 558 | user, not read | +-----------------+ 559 | yet | cur->| (being | 560 | | | prepared) | 561 | | | | 562 +-----------------+ + ------ + 563 tail->| |<-hwtail | |<-hwlease 564 | (being | ... | | ... 565 | prepared) | ... | | ... 566 +-----------------+ ... | | ... 567 | |<-hwlease +-----------------+ 568 | | tail->| |<-hwtail 569 | | | | 570 | | | | 571 | | | | 572 +-----------------+ +-----------------+ 573 574 * The cur/tail (user view) and hwcur/hwtail (kernel view) 575 * are used in the normal operation of the card. 576 * 577 * When a ring is the output of a switch port (Rx ring for 578 * a VALE port, Tx ring for the host stack or NIC), slots 579 * are reserved in blocks through 'hwlease' which points 580 * to the next unused slot. 581 * On an Rx ring, hwlease is always after hwtail, 582 * and completions cause hwtail to advance. 583 * On a Tx ring, hwlease is always between cur and hwtail, 584 * and completions cause cur to advance. 585 * 586 * nm_kr_space() returns the maximum number of slots that 587 * can be assigned. 588 * nm_kr_lease() reserves the required number of buffers, 589 * advances nkr_hwlease and also returns an entry in 590 * a circular array where completions should be reported. 591 */ 592 593 594 struct netmap_lut { 595 struct lut_entry *lut; 596 uint32_t objtotal; /* max buffer index */ 597 uint32_t objsize; /* buffer size */ 598 }; 599 600 struct netmap_vp_adapter; // forward 601 602 /* 603 * The "struct netmap_adapter" extends the "struct adapter" 604 * (or equivalent) device descriptor. 605 * It contains all base fields needed to support netmap operation. 606 * There are in fact different types of netmap adapters 607 * (native, generic, VALE switch...) so a netmap_adapter is 608 * just the first field in the derived type. 609 */ 610 struct netmap_adapter { 611 /* 612 * On linux we do not have a good way to tell if an interface 613 * is netmap-capable. So we always use the following trick: 614 * NA(ifp) points here, and the first entry (which hopefully 615 * always exists and is at least 32 bits) contains a magic 616 * value which we can use to detect that the interface is good. 617 */ 618 uint32_t magic; 619 uint32_t na_flags; /* enabled, and other flags */ 620 #define NAF_SKIP_INTR 1 /* use the regular interrupt handler. 621 * useful during initialization 622 */ 623 #define NAF_SW_ONLY 2 /* forward packets only to sw adapter */ 624 #define NAF_BDG_MAYSLEEP 4 /* the bridge is allowed to sleep when 625 * forwarding packets coming from this 626 * interface 627 */ 628 #define NAF_MEM_OWNER 8 /* the adapter uses its own memory area 629 * that cannot be changed 630 */ 631 #define NAF_NATIVE 16 /* the adapter is native. 632 * Virtual ports (non persistent vale ports, 633 * pipes, monitors...) should never use 634 * this flag. 635 */ 636 #define NAF_NETMAP_ON 32 /* netmap is active (either native or 637 * emulated). Where possible (e.g. FreeBSD) 638 * IFCAP_NETMAP also mirrors this flag. 639 */ 640 #define NAF_HOST_RINGS 64 /* the adapter supports the host rings */ 641 #define NAF_FORCE_NATIVE 128 /* the adapter is always NATIVE */ 642 #define NAF_PTNETMAP_HOST 256 /* the adapter supports ptnetmap in the host */ 643 #define NAF_ZOMBIE (1U<<30) /* the nic driver has been unloaded */ 644 #define NAF_BUSY (1U<<31) /* the adapter is used internally and 645 * cannot be registered from userspace 646 */ 647 int active_fds; /* number of user-space descriptors using this 648 interface, which is equal to the number of 649 struct netmap_if objs in the mapped region. */ 650 651 u_int num_rx_rings; /* number of adapter receive rings */ 652 u_int num_tx_rings; /* number of adapter transmit rings */ 653 654 u_int num_tx_desc; /* number of descriptor in each queue */ 655 u_int num_rx_desc; 656 657 /* tx_rings and rx_rings are private but allocated 658 * as a contiguous chunk of memory. Each array has 659 * N+1 entries, for the adapter queues and for the host queue. 660 */ 661 struct netmap_kring *tx_rings; /* array of TX rings. */ 662 struct netmap_kring *rx_rings; /* array of RX rings. */ 663 664 void *tailroom; /* space below the rings array */ 665 /* (used for leases) */ 666 667 668 NM_SELINFO_T si[NR_TXRX]; /* global wait queues */ 669 670 /* count users of the global wait queues */ 671 int si_users[NR_TXRX]; 672 673 void *pdev; /* used to store pci device */ 674 675 /* copy of if_qflush and if_transmit pointers, to intercept 676 * packets from the network stack when netmap is active. 677 */ 678 int (*if_transmit)(struct ifnet *, struct mbuf *); 679 680 /* copy of if_input for netmap_send_up() */ 681 void (*if_input)(struct ifnet *, struct mbuf *); 682 683 /* references to the ifnet and device routines, used by 684 * the generic netmap functions. 685 */ 686 struct ifnet *ifp; /* adapter is ifp->if_softc */ 687 688 /*---- callbacks for this netmap adapter -----*/ 689 /* 690 * nm_dtor() is the cleanup routine called when destroying 691 * the adapter. 692 * Called with NMG_LOCK held. 693 * 694 * nm_register() is called on NIOCREGIF and close() to enter 695 * or exit netmap mode on the NIC 696 * Called with NNG_LOCK held. 697 * 698 * nm_txsync() pushes packets to the underlying hw/switch 699 * 700 * nm_rxsync() collects packets from the underlying hw/switch 701 * 702 * nm_config() returns configuration information from the OS 703 * Called with NMG_LOCK held. 704 * 705 * nm_krings_create() create and init the tx_rings and 706 * rx_rings arrays of kring structures. In particular, 707 * set the nm_sync callbacks for each ring. 708 * There is no need to also allocate the corresponding 709 * netmap_rings, since netmap_mem_rings_create() will always 710 * be called to provide the missing ones. 711 * Called with NNG_LOCK held. 712 * 713 * nm_krings_delete() cleanup and delete the tx_rings and rx_rings 714 * arrays 715 * Called with NMG_LOCK held. 716 * 717 * nm_notify() is used to act after data have become available 718 * (or the stopped state of the ring has changed) 719 * For hw devices this is typically a selwakeup(), 720 * but for NIC/host ports attached to a switch (or vice-versa) 721 * we also need to invoke the 'txsync' code downstream. 722 * This callback pointer is actually used only to initialize 723 * kring->nm_notify. 724 * Return values are the same as for netmap_rx_irq(). 725 */ 726 void (*nm_dtor)(struct netmap_adapter *); 727 728 int (*nm_register)(struct netmap_adapter *, int onoff); 729 void (*nm_intr)(struct netmap_adapter *, int onoff); 730 731 int (*nm_txsync)(struct netmap_kring *kring, int flags); 732 int (*nm_rxsync)(struct netmap_kring *kring, int flags); 733 int (*nm_notify)(struct netmap_kring *kring, int flags); 734 #define NAF_FORCE_READ 1 735 #define NAF_FORCE_RECLAIM 2 736 /* return configuration information */ 737 int (*nm_config)(struct netmap_adapter *, 738 u_int *txr, u_int *txd, u_int *rxr, u_int *rxd); 739 int (*nm_krings_create)(struct netmap_adapter *); 740 void (*nm_krings_delete)(struct netmap_adapter *); 741 #ifdef WITH_VALE 742 /* 743 * nm_bdg_attach() initializes the na_vp field to point 744 * to an adapter that can be attached to a VALE switch. If the 745 * current adapter is already a VALE port, na_vp is simply a cast; 746 * otherwise, na_vp points to a netmap_bwrap_adapter. 747 * If applicable, this callback also initializes na_hostvp, 748 * that can be used to connect the adapter host rings to the 749 * switch. 750 * Called with NMG_LOCK held. 751 * 752 * nm_bdg_ctl() is called on the actual attach/detach to/from 753 * to/from the switch, to perform adapter-specific 754 * initializations 755 * Called with NMG_LOCK held. 756 */ 757 int (*nm_bdg_attach)(const char *bdg_name, struct netmap_adapter *); 758 int (*nm_bdg_ctl)(struct netmap_adapter *, struct nmreq *, int); 759 760 /* adapter used to attach this adapter to a VALE switch (if any) */ 761 struct netmap_vp_adapter *na_vp; 762 /* adapter used to attach the host rings of this adapter 763 * to a VALE switch (if any) */ 764 struct netmap_vp_adapter *na_hostvp; 765 #endif 766 767 /* standard refcount to control the lifetime of the adapter 768 * (it should be equal to the lifetime of the corresponding ifp) 769 */ 770 int na_refcount; 771 772 /* memory allocator (opaque) 773 * We also cache a pointer to the lut_entry for translating 774 * buffer addresses, the total number of buffers and the buffer size. 775 */ 776 struct netmap_mem_d *nm_mem; 777 struct netmap_lut na_lut; 778 779 /* additional information attached to this adapter 780 * by other netmap subsystems. Currently used by 781 * bwrap, LINUX/v1000 and ptnetmap 782 */ 783 void *na_private; 784 785 /* array of pipes that have this adapter as a parent */ 786 struct netmap_pipe_adapter **na_pipes; 787 int na_next_pipe; /* next free slot in the array */ 788 int na_max_pipes; /* size of the array */ 789 790 /* Offset of ethernet header for each packet. */ 791 u_int virt_hdr_len; 792 793 char name[64]; 794 }; 795 796 static __inline u_int 797 nma_get_ndesc(struct netmap_adapter *na, enum txrx t) 798 { 799 return (t == NR_TX ? na->num_tx_desc : na->num_rx_desc); 800 } 801 802 static __inline void 803 nma_set_ndesc(struct netmap_adapter *na, enum txrx t, u_int v) 804 { 805 if (t == NR_TX) 806 na->num_tx_desc = v; 807 else 808 na->num_rx_desc = v; 809 } 810 811 static __inline u_int 812 nma_get_nrings(struct netmap_adapter *na, enum txrx t) 813 { 814 return (t == NR_TX ? na->num_tx_rings : na->num_rx_rings); 815 } 816 817 static __inline void 818 nma_set_nrings(struct netmap_adapter *na, enum txrx t, u_int v) 819 { 820 if (t == NR_TX) 821 na->num_tx_rings = v; 822 else 823 na->num_rx_rings = v; 824 } 825 826 static __inline struct netmap_kring* 827 NMR(struct netmap_adapter *na, enum txrx t) 828 { 829 return (t == NR_TX ? na->tx_rings : na->rx_rings); 830 } 831 832 /* 833 * If the NIC is owned by the kernel 834 * (i.e., bridge), neither another bridge nor user can use it; 835 * if the NIC is owned by a user, only users can share it. 836 * Evaluation must be done under NMG_LOCK(). 837 */ 838 #define NETMAP_OWNED_BY_KERN(na) ((na)->na_flags & NAF_BUSY) 839 #define NETMAP_OWNED_BY_ANY(na) \ 840 (NETMAP_OWNED_BY_KERN(na) || ((na)->active_fds > 0)) 841 842 /* 843 * derived netmap adapters for various types of ports 844 */ 845 struct netmap_vp_adapter { /* VALE software port */ 846 struct netmap_adapter up; 847 848 /* 849 * Bridge support: 850 * 851 * bdg_port is the port number used in the bridge; 852 * na_bdg points to the bridge this NA is attached to. 853 */ 854 int bdg_port; 855 struct nm_bridge *na_bdg; 856 int retry; 857 858 /* Maximum Frame Size, used in bdg_mismatch_datapath() */ 859 u_int mfs; 860 /* Last source MAC on this port */ 861 uint64_t last_smac; 862 }; 863 864 865 struct netmap_hw_adapter { /* physical device */ 866 struct netmap_adapter up; 867 868 struct net_device_ops nm_ndo; // XXX linux only 869 struct ethtool_ops nm_eto; // XXX linux only 870 const struct ethtool_ops* save_ethtool; 871 872 int (*nm_hw_register)(struct netmap_adapter *, int onoff); 873 }; 874 875 #ifdef WITH_GENERIC 876 /* Mitigation support. */ 877 struct nm_generic_mit { 878 struct hrtimer mit_timer; 879 int mit_pending; 880 int mit_ring_idx; /* index of the ring being mitigated */ 881 struct netmap_adapter *mit_na; /* backpointer */ 882 }; 883 884 struct netmap_generic_adapter { /* emulated device */ 885 struct netmap_hw_adapter up; 886 887 /* Pointer to a previously used netmap adapter. */ 888 struct netmap_adapter *prev; 889 890 /* generic netmap adapters support: 891 * a net_device_ops struct overrides ndo_select_queue(), 892 * save_if_input saves the if_input hook (FreeBSD), 893 * mit implements rx interrupt mitigation, 894 */ 895 struct net_device_ops generic_ndo; 896 void (*save_if_input)(struct ifnet *, struct mbuf *); 897 898 struct nm_generic_mit *mit; 899 #ifdef linux 900 netdev_tx_t (*save_start_xmit)(struct mbuf *, struct ifnet *); 901 #endif 902 /* Is the adapter able to use multiple RX slots to scatter 903 * each packet pushed up by the driver? */ 904 int rxsg; 905 906 /* Is the transmission path controlled by a netmap-aware 907 * device queue (i.e. qdisc on linux)? */ 908 int txqdisc; 909 }; 910 #endif /* WITH_GENERIC */ 911 912 static __inline int 913 netmap_real_rings(struct netmap_adapter *na, enum txrx t) 914 { 915 return nma_get_nrings(na, t) + !!(na->na_flags & NAF_HOST_RINGS); 916 } 917 918 #ifdef WITH_VALE 919 struct nm_bdg_polling_state; 920 /* 921 * Bridge wrapper for non VALE ports attached to a VALE switch. 922 * 923 * The real device must already have its own netmap adapter (hwna). 924 * The bridge wrapper and the hwna adapter share the same set of 925 * netmap rings and buffers, but they have two separate sets of 926 * krings descriptors, with tx/rx meanings swapped: 927 * 928 * netmap 929 * bwrap krings rings krings hwna 930 * +------+ +------+ +-----+ +------+ +------+ 931 * |tx_rings->| |\ /| |----| |<-tx_rings| 932 * | | +------+ \ / +-----+ +------+ | | 933 * | | X | | 934 * | | / \ | | 935 * | | +------+/ \+-----+ +------+ | | 936 * |rx_rings->| | | |----| |<-rx_rings| 937 * | | +------+ +-----+ +------+ | | 938 * +------+ +------+ 939 * 940 * - packets coming from the bridge go to the brwap rx rings, 941 * which are also the hwna tx rings. The bwrap notify callback 942 * will then complete the hwna tx (see netmap_bwrap_notify). 943 * 944 * - packets coming from the outside go to the hwna rx rings, 945 * which are also the bwrap tx rings. The (overwritten) hwna 946 * notify method will then complete the bridge tx 947 * (see netmap_bwrap_intr_notify). 948 * 949 * The bridge wrapper may optionally connect the hwna 'host' rings 950 * to the bridge. This is done by using a second port in the 951 * bridge and connecting it to the 'host' netmap_vp_adapter 952 * contained in the netmap_bwrap_adapter. The brwap host adapter 953 * cross-links the hwna host rings in the same way as shown above. 954 * 955 * - packets coming from the bridge and directed to the host stack 956 * are handled by the bwrap host notify callback 957 * (see netmap_bwrap_host_notify) 958 * 959 * - packets coming from the host stack are still handled by the 960 * overwritten hwna notify callback (netmap_bwrap_intr_notify), 961 * but are diverted to the host adapter depending on the ring number. 962 * 963 */ 964 struct netmap_bwrap_adapter { 965 struct netmap_vp_adapter up; 966 struct netmap_vp_adapter host; /* for host rings */ 967 struct netmap_adapter *hwna; /* the underlying device */ 968 969 /* 970 * When we attach a physical interface to the bridge, we 971 * allow the controlling process to terminate, so we need 972 * a place to store the n_detmap_priv_d data structure. 973 * This is only done when physical interfaces 974 * are attached to a bridge. 975 */ 976 struct netmap_priv_d *na_kpriv; 977 struct nm_bdg_polling_state *na_polling_state; 978 }; 979 int netmap_bwrap_attach(const char *name, struct netmap_adapter *); 980 981 #endif /* WITH_VALE */ 982 983 #ifdef WITH_PIPES 984 985 #define NM_MAXPIPES 64 /* max number of pipes per adapter */ 986 987 struct netmap_pipe_adapter { 988 struct netmap_adapter up; 989 990 u_int id; /* pipe identifier */ 991 int role; /* either NR_REG_PIPE_MASTER or NR_REG_PIPE_SLAVE */ 992 993 struct netmap_adapter *parent; /* adapter that owns the memory */ 994 struct netmap_pipe_adapter *peer; /* the other end of the pipe */ 995 int peer_ref; /* 1 iff we are holding a ref to the peer */ 996 997 u_int parent_slot; /* index in the parent pipe array */ 998 }; 999 1000 #endif /* WITH_PIPES */ 1001 1002 1003 /* return slots reserved to rx clients; used in drivers */ 1004 static inline uint32_t 1005 nm_kr_rxspace(struct netmap_kring *k) 1006 { 1007 int space = k->nr_hwtail - k->nr_hwcur; 1008 if (space < 0) 1009 space += k->nkr_num_slots; 1010 ND("preserving %d rx slots %d -> %d", space, k->nr_hwcur, k->nr_hwtail); 1011 1012 return space; 1013 } 1014 1015 /* return slots reserved to tx clients */ 1016 #define nm_kr_txspace(_k) nm_kr_rxspace(_k) 1017 1018 1019 /* True if no space in the tx ring, only valid after txsync_prologue */ 1020 static inline int 1021 nm_kr_txempty(struct netmap_kring *kring) 1022 { 1023 return kring->rcur == kring->nr_hwtail; 1024 } 1025 1026 /* True if no more completed slots in the rx ring, only valid after 1027 * rxsync_prologue */ 1028 #define nm_kr_rxempty(_k) nm_kr_txempty(_k) 1029 1030 /* 1031 * protect against multiple threads using the same ring. 1032 * also check that the ring has not been stopped or locked 1033 */ 1034 #define NM_KR_BUSY 1 /* some other thread is syncing the ring */ 1035 #define NM_KR_STOPPED 2 /* unbounded stop (ifconfig down or driver unload) */ 1036 #define NM_KR_LOCKED 3 /* bounded, brief stop for mutual exclusion */ 1037 1038 1039 /* release the previously acquired right to use the *sync() methods of the ring */ 1040 static __inline void nm_kr_put(struct netmap_kring *kr) 1041 { 1042 NM_ATOMIC_CLEAR(&kr->nr_busy); 1043 } 1044 1045 1046 /* true if the ifp that backed the adapter has disappeared (e.g., the 1047 * driver has been unloaded) 1048 */ 1049 static inline int nm_iszombie(struct netmap_adapter *na); 1050 1051 /* try to obtain exclusive right to issue the *sync() operations on the ring. 1052 * The right is obtained and must be later relinquished via nm_kr_put() if and 1053 * only if nm_kr_tryget() returns 0. 1054 * If can_sleep is 1 there are only two other possible outcomes: 1055 * - the function returns NM_KR_BUSY 1056 * - the function returns NM_KR_STOPPED and sets the POLLERR bit in *perr 1057 * (if non-null) 1058 * In both cases the caller will typically skip the ring, possibly collecting 1059 * errors along the way. 1060 * If the calling context does not allow sleeping, the caller must pass 0 in can_sleep. 1061 * In the latter case, the function may also return NM_KR_LOCKED and leave *perr 1062 * untouched: ideally, the caller should try again at a later time. 1063 */ 1064 static __inline int nm_kr_tryget(struct netmap_kring *kr, int can_sleep, int *perr) 1065 { 1066 int busy = 1, stopped; 1067 /* check a first time without taking the lock 1068 * to avoid starvation for nm_kr_get() 1069 */ 1070 retry: 1071 stopped = kr->nkr_stopped; 1072 if (unlikely(stopped)) { 1073 goto stop; 1074 } 1075 busy = NM_ATOMIC_TEST_AND_SET(&kr->nr_busy); 1076 /* we should not return NM_KR_BUSY if the ring was 1077 * actually stopped, so check another time after 1078 * the barrier provided by the atomic operation 1079 */ 1080 stopped = kr->nkr_stopped; 1081 if (unlikely(stopped)) { 1082 goto stop; 1083 } 1084 1085 if (unlikely(nm_iszombie(kr->na))) { 1086 stopped = NM_KR_STOPPED; 1087 goto stop; 1088 } 1089 1090 return unlikely(busy) ? NM_KR_BUSY : 0; 1091 1092 stop: 1093 if (!busy) 1094 nm_kr_put(kr); 1095 if (stopped == NM_KR_STOPPED) { 1096 /* if POLLERR is defined we want to use it to simplify netmap_poll(). 1097 * Otherwise, any non-zero value will do. 1098 */ 1099 #ifdef POLLERR 1100 #define NM_POLLERR POLLERR 1101 #else 1102 #define NM_POLLERR 1 1103 #endif /* POLLERR */ 1104 if (perr) 1105 *perr |= NM_POLLERR; 1106 #undef NM_POLLERR 1107 } else if (can_sleep) { 1108 tsleep(kr, 0, "NM_KR_TRYGET", 4); 1109 goto retry; 1110 } 1111 return stopped; 1112 } 1113 1114 /* put the ring in the 'stopped' state and wait for the current user (if any) to 1115 * notice. stopped must be either NM_KR_STOPPED or NM_KR_LOCKED 1116 */ 1117 static __inline void nm_kr_stop(struct netmap_kring *kr, int stopped) 1118 { 1119 kr->nkr_stopped = stopped; 1120 while (NM_ATOMIC_TEST_AND_SET(&kr->nr_busy)) 1121 tsleep(kr, 0, "NM_KR_GET", 4); 1122 } 1123 1124 /* restart a ring after a stop */ 1125 static __inline void nm_kr_start(struct netmap_kring *kr) 1126 { 1127 kr->nkr_stopped = 0; 1128 nm_kr_put(kr); 1129 } 1130 1131 1132 /* 1133 * The following functions are used by individual drivers to 1134 * support netmap operation. 1135 * 1136 * netmap_attach() initializes a struct netmap_adapter, allocating the 1137 * struct netmap_ring's and the struct selinfo. 1138 * 1139 * netmap_detach() frees the memory allocated by netmap_attach(). 1140 * 1141 * netmap_transmit() replaces the if_transmit routine of the interface, 1142 * and is used to intercept packets coming from the stack. 1143 * 1144 * netmap_load_map/netmap_reload_map are helper routines to set/reset 1145 * the dmamap for a packet buffer 1146 * 1147 * netmap_reset() is a helper routine to be called in the hw driver 1148 * when reinitializing a ring. It should not be called by 1149 * virtual ports (vale, pipes, monitor) 1150 */ 1151 int netmap_attach(struct netmap_adapter *); 1152 void netmap_detach(struct ifnet *); 1153 int netmap_transmit(struct ifnet *, struct mbuf *); 1154 struct netmap_slot *netmap_reset(struct netmap_adapter *na, 1155 enum txrx tx, u_int n, u_int new_cur); 1156 int netmap_ring_reinit(struct netmap_kring *); 1157 1158 /* Return codes for netmap_*x_irq. */ 1159 enum { 1160 /* Driver should do normal interrupt processing, e.g. because 1161 * the interface is not in netmap mode. */ 1162 NM_IRQ_PASS = 0, 1163 /* Port is in netmap mode, and the interrupt work has been 1164 * completed. The driver does not have to notify netmap 1165 * again before the next interrupt. */ 1166 NM_IRQ_COMPLETED = -1, 1167 /* Port is in netmap mode, but the interrupt work has not been 1168 * completed. The driver has to make sure netmap will be 1169 * notified again soon, even if no more interrupts come (e.g. 1170 * on Linux the driver should not call napi_complete()). */ 1171 NM_IRQ_RESCHED = -2, 1172 }; 1173 1174 /* default functions to handle rx/tx interrupts */ 1175 int netmap_rx_irq(struct ifnet *, u_int, u_int *); 1176 #define netmap_tx_irq(_n, _q) netmap_rx_irq(_n, _q, NULL) 1177 int netmap_common_irq(struct netmap_adapter *, u_int, u_int *work_done); 1178 1179 1180 #ifdef WITH_VALE 1181 /* functions used by external modules to interface with VALE */ 1182 #define netmap_vp_to_ifp(_vp) ((_vp)->up.ifp) 1183 #define netmap_ifp_to_vp(_ifp) (NA(_ifp)->na_vp) 1184 #define netmap_ifp_to_host_vp(_ifp) (NA(_ifp)->na_hostvp) 1185 #define netmap_bdg_idx(_vp) ((_vp)->bdg_port) 1186 const char *netmap_bdg_name(struct netmap_vp_adapter *); 1187 #else /* !WITH_VALE */ 1188 #define netmap_vp_to_ifp(_vp) NULL 1189 #define netmap_ifp_to_vp(_ifp) NULL 1190 #define netmap_ifp_to_host_vp(_ifp) NULL 1191 #define netmap_bdg_idx(_vp) -1 1192 #define netmap_bdg_name(_vp) NULL 1193 #endif /* WITH_VALE */ 1194 1195 static inline int 1196 nm_netmap_on(struct netmap_adapter *na) 1197 { 1198 return na && na->na_flags & NAF_NETMAP_ON; 1199 } 1200 1201 static inline int 1202 nm_native_on(struct netmap_adapter *na) 1203 { 1204 return nm_netmap_on(na) && (na->na_flags & NAF_NATIVE); 1205 } 1206 1207 static inline int 1208 nm_iszombie(struct netmap_adapter *na) 1209 { 1210 return na == NULL || (na->na_flags & NAF_ZOMBIE); 1211 } 1212 1213 static inline void 1214 nm_update_hostrings_mode(struct netmap_adapter *na) 1215 { 1216 /* Process nr_mode and nr_pending_mode for host rings. */ 1217 na->tx_rings[na->num_tx_rings].nr_mode = 1218 na->tx_rings[na->num_tx_rings].nr_pending_mode; 1219 na->rx_rings[na->num_rx_rings].nr_mode = 1220 na->rx_rings[na->num_rx_rings].nr_pending_mode; 1221 } 1222 1223 /* set/clear native flags and if_transmit/netdev_ops */ 1224 static inline void 1225 nm_set_native_flags(struct netmap_adapter *na) 1226 { 1227 struct ifnet *ifp = na->ifp; 1228 1229 /* We do the setup for intercepting packets only if we are the 1230 * first user of this adapapter. */ 1231 if (na->active_fds > 0) { 1232 return; 1233 } 1234 1235 na->na_flags |= NAF_NETMAP_ON; 1236 #ifdef IFCAP_NETMAP /* or FreeBSD ? */ 1237 ifp->if_capenable |= IFCAP_NETMAP; 1238 #endif 1239 #if defined (__FreeBSD__) 1240 na->if_transmit = ifp->if_transmit; 1241 ifp->if_transmit = netmap_transmit; 1242 #elif defined (_WIN32) 1243 (void)ifp; /* prevent a warning */ 1244 //XXX_ale can we just comment those? 1245 //na->if_transmit = ifp->if_transmit; 1246 //ifp->if_transmit = netmap_transmit; 1247 #else 1248 na->if_transmit = (void *)ifp->netdev_ops; 1249 ifp->netdev_ops = &((struct netmap_hw_adapter *)na)->nm_ndo; 1250 ((struct netmap_hw_adapter *)na)->save_ethtool = ifp->ethtool_ops; 1251 ifp->ethtool_ops = &((struct netmap_hw_adapter*)na)->nm_eto; 1252 #endif 1253 nm_update_hostrings_mode(na); 1254 } 1255 1256 static inline void 1257 nm_clear_native_flags(struct netmap_adapter *na) 1258 { 1259 struct ifnet *ifp = na->ifp; 1260 1261 /* We undo the setup for intercepting packets only if we are the 1262 * last user of this adapapter. */ 1263 if (na->active_fds > 0) { 1264 return; 1265 } 1266 1267 nm_update_hostrings_mode(na); 1268 1269 #if defined(__FreeBSD__) 1270 ifp->if_transmit = na->if_transmit; 1271 #elif defined(_WIN32) 1272 (void)ifp; /* prevent a warning */ 1273 //XXX_ale can we just comment those? 1274 //ifp->if_transmit = na->if_transmit; 1275 #else 1276 ifp->netdev_ops = (void *)na->if_transmit; 1277 ifp->ethtool_ops = ((struct netmap_hw_adapter*)na)->save_ethtool; 1278 #endif 1279 na->na_flags &= ~NAF_NETMAP_ON; 1280 #ifdef IFCAP_NETMAP /* or FreeBSD ? */ 1281 ifp->if_capenable &= ~IFCAP_NETMAP; 1282 #endif 1283 } 1284 1285 /* 1286 * nm_*sync_prologue() functions are used in ioctl/poll and ptnetmap 1287 * kthreads. 1288 * We need netmap_ring* parameter, because in ptnetmap it is decoupled 1289 * from host kring. 1290 * The user-space ring pointers (head/cur/tail) are shared through 1291 * CSB between host and guest. 1292 */ 1293 1294 /* 1295 * validates parameters in the ring/kring, returns a value for head 1296 * If any error, returns ring_size to force a reinit. 1297 */ 1298 uint32_t nm_txsync_prologue(struct netmap_kring *, struct netmap_ring *); 1299 1300 1301 /* 1302 * validates parameters in the ring/kring, returns a value for head 1303 * If any error, returns ring_size lim to force a reinit. 1304 */ 1305 uint32_t nm_rxsync_prologue(struct netmap_kring *, struct netmap_ring *); 1306 1307 1308 /* check/fix address and len in tx rings */ 1309 #if 1 /* debug version */ 1310 #define NM_CHECK_ADDR_LEN(_na, _a, _l) do { \ 1311 if (_a == NETMAP_BUF_BASE(_na) || _l > NETMAP_BUF_SIZE(_na)) { \ 1312 RD(5, "bad addr/len ring %d slot %d idx %d len %d", \ 1313 kring->ring_id, nm_i, slot->buf_idx, len); \ 1314 if (_l > NETMAP_BUF_SIZE(_na)) \ 1315 _l = NETMAP_BUF_SIZE(_na); \ 1316 } } while (0) 1317 #else /* no debug version */ 1318 #define NM_CHECK_ADDR_LEN(_na, _a, _l) do { \ 1319 if (_l > NETMAP_BUF_SIZE(_na)) \ 1320 _l = NETMAP_BUF_SIZE(_na); \ 1321 } while (0) 1322 #endif 1323 1324 1325 /*---------------------------------------------------------------*/ 1326 /* 1327 * Support routines used by netmap subsystems 1328 * (native drivers, VALE, generic, pipes, monitors, ...) 1329 */ 1330 1331 1332 /* common routine for all functions that create a netmap adapter. It performs 1333 * two main tasks: 1334 * - if the na points to an ifp, mark the ifp as netmap capable 1335 * using na as its native adapter; 1336 * - provide defaults for the setup callbacks and the memory allocator 1337 */ 1338 int netmap_attach_common(struct netmap_adapter *); 1339 /* common actions to be performed on netmap adapter destruction */ 1340 void netmap_detach_common(struct netmap_adapter *); 1341 /* fill priv->np_[tr]xq{first,last} using the ringid and flags information 1342 * coming from a struct nmreq 1343 */ 1344 int netmap_interp_ringid(struct netmap_priv_d *priv, uint16_t ringid, uint32_t flags); 1345 /* update the ring parameters (number and size of tx and rx rings). 1346 * It calls the nm_config callback, if available. 1347 */ 1348 int netmap_update_config(struct netmap_adapter *na); 1349 /* create and initialize the common fields of the krings array. 1350 * using the information that must be already available in the na. 1351 * tailroom can be used to request the allocation of additional 1352 * tailroom bytes after the krings array. This is used by 1353 * netmap_vp_adapter's (i.e., VALE ports) to make room for 1354 * leasing-related data structures 1355 */ 1356 int netmap_krings_create(struct netmap_adapter *na, u_int tailroom); 1357 /* deletes the kring array of the adapter. The array must have 1358 * been created using netmap_krings_create 1359 */ 1360 void netmap_krings_delete(struct netmap_adapter *na); 1361 1362 int netmap_hw_krings_create(struct netmap_adapter *na); 1363 void netmap_hw_krings_delete(struct netmap_adapter *na); 1364 1365 /* set the stopped/enabled status of ring 1366 * When stopping, they also wait for all current activity on the ring to 1367 * terminate. The status change is then notified using the na nm_notify 1368 * callback. 1369 */ 1370 void netmap_set_ring(struct netmap_adapter *, u_int ring_id, enum txrx, int stopped); 1371 /* set the stopped/enabled status of all rings of the adapter. */ 1372 void netmap_set_all_rings(struct netmap_adapter *, int stopped); 1373 /* convenience wrappers for netmap_set_all_rings */ 1374 void netmap_disable_all_rings(struct ifnet *); 1375 void netmap_enable_all_rings(struct ifnet *); 1376 1377 int netmap_do_regif(struct netmap_priv_d *priv, struct netmap_adapter *na, 1378 uint16_t ringid, uint32_t flags); 1379 void netmap_do_unregif(struct netmap_priv_d *priv); 1380 1381 u_int nm_bound_var(u_int *v, u_int dflt, u_int lo, u_int hi, const char *msg); 1382 int netmap_get_na(struct nmreq *nmr, struct netmap_adapter **na, 1383 struct ifnet **ifp, int create); 1384 void netmap_unget_na(struct netmap_adapter *na, struct ifnet *ifp); 1385 int netmap_get_hw_na(struct ifnet *ifp, struct netmap_adapter **na); 1386 1387 1388 #ifdef WITH_VALE 1389 /* 1390 * The following bridge-related functions are used by other 1391 * kernel modules. 1392 * 1393 * VALE only supports unicast or broadcast. The lookup 1394 * function can return 0 .. NM_BDG_MAXPORTS-1 for regular ports, 1395 * NM_BDG_MAXPORTS for broadcast, NM_BDG_MAXPORTS+1 for unknown. 1396 * XXX in practice "unknown" might be handled same as broadcast. 1397 */ 1398 typedef u_int (*bdg_lookup_fn_t)(struct nm_bdg_fwd *ft, uint8_t *ring_nr, 1399 struct netmap_vp_adapter *); 1400 typedef int (*bdg_config_fn_t)(struct nm_ifreq *); 1401 typedef void (*bdg_dtor_fn_t)(const struct netmap_vp_adapter *); 1402 struct netmap_bdg_ops { 1403 bdg_lookup_fn_t lookup; 1404 bdg_config_fn_t config; 1405 bdg_dtor_fn_t dtor; 1406 }; 1407 1408 u_int netmap_bdg_learning(struct nm_bdg_fwd *ft, uint8_t *dst_ring, 1409 struct netmap_vp_adapter *); 1410 1411 #define NM_BRIDGES 8 /* number of bridges */ 1412 #define NM_BDG_MAXPORTS 254 /* up to 254 */ 1413 #define NM_BDG_BROADCAST NM_BDG_MAXPORTS 1414 #define NM_BDG_NOPORT (NM_BDG_MAXPORTS+1) 1415 1416 /* these are redefined in case of no VALE support */ 1417 int netmap_get_bdg_na(struct nmreq *nmr, struct netmap_adapter **na, int create); 1418 struct nm_bridge *netmap_init_bridges2(u_int); 1419 void netmap_uninit_bridges2(struct nm_bridge *, u_int); 1420 int netmap_init_bridges(void); 1421 void netmap_uninit_bridges(void); 1422 int netmap_bdg_ctl(struct nmreq *nmr, struct netmap_bdg_ops *bdg_ops); 1423 int netmap_bdg_config(struct nmreq *nmr); 1424 1425 #else /* !WITH_VALE */ 1426 #define netmap_get_bdg_na(_1, _2, _3) 0 1427 #define netmap_init_bridges(_1) 0 1428 #define netmap_uninit_bridges() 1429 #define netmap_bdg_ctl(_1, _2) EINVAL 1430 #endif /* !WITH_VALE */ 1431 1432 #ifdef WITH_PIPES 1433 /* max number of pipes per device */ 1434 #define NM_MAXPIPES 64 /* XXX how many? */ 1435 void netmap_pipe_dealloc(struct netmap_adapter *); 1436 int netmap_get_pipe_na(struct nmreq *nmr, struct netmap_adapter **na, int create); 1437 #else /* !WITH_PIPES */ 1438 #define NM_MAXPIPES 0 1439 #define netmap_pipe_alloc(_1, _2) 0 1440 #define netmap_pipe_dealloc(_1) 1441 #define netmap_get_pipe_na(nmr, _2, _3) \ 1442 ({ int role__ = (nmr)->nr_flags & NR_REG_MASK; \ 1443 (role__ == NR_REG_PIPE_MASTER || \ 1444 role__ == NR_REG_PIPE_SLAVE) ? EOPNOTSUPP : 0; }) 1445 #endif 1446 1447 #ifdef WITH_MONITOR 1448 int netmap_get_monitor_na(struct nmreq *nmr, struct netmap_adapter **na, int create); 1449 void netmap_monitor_stop(struct netmap_adapter *na); 1450 #else 1451 #define netmap_get_monitor_na(nmr, _2, _3) \ 1452 ((nmr)->nr_flags & (NR_MONITOR_TX | NR_MONITOR_RX) ? EOPNOTSUPP : 0) 1453 #endif 1454 1455 #ifdef CONFIG_NET_NS 1456 struct net *netmap_bns_get(void); 1457 void netmap_bns_put(struct net *); 1458 void netmap_bns_getbridges(struct nm_bridge **, u_int *); 1459 #else 1460 #define netmap_bns_get() 1461 #define netmap_bns_put(_1) 1462 #define netmap_bns_getbridges(b, n) \ 1463 do { *b = nm_bridges; *n = NM_BRIDGES; } while (0) 1464 #endif 1465 1466 /* Various prototypes */ 1467 int netmap_poll(struct netmap_priv_d *, int events, NM_SELRECORD_T *td); 1468 int netmap_init(void); 1469 void netmap_fini(void); 1470 int netmap_get_memory(struct netmap_priv_d* p); 1471 void netmap_dtor(void *data); 1472 1473 int netmap_ioctl(struct netmap_priv_d *priv, u_long cmd, caddr_t data, struct thread *); 1474 1475 /* netmap_adapter creation/destruction */ 1476 1477 // #define NM_DEBUG_PUTGET 1 1478 1479 #ifdef NM_DEBUG_PUTGET 1480 1481 #define NM_DBG(f) __##f 1482 1483 void __netmap_adapter_get(struct netmap_adapter *na); 1484 1485 #define netmap_adapter_get(na) \ 1486 do { \ 1487 struct netmap_adapter *__na = na; \ 1488 D("getting %p:%s (%d)", __na, (__na)->name, (__na)->na_refcount); \ 1489 __netmap_adapter_get(__na); \ 1490 } while (0) 1491 1492 int __netmap_adapter_put(struct netmap_adapter *na); 1493 1494 #define netmap_adapter_put(na) \ 1495 ({ \ 1496 struct netmap_adapter *__na = na; \ 1497 D("putting %p:%s (%d)", __na, (__na)->name, (__na)->na_refcount); \ 1498 __netmap_adapter_put(__na); \ 1499 }) 1500 1501 #else /* !NM_DEBUG_PUTGET */ 1502 1503 #define NM_DBG(f) f 1504 void netmap_adapter_get(struct netmap_adapter *na); 1505 int netmap_adapter_put(struct netmap_adapter *na); 1506 1507 #endif /* !NM_DEBUG_PUTGET */ 1508 1509 1510 /* 1511 * module variables 1512 */ 1513 #define NETMAP_BUF_BASE(_na) ((_na)->na_lut.lut[0].vaddr) 1514 #define NETMAP_BUF_SIZE(_na) ((_na)->na_lut.objsize) 1515 extern int netmap_mitigate; // XXX not really used 1516 extern int netmap_no_pendintr; 1517 extern int netmap_verbose; // XXX debugging 1518 enum { /* verbose flags */ 1519 NM_VERB_ON = 1, /* generic verbose */ 1520 NM_VERB_HOST = 0x2, /* verbose host stack */ 1521 NM_VERB_RXSYNC = 0x10, /* verbose on rxsync/txsync */ 1522 NM_VERB_TXSYNC = 0x20, 1523 NM_VERB_RXINTR = 0x100, /* verbose on rx/tx intr (driver) */ 1524 NM_VERB_TXINTR = 0x200, 1525 NM_VERB_NIC_RXSYNC = 0x1000, /* verbose on rx/tx intr (driver) */ 1526 NM_VERB_NIC_TXSYNC = 0x2000, 1527 }; 1528 1529 extern int netmap_txsync_retry; 1530 extern int netmap_adaptive_io; 1531 extern int netmap_flags; 1532 extern int netmap_generic_mit; 1533 extern int netmap_generic_ringsize; 1534 extern int netmap_generic_rings; 1535 extern int netmap_generic_txqdisc; 1536 1537 /* 1538 * NA returns a pointer to the struct netmap adapter from the ifp, 1539 * WNA is used to write it. 1540 */ 1541 #define NA(_ifp) ((struct netmap_adapter *)WNA(_ifp)) 1542 1543 /* 1544 * On old versions of FreeBSD, NA(ifp) is a pspare. On linux we 1545 * overload another pointer in the netdev. 1546 * 1547 * We check if NA(ifp) is set and its first element has a related 1548 * magic value. The capenable is within the struct netmap_adapter. 1549 */ 1550 #define NETMAP_MAGIC 0x52697a7a 1551 1552 #define NM_NA_VALID(ifp) (NA(ifp) && \ 1553 ((uint32_t)(uintptr_t)NA(ifp) ^ NA(ifp)->magic) == NETMAP_MAGIC ) 1554 1555 #define NM_ATTACH_NA(ifp, na) do { \ 1556 WNA(ifp) = na; \ 1557 if (NA(ifp)) \ 1558 NA(ifp)->magic = \ 1559 ((uint32_t)(uintptr_t)NA(ifp)) ^ NETMAP_MAGIC; \ 1560 } while(0) 1561 1562 #define NM_IS_NATIVE(ifp) (NM_NA_VALID(ifp) && NA(ifp)->nm_dtor == netmap_hw_dtor) 1563 1564 #if defined(__FreeBSD__) 1565 1566 /* Assigns the device IOMMU domain to an allocator. 1567 * Returns -ENOMEM in case the domain is different */ 1568 #define nm_iommu_group_id(dev) (0) 1569 1570 /* Callback invoked by the dma machinery after a successful dmamap_load */ 1571 static void netmap_dmamap_cb(__unused void *arg, 1572 __unused bus_dma_segment_t * segs, __unused int nseg, __unused int error) 1573 { 1574 } 1575 1576 /* bus_dmamap_load wrapper: call aforementioned function if map != NULL. 1577 * XXX can we do it without a callback ? 1578 */ 1579 static inline void 1580 netmap_load_map(struct netmap_adapter *na, 1581 bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 1582 { 1583 if (map) 1584 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE(na), 1585 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT); 1586 } 1587 1588 static inline void 1589 netmap_unload_map(struct netmap_adapter *na, 1590 bus_dma_tag_t tag, bus_dmamap_t map) 1591 { 1592 if (map) 1593 bus_dmamap_unload(tag, map); 1594 } 1595 1596 /* update the map when a buffer changes. */ 1597 static inline void 1598 netmap_reload_map(struct netmap_adapter *na, 1599 bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 1600 { 1601 if (map) { 1602 bus_dmamap_unload(tag, map); 1603 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE(na), 1604 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT); 1605 } 1606 } 1607 1608 #elif defined(_WIN32) 1609 1610 #else /* linux */ 1611 1612 int nm_iommu_group_id(bus_dma_tag_t dev); 1613 #include <linux/dma-mapping.h> 1614 1615 static inline void 1616 netmap_load_map(struct netmap_adapter *na, 1617 bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 1618 { 1619 if (0 && map) { 1620 *map = dma_map_single(na->pdev, buf, NETMAP_BUF_SIZE(na), 1621 DMA_BIDIRECTIONAL); 1622 } 1623 } 1624 1625 static inline void 1626 netmap_unload_map(struct netmap_adapter *na, 1627 bus_dma_tag_t tag, bus_dmamap_t map) 1628 { 1629 u_int sz = NETMAP_BUF_SIZE(na); 1630 1631 if (*map) { 1632 dma_unmap_single(na->pdev, *map, sz, 1633 DMA_BIDIRECTIONAL); 1634 } 1635 } 1636 1637 static inline void 1638 netmap_reload_map(struct netmap_adapter *na, 1639 bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 1640 { 1641 u_int sz = NETMAP_BUF_SIZE(na); 1642 1643 if (*map) { 1644 dma_unmap_single(na->pdev, *map, sz, 1645 DMA_BIDIRECTIONAL); 1646 } 1647 1648 *map = dma_map_single(na->pdev, buf, sz, 1649 DMA_BIDIRECTIONAL); 1650 } 1651 1652 /* 1653 * XXX How do we redefine these functions: 1654 * 1655 * on linux we need 1656 * dma_map_single(&pdev->dev, virt_addr, len, direction) 1657 * dma_unmap_single(&adapter->pdev->dev, phys_addr, len, direction 1658 * The len can be implicit (on netmap it is NETMAP_BUF_SIZE) 1659 * unfortunately the direction is not, so we need to change 1660 * something to have a cross API 1661 */ 1662 1663 #if 0 1664 struct e1000_buffer *buffer_info = &tx_ring->buffer_info[l]; 1665 /* set time_stamp *before* dma to help avoid a possible race */ 1666 buffer_info->time_stamp = jiffies; 1667 buffer_info->mapped_as_page = false; 1668 buffer_info->length = len; 1669 //buffer_info->next_to_watch = l; 1670 /* reload dma map */ 1671 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma, 1672 NETMAP_BUF_SIZE, DMA_TO_DEVICE); 1673 buffer_info->dma = dma_map_single(&adapter->pdev->dev, 1674 addr, NETMAP_BUF_SIZE, DMA_TO_DEVICE); 1675 1676 if (dma_mapping_error(&adapter->pdev->dev, buffer_info->dma)) { 1677 D("dma mapping error"); 1678 /* goto dma_error; See e1000_put_txbuf() */ 1679 /* XXX reset */ 1680 } 1681 tx_desc->buffer_addr = htole64(buffer_info->dma); //XXX 1682 1683 #endif 1684 1685 /* 1686 * The bus_dmamap_sync() can be one of wmb() or rmb() depending on direction. 1687 */ 1688 #define bus_dmamap_sync(_a, _b, _c) 1689 1690 #endif /* linux */ 1691 1692 1693 /* 1694 * functions to map NIC to KRING indexes (n2k) and vice versa (k2n) 1695 */ 1696 static inline int 1697 netmap_idx_n2k(struct netmap_kring *kr, int idx) 1698 { 1699 int n = kr->nkr_num_slots; 1700 idx += kr->nkr_hwofs; 1701 if (idx < 0) 1702 return idx + n; 1703 else if (idx < n) 1704 return idx; 1705 else 1706 return idx - n; 1707 } 1708 1709 1710 static inline int 1711 netmap_idx_k2n(struct netmap_kring *kr, int idx) 1712 { 1713 int n = kr->nkr_num_slots; 1714 idx -= kr->nkr_hwofs; 1715 if (idx < 0) 1716 return idx + n; 1717 else if (idx < n) 1718 return idx; 1719 else 1720 return idx - n; 1721 } 1722 1723 1724 /* Entries of the look-up table. */ 1725 struct lut_entry { 1726 void *vaddr; /* virtual address. */ 1727 vm_paddr_t paddr; /* physical address. */ 1728 }; 1729 1730 struct netmap_obj_pool; 1731 1732 /* 1733 * NMB return the virtual address of a buffer (buffer 0 on bad index) 1734 * PNMB also fills the physical address 1735 */ 1736 static inline void * 1737 NMB(struct netmap_adapter *na, struct netmap_slot *slot) 1738 { 1739 struct lut_entry *lut = na->na_lut.lut; 1740 uint32_t i = slot->buf_idx; 1741 return (unlikely(i >= na->na_lut.objtotal)) ? 1742 lut[0].vaddr : lut[i].vaddr; 1743 } 1744 1745 static inline void * 1746 PNMB(struct netmap_adapter *na, struct netmap_slot *slot, uint64_t *pp) 1747 { 1748 uint32_t i = slot->buf_idx; 1749 struct lut_entry *lut = na->na_lut.lut; 1750 void *ret = (i >= na->na_lut.objtotal) ? lut[0].vaddr : lut[i].vaddr; 1751 1752 #ifndef _WIN32 1753 *pp = (i >= na->na_lut.objtotal) ? lut[0].paddr : lut[i].paddr; 1754 #else 1755 *pp = (i >= na->na_lut.objtotal) ? (uint64_t)lut[0].paddr.QuadPart : (uint64_t)lut[i].paddr.QuadPart; 1756 #endif 1757 return ret; 1758 } 1759 1760 1761 /* 1762 * Structure associated to each netmap file descriptor. 1763 * It is created on open and left unbound (np_nifp == NULL). 1764 * A successful NIOCREGIF will set np_nifp and the first few fields; 1765 * this is protected by a global lock (NMG_LOCK) due to low contention. 1766 * 1767 * np_refs counts the number of references to the structure: one for the fd, 1768 * plus (on FreeBSD) one for each active mmap which we track ourselves 1769 * (linux automatically tracks them, but FreeBSD does not). 1770 * np_refs is protected by NMG_LOCK. 1771 * 1772 * Read access to the structure is lock free, because ni_nifp once set 1773 * can only go to 0 when nobody is using the entry anymore. Readers 1774 * must check that np_nifp != NULL before using the other fields. 1775 */ 1776 struct netmap_priv_d { 1777 struct netmap_if * volatile np_nifp; /* netmap if descriptor. */ 1778 1779 struct netmap_adapter *np_na; 1780 struct ifnet *np_ifp; 1781 uint32_t np_flags; /* from the ioctl */ 1782 u_int np_qfirst[NR_TXRX], 1783 np_qlast[NR_TXRX]; /* range of tx/rx rings to scan */ 1784 uint16_t np_txpoll; /* XXX and also np_rxpoll ? */ 1785 1786 int np_refs; /* use with NMG_LOCK held */ 1787 1788 /* pointers to the selinfo to be used for selrecord. 1789 * Either the local or the global one depending on the 1790 * number of rings. 1791 */ 1792 NM_SELINFO_T *np_si[NR_TXRX]; 1793 struct thread *np_td; /* kqueue, just debugging */ 1794 }; 1795 1796 struct netmap_priv_d *netmap_priv_new(void); 1797 void netmap_priv_delete(struct netmap_priv_d *); 1798 1799 static inline int nm_kring_pending(struct netmap_priv_d *np) 1800 { 1801 struct netmap_adapter *na = np->np_na; 1802 enum txrx t; 1803 int i; 1804 1805 for_rx_tx(t) { 1806 for (i = np->np_qfirst[t]; i < np->np_qlast[t]; i++) { 1807 struct netmap_kring *kring = &NMR(na, t)[i]; 1808 if (kring->nr_mode != kring->nr_pending_mode) { 1809 return 1; 1810 } 1811 } 1812 } 1813 return 0; 1814 } 1815 1816 #ifdef WITH_MONITOR 1817 1818 struct netmap_monitor_adapter { 1819 struct netmap_adapter up; 1820 1821 struct netmap_priv_d priv; 1822 uint32_t flags; 1823 }; 1824 1825 #endif /* WITH_MONITOR */ 1826 1827 1828 #ifdef WITH_GENERIC 1829 /* 1830 * generic netmap emulation for devices that do not have 1831 * native netmap support. 1832 */ 1833 int generic_netmap_attach(struct ifnet *ifp); 1834 int generic_rx_handler(struct ifnet *ifp, struct mbuf *m);; 1835 1836 int nm_os_catch_rx(struct netmap_generic_adapter *gna, int intercept); 1837 int nm_os_catch_tx(struct netmap_generic_adapter *gna, int intercept); 1838 1839 /* 1840 * the generic transmit routine is passed a structure to optionally 1841 * build a queue of descriptors, in an OS-specific way. 1842 * The payload is at addr, if non-null, and the routine should send or queue 1843 * the packet, returning 0 if successful, 1 on failure. 1844 * 1845 * At the end, if head is non-null, there will be an additional call 1846 * to the function with addr = NULL; this should tell the OS-specific 1847 * routine to send the queue and free any resources. Failure is ignored. 1848 */ 1849 struct nm_os_gen_arg { 1850 struct ifnet *ifp; 1851 void *m; /* os-specific mbuf-like object */ 1852 void *head, *tail; /* tailq, if the OS-specific routine needs to build one */ 1853 void *addr; /* payload of current packet */ 1854 u_int len; /* packet length */ 1855 u_int ring_nr; /* packet length */ 1856 u_int qevent; /* in txqdisc mode, place an event on this mbuf */ 1857 }; 1858 1859 int nm_os_generic_xmit_frame(struct nm_os_gen_arg *); 1860 int nm_os_generic_find_num_desc(struct ifnet *ifp, u_int *tx, u_int *rx); 1861 void nm_os_generic_find_num_queues(struct ifnet *ifp, u_int *txq, u_int *rxq); 1862 void nm_os_generic_set_features(struct netmap_generic_adapter *gna); 1863 1864 static inline struct ifnet* 1865 netmap_generic_getifp(struct netmap_generic_adapter *gna) 1866 { 1867 if (gna->prev) 1868 return gna->prev->ifp; 1869 1870 return gna->up.up.ifp; 1871 } 1872 1873 void netmap_generic_irq(struct netmap_adapter *na, u_int q, u_int *work_done); 1874 1875 //#define RATE_GENERIC /* Enables communication statistics for generic. */ 1876 #ifdef RATE_GENERIC 1877 void generic_rate(int txp, int txs, int txi, int rxp, int rxs, int rxi); 1878 #else 1879 #define generic_rate(txp, txs, txi, rxp, rxs, rxi) 1880 #endif 1881 1882 /* 1883 * netmap_mitigation API. This is used by the generic adapter 1884 * to reduce the number of interrupt requests/selwakeup 1885 * to clients on incoming packets. 1886 */ 1887 void nm_os_mitigation_init(struct nm_generic_mit *mit, int idx, 1888 struct netmap_adapter *na); 1889 void nm_os_mitigation_start(struct nm_generic_mit *mit); 1890 void nm_os_mitigation_restart(struct nm_generic_mit *mit); 1891 int nm_os_mitigation_active(struct nm_generic_mit *mit); 1892 void nm_os_mitigation_cleanup(struct nm_generic_mit *mit); 1893 #else /* !WITH_GENERIC */ 1894 #define generic_netmap_attach(ifp) (EOPNOTSUPP) 1895 #endif /* WITH_GENERIC */ 1896 1897 /* Shared declarations for the VALE switch. */ 1898 1899 /* 1900 * Each transmit queue accumulates a batch of packets into 1901 * a structure before forwarding. Packets to the same 1902 * destination are put in a list using ft_next as a link field. 1903 * ft_frags and ft_next are valid only on the first fragment. 1904 */ 1905 struct nm_bdg_fwd { /* forwarding entry for a bridge */ 1906 void *ft_buf; /* netmap or indirect buffer */ 1907 uint8_t ft_frags; /* how many fragments (only on 1st frag) */ 1908 uint8_t _ft_port; /* dst port (unused) */ 1909 uint16_t ft_flags; /* flags, e.g. indirect */ 1910 uint16_t ft_len; /* src fragment len */ 1911 uint16_t ft_next; /* next packet to same destination */ 1912 }; 1913 1914 /* struct 'virtio_net_hdr' from linux. */ 1915 struct nm_vnet_hdr { 1916 #define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /* Use csum_start, csum_offset */ 1917 #define VIRTIO_NET_HDR_F_DATA_VALID 2 /* Csum is valid */ 1918 uint8_t flags; 1919 #define VIRTIO_NET_HDR_GSO_NONE 0 /* Not a GSO frame */ 1920 #define VIRTIO_NET_HDR_GSO_TCPV4 1 /* GSO frame, IPv4 TCP (TSO) */ 1921 #define VIRTIO_NET_HDR_GSO_UDP 3 /* GSO frame, IPv4 UDP (UFO) */ 1922 #define VIRTIO_NET_HDR_GSO_TCPV6 4 /* GSO frame, IPv6 TCP */ 1923 #define VIRTIO_NET_HDR_GSO_ECN 0x80 /* TCP has ECN set */ 1924 uint8_t gso_type; 1925 uint16_t hdr_len; 1926 uint16_t gso_size; 1927 uint16_t csum_start; 1928 uint16_t csum_offset; 1929 }; 1930 1931 #define WORST_CASE_GSO_HEADER (14+40+60) /* IPv6 + TCP */ 1932 1933 /* Private definitions for IPv4, IPv6, UDP and TCP headers. */ 1934 1935 struct nm_iphdr { 1936 uint8_t version_ihl; 1937 uint8_t tos; 1938 uint16_t tot_len; 1939 uint16_t id; 1940 uint16_t frag_off; 1941 uint8_t ttl; 1942 uint8_t protocol; 1943 uint16_t check; 1944 uint32_t saddr; 1945 uint32_t daddr; 1946 /*The options start here. */ 1947 }; 1948 1949 struct nm_tcphdr { 1950 uint16_t source; 1951 uint16_t dest; 1952 uint32_t seq; 1953 uint32_t ack_seq; 1954 uint8_t doff; /* Data offset + Reserved */ 1955 uint8_t flags; 1956 uint16_t window; 1957 uint16_t check; 1958 uint16_t urg_ptr; 1959 }; 1960 1961 struct nm_udphdr { 1962 uint16_t source; 1963 uint16_t dest; 1964 uint16_t len; 1965 uint16_t check; 1966 }; 1967 1968 struct nm_ipv6hdr { 1969 uint8_t priority_version; 1970 uint8_t flow_lbl[3]; 1971 1972 uint16_t payload_len; 1973 uint8_t nexthdr; 1974 uint8_t hop_limit; 1975 1976 uint8_t saddr[16]; 1977 uint8_t daddr[16]; 1978 }; 1979 1980 /* Type used to store a checksum (in host byte order) that hasn't been 1981 * folded yet. 1982 */ 1983 #define rawsum_t uint32_t 1984 1985 rawsum_t nm_os_csum_raw(uint8_t *data, size_t len, rawsum_t cur_sum); 1986 uint16_t nm_os_csum_ipv4(struct nm_iphdr *iph); 1987 void nm_os_csum_tcpudp_ipv4(struct nm_iphdr *iph, void *data, 1988 size_t datalen, uint16_t *check); 1989 void nm_os_csum_tcpudp_ipv6(struct nm_ipv6hdr *ip6h, void *data, 1990 size_t datalen, uint16_t *check); 1991 uint16_t nm_os_csum_fold(rawsum_t cur_sum); 1992 1993 void bdg_mismatch_datapath(struct netmap_vp_adapter *na, 1994 struct netmap_vp_adapter *dst_na, 1995 const struct nm_bdg_fwd *ft_p, 1996 struct netmap_ring *dst_ring, 1997 u_int *j, u_int lim, u_int *howmany); 1998 1999 /* persistent virtual port routines */ 2000 int nm_os_vi_persist(const char *, struct ifnet **); 2001 void nm_os_vi_detach(struct ifnet *); 2002 void nm_os_vi_init_index(void); 2003 2004 /* 2005 * kernel thread routines 2006 */ 2007 struct nm_kthread; /* OS-specific kthread - opaque */ 2008 typedef void (*nm_kthread_worker_fn_t)(void *data); 2009 2010 /* kthread configuration */ 2011 struct nm_kthread_cfg { 2012 long type; /* kthread type/identifier */ 2013 struct ptnet_ring_cfg event; /* event/ioctl fd */ 2014 nm_kthread_worker_fn_t worker_fn; /* worker function */ 2015 void *worker_private;/* worker parameter */ 2016 int attach_user; /* attach kthread to user process */ 2017 }; 2018 /* kthread configuration */ 2019 struct nm_kthread *nm_os_kthread_create(struct nm_kthread_cfg *cfg); 2020 int nm_os_kthread_start(struct nm_kthread *); 2021 void nm_os_kthread_stop(struct nm_kthread *); 2022 void nm_os_kthread_delete(struct nm_kthread *); 2023 void nm_os_kthread_wakeup_worker(struct nm_kthread *nmk); 2024 void nm_os_kthread_send_irq(struct nm_kthread *); 2025 void nm_os_kthread_set_affinity(struct nm_kthread *, int); 2026 u_int nm_os_ncpus(void); 2027 2028 #ifdef WITH_PTNETMAP_HOST 2029 /* 2030 * netmap adapter for host ptnetmap ports 2031 */ 2032 struct netmap_pt_host_adapter { 2033 struct netmap_adapter up; 2034 2035 struct netmap_adapter *parent; 2036 int (*parent_nm_notify)(struct netmap_kring *kring, int flags); 2037 void *ptns; 2038 }; 2039 /* ptnetmap HOST routines */ 2040 int netmap_get_pt_host_na(struct nmreq *nmr, struct netmap_adapter **na, int create); 2041 int ptnetmap_ctl(struct nmreq *nmr, struct netmap_adapter *na); 2042 static inline int 2043 nm_ptnetmap_host_on(struct netmap_adapter *na) 2044 { 2045 return na && na->na_flags & NAF_PTNETMAP_HOST; 2046 } 2047 #else /* !WITH_PTNETMAP_HOST */ 2048 #define netmap_get_pt_host_na(nmr, _2, _3) \ 2049 ((nmr)->nr_flags & (NR_PTNETMAP_HOST) ? EOPNOTSUPP : 0) 2050 #define ptnetmap_ctl(_1, _2) EINVAL 2051 #define nm_ptnetmap_host_on(_1) EINVAL 2052 #endif /* !WITH_PTNETMAP_HOST */ 2053 2054 #ifdef WITH_PTNETMAP_GUEST 2055 /* ptnetmap GUEST routines */ 2056 2057 typedef uint32_t (*nm_pt_guest_ptctl_t)(struct ifnet *, uint32_t); 2058 2059 /* 2060 * netmap adapter for guest ptnetmap ports 2061 */ 2062 struct netmap_pt_guest_adapter { 2063 /* The netmap adapter to be used by netmap applications. 2064 * This field must be the first, to allow upcast. */ 2065 struct netmap_hw_adapter hwup; 2066 2067 /* The netmap adapter to be used by the driver. */ 2068 struct netmap_hw_adapter dr; 2069 2070 void *csb; 2071 2072 /* Reference counter to track users of backend netmap port: the 2073 * network stack and netmap clients. 2074 * Used to decide when we need (de)allocate krings/rings and 2075 * start (stop) ptnetmap kthreads. */ 2076 int backend_regifs; 2077 2078 }; 2079 2080 int netmap_pt_guest_attach(struct netmap_adapter *, void *, 2081 unsigned int, nm_pt_guest_ptctl_t); 2082 struct ptnet_ring; 2083 bool netmap_pt_guest_txsync(struct ptnet_ring *ptring, struct netmap_kring *kring, 2084 int flags); 2085 bool netmap_pt_guest_rxsync(struct ptnet_ring *ptring, struct netmap_kring *kring, 2086 int flags); 2087 int ptnet_nm_krings_create(struct netmap_adapter *na); 2088 void ptnet_nm_krings_delete(struct netmap_adapter *na); 2089 void ptnet_nm_dtor(struct netmap_adapter *na); 2090 #endif /* WITH_PTNETMAP_GUEST */ 2091 2092 #endif /* _NET_NETMAP_KERN_H_ */ 2093