1 /* 2 * Copyright (C) 2011-2013 Matteo Landi, Luigi Rizzo. All rights reserved. 3 * Copyright (C) 2013 Universita` di Pisa. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 /* 28 * $FreeBSD: head/sys/dev/netmap/netmap_kern.h 238985 2012-08-02 11:59:43Z luigi $ 29 * 30 * The header contains the definitions of constants and function 31 * prototypes used only in kernelspace. 32 */ 33 34 #ifndef _NET_NETMAP_KERN_H_ 35 #define _NET_NETMAP_KERN_H_ 36 37 #define WITH_VALE // comment out to disable VALE support 38 39 #define likely(x) __builtin_expect((long)!!(x), 1L) 40 #define unlikely(x) __builtin_expect((long)!!(x), 0L) 41 42 #define NM_LOCK_T struct lock 43 #define NMG_LOCK_T struct lock 44 #define NMG_LOCK_INIT() lockinit(&netmap_global_lock, \ 45 "netmap global lock", 0, LK_CANRECURSE) 46 #define NMG_LOCK_DESTROY() lockuninit(&netmap_global_lock) 47 #define NMG_LOCK() lockmgr(&netmap_global_lock, LK_EXCLUSIVE) 48 #define NMG_UNLOCK() lockmgr(&netmap_global_lock, LK_RELEASE) 49 #define NMG_LOCK_ASSERT() KKASSERT(lockstatus(&netmap_global_lock, NULL) != 0) 50 51 #define NM_SELINFO_T struct kqinfo 52 #define MBUF_LEN(m) ((m)->m_pkthdr.len) 53 #define MBUF_IFP(m) ((m)->m_pkthdr.rcvif) 54 #define NM_SEND_UP(ifp, m) ((ifp)->if_input(ifp, m, NULL, -1)) 55 56 #define NM_ATOMIC_T volatile int // XXX ? 57 /* atomic operations */ 58 #include <machine/atomic.h> 59 #define NM_ATOMIC_TEST_AND_SET(p) (!atomic_cmpset_acq_int((p), 0, 1)) 60 #define NM_ATOMIC_CLEAR(p) atomic_store_rel_int((p), 0) 61 62 #define prefetch(x) __builtin_prefetch(x) 63 64 #define mb() cpu_mfence() 65 #define rmb() cpu_lfence() 66 #define wmb() cpu_sfence() 67 68 MALLOC_DECLARE(M_NETMAP); 69 70 // XXX linux struct, not used in FreeBSD 71 struct net_device_ops { 72 }; 73 struct hrtimer { 74 }; 75 76 #define IFCAP_NETMAP 0x8000 /* XXX move to <net/if.h> */ 77 78 #define ND(format, ...) 79 #define D(format, ...) \ 80 do { \ 81 struct timeval __xxts; \ 82 microtime(&__xxts); \ 83 kprintf("%03d.%06d %s [%d] " format "\n", \ 84 (int)__xxts.tv_sec % 1000, (int)__xxts.tv_usec, \ 85 __FUNCTION__, __LINE__, ##__VA_ARGS__); \ 86 } while (0) 87 88 /* rate limited, lps indicates how many per second */ 89 #define RD(lps, format, ...) \ 90 do { \ 91 static int t0, __cnt; \ 92 if (t0 != time_second) { \ 93 t0 = time_second; \ 94 __cnt = 0; \ 95 } \ 96 if (__cnt++ < lps) \ 97 D(format, ##__VA_ARGS__); \ 98 } while (0) 99 100 struct netmap_adapter; 101 struct nm_bdg_fwd; 102 struct nm_bridge; 103 struct netmap_priv_d; 104 105 const char *nm_dump_buf(char *p, int len, int lim, char *dst); 106 107 #include <net/netmap/netmap_mbq.h> 108 109 extern NMG_LOCK_T netmap_global_lock; 110 111 /* 112 * private, kernel view of a ring. Keeps track of the status of 113 * a ring across system calls. 114 * 115 * nr_hwcur index of the next buffer to refill. 116 * It corresponds to ring->cur - ring->reserved 117 * 118 * nr_hwavail the number of slots "owned" by userspace. 119 * nr_hwavail =:= ring->avail + ring->reserved 120 * 121 * The indexes in the NIC and netmap rings are offset by nkr_hwofs slots. 122 * This is so that, on a reset, buffers owned by userspace are not 123 * modified by the kernel. In particular: 124 * RX rings: the next empty buffer (hwcur + hwavail + hwofs) coincides with 125 * the next empty buffer as known by the hardware (next_to_check or so). 126 * TX rings: hwcur + hwofs coincides with next_to_send 127 * 128 * Clients cannot issue concurrent syscall on a ring. The system 129 * detects this and reports an error using two flags, 130 * NKR_WBUSY and NKR_RBUSY 131 * For received packets, slot->flags is set to nkr_slot_flags 132 * so we can provide a proper initial value (e.g. set NS_FORWARD 133 * when operating in 'transparent' mode). 134 * 135 * The following fields are used to implement lock-free copy of packets 136 * from input to output ports in VALE switch: 137 * nkr_hwlease buffer after the last one being copied. 138 * A writer in nm_bdg_flush reserves N buffers 139 * from nr_hwlease, advances it, then does the 140 * copy outside the lock. 141 * In RX rings (used for VALE ports), 142 * nkr_hwcur + nkr_hwavail <= nkr_hwlease < nkr_hwcur+N-1 143 * In TX rings (used for NIC or host stack ports) 144 * nkr_hwcur <= nkr_hwlease < nkr_hwcur+ nkr_hwavail 145 * nkr_leases array of nkr_num_slots where writers can report 146 * completion of their block. NR_NOSLOT (~0) indicates 147 * that the writer has not finished yet 148 * nkr_lease_idx index of next free slot in nr_leases, to be assigned 149 * 150 * The kring is manipulated by txsync/rxsync and generic netmap function. 151 * q_lock is used to arbitrate access to the kring from within the netmap 152 * code, and this and other protections guarantee that there is never 153 * more than 1 concurrent call to txsync or rxsync. So we are free 154 * to manipulate the kring from within txsync/rxsync without any extra 155 * locks. 156 */ 157 struct netmap_kring { 158 struct netmap_ring *ring; 159 uint32_t nr_hwcur; 160 uint32_t nr_hwavail; 161 uint32_t nr_kflags; /* private driver flags */ 162 int32_t nr_hwreserved; 163 #define NKR_PENDINTR 0x1 // Pending interrupt. 164 uint32_t nkr_num_slots; 165 int32_t nkr_hwofs; /* offset between NIC and netmap ring */ 166 167 uint16_t nkr_slot_flags; /* initial value for flags */ 168 struct netmap_adapter *na; 169 struct nm_bdg_fwd *nkr_ft; 170 uint32_t *nkr_leases; 171 #define NR_NOSLOT ((uint32_t)~0) 172 uint32_t nkr_hwlease; 173 uint32_t nkr_lease_idx; 174 175 NM_SELINFO_T si; /* poll/select wait queue */ 176 NM_LOCK_T q_lock; /* protects kring and ring. */ 177 NM_ATOMIC_T nr_busy; /* prevent concurrent syscalls */ 178 179 volatile int nkr_stopped; 180 181 /* support for adapters without native netmap support. 182 * On tx rings we preallocate an array of tx buffers 183 * (same size as the netmap ring), on rx rings we 184 * store incoming packets in a queue. 185 * XXX who writes to the rx queue ? 186 */ 187 struct mbuf **tx_pool; 188 u_int nr_ntc; /* Emulation of a next-to-clean RX ring pointer. */ 189 struct mbq rx_queue; /* A queue for intercepted rx mbufs. */ 190 191 } __attribute__((__aligned__(64))); 192 193 194 /* return the next index, with wraparound */ 195 static inline uint32_t 196 nm_next(uint32_t i, uint32_t lim) 197 { 198 return unlikely (i == lim) ? 0 : i + 1; 199 } 200 201 /* 202 * 203 * Here is the layout for the Rx and Tx rings. 204 205 RxRING TxRING 206 207 +-----------------+ +-----------------+ 208 | | | | 209 |XXX free slot XXX| |XXX free slot XXX| 210 +-----------------+ +-----------------+ 211 | |<-hwcur | |<-hwcur 212 | reserved h | | (ready | 213 +----------- w -+ | to be | 214 cur->| a | | sent) h | 215 | v | +---------- w | 216 | a | cur->| (being a | 217 | i | | prepared) v | 218 | avail l | | a | 219 +-----------------+ + a ------ i + 220 | | ... | v l |<-hwlease 221 | (being | ... | a | ... 222 | prepared) | ... | i | ... 223 +-----------------+ ... | l | ... 224 | |<-hwlease +-----------------+ 225 | | | | 226 | | | | 227 | | | | 228 | | | | 229 +-----------------+ +-----------------+ 230 231 * The cur/avail (user view) and hwcur/hwavail (kernel view) 232 * are used in the normal operation of the card. 233 * 234 * When a ring is the output of a switch port (Rx ring for 235 * a VALE port, Tx ring for the host stack or NIC), slots 236 * are reserved in blocks through 'hwlease' which points 237 * to the next unused slot. 238 * On an Rx ring, hwlease is always after hwavail, 239 * and completions cause avail to advance. 240 * On a Tx ring, hwlease is always between cur and hwavail, 241 * and completions cause cur to advance. 242 * 243 * nm_kr_space() returns the maximum number of slots that 244 * can be assigned. 245 * nm_kr_lease() reserves the required number of buffers, 246 * advances nkr_hwlease and also returns an entry in 247 * a circular array where completions should be reported. 248 */ 249 250 251 252 253 enum txrx { NR_RX = 0, NR_TX = 1 }; 254 255 /* 256 * The "struct netmap_adapter" extends the "struct adapter" 257 * (or equivalent) device descriptor. 258 * It contains all base fields needed to support netmap operation. 259 * There are in fact different types of netmap adapters 260 * (native, generic, VALE switch...) so a netmap_adapter is 261 * just the first field in the derived type. 262 */ 263 struct netmap_adapter { 264 /* 265 * On linux we do not have a good way to tell if an interface 266 * is netmap-capable. So we use the following trick: 267 * NA(ifp) points here, and the first entry (which hopefully 268 * always exists and is at least 32 bits) contains a magic 269 * value which we can use to detect that the interface is good. 270 */ 271 uint32_t magic; 272 uint32_t na_flags; /* future place for IFCAP_NETMAP */ 273 #define NAF_SKIP_INTR 1 /* use the regular interrupt handler. 274 * useful during initialization 275 */ 276 #define NAF_SW_ONLY 2 /* forward packets only to sw adapter */ 277 #define NAF_BDG_MAYSLEEP 4 /* the bridge is allowed to sleep when 278 * forwarding packets coming from this 279 * interface 280 */ 281 #define NAF_MEM_OWNER 8 /* the adapter is responsible for the 282 * deallocation of the memory allocator 283 */ 284 #define NAF_NATIVE_ON 16 /* the adapter is native and the attached 285 * interface is in netmap mode 286 */ 287 int active_fds; /* number of user-space descriptors using this 288 interface, which is equal to the number of 289 struct netmap_if objs in the mapped region. */ 290 291 u_int num_rx_rings; /* number of adapter receive rings */ 292 u_int num_tx_rings; /* number of adapter transmit rings */ 293 294 u_int num_tx_desc; /* number of descriptor in each queue */ 295 u_int num_rx_desc; 296 297 /* tx_rings and rx_rings are private but allocated 298 * as a contiguous chunk of memory. Each array has 299 * N+1 entries, for the adapter queues and for the host queue. 300 */ 301 struct netmap_kring *tx_rings; /* array of TX rings. */ 302 struct netmap_kring *rx_rings; /* array of RX rings. */ 303 void *tailroom; /* space below the rings array */ 304 /* (used for leases) */ 305 306 307 NM_SELINFO_T tx_si, rx_si; /* global wait queues */ 308 309 /* copy of if_qflush and if_transmit pointers, to intercept 310 * packets from the network stack when netmap is active. 311 */ 312 int (*if_transmit)(struct ifnet *, struct mbuf *); 313 314 /* references to the ifnet and device routines, used by 315 * the generic netmap functions. 316 */ 317 struct ifnet *ifp; /* adapter is ifp->if_softc */ 318 319 /* private cleanup */ 320 void (*nm_dtor)(struct netmap_adapter *); 321 322 int (*nm_register)(struct netmap_adapter *, int onoff); 323 324 int (*nm_txsync)(struct netmap_adapter *, u_int ring, int flags); 325 int (*nm_rxsync)(struct netmap_adapter *, u_int ring, int flags); 326 #define NAF_FORCE_READ 1 327 #define NAF_FORCE_RECLAIM 2 328 /* return configuration information */ 329 int (*nm_config)(struct netmap_adapter *, 330 u_int *txr, u_int *txd, u_int *rxr, u_int *rxd); 331 int (*nm_krings_create)(struct netmap_adapter *); 332 void (*nm_krings_delete)(struct netmap_adapter *); 333 int (*nm_notify)(struct netmap_adapter *, 334 u_int ring, enum txrx, int flags); 335 #define NAF_GLOBAL_NOTIFY 4 336 #define NAF_DISABLE_NOTIFY 8 337 338 /* standard refcount to control the lifetime of the adapter 339 * (it should be equal to the lifetime of the corresponding ifp) 340 */ 341 int na_refcount; 342 343 /* memory allocator (opaque) 344 * We also cache a pointer to the lut_entry for translating 345 * buffer addresses, and the total number of buffers. 346 */ 347 struct netmap_mem_d *nm_mem; 348 struct lut_entry *na_lut; 349 uint32_t na_lut_objtotal; /* max buffer index */ 350 351 /* used internally. If non-null, the interface cannot be bound 352 * from userspace 353 */ 354 void *na_private; 355 }; 356 357 /* 358 * If the NIC is owned by the kernel 359 * (i.e., bridge), neither another bridge nor user can use it; 360 * if the NIC is owned by a user, only users can share it. 361 * Evaluation must be done under NMG_LOCK(). 362 */ 363 #define NETMAP_OWNED_BY_KERN(na) (na->na_private) 364 #define NETMAP_OWNED_BY_ANY(na) \ 365 (NETMAP_OWNED_BY_KERN(na) || (na->active_fds > 0)) 366 367 368 /* 369 * derived netmap adapters for various types of ports 370 */ 371 struct netmap_vp_adapter { /* VALE software port */ 372 struct netmap_adapter up; 373 374 /* 375 * Bridge support: 376 * 377 * bdg_port is the port number used in the bridge; 378 * na_bdg points to the bridge this NA is attached to. 379 */ 380 int bdg_port; 381 struct nm_bridge *na_bdg; 382 int retry; 383 }; 384 385 struct netmap_hw_adapter { /* physical device */ 386 struct netmap_adapter up; 387 388 struct net_device_ops nm_ndo; // XXX linux only 389 }; 390 391 struct netmap_generic_adapter { /* non-native device */ 392 struct netmap_hw_adapter up; 393 394 /* Pointer to a previously used netmap adapter. */ 395 struct netmap_adapter *prev; 396 397 /* generic netmap adapters support: 398 * a net_device_ops struct overrides ndo_select_queue(), 399 * save_if_input saves the if_input hook (FreeBSD), 400 * mit_timer and mit_pending implement rx interrupt mitigation, 401 */ 402 struct net_device_ops generic_ndo; 403 void (*save_if_input)(struct ifnet *, struct mbuf *, 404 const struct pktinfo *, int); 405 406 struct hrtimer mit_timer; 407 int mit_pending; 408 }; 409 410 #ifdef WITH_VALE 411 412 /* bridge wrapper for non VALE ports. It is used to connect real devices to the bridge. 413 * 414 * The real device must already have its own netmap adapter (hwna). The 415 * bridge wrapper and the hwna adapter share the same set of netmap rings and 416 * buffers, but they have two separate sets of krings descriptors, with tx/rx 417 * meanings swapped: 418 * 419 * netmap 420 * bwrap krings rings krings hwna 421 * +------+ +------+ +-----+ +------+ +------+ 422 * |tx_rings->| |\ /| |----| |<-tx_rings| 423 * | | +------+ \ / +-----+ +------+ | | 424 * | | X | | 425 * | | / \ | | 426 * | | +------+/ \+-----+ +------+ | | 427 * |rx_rings->| | | |----| |<-rx_rings| 428 * | | +------+ +-----+ +------+ | | 429 * +------+ +------+ 430 * 431 * - packets coming from the bridge go to the brwap rx rings, which are also the 432 * hwna tx rings. The bwrap notify callback will then complete the hwna tx 433 * (see netmap_bwrap_notify). 434 * - packets coming from the outside go to the hwna rx rings, which are also the 435 * bwrap tx rings. The (overwritten) hwna notify method will then complete 436 * the bridge tx (see netmap_bwrap_intr_notify). 437 * 438 * The bridge wrapper may optionally connect the hwna 'host' rings to the 439 * bridge. This is done by using a second port in the bridge and connecting it 440 * to the 'host' netmap_vp_adapter contained in the netmap_bwrap_adapter. 441 * The brwap host adapter cross-links the hwna host rings in the same way as shown above. 442 * 443 * - packets coming from the bridge and directed to host stack are handled by the 444 * bwrap host notify callback (see netmap_bwrap_host_notify) 445 * - packets coming from the host stack are still handled by the overwritten 446 * hwna notify callback (netmap_bwrap_intr_notify), but are diverted to the 447 * host adapter depending on the ring number. 448 * 449 */ 450 struct netmap_bwrap_adapter { 451 struct netmap_vp_adapter up; 452 struct netmap_vp_adapter host; /* for host rings */ 453 struct netmap_adapter *hwna; /* the underlying device */ 454 455 /* backup of the hwna notify callback */ 456 int (*save_notify)(struct netmap_adapter *, 457 u_int ring, enum txrx, int flags); 458 /* When we attach a physical interface to the bridge, we 459 * allow the controlling process to terminate, so we need 460 * a place to store the netmap_priv_d data structure. 461 * This is only done when physical interfaces are attached to a bridge. 462 */ 463 struct netmap_priv_d *na_kpriv; 464 }; 465 466 467 /* 468 * Available space in the ring. Only used in VALE code 469 */ 470 static inline uint32_t 471 nm_kr_space(struct netmap_kring *k, int is_rx) 472 { 473 int space; 474 475 if (is_rx) { 476 int busy = k->nkr_hwlease - k->nr_hwcur + k->nr_hwreserved; 477 if (busy < 0) 478 busy += k->nkr_num_slots; 479 space = k->nkr_num_slots - 1 - busy; 480 } else { 481 space = k->nr_hwcur + k->nr_hwavail - k->nkr_hwlease; 482 if (space < 0) 483 space += k->nkr_num_slots; 484 } 485 #if 0 486 // sanity check 487 if (k->nkr_hwlease >= k->nkr_num_slots || 488 k->nr_hwcur >= k->nkr_num_slots || 489 k->nr_hwavail >= k->nkr_num_slots || 490 busy < 0 || 491 busy >= k->nkr_num_slots) { 492 D("invalid kring, cur %d avail %d lease %d lease_idx %d lim %d", k->nr_hwcur, k->nr_hwavail, k->nkr_hwlease, 493 k->nkr_lease_idx, k->nkr_num_slots); 494 } 495 #endif 496 return space; 497 } 498 499 500 501 502 /* make a lease on the kring for N positions. return the 503 * lease index 504 */ 505 static inline uint32_t 506 nm_kr_lease(struct netmap_kring *k, u_int n, int is_rx) 507 { 508 uint32_t lim = k->nkr_num_slots - 1; 509 uint32_t lease_idx = k->nkr_lease_idx; 510 511 k->nkr_leases[lease_idx] = NR_NOSLOT; 512 k->nkr_lease_idx = nm_next(lease_idx, lim); 513 514 if (n > nm_kr_space(k, is_rx)) { 515 D("invalid request for %d slots", n); 516 panic("x"); 517 } 518 /* XXX verify that there are n slots */ 519 k->nkr_hwlease += n; 520 if (k->nkr_hwlease > lim) 521 k->nkr_hwlease -= lim + 1; 522 523 if (k->nkr_hwlease >= k->nkr_num_slots || 524 k->nr_hwcur >= k->nkr_num_slots || 525 k->nr_hwavail >= k->nkr_num_slots || 526 k->nkr_lease_idx >= k->nkr_num_slots) { 527 D("invalid kring %s, cur %d avail %d lease %d lease_idx %d lim %d", 528 k->na->ifp->if_xname, 529 k->nr_hwcur, k->nr_hwavail, k->nkr_hwlease, 530 k->nkr_lease_idx, k->nkr_num_slots); 531 } 532 return lease_idx; 533 } 534 535 #endif /* WITH_VALE */ 536 537 /* return update position */ 538 static inline uint32_t 539 nm_kr_rxpos(struct netmap_kring *k) 540 { 541 uint32_t pos = k->nr_hwcur + k->nr_hwavail; 542 if (pos >= k->nkr_num_slots) 543 pos -= k->nkr_num_slots; 544 #if 0 545 if (pos >= k->nkr_num_slots || 546 k->nkr_hwlease >= k->nkr_num_slots || 547 k->nr_hwcur >= k->nkr_num_slots || 548 k->nr_hwavail >= k->nkr_num_slots || 549 k->nkr_lease_idx >= k->nkr_num_slots) { 550 D("invalid kring, cur %d avail %d lease %d lease_idx %d lim %d", k->nr_hwcur, k->nr_hwavail, k->nkr_hwlease, 551 k->nkr_lease_idx, k->nkr_num_slots); 552 } 553 #endif 554 return pos; 555 } 556 557 558 /* 559 * protect against multiple threads using the same ring. 560 * also check that the ring has not been stopped. 561 * We only care for 0 or !=0 as a return code. 562 */ 563 #define NM_KR_BUSY 1 564 #define NM_KR_STOPPED 2 565 566 static __inline void nm_kr_put(struct netmap_kring *kr) 567 { 568 NM_ATOMIC_CLEAR(&kr->nr_busy); 569 } 570 571 static __inline int nm_kr_tryget(struct netmap_kring *kr) 572 { 573 /* check a first time without taking the lock 574 * to avoid starvation for nm_kr_get() 575 */ 576 if (unlikely(kr->nkr_stopped)) { 577 ND("ring %p stopped (%d)", kr, kr->nkr_stopped); 578 return NM_KR_STOPPED; 579 } 580 if (unlikely(NM_ATOMIC_TEST_AND_SET(&kr->nr_busy))) 581 return NM_KR_BUSY; 582 /* check a second time with lock held */ 583 if (unlikely(kr->nkr_stopped)) { 584 ND("ring %p stopped (%d)", kr, kr->nkr_stopped); 585 nm_kr_put(kr); 586 return NM_KR_STOPPED; 587 } 588 return 0; 589 } 590 591 592 /* 593 * The following are support routines used by individual drivers to 594 * support netmap operation. 595 * 596 * netmap_attach() initializes a struct netmap_adapter, allocating the 597 * struct netmap_ring's and the struct selinfo. 598 * 599 * netmap_detach() frees the memory allocated by netmap_attach(). 600 * 601 * netmap_transmit() replaces the if_transmit routine of the interface, 602 * and is used to intercept packets coming from the stack. 603 * 604 * netmap_load_map/netmap_reload_map are helper routines to set/reset 605 * the dmamap for a packet buffer 606 * 607 * netmap_reset() is a helper routine to be called in the driver 608 * when reinitializing a ring. 609 */ 610 int netmap_attach(struct netmap_adapter *); 611 int netmap_attach_common(struct netmap_adapter *); 612 void netmap_detach_common(struct netmap_adapter *na); 613 void netmap_detach(struct ifnet *); 614 int netmap_transmit(struct ifnet *, struct mbuf *); 615 struct netmap_slot *netmap_reset(struct netmap_adapter *na, 616 enum txrx tx, u_int n, u_int new_cur); 617 int netmap_ring_reinit(struct netmap_kring *); 618 619 620 /* 621 * Support routines to be used with the VALE switch 622 */ 623 int netmap_update_config(struct netmap_adapter *na); 624 int netmap_krings_create(struct netmap_adapter *na, u_int ntx, u_int nrx, u_int tailroom); 625 void netmap_krings_delete(struct netmap_adapter *na); 626 627 struct netmap_if * 628 netmap_do_regif(struct netmap_priv_d *priv, struct netmap_adapter *na, 629 uint16_t ringid, int *err); 630 631 632 633 u_int nm_bound_var(u_int *v, u_int dflt, u_int lo, u_int hi, const char *msg); 634 int netmap_get_na(struct nmreq *nmr, struct netmap_adapter **na, int create); 635 int netmap_get_hw_na(struct ifnet *ifp, struct netmap_adapter **na); 636 637 #ifdef WITH_VALE 638 /* 639 * The following bridge-related interfaces are used by other kernel modules 640 * In the version that only supports unicast or broadcast, the lookup 641 * function can return 0 .. NM_BDG_MAXPORTS-1 for regular ports, 642 * NM_BDG_MAXPORTS for broadcast, NM_BDG_MAXPORTS+1 for unknown. 643 * XXX in practice "unknown" might be handled same as broadcast. 644 */ 645 typedef u_int (*bdg_lookup_fn_t)(char *buf, u_int len, 646 uint8_t *ring_nr, struct netmap_vp_adapter *); 647 u_int netmap_bdg_learning(char *, u_int, uint8_t *, 648 struct netmap_vp_adapter *); 649 650 #define NM_BDG_MAXPORTS 254 /* up to 254 */ 651 #define NM_BDG_BROADCAST NM_BDG_MAXPORTS 652 #define NM_BDG_NOPORT (NM_BDG_MAXPORTS+1) 653 654 #define NM_NAME "vale" /* prefix for bridge port name */ 655 656 657 /* these are redefined in case of no VALE support */ 658 int netmap_get_bdg_na(struct nmreq *nmr, struct netmap_adapter **na, int create); 659 void netmap_init_bridges(void); 660 int netmap_bdg_ctl(struct nmreq *nmr, bdg_lookup_fn_t func); 661 662 #else /* !WITH_VALE */ 663 #define netmap_get_bdg_na(_1, _2, _3) 0 664 #define netmap_init_bridges(_1) 665 #define netmap_bdg_ctl(_1, _2) EINVAL 666 #endif /* !WITH_VALE */ 667 668 /* Various prototypes */ 669 struct dev_kqfilter_args; /* XXX this shouldn't be here */ 670 int netmap_kqfilter(struct dev_kqfilter_args *ap); 671 672 673 int netmap_init(void); 674 void netmap_fini(void); 675 int netmap_get_memory(struct netmap_priv_d* p); 676 void netmap_dtor(void *data); 677 int netmap_dtor_locked(struct netmap_priv_d *priv); 678 679 struct dev_ioctl_args; /* XXX this shouldn't be here */ 680 int netmap_ioctl(struct dev_ioctl_args *ap); 681 682 /* netmap_adapter creation/destruction */ 683 #define NM_IFPNAME(ifp) ((ifp) ? (ifp)->if_xname : "zombie") 684 #define NM_DEBUG_PUTGET 1 685 686 #ifdef NM_DEBUG_PUTGET 687 688 #define NM_DBG(f) __##f 689 690 void __netmap_adapter_get(struct netmap_adapter *na); 691 692 #define netmap_adapter_get(na) \ 693 do { \ 694 struct netmap_adapter *__na = na; \ 695 D("getting %p:%s (%d)", __na, NM_IFPNAME(__na->ifp), __na->na_refcount); \ 696 __netmap_adapter_get(__na); \ 697 } while (0) 698 699 int __netmap_adapter_put(struct netmap_adapter *na); 700 701 #define netmap_adapter_put(na) \ 702 do { \ 703 struct netmap_adapter *__na = na; \ 704 D("putting %p:%s (%d)", __na, NM_IFPNAME(__na->ifp), __na->na_refcount); \ 705 __netmap_adapter_put(__na); \ 706 } while (0) 707 708 #else /* !NM_DEBUG_PUTGET */ 709 710 #define NM_DBG(f) f 711 void netmap_adapter_get(struct netmap_adapter *na); 712 int netmap_adapter_put(struct netmap_adapter *na); 713 714 #endif /* !NM_DEBUG_PUTGET */ 715 716 717 718 extern u_int netmap_buf_size; 719 #define NETMAP_BUF_SIZE netmap_buf_size // XXX remove 720 extern int netmap_mitigate; 721 extern int netmap_no_pendintr; 722 extern u_int netmap_total_buffers; 723 extern char *netmap_buffer_base; 724 extern int netmap_verbose; // XXX debugging 725 enum { /* verbose flags */ 726 NM_VERB_ON = 1, /* generic verbose */ 727 NM_VERB_HOST = 0x2, /* verbose host stack */ 728 NM_VERB_RXSYNC = 0x10, /* verbose on rxsync/txsync */ 729 NM_VERB_TXSYNC = 0x20, 730 NM_VERB_RXINTR = 0x100, /* verbose on rx/tx intr (driver) */ 731 NM_VERB_TXINTR = 0x200, 732 NM_VERB_NIC_RXSYNC = 0x1000, /* verbose on rx/tx intr (driver) */ 733 NM_VERB_NIC_TXSYNC = 0x2000, 734 }; 735 736 extern int netmap_txsync_retry; 737 extern int netmap_generic_mit; 738 extern int netmap_generic_ringsize; 739 740 /* 741 * NA returns a pointer to the struct netmap adapter from the ifp, 742 * WNA is used to write it. 743 */ 744 #ifndef WNA 745 #define WNA(_ifp) (_ifp)->if_unused7 /* XXX better name ;) */ 746 #endif 747 #define NA(_ifp) ((struct netmap_adapter *)WNA(_ifp)) 748 749 /* 750 * Macros to determine if an interface is netmap capable or netmap enabled. 751 * See the magic field in struct netmap_adapter. 752 */ 753 /* 754 * on FreeBSD just use if_capabilities and if_capenable. 755 */ 756 #define NETMAP_CAPABLE(ifp) (NA(ifp) && \ 757 (ifp)->if_capabilities & IFCAP_NETMAP ) 758 759 #define NETMAP_SET_CAPABLE(ifp) \ 760 (ifp)->if_capabilities |= IFCAP_NETMAP 761 762 /* Callback invoked by the dma machinery after a successfull dmamap_load */ 763 static void netmap_dmamap_cb(__unused void *arg, 764 __unused bus_dma_segment_t * segs, __unused int nseg, __unused int error) 765 { 766 } 767 768 /* bus_dmamap_load wrapper: call aforementioned function if map != NULL. 769 * XXX can we do it without a callback ? 770 */ 771 static inline void 772 netmap_load_map(bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 773 { 774 if (map) 775 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE, 776 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT); 777 } 778 779 /* update the map when a buffer changes. */ 780 static inline void 781 netmap_reload_map(bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 782 { 783 if (map) { 784 bus_dmamap_unload(tag, map); 785 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE, 786 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT); 787 } 788 } 789 790 /* 791 * functions to map NIC to KRING indexes (n2k) and vice versa (k2n) 792 */ 793 static inline int 794 netmap_idx_n2k(struct netmap_kring *kr, int idx) 795 { 796 int n = kr->nkr_num_slots; 797 idx += kr->nkr_hwofs; 798 if (idx < 0) 799 return idx + n; 800 else if (idx < n) 801 return idx; 802 else 803 return idx - n; 804 } 805 806 807 static inline int 808 netmap_idx_k2n(struct netmap_kring *kr, int idx) 809 { 810 int n = kr->nkr_num_slots; 811 idx -= kr->nkr_hwofs; 812 if (idx < 0) 813 return idx + n; 814 else if (idx < n) 815 return idx; 816 else 817 return idx - n; 818 } 819 820 821 /* Entries of the look-up table. */ 822 struct lut_entry { 823 void *vaddr; /* virtual address. */ 824 vm_paddr_t paddr; /* physical address. */ 825 }; 826 827 struct netmap_obj_pool; 828 extern struct lut_entry *netmap_buffer_lut; 829 #define NMB_VA(i) (netmap_buffer_lut[i].vaddr) 830 #define NMB_PA(i) (netmap_buffer_lut[i].paddr) 831 832 /* 833 * NMB return the virtual address of a buffer (buffer 0 on bad index) 834 * PNMB also fills the physical address 835 */ 836 static inline void * 837 NMB(struct netmap_slot *slot) 838 { 839 uint32_t i = slot->buf_idx; 840 return (unlikely(i >= netmap_total_buffers)) ? NMB_VA(0) : NMB_VA(i); 841 } 842 843 static inline void * 844 PNMB(struct netmap_slot *slot, uint64_t *pp) 845 { 846 uint32_t i = slot->buf_idx; 847 void *ret = (i >= netmap_total_buffers) ? NMB_VA(0) : NMB_VA(i); 848 849 *pp = (i >= netmap_total_buffers) ? NMB_PA(0) : NMB_PA(i); 850 return ret; 851 } 852 853 /* Generic version of NMB, which uses device-specific memory. */ 854 static inline void * 855 BDG_NMB(struct netmap_adapter *na, struct netmap_slot *slot) 856 { 857 struct lut_entry *lut = na->na_lut; 858 uint32_t i = slot->buf_idx; 859 return (unlikely(i >= na->na_lut_objtotal)) ? 860 lut[0].vaddr : lut[i].vaddr; 861 } 862 863 /* default functions to handle rx/tx interrupts */ 864 int netmap_rx_irq(struct ifnet *, u_int, u_int *); 865 #define netmap_tx_irq(_n, _q) netmap_rx_irq(_n, _q, NULL) 866 int netmap_common_irq(struct ifnet *, u_int, u_int *work_done); 867 868 869 void netmap_txsync_to_host(struct netmap_adapter *na); 870 void netmap_disable_all_rings(struct ifnet *); 871 void netmap_enable_all_rings(struct ifnet *); 872 void netmap_disable_ring(struct netmap_kring *kr); 873 874 875 /* Structure associated to each thread which registered an interface. 876 * 877 * The first 4 fields of this structure are written by NIOCREGIF and 878 * read by poll() and NIOC?XSYNC. 879 * There is low contention among writers (actually, a correct user program 880 * should have no contention among writers) and among writers and readers, 881 * so we use a single global lock to protect the structure initialization. 882 * Since initialization involves the allocation of memory, we reuse the memory 883 * allocator lock. 884 * Read access to the structure is lock free. Readers must check that 885 * np_nifp is not NULL before using the other fields. 886 * If np_nifp is NULL initialization has not been performed, so they should 887 * return an error to userlevel. 888 * 889 * The ref_done field is used to regulate access to the refcount in the 890 * memory allocator. The refcount must be incremented at most once for 891 * each open("/dev/netmap"). The increment is performed by the first 892 * function that calls netmap_get_memory() (currently called by 893 * mmap(), NIOCGINFO and NIOCREGIF). 894 * If the refcount is incremented, it is then decremented when the 895 * private structure is destroyed. 896 */ 897 struct netmap_priv_d { 898 struct netmap_if * volatile np_nifp; /* netmap if descriptor. */ 899 900 struct netmap_adapter *np_na; 901 int np_ringid; /* from the ioctl */ 902 u_int np_qfirst, np_qlast; /* range of rings to scan */ 903 uint16_t np_txpoll; 904 905 struct netmap_mem_d *np_mref; /* use with NMG_LOCK held */ 906 /* np_refcount is only used on FreeBSD */ 907 int np_refcount; /* use with NMG_LOCK held */ 908 }; 909 910 911 /* 912 * generic netmap emulation for devices that do not have 913 * native netmap support. 914 * XXX generic_netmap_register() is only exported to implement 915 * nma_is_generic(). 916 */ 917 int generic_netmap_register(struct netmap_adapter *na, int enable); 918 int generic_netmap_attach(struct ifnet *ifp); 919 920 int netmap_catch_rx(struct netmap_adapter *na, int intercept); 921 void generic_rx_handler(struct ifnet *ifp, struct mbuf *m, 922 const struct pktinfo *, int); 923 void netmap_catch_packet_steering(struct netmap_generic_adapter *na, int enable); 924 int generic_xmit_frame(struct ifnet *ifp, struct mbuf *m, void *addr, u_int len, u_int ring_nr); 925 int generic_find_num_desc(struct ifnet *ifp, u_int *tx, u_int *rx); 926 void generic_find_num_queues(struct ifnet *ifp, u_int *txq, u_int *rxq); 927 928 static __inline int 929 nma_is_generic(struct netmap_adapter *na) 930 { 931 return na->nm_register == generic_netmap_register; 932 } 933 934 /* 935 * netmap_mitigation API. This is used by the generic adapter 936 * to reduce the number of interrupt requests/selwakeup 937 * to clients on incoming packets. 938 */ 939 void netmap_mitigation_init(struct netmap_generic_adapter *na); 940 void netmap_mitigation_start(struct netmap_generic_adapter *na); 941 void netmap_mitigation_restart(struct netmap_generic_adapter *na); 942 int netmap_mitigation_active(struct netmap_generic_adapter *na); 943 void netmap_mitigation_cleanup(struct netmap_generic_adapter *na); 944 945 // int generic_timer_handler(struct hrtimer *t); 946 947 #endif /* _NET_NETMAP_KERN_H_ */ 948