1 /* 2 * Copyright (C) 2011-2012 Matteo Landi, Luigi Rizzo. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23 * SUCH DAMAGE. 24 */ 25 26 /* 27 * $FreeBSD$ 28 * $Id: netmap_kern.h 11829 2012-09-26 04:06:34Z 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 NETMAP_MEM2 // use the new memory allocator 38 39 #if defined(__FreeBSD__) 40 #define likely(x) __builtin_expect(!!(x), 1) 41 #define unlikely(x) __builtin_expect(!!(x), 0) 42 43 #define NM_LOCK_T struct mtx 44 #define NM_SELINFO_T struct selinfo 45 #define MBUF_LEN(m) ((m)->m_pkthdr.len) 46 #define NM_SEND_UP(ifp, m) ((ifp)->if_input)(ifp, m) 47 #elif defined (linux) 48 #define NM_LOCK_T spinlock_t 49 #define NM_SELINFO_T wait_queue_head_t 50 #define MBUF_LEN(m) ((m)->len) 51 #define NM_SEND_UP(ifp, m) netif_rx(m) 52 53 #ifndef DEV_NETMAP 54 #define DEV_NETMAP 55 #endif 56 57 /* 58 * IFCAP_NETMAP goes into net_device's priv_flags (if_capenable). 59 * This was 16 bits up to linux 2.6.36, so we need a 16 bit value on older 60 * platforms and tolerate the clash with IFF_DYNAMIC and IFF_BRIDGE_PORT. 61 * For the 32-bit value, 0x100000 has no clashes until at least 3.5.1 62 */ 63 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37) 64 #define IFCAP_NETMAP 0x8000 65 #else 66 #define IFCAP_NETMAP 0x100000 67 #endif 68 69 #elif defined (__APPLE__) 70 #warning apple support is incomplete. 71 #define likely(x) __builtin_expect(!!(x), 1) 72 #define unlikely(x) __builtin_expect(!!(x), 0) 73 #define NM_LOCK_T IOLock * 74 #define NM_SELINFO_T struct selinfo 75 #define MBUF_LEN(m) ((m)->m_pkthdr.len) 76 #define NM_SEND_UP(ifp, m) ((ifp)->if_input)(ifp, m) 77 78 #else 79 #error unsupported platform 80 #endif 81 82 #define ND(format, ...) 83 #define D(format, ...) \ 84 do { \ 85 struct timeval __xxts; \ 86 microtime(&__xxts); \ 87 printf("%03d.%06d %s [%d] " format "\n", \ 88 (int)__xxts.tv_sec % 1000, (int)__xxts.tv_usec, \ 89 __FUNCTION__, __LINE__, ##__VA_ARGS__); \ 90 } while (0) 91 92 /* rate limited, lps indicates how many per second */ 93 #define RD(lps, format, ...) \ 94 do { \ 95 static int t0, __cnt; \ 96 if (t0 != time_second) { \ 97 t0 = time_second; \ 98 __cnt = 0; \ 99 } \ 100 if (__cnt++ < lps) \ 101 D(format, ##__VA_ARGS__); \ 102 } while (0) 103 104 struct netmap_adapter; 105 106 /* 107 * private, kernel view of a ring. Keeps track of the status of 108 * a ring across system calls. 109 * 110 * nr_hwcur index of the next buffer to refill. 111 * It corresponds to ring->cur - ring->reserved 112 * 113 * nr_hwavail the number of slots "owned" by userspace. 114 * nr_hwavail =:= ring->avail + ring->reserved 115 * 116 * The indexes in the NIC and netmap rings are offset by nkr_hwofs slots. 117 * This is so that, on a reset, buffers owned by userspace are not 118 * modified by the kernel. In particular: 119 * RX rings: the next empty buffer (hwcur + hwavail + hwofs) coincides with 120 * the next empty buffer as known by the hardware (next_to_check or so). 121 * TX rings: hwcur + hwofs coincides with next_to_send 122 * 123 * For received packets, slot->flags is set to nkr_slot_flags 124 * so we can provide a proper initial value (e.g. set NS_FORWARD 125 * when operating in 'transparent' mode). 126 */ 127 struct netmap_kring { 128 struct netmap_ring *ring; 129 u_int nr_hwcur; 130 int nr_hwavail; 131 u_int nr_kflags; /* private driver flags */ 132 #define NKR_PENDINTR 0x1 // Pending interrupt. 133 u_int nkr_num_slots; 134 135 uint16_t nkr_slot_flags; /* initial value for flags */ 136 int nkr_hwofs; /* offset between NIC and netmap ring */ 137 struct netmap_adapter *na; 138 NM_SELINFO_T si; /* poll/select wait queue */ 139 NM_LOCK_T q_lock; /* used if no device lock available */ 140 } __attribute__((__aligned__(64))); 141 142 /* 143 * This struct extends the 'struct adapter' (or 144 * equivalent) device descriptor. It contains all fields needed to 145 * support netmap operation. 146 */ 147 struct netmap_adapter { 148 /* 149 * On linux we do not have a good way to tell if an interface 150 * is netmap-capable. So we use the following trick: 151 * NA(ifp) points here, and the first entry (which hopefully 152 * always exists and is at least 32 bits) contains a magic 153 * value which we can use to detect that the interface is good. 154 */ 155 uint32_t magic; 156 uint32_t na_flags; /* future place for IFCAP_NETMAP */ 157 #define NAF_SKIP_INTR 1 /* use the regular interrupt handler. 158 * useful during initialization 159 */ 160 int refcount; /* number of user-space descriptors using this 161 interface, which is equal to the number of 162 struct netmap_if objs in the mapped region. */ 163 /* 164 * The selwakeup in the interrupt thread can use per-ring 165 * and/or global wait queues. We track how many clients 166 * of each type we have so we can optimize the drivers, 167 * and especially avoid huge contention on the locks. 168 */ 169 int na_single; /* threads attached to a single hw queue */ 170 int na_multi; /* threads attached to multiple hw queues */ 171 172 int separate_locks; /* set if the interface suports different 173 locks for rx, tx and core. */ 174 175 u_int num_rx_rings; /* number of adapter receive rings */ 176 u_int num_tx_rings; /* number of adapter transmit rings */ 177 178 u_int num_tx_desc; /* number of descriptor in each queue */ 179 u_int num_rx_desc; 180 181 /* tx_rings and rx_rings are private but allocated 182 * as a contiguous chunk of memory. Each array has 183 * N+1 entries, for the adapter queues and for the host queue. 184 */ 185 struct netmap_kring *tx_rings; /* array of TX rings. */ 186 struct netmap_kring *rx_rings; /* array of RX rings. */ 187 188 NM_SELINFO_T tx_si, rx_si; /* global wait queues */ 189 190 /* copy of if_qflush and if_transmit pointers, to intercept 191 * packets from the network stack when netmap is active. 192 */ 193 int (*if_transmit)(struct ifnet *, struct mbuf *); 194 195 /* references to the ifnet and device routines, used by 196 * the generic netmap functions. 197 */ 198 struct ifnet *ifp; /* adapter is ifp->if_softc */ 199 200 NM_LOCK_T core_lock; /* used if no device lock available */ 201 202 int (*nm_register)(struct ifnet *, int onoff); 203 void (*nm_lock)(struct ifnet *, int what, u_int ringid); 204 int (*nm_txsync)(struct ifnet *, u_int ring, int lock); 205 int (*nm_rxsync)(struct ifnet *, u_int ring, int lock); 206 /* return configuration information */ 207 int (*nm_config)(struct ifnet *, u_int *txr, u_int *txd, 208 u_int *rxr, u_int *rxd); 209 210 int bdg_port; 211 #ifdef linux 212 struct net_device_ops nm_ndo; 213 int if_refcount; // XXX additions for bridge 214 #endif /* linux */ 215 }; 216 217 /* 218 * The combination of "enable" (ifp->if_capenable & IFCAP_NETMAP) 219 * and refcount gives the status of the interface, namely: 220 * 221 * enable refcount Status 222 * 223 * FALSE 0 normal operation 224 * FALSE != 0 -- (impossible) 225 * TRUE 1 netmap mode 226 * TRUE 0 being deleted. 227 */ 228 229 #define NETMAP_DELETING(_na) ( ((_na)->refcount == 0) && \ 230 ( (_na)->ifp->if_capenable & IFCAP_NETMAP) ) 231 232 /* 233 * parameters for (*nm_lock)(adapter, what, index) 234 */ 235 enum { 236 NETMAP_NO_LOCK = 0, 237 NETMAP_CORE_LOCK, NETMAP_CORE_UNLOCK, 238 NETMAP_TX_LOCK, NETMAP_TX_UNLOCK, 239 NETMAP_RX_LOCK, NETMAP_RX_UNLOCK, 240 #ifdef __FreeBSD__ 241 #define NETMAP_REG_LOCK NETMAP_CORE_LOCK 242 #define NETMAP_REG_UNLOCK NETMAP_CORE_UNLOCK 243 #else 244 NETMAP_REG_LOCK, NETMAP_REG_UNLOCK 245 #endif 246 }; 247 248 /* 249 * The following are support routines used by individual drivers to 250 * support netmap operation. 251 * 252 * netmap_attach() initializes a struct netmap_adapter, allocating the 253 * struct netmap_ring's and the struct selinfo. 254 * 255 * netmap_detach() frees the memory allocated by netmap_attach(). 256 * 257 * netmap_start() replaces the if_transmit routine of the interface, 258 * and is used to intercept packets coming from the stack. 259 * 260 * netmap_load_map/netmap_reload_map are helper routines to set/reset 261 * the dmamap for a packet buffer 262 * 263 * netmap_reset() is a helper routine to be called in the driver 264 * when reinitializing a ring. 265 */ 266 int netmap_attach(struct netmap_adapter *, int); 267 void netmap_detach(struct ifnet *); 268 int netmap_start(struct ifnet *, struct mbuf *); 269 enum txrx { NR_RX = 0, NR_TX = 1 }; 270 struct netmap_slot *netmap_reset(struct netmap_adapter *na, 271 enum txrx tx, int n, u_int new_cur); 272 int netmap_ring_reinit(struct netmap_kring *); 273 274 extern u_int netmap_buf_size; 275 #define NETMAP_BUF_SIZE netmap_buf_size 276 extern int netmap_mitigate; 277 extern int netmap_no_pendintr; 278 extern u_int netmap_total_buffers; 279 extern char *netmap_buffer_base; 280 extern int netmap_verbose; // XXX debugging 281 enum { /* verbose flags */ 282 NM_VERB_ON = 1, /* generic verbose */ 283 NM_VERB_HOST = 0x2, /* verbose host stack */ 284 NM_VERB_RXSYNC = 0x10, /* verbose on rxsync/txsync */ 285 NM_VERB_TXSYNC = 0x20, 286 NM_VERB_RXINTR = 0x100, /* verbose on rx/tx intr (driver) */ 287 NM_VERB_TXINTR = 0x200, 288 NM_VERB_NIC_RXSYNC = 0x1000, /* verbose on rx/tx intr (driver) */ 289 NM_VERB_NIC_TXSYNC = 0x2000, 290 }; 291 292 /* 293 * NA returns a pointer to the struct netmap adapter from the ifp, 294 * WNA is used to write it. 295 */ 296 #ifndef WNA 297 #define WNA(_ifp) (_ifp)->if_pspare[0] 298 #endif 299 #define NA(_ifp) ((struct netmap_adapter *)WNA(_ifp)) 300 301 /* 302 * Macros to determine if an interface is netmap capable or netmap enabled. 303 * See the magic field in struct netmap_adapter. 304 */ 305 #ifdef __FreeBSD__ 306 /* 307 * on FreeBSD just use if_capabilities and if_capenable. 308 */ 309 #define NETMAP_CAPABLE(ifp) (NA(ifp) && \ 310 (ifp)->if_capabilities & IFCAP_NETMAP ) 311 312 #define NETMAP_SET_CAPABLE(ifp) \ 313 (ifp)->if_capabilities |= IFCAP_NETMAP 314 315 #else /* linux */ 316 317 /* 318 * on linux: 319 * we check if NA(ifp) is set and its first element has a related 320 * magic value. The capenable is within the struct netmap_adapter. 321 */ 322 #define NETMAP_MAGIC 0x52697a7a 323 324 #define NETMAP_CAPABLE(ifp) (NA(ifp) && \ 325 ((uint32_t)(uintptr_t)NA(ifp) ^ NA(ifp)->magic) == NETMAP_MAGIC ) 326 327 #define NETMAP_SET_CAPABLE(ifp) \ 328 NA(ifp)->magic = ((uint32_t)(uintptr_t)NA(ifp)) ^ NETMAP_MAGIC 329 330 #endif /* linux */ 331 332 #ifdef __FreeBSD__ 333 /* Callback invoked by the dma machinery after a successfull dmamap_load */ 334 static void netmap_dmamap_cb(__unused void *arg, 335 __unused bus_dma_segment_t * segs, __unused int nseg, __unused int error) 336 { 337 } 338 339 /* bus_dmamap_load wrapper: call aforementioned function if map != NULL. 340 * XXX can we do it without a callback ? 341 */ 342 static inline void 343 netmap_load_map(bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 344 { 345 if (map) 346 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE, 347 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT); 348 } 349 350 /* update the map when a buffer changes. */ 351 static inline void 352 netmap_reload_map(bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 353 { 354 if (map) { 355 bus_dmamap_unload(tag, map); 356 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE, 357 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT); 358 } 359 } 360 #else /* linux */ 361 362 /* 363 * XXX How do we redefine these functions: 364 * 365 * on linux we need 366 * dma_map_single(&pdev->dev, virt_addr, len, direction) 367 * dma_unmap_single(&adapter->pdev->dev, phys_addr, len, direction 368 * The len can be implicit (on netmap it is NETMAP_BUF_SIZE) 369 * unfortunately the direction is not, so we need to change 370 * something to have a cross API 371 */ 372 #define netmap_load_map(_t, _m, _b) 373 #define netmap_reload_map(_t, _m, _b) 374 #if 0 375 struct e1000_buffer *buffer_info = &tx_ring->buffer_info[l]; 376 /* set time_stamp *before* dma to help avoid a possible race */ 377 buffer_info->time_stamp = jiffies; 378 buffer_info->mapped_as_page = false; 379 buffer_info->length = len; 380 //buffer_info->next_to_watch = l; 381 /* reload dma map */ 382 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma, 383 NETMAP_BUF_SIZE, DMA_TO_DEVICE); 384 buffer_info->dma = dma_map_single(&adapter->pdev->dev, 385 addr, NETMAP_BUF_SIZE, DMA_TO_DEVICE); 386 387 if (dma_mapping_error(&adapter->pdev->dev, buffer_info->dma)) { 388 D("dma mapping error"); 389 /* goto dma_error; See e1000_put_txbuf() */ 390 /* XXX reset */ 391 } 392 tx_desc->buffer_addr = htole64(buffer_info->dma); //XXX 393 394 #endif 395 396 /* 397 * The bus_dmamap_sync() can be one of wmb() or rmb() depending on direction. 398 */ 399 #define bus_dmamap_sync(_a, _b, _c) 400 401 #endif /* linux */ 402 403 /* 404 * functions to map NIC to KRING indexes (n2k) and vice versa (k2n) 405 */ 406 static inline int 407 netmap_idx_n2k(struct netmap_kring *kr, int idx) 408 { 409 int n = kr->nkr_num_slots; 410 idx += kr->nkr_hwofs; 411 if (idx < 0) 412 return idx + n; 413 else if (idx < n) 414 return idx; 415 else 416 return idx - n; 417 } 418 419 420 static inline int 421 netmap_idx_k2n(struct netmap_kring *kr, int idx) 422 { 423 int n = kr->nkr_num_slots; 424 idx -= kr->nkr_hwofs; 425 if (idx < 0) 426 return idx + n; 427 else if (idx < n) 428 return idx; 429 else 430 return idx - n; 431 } 432 433 434 #ifdef NETMAP_MEM2 435 /* Entries of the look-up table. */ 436 struct lut_entry { 437 void *vaddr; /* virtual address. */ 438 vm_paddr_t paddr; /* phisical address. */ 439 }; 440 441 struct netmap_obj_pool; 442 extern struct lut_entry *netmap_buffer_lut; 443 #define NMB_VA(i) (netmap_buffer_lut[i].vaddr) 444 #define NMB_PA(i) (netmap_buffer_lut[i].paddr) 445 #else /* NETMAP_MEM1 */ 446 #define NMB_VA(i) (netmap_buffer_base + (i * NETMAP_BUF_SIZE) ) 447 #endif /* NETMAP_MEM2 */ 448 449 /* 450 * NMB return the virtual address of a buffer (buffer 0 on bad index) 451 * PNMB also fills the physical address 452 */ 453 static inline void * 454 NMB(struct netmap_slot *slot) 455 { 456 uint32_t i = slot->buf_idx; 457 return (unlikely(i >= netmap_total_buffers)) ? NMB_VA(0) : NMB_VA(i); 458 } 459 460 static inline void * 461 PNMB(struct netmap_slot *slot, uint64_t *pp) 462 { 463 uint32_t i = slot->buf_idx; 464 void *ret = (i >= netmap_total_buffers) ? NMB_VA(0) : NMB_VA(i); 465 #ifdef NETMAP_MEM2 466 *pp = (i >= netmap_total_buffers) ? NMB_PA(0) : NMB_PA(i); 467 #else 468 *pp = vtophys(ret); 469 #endif 470 return ret; 471 } 472 473 /* default functions to handle rx/tx interrupts */ 474 int netmap_rx_irq(struct ifnet *, int, int *); 475 #define netmap_tx_irq(_n, _q) netmap_rx_irq(_n, _q, NULL) 476 477 extern int netmap_copy; 478 #endif /* _NET_NETMAP_KERN_H_ */ 479