netmap_kern.h revision 238985
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: head/sys/dev/netmap/netmap_kern.h 238985 2012-08-02 11:59:43Z luigi $ 28 * $Id: netmap_kern.h 11343 2012-07-03 09:08:38Z 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 flags (if_capabilities) 59 * and priv_flags (if_capenable). The latter used to be 16 bits 60 * up to linux 2.6.36, so we need to use a 16 bit value on older 61 * platforms and tolerate the clash with IFF_DYNAMIC and IFF_BRIDGE_PORT. 62 * For the 32-bit value, 0x100000 (bit 20) has no clashes up to 3.3.1 63 */ 64#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37) 65#define IFCAP_NETMAP 0x8000 66#else 67#define IFCAP_NETMAP 0x100000 68#endif 69 70#elif defined (__APPLE__) 71#warning apple support is experimental 72#define likely(x) __builtin_expect(!!(x), 1) 73#define unlikely(x) __builtin_expect(!!(x), 0) 74#define NM_LOCK_T IOLock * 75#define NM_SELINFO_T struct selinfo 76#define MBUF_LEN(m) ((m)->m_pkthdr.len) 77#define NM_SEND_UP(ifp, m) ((ifp)->if_input)(ifp, m) 78 79#else 80#error unsupported platform 81#endif 82 83#define ND(format, ...) 84#define D(format, ...) \ 85 do { \ 86 struct timeval __xxts; \ 87 microtime(&__xxts); \ 88 printf("%03d.%06d %s [%d] " format "\n", \ 89 (int)__xxts.tv_sec % 1000, (int)__xxts.tv_usec, \ 90 __FUNCTION__, __LINE__, ##__VA_ARGS__); \ 91 } while (0) 92 93struct netmap_adapter; 94 95/* 96 * private, kernel view of a ring. Keeps track of the status of 97 * a ring across system calls. 98 * 99 * nr_hwcur index of the next buffer to refill. 100 * It corresponds to ring->cur - ring->reserved 101 * 102 * nr_hwavail the number of slots "owned" by userspace. 103 * nr_hwavail =:= ring->avail + ring->reserved 104 * 105 * The indexes in the NIC and netmap rings are offset by nkr_hwofs slots. 106 * This is so that, on a reset, buffers owned by userspace are not 107 * modified by the kernel. In particular: 108 * RX rings: the next empty buffer (hwcur + hwavail + hwofs) coincides with 109 * the next empty buffer as known by the hardware (next_to_check or so). 110 * TX rings: hwcur + hwofs coincides with next_to_send 111 */ 112struct netmap_kring { 113 struct netmap_ring *ring; 114 u_int nr_hwcur; 115 int nr_hwavail; 116 u_int nr_kflags; /* private driver flags */ 117#define NKR_PENDINTR 0x1 // Pending interrupt. 118 u_int nkr_num_slots; 119 120 int nkr_hwofs; /* offset between NIC and netmap ring */ 121 struct netmap_adapter *na; 122 NM_SELINFO_T si; /* poll/select wait queue */ 123 NM_LOCK_T q_lock; /* used if no device lock available */ 124} __attribute__((__aligned__(64))); 125 126/* 127 * This struct extends the 'struct adapter' (or 128 * equivalent) device descriptor. It contains all fields needed to 129 * support netmap operation. 130 */ 131struct netmap_adapter { 132 int refcount; /* number of user-space descriptors using this 133 interface, which is equal to the number of 134 struct netmap_if objs in the mapped region. */ 135 /* 136 * The selwakeup in the interrupt thread can use per-ring 137 * and/or global wait queues. We track how many clients 138 * of each type we have so we can optimize the drivers, 139 * and especially avoid huge contention on the locks. 140 */ 141 int na_single; /* threads attached to a single hw queue */ 142 int na_multi; /* threads attached to multiple hw queues */ 143 144 int separate_locks; /* set if the interface suports different 145 locks for rx, tx and core. */ 146 147 u_int num_rx_rings; /* number of tx/rx ring pairs */ 148 u_int num_tx_rings; // if nonzero, overrides num_rx_rings 149 150 u_int num_tx_desc; /* number of descriptor in each queue */ 151 u_int num_rx_desc; 152 //u_int buff_size; // XXX deprecate, use NETMAP_BUF_SIZE 153 154 /* tx_rings and rx_rings are private but allocated 155 * as a contiguous chunk of memory. Each array has 156 * N+1 entries, for the adapter queues and for the host queue. 157 */ 158 struct netmap_kring *tx_rings; /* array of TX rings. */ 159 struct netmap_kring *rx_rings; /* array of RX rings. */ 160 161 NM_SELINFO_T tx_si, rx_si; /* global wait queues */ 162 163 /* copy of if_qflush and if_transmit pointers, to intercept 164 * packets from the network stack when netmap is active. 165 */ 166 int (*if_transmit)(struct ifnet *, struct mbuf *); 167 168 /* references to the ifnet and device routines, used by 169 * the generic netmap functions. 170 */ 171 struct ifnet *ifp; /* adapter is ifp->if_softc */ 172 173 NM_LOCK_T core_lock; /* used if no device lock available */ 174 175 int (*nm_register)(struct ifnet *, int onoff); 176 void (*nm_lock)(struct ifnet *, int what, u_int ringid); 177 int (*nm_txsync)(struct ifnet *, u_int ring, int lock); 178 int (*nm_rxsync)(struct ifnet *, u_int ring, int lock); 179 180 int bdg_port; 181#ifdef linux 182 struct net_device_ops nm_ndo; 183 int if_refcount; // XXX additions for bridge 184#endif /* linux */ 185}; 186 187/* 188 * The combination of "enable" (ifp->if_capabilities &IFCAP_NETMAP) 189 * and refcount gives the status of the interface, namely: 190 * 191 * enable refcount Status 192 * 193 * FALSE 0 normal operation 194 * FALSE != 0 -- (impossible) 195 * TRUE 1 netmap mode 196 * TRUE 0 being deleted. 197 */ 198 199#define NETMAP_DELETING(_na) ( ((_na)->refcount == 0) && \ 200 ( (_na)->ifp->if_capenable & IFCAP_NETMAP) ) 201 202/* 203 * parameters for (*nm_lock)(adapter, what, index) 204 */ 205enum { 206 NETMAP_NO_LOCK = 0, 207 NETMAP_CORE_LOCK, NETMAP_CORE_UNLOCK, 208 NETMAP_TX_LOCK, NETMAP_TX_UNLOCK, 209 NETMAP_RX_LOCK, NETMAP_RX_UNLOCK, 210#ifdef __FreeBSD__ 211#define NETMAP_REG_LOCK NETMAP_CORE_LOCK 212#define NETMAP_REG_UNLOCK NETMAP_CORE_UNLOCK 213#else 214 NETMAP_REG_LOCK, NETMAP_REG_UNLOCK 215#endif 216}; 217 218/* 219 * The following are support routines used by individual drivers to 220 * support netmap operation. 221 * 222 * netmap_attach() initializes a struct netmap_adapter, allocating the 223 * struct netmap_ring's and the struct selinfo. 224 * 225 * netmap_detach() frees the memory allocated by netmap_attach(). 226 * 227 * netmap_start() replaces the if_transmit routine of the interface, 228 * and is used to intercept packets coming from the stack. 229 * 230 * netmap_load_map/netmap_reload_map are helper routines to set/reset 231 * the dmamap for a packet buffer 232 * 233 * netmap_reset() is a helper routine to be called in the driver 234 * when reinitializing a ring. 235 */ 236int netmap_attach(struct netmap_adapter *, int); 237void netmap_detach(struct ifnet *); 238int netmap_start(struct ifnet *, struct mbuf *); 239enum txrx { NR_RX = 0, NR_TX = 1 }; 240struct netmap_slot *netmap_reset(struct netmap_adapter *na, 241 enum txrx tx, int n, u_int new_cur); 242int netmap_ring_reinit(struct netmap_kring *); 243 244extern u_int netmap_buf_size; 245#define NETMAP_BUF_SIZE netmap_buf_size 246extern int netmap_mitigate; 247extern int netmap_no_pendintr; 248extern u_int netmap_total_buffers; 249extern char *netmap_buffer_base; 250extern int netmap_verbose; // XXX debugging 251enum { /* verbose flags */ 252 NM_VERB_ON = 1, /* generic verbose */ 253 NM_VERB_HOST = 0x2, /* verbose host stack */ 254 NM_VERB_RXSYNC = 0x10, /* verbose on rxsync/txsync */ 255 NM_VERB_TXSYNC = 0x20, 256 NM_VERB_RXINTR = 0x100, /* verbose on rx/tx intr (driver) */ 257 NM_VERB_TXINTR = 0x200, 258 NM_VERB_NIC_RXSYNC = 0x1000, /* verbose on rx/tx intr (driver) */ 259 NM_VERB_NIC_TXSYNC = 0x2000, 260}; 261 262/* 263 * NA returns a pointer to the struct netmap adapter from the ifp, 264 * WNA is used to write it. 265 */ 266#ifndef WNA 267#define WNA(_ifp) (_ifp)->if_pspare[0] 268#endif 269#define NA(_ifp) ((struct netmap_adapter *)WNA(_ifp)) 270 271 272#ifdef __FreeBSD__ 273/* Callback invoked by the dma machinery after a successfull dmamap_load */ 274static void netmap_dmamap_cb(__unused void *arg, 275 __unused bus_dma_segment_t * segs, __unused int nseg, __unused int error) 276{ 277} 278 279/* bus_dmamap_load wrapper: call aforementioned function if map != NULL. 280 * XXX can we do it without a callback ? 281 */ 282static inline void 283netmap_load_map(bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 284{ 285 if (map) 286 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE, 287 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT); 288} 289 290/* update the map when a buffer changes. */ 291static inline void 292netmap_reload_map(bus_dma_tag_t tag, bus_dmamap_t map, void *buf) 293{ 294 if (map) { 295 bus_dmamap_unload(tag, map); 296 bus_dmamap_load(tag, map, buf, NETMAP_BUF_SIZE, 297 netmap_dmamap_cb, NULL, BUS_DMA_NOWAIT); 298 } 299} 300#else /* linux */ 301 302/* 303 * XXX How do we redefine these functions: 304 * 305 * on linux we need 306 * dma_map_single(&pdev->dev, virt_addr, len, direction) 307 * dma_unmap_single(&adapter->pdev->dev, phys_addr, len, direction 308 * The len can be implicit (on netmap it is NETMAP_BUF_SIZE) 309 * unfortunately the direction is not, so we need to change 310 * something to have a cross API 311 */ 312#define netmap_load_map(_t, _m, _b) 313#define netmap_reload_map(_t, _m, _b) 314#if 0 315 struct e1000_buffer *buffer_info = &tx_ring->buffer_info[l]; 316 /* set time_stamp *before* dma to help avoid a possible race */ 317 buffer_info->time_stamp = jiffies; 318 buffer_info->mapped_as_page = false; 319 buffer_info->length = len; 320 //buffer_info->next_to_watch = l; 321 /* reload dma map */ 322 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma, 323 NETMAP_BUF_SIZE, DMA_TO_DEVICE); 324 buffer_info->dma = dma_map_single(&adapter->pdev->dev, 325 addr, NETMAP_BUF_SIZE, DMA_TO_DEVICE); 326 327 if (dma_mapping_error(&adapter->pdev->dev, buffer_info->dma)) { 328 D("dma mapping error"); 329 /* goto dma_error; See e1000_put_txbuf() */ 330 /* XXX reset */ 331 } 332 tx_desc->buffer_addr = htole64(buffer_info->dma); //XXX 333 334#endif 335 336/* 337 * The bus_dmamap_sync() can be one of wmb() or rmb() depending on direction. 338 */ 339#define bus_dmamap_sync(_a, _b, _c) 340 341#endif /* linux */ 342 343/* 344 * functions to map NIC to KRING indexes (n2k) and vice versa (k2n) 345 */ 346static inline int 347netmap_idx_n2k(struct netmap_kring *kr, int idx) 348{ 349 int n = kr->nkr_num_slots; 350 idx += kr->nkr_hwofs; 351 if (idx < 0) 352 return idx + n; 353 else if (idx < n) 354 return idx; 355 else 356 return idx - n; 357} 358 359 360static inline int 361netmap_idx_k2n(struct netmap_kring *kr, int idx) 362{ 363 int n = kr->nkr_num_slots; 364 idx -= kr->nkr_hwofs; 365 if (idx < 0) 366 return idx + n; 367 else if (idx < n) 368 return idx; 369 else 370 return idx - n; 371} 372 373 374#ifdef NETMAP_MEM2 375/* Entries of the look-up table. */ 376struct lut_entry { 377 void *vaddr; /* virtual address. */ 378 vm_paddr_t paddr; /* phisical address. */ 379}; 380 381struct netmap_obj_pool; 382extern struct lut_entry *netmap_buffer_lut; 383#define NMB_VA(i) (netmap_buffer_lut[i].vaddr) 384#define NMB_PA(i) (netmap_buffer_lut[i].paddr) 385#else /* NETMAP_MEM1 */ 386#define NMB_VA(i) (netmap_buffer_base + (i * NETMAP_BUF_SIZE) ) 387#endif /* NETMAP_MEM2 */ 388 389/* 390 * NMB return the virtual address of a buffer (buffer 0 on bad index) 391 * PNMB also fills the physical address 392 */ 393static inline void * 394NMB(struct netmap_slot *slot) 395{ 396 uint32_t i = slot->buf_idx; 397 return (unlikely(i >= netmap_total_buffers)) ? NMB_VA(0) : NMB_VA(i); 398} 399 400static inline void * 401PNMB(struct netmap_slot *slot, uint64_t *pp) 402{ 403 uint32_t i = slot->buf_idx; 404 void *ret = (i >= netmap_total_buffers) ? NMB_VA(0) : NMB_VA(i); 405#ifdef NETMAP_MEM2 406 *pp = (i >= netmap_total_buffers) ? NMB_PA(0) : NMB_PA(i); 407#else 408 *pp = vtophys(ret); 409#endif 410 return ret; 411} 412 413/* default functions to handle rx/tx interrupts */ 414int netmap_rx_irq(struct ifnet *, int, int *); 415#define netmap_tx_irq(_n, _q) netmap_rx_irq(_n, _q, NULL) 416 417extern int netmap_copy; 418#endif /* _NET_NETMAP_KERN_H_ */ 419