| 1 | |
| 2 | |
| 3 | * Copyright (C) 2017 - 2019 Cambridge Greys Limited
|
| 4 | * Copyright (C) 2011 - 2014 Cisco Systems Inc
|
| 5 | * Copyright (C) 2001 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
|
| 6 | * Copyright (C) 2001 Lennert Buytenhek ([email protected]) and
|
| 7 | * James Leu ([email protected]).
|
| 8 | * Copyright (C) 2001 by various other people who didn't put their name here.
|
| 9 |
|
| 10 |
|
| 11 | #define pr_fmt(fmt) "uml-vector: " fmt
|
| 12 |
|
| 13 | #include <linux/memblock.h>
|
| 14 | #include <linux/etherdevice.h>
|
| 15 | #include <linux/ethtool.h>
|
| 16 | #include <linux/hex.h>
|
| 17 | #include <linux/inetdevice.h>
|
| 18 | #include <linux/init.h>
|
| 19 | #include <linux/list.h>
|
| 20 | #include <linux/netdevice.h>
|
| 21 | #include <linux/platform_device.h>
|
| 22 | #include <linux/rtnetlink.h>
|
| 23 | #include <linux/skbuff.h>
|
| 24 | #include <linux/slab.h>
|
| 25 | #include <linux/interrupt.h>
|
| 26 | #include <linux/firmware.h>
|
| 27 | #include <linux/fs.h>
|
| 28 | #include <asm/atomic.h>
|
| 29 | #include <uapi/linux/filter.h>
|
| 30 | #include <init.h>
|
| 31 | #include <irq_kern.h>
|
| 32 | #include <irq_user.h>
|
| 33 | #include <os.h>
|
| 34 | #include "mconsole_kern.h"
|
| 35 | #include "vector_user.h"
|
| 36 | #include "vector_kern.h"
|
| 37 |
|
| 38 | |
| 39 | * Adapted from network devices with the following major changes:
|
| 40 | * All transports are static - simplifies the code significantly
|
| 41 | * Multiple FDs/IRQs per device
|
| 42 | * Vector IO optionally used for read/write, falling back to legacy
|
| 43 | * based on configuration and/or availability
|
| 44 | * Configuration is no longer positional - L2TPv3 and GRE require up to
|
| 45 | * 10 parameters, passing this as positional is not fit for purpose.
|
| 46 | * Only socket transports are supported
|
| 47 |
|
| 48 |
|
| 49 |
|
| 50 | #define DRIVER_NAME "uml-vector"
|
| 51 | struct vector_cmd_line_arg {
|
| 52 | struct list_head list;
|
| 53 | int unit;
|
| 54 | char *arguments;
|
| 55 | };
|
| 56 |
|
| 57 | struct vector_device {
|
| 58 | struct list_head list;
|
| 59 | struct net_device *dev;
|
| 60 | struct platform_device pdev;
|
| 61 | int unit;
|
| 62 | int opened;
|
| 63 | };
|
| 64 |
|
| 65 | static LIST_HEAD(vec_cmd_line);
|
| 66 |
|
| 67 | static DEFINE_SPINLOCK(vector_devices_lock);
|
| 68 | static LIST_HEAD(vector_devices);
|
| 69 |
|
| 70 | static int driver_registered;
|
| 71 |
|
| 72 | static void vector_eth_configure(int n, struct arglist *def);
|
| 73 | static int vector_mmsg_rx(struct vector_private *vp, int budget);
|
| 74 |
|
| 75 | |
| 76 | * mtu, buffer sizing, default headroom, etc
|
| 77 |
|
| 78 |
|
| 79 | #define DEFAULT_HEADROOM 2
|
| 80 | #define SAFETY_MARGIN 32
|
| 81 | #define DEFAULT_VECTOR_SIZE 64
|
| 82 | #define TX_SMALL_PACKET 128
|
| 83 | #define MAX_IOV_SIZE (MAX_SKB_FRAGS + 1)
|
| 84 |
|
| 85 | static const struct {
|
| 86 | const char string[ETH_GSTRING_LEN];
|
| 87 | } ethtool_stats_keys[] = {
|
| 88 | { "rx_queue_max" },
|
| 89 | { "rx_queue_running_average" },
|
| 90 | { "tx_queue_max" },
|
| 91 | { "tx_queue_running_average" },
|
| 92 | { "rx_encaps_errors" },
|
| 93 | { "tx_timeout_count" },
|
| 94 | { "tx_restart_queue" },
|
| 95 | { "tx_kicks" },
|
| 96 | { "tx_flow_control_xon" },
|
| 97 | { "tx_flow_control_xoff" },
|
| 98 | { "rx_csum_offload_good" },
|
| 99 | { "rx_csum_offload_errors"},
|
| 100 | { "sg_ok"},
|
| 101 | { "sg_linearized"},
|
| 102 | };
|
| 103 |
|
| 104 | #define VECTOR_NUM_STATS ARRAY_SIZE(ethtool_stats_keys)
|
| 105 |
|
| 106 | static void vector_reset_stats(struct vector_private *vp)
|
| 107 | {
|
| 108 |
|
| 109 |
|
| 110 | |
| 111 | * in vector_poll.
|
| 112 |
|
| 113 |
|
| 114 | spin_lock(&vp->rx_queue->head_lock);
|
| 115 | vp->estats.rx_queue_max = 0;
|
| 116 | vp->estats.rx_queue_running_average = 0;
|
| 117 | vp->estats.rx_encaps_errors = 0;
|
| 118 | vp->estats.sg_ok = 0;
|
| 119 | vp->estats.sg_linearized = 0;
|
| 120 | spin_unlock(&vp->rx_queue->head_lock);
|
| 121 |
|
| 122 | |
| 123 | * in vector_send.
|
| 124 |
|
| 125 |
|
| 126 | spin_lock(&vp->tx_queue->head_lock);
|
| 127 | vp->estats.tx_timeout_count = 0;
|
| 128 | vp->estats.tx_restart_queue = 0;
|
| 129 | vp->estats.tx_kicks = 0;
|
| 130 | vp->estats.tx_flow_control_xon = 0;
|
| 131 | vp->estats.tx_flow_control_xoff = 0;
|
| 132 | vp->estats.tx_queue_max = 0;
|
| 133 | vp->estats.tx_queue_running_average = 0;
|
| 134 | spin_unlock(&vp->tx_queue->head_lock);
|
| 135 | }
|
| 136 |
|
| 137 | static int get_mtu(struct arglist *def)
|
| 138 | {
|
| 139 | char *mtu = uml_vector_fetch_arg(def, "mtu");
|
| 140 | long result;
|
| 141 |
|
| 142 | if (mtu != NULL) {
|
| 143 | if (kstrtoul(mtu, 10, &result) == 0)
|
| 144 | if ((result < (1 << 16) - 1) && (result >= 576))
|
| 145 | return result;
|
| 146 | }
|
| 147 | return ETH_MAX_PACKET;
|
| 148 | }
|
| 149 |
|
| 150 | static char *get_bpf_file(struct arglist *def)
|
| 151 | {
|
| 152 | return uml_vector_fetch_arg(def, "bpffile");
|
| 153 | }
|
| 154 |
|
| 155 | static bool get_bpf_flash(struct arglist *def)
|
| 156 | {
|
| 157 | char *allow = uml_vector_fetch_arg(def, "bpfflash");
|
| 158 | long result;
|
| 159 |
|
| 160 | if (allow != NULL) {
|
| 161 | if (kstrtoul(allow, 10, &result) == 0)
|
| 162 | return result > 0;
|
| 163 | }
|
| 164 | return false;
|
| 165 | }
|
| 166 |
|
| 167 | static int get_depth(struct arglist *def)
|
| 168 | {
|
| 169 | char *mtu = uml_vector_fetch_arg(def, "depth");
|
| 170 | long result;
|
| 171 |
|
| 172 | if (mtu != NULL) {
|
| 173 | if (kstrtoul(mtu, 10, &result) == 0)
|
| 174 | return result;
|
| 175 | }
|
| 176 | return DEFAULT_VECTOR_SIZE;
|
| 177 | }
|
| 178 |
|
| 179 | static int get_headroom(struct arglist *def)
|
| 180 | {
|
| 181 | char *mtu = uml_vector_fetch_arg(def, "headroom");
|
| 182 | long result;
|
| 183 |
|
| 184 | if (mtu != NULL) {
|
| 185 | if (kstrtoul(mtu, 10, &result) == 0)
|
| 186 | return result;
|
| 187 | }
|
| 188 | return DEFAULT_HEADROOM;
|
| 189 | }
|
| 190 |
|
| 191 | static int get_req_size(struct arglist *def)
|
| 192 | {
|
| 193 | char *gro = uml_vector_fetch_arg(def, "gro");
|
| 194 | long result;
|
| 195 |
|
| 196 | if (gro != NULL) {
|
| 197 | if (kstrtoul(gro, 10, &result) == 0) {
|
| 198 | if (result > 0)
|
| 199 | return 65536;
|
| 200 | }
|
| 201 | }
|
| 202 | return get_mtu(def) + ETH_HEADER_OTHER +
|
| 203 | get_headroom(def) + SAFETY_MARGIN;
|
| 204 | }
|
| 205 |
|
| 206 |
|
| 207 | static int get_transport_options(struct arglist *def)
|
| 208 | {
|
| 209 | char *transport = uml_vector_fetch_arg(def, "transport");
|
| 210 | char *vector = uml_vector_fetch_arg(def, "vec");
|
| 211 |
|
| 212 | int vec_rx = VECTOR_RX;
|
| 213 | int vec_tx = VECTOR_TX;
|
| 214 | long parsed;
|
| 215 | int result = 0;
|
| 216 |
|
| 217 | if (transport == NULL)
|
| 218 | return -EINVAL;
|
| 219 |
|
| 220 | if (vector != NULL) {
|
| 221 | if (kstrtoul(vector, 10, &parsed) == 0) {
|
| 222 | if (parsed == 0) {
|
| 223 | vec_rx = 0;
|
| 224 | vec_tx = 0;
|
| 225 | }
|
| 226 | }
|
| 227 | }
|
| 228 |
|
| 229 | if (get_bpf_flash(def))
|
| 230 | result = VECTOR_BPF_FLASH;
|
| 231 |
|
| 232 | if (strncmp(transport, TRANS_TAP, TRANS_TAP_LEN) == 0)
|
| 233 | return result;
|
| 234 | if (strncmp(transport, TRANS_HYBRID, TRANS_HYBRID_LEN) == 0)
|
| 235 | return (result | vec_rx | VECTOR_BPF);
|
| 236 | if (strncmp(transport, TRANS_RAW, TRANS_RAW_LEN) == 0)
|
| 237 | return (result | vec_rx | vec_tx | VECTOR_QDISC_BYPASS);
|
| 238 | return (result | vec_rx | vec_tx);
|
| 239 | }
|
| 240 |
|
| 241 |
|
| 242 | |
| 243 | * All of our supported transports are datagram oriented and we always
|
| 244 | * read using recvmsg or recvmmsg. If we pass a buffer which is smaller
|
| 245 | * than the packet size it still counts as full packet read and will
|
| 246 | * clean the incoming stream to keep sigio/epoll happy
|
| 247 |
|
| 248 |
|
| 249 | #define DROP_BUFFER_SIZE 32
|
| 250 |
|
| 251 | static char *drop_buffer;
|
| 252 |
|
| 253 |
|
| 254 | |
| 255 | * Advance the mmsg queue head by n = advance. Resets the queue to
|
| 256 | * maximum enqueue/dequeue-at-once capacity if possible. Called by
|
| 257 | * dequeuers. Caller must hold the head_lock!
|
| 258 |
|
| 259 |
|
| 260 | static int vector_advancehead(struct vector_queue *qi, int advance)
|
| 261 | {
|
| 262 | qi->head =
|
| 263 | (qi->head + advance)
|
| 264 | % qi->max_depth;
|
| 265 |
|
| 266 |
|
| 267 | atomic_sub(advance, &qi->queue_depth);
|
| 268 | return atomic_read(&qi->queue_depth);
|
| 269 | }
|
| 270 |
|
| 271 | |
| 272 | * This is called by enqueuers which should hold the
|
| 273 | * head lock already
|
| 274 |
|
| 275 |
|
| 276 | static int vector_advancetail(struct vector_queue *qi, int advance)
|
| 277 | {
|
| 278 | qi->tail =
|
| 279 | (qi->tail + advance)
|
| 280 | % qi->max_depth;
|
| 281 | atomic_add(advance, &qi->queue_depth);
|
| 282 | return atomic_read(&qi->queue_depth);
|
| 283 | }
|
| 284 |
|
| 285 | static int prep_msg(struct vector_private *vp,
|
| 286 | struct sk_buff *skb,
|
| 287 | struct iovec *iov)
|
| 288 | {
|
| 289 | int iov_index = 0;
|
| 290 | int nr_frags, frag;
|
| 291 | skb_frag_t *skb_frag;
|
| 292 |
|
| 293 | nr_frags = skb_shinfo(skb)->nr_frags;
|
| 294 | if (nr_frags > MAX_IOV_SIZE) {
|
| 295 | if (skb_linearize(skb) != 0)
|
| 296 | goto drop;
|
| 297 | }
|
| 298 | if (vp->header_size > 0) {
|
| 299 | iov[iov_index].iov_len = vp->header_size;
|
| 300 | vp->form_header(iov[iov_index].iov_base, skb, vp);
|
| 301 | iov_index++;
|
| 302 | }
|
| 303 | iov[iov_index].iov_base = skb->data;
|
| 304 | if (nr_frags > 0) {
|
| 305 | iov[iov_index].iov_len = skb->len - skb->data_len;
|
| 306 | vp->estats.sg_ok++;
|
| 307 | } else
|
| 308 | iov[iov_index].iov_len = skb->len;
|
| 309 | iov_index++;
|
| 310 | for (frag = 0; frag < nr_frags; frag++) {
|
| 311 | skb_frag = &skb_shinfo(skb)->frags[frag];
|
| 312 | iov[iov_index].iov_base = skb_frag_address_safe(skb_frag);
|
| 313 | iov[iov_index].iov_len = skb_frag_size(skb_frag);
|
| 314 | iov_index++;
|
| 315 | }
|
| 316 | return iov_index;
|
| 317 | drop:
|
| 318 | return -1;
|
| 319 | }
|
| 320 | |
| 321 | * Generic vector enqueue with support for forming headers using transport
|
| 322 | * specific callback. Allows GRE, L2TPv3, RAW and other transports
|
| 323 | * to use a common enqueue procedure in vector mode
|
| 324 |
|
| 325 |
|
| 326 | static int vector_enqueue(struct vector_queue *qi, struct sk_buff *skb)
|
| 327 | {
|
| 328 | struct vector_private *vp = netdev_priv(qi->dev);
|
| 329 | int queue_depth;
|
| 330 | int packet_len;
|
| 331 | struct mmsghdr *mmsg_vector = qi->mmsg_vector;
|
| 332 | int iov_count;
|
| 333 |
|
| 334 | spin_lock(&qi->tail_lock);
|
| 335 | queue_depth = atomic_read(&qi->queue_depth);
|
| 336 |
|
| 337 | if (skb)
|
| 338 | packet_len = skb->len;
|
| 339 |
|
| 340 | if (queue_depth < qi->max_depth) {
|
| 341 |
|
| 342 | *(qi->skbuff_vector + qi->tail) = skb;
|
| 343 | mmsg_vector += qi->tail;
|
| 344 | iov_count = prep_msg(
|
| 345 | vp,
|
| 346 | skb,
|
| 347 | mmsg_vector->msg_hdr.msg_iov
|
| 348 | );
|
| 349 | if (iov_count < 1)
|
| 350 | goto drop;
|
| 351 | mmsg_vector->msg_hdr.msg_iovlen = iov_count;
|
| 352 | mmsg_vector->msg_hdr.msg_name = vp->fds->remote_addr;
|
| 353 | mmsg_vector->msg_hdr.msg_namelen = vp->fds->remote_addr_size;
|
| 354 | wmb();
|
| 355 | queue_depth = vector_advancetail(qi, 1);
|
| 356 | } else
|
| 357 | goto drop;
|
| 358 | spin_unlock(&qi->tail_lock);
|
| 359 | return queue_depth;
|
| 360 | drop:
|
| 361 | qi->dev->stats.tx_dropped++;
|
| 362 | if (skb != NULL) {
|
| 363 | packet_len = skb->len;
|
| 364 | dev_consume_skb_any(skb);
|
| 365 | netdev_completed_queue(qi->dev, 1, packet_len);
|
| 366 | }
|
| 367 | spin_unlock(&qi->tail_lock);
|
| 368 | return queue_depth;
|
| 369 | }
|
| 370 |
|
| 371 | static int consume_vector_skbs(struct vector_queue *qi, int count)
|
| 372 | {
|
| 373 | struct sk_buff *skb;
|
| 374 | int skb_index;
|
| 375 | int bytes_compl = 0;
|
| 376 |
|
| 377 | for (skb_index = qi->head; skb_index < qi->head + count; skb_index++) {
|
| 378 | skb = *(qi->skbuff_vector + skb_index);
|
| 379 | |
| 380 | * needed
|
| 381 |
|
| 382 | bytes_compl += skb->len;
|
| 383 | *(qi->skbuff_vector + skb_index) = NULL;
|
| 384 | dev_consume_skb_any(skb);
|
| 385 | }
|
| 386 | qi->dev->stats.tx_bytes += bytes_compl;
|
| 387 | qi->dev->stats.tx_packets += count;
|
| 388 | netdev_completed_queue(qi->dev, count, bytes_compl);
|
| 389 | return vector_advancehead(qi, count);
|
| 390 | }
|
| 391 |
|
| 392 | |
| 393 | * Generic vector dequeue via sendmmsg with support for forming headers
|
| 394 | * using transport specific callback. Allows GRE, L2TPv3, RAW and
|
| 395 | * other transports to use a common dequeue procedure in vector mode
|
| 396 |
|
| 397 |
|
| 398 |
|
| 399 | static int vector_send(struct vector_queue *qi)
|
| 400 | {
|
| 401 | struct vector_private *vp = netdev_priv(qi->dev);
|
| 402 | struct mmsghdr *send_from;
|
| 403 | int result = 0, send_len;
|
| 404 |
|
| 405 | if (spin_trylock(&qi->head_lock)) {
|
| 406 |
|
| 407 | while (atomic_read(&qi->queue_depth) > 0) {
|
| 408 |
|
| 409 | send_len = atomic_read(&qi->queue_depth);
|
| 410 | send_from = qi->mmsg_vector;
|
| 411 | send_from += qi->head;
|
| 412 |
|
| 413 | if (send_len + qi->head > qi->max_depth)
|
| 414 | send_len = qi->max_depth - qi->head;
|
| 415 |
|
| 416 | if (send_len > 0) {
|
| 417 | result = uml_vector_sendmmsg(
|
| 418 | vp->fds->tx_fd,
|
| 419 | send_from,
|
| 420 | send_len,
|
| 421 | 0
|
| 422 | );
|
| 423 | vp->in_write_poll =
|
| 424 | (result != send_len);
|
| 425 | }
|
| 426 | |
| 427 | * we may end being unsure in the TX success
|
| 428 | * for all packets. It is safer to declare
|
| 429 | * them all TX-ed and blame the network.
|
| 430 |
|
| 431 | if (result < 0) {
|
| 432 | if (net_ratelimit())
|
| 433 | netdev_err(vp->dev, "sendmmsg err=%i\n",
|
| 434 | result);
|
| 435 | vp->in_error = true;
|
| 436 | result = send_len;
|
| 437 | }
|
| 438 | if (result > 0) {
|
| 439 | consume_vector_skbs(qi, result);
|
| 440 | |
| 441 | * Restart the upper layers to feed us
|
| 442 | * more packets.
|
| 443 |
|
| 444 | if (result > vp->estats.tx_queue_max)
|
| 445 | vp->estats.tx_queue_max = result;
|
| 446 | vp->estats.tx_queue_running_average =
|
| 447 | (vp->estats.tx_queue_running_average + result) >> 1;
|
| 448 | }
|
| 449 | netif_wake_queue(qi->dev);
|
| 450 | |
| 451 | * poll write IRQ will reschedule xmit for us.
|
| 452 |
|
| 453 | if (result != send_len) {
|
| 454 | vp->estats.tx_restart_queue++;
|
| 455 | break;
|
| 456 | }
|
| 457 | }
|
| 458 | spin_unlock(&qi->head_lock);
|
| 459 | }
|
| 460 | return atomic_read(&qi->queue_depth);
|
| 461 | }
|
| 462 |
|
| 463 | |
| 464 | * it in queue cleanup if initialization fails.
|
| 465 |
|
| 466 |
|
| 467 | static void destroy_queue(struct vector_queue *qi)
|
| 468 | {
|
| 469 | int i;
|
| 470 | struct iovec *iov;
|
| 471 | struct vector_private *vp = netdev_priv(qi->dev);
|
| 472 | struct mmsghdr *mmsg_vector;
|
| 473 |
|
| 474 | if (qi == NULL)
|
| 475 | return;
|
| 476 | |
| 477 | * set to NULL.
|
| 478 |
|
| 479 | if (qi->skbuff_vector != NULL) {
|
| 480 | for (i = 0; i < qi->max_depth; i++) {
|
| 481 | if (*(qi->skbuff_vector + i) != NULL)
|
| 482 | dev_kfree_skb_any(*(qi->skbuff_vector + i));
|
| 483 | }
|
| 484 | kfree(qi->skbuff_vector);
|
| 485 | }
|
| 486 |
|
| 487 | if (qi->mmsg_vector != NULL) {
|
| 488 | mmsg_vector = qi->mmsg_vector;
|
| 489 | for (i = 0; i < qi->max_depth; i++) {
|
| 490 | iov = mmsg_vector->msg_hdr.msg_iov;
|
| 491 | if (iov != NULL) {
|
| 492 | if ((vp->header_size > 0) &&
|
| 493 | (iov->iov_base != NULL))
|
| 494 | kfree(iov->iov_base);
|
| 495 | kfree(iov);
|
| 496 | }
|
| 497 | mmsg_vector++;
|
| 498 | }
|
| 499 | kfree(qi->mmsg_vector);
|
| 500 | }
|
| 501 | kfree(qi);
|
| 502 | }
|
| 503 |
|
| 504 | |
| 505 | * Queue constructor. Create a queue with a given side.
|
| 506 |
|
| 507 | static struct vector_queue *create_queue(
|
| 508 | struct vector_private *vp,
|
| 509 | int max_size,
|
| 510 | int header_size,
|
| 511 | int num_extra_frags)
|
| 512 | {
|
| 513 | struct vector_queue *result;
|
| 514 | int i;
|
| 515 | struct iovec *iov;
|
| 516 | struct mmsghdr *mmsg_vector;
|
| 517 |
|
| 518 | result = kmalloc_obj(struct vector_queue);
|
| 519 | if (result == NULL)
|
| 520 | return NULL;
|
| 521 | result->max_depth = max_size;
|
| 522 | result->dev = vp->dev;
|
| 523 | result->mmsg_vector = kmalloc(
|
| 524 | (sizeof(struct mmsghdr) * max_size), GFP_KERNEL);
|
| 525 | if (result->mmsg_vector == NULL)
|
| 526 | goto out_mmsg_fail;
|
| 527 | result->skbuff_vector = kmalloc(
|
| 528 | (sizeof(void *) * max_size), GFP_KERNEL);
|
| 529 | if (result->skbuff_vector == NULL)
|
| 530 | goto out_skb_fail;
|
| 531 |
|
| 532 |
|
| 533 |
|
| 534 | mmsg_vector = result->mmsg_vector;
|
| 535 | for (i = 0; i < max_size; i++) {
|
| 536 | |
| 537 | * what to free on destruction
|
| 538 |
|
| 539 | *(result->skbuff_vector + i) = NULL;
|
| 540 | mmsg_vector->msg_hdr.msg_iov = NULL;
|
| 541 | mmsg_vector++;
|
| 542 | }
|
| 543 | mmsg_vector = result->mmsg_vector;
|
| 544 | result->max_iov_frags = num_extra_frags;
|
| 545 | for (i = 0; i < max_size; i++) {
|
| 546 | if (vp->header_size > 0)
|
| 547 | iov = kmalloc_objs(struct iovec, 3 + num_extra_frags);
|
| 548 | else
|
| 549 | iov = kmalloc_objs(struct iovec, 2 + num_extra_frags);
|
| 550 | if (iov == NULL)
|
| 551 | goto out_fail;
|
| 552 | mmsg_vector->msg_hdr.msg_iov = iov;
|
| 553 | mmsg_vector->msg_hdr.msg_iovlen = 1;
|
| 554 | mmsg_vector->msg_hdr.msg_control = NULL;
|
| 555 | mmsg_vector->msg_hdr.msg_controllen = 0;
|
| 556 | mmsg_vector->msg_hdr.msg_flags = MSG_DONTWAIT;
|
| 557 | mmsg_vector->msg_hdr.msg_name = NULL;
|
| 558 | mmsg_vector->msg_hdr.msg_namelen = 0;
|
| 559 | if (vp->header_size > 0) {
|
| 560 | iov->iov_base = kmalloc(header_size, GFP_KERNEL);
|
| 561 | if (iov->iov_base == NULL)
|
| 562 | goto out_fail;
|
| 563 | iov->iov_len = header_size;
|
| 564 | mmsg_vector->msg_hdr.msg_iovlen = 2;
|
| 565 | iov++;
|
| 566 | }
|
| 567 | iov->iov_base = NULL;
|
| 568 | iov->iov_len = 0;
|
| 569 | mmsg_vector++;
|
| 570 | }
|
| 571 | spin_lock_init(&result->head_lock);
|
| 572 | spin_lock_init(&result->tail_lock);
|
| 573 | atomic_set(&result->queue_depth, 0);
|
| 574 | result->head = 0;
|
| 575 | result->tail = 0;
|
| 576 | return result;
|
| 577 | out_skb_fail:
|
| 578 | kfree(result->mmsg_vector);
|
| 579 | out_mmsg_fail:
|
| 580 | kfree(result);
|
| 581 | return NULL;
|
| 582 | out_fail:
|
| 583 | destroy_queue(result);
|
| 584 | return NULL;
|
| 585 | }
|
| 586 |
|
| 587 | |
| 588 | * We do not use the RX queue as a proper wraparound queue for now
|
| 589 | * This is not necessary because the consumption via napi_gro_receive()
|
| 590 | * happens in-line. While we can try using the return code of
|
| 591 | * netif_rx() for flow control there are no drivers doing this today.
|
| 592 | * For this RX specific use we ignore the tail/head locks and
|
| 593 | * just read into a prepared queue filled with skbuffs.
|
| 594 |
|
| 595 |
|
| 596 | static struct sk_buff *prep_skb(
|
| 597 | struct vector_private *vp,
|
| 598 | struct user_msghdr *msg)
|
| 599 | {
|
| 600 | int linear = vp->max_packet + vp->headroom + SAFETY_MARGIN;
|
| 601 | struct sk_buff *result;
|
| 602 | int iov_index = 0, len;
|
| 603 | struct iovec *iov = msg->msg_iov;
|
| 604 | int err, nr_frags, frag;
|
| 605 | skb_frag_t *skb_frag;
|
| 606 |
|
| 607 | if (vp->req_size <= linear)
|
| 608 | len = linear;
|
| 609 | else
|
| 610 | len = vp->req_size;
|
| 611 | result = alloc_skb_with_frags(
|
| 612 | linear,
|
| 613 | len - vp->max_packet,
|
| 614 | 3,
|
| 615 | &err,
|
| 616 | GFP_ATOMIC
|
| 617 | );
|
| 618 | if (vp->header_size > 0)
|
| 619 | iov_index++;
|
| 620 | if (result == NULL) {
|
| 621 | iov[iov_index].iov_base = NULL;
|
| 622 | iov[iov_index].iov_len = 0;
|
| 623 | goto done;
|
| 624 | }
|
| 625 | skb_reserve(result, vp->headroom);
|
| 626 | result->dev = vp->dev;
|
| 627 | skb_put(result, vp->max_packet);
|
| 628 | result->data_len = len - vp->max_packet;
|
| 629 | result->len += len - vp->max_packet;
|
| 630 | skb_reset_mac_header(result);
|
| 631 | result->ip_summed = CHECKSUM_NONE;
|
| 632 | iov[iov_index].iov_base = result->data;
|
| 633 | iov[iov_index].iov_len = vp->max_packet;
|
| 634 | iov_index++;
|
| 635 |
|
| 636 | nr_frags = skb_shinfo(result)->nr_frags;
|
| 637 | for (frag = 0; frag < nr_frags; frag++) {
|
| 638 | skb_frag = &skb_shinfo(result)->frags[frag];
|
| 639 | iov[iov_index].iov_base = skb_frag_address_safe(skb_frag);
|
| 640 | if (iov[iov_index].iov_base != NULL)
|
| 641 | iov[iov_index].iov_len = skb_frag_size(skb_frag);
|
| 642 | else
|
| 643 | iov[iov_index].iov_len = 0;
|
| 644 | iov_index++;
|
| 645 | }
|
| 646 | done:
|
| 647 | msg->msg_iovlen = iov_index;
|
| 648 | return result;
|
| 649 | }
|
| 650 |
|
| 651 |
|
| 652 |
|
| 653 |
|
| 654 | static void prep_queue_for_rx(struct vector_queue *qi)
|
| 655 | {
|
| 656 | struct vector_private *vp = netdev_priv(qi->dev);
|
| 657 | struct mmsghdr *mmsg_vector = qi->mmsg_vector;
|
| 658 | void **skbuff_vector = qi->skbuff_vector;
|
| 659 | int i, queue_depth;
|
| 660 |
|
| 661 | queue_depth = atomic_read(&qi->queue_depth);
|
| 662 |
|
| 663 | if (queue_depth == 0)
|
| 664 | return;
|
| 665 |
|
| 666 | |
| 667 | * have to do the tail wraparound math for it.
|
| 668 |
|
| 669 |
|
| 670 | qi->head = qi->tail = 0;
|
| 671 |
|
| 672 | for (i = 0; i < queue_depth; i++) {
|
| 673 | |
| 674 | * iov argument still performs an RX, just drops the packet
|
| 675 | * This allows us stop faffing around with a "drop buffer"
|
| 676 |
|
| 677 |
|
| 678 | *skbuff_vector = prep_skb(vp, &mmsg_vector->msg_hdr);
|
| 679 | skbuff_vector++;
|
| 680 | mmsg_vector++;
|
| 681 | }
|
| 682 | atomic_set(&qi->queue_depth, 0);
|
| 683 | }
|
| 684 |
|
| 685 | static struct vector_device *find_device(int n)
|
| 686 | {
|
| 687 | struct vector_device *device;
|
| 688 | struct list_head *ele;
|
| 689 |
|
| 690 | spin_lock(&vector_devices_lock);
|
| 691 | list_for_each(ele, &vector_devices) {
|
| 692 | device = list_entry(ele, struct vector_device, list);
|
| 693 | if (device->unit == n)
|
| 694 | goto out;
|
| 695 | }
|
| 696 | device = NULL;
|
| 697 | out:
|
| 698 | spin_unlock(&vector_devices_lock);
|
| 699 | return device;
|
| 700 | }
|
| 701 |
|
| 702 | static int vector_parse(char *str, int *index_out, char **str_out,
|
| 703 | char **error_out)
|
| 704 | {
|
| 705 | int n, err;
|
| 706 | char *start = str;
|
| 707 |
|
| 708 | while ((*str != ':') && (strlen(str) > 1))
|
| 709 | str++;
|
| 710 | if (*str != ':') {
|
| 711 | *error_out = "Expected ':' after device number";
|
| 712 | return -EINVAL;
|
| 713 | }
|
| 714 | *str = '\0';
|
| 715 |
|
| 716 | err = kstrtouint(start, 0, &n);
|
| 717 | if (err < 0) {
|
| 718 | *error_out = "Bad device number";
|
| 719 | return err;
|
| 720 | }
|
| 721 |
|
| 722 | str++;
|
| 723 | if (find_device(n)) {
|
| 724 | *error_out = "Device already configured";
|
| 725 | return -EINVAL;
|
| 726 | }
|
| 727 |
|
| 728 | *index_out = n;
|
| 729 | *str_out = str;
|
| 730 | return 0;
|
| 731 | }
|
| 732 |
|
| 733 | static int vector_config(char *str, char **error_out)
|
| 734 | {
|
| 735 | int err, n;
|
| 736 | char *params;
|
| 737 | struct arglist *parsed;
|
| 738 |
|
| 739 | err = vector_parse(str, &n, ¶ms, error_out);
|
| 740 | if (err != 0)
|
| 741 | return err;
|
| 742 |
|
| 743 | |
| 744 | * driver. We should copy it to make sure things do not go wrong
|
| 745 | * later.
|
| 746 |
|
| 747 |
|
| 748 | params = kstrdup(params, GFP_KERNEL);
|
| 749 | if (params == NULL) {
|
| 750 | *error_out = "vector_config failed to strdup string";
|
| 751 | return -ENOMEM;
|
| 752 | }
|
| 753 |
|
| 754 | parsed = uml_parse_vector_ifspec(params);
|
| 755 |
|
| 756 | if (parsed == NULL) {
|
| 757 | *error_out = "vector_config failed to parse parameters";
|
| 758 | kfree(params);
|
| 759 | return -EINVAL;
|
| 760 | }
|
| 761 |
|
| 762 | vector_eth_configure(n, parsed);
|
| 763 | return 0;
|
| 764 | }
|
| 765 |
|
| 766 | static int vector_id(char **str, int *start_out, int *end_out)
|
| 767 | {
|
| 768 | char *end;
|
| 769 | int n;
|
| 770 |
|
| 771 | n = simple_strtoul(*str, &end, 0);
|
| 772 | if ((*end != '\0') || (end == *str))
|
| 773 | return -1;
|
| 774 |
|
| 775 | *start_out = n;
|
| 776 | *end_out = n;
|
| 777 | *str = end;
|
| 778 | return n;
|
| 779 | }
|
| 780 |
|
| 781 | static int vector_remove(int n, char **error_out)
|
| 782 | {
|
| 783 | struct vector_device *vec_d;
|
| 784 | struct net_device *dev;
|
| 785 | struct vector_private *vp;
|
| 786 |
|
| 787 | vec_d = find_device(n);
|
| 788 | if (vec_d == NULL)
|
| 789 | return -ENODEV;
|
| 790 | dev = vec_d->dev;
|
| 791 | vp = netdev_priv(dev);
|
| 792 | if (vp->fds != NULL)
|
| 793 | return -EBUSY;
|
| 794 | unregister_netdev(dev);
|
| 795 | platform_device_unregister(&vec_d->pdev);
|
| 796 | return 0;
|
| 797 | }
|
| 798 |
|
| 799 | |
| 800 | * There is no shared per-transport initialization code, so
|
| 801 | * we will just initialize each interface one by one and
|
| 802 | * add them to a list
|
| 803 |
|
| 804 |
|
| 805 | static struct platform_driver uml_net_driver = {
|
| 806 | .driver = {
|
| 807 | .name = DRIVER_NAME,
|
| 808 | },
|
| 809 | };
|
| 810 |
|
| 811 |
|
| 812 | static void vector_device_release(struct device *dev)
|
| 813 | {
|
| 814 | struct vector_device *device =
|
| 815 | container_of(dev, struct vector_device, pdev.dev);
|
| 816 | struct net_device *netdev = device->dev;
|
| 817 |
|
| 818 | list_del(&device->list);
|
| 819 | kfree(device);
|
| 820 | free_netdev(netdev);
|
| 821 | }
|
| 822 |
|
| 823 | |
| 824 | * explicitly specifies not to use recvmmsg vector RX.
|
| 825 |
|
| 826 |
|
| 827 | static int vector_legacy_rx(struct vector_private *vp)
|
| 828 | {
|
| 829 | int pkt_len;
|
| 830 | struct user_msghdr hdr;
|
| 831 | struct iovec iov[2 + MAX_IOV_SIZE];
|
| 832 | int iovpos = 0;
|
| 833 | struct sk_buff *skb;
|
| 834 | int header_check;
|
| 835 |
|
| 836 | hdr.msg_name = NULL;
|
| 837 | hdr.msg_namelen = 0;
|
| 838 | hdr.msg_iov = (struct iovec *) &iov;
|
| 839 | hdr.msg_control = NULL;
|
| 840 | hdr.msg_controllen = 0;
|
| 841 | hdr.msg_flags = 0;
|
| 842 |
|
| 843 | if (vp->header_size > 0) {
|
| 844 | iov[0].iov_base = vp->header_rxbuffer;
|
| 845 | iov[0].iov_len = vp->header_size;
|
| 846 | }
|
| 847 |
|
| 848 | skb = prep_skb(vp, &hdr);
|
| 849 |
|
| 850 | if (skb == NULL) {
|
| 851 | |
| 852 | * anything with it.
|
| 853 |
|
| 854 | iov[iovpos].iov_base = drop_buffer;
|
| 855 | iov[iovpos].iov_len = DROP_BUFFER_SIZE;
|
| 856 | hdr.msg_iovlen = 1;
|
| 857 | vp->dev->stats.rx_dropped++;
|
| 858 | }
|
| 859 |
|
| 860 | pkt_len = uml_vector_recvmsg(vp->fds->rx_fd, &hdr, 0);
|
| 861 | if (pkt_len < 0) {
|
| 862 | vp->in_error = true;
|
| 863 | return pkt_len;
|
| 864 | }
|
| 865 |
|
| 866 | if (skb != NULL) {
|
| 867 | if (pkt_len > vp->header_size) {
|
| 868 | if (vp->header_size > 0) {
|
| 869 | header_check = vp->verify_header(
|
| 870 | vp->header_rxbuffer, skb, vp);
|
| 871 | if (header_check < 0) {
|
| 872 | dev_kfree_skb_irq(skb);
|
| 873 | vp->dev->stats.rx_dropped++;
|
| 874 | vp->estats.rx_encaps_errors++;
|
| 875 | return 0;
|
| 876 | }
|
| 877 | if (header_check > 0) {
|
| 878 | vp->estats.rx_csum_offload_good++;
|
| 879 | skb->ip_summed = CHECKSUM_UNNECESSARY;
|
| 880 | }
|
| 881 | }
|
| 882 | pskb_trim(skb, pkt_len - vp->rx_header_size);
|
| 883 | skb->protocol = eth_type_trans(skb, skb->dev);
|
| 884 | vp->dev->stats.rx_bytes += skb->len;
|
| 885 | vp->dev->stats.rx_packets++;
|
| 886 | napi_gro_receive(&vp->napi, skb);
|
| 887 | } else {
|
| 888 | dev_kfree_skb_irq(skb);
|
| 889 | }
|
| 890 | }
|
| 891 | return pkt_len;
|
| 892 | }
|
| 893 |
|
| 894 | |
| 895 | * Packet at a time TX which falls back to vector TX if the
|
| 896 | * underlying transport is busy.
|
| 897 |
|
| 898 |
|
| 899 |
|
| 900 |
|
| 901 | static int writev_tx(struct vector_private *vp, struct sk_buff *skb)
|
| 902 | {
|
| 903 | struct iovec iov[3 + MAX_IOV_SIZE];
|
| 904 | int iov_count, pkt_len = 0;
|
| 905 |
|
| 906 | iov[0].iov_base = vp->header_txbuffer;
|
| 907 | iov_count = prep_msg(vp, skb, (struct iovec *) &iov);
|
| 908 |
|
| 909 | if (iov_count < 1)
|
| 910 | goto drop;
|
| 911 |
|
| 912 | pkt_len = uml_vector_writev(
|
| 913 | vp->fds->tx_fd,
|
| 914 | (struct iovec *) &iov,
|
| 915 | iov_count
|
| 916 | );
|
| 917 |
|
| 918 | if (pkt_len < 0)
|
| 919 | goto drop;
|
| 920 |
|
| 921 | netif_trans_update(vp->dev);
|
| 922 | netif_wake_queue(vp->dev);
|
| 923 |
|
| 924 | if (pkt_len > 0) {
|
| 925 | vp->dev->stats.tx_bytes += skb->len;
|
| 926 | vp->dev->stats.tx_packets++;
|
| 927 | } else {
|
| 928 | vp->dev->stats.tx_dropped++;
|
| 929 | }
|
| 930 | consume_skb(skb);
|
| 931 | return pkt_len;
|
| 932 | drop:
|
| 933 | vp->dev->stats.tx_dropped++;
|
| 934 | consume_skb(skb);
|
| 935 | if (pkt_len < 0)
|
| 936 | vp->in_error = true;
|
| 937 | return pkt_len;
|
| 938 | }
|
| 939 |
|
| 940 | |
| 941 | * Receive as many messages as we can in one call using the special
|
| 942 | * mmsg vector matched to an skb vector which we prepared earlier.
|
| 943 |
|
| 944 |
|
| 945 | static int vector_mmsg_rx(struct vector_private *vp, int budget)
|
| 946 | {
|
| 947 | int packet_count, i;
|
| 948 | struct vector_queue *qi = vp->rx_queue;
|
| 949 | struct sk_buff *skb;
|
| 950 | struct mmsghdr *mmsg_vector = qi->mmsg_vector;
|
| 951 | void **skbuff_vector = qi->skbuff_vector;
|
| 952 | int header_check;
|
| 953 |
|
| 954 | |
| 955 | * iovs are updated to point to them.
|
| 956 |
|
| 957 |
|
| 958 | prep_queue_for_rx(qi);
|
| 959 |
|
| 960 |
|
| 961 |
|
| 962 | if (budget > qi->max_depth)
|
| 963 | budget = qi->max_depth;
|
| 964 |
|
| 965 | packet_count = uml_vector_recvmmsg(
|
| 966 | vp->fds->rx_fd, qi->mmsg_vector, budget, 0);
|
| 967 |
|
| 968 | if (packet_count < 0)
|
| 969 | vp->in_error = true;
|
| 970 |
|
| 971 | if (packet_count <= 0)
|
| 972 | return packet_count;
|
| 973 |
|
| 974 | |
| 975 | * The queue_depth tells us how many buffers have been used and how
|
| 976 | * many do we need to prep the next time prep_queue_for_rx() is called.
|
| 977 |
|
| 978 |
|
| 979 | atomic_add(packet_count, &qi->queue_depth);
|
| 980 |
|
| 981 | for (i = 0; i < packet_count; i++) {
|
| 982 | skb = (*skbuff_vector);
|
| 983 | if (mmsg_vector->msg_len > vp->header_size) {
|
| 984 | if (vp->header_size > 0) {
|
| 985 | header_check = vp->verify_header(
|
| 986 | mmsg_vector->msg_hdr.msg_iov->iov_base,
|
| 987 | skb,
|
| 988 | vp
|
| 989 | );
|
| 990 | if (header_check < 0) {
|
| 991 | |
| 992 | * We can actually keep this skb and reuse it,
|
| 993 | * but that will make the prep logic too
|
| 994 | * complex.
|
| 995 |
|
| 996 | dev_kfree_skb_irq(skb);
|
| 997 | vp->estats.rx_encaps_errors++;
|
| 998 | (*skbuff_vector) = NULL;
|
| 999 | mmsg_vector++;
|
| 1000 | skbuff_vector++;
|
| 1001 | continue;
|
| 1002 | }
|
| 1003 | if (header_check > 0) {
|
| 1004 | vp->estats.rx_csum_offload_good++;
|
| 1005 | skb->ip_summed = CHECKSUM_UNNECESSARY;
|
| 1006 | }
|
| 1007 | }
|
| 1008 | pskb_trim(skb,
|
| 1009 | mmsg_vector->msg_len - vp->rx_header_size);
|
| 1010 | skb->protocol = eth_type_trans(skb, skb->dev);
|
| 1011 | |
| 1012 | * We do not need to lock on updating stats here
|
| 1013 | * The interrupt loop is non-reentrant.
|
| 1014 |
|
| 1015 | vp->dev->stats.rx_bytes += skb->len;
|
| 1016 | vp->dev->stats.rx_packets++;
|
| 1017 | napi_gro_receive(&vp->napi, skb);
|
| 1018 | } else {
|
| 1019 | |
| 1020 | * We can actually keep this skb and reuse it,
|
| 1021 | * but that will make the prep logic too complex.
|
| 1022 |
|
| 1023 | if (skb != NULL)
|
| 1024 | dev_kfree_skb_irq(skb);
|
| 1025 | }
|
| 1026 | (*skbuff_vector) = NULL;
|
| 1027 |
|
| 1028 | mmsg_vector++;
|
| 1029 | skbuff_vector++;
|
| 1030 | }
|
| 1031 | if (packet_count > 0) {
|
| 1032 | if (vp->estats.rx_queue_max < packet_count)
|
| 1033 | vp->estats.rx_queue_max = packet_count;
|
| 1034 | vp->estats.rx_queue_running_average =
|
| 1035 | (vp->estats.rx_queue_running_average + packet_count) >> 1;
|
| 1036 | }
|
| 1037 | return packet_count;
|
| 1038 | }
|
| 1039 |
|
| 1040 | static int vector_net_start_xmit(struct sk_buff *skb, struct net_device *dev)
|
| 1041 | {
|
| 1042 | struct vector_private *vp = netdev_priv(dev);
|
| 1043 | int queue_depth = 0;
|
| 1044 |
|
| 1045 | if (vp->in_error) {
|
| 1046 | deactivate_fd(vp->fds->rx_fd, vp->rx_irq);
|
| 1047 | if ((vp->fds->rx_fd != vp->fds->tx_fd) && (vp->tx_irq != 0))
|
| 1048 | deactivate_fd(vp->fds->tx_fd, vp->tx_irq);
|
| 1049 | return NETDEV_TX_BUSY;
|
| 1050 | }
|
| 1051 |
|
| 1052 | if ((vp->options & VECTOR_TX) == 0) {
|
| 1053 | writev_tx(vp, skb);
|
| 1054 | return NETDEV_TX_OK;
|
| 1055 | }
|
| 1056 |
|
| 1057 | |
| 1058 | * packet at a time mode as there is no device queue
|
| 1059 |
|
| 1060 |
|
| 1061 | netdev_sent_queue(vp->dev, skb->len);
|
| 1062 | queue_depth = vector_enqueue(vp->tx_queue, skb);
|
| 1063 |
|
| 1064 | if (queue_depth < vp->tx_queue->max_depth && netdev_xmit_more()) {
|
| 1065 | mod_timer(&vp->tl, vp->coalesce);
|
| 1066 | return NETDEV_TX_OK;
|
| 1067 | } else {
|
| 1068 | queue_depth = vector_send(vp->tx_queue);
|
| 1069 | if (queue_depth > 0)
|
| 1070 | napi_schedule(&vp->napi);
|
| 1071 | }
|
| 1072 |
|
| 1073 | return NETDEV_TX_OK;
|
| 1074 | }
|
| 1075 |
|
| 1076 | static irqreturn_t vector_rx_interrupt(int irq, void *dev_id)
|
| 1077 | {
|
| 1078 | struct net_device *dev = dev_id;
|
| 1079 | struct vector_private *vp = netdev_priv(dev);
|
| 1080 |
|
| 1081 | if (!netif_running(dev))
|
| 1082 | return IRQ_NONE;
|
| 1083 | napi_schedule(&vp->napi);
|
| 1084 | return IRQ_HANDLED;
|
| 1085 |
|
| 1086 | }
|
| 1087 |
|
| 1088 | static irqreturn_t vector_tx_interrupt(int irq, void *dev_id)
|
| 1089 | {
|
| 1090 | struct net_device *dev = dev_id;
|
| 1091 | struct vector_private *vp = netdev_priv(dev);
|
| 1092 |
|
| 1093 | if (!netif_running(dev))
|
| 1094 | return IRQ_NONE;
|
| 1095 | |
| 1096 | * -EAGAIN or -ENOBUFFS from sendmmsg. Otherwise
|
| 1097 | * we ignore it. In the future, it may be worth
|
| 1098 | * it to improve the IRQ controller a bit to make
|
| 1099 | * tweaking the IRQ mask less costly
|
| 1100 |
|
| 1101 |
|
| 1102 | napi_schedule(&vp->napi);
|
| 1103 | return IRQ_HANDLED;
|
| 1104 |
|
| 1105 | }
|
| 1106 |
|
| 1107 | static int irq_rr;
|
| 1108 |
|
| 1109 | static int vector_net_close(struct net_device *dev)
|
| 1110 | {
|
| 1111 | struct vector_private *vp = netdev_priv(dev);
|
| 1112 |
|
| 1113 | netif_stop_queue(dev);
|
| 1114 | timer_delete(&vp->tl);
|
| 1115 |
|
| 1116 | vp->opened = false;
|
| 1117 |
|
| 1118 | if (vp->fds == NULL)
|
| 1119 | return 0;
|
| 1120 |
|
| 1121 |
|
| 1122 | if (vp->rx_irq > 0) {
|
| 1123 | um_free_irq(vp->rx_irq, dev);
|
| 1124 | vp->rx_irq = 0;
|
| 1125 | }
|
| 1126 | if (vp->tx_irq > 0) {
|
| 1127 | um_free_irq(vp->tx_irq, dev);
|
| 1128 | vp->tx_irq = 0;
|
| 1129 | }
|
| 1130 | napi_disable(&vp->napi);
|
| 1131 | netif_napi_del(&vp->napi);
|
| 1132 | if (vp->fds->rx_fd > 0) {
|
| 1133 | if (vp->bpf)
|
| 1134 | uml_vector_detach_bpf(vp->fds->rx_fd, vp->bpf);
|
| 1135 | os_close_file(vp->fds->rx_fd);
|
| 1136 | vp->fds->rx_fd = -1;
|
| 1137 | }
|
| 1138 | if (vp->fds->tx_fd > 0) {
|
| 1139 | os_close_file(vp->fds->tx_fd);
|
| 1140 | vp->fds->tx_fd = -1;
|
| 1141 | }
|
| 1142 | if (vp->bpf != NULL)
|
| 1143 | kfree(vp->bpf->filter);
|
| 1144 | kfree(vp->bpf);
|
| 1145 | vp->bpf = NULL;
|
| 1146 | kfree(vp->fds->remote_addr);
|
| 1147 | kfree(vp->transport_data);
|
| 1148 | kfree(vp->header_rxbuffer);
|
| 1149 | kfree(vp->header_txbuffer);
|
| 1150 | if (vp->rx_queue != NULL)
|
| 1151 | destroy_queue(vp->rx_queue);
|
| 1152 | if (vp->tx_queue != NULL)
|
| 1153 | destroy_queue(vp->tx_queue);
|
| 1154 | kfree(vp->fds);
|
| 1155 | vp->fds = NULL;
|
| 1156 | vp->in_error = false;
|
| 1157 | return 0;
|
| 1158 | }
|
| 1159 |
|
| 1160 | static int vector_poll(struct napi_struct *napi, int budget)
|
| 1161 | {
|
| 1162 | struct vector_private *vp = container_of(napi, struct vector_private, napi);
|
| 1163 | int work_done = 0;
|
| 1164 | int err;
|
| 1165 | bool tx_enqueued = false;
|
| 1166 |
|
| 1167 | if ((vp->options & VECTOR_TX) != 0)
|
| 1168 | tx_enqueued = (vector_send(vp->tx_queue) > 0);
|
| 1169 | spin_lock(&vp->rx_queue->head_lock);
|
| 1170 | if ((vp->options & VECTOR_RX) > 0)
|
| 1171 | err = vector_mmsg_rx(vp, budget);
|
| 1172 | else {
|
| 1173 | err = vector_legacy_rx(vp);
|
| 1174 | if (err > 0)
|
| 1175 | err = 1;
|
| 1176 | }
|
| 1177 | spin_unlock(&vp->rx_queue->head_lock);
|
| 1178 | if (err > 0)
|
| 1179 | work_done += err;
|
| 1180 |
|
| 1181 | if (tx_enqueued || err > 0)
|
| 1182 | napi_schedule(napi);
|
| 1183 | if (work_done <= budget)
|
| 1184 | napi_complete_done(napi, work_done);
|
| 1185 | return work_done;
|
| 1186 | }
|
| 1187 |
|
| 1188 | static void vector_reset_tx(struct work_struct *work)
|
| 1189 | {
|
| 1190 | struct vector_private *vp =
|
| 1191 | container_of(work, struct vector_private, reset_tx);
|
| 1192 | netdev_reset_queue(vp->dev);
|
| 1193 | netif_start_queue(vp->dev);
|
| 1194 | netif_wake_queue(vp->dev);
|
| 1195 | }
|
| 1196 |
|
| 1197 | static int vector_net_open(struct net_device *dev)
|
| 1198 | {
|
| 1199 | struct vector_private *vp = netdev_priv(dev);
|
| 1200 | int err = -EINVAL;
|
| 1201 | struct vector_device *vdevice;
|
| 1202 |
|
| 1203 | if (vp->opened)
|
| 1204 | return -ENXIO;
|
| 1205 | vp->opened = true;
|
| 1206 |
|
| 1207 | vp->bpf = uml_vector_user_bpf(get_bpf_file(vp->parsed));
|
| 1208 |
|
| 1209 | vp->fds = uml_vector_user_open(vp->unit, vp->parsed);
|
| 1210 |
|
| 1211 | if (vp->fds == NULL)
|
| 1212 | goto out_close;
|
| 1213 |
|
| 1214 | if (build_transport_data(vp) < 0)
|
| 1215 | goto out_close;
|
| 1216 |
|
| 1217 | if ((vp->options & VECTOR_RX) > 0) {
|
| 1218 | vp->rx_queue = create_queue(
|
| 1219 | vp,
|
| 1220 | get_depth(vp->parsed),
|
| 1221 | vp->rx_header_size,
|
| 1222 | MAX_IOV_SIZE
|
| 1223 | );
|
| 1224 | atomic_set(&vp->rx_queue->queue_depth, get_depth(vp->parsed));
|
| 1225 | } else {
|
| 1226 | vp->header_rxbuffer = kmalloc(
|
| 1227 | vp->rx_header_size,
|
| 1228 | GFP_KERNEL
|
| 1229 | );
|
| 1230 | if (vp->header_rxbuffer == NULL)
|
| 1231 | goto out_close;
|
| 1232 | }
|
| 1233 | if ((vp->options & VECTOR_TX) > 0) {
|
| 1234 | vp->tx_queue = create_queue(
|
| 1235 | vp,
|
| 1236 | get_depth(vp->parsed),
|
| 1237 | vp->header_size,
|
| 1238 | MAX_IOV_SIZE
|
| 1239 | );
|
| 1240 | } else {
|
| 1241 | vp->header_txbuffer = kmalloc(vp->header_size, GFP_KERNEL);
|
| 1242 | if (vp->header_txbuffer == NULL)
|
| 1243 | goto out_close;
|
| 1244 | }
|
| 1245 |
|
| 1246 | netif_napi_add_weight(vp->dev, &vp->napi, vector_poll,
|
| 1247 | get_depth(vp->parsed));
|
| 1248 | napi_enable(&vp->napi);
|
| 1249 |
|
| 1250 |
|
| 1251 | err = um_request_irq(
|
| 1252 | irq_rr + VECTOR_BASE_IRQ, vp->fds->rx_fd,
|
| 1253 | IRQ_READ, vector_rx_interrupt,
|
| 1254 | IRQF_SHARED, dev->name, dev);
|
| 1255 | if (err < 0) {
|
| 1256 | netdev_err(dev, "vector_open: failed to get rx irq(%d)\n", err);
|
| 1257 | err = -ENETUNREACH;
|
| 1258 | goto out_close;
|
| 1259 | }
|
| 1260 | vp->rx_irq = irq_rr + VECTOR_BASE_IRQ;
|
| 1261 | dev->irq = irq_rr + VECTOR_BASE_IRQ;
|
| 1262 | irq_rr = (irq_rr + 1) % VECTOR_IRQ_SPACE;
|
| 1263 |
|
| 1264 |
|
| 1265 | if ((vp->options & VECTOR_TX) > 0) {
|
| 1266 | err = um_request_irq(
|
| 1267 | irq_rr + VECTOR_BASE_IRQ, vp->fds->tx_fd,
|
| 1268 | IRQ_WRITE, vector_tx_interrupt,
|
| 1269 | IRQF_SHARED, dev->name, dev);
|
| 1270 | if (err < 0) {
|
| 1271 | netdev_err(dev,
|
| 1272 | "vector_open: failed to get tx irq(%d)\n", err);
|
| 1273 | err = -ENETUNREACH;
|
| 1274 | goto out_close;
|
| 1275 | }
|
| 1276 | vp->tx_irq = irq_rr + VECTOR_BASE_IRQ;
|
| 1277 | irq_rr = (irq_rr + 1) % VECTOR_IRQ_SPACE;
|
| 1278 | }
|
| 1279 |
|
| 1280 | if ((vp->options & VECTOR_QDISC_BYPASS) != 0) {
|
| 1281 | if (!uml_raw_enable_qdisc_bypass(vp->fds->rx_fd))
|
| 1282 | vp->options |= VECTOR_BPF;
|
| 1283 | }
|
| 1284 | if (((vp->options & VECTOR_BPF) != 0) && (vp->bpf == NULL))
|
| 1285 | vp->bpf = uml_vector_default_bpf(dev->dev_addr);
|
| 1286 |
|
| 1287 | if (vp->bpf != NULL)
|
| 1288 | uml_vector_attach_bpf(vp->fds->rx_fd, vp->bpf);
|
| 1289 |
|
| 1290 | netif_start_queue(dev);
|
| 1291 | vector_reset_stats(vp);
|
| 1292 |
|
| 1293 | |
| 1294 | * is full when we get here. In this case, new data is never queued,
|
| 1295 | * SIGIOs never arrive, and the net never works.
|
| 1296 |
|
| 1297 |
|
| 1298 | napi_schedule(&vp->napi);
|
| 1299 |
|
| 1300 | vdevice = find_device(vp->unit);
|
| 1301 | vdevice->opened = 1;
|
| 1302 |
|
| 1303 | if ((vp->options & VECTOR_TX) != 0)
|
| 1304 | add_timer(&vp->tl);
|
| 1305 | return 0;
|
| 1306 | out_close:
|
| 1307 | vector_net_close(dev);
|
| 1308 | return err;
|
| 1309 | }
|
| 1310 |
|
| 1311 |
|
| 1312 | static void vector_net_set_multicast_list(struct net_device *dev)
|
| 1313 | {
|
| 1314 |
|
| 1315 | return;
|
| 1316 | }
|
| 1317 |
|
| 1318 | static void vector_net_tx_timeout(struct net_device *dev, unsigned int txqueue)
|
| 1319 | {
|
| 1320 | struct vector_private *vp = netdev_priv(dev);
|
| 1321 |
|
| 1322 | vp->estats.tx_timeout_count++;
|
| 1323 | netif_trans_update(dev);
|
| 1324 | schedule_work(&vp->reset_tx);
|
| 1325 | }
|
| 1326 |
|
| 1327 | static netdev_features_t vector_fix_features(struct net_device *dev,
|
| 1328 | netdev_features_t features)
|
| 1329 | {
|
| 1330 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
|
| 1331 | return features;
|
| 1332 | }
|
| 1333 |
|
| 1334 | static int vector_set_features(struct net_device *dev,
|
| 1335 | netdev_features_t features)
|
| 1336 | {
|
| 1337 | struct vector_private *vp = netdev_priv(dev);
|
| 1338 | |
| 1339 | * no way to negotiate it on raw sockets, so we can change
|
| 1340 | * only our side.
|
| 1341 |
|
| 1342 | if (features & NETIF_F_GRO)
|
| 1343 |
|
| 1344 | vp->req_size = 65536;
|
| 1345 | else
|
| 1346 |
|
| 1347 | vp->req_size = vp->max_packet + vp->headroom + SAFETY_MARGIN;
|
| 1348 | return 0;
|
| 1349 | }
|
| 1350 |
|
| 1351 | #ifdef CONFIG_NET_POLL_CONTROLLER
|
| 1352 | static void vector_net_poll_controller(struct net_device *dev)
|
| 1353 | {
|
| 1354 | disable_irq(dev->irq);
|
| 1355 | vector_rx_interrupt(dev->irq, dev);
|
| 1356 | enable_irq(dev->irq);
|
| 1357 | }
|
| 1358 | #endif
|
| 1359 |
|
| 1360 | static void vector_net_get_drvinfo(struct net_device *dev,
|
| 1361 | struct ethtool_drvinfo *info)
|
| 1362 | {
|
| 1363 | strscpy(info->driver, DRIVER_NAME);
|
| 1364 | }
|
| 1365 |
|
| 1366 | static int vector_net_load_bpf_flash(struct net_device *dev,
|
| 1367 | struct ethtool_flash *efl)
|
| 1368 | {
|
| 1369 | struct vector_private *vp = netdev_priv(dev);
|
| 1370 | struct vector_device *vdevice;
|
| 1371 | const struct firmware *fw;
|
| 1372 | int result = 0;
|
| 1373 |
|
| 1374 | if (!(vp->options & VECTOR_BPF_FLASH)) {
|
| 1375 | netdev_err(dev, "loading firmware not permitted: %s\n", efl->data);
|
| 1376 | return -1;
|
| 1377 | }
|
| 1378 |
|
| 1379 | if (vp->bpf != NULL) {
|
| 1380 | if (vp->opened)
|
| 1381 | uml_vector_detach_bpf(vp->fds->rx_fd, vp->bpf);
|
| 1382 | kfree(vp->bpf->filter);
|
| 1383 | vp->bpf->filter = NULL;
|
| 1384 | } else {
|
| 1385 | vp->bpf = kmalloc_obj(struct sock_fprog, GFP_ATOMIC);
|
| 1386 | if (vp->bpf == NULL) {
|
| 1387 | netdev_err(dev, "failed to allocate memory for firmware\n");
|
| 1388 | goto flash_fail;
|
| 1389 | }
|
| 1390 | }
|
| 1391 |
|
| 1392 | vdevice = find_device(vp->unit);
|
| 1393 |
|
| 1394 | if (request_firmware(&fw, efl->data, &vdevice->pdev.dev))
|
| 1395 | goto flash_fail;
|
| 1396 |
|
| 1397 | vp->bpf->filter = kmemdup(fw->data, fw->size, GFP_ATOMIC);
|
| 1398 | if (!vp->bpf->filter)
|
| 1399 | goto free_buffer;
|
| 1400 |
|
| 1401 | vp->bpf->len = fw->size / sizeof(struct sock_filter);
|
| 1402 | release_firmware(fw);
|
| 1403 |
|
| 1404 | if (vp->opened)
|
| 1405 | result = uml_vector_attach_bpf(vp->fds->rx_fd, vp->bpf);
|
| 1406 |
|
| 1407 | return result;
|
| 1408 |
|
| 1409 | free_buffer:
|
| 1410 | release_firmware(fw);
|
| 1411 |
|
| 1412 | flash_fail:
|
| 1413 | if (vp->bpf != NULL)
|
| 1414 | kfree(vp->bpf->filter);
|
| 1415 | kfree(vp->bpf);
|
| 1416 | vp->bpf = NULL;
|
| 1417 | return -1;
|
| 1418 | }
|
| 1419 |
|
| 1420 | static void vector_get_ringparam(struct net_device *netdev,
|
| 1421 | struct ethtool_ringparam *ring,
|
| 1422 | struct kernel_ethtool_ringparam *kernel_ring,
|
| 1423 | struct netlink_ext_ack *extack)
|
| 1424 | {
|
| 1425 | struct vector_private *vp = netdev_priv(netdev);
|
| 1426 |
|
| 1427 | ring->rx_max_pending = vp->rx_queue->max_depth;
|
| 1428 | ring->tx_max_pending = vp->tx_queue->max_depth;
|
| 1429 | ring->rx_pending = vp->rx_queue->max_depth;
|
| 1430 | ring->tx_pending = vp->tx_queue->max_depth;
|
| 1431 | }
|
| 1432 |
|
| 1433 | static void vector_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
|
| 1434 | {
|
| 1435 | switch (stringset) {
|
| 1436 | case ETH_SS_TEST:
|
| 1437 | *buf = '\0';
|
| 1438 | break;
|
| 1439 | case ETH_SS_STATS:
|
| 1440 | memcpy(buf, ðtool_stats_keys, sizeof(ethtool_stats_keys));
|
| 1441 | break;
|
| 1442 | default:
|
| 1443 | WARN_ON(1);
|
| 1444 | break;
|
| 1445 | }
|
| 1446 | }
|
| 1447 |
|
| 1448 | static int vector_get_sset_count(struct net_device *dev, int sset)
|
| 1449 | {
|
| 1450 | switch (sset) {
|
| 1451 | case ETH_SS_TEST:
|
| 1452 | return 0;
|
| 1453 | case ETH_SS_STATS:
|
| 1454 | return VECTOR_NUM_STATS;
|
| 1455 | default:
|
| 1456 | return -EOPNOTSUPP;
|
| 1457 | }
|
| 1458 | }
|
| 1459 |
|
| 1460 | static void vector_get_ethtool_stats(struct net_device *dev,
|
| 1461 | struct ethtool_stats *estats,
|
| 1462 | u64 *tmp_stats)
|
| 1463 | {
|
| 1464 | struct vector_private *vp = netdev_priv(dev);
|
| 1465 |
|
| 1466 | |
| 1467 | * rx/tx which are protected by the head locks
|
| 1468 | * grabbing these locks here ensures they are up
|
| 1469 | * to date.
|
| 1470 |
|
| 1471 |
|
| 1472 | spin_lock(&vp->tx_queue->head_lock);
|
| 1473 | spin_lock(&vp->rx_queue->head_lock);
|
| 1474 | memcpy(tmp_stats, &vp->estats, sizeof(struct vector_estats));
|
| 1475 | spin_unlock(&vp->rx_queue->head_lock);
|
| 1476 | spin_unlock(&vp->tx_queue->head_lock);
|
| 1477 | }
|
| 1478 |
|
| 1479 | static int vector_get_coalesce(struct net_device *netdev,
|
| 1480 | struct ethtool_coalesce *ec,
|
| 1481 | struct kernel_ethtool_coalesce *kernel_coal,
|
| 1482 | struct netlink_ext_ack *extack)
|
| 1483 | {
|
| 1484 | struct vector_private *vp = netdev_priv(netdev);
|
| 1485 |
|
| 1486 | ec->tx_coalesce_usecs = (vp->coalesce * 1000000) / HZ;
|
| 1487 | return 0;
|
| 1488 | }
|
| 1489 |
|
| 1490 | static int vector_set_coalesce(struct net_device *netdev,
|
| 1491 | struct ethtool_coalesce *ec,
|
| 1492 | struct kernel_ethtool_coalesce *kernel_coal,
|
| 1493 | struct netlink_ext_ack *extack)
|
| 1494 | {
|
| 1495 | struct vector_private *vp = netdev_priv(netdev);
|
| 1496 |
|
| 1497 | vp->coalesce = (ec->tx_coalesce_usecs * HZ) / 1000000;
|
| 1498 | if (vp->coalesce == 0)
|
| 1499 | vp->coalesce = 1;
|
| 1500 | return 0;
|
| 1501 | }
|
| 1502 |
|
| 1503 | static const struct ethtool_ops vector_net_ethtool_ops = {
|
| 1504 | .supported_coalesce_params = ETHTOOL_COALESCE_TX_USECS,
|
| 1505 | .get_drvinfo = vector_net_get_drvinfo,
|
| 1506 | .get_link = ethtool_op_get_link,
|
| 1507 | .get_ts_info = ethtool_op_get_ts_info,
|
| 1508 | .get_ringparam = vector_get_ringparam,
|
| 1509 | .get_strings = vector_get_strings,
|
| 1510 | .get_sset_count = vector_get_sset_count,
|
| 1511 | .get_ethtool_stats = vector_get_ethtool_stats,
|
| 1512 | .get_coalesce = vector_get_coalesce,
|
| 1513 | .set_coalesce = vector_set_coalesce,
|
| 1514 | .flash_device = vector_net_load_bpf_flash,
|
| 1515 | };
|
| 1516 |
|
| 1517 |
|
| 1518 | static const struct net_device_ops vector_netdev_ops = {
|
| 1519 | .ndo_open = vector_net_open,
|
| 1520 | .ndo_stop = vector_net_close,
|
| 1521 | .ndo_start_xmit = vector_net_start_xmit,
|
| 1522 | .ndo_set_rx_mode = vector_net_set_multicast_list,
|
| 1523 | .ndo_tx_timeout = vector_net_tx_timeout,
|
| 1524 | .ndo_set_mac_address = eth_mac_addr,
|
| 1525 | .ndo_validate_addr = eth_validate_addr,
|
| 1526 | .ndo_fix_features = vector_fix_features,
|
| 1527 | .ndo_set_features = vector_set_features,
|
| 1528 | #ifdef CONFIG_NET_POLL_CONTROLLER
|
| 1529 | .ndo_poll_controller = vector_net_poll_controller,
|
| 1530 | #endif
|
| 1531 | };
|
| 1532 |
|
| 1533 | static void vector_timer_expire(struct timer_list *t)
|
| 1534 | {
|
| 1535 | struct vector_private *vp = timer_container_of(vp, t, tl);
|
| 1536 |
|
| 1537 | vp->estats.tx_kicks++;
|
| 1538 | napi_schedule(&vp->napi);
|
| 1539 | }
|
| 1540 |
|
| 1541 | static void vector_setup_etheraddr(struct net_device *dev, char *str)
|
| 1542 | {
|
| 1543 | u8 addr[ETH_ALEN];
|
| 1544 |
|
| 1545 | if (str == NULL)
|
| 1546 | goto random;
|
| 1547 |
|
| 1548 | if (!mac_pton(str, addr)) {
|
| 1549 | netdev_err(dev,
|
| 1550 | "Failed to parse '%s' as an ethernet address\n", str);
|
| 1551 | goto random;
|
| 1552 | }
|
| 1553 | if (is_multicast_ether_addr(addr)) {
|
| 1554 | netdev_err(dev,
|
| 1555 | "Attempt to assign a multicast ethernet address to a device disallowed\n");
|
| 1556 | goto random;
|
| 1557 | }
|
| 1558 | if (!is_valid_ether_addr(addr)) {
|
| 1559 | netdev_err(dev,
|
| 1560 | "Attempt to assign an invalid ethernet address to a device disallowed\n");
|
| 1561 | goto random;
|
| 1562 | }
|
| 1563 | if (!is_local_ether_addr(addr)) {
|
| 1564 | netdev_warn(dev, "Warning: Assigning a globally valid ethernet address to a device\n");
|
| 1565 | netdev_warn(dev, "You should set the 2nd rightmost bit in the first byte of the MAC,\n");
|
| 1566 | netdev_warn(dev, "i.e. %02x:%02x:%02x:%02x:%02x:%02x\n",
|
| 1567 | addr[0] | 0x02, addr[1], addr[2], addr[3], addr[4], addr[5]);
|
| 1568 | }
|
| 1569 | eth_hw_addr_set(dev, addr);
|
| 1570 | return;
|
| 1571 |
|
| 1572 | random:
|
| 1573 | netdev_info(dev, "Choosing a random ethernet address\n");
|
| 1574 | eth_hw_addr_random(dev);
|
| 1575 | }
|
| 1576 |
|
| 1577 | static void vector_eth_configure(
|
| 1578 | int n,
|
| 1579 | struct arglist *def
|
| 1580 | )
|
| 1581 | {
|
| 1582 | struct vector_device *device;
|
| 1583 | struct net_device *dev;
|
| 1584 | struct vector_private *vp;
|
| 1585 | int err;
|
| 1586 |
|
| 1587 | device = kzalloc_obj(*device);
|
| 1588 | if (device == NULL) {
|
| 1589 | pr_err("Failed to allocate struct vector_device for vec%d\n", n);
|
| 1590 | return;
|
| 1591 | }
|
| 1592 | dev = alloc_etherdev(sizeof(struct vector_private));
|
| 1593 | if (dev == NULL) {
|
| 1594 | pr_err("Failed to allocate struct net_device for vec%d\n", n);
|
| 1595 | goto out_free_device;
|
| 1596 | }
|
| 1597 |
|
| 1598 | dev->mtu = get_mtu(def);
|
| 1599 |
|
| 1600 | INIT_LIST_HEAD(&device->list);
|
| 1601 | device->unit = n;
|
| 1602 |
|
| 1603 | |
| 1604 | * netdevice, that is OK, register_netdev{,ice}() will notice this
|
| 1605 | * and fail.
|
| 1606 |
|
| 1607 | snprintf(dev->name, sizeof(dev->name), "vec%d", n);
|
| 1608 | vector_setup_etheraddr(dev, uml_vector_fetch_arg(def, "mac"));
|
| 1609 | vp = netdev_priv(dev);
|
| 1610 |
|
| 1611 |
|
| 1612 | if (!driver_registered) {
|
| 1613 | platform_driver_register(¨_net_driver);
|
| 1614 | driver_registered = 1;
|
| 1615 | }
|
| 1616 | device->pdev.id = n;
|
| 1617 | device->pdev.name = DRIVER_NAME;
|
| 1618 | device->pdev.dev.release = vector_device_release;
|
| 1619 | dev_set_drvdata(&device->pdev.dev, device);
|
| 1620 | if (platform_device_register(&device->pdev))
|
| 1621 | goto out_free_netdev;
|
| 1622 | SET_NETDEV_DEV(dev, &device->pdev.dev);
|
| 1623 |
|
| 1624 | device->dev = dev;
|
| 1625 |
|
| 1626 | INIT_LIST_HEAD(&vp->list);
|
| 1627 | vp->dev = dev;
|
| 1628 | vp->unit = n;
|
| 1629 | vp->options = get_transport_options(def);
|
| 1630 | vp->parsed = def;
|
| 1631 | vp->max_packet = get_mtu(def) + ETH_HEADER_OTHER;
|
| 1632 | |
| 1633 | * TODO - we need to calculate headroom so that ip header
|
| 1634 | * is 16 byte aligned all the time
|
| 1635 |
|
| 1636 | vp->headroom = get_headroom(def);
|
| 1637 | vp->coalesce = 2;
|
| 1638 | vp->req_size = get_req_size(def);
|
| 1639 |
|
| 1640 | dev->features = dev->hw_features = (NETIF_F_SG | NETIF_F_FRAGLIST);
|
| 1641 | INIT_WORK(&vp->reset_tx, vector_reset_tx);
|
| 1642 |
|
| 1643 | timer_setup(&vp->tl, vector_timer_expire, 0);
|
| 1644 |
|
| 1645 |
|
| 1646 | dev->netdev_ops = &vector_netdev_ops;
|
| 1647 | dev->ethtool_ops = &vector_net_ethtool_ops;
|
| 1648 | dev->watchdog_timeo = (HZ >> 1);
|
| 1649 |
|
| 1650 | dev->irq = 0;
|
| 1651 |
|
| 1652 | rtnl_lock();
|
| 1653 | err = register_netdevice(dev);
|
| 1654 | rtnl_unlock();
|
| 1655 | if (err)
|
| 1656 | goto out_undo_user_init;
|
| 1657 |
|
| 1658 | spin_lock(&vector_devices_lock);
|
| 1659 | list_add(&device->list, &vector_devices);
|
| 1660 | spin_unlock(&vector_devices_lock);
|
| 1661 |
|
| 1662 | return;
|
| 1663 |
|
| 1664 | out_undo_user_init:
|
| 1665 | return;
|
| 1666 | out_free_netdev:
|
| 1667 | free_netdev(dev);
|
| 1668 | out_free_device:
|
| 1669 | kfree(device);
|
| 1670 | }
|
| 1671 |
|
| 1672 |
|
| 1673 |
|
| 1674 |
|
| 1675 | |
| 1676 | * Invoked late in the init
|
| 1677 |
|
| 1678 |
|
| 1679 | static int __init vector_init(void)
|
| 1680 | {
|
| 1681 | struct list_head *ele;
|
| 1682 | struct vector_cmd_line_arg *def;
|
| 1683 | struct arglist *parsed;
|
| 1684 |
|
| 1685 | list_for_each(ele, &vec_cmd_line) {
|
| 1686 | def = list_entry(ele, struct vector_cmd_line_arg, list);
|
| 1687 | parsed = uml_parse_vector_ifspec(def->arguments);
|
| 1688 | if (parsed != NULL)
|
| 1689 | vector_eth_configure(def->unit, parsed);
|
| 1690 | }
|
| 1691 | return 0;
|
| 1692 | }
|
| 1693 |
|
| 1694 |
|
| 1695 | |
| 1696 | * arguments until a proper vector_init is called
|
| 1697 | * later
|
| 1698 |
|
| 1699 |
|
| 1700 | static int __init vector_setup(char *str)
|
| 1701 | {
|
| 1702 | char *error;
|
| 1703 | int n, err;
|
| 1704 | struct vector_cmd_line_arg *new;
|
| 1705 |
|
| 1706 | err = vector_parse(str, &n, &str, &error);
|
| 1707 | if (err) {
|
| 1708 | pr_err("Couldn't parse '%s': %s\n", str, error);
|
| 1709 | return 1;
|
| 1710 | }
|
| 1711 | new = memblock_alloc_or_panic(sizeof(*new), SMP_CACHE_BYTES);
|
| 1712 | INIT_LIST_HEAD(&new->list);
|
| 1713 | new->unit = n;
|
| 1714 | new->arguments = str;
|
| 1715 | list_add_tail(&new->list, &vec_cmd_line);
|
| 1716 | return 1;
|
| 1717 | }
|
| 1718 |
|
| 1719 | __setup("vec", vector_setup);
|
| 1720 | __uml_help(vector_setup,
|
| 1721 | "vec[0-9]+:<option>=<value>,<option>=<value>\n"
|
| 1722 | " Configure a vector io network device.\n\n"
|
| 1723 | );
|
| 1724 |
|
| 1725 | late_initcall(vector_init);
|
| 1726 |
|
| 1727 | static struct mc_device vector_mc = {
|
| 1728 | .list = LIST_HEAD_INIT(vector_mc.list),
|
| 1729 | .name = "vec",
|
| 1730 | .config = vector_config,
|
| 1731 | .get_config = NULL,
|
| 1732 | .id = vector_id,
|
| 1733 | .remove = vector_remove,
|
| 1734 | };
|
| 1735 |
|
| 1736 | #ifdef CONFIG_INET
|
| 1737 | static int vector_inetaddr_event(
|
| 1738 | struct notifier_block *this,
|
| 1739 | unsigned long event,
|
| 1740 | void *ptr)
|
| 1741 | {
|
| 1742 | return NOTIFY_DONE;
|
| 1743 | }
|
| 1744 |
|
| 1745 | static struct notifier_block vector_inetaddr_notifier = {
|
| 1746 | .notifier_call = vector_inetaddr_event,
|
| 1747 | };
|
| 1748 |
|
| 1749 | static void inet_register(void)
|
| 1750 | {
|
| 1751 | register_inetaddr_notifier(&vector_inetaddr_notifier);
|
| 1752 | }
|
| 1753 | #else
|
| 1754 | static inline void inet_register(void)
|
| 1755 | {
|
| 1756 | }
|
| 1757 | #endif
|
| 1758 |
|
| 1759 | static int vector_net_init(void)
|
| 1760 | {
|
| 1761 | mconsole_register_dev(&vector_mc);
|
| 1762 | inet_register();
|
| 1763 | return 0;
|
| 1764 | }
|
| 1765 |
|
| 1766 | __initcall(vector_net_init);
|
| 1767 |
|
| 1768 |
|
| 1769 |
|