target/linux/layerscape/patches-5.4/809-jailhouse-0001-ivshmem-net-virtual-network-device-for-Jailhouse.patch

   1 From 7f48bab7c7b468961cf70efa1d86a75173e3987a Mon Sep 17 00:00:00 2001
   2 From: Mans Rullgard <mans@mansr.com>
   3 Date: Thu, 26 May 2016 16:04:02 +0100
   4 Subject: [PATCH] ivshmem-net: virtual network device for Jailhouse
   5
   6 Work in progress.
   7
   8 (cherry picked from commit ed818547b45e652db57d5966efe336ed646feb45)
   9 ---
  10  drivers/net/Kconfig       |   4 +
  11  drivers/net/Makefile      |   2 +
  12  drivers/net/ivshmem-net.c | 923 ++++++++++++++++++++++++++++++++++++++++++++++
  13  3 files changed, 929 insertions(+)
  14  create mode 100644 drivers/net/ivshmem-net.c
  15
  16 --- a/drivers/net/Kconfig
  17 +++ b/drivers/net/Kconfig
  18 @@ -528,4 +528,8 @@ config NET_FAILOVER
  19           a VM with direct attached VF by failing over to the paravirtual
  20           datapath when the VF is unplugged.
  21
  22 +config IVSHMEM_NET
  23 +       tristate "IVSHMEM virtual network device"
  24 +       depends on PCI
  25 +
  26  endif # NETDEVICES
  27 --- a/drivers/net/Makefile
  28 +++ b/drivers/net/Makefile
  29 @@ -79,3 +79,5 @@ thunderbolt-net-y += thunderbolt.o
  30  obj-$(CONFIG_THUNDERBOLT_NET) += thunderbolt-net.o
  31  obj-$(CONFIG_NETDEVSIM) += netdevsim/
  32  obj-$(CONFIG_NET_FAILOVER) += net_failover.o
  33 +
  34 +obj-$(CONFIG_IVSHMEM_NET) += ivshmem-net.o
  35 --- /dev/null
  36 +++ b/drivers/net/ivshmem-net.c
  37 @@ -0,0 +1,923 @@
  38 +/*
  39 + * Copyright 2016 Mans Rullgard <mans@mansr.com>
  40 + *
  41 + * This program is free software; you can redistribute it and/or modify
  42 + * it under the terms of the GNU General Public License as published by
  43 + * the Free Software Foundation; either version 2 of the License, or
  44 + * (at your option) any later version.
  45 + *
  46 + * This program is distributed in the hope that it will be useful,
  47 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
  48 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  49 + * GNU General Public License for more details.
  50 + *
  51 + * You should have received a copy of the GNU General Public License
  52 + * along with this program; if not, see <http://www.gnu.org/licenses/>.
  53 + */
  54 +
  55 +#include <linux/kernel.h>
  56 +#include <linux/module.h>
  57 +#include <linux/pci.h>
  58 +#include <linux/io.h>
  59 +#include <linux/bitops.h>
  60 +#include <linux/interrupt.h>
  61 +#include <linux/netdevice.h>
  62 +#include <linux/etherdevice.h>
  63 +#include <linux/rtnetlink.h>
  64 +#include <linux/virtio_ring.h>
  65 +
  66 +#define DRV_NAME "ivshmem-net"
  67 +
  68 +#define JAILHOUSE_CFG_SHMEM_PTR        0x40
  69 +#define JAILHOUSE_CFG_SHMEM_SZ 0x48
  70 +
  71 +#define IVSHM_NET_STATE_RESET  0
  72 +#define IVSHM_NET_STATE_INIT   1
  73 +#define IVSHM_NET_STATE_READY  2
  74 +#define IVSHM_NET_STATE_RUN    3
  75 +
  76 +#define IVSHM_NET_MTU_MIN 256
  77 +#define IVSHM_NET_MTU_MAX 65535
  78 +#define IVSHM_NET_MTU_DEF 16384
  79 +
  80 +#define IVSHM_NET_FRAME_SIZE(s) ALIGN(18 + (s), SMP_CACHE_BYTES)
  81 +
  82 +#define IVSHM_NET_VQ_ALIGN 64
  83 +
  84 +struct ivshmem_regs {
  85 +       u32 imask;
  86 +       u32 istat;
  87 +       u32 ivpos;
  88 +       u32 doorbell;
  89 +       u32 lstate;
  90 +       u32 rstate;
  91 +};
  92 +
  93 +struct ivshm_net_queue {
  94 +       struct vring vr;
  95 +       u32 free_head;
  96 +       u32 num_free;
  97 +       u32 num_added;
  98 +       u16 last_avail_idx;
  99 +       u16 last_used_idx;
 100 +
 101 +       void *data;
 102 +       void *end;
 103 +       u32 size;
 104 +       u32 head;
 105 +       u32 tail;
 106 +};
 107 +
 108 +struct ivshm_net_stats {
 109 +       u32 interrupts;
 110 +       u32 tx_packets;
 111 +       u32 tx_notify;
 112 +       u32 tx_pause;
 113 +       u32 rx_packets;
 114 +       u32 rx_notify;
 115 +       u32 napi_poll;
 116 +       u32 napi_complete;
 117 +       u32 napi_poll_n[10];
 118 +};
 119 +
 120 +struct ivshm_net {
 121 +       struct ivshm_net_queue rx;
 122 +       struct ivshm_net_queue tx;
 123 +
 124 +       u32 vrsize;
 125 +       u32 qlen;
 126 +       u32 qsize;
 127 +
 128 +       spinlock_t tx_free_lock;
 129 +       spinlock_t tx_clean_lock;
 130 +
 131 +       struct napi_struct napi;
 132 +
 133 +       u32 lstate;
 134 +       u32 rstate;
 135 +
 136 +       struct workqueue_struct *state_wq;
 137 +       struct work_struct state_work;
 138 +
 139 +       struct ivshm_net_stats stats;
 140 +
 141 +       struct ivshmem_regs __iomem *ivshm_regs;
 142 +       void *shm;
 143 +       phys_addr_t shmaddr;
 144 +       resource_size_t shmlen;
 145 +       u32 peer_id;
 146 +
 147 +       struct pci_dev *pdev;
 148 +       struct msix_entry msix;
 149 +       bool using_msix;
 150 +};
 151 +
 152 +static void *ivshm_net_desc_data(struct ivshm_net *in,
 153 +                                struct ivshm_net_queue *q,
 154 +                                struct vring_desc *desc,
 155 +                                u32 *len)
 156 +{
 157 +       u64 addr = READ_ONCE(desc->addr);
 158 +       u32 dlen = READ_ONCE(desc->len);
 159 +       void *data;
 160 +
 161 +       if (addr < in->shmaddr || desc->addr > in->shmaddr + in->shmlen)
 162 +               return NULL;
 163 +
 164 +       data = in->shm + (addr - in->shmaddr);
 165 +
 166 +       if (data < q->data || data >= q->end)
 167 +               return NULL;
 168 +
 169 +       if (dlen > q->end - data)
 170 +               return NULL;
 171 +
 172 +       *len = dlen;
 173 +
 174 +       return data;
 175 +}
 176 +
 177 +static void ivshm_net_init_queue(struct ivshm_net *in,
 178 +                                struct ivshm_net_queue *q,
 179 +                                void *mem, unsigned int len)
 180 +{
 181 +       memset(q, 0, sizeof(*q));
 182 +
 183 +       vring_init(&q->vr, len, mem, IVSHM_NET_VQ_ALIGN);
 184 +       q->data = mem + in->vrsize;
 185 +       q->end = q->data + in->qsize;
 186 +       q->size = in->qsize;
 187 +}
 188 +
 189 +static void ivshm_net_init_queues(struct net_device *ndev)
 190 +{
 191 +       struct ivshm_net *in = netdev_priv(ndev);
 192 +       int ivpos = readl(&in->ivshm_regs->ivpos);
 193 +       void *tx;
 194 +       void *rx;
 195 +       int i;
 196 +
 197 +       tx = in->shm +  ivpos * in->shmlen / 2;
 198 +       rx = in->shm + !ivpos * in->shmlen / 2;
 199 +
 200 +       memset(tx, 0, in->shmlen / 2);
 201 +
 202 +       ivshm_net_init_queue(in, &in->rx, rx, in->qlen);
 203 +       ivshm_net_init_queue(in, &in->tx, tx, in->qlen);
 204 +
 205 +       swap(in->rx.vr.used, in->tx.vr.used);
 206 +
 207 +       in->tx.num_free = in->tx.vr.num;
 208 +
 209 +       for (i = 0; i < in->tx.vr.num - 1; i++)
 210 +               in->tx.vr.desc[i].next = i + 1;
 211 +}
 212 +
 213 +static int ivshm_net_calc_qsize(struct net_device *ndev)
 214 +{
 215 +       struct ivshm_net *in = netdev_priv(ndev);
 216 +       unsigned int vrsize;
 217 +       unsigned int qsize;
 218 +       unsigned int qlen;
 219 +
 220 +       for (qlen = 4096; qlen > 32; qlen >>= 1) {
 221 +               vrsize = vring_size(qlen, IVSHM_NET_VQ_ALIGN);
 222 +               vrsize = ALIGN(vrsize, IVSHM_NET_VQ_ALIGN);
 223 +               if (vrsize < in->shmlen / 16)
 224 +                       break;
 225 +       }
 226 +
 227 +       if (vrsize > in->shmlen / 2)
 228 +               return -EINVAL;
 229 +
 230 +       qsize = in->shmlen / 2 - vrsize;
 231 +
 232 +       if (qsize < 4 * IVSHM_NET_MTU_MIN)
 233 +               return -EINVAL;
 234 +
 235 +       in->vrsize = vrsize;
 236 +       in->qlen = qlen;
 237 +       in->qsize = qsize;
 238 +
 239 +       return 0;
 240 +}
 241 +
 242 +static void ivshm_net_notify_tx(struct ivshm_net *in, unsigned int num)
 243 +{
 244 +       u16 evt, old, new;
 245 +
 246 +       virt_mb();
 247 +
 248 +       evt = READ_ONCE(vring_avail_event(&in->tx.vr));
 249 +       old = in->tx.last_avail_idx - num;
 250 +       new = in->tx.last_avail_idx;
 251 +
 252 +       if (vring_need_event(evt, new, old)) {
 253 +               writel(in->peer_id << 16, &in->ivshm_regs->doorbell);
 254 +               in->stats.tx_notify++;
 255 +       }
 256 +}
 257 +
 258 +static void ivshm_net_enable_rx_irq(struct ivshm_net *in)
 259 +{
 260 +       vring_avail_event(&in->rx.vr) = in->rx.last_avail_idx;
 261 +       virt_wmb();
 262 +}
 263 +
 264 +static void ivshm_net_notify_rx(struct ivshm_net *in, unsigned int num)
 265 +{
 266 +       u16 evt, old, new;
 267 +
 268 +       virt_mb();
 269 +
 270 +       evt = vring_used_event(&in->rx.vr);
 271 +       old = in->rx.last_used_idx - num;
 272 +       new = in->rx.last_used_idx;
 273 +
 274 +       if (vring_need_event(evt, new, old)) {
 275 +               writel(in->peer_id << 16, &in->ivshm_regs->doorbell);
 276 +               in->stats.rx_notify++;
 277 +       }
 278 +}
 279 +
 280 +static void ivshm_net_enable_tx_irq(struct ivshm_net *in)
 281 +{
 282 +       vring_used_event(&in->tx.vr) = in->tx.last_used_idx;
 283 +       virt_wmb();
 284 +}
 285 +
 286 +static bool ivshm_net_rx_avail(struct ivshm_net *in)
 287 +{
 288 +       virt_mb();
 289 +       return READ_ONCE(in->rx.vr.avail->idx) != in->rx.last_avail_idx;
 290 +}
 291 +
 292 +static size_t ivshm_net_tx_space(struct ivshm_net *in)
 293 +{
 294 +       struct ivshm_net_queue *tx = &in->tx;
 295 +       u32 tail = tx->tail;
 296 +       u32 head = tx->head;
 297 +       u32 space;
 298 +
 299 +       if (head < tail)
 300 +               space = tail - head;
 301 +       else
 302 +               space = max(tx->size - head, tail);
 303 +
 304 +       return space;
 305 +}
 306 +
 307 +static bool ivshm_net_tx_ok(struct ivshm_net *in, unsigned int mtu)
 308 +{
 309 +       return in->tx.num_free >= 2 &&
 310 +               ivshm_net_tx_space(in) >= 2 * IVSHM_NET_FRAME_SIZE(mtu);
 311 +}
 312 +
 313 +static u32 ivshm_net_tx_advance(struct ivshm_net_queue *q, u32 *pos, u32 len)
 314 +{
 315 +       u32 p = *pos;
 316 +
 317 +       len = IVSHM_NET_FRAME_SIZE(len);
 318 +
 319 +       if (q->size - p < len)
 320 +               p = 0;
 321 +       *pos = p + len;
 322 +
 323 +       return p;
 324 +}
 325 +
 326 +static int ivshm_net_tx_frame(struct net_device *ndev, struct sk_buff *skb)
 327 +{
 328 +       struct ivshm_net *in = netdev_priv(ndev);
 329 +       struct ivshm_net_queue *tx = &in->tx;
 330 +       struct vring *vr = &tx->vr;
 331 +       struct vring_desc *desc;
 332 +       unsigned int desc_idx;
 333 +       unsigned int avail;
 334 +       u32 head;
 335 +       void *buf;
 336 +
 337 +       BUG_ON(tx->num_free < 1);
 338 +
 339 +       spin_lock(&in->tx_free_lock);
 340 +       desc_idx = tx->free_head;
 341 +       desc = &vr->desc[desc_idx];
 342 +       tx->free_head = desc->next;
 343 +       tx->num_free--;
 344 +       spin_unlock(&in->tx_free_lock);
 345 +
 346 +       head = ivshm_net_tx_advance(tx, &tx->head, skb->len);
 347 +
 348 +       buf = tx->data + head;
 349 +       skb_copy_and_csum_dev(skb, buf);
 350 +
 351 +       desc->addr = in->shmaddr + (buf - in->shm);
 352 +       desc->len = skb->len;
 353 +
 354 +       avail = tx->last_avail_idx++ & (vr->num - 1);
 355 +       vr->avail->ring[avail] = desc_idx;
 356 +       tx->num_added++;
 357 +
 358 +       if (!skb->xmit_more) {
 359 +               virt_store_release(&vr->avail->idx, tx->last_avail_idx);
 360 +               ivshm_net_notify_tx(in, tx->num_added);
 361 +               tx->num_added = 0;
 362 +       }
 363 +
 364 +       return 0;
 365 +}
 366 +
 367 +static void ivshm_net_tx_clean(struct net_device *ndev)
 368 +{
 369 +       struct ivshm_net *in = netdev_priv(ndev);
 370 +       struct ivshm_net_queue *tx = &in->tx;
 371 +       struct vring *vr = &tx->vr;
 372 +       struct vring_desc *desc;
 373 +       struct vring_desc *fdesc;
 374 +       unsigned int used;
 375 +       unsigned int num;
 376 +       u16 used_idx;
 377 +       u16 last;
 378 +       u32 fhead;
 379 +
 380 +       if (!spin_trylock(&in->tx_clean_lock))
 381 +               return;
 382 +
 383 +       used_idx = virt_load_acquire(&vr->used->idx);
 384 +       last = tx->last_used_idx;
 385 +
 386 +       fdesc = NULL;
 387 +       num = 0;
 388 +
 389 +       while (last != used_idx) {
 390 +               void *data;
 391 +               u32 len;
 392 +               u32 tail;
 393 +
 394 +               used = vr->used->ring[last & (vr->num - 1)].id;
 395 +               if (used >= vr->num) {
 396 +                       netdev_err(ndev, "invalid tx used %d\n", used);
 397 +                       break;
 398 +               }
 399 +
 400 +               desc = &vr->desc[used];
 401 +
 402 +               data = ivshm_net_desc_data(in, &in->tx, desc, &len);
 403 +               if (!data) {
 404 +                       netdev_err(ndev, "bad tx descriptor\n");
 405 +                       break;
 406 +               }
 407 +
 408 +               tail = ivshm_net_tx_advance(tx, &tx->tail, len);
 409 +               if (data != tx->data + tail) {
 410 +                       netdev_err(ndev, "bad tx descriptor\n");
 411 +                       break;
 412 +               }
 413 +
 414 +               if (!num)
 415 +                       fdesc = desc;
 416 +               else
 417 +                       desc->next = fhead;
 418 +
 419 +               fhead = used;
 420 +               last++;
 421 +               num++;
 422 +       }
 423 +
 424 +       tx->last_used_idx = last;
 425 +
 426 +       spin_unlock(&in->tx_clean_lock);
 427 +
 428 +       if (num) {
 429 +               spin_lock(&in->tx_free_lock);
 430 +               fdesc->next = tx->free_head;
 431 +               tx->free_head = fhead;
 432 +               tx->num_free += num;
 433 +               BUG_ON(tx->num_free > vr->num);
 434 +               spin_unlock(&in->tx_free_lock);
 435 +       }
 436 +}
 437 +
 438 +static struct vring_desc *ivshm_net_rx_desc(struct net_device *ndev)
 439 +{
 440 +       struct ivshm_net *in = netdev_priv(ndev);
 441 +       struct ivshm_net_queue *rx = &in->rx;
 442 +       struct vring *vr = &rx->vr;
 443 +       unsigned int avail;
 444 +       u16 avail_idx;
 445 +
 446 +       avail_idx = virt_load_acquire(&vr->avail->idx);
 447 +
 448 +       if (avail_idx == rx->last_avail_idx)
 449 +               return NULL;
 450 +
 451 +       avail = vr->avail->ring[rx->last_avail_idx++ & (vr->num - 1)];
 452 +       if (avail >= vr->num) {
 453 +               netdev_err(ndev, "invalid rx avail %d\n", avail);
 454 +               return NULL;
 455 +       }
 456 +
 457 +       return &vr->desc[avail];
 458 +}
 459 +
 460 +static void ivshm_net_rx_finish(struct ivshm_net *in, struct vring_desc *desc)
 461 +{
 462 +       struct ivshm_net_queue *rx = &in->rx;
 463 +       struct vring *vr = &rx->vr;
 464 +       unsigned int desc_id = desc - vr->desc;
 465 +       unsigned int used;
 466 +
 467 +       used = rx->last_used_idx++ & (vr->num - 1);
 468 +       vr->used->ring[used].id = desc_id;
 469 +
 470 +       virt_store_release(&vr->used->idx, rx->last_used_idx);
 471 +}
 472 +
 473 +static int ivshm_net_poll(struct napi_struct *napi, int budget)
 474 +{
 475 +       struct net_device *ndev = napi->dev;
 476 +       struct ivshm_net *in = container_of(napi, struct ivshm_net, napi);
 477 +       int received = 0;
 478 +
 479 +       in->stats.napi_poll++;
 480 +
 481 +       ivshm_net_tx_clean(ndev);
 482 +
 483 +       while (received < budget) {
 484 +               struct vring_desc *desc;
 485 +               struct sk_buff *skb;
 486 +               void *data;
 487 +               u32 len;
 488 +
 489 +               desc = ivshm_net_rx_desc(ndev);
 490 +               if (!desc)
 491 +                       break;
 492 +
 493 +               data = ivshm_net_desc_data(in, &in->rx, desc, &len);
 494 +               if (!data) {
 495 +                       netdev_err(ndev, "bad rx descriptor\n");
 496 +                       break;
 497 +               }
 498 +
 499 +               skb = napi_alloc_skb(napi, len);
 500 +
 501 +               if (skb) {
 502 +                       memcpy(skb_put(skb, len), data, len);
 503 +                       skb->protocol = eth_type_trans(skb, ndev);
 504 +                       napi_gro_receive(napi, skb);
 505 +               }
 506 +
 507 +               ndev->stats.rx_packets++;
 508 +               ndev->stats.rx_bytes += len;
 509 +
 510 +               ivshm_net_rx_finish(in, desc);
 511 +               received++;
 512 +       }
 513 +
 514 +       if (received < budget) {
 515 +               in->stats.napi_complete++;
 516 +               napi_complete_done(napi, received);
 517 +               ivshm_net_enable_rx_irq(in);
 518 +               if (ivshm_net_rx_avail(in))
 519 +                       napi_schedule(napi);
 520 +       }
 521 +
 522 +       if (received)
 523 +               ivshm_net_notify_rx(in, received);
 524 +
 525 +       in->stats.rx_packets += received;
 526 +       in->stats.napi_poll_n[received ? 1 + min(ilog2(received), 8) : 0]++;
 527 +
 528 +       if (ivshm_net_tx_ok(in, ndev->mtu))
 529 +               netif_wake_queue(ndev);
 530 +
 531 +       return received;
 532 +}
 533 +
 534 +static netdev_tx_t ivshm_net_xmit(struct sk_buff *skb, struct net_device *ndev)
 535 +{
 536 +       struct ivshm_net *in = netdev_priv(ndev);
 537 +
 538 +       ivshm_net_tx_clean(ndev);
 539 +
 540 +       if (!ivshm_net_tx_ok(in, ndev->mtu)) {
 541 +               ivshm_net_enable_tx_irq(in);
 542 +               netif_stop_queue(ndev);
 543 +               skb->xmit_more = 0;
 544 +               in->stats.tx_pause++;
 545 +       }
 546 +
 547 +       ivshm_net_tx_frame(ndev, skb);
 548 +
 549 +       in->stats.tx_packets++;
 550 +       ndev->stats.tx_packets++;
 551 +       ndev->stats.tx_bytes += skb->len;
 552 +
 553 +       dev_consume_skb_any(skb);
 554 +
 555 +       return NETDEV_TX_OK;
 556 +}
 557 +
 558 +static void ivshm_net_set_state(struct ivshm_net *in, u32 state)
 559 +{
 560 +       virt_wmb();
 561 +       WRITE_ONCE(in->lstate, state);
 562 +       writel(state, &in->ivshm_regs->lstate);
 563 +}
 564 +
 565 +static void ivshm_net_run(struct net_device *ndev)
 566 +{
 567 +       struct ivshm_net *in = netdev_priv(ndev);
 568 +
 569 +       netif_start_queue(ndev);
 570 +       napi_enable(&in->napi);
 571 +       napi_schedule(&in->napi);
 572 +       ivshm_net_set_state(in, IVSHM_NET_STATE_RUN);
 573 +}
 574 +
 575 +static void ivshm_net_state_change(struct work_struct *work)
 576 +{
 577 +       struct ivshm_net *in = container_of(work, struct ivshm_net, state_work);
 578 +       struct net_device *ndev = in->napi.dev;
 579 +       u32 rstate = readl(&in->ivshm_regs->rstate);
 580 +
 581 +
 582 +       switch (in->lstate) {
 583 +       case IVSHM_NET_STATE_RESET:
 584 +               if (rstate < IVSHM_NET_STATE_READY)
 585 +                       ivshm_net_set_state(in, IVSHM_NET_STATE_INIT);
 586 +               break;
 587 +
 588 +       case IVSHM_NET_STATE_INIT:
 589 +               if (rstate > IVSHM_NET_STATE_RESET) {
 590 +                       ivshm_net_init_queues(ndev);
 591 +                       ivshm_net_set_state(in, IVSHM_NET_STATE_READY);
 592 +
 593 +                       rtnl_lock();
 594 +                       call_netdevice_notifiers(NETDEV_CHANGEADDR, ndev);
 595 +                       rtnl_unlock();
 596 +               }
 597 +               break;
 598 +
 599 +       case IVSHM_NET_STATE_READY:
 600 +               if (rstate >= IVSHM_NET_STATE_READY) {
 601 +                       netif_carrier_on(ndev);
 602 +                       if (ndev->flags & IFF_UP)
 603 +                               ivshm_net_run(ndev);
 604 +               } else {
 605 +                       netif_carrier_off(ndev);
 606 +                       ivshm_net_set_state(in, IVSHM_NET_STATE_RESET);
 607 +               }
 608 +               break;
 609 +
 610 +       case IVSHM_NET_STATE_RUN:
 611 +               if (rstate < IVSHM_NET_STATE_READY) {
 612 +                       netif_stop_queue(ndev);
 613 +                       napi_disable(&in->napi);
 614 +                       netif_carrier_off(ndev);
 615 +                       ivshm_net_set_state(in, IVSHM_NET_STATE_RESET);
 616 +               }
 617 +               break;
 618 +       }
 619 +
 620 +       virt_wmb();
 621 +       WRITE_ONCE(in->rstate, rstate);
 622 +}
 623 +
 624 +static bool ivshm_net_check_state(struct net_device *ndev)
 625 +{
 626 +       struct ivshm_net *in = netdev_priv(ndev);
 627 +       u32 rstate = readl(&in->ivshm_regs->rstate);
 628 +
 629 +       if (rstate != READ_ONCE(in->rstate) ||
 630 +           in->lstate != IVSHM_NET_STATE_RUN) {
 631 +               queue_work(in->state_wq, &in->state_work);
 632 +               return false;
 633 +       }
 634 +
 635 +       return true;
 636 +}
 637 +
 638 +static irqreturn_t ivshm_net_int(int irq, void *data)
 639 +{
 640 +       struct net_device *ndev = data;
 641 +       struct ivshm_net *in = netdev_priv(ndev);
 642 +
 643 +       in->stats.interrupts++;
 644 +
 645 +       ivshm_net_check_state(ndev);
 646 +       napi_schedule_irqoff(&in->napi);
 647 +
 648 +       return IRQ_HANDLED;
 649 +}
 650 +
 651 +static int ivshm_net_open(struct net_device *ndev)
 652 +{
 653 +       struct ivshm_net *in = netdev_priv(ndev);
 654 +
 655 +       netdev_reset_queue(ndev);
 656 +       ndev->operstate = IF_OPER_UP;
 657 +
 658 +       if (in->lstate == IVSHM_NET_STATE_READY)
 659 +               ivshm_net_run(ndev);
 660 +
 661 +       return 0;
 662 +}
 663 +
 664 +static int ivshm_net_stop(struct net_device *ndev)
 665 +{
 666 +       struct ivshm_net *in = netdev_priv(ndev);
 667 +
 668 +       ndev->operstate = IF_OPER_DOWN;
 669 +
 670 +       if (in->lstate == IVSHM_NET_STATE_RUN) {
 671 +               napi_disable(&in->napi);
 672 +               netif_stop_queue(ndev);
 673 +               ivshm_net_set_state(in, IVSHM_NET_STATE_READY);
 674 +       }
 675 +
 676 +       return 0;
 677 +}
 678 +
 679 +static int ivshm_net_change_mtu(struct net_device *ndev, int mtu)
 680 +{
 681 +       struct ivshm_net *in = netdev_priv(ndev);
 682 +       struct ivshm_net_queue *tx = &in->tx;
 683 +
 684 +       if (mtu < IVSHM_NET_MTU_MIN || mtu > IVSHM_NET_MTU_MAX)
 685 +               return -EINVAL;
 686 +
 687 +       if (in->tx.size / mtu < 4)
 688 +               return -EINVAL;
 689 +
 690 +       if (ivshm_net_tx_space(in) < 2 * IVSHM_NET_FRAME_SIZE(mtu))
 691 +               return -EBUSY;
 692 +
 693 +       if (in->tx.size - tx->head < IVSHM_NET_FRAME_SIZE(mtu) &&
 694 +           tx->head < tx->tail)
 695 +               return -EBUSY;
 696 +
 697 +       netif_tx_lock_bh(ndev);
 698 +       if (in->tx.size - tx->head < IVSHM_NET_FRAME_SIZE(mtu))
 699 +               tx->head = 0;
 700 +       netif_tx_unlock_bh(ndev);
 701 +
 702 +       ndev->mtu = mtu;
 703 +
 704 +       return 0;
 705 +}
 706 +
 707 +#ifdef CONFIG_NET_POLL_CONTROLLER
 708 +static void ivshm_net_poll_controller(struct net_device *ndev)
 709 +{
 710 +       struct ivshm_net *in = netdev_priv(ndev);
 711 +
 712 +       napi_schedule(&in->napi);
 713 +}
 714 +#endif
 715 +
 716 +static const struct net_device_ops ivshm_net_ops = {
 717 +       .ndo_open       = ivshm_net_open,
 718 +       .ndo_stop       = ivshm_net_stop,
 719 +       .ndo_start_xmit = ivshm_net_xmit,
 720 +       .ndo_change_mtu = ivshm_net_change_mtu,
 721 +#ifdef CONFIG_NET_POLL_CONTROLLER
 722 +       .ndo_poll_controller = ivshm_net_poll_controller,
 723 +#endif
 724 +};
 725 +
 726 +static const char ivshm_net_stats[][ETH_GSTRING_LEN] = {
 727 +       "interrupts",
 728 +       "tx_packets",
 729 +       "tx_notify",
 730 +       "tx_pause",
 731 +       "rx_packets",
 732 +       "rx_notify",
 733 +       "napi_poll",
 734 +       "napi_complete",
 735 +       "napi_poll_0",
 736 +       "napi_poll_1",
 737 +       "napi_poll_2",
 738 +       "napi_poll_4",
 739 +       "napi_poll_8",
 740 +       "napi_poll_16",
 741 +       "napi_poll_32",
 742 +       "napi_poll_64",
 743 +       "napi_poll_128",
 744 +       "napi_poll_256",
 745 +};
 746 +
 747 +#define NUM_STATS ARRAY_SIZE(ivshm_net_stats)
 748 +
 749 +static int ivshm_net_get_sset_count(struct net_device *ndev, int sset)
 750 +{
 751 +       if (sset == ETH_SS_STATS)
 752 +               return NUM_STATS;
 753 +
 754 +       return -EOPNOTSUPP;
 755 +}
 756 +
 757 +static void ivshm_net_get_strings(struct net_device *ndev, u32 sset, u8 *buf)
 758 +{
 759 +       if (sset == ETH_SS_STATS)
 760 +               memcpy(buf, &ivshm_net_stats, sizeof(ivshm_net_stats));
 761 +}
 762 +
 763 +static void ivshm_net_get_ethtool_stats(struct net_device *ndev,
 764 +                                       struct ethtool_stats *estats, u64 *st)
 765 +{
 766 +       struct ivshm_net *in = netdev_priv(ndev);
 767 +       unsigned int n = 0;
 768 +       unsigned int i;
 769 +
 770 +       st[n++] = in->stats.interrupts;
 771 +       st[n++] = in->stats.tx_packets;
 772 +       st[n++] = in->stats.tx_notify;
 773 +       st[n++] = in->stats.tx_pause;
 774 +       st[n++] = in->stats.rx_packets;
 775 +       st[n++] = in->stats.rx_notify;
 776 +       st[n++] = in->stats.napi_poll;
 777 +       st[n++] = in->stats.napi_complete;
 778 +
 779 +       for (i = 0; i < ARRAY_SIZE(in->stats.napi_poll_n); i++)
 780 +               st[n++] = in->stats.napi_poll_n[i];
 781 +
 782 +       memset(&in->stats, 0, sizeof(in->stats));
 783 +}
 784 +
 785 +static const struct ethtool_ops ivshm_net_ethtool_ops = {
 786 +       .get_sset_count         = ivshm_net_get_sset_count,
 787 +       .get_strings            = ivshm_net_get_strings,
 788 +       .get_ethtool_stats      = ivshm_net_get_ethtool_stats,
 789 +};
 790 +
 791 +static int ivshm_net_probe(struct pci_dev *pdev,
 792 +                          const struct pci_device_id *id)
 793 +{
 794 +       struct net_device *ndev;
 795 +       struct ivshm_net *in;
 796 +       struct ivshmem_regs __iomem *regs;
 797 +       resource_size_t shmaddr;
 798 +       resource_size_t shmlen;
 799 +       int interrupt;
 800 +       void *shm;
 801 +       u32 ivpos;
 802 +       int err;
 803 +
 804 +       err = pcim_enable_device(pdev);
 805 +       if (err) {
 806 +               dev_err(&pdev->dev, "pci_enable_device: %d\n", err);
 807 +               return err;
 808 +       }
 809 +
 810 +       err = pcim_iomap_regions(pdev, BIT(0), DRV_NAME);
 811 +       if (err) {
 812 +               dev_err(&pdev->dev, "pcim_iomap_regions: %d\n", err);
 813 +               return err;
 814 +       }
 815 +
 816 +       regs = pcim_iomap_table(pdev)[0];
 817 +
 818 +       shmlen = pci_resource_len(pdev, 2);
 819 +
 820 +       if (shmlen) {
 821 +               shmaddr = pci_resource_start(pdev, 2);
 822 +       } else {
 823 +               union { u64 v; u32 hl[2]; } val;
 824 +
 825 +               pci_read_config_dword(pdev, JAILHOUSE_CFG_SHMEM_PTR,
 826 +                                     &val.hl[0]);
 827 +               pci_read_config_dword(pdev, JAILHOUSE_CFG_SHMEM_PTR + 4,
 828 +                                     &val.hl[1]);
 829 +               shmaddr = val.v;
 830 +
 831 +               pci_read_config_dword(pdev, JAILHOUSE_CFG_SHMEM_SZ,
 832 +                                     &val.hl[0]);
 833 +               pci_read_config_dword(pdev, JAILHOUSE_CFG_SHMEM_SZ + 4,
 834 +                                     &val.hl[1]);
 835 +               shmlen = val.v;
 836 +       }
 837 +
 838 +
 839 +       if (!devm_request_mem_region(&pdev->dev, shmaddr, shmlen, DRV_NAME))
 840 +               return -EBUSY;
 841 +
 842 +       shm = devm_memremap(&pdev->dev, shmaddr, shmlen, MEMREMAP_WC);
 843 +       if (!shm)
 844 +               return -ENOMEM;
 845 +
 846 +       ivpos = readl(&regs->ivpos);
 847 +       if (ivpos > 1) {
 848 +               dev_err(&pdev->dev, "invalid IVPosition %d\n", ivpos);
 849 +               return -EINVAL;
 850 +       }
 851 +
 852 +       dev_info(&pdev->dev, "shared memory size %pa\n", &shmlen);
 853 +
 854 +       ndev = alloc_etherdev(sizeof(*in));
 855 +       if (!ndev)
 856 +               return -ENOMEM;
 857 +
 858 +       pci_set_drvdata(pdev, ndev);
 859 +       SET_NETDEV_DEV(ndev, &pdev->dev);
 860 +
 861 +       in = netdev_priv(ndev);
 862 +       in->ivshm_regs = regs;
 863 +       in->shm = shm;
 864 +       in->shmaddr = shmaddr;
 865 +       in->shmlen = shmlen;
 866 +       in->peer_id = !ivpos;
 867 +       in->pdev = pdev;
 868 +       spin_lock_init(&in->tx_free_lock);
 869 +       spin_lock_init(&in->tx_clean_lock);
 870 +
 871 +       err = ivshm_net_calc_qsize(ndev);
 872 +       if (err)
 873 +               goto err_free;
 874 +
 875 +       in->state_wq = alloc_ordered_workqueue(DRV_NAME, 0);
 876 +       if (!in->state_wq)
 877 +               goto err_free;
 878 +
 879 +       INIT_WORK(&in->state_work, ivshm_net_state_change);
 880 +
 881 +       eth_random_addr(ndev->dev_addr);
 882 +       ndev->netdev_ops = &ivshm_net_ops;
 883 +       ndev->ethtool_ops = &ivshm_net_ethtool_ops;
 884 +       ndev->mtu = min_t(u32, IVSHM_NET_MTU_DEF, in->qsize / 16);
 885 +       ndev->hw_features = NETIF_F_HW_CSUM | NETIF_F_SG;
 886 +       ndev->features = ndev->hw_features;
 887 +
 888 +       netif_carrier_off(ndev);
 889 +       netif_napi_add(ndev, &in->napi, ivshm_net_poll, NAPI_POLL_WEIGHT);
 890 +
 891 +       err = register_netdev(ndev);
 892 +       if (err)
 893 +               goto err_wq;
 894 +
 895 +       err = pci_enable_msix(pdev, &in->msix, 1);
 896 +       if (!err) {
 897 +               interrupt = in->msix.vector;
 898 +               in->using_msix = true;
 899 +       } else {
 900 +               interrupt = pdev->irq;
 901 +               in->using_msix = false;
 902 +       }
 903 +
 904 +       err = request_irq(interrupt, ivshm_net_int, 0, DRV_NAME, ndev);
 905 +       if (err)
 906 +               goto err_int;
 907 +
 908 +       pci_set_master(pdev);
 909 +
 910 +       writel(IVSHM_NET_STATE_RESET, &in->ivshm_regs->lstate);
 911 +
 912 +       return 0;
 913 +
 914 +err_int:
 915 +       if (in->using_msix)
 916 +               pci_disable_msix(pdev);
 917 +       unregister_netdev(ndev);
 918 +err_wq:
 919 +       destroy_workqueue(in->state_wq);
 920 +err_free:
 921 +       free_netdev(ndev);
 922 +
 923 +       return err;
 924 +}
 925 +
 926 +static void ivshm_net_remove(struct pci_dev *pdev)
 927 +{
 928 +       struct net_device *ndev = pci_get_drvdata(pdev);
 929 +       struct ivshm_net *in = netdev_priv(ndev);
 930 +
 931 +       if (in->using_msix)  {
 932 +               free_irq(in->msix.vector, ndev);
 933 +               pci_disable_msix(pdev);
 934 +       } else {
 935 +               free_irq(pdev->irq, ndev);
 936 +       }
 937 +
 938 +       unregister_netdev(ndev);
 939 +       cancel_work_sync(&in->state_work);
 940 +       destroy_workqueue(in->state_wq);
 941 +       free_netdev(ndev);
 942 +}
 943 +
 944 +static const struct pci_device_id ivshm_net_id_table[] = {
 945 +       { PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET, 0x1110),
 946 +               (PCI_CLASS_OTHERS << 16) | (0x01 << 8), 0xffff00 },
 947 +       { 0 }
 948 +};
 949 +MODULE_DEVICE_TABLE(pci, ivshm_net_id_table);
 950 +
 951 +static struct pci_driver ivshm_net_driver = {
 952 +       .name           = DRV_NAME,
 953 +       .id_table       = ivshm_net_id_table,
 954 +       .probe          = ivshm_net_probe,
 955 +       .remove         = ivshm_net_remove,
 956 +};
 957 +module_pci_driver(ivshm_net_driver);
 958 +
 959 +MODULE_AUTHOR("Mans Rullgard <mans@mansr.com>");
 960 +MODULE_LICENSE("GPL");