1/*-
2 * Copyright (c) 2011 NetApp, Inc.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_net.c 268953 2014-07-21 19:08:02Z jhb $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_net.c 268953 2014-07-21 19:08:02Z jhb $");
31
32#include <sys/param.h>
33#include <sys/linker_set.h>
34#include <sys/select.h>
35#include <sys/uio.h>
36#include <sys/ioctl.h>
37#include <net/ethernet.h>
38
39#include <errno.h>
40#include <fcntl.h>
41#include <stdio.h>
42#include <stdlib.h>
43#include <stdint.h>
44#include <string.h>
45#include <strings.h>
46#include <unistd.h>
47#include <assert.h>
48#include <md5.h>
49#include <pthread.h>
50#include <pthread_np.h>
51
52#include "bhyverun.h"
53#include "pci_emul.h"
54#include "mevent.h"
55#include "virtio.h"
56
57#define VTNET_RINGSZ 1024
58
59#define VTNET_MAXSEGS 32
60
61/*
62 * Host capabilities. Note that we only offer a few of these.
63 */
64#define VIRTIO_NET_F_CSUM (1 << 0) /* host handles partial cksum */
65#define VIRTIO_NET_F_GUEST_CSUM (1 << 1) /* guest handles partial cksum */
66#define VIRTIO_NET_F_MAC (1 << 5) /* host supplies MAC */
67#define VIRTIO_NET_F_GSO_DEPREC (1 << 6) /* deprecated: host handles GSO */
68#define VIRTIO_NET_F_GUEST_TSO4 (1 << 7) /* guest can rcv TSOv4 */
69#define VIRTIO_NET_F_GUEST_TSO6 (1 << 8) /* guest can rcv TSOv6 */
70#define VIRTIO_NET_F_GUEST_ECN (1 << 9) /* guest can rcv TSO with ECN */
71#define VIRTIO_NET_F_GUEST_UFO (1 << 10) /* guest can rcv UFO */
72#define VIRTIO_NET_F_HOST_TSO4 (1 << 11) /* host can rcv TSOv4 */
73#define VIRTIO_NET_F_HOST_TSO6 (1 << 12) /* host can rcv TSOv6 */
74#define VIRTIO_NET_F_HOST_ECN (1 << 13) /* host can rcv TSO with ECN */
75#define VIRTIO_NET_F_HOST_UFO (1 << 14) /* host can rcv UFO */
76#define VIRTIO_NET_F_MRG_RXBUF (1 << 15) /* host can merge RX buffers */
77#define VIRTIO_NET_F_STATUS (1 << 16) /* config status field available */
78#define VIRTIO_NET_F_CTRL_VQ (1 << 17) /* control channel available */
79#define VIRTIO_NET_F_CTRL_RX (1 << 18) /* control channel RX mode support */
80#define VIRTIO_NET_F_CTRL_VLAN (1 << 19) /* control channel VLAN filtering */
81#define VIRTIO_NET_F_GUEST_ANNOUNCE \
82 (1 << 21) /* guest can send gratuitous pkts */
83
84#define VTNET_S_HOSTCAPS \
85 ( VIRTIO_NET_F_MAC | VIRTIO_NET_F_MRG_RXBUF | VIRTIO_NET_F_STATUS | \
86 VIRTIO_F_NOTIFY_ON_EMPTY)
87
88/*
89 * PCI config-space "registers"
90 */
91struct virtio_net_config {
92 uint8_t mac[6];
93 uint16_t status;
94} __packed;
95
96/*
97 * Queue definitions.
98 */
99#define VTNET_RXQ 0
100#define VTNET_TXQ 1
101#define VTNET_CTLQ 2 /* NB: not yet supported */
102
103#define VTNET_MAXQ 3
104
105/*
106 * Fixed network header size
107 */
108struct virtio_net_rxhdr {
109 uint8_t vrh_flags;
110 uint8_t vrh_gso_type;
111 uint16_t vrh_hdr_len;
112 uint16_t vrh_gso_size;
113 uint16_t vrh_csum_start;
114 uint16_t vrh_csum_offset;
115 uint16_t vrh_bufs;
116} __packed;
117
118/*
119 * Debug printf
120 */
121static int pci_vtnet_debug;
122#define DPRINTF(params) if (pci_vtnet_debug) printf params
123#define WPRINTF(params) printf params
124
125/*
126 * Per-device softc
127 */
128struct pci_vtnet_softc {
129 struct virtio_softc vsc_vs;
130 struct vqueue_info vsc_queues[VTNET_MAXQ - 1];
131 pthread_mutex_t vsc_mtx;
132 struct mevent *vsc_mevp;
133
134 int vsc_tapfd;
135 int vsc_rx_ready;
136 volatile int resetting; /* set and checked outside lock */
137
138 uint32_t vsc_features;
139 struct virtio_net_config vsc_config;
140
141 pthread_mutex_t rx_mtx;
142 int rx_in_progress;
143
144 pthread_t tx_tid;
145 pthread_mutex_t tx_mtx;
146 pthread_cond_t tx_cond;
147 int tx_in_progress;
148};
149
150static void pci_vtnet_reset(void *);
151/* static void pci_vtnet_notify(void *, struct vqueue_info *); */
152static int pci_vtnet_cfgread(void *, int, int, uint32_t *);
153static int pci_vtnet_cfgwrite(void *, int, int, uint32_t);
154
155static struct virtio_consts vtnet_vi_consts = {
156 "vtnet", /* our name */
157 VTNET_MAXQ - 1, /* we currently support 2 virtqueues */
158 sizeof(struct virtio_net_config), /* config reg size */
159 pci_vtnet_reset, /* reset */
160 NULL, /* device-wide qnotify -- not used */
161 pci_vtnet_cfgread, /* read PCI config */
162 pci_vtnet_cfgwrite, /* write PCI config */
163 VTNET_S_HOSTCAPS, /* our capabilities */
164};
165
166/*
167 * If the transmit thread is active then stall until it is done.
168 */
169static void
170pci_vtnet_txwait(struct pci_vtnet_softc *sc)
171{
172
173 pthread_mutex_lock(&sc->tx_mtx);
174 while (sc->tx_in_progress) {
175 pthread_mutex_unlock(&sc->tx_mtx);
176 usleep(10000);
177 pthread_mutex_lock(&sc->tx_mtx);
178 }
179 pthread_mutex_unlock(&sc->tx_mtx);
180}
181
182/*
183 * If the receive thread is active then stall until it is done.
184 */
185static void
186pci_vtnet_rxwait(struct pci_vtnet_softc *sc)
187{
188
189 pthread_mutex_lock(&sc->rx_mtx);
190 while (sc->rx_in_progress) {
191 pthread_mutex_unlock(&sc->rx_mtx);
192 usleep(10000);
193 pthread_mutex_lock(&sc->rx_mtx);
194 }
195 pthread_mutex_unlock(&sc->rx_mtx);
196}
197
198static void
199pci_vtnet_reset(void *vsc)
200{
201 struct pci_vtnet_softc *sc = vsc;
202
203 DPRINTF(("vtnet: device reset requested !\n"));
204
205 sc->resetting = 1;
206
207 /*
208 * Wait for the transmit and receive threads to finish their
209 * processing.
210 */
211 pci_vtnet_txwait(sc);
212 pci_vtnet_rxwait(sc);
213
214 sc->vsc_rx_ready = 0;
215
216 /* now reset rings, MSI-X vectors, and negotiated capabilities */
217 vi_reset_dev(&sc->vsc_vs);
218
219 sc->resetting = 0;
220}
221
222/*
223 * Called to send a buffer chain out to the tap device
224 */
225static void
226pci_vtnet_tap_tx(struct pci_vtnet_softc *sc, struct iovec *iov, int iovcnt,
227 int len)
228{
229 static char pad[60]; /* all zero bytes */
230
231 if (sc->vsc_tapfd == -1)
232 return;
233
234 /*
235 * If the length is < 60, pad out to that and add the
236 * extra zero'd segment to the iov. It is guaranteed that
237 * there is always an extra iov available by the caller.
238 */
239 if (len < 60) {
240 iov[iovcnt].iov_base = pad;
241 iov[iovcnt].iov_len = 60 - len;
242 iovcnt++;
243 }
244 (void) writev(sc->vsc_tapfd, iov, iovcnt);
245}
246
247/*
248 * Called when there is read activity on the tap file descriptor.
249 * Each buffer posted by the guest is assumed to be able to contain
250 * an entire ethernet frame + rx header.
251 * MP note: the dummybuf is only used for discarding frames, so there
252 * is no need for it to be per-vtnet or locked.
253 */
254static uint8_t dummybuf[2048];
255
256static void
257pci_vtnet_tap_rx(struct pci_vtnet_softc *sc)
258{
259 struct vqueue_info *vq;
260 struct virtio_net_rxhdr *vrx;
261 uint8_t *buf;
262 int len;
263 struct iovec iov;
264
265 /*
266 * Should never be called without a valid tap fd
267 */
268 assert(sc->vsc_tapfd != -1);
269
270 /*
271 * But, will be called when the rx ring hasn't yet
272 * been set up or the guest is resetting the device.
273 */
274 if (!sc->vsc_rx_ready || sc->resetting) {
275 /*
276 * Drop the packet and try later.
277 */
278 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
279 return;
280 }
281
282 /*
283 * Check for available rx buffers
284 */
285 vq = &sc->vsc_queues[VTNET_RXQ];
286 vq_startchains(vq);
287 if (!vq_has_descs(vq)) {
288 /*
289 * Drop the packet and try later. Interrupt on
290 * empty, if that's negotiated.
291 */
292 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
293 vq_endchains(vq, 1);
294 return;
295 }
296
297 do {
298 /*
299 * Get descriptor chain, which should have just
300 * one descriptor in it.
301 * ??? allow guests to use multiple descs?
302 */
303 assert(vq_getchain(vq, &iov, 1, NULL) == 1);
304
305 /*
306 * Get a pointer to the rx header, and use the
307 * data immediately following it for the packet buffer.
308 */
309 vrx = iov.iov_base;
310 buf = (uint8_t *)(vrx + 1);
311
312 len = read(sc->vsc_tapfd, buf,
313 iov.iov_len - sizeof(struct virtio_net_rxhdr));
314
315 if (len < 0 && errno == EWOULDBLOCK) {
316 /*
317 * No more packets, but still some avail ring
318 * entries. Interrupt if needed/appropriate.
319 */
320 vq_endchains(vq, 0);
321 return;
322 }
323
324 /*
325 * The only valid field in the rx packet header is the
326 * number of buffers, which is always 1 without TSO
327 * support.
328 */
329 memset(vrx, 0, sizeof(struct virtio_net_rxhdr));
330 vrx->vrh_bufs = 1;
331
332 /*
333 * Release this chain and handle more chains.
334 */
335 vq_relchain(vq, len + sizeof(struct virtio_net_rxhdr));
336 } while (vq_has_descs(vq));
337
338 /* Interrupt if needed, including for NOTIFY_ON_EMPTY. */
339 vq_endchains(vq, 1);
340}
341
342static void
343pci_vtnet_tap_callback(int fd, enum ev_type type, void *param)
344{
345 struct pci_vtnet_softc *sc = param;
346
347 pthread_mutex_lock(&sc->rx_mtx);
348 sc->rx_in_progress = 1;
349 pci_vtnet_tap_rx(sc);
350 sc->rx_in_progress = 0;
351 pthread_mutex_unlock(&sc->rx_mtx);
352
353}
354
355static void
356pci_vtnet_ping_rxq(void *vsc, struct vqueue_info *vq)
357{
358 struct pci_vtnet_softc *sc = vsc;
359
360 /*
361 * A qnotify means that the rx process can now begin
362 */
363 if (sc->vsc_rx_ready == 0) {
364 sc->vsc_rx_ready = 1;
365 }
366}
367
368static void
369pci_vtnet_proctx(struct pci_vtnet_softc *sc, struct vqueue_info *vq)
370{
371 struct iovec iov[VTNET_MAXSEGS + 1];
372 int i, n;
373 int plen, tlen;
374
375 /*
376 * Obtain chain of descriptors. The first one is
377 * really the header descriptor, so we need to sum
378 * up two lengths: packet length and transfer length.
379 */
380 n = vq_getchain(vq, iov, VTNET_MAXSEGS, NULL);
381 assert(n >= 1 && n <= VTNET_MAXSEGS);
382 plen = 0;
383 tlen = iov[0].iov_len;
384 for (i = 1; i < n; i++) {
385 plen += iov[i].iov_len;
386 tlen += iov[i].iov_len;
387 }
388
389 DPRINTF(("virtio: packet send, %d bytes, %d segs\n\r", plen, n));
390 pci_vtnet_tap_tx(sc, &iov[1], n - 1, plen);
391
392 /* chain is processed, release it and set tlen */
393 vq_relchain(vq, tlen);
394}
395
396static void
397pci_vtnet_ping_txq(void *vsc, struct vqueue_info *vq)
398{
399 struct pci_vtnet_softc *sc = vsc;
400
401 /*
402 * Any ring entries to process?
403 */
404 if (!vq_has_descs(vq))
405 return;
406
407 /* Signal the tx thread for processing */
408 pthread_mutex_lock(&sc->tx_mtx);
409 if (sc->tx_in_progress == 0)
410 pthread_cond_signal(&sc->tx_cond);
411 pthread_mutex_unlock(&sc->tx_mtx);
412}
413
414/*
415 * Thread which will handle processing of TX desc
416 */
417static void *
418pci_vtnet_tx_thread(void *param)
419{
420 struct pci_vtnet_softc *sc = param;
421 struct vqueue_info *vq;
422 int have_work, error;
423
424 vq = &sc->vsc_queues[VTNET_TXQ];
425
426 /*
427 * Let us wait till the tx queue pointers get initialised &
428 * first tx signaled
429 */
430 pthread_mutex_lock(&sc->tx_mtx);
431 error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
432 assert(error == 0);
433
434 for (;;) {
435 /* note - tx mutex is locked here */
436 do {
437 if (sc->resetting)
438 have_work = 0;
439 else
440 have_work = vq_has_descs(vq);
441
442 if (!have_work) {
443 sc->tx_in_progress = 0;
444 error = pthread_cond_wait(&sc->tx_cond,
445 &sc->tx_mtx);
446 assert(error == 0);
447 }
448 } while (!have_work);
449 sc->tx_in_progress = 1;
450 pthread_mutex_unlock(&sc->tx_mtx);
451
452 vq_startchains(vq);
453 do {
454 /*
455 * Run through entries, placing them into
456 * iovecs and sending when an end-of-packet
457 * is found
458 */
459 pci_vtnet_proctx(sc, vq);
460 } while (vq_has_descs(vq));
461
462 /*
463 * Generate an interrupt if needed.
464 */
465 vq_endchains(vq, 1);
466
467 pthread_mutex_lock(&sc->tx_mtx);
468 }
469}
470
471#ifdef notyet
472static void
473pci_vtnet_ping_ctlq(void *vsc, struct vqueue_info *vq)
474{
475
476 DPRINTF(("vtnet: control qnotify!\n\r"));
477}
478#endif
479
480static int
481pci_vtnet_parsemac(char *mac_str, uint8_t *mac_addr)
482{
483 struct ether_addr *ea;
484 char *tmpstr;
485 char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
486
487 tmpstr = strsep(&mac_str,"=");
488
489 if ((mac_str != NULL) && (!strcmp(tmpstr,"mac"))) {
490 ea = ether_aton(mac_str);
491
492 if (ea == NULL || ETHER_IS_MULTICAST(ea->octet) ||
493 memcmp(ea->octet, zero_addr, ETHER_ADDR_LEN) == 0) {
494 fprintf(stderr, "Invalid MAC %s\n", mac_str);
495 return (EINVAL);
496 } else
497 memcpy(mac_addr, ea->octet, ETHER_ADDR_LEN);
498 }
499
500 return (0);
501}
502
503
504static int
505pci_vtnet_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
506{
507 MD5_CTX mdctx;
508 unsigned char digest[16];
509 char nstr[80];
510 char tname[MAXCOMLEN + 1];
511 struct pci_vtnet_softc *sc;
512 char *devname;
513 char *vtopts;
514 int mac_provided;
515
516 sc = calloc(1, sizeof(struct pci_vtnet_softc));
517
518 pthread_mutex_init(&sc->vsc_mtx, NULL);
519
520 vi_softc_linkup(&sc->vsc_vs, &vtnet_vi_consts, sc, pi, sc->vsc_queues);
521 sc->vsc_vs.vs_mtx = &sc->vsc_mtx;
522
523 sc->vsc_queues[VTNET_RXQ].vq_qsize = VTNET_RINGSZ;
524 sc->vsc_queues[VTNET_RXQ].vq_notify = pci_vtnet_ping_rxq;
525 sc->vsc_queues[VTNET_TXQ].vq_qsize = VTNET_RINGSZ;
526 sc->vsc_queues[VTNET_TXQ].vq_notify = pci_vtnet_ping_txq;
527#ifdef notyet
528 sc->vsc_queues[VTNET_CTLQ].vq_qsize = VTNET_RINGSZ;
529 sc->vsc_queues[VTNET_CTLQ].vq_notify = pci_vtnet_ping_ctlq;
530#endif
531
532 /*
533 * Attempt to open the tap device and read the MAC address
534 * if specified
535 */
536 mac_provided = 0;
537 sc->vsc_tapfd = -1;
538 if (opts != NULL) {
539 char tbuf[80];
540 int err;
541
542 devname = vtopts = strdup(opts);
543 (void) strsep(&vtopts, ",");
544
545 if (vtopts != NULL) {
546 err = pci_vtnet_parsemac(vtopts, sc->vsc_config.mac);
547 if (err != 0) {
548 free(devname);
549 return (err);
550 }
551 mac_provided = 1;
552 }
553
554 strcpy(tbuf, "/dev/");
555 strlcat(tbuf, devname, sizeof(tbuf));
556
557 free(devname);
558
559 sc->vsc_tapfd = open(tbuf, O_RDWR);
560 if (sc->vsc_tapfd == -1) {
561 WPRINTF(("open of tap device %s failed\n", tbuf));
562 } else {
563 /*
564 * Set non-blocking and register for read
565 * notifications with the event loop
566 */
567 int opt = 1;
568 if (ioctl(sc->vsc_tapfd, FIONBIO, &opt) < 0) {
569 WPRINTF(("tap device O_NONBLOCK failed\n"));
570 close(sc->vsc_tapfd);
571 sc->vsc_tapfd = -1;
572 }
573
574 sc->vsc_mevp = mevent_add(sc->vsc_tapfd,
575 EVF_READ,
576 pci_vtnet_tap_callback,
577 sc);
578 if (sc->vsc_mevp == NULL) {
579 WPRINTF(("Could not register event\n"));
580 close(sc->vsc_tapfd);
581 sc->vsc_tapfd = -1;
582 }
583 }
584 }
585
586 /*
587 * The default MAC address is the standard NetApp OUI of 00-a0-98,
588 * followed by an MD5 of the PCI slot/func number and dev name
589 */
590 if (!mac_provided) {
591 snprintf(nstr, sizeof(nstr), "%d-%d-%s", pi->pi_slot,
592 pi->pi_func, vmname);
593
594 MD5Init(&mdctx);
595 MD5Update(&mdctx, nstr, strlen(nstr));
596 MD5Final(digest, &mdctx);
597
598 sc->vsc_config.mac[0] = 0x00;
599 sc->vsc_config.mac[1] = 0xa0;
600 sc->vsc_config.mac[2] = 0x98;
601 sc->vsc_config.mac[3] = digest[0];
602 sc->vsc_config.mac[4] = digest[1];
603 sc->vsc_config.mac[5] = digest[2];
604 }
605
606 /* initialize config space */
607 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_NET);
608 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
609 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_NETWORK);
610 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_NET);
611
612 pci_lintr_request(pi);
613
614 /* link always up */
615 sc->vsc_config.status = 1;
616
617 /* use BAR 1 to map MSI-X table and PBA, if we're using MSI-X */
618 if (vi_intr_init(&sc->vsc_vs, 1, fbsdrun_virtio_msix()))
619 return (1);
620
621 /* use BAR 0 to map config regs in IO space */
622 vi_set_io_bar(&sc->vsc_vs, 0);
623
624 sc->resetting = 0;
625
626 sc->rx_in_progress = 0;
627 pthread_mutex_init(&sc->rx_mtx, NULL);
628
629 /*
630 * Initialize tx semaphore & spawn TX processing thread.
631 * As of now, only one thread for TX desc processing is
632 * spawned.
633 */
634 sc->tx_in_progress = 0;
635 pthread_mutex_init(&sc->tx_mtx, NULL);
636 pthread_cond_init(&sc->tx_cond, NULL);
637 pthread_create(&sc->tx_tid, NULL, pci_vtnet_tx_thread, (void *)sc);
638 snprintf(tname, sizeof(tname), "vtnet-%d:%d tx", pi->pi_slot,
639 pi->pi_func);
640 pthread_set_name_np(sc->tx_tid, tname);
641
642 return (0);
643}
644
645static int
646pci_vtnet_cfgwrite(void *vsc, int offset, int size, uint32_t value)
647{
648 struct pci_vtnet_softc *sc = vsc;
649 void *ptr;
650
651 if (offset < 6) {
652 assert(offset + size <= 6);
653 /*
654 * The driver is allowed to change the MAC address
655 */
656 ptr = &sc->vsc_config.mac[offset];
657 memcpy(ptr, &value, size);
658 } else {
659 DPRINTF(("vtnet: write to readonly reg %d\n\r", offset));
660 return (1);
661 }
662 return (0);
663}
664
665static int
666pci_vtnet_cfgread(void *vsc, int offset, int size, uint32_t *retval)
667{
668 struct pci_vtnet_softc *sc = vsc;
669 void *ptr;
670
671 ptr = (uint8_t *)&sc->vsc_config + offset;
672 memcpy(retval, ptr, size);
673 return (0);
674}
675
676struct pci_devemu pci_de_vnet = {
677 .pe_emu = "virtio-net",
678 .pe_init = pci_vtnet_init,
679 .pe_barwrite = vi_pci_write,
680 .pe_barread = vi_pci_read
681};
682PCI_EMUL_SET(pci_de_vnet);
2 * Copyright (c) 2011 NetApp, Inc.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_net.c 268953 2014-07-21 19:08:02Z jhb $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/pci_virtio_net.c 268953 2014-07-21 19:08:02Z jhb $");
31
32#include <sys/param.h>
33#include <sys/linker_set.h>
34#include <sys/select.h>
35#include <sys/uio.h>
36#include <sys/ioctl.h>
37#include <net/ethernet.h>
38
39#include <errno.h>
40#include <fcntl.h>
41#include <stdio.h>
42#include <stdlib.h>
43#include <stdint.h>
44#include <string.h>
45#include <strings.h>
46#include <unistd.h>
47#include <assert.h>
48#include <md5.h>
49#include <pthread.h>
50#include <pthread_np.h>
51
52#include "bhyverun.h"
53#include "pci_emul.h"
54#include "mevent.h"
55#include "virtio.h"
56
57#define VTNET_RINGSZ 1024
58
59#define VTNET_MAXSEGS 32
60
61/*
62 * Host capabilities. Note that we only offer a few of these.
63 */
64#define VIRTIO_NET_F_CSUM (1 << 0) /* host handles partial cksum */
65#define VIRTIO_NET_F_GUEST_CSUM (1 << 1) /* guest handles partial cksum */
66#define VIRTIO_NET_F_MAC (1 << 5) /* host supplies MAC */
67#define VIRTIO_NET_F_GSO_DEPREC (1 << 6) /* deprecated: host handles GSO */
68#define VIRTIO_NET_F_GUEST_TSO4 (1 << 7) /* guest can rcv TSOv4 */
69#define VIRTIO_NET_F_GUEST_TSO6 (1 << 8) /* guest can rcv TSOv6 */
70#define VIRTIO_NET_F_GUEST_ECN (1 << 9) /* guest can rcv TSO with ECN */
71#define VIRTIO_NET_F_GUEST_UFO (1 << 10) /* guest can rcv UFO */
72#define VIRTIO_NET_F_HOST_TSO4 (1 << 11) /* host can rcv TSOv4 */
73#define VIRTIO_NET_F_HOST_TSO6 (1 << 12) /* host can rcv TSOv6 */
74#define VIRTIO_NET_F_HOST_ECN (1 << 13) /* host can rcv TSO with ECN */
75#define VIRTIO_NET_F_HOST_UFO (1 << 14) /* host can rcv UFO */
76#define VIRTIO_NET_F_MRG_RXBUF (1 << 15) /* host can merge RX buffers */
77#define VIRTIO_NET_F_STATUS (1 << 16) /* config status field available */
78#define VIRTIO_NET_F_CTRL_VQ (1 << 17) /* control channel available */
79#define VIRTIO_NET_F_CTRL_RX (1 << 18) /* control channel RX mode support */
80#define VIRTIO_NET_F_CTRL_VLAN (1 << 19) /* control channel VLAN filtering */
81#define VIRTIO_NET_F_GUEST_ANNOUNCE \
82 (1 << 21) /* guest can send gratuitous pkts */
83
84#define VTNET_S_HOSTCAPS \
85 ( VIRTIO_NET_F_MAC | VIRTIO_NET_F_MRG_RXBUF | VIRTIO_NET_F_STATUS | \
86 VIRTIO_F_NOTIFY_ON_EMPTY)
87
88/*
89 * PCI config-space "registers"
90 */
91struct virtio_net_config {
92 uint8_t mac[6];
93 uint16_t status;
94} __packed;
95
96/*
97 * Queue definitions.
98 */
99#define VTNET_RXQ 0
100#define VTNET_TXQ 1
101#define VTNET_CTLQ 2 /* NB: not yet supported */
102
103#define VTNET_MAXQ 3
104
105/*
106 * Fixed network header size
107 */
108struct virtio_net_rxhdr {
109 uint8_t vrh_flags;
110 uint8_t vrh_gso_type;
111 uint16_t vrh_hdr_len;
112 uint16_t vrh_gso_size;
113 uint16_t vrh_csum_start;
114 uint16_t vrh_csum_offset;
115 uint16_t vrh_bufs;
116} __packed;
117
118/*
119 * Debug printf
120 */
121static int pci_vtnet_debug;
122#define DPRINTF(params) if (pci_vtnet_debug) printf params
123#define WPRINTF(params) printf params
124
125/*
126 * Per-device softc
127 */
128struct pci_vtnet_softc {
129 struct virtio_softc vsc_vs;
130 struct vqueue_info vsc_queues[VTNET_MAXQ - 1];
131 pthread_mutex_t vsc_mtx;
132 struct mevent *vsc_mevp;
133
134 int vsc_tapfd;
135 int vsc_rx_ready;
136 volatile int resetting; /* set and checked outside lock */
137
138 uint32_t vsc_features;
139 struct virtio_net_config vsc_config;
140
141 pthread_mutex_t rx_mtx;
142 int rx_in_progress;
143
144 pthread_t tx_tid;
145 pthread_mutex_t tx_mtx;
146 pthread_cond_t tx_cond;
147 int tx_in_progress;
148};
149
150static void pci_vtnet_reset(void *);
151/* static void pci_vtnet_notify(void *, struct vqueue_info *); */
152static int pci_vtnet_cfgread(void *, int, int, uint32_t *);
153static int pci_vtnet_cfgwrite(void *, int, int, uint32_t);
154
155static struct virtio_consts vtnet_vi_consts = {
156 "vtnet", /* our name */
157 VTNET_MAXQ - 1, /* we currently support 2 virtqueues */
158 sizeof(struct virtio_net_config), /* config reg size */
159 pci_vtnet_reset, /* reset */
160 NULL, /* device-wide qnotify -- not used */
161 pci_vtnet_cfgread, /* read PCI config */
162 pci_vtnet_cfgwrite, /* write PCI config */
163 VTNET_S_HOSTCAPS, /* our capabilities */
164};
165
166/*
167 * If the transmit thread is active then stall until it is done.
168 */
169static void
170pci_vtnet_txwait(struct pci_vtnet_softc *sc)
171{
172
173 pthread_mutex_lock(&sc->tx_mtx);
174 while (sc->tx_in_progress) {
175 pthread_mutex_unlock(&sc->tx_mtx);
176 usleep(10000);
177 pthread_mutex_lock(&sc->tx_mtx);
178 }
179 pthread_mutex_unlock(&sc->tx_mtx);
180}
181
182/*
183 * If the receive thread is active then stall until it is done.
184 */
185static void
186pci_vtnet_rxwait(struct pci_vtnet_softc *sc)
187{
188
189 pthread_mutex_lock(&sc->rx_mtx);
190 while (sc->rx_in_progress) {
191 pthread_mutex_unlock(&sc->rx_mtx);
192 usleep(10000);
193 pthread_mutex_lock(&sc->rx_mtx);
194 }
195 pthread_mutex_unlock(&sc->rx_mtx);
196}
197
198static void
199pci_vtnet_reset(void *vsc)
200{
201 struct pci_vtnet_softc *sc = vsc;
202
203 DPRINTF(("vtnet: device reset requested !\n"));
204
205 sc->resetting = 1;
206
207 /*
208 * Wait for the transmit and receive threads to finish their
209 * processing.
210 */
211 pci_vtnet_txwait(sc);
212 pci_vtnet_rxwait(sc);
213
214 sc->vsc_rx_ready = 0;
215
216 /* now reset rings, MSI-X vectors, and negotiated capabilities */
217 vi_reset_dev(&sc->vsc_vs);
218
219 sc->resetting = 0;
220}
221
222/*
223 * Called to send a buffer chain out to the tap device
224 */
225static void
226pci_vtnet_tap_tx(struct pci_vtnet_softc *sc, struct iovec *iov, int iovcnt,
227 int len)
228{
229 static char pad[60]; /* all zero bytes */
230
231 if (sc->vsc_tapfd == -1)
232 return;
233
234 /*
235 * If the length is < 60, pad out to that and add the
236 * extra zero'd segment to the iov. It is guaranteed that
237 * there is always an extra iov available by the caller.
238 */
239 if (len < 60) {
240 iov[iovcnt].iov_base = pad;
241 iov[iovcnt].iov_len = 60 - len;
242 iovcnt++;
243 }
244 (void) writev(sc->vsc_tapfd, iov, iovcnt);
245}
246
247/*
248 * Called when there is read activity on the tap file descriptor.
249 * Each buffer posted by the guest is assumed to be able to contain
250 * an entire ethernet frame + rx header.
251 * MP note: the dummybuf is only used for discarding frames, so there
252 * is no need for it to be per-vtnet or locked.
253 */
254static uint8_t dummybuf[2048];
255
256static void
257pci_vtnet_tap_rx(struct pci_vtnet_softc *sc)
258{
259 struct vqueue_info *vq;
260 struct virtio_net_rxhdr *vrx;
261 uint8_t *buf;
262 int len;
263 struct iovec iov;
264
265 /*
266 * Should never be called without a valid tap fd
267 */
268 assert(sc->vsc_tapfd != -1);
269
270 /*
271 * But, will be called when the rx ring hasn't yet
272 * been set up or the guest is resetting the device.
273 */
274 if (!sc->vsc_rx_ready || sc->resetting) {
275 /*
276 * Drop the packet and try later.
277 */
278 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
279 return;
280 }
281
282 /*
283 * Check for available rx buffers
284 */
285 vq = &sc->vsc_queues[VTNET_RXQ];
286 vq_startchains(vq);
287 if (!vq_has_descs(vq)) {
288 /*
289 * Drop the packet and try later. Interrupt on
290 * empty, if that's negotiated.
291 */
292 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
293 vq_endchains(vq, 1);
294 return;
295 }
296
297 do {
298 /*
299 * Get descriptor chain, which should have just
300 * one descriptor in it.
301 * ??? allow guests to use multiple descs?
302 */
303 assert(vq_getchain(vq, &iov, 1, NULL) == 1);
304
305 /*
306 * Get a pointer to the rx header, and use the
307 * data immediately following it for the packet buffer.
308 */
309 vrx = iov.iov_base;
310 buf = (uint8_t *)(vrx + 1);
311
312 len = read(sc->vsc_tapfd, buf,
313 iov.iov_len - sizeof(struct virtio_net_rxhdr));
314
315 if (len < 0 && errno == EWOULDBLOCK) {
316 /*
317 * No more packets, but still some avail ring
318 * entries. Interrupt if needed/appropriate.
319 */
320 vq_endchains(vq, 0);
321 return;
322 }
323
324 /*
325 * The only valid field in the rx packet header is the
326 * number of buffers, which is always 1 without TSO
327 * support.
328 */
329 memset(vrx, 0, sizeof(struct virtio_net_rxhdr));
330 vrx->vrh_bufs = 1;
331
332 /*
333 * Release this chain and handle more chains.
334 */
335 vq_relchain(vq, len + sizeof(struct virtio_net_rxhdr));
336 } while (vq_has_descs(vq));
337
338 /* Interrupt if needed, including for NOTIFY_ON_EMPTY. */
339 vq_endchains(vq, 1);
340}
341
342static void
343pci_vtnet_tap_callback(int fd, enum ev_type type, void *param)
344{
345 struct pci_vtnet_softc *sc = param;
346
347 pthread_mutex_lock(&sc->rx_mtx);
348 sc->rx_in_progress = 1;
349 pci_vtnet_tap_rx(sc);
350 sc->rx_in_progress = 0;
351 pthread_mutex_unlock(&sc->rx_mtx);
352
353}
354
355static void
356pci_vtnet_ping_rxq(void *vsc, struct vqueue_info *vq)
357{
358 struct pci_vtnet_softc *sc = vsc;
359
360 /*
361 * A qnotify means that the rx process can now begin
362 */
363 if (sc->vsc_rx_ready == 0) {
364 sc->vsc_rx_ready = 1;
365 }
366}
367
368static void
369pci_vtnet_proctx(struct pci_vtnet_softc *sc, struct vqueue_info *vq)
370{
371 struct iovec iov[VTNET_MAXSEGS + 1];
372 int i, n;
373 int plen, tlen;
374
375 /*
376 * Obtain chain of descriptors. The first one is
377 * really the header descriptor, so we need to sum
378 * up two lengths: packet length and transfer length.
379 */
380 n = vq_getchain(vq, iov, VTNET_MAXSEGS, NULL);
381 assert(n >= 1 && n <= VTNET_MAXSEGS);
382 plen = 0;
383 tlen = iov[0].iov_len;
384 for (i = 1; i < n; i++) {
385 plen += iov[i].iov_len;
386 tlen += iov[i].iov_len;
387 }
388
389 DPRINTF(("virtio: packet send, %d bytes, %d segs\n\r", plen, n));
390 pci_vtnet_tap_tx(sc, &iov[1], n - 1, plen);
391
392 /* chain is processed, release it and set tlen */
393 vq_relchain(vq, tlen);
394}
395
396static void
397pci_vtnet_ping_txq(void *vsc, struct vqueue_info *vq)
398{
399 struct pci_vtnet_softc *sc = vsc;
400
401 /*
402 * Any ring entries to process?
403 */
404 if (!vq_has_descs(vq))
405 return;
406
407 /* Signal the tx thread for processing */
408 pthread_mutex_lock(&sc->tx_mtx);
409 if (sc->tx_in_progress == 0)
410 pthread_cond_signal(&sc->tx_cond);
411 pthread_mutex_unlock(&sc->tx_mtx);
412}
413
414/*
415 * Thread which will handle processing of TX desc
416 */
417static void *
418pci_vtnet_tx_thread(void *param)
419{
420 struct pci_vtnet_softc *sc = param;
421 struct vqueue_info *vq;
422 int have_work, error;
423
424 vq = &sc->vsc_queues[VTNET_TXQ];
425
426 /*
427 * Let us wait till the tx queue pointers get initialised &
428 * first tx signaled
429 */
430 pthread_mutex_lock(&sc->tx_mtx);
431 error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
432 assert(error == 0);
433
434 for (;;) {
435 /* note - tx mutex is locked here */
436 do {
437 if (sc->resetting)
438 have_work = 0;
439 else
440 have_work = vq_has_descs(vq);
441
442 if (!have_work) {
443 sc->tx_in_progress = 0;
444 error = pthread_cond_wait(&sc->tx_cond,
445 &sc->tx_mtx);
446 assert(error == 0);
447 }
448 } while (!have_work);
449 sc->tx_in_progress = 1;
450 pthread_mutex_unlock(&sc->tx_mtx);
451
452 vq_startchains(vq);
453 do {
454 /*
455 * Run through entries, placing them into
456 * iovecs and sending when an end-of-packet
457 * is found
458 */
459 pci_vtnet_proctx(sc, vq);
460 } while (vq_has_descs(vq));
461
462 /*
463 * Generate an interrupt if needed.
464 */
465 vq_endchains(vq, 1);
466
467 pthread_mutex_lock(&sc->tx_mtx);
468 }
469}
470
471#ifdef notyet
472static void
473pci_vtnet_ping_ctlq(void *vsc, struct vqueue_info *vq)
474{
475
476 DPRINTF(("vtnet: control qnotify!\n\r"));
477}
478#endif
479
480static int
481pci_vtnet_parsemac(char *mac_str, uint8_t *mac_addr)
482{
483 struct ether_addr *ea;
484 char *tmpstr;
485 char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
486
487 tmpstr = strsep(&mac_str,"=");
488
489 if ((mac_str != NULL) && (!strcmp(tmpstr,"mac"))) {
490 ea = ether_aton(mac_str);
491
492 if (ea == NULL || ETHER_IS_MULTICAST(ea->octet) ||
493 memcmp(ea->octet, zero_addr, ETHER_ADDR_LEN) == 0) {
494 fprintf(stderr, "Invalid MAC %s\n", mac_str);
495 return (EINVAL);
496 } else
497 memcpy(mac_addr, ea->octet, ETHER_ADDR_LEN);
498 }
499
500 return (0);
501}
502
503
504static int
505pci_vtnet_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
506{
507 MD5_CTX mdctx;
508 unsigned char digest[16];
509 char nstr[80];
510 char tname[MAXCOMLEN + 1];
511 struct pci_vtnet_softc *sc;
512 char *devname;
513 char *vtopts;
514 int mac_provided;
515
516 sc = calloc(1, sizeof(struct pci_vtnet_softc));
517
518 pthread_mutex_init(&sc->vsc_mtx, NULL);
519
520 vi_softc_linkup(&sc->vsc_vs, &vtnet_vi_consts, sc, pi, sc->vsc_queues);
521 sc->vsc_vs.vs_mtx = &sc->vsc_mtx;
522
523 sc->vsc_queues[VTNET_RXQ].vq_qsize = VTNET_RINGSZ;
524 sc->vsc_queues[VTNET_RXQ].vq_notify = pci_vtnet_ping_rxq;
525 sc->vsc_queues[VTNET_TXQ].vq_qsize = VTNET_RINGSZ;
526 sc->vsc_queues[VTNET_TXQ].vq_notify = pci_vtnet_ping_txq;
527#ifdef notyet
528 sc->vsc_queues[VTNET_CTLQ].vq_qsize = VTNET_RINGSZ;
529 sc->vsc_queues[VTNET_CTLQ].vq_notify = pci_vtnet_ping_ctlq;
530#endif
531
532 /*
533 * Attempt to open the tap device and read the MAC address
534 * if specified
535 */
536 mac_provided = 0;
537 sc->vsc_tapfd = -1;
538 if (opts != NULL) {
539 char tbuf[80];
540 int err;
541
542 devname = vtopts = strdup(opts);
543 (void) strsep(&vtopts, ",");
544
545 if (vtopts != NULL) {
546 err = pci_vtnet_parsemac(vtopts, sc->vsc_config.mac);
547 if (err != 0) {
548 free(devname);
549 return (err);
550 }
551 mac_provided = 1;
552 }
553
554 strcpy(tbuf, "/dev/");
555 strlcat(tbuf, devname, sizeof(tbuf));
556
557 free(devname);
558
559 sc->vsc_tapfd = open(tbuf, O_RDWR);
560 if (sc->vsc_tapfd == -1) {
561 WPRINTF(("open of tap device %s failed\n", tbuf));
562 } else {
563 /*
564 * Set non-blocking and register for read
565 * notifications with the event loop
566 */
567 int opt = 1;
568 if (ioctl(sc->vsc_tapfd, FIONBIO, &opt) < 0) {
569 WPRINTF(("tap device O_NONBLOCK failed\n"));
570 close(sc->vsc_tapfd);
571 sc->vsc_tapfd = -1;
572 }
573
574 sc->vsc_mevp = mevent_add(sc->vsc_tapfd,
575 EVF_READ,
576 pci_vtnet_tap_callback,
577 sc);
578 if (sc->vsc_mevp == NULL) {
579 WPRINTF(("Could not register event\n"));
580 close(sc->vsc_tapfd);
581 sc->vsc_tapfd = -1;
582 }
583 }
584 }
585
586 /*
587 * The default MAC address is the standard NetApp OUI of 00-a0-98,
588 * followed by an MD5 of the PCI slot/func number and dev name
589 */
590 if (!mac_provided) {
591 snprintf(nstr, sizeof(nstr), "%d-%d-%s", pi->pi_slot,
592 pi->pi_func, vmname);
593
594 MD5Init(&mdctx);
595 MD5Update(&mdctx, nstr, strlen(nstr));
596 MD5Final(digest, &mdctx);
597
598 sc->vsc_config.mac[0] = 0x00;
599 sc->vsc_config.mac[1] = 0xa0;
600 sc->vsc_config.mac[2] = 0x98;
601 sc->vsc_config.mac[3] = digest[0];
602 sc->vsc_config.mac[4] = digest[1];
603 sc->vsc_config.mac[5] = digest[2];
604 }
605
606 /* initialize config space */
607 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_NET);
608 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
609 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_NETWORK);
610 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_NET);
611
612 pci_lintr_request(pi);
613
614 /* link always up */
615 sc->vsc_config.status = 1;
616
617 /* use BAR 1 to map MSI-X table and PBA, if we're using MSI-X */
618 if (vi_intr_init(&sc->vsc_vs, 1, fbsdrun_virtio_msix()))
619 return (1);
620
621 /* use BAR 0 to map config regs in IO space */
622 vi_set_io_bar(&sc->vsc_vs, 0);
623
624 sc->resetting = 0;
625
626 sc->rx_in_progress = 0;
627 pthread_mutex_init(&sc->rx_mtx, NULL);
628
629 /*
630 * Initialize tx semaphore & spawn TX processing thread.
631 * As of now, only one thread for TX desc processing is
632 * spawned.
633 */
634 sc->tx_in_progress = 0;
635 pthread_mutex_init(&sc->tx_mtx, NULL);
636 pthread_cond_init(&sc->tx_cond, NULL);
637 pthread_create(&sc->tx_tid, NULL, pci_vtnet_tx_thread, (void *)sc);
638 snprintf(tname, sizeof(tname), "vtnet-%d:%d tx", pi->pi_slot,
639 pi->pi_func);
640 pthread_set_name_np(sc->tx_tid, tname);
641
642 return (0);
643}
644
645static int
646pci_vtnet_cfgwrite(void *vsc, int offset, int size, uint32_t value)
647{
648 struct pci_vtnet_softc *sc = vsc;
649 void *ptr;
650
651 if (offset < 6) {
652 assert(offset + size <= 6);
653 /*
654 * The driver is allowed to change the MAC address
655 */
656 ptr = &sc->vsc_config.mac[offset];
657 memcpy(ptr, &value, size);
658 } else {
659 DPRINTF(("vtnet: write to readonly reg %d\n\r", offset));
660 return (1);
661 }
662 return (0);
663}
664
665static int
666pci_vtnet_cfgread(void *vsc, int offset, int size, uint32_t *retval)
667{
668 struct pci_vtnet_softc *sc = vsc;
669 void *ptr;
670
671 ptr = (uint8_t *)&sc->vsc_config + offset;
672 memcpy(retval, ptr, size);
673 return (0);
674}
675
676struct pci_devemu pci_de_vnet = {
677 .pe_emu = "virtio-net",
678 .pe_init = pci_vtnet_init,
679 .pe_barwrite = vi_pci_write,
680 .pe_barread = vi_pci_read
681};
682PCI_EMUL_SET(pci_de_vnet);