3 * This file is part of the Palacios Virtual Machine Monitor developed
4 * by the V3VEE Project with funding from the United States National
5 * Science Foundation and the Department of Energy.
7 * The V3VEE Project is a joint project between Northwestern University
8 * and the University of New Mexico. You can find out more at
11 * Copyright (c) 2010, Lei Xia <lxia@northwestern.edu>
12 * Copyright (c) 2010, Cui Zheng <cuizheng@cs.unm.edu>
13 * Copyright (c) 2010, The V3VEE Project <http://www.v3vee.org>
14 * All rights reserved.
16 * Author: Lei Xia <lxia@northwestern.edu>
17 * Cui Zheng <cuizheng@cs.unm.edu>
20 * This is free software. You are permitted to use,
21 * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
24 #include <palacios/vmm.h>
25 #include <palacios/vmm_dev_mgr.h>
26 #include <devices/lnx_virtio_pci.h>
27 #include <palacios/vm_guest_mem.h>
28 #include <palacios/vmm_sprintf.h>
29 #include <vnet/vnet.h>
30 #include <palacios/vmm_lock.h>
31 #include <palacios/vmm_util.h>
32 #include <devices/pci.h>
33 #include <palacios/vmm_ethernet.h>
34 #include <palacios/vmm_time.h>
37 #ifndef V3_CONFIG_DEBUG_VIRTIO_NET
39 #define PrintDebug(fmt, args...)
42 #ifndef V3_CONFIG_VNET
43 static int net_debug = 0;
46 #define TX_QUEUE_SIZE 4096
47 #define RX_QUEUE_SIZE 4096
48 #define CTRL_QUEUE_SIZE 64
50 /* The feature bitmap for virtio nic
52 #define VIRTIO_NET_F_CSUM 0 /* Host handles pkts w/ partial csum */
53 #define VIRTIO_NET_F_GUEST_CSUM 1 /* Guest handles pkts w/ partial csum */
54 #define VIRTIO_NET_F_MAC 5 /* Host has given MAC address. */
55 #define VIRTIO_NET_F_GSO 6 /* Host handles pkts w/ any GSO type */
56 #define VIRTIO_NET_F_GUEST_TSO4 7 /* Guest can handle TSOv4 in. */
57 #define VIRTIO_NET_F_GUEST_TSO6 8 /* Guest can handle TSOv6 in. */
58 #define VIRTIO_NET_F_GUEST_ECN 9 /* Guest can handle TSO[6] w/ ECN in. */
59 #define VIRTIO_NET_F_GUEST_UFO 10 /* Guest can handle UFO in. */
60 #define VIRTIO_NET_F_HOST_TSO4 11 /* Host can handle TSOv4 in. */
61 #define VIRTIO_NET_F_HOST_TSO6 12 /* Host can handle TSOv6 in. */
62 #define VIRTIO_NET_F_HOST_ECN 13 /* Host can handle TSO[6] w/ ECN in. */
63 #define VIRTIO_NET_F_HOST_UFO 14 /* Host can handle UFO in. */
64 #define VIRTIO_NET_F_MRG_RXBUF 15 /* Host can merge receive buffers. */
65 #define VIRTIO_NET_F_STATUS 16 /* virtio_net_config.status available */
67 /* Port to get virtio config */
68 #define VIRTIO_NET_CONFIG 20
70 #define VIRTIO_NET_MAX_BUFSIZE (sizeof(struct virtio_net_hdr) + (64 << 10))
72 /* for gso_type in virtio_net_hdr */
73 #define VIRTIO_NET_HDR_GSO_NONE 0
74 #define VIRTIO_NET_HDR_GSO_TCPV4 1 /* GSO frame, IPv4 TCP (TSO) */
75 #define VIRTIO_NET_HDR_GSO_UDP 3 /* GSO frame, IPv4 UDP (UFO) */
76 #define VIRTIO_NET_HDR_GSO_TCPV6 4 /* GSO frame, IPv6 TCP */
77 #define VIRTIO_NET_HDR_GSO_ECN 0x80 /* TCP has ECN set */
80 /* for flags in virtio_net_hdr */
81 #define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /* Use csum_start, csum_offset */
84 /* First element of the scatter-gather list, used with GSO or CSUM features */
89 uint16_t hdr_len; /* Ethernet + IP + tcp/udp hdrs */
90 uint16_t gso_size; /* Bytes to append to hdr_len per frame */
91 uint16_t csum_start; /* Position to start checksumming from */
92 uint16_t csum_offset; /* Offset after that to place checksum */
93 }__attribute__((packed));
96 /* The header to use when the MRG_RXBUF
97 * feature has been negotiated. */
98 struct virtio_net_hdr_mrg_rxbuf {
99 struct virtio_net_hdr hdr;
100 uint16_t num_buffers; /* Number of merged rx buffers */
103 struct virtio_net_config
105 uint8_t mac[ETH_ALEN]; /* VIRTIO_NET_F_MAC */
107 } __attribute__((packed));
109 struct virtio_dev_state {
111 struct vm_device * pci_bus;
112 struct list_head dev_list;
113 struct v3_vm_info *vm;
115 enum {GUEST_DRIVEN=0, VMM_DRIVEN, ADAPTIVE} model;
116 uint64_t lower_thresh_pps, upper_thresh_pps, period_us;
118 uint8_t mac[ETH_ALEN];
121 struct virtio_net_state {
123 struct virtio_net_config net_cfg;
124 struct virtio_config virtio_cfg;
126 struct v3_vm_info * vm;
127 struct vm_device * dev;
128 struct pci_device * pci_dev;
133 struct virtio_queue rx_vq; /* idx 0*/
134 struct virtio_queue tx_vq; /* idx 1*/
135 struct virtio_queue ctrl_vq; /* idx 2*/
137 uint8_t mergeable_rx_bufs;
139 struct v3_timer * timer;
140 struct nic_statistics stats;
142 struct v3_dev_net_ops * net_ops;
143 v3_lock_t rx_lock, tx_lock;
145 uint8_t tx_notify, rx_notify;
146 uint32_t tx_pkts, rx_pkts;
150 struct virtio_dev_state * virtio_dev;
151 struct list_head dev_link;
155 static int virtio_init_state(struct virtio_net_state * virtio)
157 virtio->rx_vq.queue_size = RX_QUEUE_SIZE;
158 virtio->tx_vq.queue_size = TX_QUEUE_SIZE;
159 virtio->ctrl_vq.queue_size = CTRL_QUEUE_SIZE;
161 virtio->rx_vq.ring_desc_addr = 0;
162 virtio->rx_vq.ring_avail_addr = 0;
163 virtio->rx_vq.ring_used_addr = 0;
164 virtio->rx_vq.pfn = 0;
165 virtio->rx_vq.cur_avail_idx = 0;
167 virtio->tx_vq.ring_desc_addr = 0;
168 virtio->tx_vq.ring_avail_addr = 0;
169 virtio->tx_vq.ring_used_addr = 0;
170 virtio->tx_vq.pfn = 0;
171 virtio->tx_vq.cur_avail_idx = 0;
173 virtio->ctrl_vq.ring_desc_addr = 0;
174 virtio->ctrl_vq.ring_avail_addr = 0;
175 virtio->ctrl_vq.ring_used_addr = 0;
176 virtio->ctrl_vq.pfn = 0;
177 virtio->ctrl_vq.cur_avail_idx = 0;
179 virtio->virtio_cfg.pci_isr = 0;
181 virtio->mergeable_rx_bufs = 1;
183 virtio->virtio_cfg.host_features = 0 | (1 << VIRTIO_NET_F_MAC);
184 if(virtio->mergeable_rx_bufs) {
185 virtio->virtio_cfg.host_features |= (1 << VIRTIO_NET_F_MRG_RXBUF);
188 if ((v3_lock_init(&(virtio->rx_lock)) == -1) ||
189 (v3_lock_init(&(virtio->tx_lock)) == -1)){
190 PrintError(VM_NONE, VCORE_NONE, "Virtio NIC: Failure to init locks for net_state\n");
196 static int virtio_deinit_state(struct guest_info *core, struct virtio_net_state *ns)
199 v3_remove_timer(core,ns->timer);
204 static int tx_one_pkt(struct guest_info * core,
205 struct virtio_net_state * virtio,
206 struct vring_desc * buf_desc)
208 uint8_t * buf = NULL;
209 uint32_t len = buf_desc->length;
211 if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
212 PrintDebug(core->vm_info, core, "Could not translate buffer address\n");
216 V3_Net_Print(2, "Virtio-NIC: virtio_tx: size: %d\n", len);
218 v3_hexdump(buf, len, NULL, 0);
221 if(virtio->net_ops->send(buf, len, virtio->backend_data) < 0){
222 virtio->stats.tx_dropped ++;
226 virtio->stats.tx_pkts ++;
227 virtio->stats.tx_bytes += len;
233 /*copy data into ring buffer */
234 static inline int copy_data_to_desc(struct guest_info * core,
235 struct virtio_net_state * virtio_state,
236 struct vring_desc * desc,
241 uint8_t * desc_buf = NULL;
243 if (v3_gpa_to_hva(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
244 PrintDebug(core->vm_info, core, "Could not translate buffer address\n");
247 len = (desc->length < (buf_len+dst_offset))?(desc->length - dst_offset):buf_len;
248 memcpy(desc_buf + dst_offset, buf, len);
254 static inline int get_desc_count(struct virtio_queue * q, int index) {
255 struct vring_desc * tmp_desc = &(q->desc[index]);
258 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
259 tmp_desc = &(q->desc[tmp_desc->next]);
266 static inline void enable_cb(struct virtio_queue *queue){
268 queue->used->flags &= ~ VRING_NO_NOTIFY_FLAG;
272 static inline void disable_cb(struct virtio_queue *queue) {
274 queue->used->flags |= VRING_NO_NOTIFY_FLAG;
278 static int handle_pkt_tx(struct guest_info * core,
279 struct virtio_net_state * virtio_state,
282 struct virtio_queue * q;
283 int txed = 0, left = 0;
286 q = &(virtio_state->tx_vq);
287 if (!q->ring_avail_addr) {
292 struct vring_desc * hdr_desc = NULL;
294 uint16_t desc_idx, tmp_idx;
297 flags = v3_lock_irqsave(virtio_state->tx_lock);
299 if(q->cur_avail_idx == q->avail->index ||
300 (quote > 0 && txed >= quote)) {
301 left = (q->cur_avail_idx != q->avail->index);
302 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
306 desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
307 tmp_idx = q->cur_avail_idx ++;
309 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
311 desc_cnt = get_desc_count(q, desc_idx);
313 PrintError(core->vm_info, core, "VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt);
316 hdr_desc = &(q->desc[desc_idx]);
317 if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) != -1) {
318 struct virtio_net_hdr_mrg_rxbuf * hdr;
319 struct vring_desc * buf_desc;
321 hdr = (struct virtio_net_hdr_mrg_rxbuf *)hdr_addr;
322 desc_idx = hdr_desc->next;
324 /* here we assumed that one ethernet pkt is not splitted into multiple buffer */
325 buf_desc = &(q->desc[desc_idx]);
326 if (tx_one_pkt(core, virtio_state, buf_desc) == -1) {
327 PrintError(core->vm_info, core, "Virtio NIC: Fails to send packet\n");
330 PrintError(core->vm_info, core, "Could not translate block header address\n");
333 flags = v3_lock_irqsave(virtio_state->tx_lock);
335 q->used->ring[q->used->index % q->queue_size].id =
336 q->avail->ring[tmp_idx % q->queue_size];
340 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
345 if (txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
346 v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus,
347 virtio_state->pci_dev, 0);
348 virtio_state->virtio_cfg.pci_isr = 0x1;
349 virtio_state->stats.rx_interrupts ++;
356 static int virtio_setup_queue(struct guest_info *core,
357 struct virtio_net_state * virtio_state,
358 struct virtio_queue * queue,
359 addr_t pfn, addr_t page_addr) {
362 queue->ring_desc_addr = page_addr;
363 queue->ring_avail_addr = page_addr + (queue->queue_size * sizeof(struct vring_desc));
364 queue->ring_used_addr = ((queue->ring_avail_addr) +
365 (sizeof(struct vring_avail)) +
366 (queue->queue_size * sizeof(uint16_t)));
368 // round up to next page boundary.
369 queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
370 if (v3_gpa_to_hva(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
371 PrintError(core->vm_info, core, "Could not translate ring descriptor address\n");
375 if (v3_gpa_to_hva(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
376 PrintError(core->vm_info, core, "Could not translate ring available address\n");
380 if (v3_gpa_to_hva(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
381 PrintError(core->vm_info, core, "Could not translate ring used address\n");
385 PrintDebug(core->vm_info, core, "RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
386 (void *)(queue->ring_desc_addr),
387 (void *)(queue->ring_avail_addr),
388 (void *)(queue->ring_used_addr));
390 PrintDebug(core->vm_info, core, "RingDesc=%p, Avail=%p, Used=%p\n",
391 queue->desc, queue->avail, queue->used);
396 static int virtio_io_write(struct guest_info *core,
397 uint16_t port, void * src,
398 uint_t length, void * private_data)
400 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
401 int port_idx = port % virtio->io_range_size;
403 PrintDebug(core->vm_info, core, "VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n",
404 private_data, port, port_idx,
405 length, *(uint32_t *)src);
408 case GUEST_FEATURES_PORT:
410 PrintError(core->vm_info, core, "Illegal write length for guest features\n");
413 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
416 case VRING_PG_NUM_PORT:
418 PrintError(core->vm_info, core, "Illegal write length for page frame number\n");
421 addr_t pfn = *(uint32_t *)src;
422 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
423 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
426 virtio_setup_queue(core, virtio,
431 virtio_setup_queue(core, virtio,
434 if(virtio->tx_notify == 0){
435 disable_cb(&virtio->tx_vq);
440 virtio_setup_queue(core, virtio,
449 case VRING_Q_SEL_PORT:
450 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
451 if (virtio->virtio_cfg.vring_queue_selector > 2) {
452 PrintError(core->vm_info, core, "Virtio NIC: wrong queue idx: %d\n",
453 virtio->virtio_cfg.vring_queue_selector);
458 case VRING_Q_NOTIFY_PORT:
460 uint16_t queue_idx = *(uint16_t *)src;
462 /* receive queue refill */
463 virtio->stats.tx_interrupts ++;
464 } else if (queue_idx == 1){
465 if (handle_pkt_tx(core, virtio, 0) < 0) {
466 PrintError(core->vm_info, core, "Virtio NIC: Error to handle packet TX\n");
469 virtio->stats.tx_interrupts ++;
470 } else if (queue_idx == 2){
473 PrintError(core->vm_info, core, "Virtio NIC: Wrong queue index %d\n", queue_idx);
478 case VIRTIO_STATUS_PORT:
479 virtio->virtio_cfg.status = *(uint8_t *)src;
480 if (virtio->virtio_cfg.status == 0) {
481 virtio_init_state(virtio);
485 case VIRTIO_ISR_PORT:
486 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
497 static int virtio_io_read(struct guest_info *core,
498 uint16_t port, void * dst,
499 uint_t length, void * private_data)
501 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
502 int port_idx = port % virtio->io_range_size;
503 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
505 PrintDebug(core->vm_info, core, "Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n",
506 private_data, port, port_idx, length);
509 case HOST_FEATURES_PORT:
511 PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for host features\n");
514 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
517 case VRING_PG_NUM_PORT:
519 PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for page frame number\n");
524 *(uint32_t *)dst = virtio->rx_vq.pfn;
527 *(uint32_t *)dst = virtio->tx_vq.pfn;
530 *(uint32_t *)dst = virtio->ctrl_vq.pfn;
537 case VRING_SIZE_PORT:
539 PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for vring size\n");
544 *(uint16_t *)dst = virtio->rx_vq.queue_size;
547 *(uint16_t *)dst = virtio->tx_vq.queue_size;
550 *(uint16_t *)dst = virtio->ctrl_vq.queue_size;
557 case VIRTIO_STATUS_PORT:
559 PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for status\n");
562 *(uint8_t *)dst = virtio->virtio_cfg.status;
565 case VIRTIO_ISR_PORT:
566 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
567 virtio->virtio_cfg.pci_isr = 0;
568 v3_pci_lower_irq(virtio->virtio_dev->pci_bus,
572 case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN:
573 *(uint8_t *)dst = virtio->net_cfg.mac[port_idx-VIRTIO_NET_CONFIG];
577 PrintError(core->vm_info, core, "Virtio NIC: Read of Unhandled Virtio Read:%d\n",
586 /* receiving raw ethernet pkt from backend */
587 static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) {
588 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
589 struct virtio_queue * q = &(virtio->rx_vq);
590 struct virtio_net_hdr_mrg_rxbuf hdr;
592 uint8_t kick_guest = 0;
594 V3_Net_Print(2, "Virtio NIC: virtio_rx: size: %d\n", size);
596 if (!q->ring_avail_addr) {
597 V3_Net_Print(2, "Virtio NIC: RX Queue not set\n");
598 virtio->stats.rx_dropped ++;
603 memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
605 flags = v3_lock_irqsave(virtio->rx_lock);
607 if (q->cur_avail_idx != q->avail->index){
609 struct vring_desc * buf_desc;
610 uint32_t hdr_len, len;
613 hdr_len = (virtio->mergeable_rx_bufs)?
614 sizeof(struct virtio_net_hdr_mrg_rxbuf):
615 sizeof(struct virtio_net_hdr);
617 if(virtio->mergeable_rx_bufs){/* merged buffer */
618 struct vring_desc * hdr_desc;
619 uint16_t old_idx = q->cur_avail_idx;
621 buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
622 hdr_desc = &(q->desc[buf_idx]);
624 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
625 virtio, hdr_desc, buf, size, hdr_len);
631 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
632 q->used->ring[q->used->index % q->queue_size].length = hdr_len + offset;
636 while(offset < size) {
637 buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
638 buf_desc = &(q->desc[buf_idx]);
640 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
641 virtio, buf_desc, buf+offset, size-offset, 0);
643 V3_Net_Print(2, "Virtio NIC: merged buffer, %d buffer size %d\n",
644 hdr.num_buffers, len);
645 q->cur_avail_idx = old_idx;
649 buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
651 q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
652 q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].length = len;
658 copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
659 virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0);
660 q->used->index += hdr.num_buffers;
662 buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
663 buf_desc = &(q->desc[buf_idx]);
666 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
667 virtio, buf_desc, (uchar_t *)&(hdr.hdr), hdr_len, 0);
669 V3_Net_Print(2, "Virtio NIC: rx copy header error %d, hdr_len %d\n",
674 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
675 virtio, buf_desc, buf, size, hdr_len);
677 V3_Net_Print(2, "Virtio NIC: rx copy data error %d\n", len);
682 /* copy rest of data */
683 while(offset < size &&
684 (buf_desc->flags & VIRTIO_NEXT_FLAG)){
685 buf_desc = &(q->desc[buf_desc->next]);
686 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
687 virtio, buf_desc, buf+offset, size-offset, 0);
693 buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
696 V3_Net_Print(2, "Virtio NIC: rx not enough ring buffer, buffer size %d\n",
701 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
702 q->used->ring[q->used->index % q->queue_size].length = size + hdr_len; /* This should be the total length of data sent to guest (header+pkt_data) */
707 virtio->stats.rx_pkts ++;
708 virtio->stats.rx_bytes += size;
710 V3_Net_Print(2, "Virtio NIC: Guest RX queue is full\n");
711 virtio->stats.rx_dropped ++;
713 /* kick guest to refill RX queue */
717 v3_unlock_irqrestore(virtio->rx_lock, flags);
719 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG) || kick_guest) {
720 V3_Net_Print(2, "Virtio NIC: RX Raising IRQ %d\n",
721 virtio->pci_dev->config_header.intr_line);
723 virtio->virtio_cfg.pci_isr = 0x1;
724 v3_pci_raise_irq(virtio->virtio_dev->pci_bus, virtio->pci_dev, 0);
725 virtio->stats.rx_interrupts ++;
728 /* notify guest if it is in guest mode */
729 if((kick_guest || virtio->rx_notify == 1) &&
730 V3_Get_CPU() != virtio->virtio_dev->vm->cores[0].pcpu_id){
731 v3_interrupt_cpu(virtio->virtio_dev->vm,
732 virtio->virtio_dev->vm->cores[0].pcpu_id,
739 virtio->stats.rx_dropped ++;
740 v3_unlock_irqrestore(virtio->rx_lock, flags);
745 static int virtio_free(struct virtio_dev_state * virtio) {
746 struct virtio_net_state * backend = NULL;
747 struct virtio_net_state * tmp = NULL;
750 list_for_each_entry_safe(backend, tmp, &(virtio->dev_list), dev_link) {
751 virtio_deinit_state(&(virtio->vm->cores[0]),backend);
752 list_del(&(backend->dev_link));
762 static struct v3_device_ops dev_ops = {
763 .free = (int (*)(void *))virtio_free,
767 static int virtio_poll(int quote, void * data){
768 struct virtio_net_state * virtio = (struct virtio_net_state *)data;
770 if (virtio->status) {
772 return handle_pkt_tx(&(virtio->vm->cores[0]), virtio, quote);
778 static int register_dev(struct virtio_dev_state * virtio,
779 struct virtio_net_state * net_state)
781 struct pci_device * pci_dev = NULL;
782 struct v3_pci_bar bars[6];
783 int num_ports = sizeof(struct virtio_config);
784 int tmp_ports = num_ports;
787 /* This gets the number of ports, rounded up to a power of 2 */
788 net_state->io_range_size = 1;
789 while (tmp_ports > 0) {
791 net_state->io_range_size <<= 1;
794 /* this is to account for any low order bits being set in num_ports
795 * if there are none, then num_ports was already a power of 2 so we shift right to reset it
797 if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
798 net_state->io_range_size >>= 1;
801 for (i = 0; i < 6; i++) {
802 bars[i].type = PCI_BAR_NONE;
805 PrintDebug(VM_NONE, VCORE_NONE, "Virtio NIC: io_range_size = %d\n",
806 net_state->io_range_size);
808 bars[0].type = PCI_BAR_IO;
809 bars[0].default_base_port = -1;
810 bars[0].num_ports = net_state->io_range_size;
811 bars[0].io_read = virtio_io_read;
812 bars[0].io_write = virtio_io_write;
813 bars[0].private_data = net_state;
815 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
816 0, PCI_AUTO_DEV_NUM, 0,
817 "LNX_VIRTIO_NIC", bars,
818 NULL, NULL, NULL, NULL, net_state);
821 PrintError(VM_NONE, VCORE_NONE, "Virtio NIC: Could not register PCI Device\n");
825 PrintDebug(VM_NONE, VCORE_NONE, "Virtio NIC: registered to PCI bus\n");
827 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
828 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
831 pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
832 pci_dev->config_header.class = PCI_CLASS_NETWORK;
833 pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
834 pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;
835 pci_dev->config_header.intr_pin = 1;
836 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
838 net_state->pci_dev = pci_dev;
839 net_state->virtio_dev = virtio;
841 memcpy(net_state->net_cfg.mac, virtio->mac, 6);
843 virtio_init_state(net_state);
845 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: Registered Intr Line %d\n", pci_dev->config_header.intr_line);
847 /* Add backend to list of devices */
848 list_add(&(net_state->dev_link), &(virtio->dev_list));
853 #define RATE_UPPER_THRESHOLD_DEFAULT 10000 /* 10000 pkts per second, around 100Mbits */
854 #define RATE_LOWER_THRESHOLD_DEFAULT 1000 /* 1000 pkts per second, around 10Mbits */
855 #define PROFILE_PERIOD_DEFAULT 10000 /* us */
857 static void virtio_nic_timer(struct guest_info * core,
858 uint64_t cpu_cycles, uint64_t cpu_freq,
860 struct virtio_net_state * net_state = (struct virtio_net_state *)priv_data;
862 static int profile_ms = 0;
863 uint64_t target_period_us = net_state->virtio_dev->period_us;
864 uint64_t upper_thresh_pps = net_state->virtio_dev->upper_thresh_pps;
865 uint64_t lower_thresh_pps = net_state->virtio_dev->lower_thresh_pps;
868 if(!net_state->status){ /* VNIC is not in working status */
872 period_us = (1000*cpu_cycles)/cpu_freq;
873 net_state->past_us += period_us;
875 if (net_state->past_us > target_period_us) {
877 uint64_t tx_count, rx_count;
878 uint64_t lb_tx_count, lb_rx_count;
879 uint64_t ub_tx_count, ub_rx_count;
881 lb_tx_count = lb_rx_count = (lower_thresh_pps * 1000000) / net_state->past_us; // packets expected in this interval
882 ub_tx_count = ub_rx_count = (upper_thresh_pps * 1000000) / net_state->past_us;
884 tx_count = net_state->stats.tx_pkts - net_state->tx_pkts;
885 rx_count = net_state->stats.rx_pkts - net_state->rx_pkts;
887 net_state->tx_pkts = net_state->stats.tx_pkts;
888 net_state->rx_pkts = net_state->stats.rx_pkts;
890 if(tx_count > ub_tx_count && net_state->tx_notify == 1) {
891 PrintDebug(core->vm_info, core, "Virtio NIC: Switch TX to VMM driven mode\n");
892 disable_cb(&(net_state->tx_vq));
893 net_state->tx_notify = 0;
896 if(tx_count < lb_tx_count && net_state->tx_notify == 0) {
897 PrintDebug(core->vm_info, core, "Virtio NIC: Switch TX to Guest driven mode\n");
898 enable_cb(&(net_state->tx_vq));
899 net_state->tx_notify = 1;
902 if(rx_count > ub_rx_count && net_state->rx_notify == 1) {
903 PrintDebug(core->vm_info, core, "Virtio NIC: Switch RX to VMM None notify mode\n");
904 net_state->rx_notify = 0;
907 if(rx_count < lb_rx_count && net_state->rx_notify == 0) {
908 V3_Print(core->vm_info, core, "Virtio NIC: Switch RX to VMM notify mode\n");
909 net_state->rx_notify = 1;
912 net_state->past_us = 0;
915 profile_ms += period_us/1000;
916 if(profile_ms > 20000){
917 PrintDebug(core->vm_info, core, "Virtio NIC: TX: Pkt: %lld, Bytes: %lld\n\t\tRX Pkt: %lld. Bytes: %lld\n\t\tDropped: tx %lld, rx %lld\nInterrupts: tx %d, rx %d\nTotal Exit: %lld\n",
918 net_state->stats.tx_pkts, net_state->stats.tx_bytes,
919 net_state->stats.rx_pkts, net_state->stats.rx_bytes,
920 net_state->stats.tx_dropped, net_state->stats.rx_dropped,
921 net_state->stats.tx_interrupts, net_state->stats.rx_interrupts,
922 net_state->vm->cores[0].num_exits);
927 static struct v3_timer_ops timer_ops = {
928 .update_timer = virtio_nic_timer,
931 static int connect_fn(struct v3_vm_info * info,
932 void * frontend_data,
933 struct v3_dev_net_ops * ops,
935 void * private_data) {
936 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
937 struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
940 PrintError(info, VCORE_NONE, "Cannot allocate in connect\n");
944 memset(net_state, 0, sizeof(struct virtio_net_state));
945 register_dev(virtio, net_state);
947 net_state->vm = info;
948 net_state->net_ops = ops;
949 net_state->backend_data = private_data;
950 net_state->virtio_dev = virtio;
952 switch (virtio->model) {
954 V3_Print(info, VCORE_NONE, "Virtio NIC: Guest-driven operation\n");
955 net_state->tx_notify = 1;
956 net_state->rx_notify = 1;
959 V3_Print(info, VCORE_NONE, "Virtio NIC: VMM-driven operation\n");
960 net_state->tx_notify = 0;
961 net_state->rx_notify = 0;
964 V3_Print(info, VCORE_NONE, "Virtio NIC: Adaptive operation (begins in guest-driven mode)\n");
965 net_state->tx_notify = 1;
966 net_state->rx_notify = 1;
968 net_state->timer = v3_add_timer(&(info->cores[0]), &timer_ops,net_state);
974 V3_Print(info, VCORE_NONE, "Virtio NIC: Unknown model, using GUEST_DRIVEN\n");
975 net_state->tx_notify = 1;
976 net_state->rx_notify = 1;
981 ops->recv = virtio_rx;
982 ops->poll = virtio_poll;
983 ops->config.frontend_data = net_state;
984 ops->config.poll = 1;
985 ops->config.quote = 64;
986 ops->config.fnt_mac = V3_Malloc(ETH_ALEN);
988 if (!ops->config.fnt_mac) {
989 PrintError(info, VCORE_NONE, "Cannot allocate in connect\n");
990 // should unregister here
994 memcpy(ops->config.fnt_mac, virtio->mac, ETH_ALEN);
999 static int setup_perf_model(struct virtio_dev_state *virtio_state, v3_cfg_tree_t *t)
1001 char *mode = v3_cfg_val(t,"mode");
1004 virtio_state->model = GUEST_DRIVEN;
1005 virtio_state->lower_thresh_pps = RATE_LOWER_THRESHOLD_DEFAULT;
1006 virtio_state->upper_thresh_pps = RATE_UPPER_THRESHOLD_DEFAULT;
1007 virtio_state->period_us = PROFILE_PERIOD_DEFAULT;
1012 if (!strcasecmp(mode,"guest-driven")) {
1013 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: Setting static GUEST_DRIVEN mode of operation (latency optimized)\n");
1014 virtio_state->model=GUEST_DRIVEN;
1015 } else if (!strcasecmp(mode, "vmm-driven")) {
1016 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: Setting static VMM_DRIVEN mode of operation (throughput optimized)\n");
1017 virtio_state->model=VMM_DRIVEN;
1018 } else if (!strcasecmp(mode, "adaptive")) {
1021 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: Setting dynamic ADAPTIVE mode of operation\n");
1022 virtio_state->model=ADAPTIVE;
1024 if (!(s=v3_cfg_val(t,"upper"))) {
1025 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No upper bound given, using default\n");
1027 virtio_state->upper_thresh_pps = atoi(s);
1029 if (!(s=v3_cfg_val(t,"lower"))) {
1030 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No lower bound given, using default\n");
1032 virtio_state->lower_thresh_pps = atoi(s);
1034 if (!(s=v3_cfg_val(t,"period"))) {
1035 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No period given, using default\n");
1037 virtio_state->period_us = atoi(s);
1040 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: lower_thresh_pps=%llu, upper_thresh_pps=%llu, period_us=%llu\n",
1041 virtio_state->lower_thresh_pps,
1042 virtio_state->upper_thresh_pps,
1043 virtio_state->period_us);
1045 PrintError(VM_NONE, VCORE_NONE, "Virtio NIC: Unknown mode of operation '%s', using default (guest-driven)\n",mode);
1046 virtio_state->model=GUEST_DRIVEN;
1049 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No model given, using default (guest-driven)\n");
1057 <device class="LNX_VIRTIO_NIC" id="nic">
1058 <bus>pci-bus-to-attach-to</bus> // required
1059 <mac>mac address</mac> // if ommited with pic one
1060 <model mode="guest-driven|vmm-driven|adaptive" upper="pkts_per_sec" lower="pkts" period="us" />
1063 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
1064 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
1065 struct virtio_dev_state * virtio_state = NULL;
1066 char * dev_id = v3_cfg_val(cfg, "ID");
1067 char * mac = v3_cfg_val(cfg, "mac");
1068 v3_cfg_tree_t *model = v3_cfg_subtree(cfg,"model");
1070 if (pci_bus == NULL) {
1071 PrintError(vm, VCORE_NONE, "Virtio NIC: Virtio device require a PCI Bus");
1075 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
1077 if (!virtio_state) {
1078 PrintError(vm, VCORE_NONE, "Cannot allocate in init\n");
1082 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
1084 INIT_LIST_HEAD(&(virtio_state->dev_list));
1085 virtio_state->pci_bus = pci_bus;
1086 virtio_state->vm = vm;
1089 if (!str2mac(mac, virtio_state->mac)) {
1090 PrintDebug(vm, VCORE_NONE, "Virtio NIC: Mac specified %s\n", mac);
1092 PrintError(vm, VCORE_NONE, "Virtio NIC: Mac specified is incorrect, picking a randoom mac\n");
1093 random_ethaddr(virtio_state->mac);
1096 PrintDebug(vm, VCORE_NONE, "Virtio NIC: no mac specified, so picking a random mac\n");
1097 random_ethaddr(virtio_state->mac);
1100 if (setup_perf_model(virtio_state,model)<0) {
1101 PrintError(vm, VCORE_NONE, "Cannnot setup performance model\n");
1102 V3_Free(virtio_state);
1106 struct vm_device * dev = v3_add_device(vm, dev_id, &dev_ops, virtio_state);
1109 PrintError(vm, VCORE_NONE, "Virtio NIC: Could not attach device %s\n", dev_id);
1110 V3_Free(virtio_state);
1114 if (v3_dev_add_net_frontend(vm, dev_id, connect_fn, (void *)virtio_state) == -1) {
1115 PrintError(vm, VCORE_NONE, "Virtio NIC: Could not register %s as net frontend\n", dev_id);
1116 v3_remove_device(dev);
1123 device_register("LNX_VIRTIO_NIC", virtio_init)