2 * This file is part of the Palacios Virtual Machine Monitor developed
3 * by the V3VEE Project with funding from the United States National
4 * Science Foundation and the Department of Energy.
6 * The V3VEE Project is a joint project between Northwestern University
7 * and the University of New Mexico. You can find out more at
10 * Copyright (c) 2010, Lei Xia <lxia@northwestern.edu>
11 * Copyright (c) 2010, Cui Zheng <cuizheng@cs.unm.edu>
12 * Copyright (c) 2010, The V3VEE Project <http://www.v3vee.org>
13 * All rights reserved.
15 * Author: Lei Xia <lxia@northwestern.edu>
16 * Cui Zheng <cuizheng@cs.unm.edu>
19 * This is free software. You are permitted to use,
20 * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
23 #include <palacios/vmm.h>
24 #include <palacios/vmm_dev_mgr.h>
25 #include <devices/lnx_virtio_pci.h>
26 #include <palacios/vm_guest_mem.h>
27 #include <palacios/vmm_sprintf.h>
28 #include <vnet/vnet.h>
29 #include <palacios/vmm_lock.h>
30 #include <palacios/vmm_util.h>
31 #include <devices/pci.h>
32 #include <palacios/vmm_ethernet.h>
33 #include <palacios/vmm_time.h>
36 #ifndef V3_CONFIG_DEBUG_VIRTIO_NET
38 #define PrintDebug(fmt, args...)
41 #ifndef V3_CONFIG_VNET
42 static int net_debug = 0;
45 #define TX_QUEUE_SIZE 4096
46 #define RX_QUEUE_SIZE 4096
47 #define CTRL_QUEUE_SIZE 64
49 /* The feature bitmap for virtio nic
51 #define VIRTIO_NET_F_CSUM 0 /* Host handles pkts w/ partial csum */
52 #define VIRTIO_NET_F_GUEST_CSUM 1 /* Guest handles pkts w/ partial csum */
53 #define VIRTIO_NET_F_MAC 5 /* Host has given MAC address. */
54 #define VIRTIO_NET_F_GSO 6 /* Host handles pkts w/ any GSO type */
55 #define VIRTIO_NET_F_GUEST_TSO4 7 /* Guest can handle TSOv4 in. */
56 #define VIRTIO_NET_F_GUEST_TSO6 8 /* Guest can handle TSOv6 in. */
57 #define VIRTIO_NET_F_GUEST_ECN 9 /* Guest can handle TSO[6] w/ ECN in. */
58 #define VIRTIO_NET_F_GUEST_UFO 10 /* Guest can handle UFO in. */
59 #define VIRTIO_NET_F_HOST_TSO4 11 /* Host can handle TSOv4 in. */
60 #define VIRTIO_NET_F_HOST_TSO6 12 /* Host can handle TSOv6 in. */
61 #define VIRTIO_NET_F_HOST_ECN 13 /* Host can handle TSO[6] w/ ECN in. */
62 #define VIRTIO_NET_F_HOST_UFO 14 /* Host can handle UFO in. */
63 #define VIRTIO_NET_F_MRG_RXBUF 15 /* Host can merge receive buffers. */
64 #define VIRTIO_NET_F_STATUS 16 /* virtio_net_config.status available */
66 /* Port to get virtio config */
67 #define VIRTIO_NET_CONFIG 20
69 #define VIRTIO_NET_MAX_BUFSIZE (sizeof(struct virtio_net_hdr) + (64 << 10))
71 /* for gso_type in virtio_net_hdr */
72 #define VIRTIO_NET_HDR_GSO_NONE 0
73 #define VIRTIO_NET_HDR_GSO_TCPV4 1 /* GSO frame, IPv4 TCP (TSO) */
74 #define VIRTIO_NET_HDR_GSO_UDP 3 /* GSO frame, IPv4 UDP (UFO) */
75 #define VIRTIO_NET_HDR_GSO_TCPV6 4 /* GSO frame, IPv6 TCP */
76 #define VIRTIO_NET_HDR_GSO_ECN 0x80 /* TCP has ECN set */
79 /* for flags in virtio_net_hdr */
80 #define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /* Use csum_start, csum_offset */
83 /* First element of the scatter-gather list, used with GSO or CSUM features */
88 uint16_t hdr_len; /* Ethernet + IP + tcp/udp hdrs */
89 uint16_t gso_size; /* Bytes to append to hdr_len per frame */
90 uint16_t csum_start; /* Position to start checksumming from */
91 uint16_t csum_offset; /* Offset after that to place checksum */
92 }__attribute__((packed));
95 /* The header to use when the MRG_RXBUF
96 * feature has been negotiated. */
97 struct virtio_net_hdr_mrg_rxbuf {
98 struct virtio_net_hdr hdr;
99 uint16_t num_buffers; /* Number of merged rx buffers */
102 struct virtio_net_config
104 uint8_t mac[ETH_ALEN]; /* VIRTIO_NET_F_MAC */
106 } __attribute__((packed));
108 struct virtio_dev_state {
109 struct vm_device * pci_bus;
110 struct list_head dev_list;
111 struct v3_vm_info *vm;
113 uint8_t mac[ETH_ALEN];
116 struct virtio_net_state {
117 struct virtio_net_config net_cfg;
118 struct virtio_config virtio_cfg;
120 struct v3_vm_info * vm;
121 struct vm_device * dev;
122 struct pci_device * pci_dev;
127 struct virtio_queue rx_vq; /* idx 0*/
128 struct virtio_queue tx_vq; /* idx 1*/
129 struct virtio_queue ctrl_vq; /* idx 2*/
131 uint8_t mergeable_rx_bufs;
133 struct v3_timer * timer;
134 struct vnet_thread * poll_thread;
136 struct nic_statistics stats;
138 struct v3_dev_net_ops * net_ops;
139 v3_lock_t rx_lock, tx_lock;
141 uint8_t tx_notify, rx_notify;
142 uint32_t tx_pkts, rx_pkts;
146 struct virtio_dev_state * virtio_dev;
147 struct list_head dev_link;
151 static int virtio_init_state(struct virtio_net_state * virtio)
153 virtio->rx_vq.queue_size = RX_QUEUE_SIZE;
154 virtio->tx_vq.queue_size = TX_QUEUE_SIZE;
155 virtio->ctrl_vq.queue_size = CTRL_QUEUE_SIZE;
157 virtio->rx_vq.ring_desc_addr = 0;
158 virtio->rx_vq.ring_avail_addr = 0;
159 virtio->rx_vq.ring_used_addr = 0;
160 virtio->rx_vq.pfn = 0;
161 virtio->rx_vq.cur_avail_idx = 0;
163 virtio->tx_vq.ring_desc_addr = 0;
164 virtio->tx_vq.ring_avail_addr = 0;
165 virtio->tx_vq.ring_used_addr = 0;
166 virtio->tx_vq.pfn = 0;
167 virtio->tx_vq.cur_avail_idx = 0;
169 virtio->ctrl_vq.ring_desc_addr = 0;
170 virtio->ctrl_vq.ring_avail_addr = 0;
171 virtio->ctrl_vq.ring_used_addr = 0;
172 virtio->ctrl_vq.pfn = 0;
173 virtio->ctrl_vq.cur_avail_idx = 0;
175 virtio->virtio_cfg.pci_isr = 0;
177 virtio->mergeable_rx_bufs = 0;
179 virtio->virtio_cfg.host_features = 0 | (1 << VIRTIO_NET_F_MAC);
180 if(virtio->mergeable_rx_bufs) {
181 virtio->virtio_cfg.host_features |= (1 << VIRTIO_NET_F_MRG_RXBUF);
184 if ((v3_lock_init(&(virtio->rx_lock)) == -1) ||
185 (v3_lock_init(&(virtio->tx_lock)) == -1)){
186 PrintError("Virtio NIC: Failure to init locks for net_state\n");
192 static int tx_one_pkt(struct guest_info * core,
193 struct virtio_net_state * virtio,
194 struct vring_desc * buf_desc)
196 uint8_t * buf = NULL;
197 uint32_t len = buf_desc->length;
198 int synchronize = virtio->tx_notify;
200 if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
201 PrintDebug("Could not translate buffer address\n");
205 V3_Net_Print(2, "Virtio-NIC: virtio_tx: size: %d\n", len);
207 v3_hexdump(buf, len, NULL, 0);
210 if(virtio->net_ops->send(buf, len, synchronize, virtio->backend_data) < 0){
211 virtio->stats.tx_dropped ++;
215 virtio->stats.tx_pkts ++;
216 virtio->stats.tx_bytes += len;
222 static inline int copy_data_to_desc(struct guest_info * core,
223 struct virtio_net_state * virtio_state,
224 struct vring_desc * desc,
229 uint8_t * desc_buf = NULL;
231 if (v3_gpa_to_hva(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
232 PrintDebug("Could not translate buffer address\n");
235 len = (desc->length < buf_len)?(desc->length - dst_offset):buf_len;
236 memcpy(desc_buf+dst_offset, buf, len);
242 static inline int get_desc_count(struct virtio_queue * q, int index) {
243 struct vring_desc * tmp_desc = &(q->desc[index]);
246 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
247 tmp_desc = &(q->desc[tmp_desc->next]);
254 static inline void enable_cb(struct virtio_queue *queue){
256 queue->used->flags &= ~ VRING_NO_NOTIFY_FLAG;
260 static inline void disable_cb(struct virtio_queue *queue) {
262 queue->used->flags |= VRING_NO_NOTIFY_FLAG;
266 static int handle_pkt_tx(struct guest_info * core,
267 struct virtio_net_state * virtio_state)
269 struct virtio_queue *q = &(virtio_state->tx_vq);
273 if (!q->ring_avail_addr) {
277 flags = v3_lock_irqsave(virtio_state->tx_lock);
278 while (q->cur_avail_idx != q->avail->index) {
279 struct virtio_net_hdr_mrg_rxbuf * hdr = NULL;
280 struct vring_desc * hdr_desc = NULL;
282 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
283 int desc_cnt = get_desc_count(q, desc_idx);
286 PrintError("VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt);
290 hdr_desc = &(q->desc[desc_idx]);
291 if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
292 PrintError("Could not translate block header address\n");
296 hdr = (struct virtio_net_hdr_mrg_rxbuf *)hdr_addr;
297 desc_idx = hdr_desc->next;
299 V3_Net_Print(2, "Virtio NIC: TX hdr count : %d\n", hdr->num_buffers);
301 /* here we assumed that one ethernet pkt is not splitted into multiple buffer */
302 struct vring_desc * buf_desc = &(q->desc[desc_idx]);
303 if (tx_one_pkt(core, virtio_state, buf_desc) == -1) {
304 PrintError("Virtio NIC: Error handling nic operation\n");
307 if(buf_desc->next & VIRTIO_NEXT_FLAG){
308 V3_Net_Print(2, "Virtio NIC: TX more buffer need to read\n");
311 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
312 q->used->ring[q->used->index % q->queue_size].length = buf_desc->length; /* What do we set this to???? */
320 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
322 if (txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
323 v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev);
324 virtio_state->virtio_cfg.pci_isr = 0x1;
326 virtio_state->stats.rx_interrupts ++;
330 V3_Net_Print(2, "Virtio Handle TX: txed pkts: %d\n", txed);
337 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
342 static int virtio_setup_queue(struct guest_info *core,
343 struct virtio_net_state * virtio_state,
344 struct virtio_queue * queue,
345 addr_t pfn, addr_t page_addr) {
348 queue->ring_desc_addr = page_addr;
349 queue->ring_avail_addr = page_addr + (queue->queue_size * sizeof(struct vring_desc));
350 queue->ring_used_addr = ((queue->ring_avail_addr) +
351 (sizeof(struct vring_avail)) +
352 (queue->queue_size * sizeof(uint16_t)));
354 // round up to next page boundary.
355 queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
356 if (v3_gpa_to_hva(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
357 PrintError("Could not translate ring descriptor address\n");
361 if (v3_gpa_to_hva(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
362 PrintError("Could not translate ring available address\n");
366 if (v3_gpa_to_hva(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
367 PrintError("Could not translate ring used address\n");
371 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
372 (void *)(queue->ring_desc_addr),
373 (void *)(queue->ring_avail_addr),
374 (void *)(queue->ring_used_addr));
376 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
377 queue->desc, queue->avail, queue->used);
382 static int virtio_io_write(struct guest_info *core,
383 uint16_t port, void * src,
384 uint_t length, void * private_data)
386 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
387 int port_idx = port % virtio->io_range_size;
389 PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n", private_data,
390 port, port_idx, length, *(uint32_t *)src);
393 case GUEST_FEATURES_PORT:
395 PrintError("Illegal write length for guest features\n");
398 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
401 case VRING_PG_NUM_PORT:
403 PrintError("Illegal write length for page frame number\n");
406 addr_t pfn = *(uint32_t *)src;
407 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
408 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
411 virtio_setup_queue(core, virtio, &virtio->rx_vq, pfn, page_addr);
414 virtio_setup_queue(core, virtio, &virtio->tx_vq, pfn, page_addr);
415 if(virtio->tx_notify == 0){
416 disable_cb(&virtio->tx_vq);
417 vnet_thread_wakeup(virtio->poll_thread);
421 virtio_setup_queue(core, virtio, &virtio->ctrl_vq, pfn, page_addr);
428 case VRING_Q_SEL_PORT:
429 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
430 if (virtio->virtio_cfg.vring_queue_selector > 2) {
431 PrintError("Virtio NIC: wrong queue idx: %d\n",
432 virtio->virtio_cfg.vring_queue_selector);
437 case VRING_Q_NOTIFY_PORT:
439 uint16_t queue_idx = *(uint16_t *)src;
441 /* receive queue refill */
442 virtio->stats.tx_interrupts ++;
443 } else if (queue_idx == 1){
444 if (handle_pkt_tx(core, virtio) == -1) {
445 PrintError("Could not handle Virtio NIC tx kick\n");
448 virtio->stats.tx_interrupts ++;
449 } else if (queue_idx == 2){
452 PrintError("Wrong queue index %d\n", queue_idx);
457 case VIRTIO_STATUS_PORT:
458 virtio->virtio_cfg.status = *(uint8_t *)src;
459 if (virtio->virtio_cfg.status == 0) {
460 virtio_init_state(virtio);
464 case VIRTIO_ISR_PORT:
465 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
476 static int virtio_io_read(struct guest_info *core,
477 uint16_t port, void * dst,
478 uint_t length, void * private_data)
480 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
481 int port_idx = port % virtio->io_range_size;
482 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
484 PrintDebug("Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n", private_data,
485 port, port_idx, length);
488 case HOST_FEATURES_PORT:
490 PrintError("Illegal read length for host features\n");
493 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
496 case VRING_PG_NUM_PORT:
498 PrintError("Illegal read length for page frame number\n");
503 *(uint32_t *)dst = virtio->rx_vq.pfn;
506 *(uint32_t *)dst = virtio->tx_vq.pfn;
509 *(uint32_t *)dst = virtio->ctrl_vq.pfn;
516 case VRING_SIZE_PORT:
518 PrintError("Illegal read length for vring size\n");
523 *(uint16_t *)dst = virtio->rx_vq.queue_size;
526 *(uint16_t *)dst = virtio->tx_vq.queue_size;
529 *(uint16_t *)dst = virtio->ctrl_vq.queue_size;
536 case VIRTIO_STATUS_PORT:
538 PrintError("Illegal read length for status\n");
541 *(uint8_t *)dst = virtio->virtio_cfg.status;
544 case VIRTIO_ISR_PORT:
545 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
546 virtio->virtio_cfg.pci_isr = 0;
547 v3_pci_lower_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
550 case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN:
551 *(uint8_t *)dst = virtio->net_cfg.mac[port_idx-VIRTIO_NET_CONFIG];
555 PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n", port_idx);
563 /* receiving raw ethernet pkt from backend */
564 static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) {
565 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
566 struct virtio_queue * q = &(virtio->rx_vq);
567 struct virtio_net_hdr_mrg_rxbuf hdr;
571 V3_Net_Print(2, "Virtio-NIC: virtio_rx: size: %d\n", size);
573 v3_hexdump(buf, size, NULL, 0);
576 flags = v3_lock_irqsave(virtio->rx_lock);
579 memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
581 if (q->ring_avail_addr == 0) {
582 V3_Net_Print(2, "Virtio NIC: RX Queue not set\n");
583 virtio->stats.rx_dropped ++;
587 if (q->cur_avail_idx != q->avail->index){
589 uint16_t buf_idx = 0;
590 uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
591 struct vring_desc * hdr_desc = NULL;
592 struct vring_desc * buf_desc = NULL;
593 uint32_t hdr_len = 0;
596 hdr_desc = &(q->desc[hdr_idx]);
597 if (v3_gpa_to_hva(&(virtio->virtio_dev->vm->cores[0]), hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
598 V3_Net_Print(2, "Virtio NIC: Could not translate receive buffer address\n");
599 virtio->stats.rx_dropped ++;
603 hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
605 if(virtio->mergeable_rx_bufs){/* merged buffer */
610 hdr_desc = &(q->desc[buf_idx]);
611 hdr_desc->flags &= ~VIRTIO_NEXT_FLAG;
613 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, buf, data_len, hdr_len);
617 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
618 q->used->ring[q->used->index % q->queue_size].length = hdr_len + len;
621 while(offset < data_len) {
622 buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
623 buf_desc = &(q->desc[buf_idx]);
625 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf + offset, data_len - offset, 0);
627 V3_Net_Print(2, "Virtio NIC:merged buffer, %d buffer size %d\n", hdr.num_buffers, data_len);
628 virtio->stats.rx_dropped ++;
632 buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
635 q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
636 q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].length = len;
639 q->used->index += hdr.num_buffers;
640 copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0);
642 hdr_desc = &(q->desc[buf_idx]);
643 copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0);
645 buf_idx = hdr_desc->next;
646 buf_desc = &(q->desc[buf_idx]);
647 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf, data_len, 0);
648 if (len < data_len) {
649 V3_Net_Print(2, "Virtio NIC: ring buffer len less than pkt size, merged buffer not supported, buffer size %d\n", len);
650 virtio->stats.rx_dropped ++;
654 buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
656 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
657 q->used->ring[q->used->index % q->queue_size].length = data_len + hdr_len; /* This should be the total length of data sent to guest (header+pkt_data) */
662 virtio->stats.rx_pkts ++;
663 virtio->stats.rx_bytes += size;
665 V3_Net_Print(2, "Virtio NIC: Guest RX queue is full\n");
666 virtio->stats.rx_dropped ++;
668 /* kick guest to refill the queue */
669 virtio->virtio_cfg.pci_isr = 0x1;
670 v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
671 v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].pcpu_id, 0);
672 virtio->stats.rx_interrupts ++;
677 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
678 V3_Net_Print(2, "Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line);
680 virtio->virtio_cfg.pci_isr = 0x1;
681 v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
683 virtio->stats.rx_interrupts ++;
686 v3_unlock_irqrestore(virtio->rx_lock, flags);
688 /* notify guest if it is in guest mode */
689 if(virtio->rx_notify == 1 &&
690 V3_Get_CPU() != virtio->virtio_dev->vm->cores[0].pcpu_id){
691 v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].pcpu_id, 0);
698 v3_unlock_irqrestore(virtio->rx_lock, flags);
703 static int virtio_free(struct virtio_dev_state * virtio) {
704 struct virtio_net_state * backend = NULL;
705 struct virtio_net_state * tmp = NULL;
708 list_for_each_entry_safe(backend, tmp, &(virtio->dev_list), dev_link) {
710 // unregister from PCI
712 list_del(&(backend->dev_link));
721 static struct v3_device_ops dev_ops = {
722 .free = (int (*)(void *))virtio_free,
726 static int virtio_tx_flush(void * args){
727 struct virtio_net_state *virtio = (struct virtio_net_state *)args;
729 V3_Print("Virtio TX Poll Thread Starting for %s\n", virtio->vm->name);
732 if(virtio->tx_notify == 0){
733 handle_pkt_tx(&(virtio->vm->cores[0]), virtio);
736 vnet_thread_sleep(-1);
743 static int register_dev(struct virtio_dev_state * virtio,
744 struct virtio_net_state * net_state)
746 struct pci_device * pci_dev = NULL;
747 struct v3_pci_bar bars[6];
748 int num_ports = sizeof(struct virtio_config);
749 int tmp_ports = num_ports;
752 // This gets the number of ports, rounded up to a power of 2
753 net_state->io_range_size = 1; // must be a power of 2
754 while (tmp_ports > 0) {
756 net_state->io_range_size <<= 1;
759 /* this is to account for any low order bits being set in num_ports
760 * if there are none, then num_ports was already a power of 2 so we shift right to reset it
762 if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
763 net_state->io_range_size >>= 1;
766 for (i = 0; i < 6; i++) {
767 bars[i].type = PCI_BAR_NONE;
770 PrintDebug("Virtio-NIC io_range_size = %d\n", net_state->io_range_size);
772 bars[0].type = PCI_BAR_IO;
773 bars[0].default_base_port = -1;
774 bars[0].num_ports = net_state->io_range_size;
775 bars[0].io_read = virtio_io_read;
776 bars[0].io_write = virtio_io_write;
777 bars[0].private_data = net_state;
779 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
780 0, 4/*PCI_AUTO_DEV_NUM*/, 0,
781 "LNX_VIRTIO_NIC", bars,
782 NULL, NULL, NULL, net_state);
785 PrintError("Virtio NIC: Could not register PCI Device\n");
789 PrintDebug("Virtio NIC: registered to PCI bus\n");
791 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
792 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
795 pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
796 pci_dev->config_header.class = PCI_CLASS_NETWORK;
797 pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
798 pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;
799 pci_dev->config_header.intr_pin = 1;
800 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
802 net_state->pci_dev = pci_dev;
803 net_state->virtio_dev = virtio;
805 memcpy(net_state->net_cfg.mac, virtio->mac, 6);
807 virtio_init_state(net_state);
809 /* Add backend to list of devices */
810 list_add(&(net_state->dev_link), &(virtio->dev_list));
816 #define RATE_UPPER_THRESHOLD 10 /* 10000 pkts per second, around 100Mbits */
817 #define RATE_LOWER_THRESHOLD 1
818 #define PROFILE_PERIOD 10000 /*us*/
820 static void virtio_nic_timer(struct guest_info * core,
821 uint64_t cpu_cycles, uint64_t cpu_freq,
823 struct virtio_net_state * net_state = (struct virtio_net_state *)priv_data;
825 static int profile_ms = 0;
827 if(!net_state->status){ /* VNIC is not in working status */
831 period_us = (1000*cpu_cycles)/cpu_freq;
832 net_state->past_us += period_us;
834 if(net_state->past_us > PROFILE_PERIOD){
835 uint32_t tx_rate, rx_rate;
837 tx_rate = (net_state->stats.tx_pkts - net_state->tx_pkts)/(net_state->past_us/1000); /* pkts/per ms */
838 rx_rate = (net_state->stats.rx_pkts - net_state->rx_pkts)/(net_state->past_us/1000);
840 net_state->tx_pkts = net_state->stats.tx_pkts;
841 net_state->rx_pkts = net_state->stats.rx_pkts;
843 if(tx_rate > RATE_UPPER_THRESHOLD && net_state->tx_notify == 1){
844 V3_Print("Virtio NIC: Switch TX to VMM driven mode\n");
845 disable_cb(&(net_state->tx_vq));
846 net_state->tx_notify = 0;
847 vnet_thread_wakeup(net_state->poll_thread);
850 if(tx_rate < RATE_LOWER_THRESHOLD && net_state->tx_notify == 0){
851 V3_Print("Virtio NIC: Switch TX to Guest driven mode\n");
852 enable_cb(&(net_state->tx_vq));
853 net_state->tx_notify = 1;
856 if(rx_rate > RATE_UPPER_THRESHOLD && net_state->rx_notify == 1){
857 V3_Print("Virtio NIC: Switch RX to VMM None notify mode\n");
858 net_state->rx_notify = 0;
861 if(rx_rate < RATE_LOWER_THRESHOLD && net_state->rx_notify == 0){
862 V3_Print("Virtio NIC: Switch RX to VMM notify mode\n");
863 net_state->rx_notify = 1;
866 net_state->past_us = 0;
869 profile_ms += period_us/1000;
870 if(profile_ms > 20000){
871 V3_Net_Print(1, "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",
872 net_state->stats.tx_pkts, net_state->stats.tx_bytes,
873 net_state->stats.rx_pkts, net_state->stats.rx_bytes,
874 net_state->stats.tx_dropped, net_state->stats.rx_dropped,
875 net_state->stats.tx_interrupts, net_state->stats.rx_interrupts,
876 net_state->vm->cores[0].num_exits);
881 static struct v3_timer_ops timer_ops = {
882 .update_timer = virtio_nic_timer,
886 static int connect_fn(struct v3_vm_info * info,
887 void * frontend_data,
888 struct v3_dev_net_ops * ops,
890 void * private_data) {
891 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
892 struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
894 memset(net_state, 0, sizeof(struct virtio_net_state));
895 register_dev(virtio, net_state);
897 net_state->vm = info;
898 net_state->net_ops = ops;
899 net_state->backend_data = private_data;
900 net_state->virtio_dev = virtio;
901 net_state->tx_notify = 0;
902 net_state->rx_notify = 0;
904 net_state->timer = v3_add_timer(&(info->cores[0]),&timer_ops,net_state);
906 PrintError("net_state 0x%p\n", (void *)net_state);
908 ops->recv = virtio_rx;
909 ops->frontend_data = net_state;
910 memcpy(ops->fnt_mac, virtio->mac, ETH_ALEN);
912 net_state->poll_thread = vnet_start_thread(virtio_tx_flush, (void *)net_state, "Virtio_Poll");
914 net_state->status = 1;
919 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
920 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
921 struct virtio_dev_state * virtio_state = NULL;
922 char * dev_id = v3_cfg_val(cfg, "ID");
924 char * str = v3_cfg_val(cfg, "mac");
925 memcpy(macstr, str, strlen(str));
927 if (pci_bus == NULL) {
928 PrintError("Virtio NIC: VirtIO devices require a PCI Bus");
932 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
933 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
935 INIT_LIST_HEAD(&(virtio_state->dev_list));
936 virtio_state->pci_bus = pci_bus;
937 virtio_state->vm = vm;
939 if (macstr != NULL && !str2mac(macstr, virtio_state->mac)) {
940 PrintDebug("Virtio NIC: Mac specified %s\n", macstr);
942 random_ethaddr(virtio_state->mac);
945 struct vm_device * dev = v3_add_device(vm, dev_id, &dev_ops, virtio_state);
948 PrintError("Virtio NIC: Could not attach device %s\n", dev_id);
949 V3_Free(virtio_state);
953 if (v3_dev_add_net_frontend(vm, dev_id, connect_fn, (void *)virtio_state) == -1) {
954 PrintError("Virtio NIC: Could not register %s as net frontend\n", dev_id);
955 v3_remove_device(dev);
962 device_register("LNX_VIRTIO_NIC", virtio_init)