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 {
110 struct vm_device * pci_bus;
111 struct list_head dev_list;
112 struct v3_vm_info *vm;
114 uint8_t mac[ETH_ALEN];
117 struct virtio_net_state {
118 struct virtio_net_config net_cfg;
119 struct virtio_config virtio_cfg;
121 struct v3_vm_info * vm;
122 struct vm_device * dev;
123 struct pci_device * pci_dev;
128 struct virtio_queue rx_vq; /* idx 0*/
129 struct virtio_queue tx_vq; /* idx 1*/
130 struct virtio_queue ctrl_vq; /* idx 2*/
132 uint8_t mergeable_rx_bufs;
134 struct v3_timer * timer;
135 struct nic_statistics stats;
137 struct v3_dev_net_ops * net_ops;
138 v3_lock_t rx_lock, tx_lock;
140 uint8_t tx_notify, rx_notify;
141 uint32_t tx_pkts, rx_pkts;
145 struct virtio_dev_state * virtio_dev;
146 struct list_head dev_link;
150 static int virtio_init_state(struct virtio_net_state * virtio)
152 virtio->rx_vq.queue_size = RX_QUEUE_SIZE;
153 virtio->tx_vq.queue_size = TX_QUEUE_SIZE;
154 virtio->ctrl_vq.queue_size = CTRL_QUEUE_SIZE;
156 virtio->rx_vq.ring_desc_addr = 0;
157 virtio->rx_vq.ring_avail_addr = 0;
158 virtio->rx_vq.ring_used_addr = 0;
159 virtio->rx_vq.pfn = 0;
160 virtio->rx_vq.cur_avail_idx = 0;
162 virtio->tx_vq.ring_desc_addr = 0;
163 virtio->tx_vq.ring_avail_addr = 0;
164 virtio->tx_vq.ring_used_addr = 0;
165 virtio->tx_vq.pfn = 0;
166 virtio->tx_vq.cur_avail_idx = 0;
168 virtio->ctrl_vq.ring_desc_addr = 0;
169 virtio->ctrl_vq.ring_avail_addr = 0;
170 virtio->ctrl_vq.ring_used_addr = 0;
171 virtio->ctrl_vq.pfn = 0;
172 virtio->ctrl_vq.cur_avail_idx = 0;
174 virtio->virtio_cfg.pci_isr = 0;
176 virtio->mergeable_rx_bufs = 1;
178 virtio->virtio_cfg.host_features = 0 | (1 << VIRTIO_NET_F_MAC);
179 if(virtio->mergeable_rx_bufs) {
180 virtio->virtio_cfg.host_features |= (1 << VIRTIO_NET_F_MRG_RXBUF);
183 if ((v3_lock_init(&(virtio->rx_lock)) == -1) ||
184 (v3_lock_init(&(virtio->tx_lock)) == -1)){
185 PrintError("Virtio NIC: Failure to init locks for net_state\n");
191 static int tx_one_pkt(struct guest_info * core,
192 struct virtio_net_state * virtio,
193 struct vring_desc * buf_desc)
195 uint8_t * buf = NULL;
196 uint32_t len = buf_desc->length;
198 if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
199 PrintDebug("Could not translate buffer address\n");
203 V3_Net_Print(2, "Virtio-NIC: virtio_tx: size: %d\n", len);
205 v3_hexdump(buf, len, NULL, 0);
208 if(virtio->net_ops->send(buf, len, virtio->backend_data) < 0){
209 virtio->stats.tx_dropped ++;
213 virtio->stats.tx_pkts ++;
214 virtio->stats.tx_bytes += len;
220 /*copy data into ring buffer */
221 static inline int copy_data_to_desc(struct guest_info * core,
222 struct virtio_net_state * virtio_state,
223 struct vring_desc * desc,
228 uint8_t * desc_buf = NULL;
230 if (v3_gpa_to_hva(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
231 PrintDebug("Could not translate buffer address\n");
234 len = (desc->length < (buf_len+dst_offset))?(desc->length - dst_offset):buf_len;
235 memcpy(desc_buf + dst_offset, buf, len);
241 static inline int get_desc_count(struct virtio_queue * q, int index) {
242 struct vring_desc * tmp_desc = &(q->desc[index]);
245 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
246 tmp_desc = &(q->desc[tmp_desc->next]);
253 static inline void enable_cb(struct virtio_queue *queue){
255 queue->used->flags &= ~ VRING_NO_NOTIFY_FLAG;
259 static inline void disable_cb(struct virtio_queue *queue) {
261 queue->used->flags |= VRING_NO_NOTIFY_FLAG;
265 static int handle_pkt_tx(struct guest_info * core,
266 struct virtio_net_state * virtio_state,
269 struct virtio_queue *q = &(virtio_state->tx_vq);
270 int txed = 0, left = 0;
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: Fails to send packet\n");
306 if(buf_desc->next & VIRTIO_NEXT_FLAG){
307 PrintError("Virtio NIC: TX more buffer need to read\n");
310 q->used->ring[q->used->index % q->queue_size].id =
311 q->avail->ring[q->cur_avail_idx % q->queue_size];
313 q->used->ring[q->used->index % q->queue_size].length =
314 buf_desc->length; /* What do we set this to???? */
319 if(++txed >= quote && quote > 0){
320 left = (q->cur_avail_idx != q->avail->index);
325 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
327 if (txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
328 v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus,
329 0, virtio_state->pci_dev);
330 virtio_state->virtio_cfg.pci_isr = 0x1;
331 virtio_state->stats.rx_interrupts ++;
335 V3_Net_Print(2, "Virtio Handle TX: txed pkts: %d, left %d\n", txed, left);
342 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
347 static int virtio_setup_queue(struct guest_info *core,
348 struct virtio_net_state * virtio_state,
349 struct virtio_queue * queue,
350 addr_t pfn, addr_t page_addr) {
353 queue->ring_desc_addr = page_addr;
354 queue->ring_avail_addr = page_addr + (queue->queue_size * sizeof(struct vring_desc));
355 queue->ring_used_addr = ((queue->ring_avail_addr) +
356 (sizeof(struct vring_avail)) +
357 (queue->queue_size * sizeof(uint16_t)));
359 // round up to next page boundary.
360 queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
361 if (v3_gpa_to_hva(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
362 PrintError("Could not translate ring descriptor address\n");
366 if (v3_gpa_to_hva(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
367 PrintError("Could not translate ring available address\n");
371 if (v3_gpa_to_hva(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
372 PrintError("Could not translate ring used address\n");
376 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
377 (void *)(queue->ring_desc_addr),
378 (void *)(queue->ring_avail_addr),
379 (void *)(queue->ring_used_addr));
381 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
382 queue->desc, queue->avail, queue->used);
387 static int virtio_io_write(struct guest_info *core,
388 uint16_t port, void * src,
389 uint_t length, void * private_data)
391 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
392 int port_idx = port % virtio->io_range_size;
394 PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n",
395 private_data, port, port_idx,
396 length, *(uint32_t *)src);
399 case GUEST_FEATURES_PORT:
401 PrintError("Illegal write length for guest features\n");
404 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
407 case VRING_PG_NUM_PORT:
409 PrintError("Illegal write length for page frame number\n");
412 addr_t pfn = *(uint32_t *)src;
413 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
414 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
417 virtio_setup_queue(core, virtio,
422 virtio_setup_queue(core, virtio,
425 if(virtio->tx_notify == 0){
426 disable_cb(&virtio->tx_vq);
431 virtio_setup_queue(core, virtio,
440 case VRING_Q_SEL_PORT:
441 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
442 if (virtio->virtio_cfg.vring_queue_selector > 2) {
443 PrintError("Virtio NIC: wrong queue idx: %d\n",
444 virtio->virtio_cfg.vring_queue_selector);
449 case VRING_Q_NOTIFY_PORT:
451 uint16_t queue_idx = *(uint16_t *)src;
453 /* receive queue refill */
454 virtio->stats.tx_interrupts ++;
455 } else if (queue_idx == 1){
456 if (handle_pkt_tx(core, virtio, 0) < 0) {
457 PrintError("Virtio NIC: Error to handle packet TX\n");
460 virtio->stats.tx_interrupts ++;
461 } else if (queue_idx == 2){
464 PrintError("Virtio NIC: Wrong queue index %d\n", queue_idx);
469 case VIRTIO_STATUS_PORT:
470 virtio->virtio_cfg.status = *(uint8_t *)src;
471 if (virtio->virtio_cfg.status == 0) {
472 virtio_init_state(virtio);
476 case VIRTIO_ISR_PORT:
477 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
488 static int virtio_io_read(struct guest_info *core,
489 uint16_t port, void * dst,
490 uint_t length, void * private_data)
492 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
493 int port_idx = port % virtio->io_range_size;
494 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
496 PrintDebug("Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n",
497 private_data, port, port_idx, length);
500 case HOST_FEATURES_PORT:
502 PrintError("Virtio NIC: Illegal read length for host features\n");
505 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
508 case VRING_PG_NUM_PORT:
510 PrintError("Virtio NIC: Illegal read length for page frame number\n");
515 *(uint32_t *)dst = virtio->rx_vq.pfn;
518 *(uint32_t *)dst = virtio->tx_vq.pfn;
521 *(uint32_t *)dst = virtio->ctrl_vq.pfn;
528 case VRING_SIZE_PORT:
530 PrintError("Virtio NIC: Illegal read length for vring size\n");
535 *(uint16_t *)dst = virtio->rx_vq.queue_size;
538 *(uint16_t *)dst = virtio->tx_vq.queue_size;
541 *(uint16_t *)dst = virtio->ctrl_vq.queue_size;
548 case VIRTIO_STATUS_PORT:
550 PrintError("Virtio NIC: Illegal read length for status\n");
553 *(uint8_t *)dst = virtio->virtio_cfg.status;
556 case VIRTIO_ISR_PORT:
557 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
558 virtio->virtio_cfg.pci_isr = 0;
559 v3_pci_lower_irq(virtio->virtio_dev->pci_bus,
563 case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN:
564 *(uint8_t *)dst = virtio->net_cfg.mac[port_idx-VIRTIO_NET_CONFIG];
568 PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n",
577 /* receiving raw ethernet pkt from backend */
578 static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) {
579 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
580 struct virtio_queue * q = &(virtio->rx_vq);
581 struct virtio_net_hdr_mrg_rxbuf hdr;
583 uint8_t kick_guest = 0;
585 V3_Net_Print(2, "Virtio NIC: virtio_rx: size: %d\n", size);
587 if (!q->ring_avail_addr) {
588 V3_Net_Print(2, "Virtio NIC: RX Queue not set\n");
589 virtio->stats.rx_dropped ++;
594 memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
596 flags = v3_lock_irqsave(virtio->rx_lock);
598 if (q->cur_avail_idx != q->avail->index){
600 struct vring_desc * buf_desc;
601 uint32_t hdr_len, len;
604 hdr_len = (virtio->mergeable_rx_bufs)?
605 sizeof(struct virtio_net_hdr_mrg_rxbuf):
606 sizeof(struct virtio_net_hdr);
608 if(virtio->mergeable_rx_bufs){/* merged buffer */
609 struct vring_desc * hdr_desc;
610 uint16_t old_idx = q->cur_avail_idx;
612 buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
613 hdr_desc = &(q->desc[buf_idx]);
615 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
616 virtio, hdr_desc, buf, size, hdr_len);
622 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
623 q->used->ring[q->used->index % q->queue_size].length = hdr_len + offset;
627 while(offset < size) {
628 buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
629 buf_desc = &(q->desc[buf_idx]);
631 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
632 virtio, buf_desc, buf+offset, size-offset, 0);
634 V3_Net_Print(2, "Virtio NIC: merged buffer, %d buffer size %d\n",
635 hdr.num_buffers, len);
636 q->cur_avail_idx = old_idx;
640 buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
642 q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
643 q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].length = len;
649 copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
650 virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0);
651 q->used->index += hdr.num_buffers;
653 buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
654 buf_desc = &(q->desc[buf_idx]);
657 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
658 virtio, buf_desc, (uchar_t *)&(hdr.hdr), hdr_len, 0);
660 V3_Net_Print(2, "Virtio NIC: rx copy header error %d, hdr_len %d\n",
665 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
666 virtio, buf_desc, buf, size, hdr_len);
668 V3_Net_Print(2, "Virtio NIC: rx copy data error %d\n", len);
673 /* copy rest of data */
674 while(offset < size &&
675 (buf_desc->flags & VIRTIO_NEXT_FLAG)){
676 buf_desc = &(q->desc[buf_desc->next]);
677 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
678 virtio, buf_desc, buf+offset, size-offset, 0);
684 buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
687 V3_Net_Print(2, "Virtio NIC: rx not enough ring buffer, buffer size %d\n",
692 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
693 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) */
698 virtio->stats.rx_pkts ++;
699 virtio->stats.rx_bytes += size;
701 V3_Net_Print(2, "Virtio NIC: Guest RX queue is full\n");
702 virtio->stats.rx_dropped ++;
704 /* kick guest to refill RX queue */
708 v3_unlock_irqrestore(virtio->rx_lock, flags);
710 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG) || kick_guest) {
711 V3_Net_Print(2, "Virtio NIC: RX Raising IRQ %d\n",
712 virtio->pci_dev->config_header.intr_line);
714 virtio->virtio_cfg.pci_isr = 0x1;
715 v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
716 virtio->stats.rx_interrupts ++;
719 /* notify guest if it is in guest mode */
720 if((kick_guest || virtio->rx_notify == 1) &&
721 V3_Get_CPU() != virtio->virtio_dev->vm->cores[0].pcpu_id){
722 v3_interrupt_cpu(virtio->virtio_dev->vm,
723 virtio->virtio_dev->vm->cores[0].pcpu_id,
730 virtio->stats.rx_dropped ++;
731 v3_unlock_irqrestore(virtio->rx_lock, flags);
736 static int virtio_free(struct virtio_dev_state * virtio) {
737 struct virtio_net_state * backend = NULL;
738 struct virtio_net_state * tmp = NULL;
741 list_for_each_entry_safe(backend, tmp, &(virtio->dev_list), dev_link) {
743 // unregister from PCI
745 list_del(&(backend->dev_link));
754 static struct v3_device_ops dev_ops = {
755 .free = (int (*)(void *))virtio_free,
759 static int virtio_poll(int quote, void * data){
760 struct virtio_net_state * virtio = (struct virtio_net_state *)data;
762 if (virtio->status) {
764 return handle_pkt_tx(&(virtio->vm->cores[0]), virtio, quote);
770 static int register_dev(struct virtio_dev_state * virtio,
771 struct virtio_net_state * net_state)
773 struct pci_device * pci_dev = NULL;
774 struct v3_pci_bar bars[6];
775 int num_ports = sizeof(struct virtio_config);
776 int tmp_ports = num_ports;
779 /* This gets the number of ports, rounded up to a power of 2 */
780 net_state->io_range_size = 1;
781 while (tmp_ports > 0) {
783 net_state->io_range_size <<= 1;
786 /* this is to account for any low order bits being set in num_ports
787 * if there are none, then num_ports was already a power of 2 so we shift right to reset it
789 if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
790 net_state->io_range_size >>= 1;
793 for (i = 0; i < 6; i++) {
794 bars[i].type = PCI_BAR_NONE;
797 PrintDebug("Virtio NIC: io_range_size = %d\n",
798 net_state->io_range_size);
800 bars[0].type = PCI_BAR_IO;
801 bars[0].default_base_port = -1;
802 bars[0].num_ports = net_state->io_range_size;
803 bars[0].io_read = virtio_io_read;
804 bars[0].io_write = virtio_io_write;
805 bars[0].private_data = net_state;
807 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
808 0, PCI_AUTO_DEV_NUM, 0,
809 "LNX_VIRTIO_NIC", bars,
810 NULL, NULL, NULL, net_state);
813 PrintError("Virtio NIC: Could not register PCI Device\n");
817 PrintDebug("Virtio NIC: registered to PCI bus\n");
819 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
820 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
823 pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
824 pci_dev->config_header.class = PCI_CLASS_NETWORK;
825 pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
826 pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;
827 pci_dev->config_header.intr_pin = 1;
828 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
830 net_state->pci_dev = pci_dev;
831 net_state->virtio_dev = virtio;
833 memcpy(net_state->net_cfg.mac, virtio->mac, 6);
835 virtio_init_state(net_state);
837 V3_Print("Virtio NIC: Registered Intr Line %d\n", pci_dev->config_header.intr_line);
839 /* Add backend to list of devices */
840 list_add(&(net_state->dev_link), &(virtio->dev_list));
845 #define RATE_UPPER_THRESHOLD 10 /* 10000 pkts per second, around 100Mbits */
846 #define RATE_LOWER_THRESHOLD 1
847 #define PROFILE_PERIOD 10000 /*us*/
849 static void virtio_nic_timer(struct guest_info * core,
850 uint64_t cpu_cycles, uint64_t cpu_freq,
852 struct virtio_net_state * net_state = (struct virtio_net_state *)priv_data;
854 static int profile_ms = 0;
856 if(!net_state->status){ /* VNIC is not in working status */
860 period_us = (1000*cpu_cycles)/cpu_freq;
861 net_state->past_us += period_us;
863 if(net_state->past_us > PROFILE_PERIOD){
864 uint32_t tx_rate, rx_rate;
866 tx_rate = (net_state->stats.tx_pkts - net_state->tx_pkts)/(net_state->past_us/1000); /* pkts/per ms */
867 rx_rate = (net_state->stats.rx_pkts - net_state->rx_pkts)/(net_state->past_us/1000);
869 net_state->tx_pkts = net_state->stats.tx_pkts;
870 net_state->rx_pkts = net_state->stats.rx_pkts;
872 if(tx_rate > RATE_UPPER_THRESHOLD && net_state->tx_notify == 1){
873 V3_Print("Virtio NIC: Switch TX to VMM driven mode\n");
874 disable_cb(&(net_state->tx_vq));
875 net_state->tx_notify = 0;
878 if(tx_rate < RATE_LOWER_THRESHOLD && net_state->tx_notify == 0){
879 V3_Print("Virtio NIC: Switch TX to Guest driven mode\n");
880 enable_cb(&(net_state->tx_vq));
881 net_state->tx_notify = 1;
884 if(rx_rate > RATE_UPPER_THRESHOLD && net_state->rx_notify == 1){
885 V3_Print("Virtio NIC: Switch RX to VMM None notify mode\n");
886 net_state->rx_notify = 0;
889 if(rx_rate < RATE_LOWER_THRESHOLD && net_state->rx_notify == 0){
890 V3_Print("Virtio NIC: Switch RX to VMM notify mode\n");
891 net_state->rx_notify = 1;
894 net_state->past_us = 0;
897 profile_ms += period_us/1000;
898 if(profile_ms > 20000){
899 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",
900 net_state->stats.tx_pkts, net_state->stats.tx_bytes,
901 net_state->stats.rx_pkts, net_state->stats.rx_bytes,
902 net_state->stats.tx_dropped, net_state->stats.rx_dropped,
903 net_state->stats.tx_interrupts, net_state->stats.rx_interrupts,
904 net_state->vm->cores[0].num_exits);
909 static struct v3_timer_ops timer_ops = {
910 .update_timer = virtio_nic_timer,
914 static int connect_fn(struct v3_vm_info * info,
915 void * frontend_data,
916 struct v3_dev_net_ops * ops,
918 void * private_data) {
919 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
920 struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
922 memset(net_state, 0, sizeof(struct virtio_net_state));
923 register_dev(virtio, net_state);
925 net_state->vm = info;
926 net_state->net_ops = ops;
927 net_state->backend_data = private_data;
928 net_state->virtio_dev = virtio;
930 net_state->tx_notify = 1;
931 net_state->rx_notify = 1;
933 net_state->timer = v3_add_timer(&(info->cores[0]),
934 &timer_ops,net_state);
936 ops->recv = virtio_rx;
937 ops->poll = virtio_poll;
938 ops->config.frontend_data = net_state;
939 ops->config.poll = 1;
940 ops->config.quote = 64;
941 ops->config.fnt_mac = V3_Malloc(ETH_ALEN);
942 memcpy(ops->config.fnt_mac, virtio->mac, ETH_ALEN);
947 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
948 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
949 struct virtio_dev_state * virtio_state = NULL;
950 char * dev_id = v3_cfg_val(cfg, "ID");
952 char * str = v3_cfg_val(cfg, "mac");
953 memcpy(macstr, str, strlen(str));
955 if (pci_bus == NULL) {
956 PrintError("Virtio NIC: Virtio device require a PCI Bus");
960 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
961 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
963 INIT_LIST_HEAD(&(virtio_state->dev_list));
964 virtio_state->pci_bus = pci_bus;
965 virtio_state->vm = vm;
967 if (macstr != NULL && !str2mac(macstr, virtio_state->mac)) {
968 PrintDebug("Virtio NIC: Mac specified %s\n", macstr);
970 random_ethaddr(virtio_state->mac);
973 struct vm_device * dev = v3_add_device(vm, dev_id, &dev_ops, virtio_state);
976 PrintError("Virtio NIC: Could not attach device %s\n", dev_id);
977 V3_Free(virtio_state);
981 if (v3_dev_add_net_frontend(vm, dev_id, connect_fn, (void *)virtio_state) == -1) {
982 PrintError("Virtio NIC: Could not register %s as net frontend\n", dev_id);
983 v3_remove_device(dev);
990 device_register("LNX_VIRTIO_NIC", virtio_init)