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 #define TX_QUEUE_SIZE 4096
42 #define RX_QUEUE_SIZE 4096
43 #define CTRL_QUEUE_SIZE 64
45 /* The feature bitmap for virtio nic
47 #define VIRTIO_NET_F_CSUM 0 /* Host handles pkts w/ partial csum */
48 #define VIRTIO_NET_F_GUEST_CSUM 1 /* Guest handles pkts w/ partial csum */
49 #define VIRTIO_NET_F_MAC 5 /* Host has given MAC address. */
50 #define VIRTIO_NET_F_GSO 6 /* Host handles pkts w/ any GSO type */
51 #define VIRTIO_NET_F_GUEST_TSO4 7 /* Guest can handle TSOv4 in. */
52 #define VIRTIO_NET_F_GUEST_TSO6 8 /* Guest can handle TSOv6 in. */
53 #define VIRTIO_NET_F_GUEST_ECN 9 /* Guest can handle TSO[6] w/ ECN in. */
54 #define VIRTIO_NET_F_GUEST_UFO 10 /* Guest can handle UFO in. */
55 #define VIRTIO_NET_F_HOST_TSO4 11 /* Host can handle TSOv4 in. */
56 #define VIRTIO_NET_F_HOST_TSO6 12 /* Host can handle TSOv6 in. */
57 #define VIRTIO_NET_F_HOST_ECN 13 /* Host can handle TSO[6] w/ ECN in. */
58 #define VIRTIO_NET_F_HOST_UFO 14 /* Host can handle UFO in. */
59 #define VIRTIO_NET_F_MRG_RXBUF 15 /* Host can merge receive buffers. */
60 #define VIRTIO_NET_F_STATUS 16 /* virtio_net_config.status available */
62 /* Port to get virtio config */
63 #define VIRTIO_NET_CONFIG 20
65 #define VIRTIO_NET_MAX_BUFSIZE (sizeof(struct virtio_net_hdr) + (64 << 10))
67 /* for gso_type in virtio_net_hdr */
68 #define VIRTIO_NET_HDR_GSO_NONE 0
69 #define VIRTIO_NET_HDR_GSO_TCPV4 1 /* GSO frame, IPv4 TCP (TSO) */
70 #define VIRTIO_NET_HDR_GSO_UDP 3 /* GSO frame, IPv4 UDP (UFO) */
71 #define VIRTIO_NET_HDR_GSO_TCPV6 4 /* GSO frame, IPv6 TCP */
72 #define VIRTIO_NET_HDR_GSO_ECN 0x80 /* TCP has ECN set */
75 /* for flags in virtio_net_hdr */
76 #define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /* Use csum_start, csum_offset */
79 /* First element of the scatter-gather list, used with GSO or CSUM features */
84 uint16_t hdr_len; /* Ethernet + IP + tcp/udp hdrs */
85 uint16_t gso_size; /* Bytes to append to hdr_len per frame */
86 uint16_t csum_start; /* Position to start checksumming from */
87 uint16_t csum_offset; /* Offset after that to place checksum */
88 }__attribute__((packed));
91 /* The header to use when the MRG_RXBUF
92 * feature has been negotiated. */
93 struct virtio_net_hdr_mrg_rxbuf {
94 struct virtio_net_hdr hdr;
95 uint16_t num_buffers; /* Number of merged rx buffers */
98 struct virtio_net_config
100 uint8_t mac[ETH_ALEN]; /* VIRTIO_NET_F_MAC */
102 } __attribute__((packed));
104 struct virtio_dev_state {
105 struct vm_device * pci_bus;
106 struct list_head dev_list;
107 struct v3_vm_info *vm;
109 uint8_t mac[ETH_ALEN];
112 struct virtio_net_state {
113 struct virtio_net_config net_cfg;
114 struct virtio_config virtio_cfg;
116 struct v3_vm_info * vm;
117 struct vm_device * dev;
118 struct pci_device * pci_dev;
123 struct virtio_queue rx_vq; /* idx 0*/
124 struct virtio_queue tx_vq; /* idx 1*/
125 struct virtio_queue ctrl_vq; /* idx 2*/
127 uint8_t mergeable_rx_bufs;
129 struct v3_timer * timer;
130 struct vnet_thread * poll_thread;
132 struct nic_statistics stats;
134 struct v3_dev_net_ops * net_ops;
135 v3_lock_t rx_lock, tx_lock;
137 uint8_t tx_notify, rx_notify;
138 uint32_t tx_pkts, rx_pkts;
142 struct virtio_dev_state * virtio_dev;
143 struct list_head dev_link;
147 static int virtio_init_state(struct virtio_net_state * virtio)
149 virtio->rx_vq.queue_size = RX_QUEUE_SIZE;
150 virtio->tx_vq.queue_size = TX_QUEUE_SIZE;
151 virtio->ctrl_vq.queue_size = CTRL_QUEUE_SIZE;
153 virtio->rx_vq.ring_desc_addr = 0;
154 virtio->rx_vq.ring_avail_addr = 0;
155 virtio->rx_vq.ring_used_addr = 0;
156 virtio->rx_vq.pfn = 0;
157 virtio->rx_vq.cur_avail_idx = 0;
159 virtio->tx_vq.ring_desc_addr = 0;
160 virtio->tx_vq.ring_avail_addr = 0;
161 virtio->tx_vq.ring_used_addr = 0;
162 virtio->tx_vq.pfn = 0;
163 virtio->tx_vq.cur_avail_idx = 0;
165 virtio->ctrl_vq.ring_desc_addr = 0;
166 virtio->ctrl_vq.ring_avail_addr = 0;
167 virtio->ctrl_vq.ring_used_addr = 0;
168 virtio->ctrl_vq.pfn = 0;
169 virtio->ctrl_vq.cur_avail_idx = 0;
171 virtio->virtio_cfg.pci_isr = 0;
173 virtio->mergeable_rx_bufs = 0;
175 virtio->virtio_cfg.host_features = 0 | (1 << VIRTIO_NET_F_MAC);
176 if(virtio->mergeable_rx_bufs) {
177 virtio->virtio_cfg.host_features |= (1 << VIRTIO_NET_F_MRG_RXBUF);
180 if ((v3_lock_init(&(virtio->rx_lock)) == -1) ||
181 (v3_lock_init(&(virtio->tx_lock)) == -1)){
182 PrintError("Virtio NIC: Failure to init locks for net_state\n");
188 static int tx_one_pkt(struct guest_info * core,
189 struct virtio_net_state * virtio,
190 struct vring_desc * buf_desc)
192 uint8_t * buf = NULL;
193 uint32_t len = buf_desc->length;
194 int synchronize = virtio->tx_notify;
196 if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
197 PrintDebug("Could not translate buffer address\n");
201 V3_Net_Print(2, "Virtio-NIC: virtio_tx: size: %d\n", len);
202 if(v3_net_debug >= 4){
203 v3_hexdump(buf, len, NULL, 0);
206 if(virtio->net_ops->send(buf, len, synchronize, virtio->backend_data) < 0){
207 virtio->stats.tx_dropped ++;
211 virtio->stats.tx_pkts ++;
212 virtio->stats.tx_bytes += len;
218 static inline int copy_data_to_desc(struct guest_info * core,
219 struct virtio_net_state * virtio_state,
220 struct vring_desc * desc,
225 uint8_t * desc_buf = NULL;
227 if (v3_gpa_to_hva(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
228 PrintDebug("Could not translate buffer address\n");
231 len = (desc->length < buf_len)?(desc->length - dst_offset):buf_len;
232 memcpy(desc_buf+dst_offset, buf, len);
238 static inline int get_desc_count(struct virtio_queue * q, int index) {
239 struct vring_desc * tmp_desc = &(q->desc[index]);
242 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
243 tmp_desc = &(q->desc[tmp_desc->next]);
250 static inline void enable_cb(struct virtio_queue *queue){
251 queue->used->flags &= ~ VRING_NO_NOTIFY_FLAG;
254 static inline void disable_cb(struct virtio_queue *queue) {
255 queue->used->flags |= VRING_NO_NOTIFY_FLAG;
258 static int handle_pkt_tx(struct guest_info * core,
259 struct virtio_net_state * virtio_state)
261 struct virtio_queue *q = &(virtio_state->tx_vq);
265 if (!q->ring_avail_addr) {
269 flags = v3_lock_irqsave(virtio_state->tx_lock);
270 while (q->cur_avail_idx != q->avail->index) {
271 struct virtio_net_hdr_mrg_rxbuf * hdr = NULL;
272 struct vring_desc * hdr_desc = NULL;
274 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
275 int desc_cnt = get_desc_count(q, desc_idx);
278 PrintError("VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt);
282 hdr_desc = &(q->desc[desc_idx]);
283 if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
284 PrintError("Could not translate block header address\n");
288 hdr = (struct virtio_net_hdr_mrg_rxbuf *)hdr_addr;
289 desc_idx = hdr_desc->next;
291 V3_Net_Print(2, "Virtio NIC: TX hdr count : %d\n", hdr->num_buffers);
293 /* here we assumed that one ethernet pkt is not splitted into multiple buffer */
294 struct vring_desc * buf_desc = &(q->desc[desc_idx]);
295 if (tx_one_pkt(core, virtio_state, buf_desc) == -1) {
296 PrintError("Virtio NIC: Error handling nic operation\n");
299 if(buf_desc->next & VIRTIO_NEXT_FLAG){
300 V3_Net_Print(2, "Virtio NIC: TX more buffer need to read\n");
303 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
304 q->used->ring[q->used->index % q->queue_size].length = buf_desc->length; /* What do we set this to???? */
312 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
314 if (txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
315 v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev);
316 virtio_state->virtio_cfg.pci_isr = 0x1;
318 virtio_state->stats.rx_interrupts ++;
322 V3_Net_Print(2, "Virtio Handle TX: txed pkts: %d\n", txed);
329 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
334 static int virtio_setup_queue(struct guest_info *core,
335 struct virtio_net_state * virtio_state,
336 struct virtio_queue * queue,
337 addr_t pfn, addr_t page_addr) {
340 queue->ring_desc_addr = page_addr;
341 queue->ring_avail_addr = page_addr + (queue->queue_size * sizeof(struct vring_desc));
342 queue->ring_used_addr = ((queue->ring_avail_addr) +
343 (sizeof(struct vring_avail)) +
344 (queue->queue_size * sizeof(uint16_t)));
346 // round up to next page boundary.
347 queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
348 if (v3_gpa_to_hva(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
349 PrintError("Could not translate ring descriptor address\n");
353 if (v3_gpa_to_hva(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
354 PrintError("Could not translate ring available address\n");
358 if (v3_gpa_to_hva(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
359 PrintError("Could not translate ring used address\n");
363 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
364 (void *)(queue->ring_desc_addr),
365 (void *)(queue->ring_avail_addr),
366 (void *)(queue->ring_used_addr));
368 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
369 queue->desc, queue->avail, queue->used);
374 static int virtio_io_write(struct guest_info *core,
375 uint16_t port, void * src,
376 uint_t length, void * private_data)
378 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
379 int port_idx = port % virtio->io_range_size;
381 PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n", private_data,
382 port, port_idx, length, *(uint32_t *)src);
385 case GUEST_FEATURES_PORT:
387 PrintError("Illegal write length for guest features\n");
390 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
393 case VRING_PG_NUM_PORT:
395 PrintError("Illegal write length for page frame number\n");
398 addr_t pfn = *(uint32_t *)src;
399 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
400 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
403 virtio_setup_queue(core, virtio, &virtio->rx_vq, pfn, page_addr);
406 virtio_setup_queue(core, virtio, &virtio->tx_vq, pfn, page_addr);
407 if(virtio->tx_notify == 0){
408 disable_cb(&virtio->tx_vq);
409 vnet_thread_wakeup(virtio->poll_thread);
413 virtio_setup_queue(core, virtio, &virtio->ctrl_vq, pfn, page_addr);
420 case VRING_Q_SEL_PORT:
421 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
422 if (virtio->virtio_cfg.vring_queue_selector > 2) {
423 PrintError("Virtio NIC: wrong queue idx: %d\n",
424 virtio->virtio_cfg.vring_queue_selector);
429 case VRING_Q_NOTIFY_PORT:
431 uint16_t queue_idx = *(uint16_t *)src;
433 /* receive queue refill */
434 virtio->stats.tx_interrupts ++;
435 } else if (queue_idx == 1){
436 if (handle_pkt_tx(core, virtio) == -1) {
437 PrintError("Could not handle Virtio NIC tx kick\n");
440 virtio->stats.tx_interrupts ++;
441 } else if (queue_idx == 2){
444 PrintError("Wrong queue index %d\n", queue_idx);
449 case VIRTIO_STATUS_PORT:
450 virtio->virtio_cfg.status = *(uint8_t *)src;
451 if (virtio->virtio_cfg.status == 0) {
452 virtio_init_state(virtio);
456 case VIRTIO_ISR_PORT:
457 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
468 static int virtio_io_read(struct guest_info *core,
469 uint16_t port, void * dst,
470 uint_t length, void * private_data)
472 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
473 int port_idx = port % virtio->io_range_size;
474 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
476 PrintDebug("Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n", private_data,
477 port, port_idx, length);
480 case HOST_FEATURES_PORT:
482 PrintError("Illegal read length for host features\n");
485 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
488 case VRING_PG_NUM_PORT:
490 PrintError("Illegal read length for page frame number\n");
495 *(uint32_t *)dst = virtio->rx_vq.pfn;
498 *(uint32_t *)dst = virtio->tx_vq.pfn;
501 *(uint32_t *)dst = virtio->ctrl_vq.pfn;
508 case VRING_SIZE_PORT:
510 PrintError("Illegal read length for vring size\n");
515 *(uint16_t *)dst = virtio->rx_vq.queue_size;
518 *(uint16_t *)dst = virtio->tx_vq.queue_size;
521 *(uint16_t *)dst = virtio->ctrl_vq.queue_size;
528 case VIRTIO_STATUS_PORT:
530 PrintError("Illegal read length for status\n");
533 *(uint8_t *)dst = virtio->virtio_cfg.status;
536 case VIRTIO_ISR_PORT:
537 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
538 virtio->virtio_cfg.pci_isr = 0;
539 v3_pci_lower_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
542 case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN:
543 *(uint8_t *)dst = virtio->net_cfg.mac[port_idx-VIRTIO_NET_CONFIG];
547 PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n", port_idx);
555 /* receiving raw ethernet pkt from backend */
556 static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) {
557 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
558 struct virtio_queue * q = &(virtio->rx_vq);
559 struct virtio_net_hdr_mrg_rxbuf hdr;
563 V3_Net_Print(2, "Virtio-NIC: virtio_rx: size: %d\n", size);
564 if(v3_net_debug >= 4){
565 v3_hexdump(buf, size, NULL, 0);
568 flags = v3_lock_irqsave(virtio->rx_lock);
571 memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
573 if (q->ring_avail_addr == 0) {
574 V3_Net_Print(2, "Virtio NIC: RX Queue not set\n");
575 virtio->stats.rx_dropped ++;
579 if (q->cur_avail_idx != q->avail->index){
581 uint16_t buf_idx = 0;
582 uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
583 struct vring_desc * hdr_desc = NULL;
584 struct vring_desc * buf_desc = NULL;
585 uint32_t hdr_len = 0;
588 hdr_desc = &(q->desc[hdr_idx]);
589 if (v3_gpa_to_hva(&(virtio->virtio_dev->vm->cores[0]), hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
590 V3_Net_Print(2, "Virtio NIC: Could not translate receive buffer address\n");
591 virtio->stats.rx_dropped ++;
595 hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
597 if(virtio->mergeable_rx_bufs){/* merged buffer */
602 hdr_desc = &(q->desc[buf_idx]);
603 hdr_desc->flags &= ~VIRTIO_NEXT_FLAG;
605 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, buf, data_len, hdr_len);
609 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
610 q->used->ring[q->used->index % q->queue_size].length = hdr_len + len;
613 while(offset < data_len) {
614 buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
615 buf_desc = &(q->desc[buf_idx]);
617 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf + offset, data_len - offset, 0);
619 V3_Net_Print(2, "Virtio NIC:merged buffer, %d buffer size %d\n", hdr.num_buffers, data_len);
620 virtio->stats.rx_dropped ++;
624 buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
627 q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
628 q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].length = len;
631 q->used->index += hdr.num_buffers;
632 copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0);
634 hdr_desc = &(q->desc[buf_idx]);
635 copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0);
637 buf_idx = hdr_desc->next;
638 buf_desc = &(q->desc[buf_idx]);
639 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf, data_len, 0);
640 if (len < data_len) {
641 V3_Net_Print(2, "Virtio NIC: ring buffer len less than pkt size, merged buffer not supported, buffer size %d\n", len);
642 virtio->stats.rx_dropped ++;
646 buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
648 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
649 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) */
654 virtio->stats.rx_pkts ++;
655 virtio->stats.rx_bytes += size;
657 V3_Net_Print(2, "Virtio NIC: Guest RX queue is full\n");
658 virtio->stats.rx_dropped ++;
660 /* kick guest to refill the queue */
661 virtio->virtio_cfg.pci_isr = 0x1;
662 v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
663 v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].cpu_id, 0);
664 virtio->stats.rx_interrupts ++;
669 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
670 V3_Net_Print(2, "Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line);
672 virtio->virtio_cfg.pci_isr = 0x1;
673 v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
675 virtio->stats.rx_interrupts ++;
678 v3_unlock_irqrestore(virtio->rx_lock, flags);
680 /* notify guest if it is in guest mode */
681 if(virtio->rx_notify == 1 &&
682 V3_Get_CPU() != virtio->virtio_dev->vm->cores[0].cpu_id){
683 v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].cpu_id, 0);
690 v3_unlock_irqrestore(virtio->rx_lock, flags);
695 static int virtio_free(struct virtio_dev_state * virtio) {
696 struct virtio_net_state * backend = NULL;
697 struct virtio_net_state * tmp = NULL;
700 list_for_each_entry_safe(backend, tmp, &(virtio->dev_list), dev_link) {
702 // unregister from PCI
704 list_del(&(backend->dev_link));
713 static struct v3_device_ops dev_ops = {
714 .free = (int (*)(void *))virtio_free,
718 static int virtio_tx_flush(void * args){
719 struct virtio_net_state *virtio = (struct virtio_net_state *)args;
721 V3_Print("Virtio TX Poll Thread Starting for %s\n", virtio->vm->name);
724 if(virtio->tx_notify == 0){
725 handle_pkt_tx(&(virtio->vm->cores[0]), virtio);
728 vnet_thread_sleep(0);
735 static int register_dev(struct virtio_dev_state * virtio,
736 struct virtio_net_state * net_state)
738 struct pci_device * pci_dev = NULL;
739 struct v3_pci_bar bars[6];
740 int num_ports = sizeof(struct virtio_config);
741 int tmp_ports = num_ports;
744 // This gets the number of ports, rounded up to a power of 2
745 net_state->io_range_size = 1; // must be a power of 2
746 while (tmp_ports > 0) {
748 net_state->io_range_size <<= 1;
751 /* this is to account for any low order bits being set in num_ports
752 * if there are none, then num_ports was already a power of 2 so we shift right to reset it
754 if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
755 net_state->io_range_size >>= 1;
758 for (i = 0; i < 6; i++) {
759 bars[i].type = PCI_BAR_NONE;
762 PrintDebug("Virtio-NIC io_range_size = %d\n", net_state->io_range_size);
764 bars[0].type = PCI_BAR_IO;
765 bars[0].default_base_port = -1;
766 bars[0].num_ports = net_state->io_range_size;
767 bars[0].io_read = virtio_io_read;
768 bars[0].io_write = virtio_io_write;
769 bars[0].private_data = net_state;
771 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
772 0, 4/*PCI_AUTO_DEV_NUM*/, 0,
773 "LNX_VIRTIO_NIC", bars,
774 NULL, NULL, NULL, net_state);
777 PrintError("Virtio NIC: Could not register PCI Device\n");
781 PrintDebug("Virtio NIC: registered to PCI bus\n");
783 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
784 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
787 pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
788 pci_dev->config_header.class = PCI_CLASS_NETWORK;
789 pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
790 pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;
791 pci_dev->config_header.intr_pin = 1;
792 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
794 net_state->pci_dev = pci_dev;
795 net_state->virtio_dev = virtio;
797 memcpy(net_state->net_cfg.mac, virtio->mac, 6);
799 virtio_init_state(net_state);
801 /* Add backend to list of devices */
802 list_add(&(net_state->dev_link), &(virtio->dev_list));
807 #define RATE_UPPER_THRESHOLD 10 /* 10000 pkts per second, around 100Mbits */
808 #define RATE_LOWER_THRESHOLD 1
809 #define PROFILE_PERIOD 10000 /*us*/
811 static void virtio_nic_timer(struct guest_info * core,
812 uint64_t cpu_cycles, uint64_t cpu_freq,
814 struct virtio_net_state * net_state = (struct virtio_net_state *)priv_data;
816 static int profile_ms = 0;
818 period_us = (1000*cpu_cycles)/cpu_freq;
819 net_state->past_us += period_us;
821 if(net_state->past_us > PROFILE_PERIOD){
822 uint32_t tx_rate, rx_rate;
824 tx_rate = (net_state->stats.tx_pkts - net_state->tx_pkts)/(net_state->past_us/1000); /* pkts/per ms */
825 rx_rate = (net_state->stats.rx_pkts - net_state->rx_pkts)/(net_state->past_us/1000);
827 net_state->tx_pkts = net_state->stats.tx_pkts;
828 net_state->rx_pkts = net_state->stats.rx_pkts;
830 if(tx_rate > RATE_UPPER_THRESHOLD && net_state->tx_notify == 1){
831 V3_Print("Virtio NIC: Switch TX to VMM driven mode\n");
832 disable_cb(&(net_state->tx_vq));
833 net_state->tx_notify = 0;
834 vnet_thread_wakeup(net_state->poll_thread);
837 if(tx_rate < RATE_LOWER_THRESHOLD && net_state->tx_notify == 0){
838 V3_Print("Virtio NIC: Switch TX to Guest driven mode\n");
839 enable_cb(&(net_state->tx_vq));
840 net_state->tx_notify = 1;
843 if(rx_rate > RATE_UPPER_THRESHOLD && net_state->rx_notify == 1){
844 V3_Print("Virtio NIC: Switch RX to VMM None notify mode\n");
845 net_state->rx_notify = 0;
848 if(rx_rate < RATE_LOWER_THRESHOLD && net_state->rx_notify == 0){
849 V3_Print("Virtio NIC: Switch RX to VMM notify mode\n");
850 net_state->rx_notify = 1;
853 net_state->past_us = 0;
856 profile_ms += period_us/1000;
857 if(profile_ms > 20000){
858 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",
859 net_state->stats.tx_pkts, net_state->stats.tx_bytes,
860 net_state->stats.rx_pkts, net_state->stats.rx_bytes,
861 net_state->stats.tx_dropped, net_state->stats.rx_dropped,
862 net_state->stats.tx_interrupts, net_state->stats.rx_interrupts,
863 net_state->vm->cores[0].num_exits);
868 static struct v3_timer_ops timer_ops = {
869 .update_timer = virtio_nic_timer,
873 static int connect_fn(struct v3_vm_info * info,
874 void * frontend_data,
875 struct v3_dev_net_ops * ops,
877 void * private_data) {
878 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
879 struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
881 memset(net_state, 0, sizeof(struct virtio_net_state));
882 register_dev(virtio, net_state);
884 net_state->vm = info;
885 net_state->net_ops = ops;
886 net_state->backend_data = private_data;
887 net_state->virtio_dev = virtio;
888 net_state->tx_notify = 0;
889 net_state->rx_notify = 0;
891 net_state->timer = v3_add_timer(&(info->cores[0]),&timer_ops,net_state);
893 ops->recv = virtio_rx;
894 ops->frontend_data = net_state;
895 memcpy(ops->fnt_mac, virtio->mac, ETH_ALEN);
897 net_state->poll_thread = vnet_thread_create(virtio_tx_flush, (void *)net_state, "Virtio_Poll");
902 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
903 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
904 struct virtio_dev_state * virtio_state = NULL;
905 char * dev_id = v3_cfg_val(cfg, "ID");
907 char * str = v3_cfg_val(cfg, "mac");
908 memcpy(macstr, str, strlen(str));
910 if (pci_bus == NULL) {
911 PrintError("Virtio NIC: VirtIO devices require a PCI Bus");
915 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
916 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
918 INIT_LIST_HEAD(&(virtio_state->dev_list));
919 virtio_state->pci_bus = pci_bus;
920 virtio_state->vm = vm;
922 if (macstr != NULL && !str2mac(macstr, virtio_state->mac)) {
923 PrintDebug("Virtio NIC: Mac specified %s\n", macstr);
925 random_ethaddr(virtio_state->mac);
928 struct vm_device * dev = v3_add_device(vm, dev_id, &dev_ops, virtio_state);
931 PrintError("Virtio NIC: Could not attach device %s\n", dev_id);
932 V3_Free(virtio_state);
936 if (v3_dev_add_net_frontend(vm, dev_id, connect_fn, (void *)virtio_state) == -1) {
937 PrintError("Virtio NIC: Could not register %s as net frontend\n", dev_id);
938 v3_remove_device(dev);
945 device_register("LNX_VIRTIO_NIC", virtio_init)