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);
202 v3_lock_deinit(&(ns->rx_lock));
203 v3_lock_deinit(&(ns->tx_lock));
209 static int tx_one_pkt(struct guest_info * core,
210 struct virtio_net_state * virtio,
211 struct vring_desc * buf_desc)
213 uint8_t * buf = NULL;
214 uint32_t len = buf_desc->length;
216 if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
217 PrintDebug(core->vm_info, core, "Could not translate buffer address\n");
221 V3_Net_Print(2, "Virtio-NIC: virtio_tx: size: %d\n", len);
223 v3_hexdump(buf, len, NULL, 0);
226 if(virtio->net_ops->send(buf, len, virtio->backend_data) < 0){
227 virtio->stats.tx_dropped ++;
231 virtio->stats.tx_pkts ++;
232 virtio->stats.tx_bytes += len;
238 /*copy data into ring buffer */
239 static inline int copy_data_to_desc(struct guest_info * core,
240 struct virtio_net_state * virtio_state,
241 struct vring_desc * desc,
246 uint8_t * desc_buf = NULL;
248 if (v3_gpa_to_hva(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
249 PrintDebug(core->vm_info, core, "Could not translate buffer address\n");
252 len = (desc->length < (buf_len+dst_offset))?(desc->length - dst_offset):buf_len;
253 memcpy(desc_buf + dst_offset, buf, len);
259 static inline int get_desc_count(struct virtio_queue * q, int index) {
260 struct vring_desc * tmp_desc = &(q->desc[index]);
263 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
264 tmp_desc = &(q->desc[tmp_desc->next]);
271 static inline void enable_cb(struct virtio_queue *queue){
273 queue->used->flags &= ~ VRING_NO_NOTIFY_FLAG;
277 static inline void disable_cb(struct virtio_queue *queue) {
279 queue->used->flags |= VRING_NO_NOTIFY_FLAG;
283 static int handle_pkt_tx(struct guest_info * core,
284 struct virtio_net_state * virtio_state,
287 struct virtio_queue * q;
288 int txed = 0, left = 0;
291 q = &(virtio_state->tx_vq);
292 if (!q->ring_avail_addr) {
297 struct vring_desc * hdr_desc = NULL;
299 uint16_t desc_idx, tmp_idx;
302 flags = v3_lock_irqsave(virtio_state->tx_lock);
304 if(q->cur_avail_idx == q->avail->index ||
305 (quote > 0 && txed >= quote)) {
306 left = (q->cur_avail_idx != q->avail->index);
307 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
311 desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
312 tmp_idx = q->cur_avail_idx ++;
314 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
316 desc_cnt = get_desc_count(q, desc_idx);
318 PrintError(core->vm_info, core, "VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt);
321 hdr_desc = &(q->desc[desc_idx]);
322 if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) != -1) {
323 struct virtio_net_hdr_mrg_rxbuf * hdr;
324 struct vring_desc * buf_desc;
326 hdr = (struct virtio_net_hdr_mrg_rxbuf *)hdr_addr;
327 desc_idx = hdr_desc->next;
329 /* here we assumed that one ethernet pkt is not splitted into multiple buffer */
330 buf_desc = &(q->desc[desc_idx]);
331 if (tx_one_pkt(core, virtio_state, buf_desc) == -1) {
332 PrintError(core->vm_info, core, "Virtio NIC: Fails to send packet\n");
335 PrintError(core->vm_info, core, "Could not translate block header address\n");
338 flags = v3_lock_irqsave(virtio_state->tx_lock);
340 q->used->ring[q->used->index % q->queue_size].id =
341 q->avail->ring[tmp_idx % q->queue_size];
345 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
350 if (txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
351 v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus,
352 virtio_state->pci_dev, 0);
353 virtio_state->virtio_cfg.pci_isr = 0x1;
354 virtio_state->stats.rx_interrupts ++;
361 static int virtio_setup_queue(struct guest_info *core,
362 struct virtio_net_state * virtio_state,
363 struct virtio_queue * queue,
364 addr_t pfn, addr_t page_addr) {
367 queue->ring_desc_addr = page_addr;
368 queue->ring_avail_addr = page_addr + (queue->queue_size * sizeof(struct vring_desc));
369 queue->ring_used_addr = ((queue->ring_avail_addr) +
370 (sizeof(struct vring_avail)) +
371 (queue->queue_size * sizeof(uint16_t)));
373 // round up to next page boundary.
374 queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
375 if (v3_gpa_to_hva(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
376 PrintError(core->vm_info, core, "Could not translate ring descriptor address\n");
380 if (v3_gpa_to_hva(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
381 PrintError(core->vm_info, core, "Could not translate ring available address\n");
385 if (v3_gpa_to_hva(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
386 PrintError(core->vm_info, core, "Could not translate ring used address\n");
390 PrintDebug(core->vm_info, core, "RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
391 (void *)(queue->ring_desc_addr),
392 (void *)(queue->ring_avail_addr),
393 (void *)(queue->ring_used_addr));
395 PrintDebug(core->vm_info, core, "RingDesc=%p, Avail=%p, Used=%p\n",
396 queue->desc, queue->avail, queue->used);
401 static int virtio_io_write(struct guest_info *core,
402 uint16_t port, void * src,
403 uint_t length, void * private_data)
405 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
406 int port_idx = port % virtio->io_range_size;
408 PrintDebug(core->vm_info, core, "VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n",
409 private_data, port, port_idx,
410 length, *(uint32_t *)src);
413 case GUEST_FEATURES_PORT:
415 PrintError(core->vm_info, core, "Illegal write length for guest features\n");
418 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
421 case VRING_PG_NUM_PORT:
423 PrintError(core->vm_info, core, "Illegal write length for page frame number\n");
426 addr_t pfn = *(uint32_t *)src;
427 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
428 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
431 virtio_setup_queue(core, virtio,
436 virtio_setup_queue(core, virtio,
439 if(virtio->tx_notify == 0){
440 disable_cb(&virtio->tx_vq);
445 virtio_setup_queue(core, virtio,
454 case VRING_Q_SEL_PORT:
455 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
456 if (virtio->virtio_cfg.vring_queue_selector > 2) {
457 PrintError(core->vm_info, core, "Virtio NIC: wrong queue idx: %d\n",
458 virtio->virtio_cfg.vring_queue_selector);
463 case VRING_Q_NOTIFY_PORT:
465 uint16_t queue_idx = *(uint16_t *)src;
467 /* receive queue refill */
468 virtio->stats.tx_interrupts ++;
469 } else if (queue_idx == 1){
470 if (handle_pkt_tx(core, virtio, 0) < 0) {
471 PrintError(core->vm_info, core, "Virtio NIC: Error to handle packet TX\n");
474 virtio->stats.tx_interrupts ++;
475 } else if (queue_idx == 2){
478 PrintError(core->vm_info, core, "Virtio NIC: Wrong queue index %d\n", queue_idx);
483 case VIRTIO_STATUS_PORT:
484 virtio->virtio_cfg.status = *(uint8_t *)src;
485 if (virtio->virtio_cfg.status == 0) {
486 virtio_init_state(virtio);
490 case VIRTIO_ISR_PORT:
491 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
502 static int virtio_io_read(struct guest_info *core,
503 uint16_t port, void * dst,
504 uint_t length, void * private_data)
506 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
507 int port_idx = port % virtio->io_range_size;
508 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
510 PrintDebug(core->vm_info, core, "Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n",
511 private_data, port, port_idx, length);
514 case HOST_FEATURES_PORT:
516 PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for host features\n");
519 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
522 case VRING_PG_NUM_PORT:
524 PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for page frame number\n");
529 *(uint32_t *)dst = virtio->rx_vq.pfn;
532 *(uint32_t *)dst = virtio->tx_vq.pfn;
535 *(uint32_t *)dst = virtio->ctrl_vq.pfn;
542 case VRING_SIZE_PORT:
544 PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for vring size\n");
549 *(uint16_t *)dst = virtio->rx_vq.queue_size;
552 *(uint16_t *)dst = virtio->tx_vq.queue_size;
555 *(uint16_t *)dst = virtio->ctrl_vq.queue_size;
562 case VIRTIO_STATUS_PORT:
564 PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for status\n");
567 *(uint8_t *)dst = virtio->virtio_cfg.status;
570 case VIRTIO_ISR_PORT:
571 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
572 virtio->virtio_cfg.pci_isr = 0;
573 v3_pci_lower_irq(virtio->virtio_dev->pci_bus,
577 case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN - 1:
578 *(uint8_t *)dst = virtio->net_cfg.mac[port_idx-VIRTIO_NET_CONFIG];
582 PrintError(core->vm_info, core, "Virtio NIC: Read of Unhandled Virtio Read:%d\n",
591 /* receiving raw ethernet pkt from backend */
592 static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) {
593 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
594 struct virtio_queue * q = &(virtio->rx_vq);
595 struct virtio_net_hdr_mrg_rxbuf hdr;
597 uint8_t kick_guest = 0;
599 V3_Net_Print(2, "Virtio NIC: virtio_rx: size: %d\n", size);
601 if (!q->ring_avail_addr) {
602 V3_Net_Print(2, "Virtio NIC: RX Queue not set\n");
603 virtio->stats.rx_dropped ++;
608 memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
610 flags = v3_lock_irqsave(virtio->rx_lock);
612 if (q->cur_avail_idx != q->avail->index){
614 struct vring_desc * buf_desc;
619 hdr_len = (virtio->mergeable_rx_bufs)?
620 sizeof(struct virtio_net_hdr_mrg_rxbuf):
621 sizeof(struct virtio_net_hdr);
623 if(virtio->mergeable_rx_bufs){/* merged buffer */
624 struct vring_desc * hdr_desc;
625 uint16_t old_idx = q->cur_avail_idx;
627 buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
628 hdr_desc = &(q->desc[buf_idx]);
630 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
631 virtio, hdr_desc, buf, size, hdr_len);
637 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
638 q->used->ring[q->used->index % q->queue_size].length = hdr_len + offset;
642 while(offset < size) {
643 buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
644 buf_desc = &(q->desc[buf_idx]);
646 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
647 virtio, buf_desc, buf+offset, size-offset, 0);
649 V3_Net_Print(2, "Virtio NIC: merged buffer, %d buffer size %d\n",
650 hdr.num_buffers, len);
651 q->cur_avail_idx = old_idx;
655 buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
657 q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
658 q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].length = len;
664 copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
665 virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0);
666 q->used->index += hdr.num_buffers;
668 buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
669 buf_desc = &(q->desc[buf_idx]);
672 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
673 virtio, buf_desc, (uchar_t *)&(hdr.hdr), hdr_len, 0);
675 V3_Net_Print(2, "Virtio NIC: rx copy header error %d, hdr_len %d\n",
680 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
681 virtio, buf_desc, buf, size, hdr_len);
683 V3_Net_Print(2, "Virtio NIC: rx copy data error %d\n", len);
688 /* copy rest of data */
689 while(offset < size &&
690 (buf_desc->flags & VIRTIO_NEXT_FLAG)){
691 buf_desc = &(q->desc[buf_desc->next]);
692 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
693 virtio, buf_desc, buf+offset, size-offset, 0);
699 buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
702 V3_Net_Print(2, "Virtio NIC: rx not enough ring buffer, buffer size %d\n",
707 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
708 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) */
713 virtio->stats.rx_pkts ++;
714 virtio->stats.rx_bytes += size;
716 V3_Net_Print(2, "Virtio NIC: Guest RX queue is full\n");
717 virtio->stats.rx_dropped ++;
719 /* kick guest to refill RX queue */
723 v3_unlock_irqrestore(virtio->rx_lock, flags);
725 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG) || kick_guest) {
726 V3_Net_Print(2, "Virtio NIC: RX Raising IRQ %d\n",
727 virtio->pci_dev->config_header.intr_line);
729 virtio->virtio_cfg.pci_isr = 0x1;
730 v3_pci_raise_irq(virtio->virtio_dev->pci_bus, virtio->pci_dev, 0);
731 virtio->stats.rx_interrupts ++;
734 /* notify guest if it is in guest mode */
735 if((kick_guest || virtio->rx_notify == 1) &&
736 V3_Get_CPU() != virtio->virtio_dev->vm->cores[0].pcpu_id){
737 v3_interrupt_cpu(virtio->virtio_dev->vm,
738 virtio->virtio_dev->vm->cores[0].pcpu_id,
745 virtio->stats.rx_dropped ++;
746 v3_unlock_irqrestore(virtio->rx_lock, flags);
751 static int virtio_free(struct virtio_dev_state * virtio) {
752 struct virtio_net_state * backend = NULL;
753 struct virtio_net_state * tmp = NULL;
756 list_for_each_entry_safe(backend, tmp, &(virtio->dev_list), dev_link) {
757 virtio_deinit_state(&(virtio->vm->cores[0]),backend);
758 list_del(&(backend->dev_link));
768 static struct v3_device_ops dev_ops = {
769 .free = (int (*)(void *))virtio_free,
773 static int virtio_poll(int quote, void * data){
774 struct virtio_net_state * virtio = (struct virtio_net_state *)data;
776 if (virtio->status) {
778 return handle_pkt_tx(&(virtio->vm->cores[0]), virtio, quote);
784 static int register_dev(struct virtio_dev_state * virtio,
785 struct virtio_net_state * net_state)
787 struct pci_device * pci_dev = NULL;
788 struct v3_pci_bar bars[6];
789 int num_ports = sizeof(struct virtio_config);
790 int tmp_ports = num_ports;
793 /* This gets the number of ports, rounded up to a power of 2 */
794 net_state->io_range_size = 1;
795 while (tmp_ports > 0) {
797 net_state->io_range_size <<= 1;
800 /* this is to account for any low order bits being set in num_ports
801 * if there are none, then num_ports was already a power of 2 so we shift right to reset it
803 if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
804 net_state->io_range_size >>= 1;
807 for (i = 0; i < 6; i++) {
808 bars[i].type = PCI_BAR_NONE;
811 PrintDebug(VM_NONE, VCORE_NONE, "Virtio NIC: io_range_size = %d\n",
812 net_state->io_range_size);
814 bars[0].type = PCI_BAR_IO;
815 bars[0].default_base_port = -1;
816 bars[0].num_ports = net_state->io_range_size;
817 bars[0].io_read = virtio_io_read;
818 bars[0].io_write = virtio_io_write;
819 bars[0].private_data = net_state;
821 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
822 0, PCI_AUTO_DEV_NUM, 0,
823 "LNX_VIRTIO_NIC", bars,
824 NULL, NULL, NULL, NULL, net_state);
827 PrintError(VM_NONE, VCORE_NONE, "Virtio NIC: Could not register PCI Device\n");
831 PrintDebug(VM_NONE, VCORE_NONE, "Virtio NIC: registered to PCI bus\n");
833 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
834 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
837 pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
838 pci_dev->config_header.class = PCI_CLASS_NETWORK;
839 pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
840 pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;
841 pci_dev->config_header.intr_pin = 1;
842 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
844 net_state->pci_dev = pci_dev;
845 net_state->virtio_dev = virtio;
847 memcpy(net_state->net_cfg.mac, virtio->mac, 6);
849 virtio_init_state(net_state);
851 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: Registered Intr Line %d\n", pci_dev->config_header.intr_line);
853 /* Add backend to list of devices */
854 list_add(&(net_state->dev_link), &(virtio->dev_list));
859 #define RATE_UPPER_THRESHOLD_DEFAULT 10000 /* 10000 pkts per second, around 100Mbits */
860 #define RATE_LOWER_THRESHOLD_DEFAULT 1000 /* 1000 pkts per second, around 10Mbits */
861 #define PROFILE_PERIOD_DEFAULT 10000 /* us */
863 static void virtio_nic_timer(struct guest_info * core,
864 uint64_t cpu_cycles, uint64_t cpu_freq,
866 struct virtio_net_state * net_state = (struct virtio_net_state *)priv_data;
868 static int profile_ms = 0;
869 uint64_t target_period_us = net_state->virtio_dev->period_us;
870 uint64_t upper_thresh_pps = net_state->virtio_dev->upper_thresh_pps;
871 uint64_t lower_thresh_pps = net_state->virtio_dev->lower_thresh_pps;
874 if(!net_state->status){ /* VNIC is not in working status */
878 period_us = (1000*cpu_cycles)/cpu_freq;
879 net_state->past_us += period_us;
881 if (net_state->past_us > target_period_us) {
883 uint64_t tx_count, rx_count;
884 uint64_t lb_tx_count, lb_rx_count;
885 uint64_t ub_tx_count, ub_rx_count;
887 lb_tx_count = lb_rx_count = (lower_thresh_pps * 1000000) / net_state->past_us; // packets expected in this interval
888 ub_tx_count = ub_rx_count = (upper_thresh_pps * 1000000) / net_state->past_us;
890 tx_count = net_state->stats.tx_pkts - net_state->tx_pkts;
891 rx_count = net_state->stats.rx_pkts - net_state->rx_pkts;
893 net_state->tx_pkts = net_state->stats.tx_pkts;
894 net_state->rx_pkts = net_state->stats.rx_pkts;
896 if(tx_count > ub_tx_count && net_state->tx_notify == 1) {
897 PrintDebug(core->vm_info, core, "Virtio NIC: Switch TX to VMM driven mode\n");
898 disable_cb(&(net_state->tx_vq));
899 net_state->tx_notify = 0;
902 if(tx_count < lb_tx_count && net_state->tx_notify == 0) {
903 PrintDebug(core->vm_info, core, "Virtio NIC: Switch TX to Guest driven mode\n");
904 enable_cb(&(net_state->tx_vq));
905 net_state->tx_notify = 1;
908 if(rx_count > ub_rx_count && net_state->rx_notify == 1) {
909 PrintDebug(core->vm_info, core, "Virtio NIC: Switch RX to VMM None notify mode\n");
910 net_state->rx_notify = 0;
913 if(rx_count < lb_rx_count && net_state->rx_notify == 0) {
914 V3_Print(core->vm_info, core, "Virtio NIC: Switch RX to VMM notify mode\n");
915 net_state->rx_notify = 1;
918 net_state->past_us = 0;
921 profile_ms += period_us/1000;
922 if(profile_ms > 20000){
923 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",
924 net_state->stats.tx_pkts, net_state->stats.tx_bytes,
925 net_state->stats.rx_pkts, net_state->stats.rx_bytes,
926 net_state->stats.tx_dropped, net_state->stats.rx_dropped,
927 net_state->stats.tx_interrupts, net_state->stats.rx_interrupts,
928 net_state->vm->cores[0].num_exits);
933 static struct v3_timer_ops timer_ops = {
934 .update_timer = virtio_nic_timer,
937 static int connect_fn(struct v3_vm_info * info,
938 void * frontend_data,
939 struct v3_dev_net_ops * ops,
941 void * private_data) {
942 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
943 struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
946 PrintError(info, VCORE_NONE, "Cannot allocate in connect\n");
950 memset(net_state, 0, sizeof(struct virtio_net_state));
951 register_dev(virtio, net_state);
953 net_state->vm = info;
954 net_state->net_ops = ops;
955 net_state->backend_data = private_data;
956 net_state->virtio_dev = virtio;
958 switch (virtio->model) {
960 V3_Print(info, VCORE_NONE, "Virtio NIC: Guest-driven operation\n");
961 net_state->tx_notify = 1;
962 net_state->rx_notify = 1;
965 V3_Print(info, VCORE_NONE, "Virtio NIC: VMM-driven operation\n");
966 net_state->tx_notify = 0;
967 net_state->rx_notify = 0;
970 V3_Print(info, VCORE_NONE, "Virtio NIC: Adaptive operation (begins in guest-driven mode)\n");
971 net_state->tx_notify = 1;
972 net_state->rx_notify = 1;
974 net_state->timer = v3_add_timer(&(info->cores[0]), &timer_ops,net_state);
980 V3_Print(info, VCORE_NONE, "Virtio NIC: Unknown model, using GUEST_DRIVEN\n");
981 net_state->tx_notify = 1;
982 net_state->rx_notify = 1;
987 ops->recv = virtio_rx;
988 ops->poll = virtio_poll;
989 ops->config.frontend_data = net_state;
990 ops->config.poll = 1;
991 ops->config.quote = 64;
992 ops->config.fnt_mac = V3_Malloc(ETH_ALEN);
994 if (!ops->config.fnt_mac) {
995 PrintError(info, VCORE_NONE, "Cannot allocate in connect\n");
996 // should unregister here
1000 memcpy(ops->config.fnt_mac, virtio->mac, ETH_ALEN);
1005 static int setup_perf_model(struct virtio_dev_state *virtio_state, v3_cfg_tree_t *t)
1007 char *mode = v3_cfg_val(t,"mode");
1010 virtio_state->model = GUEST_DRIVEN;
1011 virtio_state->lower_thresh_pps = RATE_LOWER_THRESHOLD_DEFAULT;
1012 virtio_state->upper_thresh_pps = RATE_UPPER_THRESHOLD_DEFAULT;
1013 virtio_state->period_us = PROFILE_PERIOD_DEFAULT;
1018 if (!strcasecmp(mode,"guest-driven")) {
1019 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: Setting static GUEST_DRIVEN mode of operation (latency optimized)\n");
1020 virtio_state->model=GUEST_DRIVEN;
1021 } else if (!strcasecmp(mode, "vmm-driven")) {
1022 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: Setting static VMM_DRIVEN mode of operation (throughput optimized)\n");
1023 virtio_state->model=VMM_DRIVEN;
1024 } else if (!strcasecmp(mode, "adaptive")) {
1027 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: Setting dynamic ADAPTIVE mode of operation\n");
1028 virtio_state->model=ADAPTIVE;
1030 if (!(s=v3_cfg_val(t,"upper"))) {
1031 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No upper bound given, using default\n");
1033 virtio_state->upper_thresh_pps = atoi(s);
1035 if (!(s=v3_cfg_val(t,"lower"))) {
1036 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No lower bound given, using default\n");
1038 virtio_state->lower_thresh_pps = atoi(s);
1040 if (!(s=v3_cfg_val(t,"period"))) {
1041 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No period given, using default\n");
1043 virtio_state->period_us = atoi(s);
1046 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: lower_thresh_pps=%llu, upper_thresh_pps=%llu, period_us=%llu\n",
1047 virtio_state->lower_thresh_pps,
1048 virtio_state->upper_thresh_pps,
1049 virtio_state->period_us);
1051 PrintError(VM_NONE, VCORE_NONE, "Virtio NIC: Unknown mode of operation '%s', using default (guest-driven)\n",mode);
1052 virtio_state->model=GUEST_DRIVEN;
1055 V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No model given, using default (guest-driven)\n");
1063 <device class="LNX_VIRTIO_NIC" id="nic">
1064 <bus>pci-bus-to-attach-to</bus> // required
1065 <mac>mac address</mac> // if ommited with pic one
1066 <model mode="guest-driven|vmm-driven|adaptive" upper="pkts_per_sec" lower="pkts" period="us" />
1069 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
1070 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
1071 struct virtio_dev_state * virtio_state = NULL;
1072 char * dev_id = v3_cfg_val(cfg, "ID");
1073 char * mac = v3_cfg_val(cfg, "mac");
1074 v3_cfg_tree_t *model = v3_cfg_subtree(cfg,"model");
1076 if (pci_bus == NULL) {
1077 PrintError(vm, VCORE_NONE, "Virtio NIC: Virtio device require a PCI Bus");
1081 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
1083 if (!virtio_state) {
1084 PrintError(vm, VCORE_NONE, "Cannot allocate in init\n");
1088 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
1090 INIT_LIST_HEAD(&(virtio_state->dev_list));
1091 virtio_state->pci_bus = pci_bus;
1092 virtio_state->vm = vm;
1095 if (!str2mac(mac, virtio_state->mac)) {
1096 PrintDebug(vm, VCORE_NONE, "Virtio NIC: Mac specified %s\n", mac);
1098 PrintError(vm, VCORE_NONE, "Virtio NIC: Mac specified is incorrect, picking a randoom mac\n");
1099 random_ethaddr(virtio_state->mac);
1102 PrintDebug(vm, VCORE_NONE, "Virtio NIC: no mac specified, so picking a random mac\n");
1103 random_ethaddr(virtio_state->mac);
1106 if (setup_perf_model(virtio_state,model)<0) {
1107 PrintError(vm, VCORE_NONE, "Cannnot setup performance model\n");
1108 V3_Free(virtio_state);
1112 struct vm_device * dev = v3_add_device(vm, dev_id, &dev_ops, virtio_state);
1115 PrintError(vm, VCORE_NONE, "Virtio NIC: Could not attach device %s\n", dev_id);
1116 V3_Free(virtio_state);
1120 if (v3_dev_add_net_frontend(vm, dev_id, connect_fn, (void *)virtio_state) == -1) {
1121 PrintError(vm, VCORE_NONE, "Virtio NIC: Could not register %s as net frontend\n", dev_id);
1122 v3_remove_device(dev);
1129 device_register("LNX_VIRTIO_NIC", virtio_init)