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 <palacios/vmm_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>
35 #ifndef CONFIG_DEBUG_VIRTIO_NET
37 #define PrintDebug(fmt, args...)
40 #define VIRTIO_NET_MAX_BUFSIZE (sizeof(struct virtio_net_hdr) + (64 << 10))
43 struct virtio_net_hdr {
47 uint16_t hdr_len; /* Ethernet + IP + tcp/udp hdrs */
48 uint16_t gso_size; /* Bytes to append to hdr_len per frame */
49 uint16_t csum_start; /* Position to start checksumming from */
50 uint16_t csum_offset; /* Offset after that to place checksum */
51 }__attribute__((packed));
54 struct virtio_net_hdr_mrg_rxbuf {
55 struct virtio_net_hdr hdr;
56 uint16_t num_buffers; /* Number of merged rx buffers */
60 #define TX_QUEUE_SIZE 64
61 #define RX_QUEUE_SIZE 1024
62 #define CTRL_QUEUE_SIZE 64
64 #define VIRTIO_NET_F_MRG_RXBUF 15 /* Host can merge receive buffers. */
65 #define VIRTIO_NET_F_MAC 5 /* Host has given MAC address. */
66 #define VIRTIO_NET_F_GSO 6 /* Host handles pkts w/ any GSO type */
67 #define VIRTIO_NET_F_HOST_TSO4 11 /* Host can handle TSOv4 in. */
69 /* Port to get virtio config */
70 #define VIRTIO_NET_CONFIG 20
72 struct virtio_net_config
74 uint8_t mac[ETH_ALEN]; /* VIRTIO_NET_F_MAC */
76 } __attribute__((packed));
78 struct virtio_dev_state {
79 struct vm_device * pci_bus;
80 struct list_head dev_list;
81 struct v3_vm_info *vm;
83 uint8_t mac[ETH_ALEN];
86 struct virtio_net_state {
87 struct virtio_net_config net_cfg;
88 struct virtio_config virtio_cfg;
90 struct vm_device * dev;
91 struct pci_device * pci_dev;
94 struct virtio_queue rx_vq; /* idx 0*/
95 struct virtio_queue tx_vq; /* idx 1*/
96 struct virtio_queue ctrl_vq; /* idx 2*/
98 struct nic_statistics statistics;
100 struct v3_dev_net_ops * net_ops;
101 v3_lock_t rx_lock, tx_lock;
104 struct virtio_dev_state * virtio_dev;
105 struct list_head dev_link;
109 static int virtio_init_state(struct virtio_net_state * virtio)
111 virtio->rx_vq.queue_size = RX_QUEUE_SIZE;
112 virtio->tx_vq.queue_size = TX_QUEUE_SIZE;
113 virtio->ctrl_vq.queue_size = CTRL_QUEUE_SIZE;
115 virtio->rx_vq.ring_desc_addr = 0;
116 virtio->rx_vq.ring_avail_addr = 0;
117 virtio->rx_vq.ring_used_addr = 0;
118 virtio->rx_vq.pfn = 0;
119 virtio->rx_vq.cur_avail_idx = 0;
121 virtio->tx_vq.ring_desc_addr = 0;
122 virtio->tx_vq.ring_avail_addr = 0;
123 virtio->tx_vq.ring_used_addr = 0;
124 virtio->tx_vq.pfn = 0;
125 virtio->tx_vq.cur_avail_idx = 0;
127 virtio->ctrl_vq.ring_desc_addr = 0;
128 virtio->ctrl_vq.ring_avail_addr = 0;
129 virtio->ctrl_vq.ring_used_addr = 0;
130 virtio->ctrl_vq.pfn = 0;
131 virtio->ctrl_vq.cur_avail_idx = 0;
133 virtio->virtio_cfg.pci_isr = 0;
135 virtio->virtio_cfg.host_features = 0 | (1 << VIRTIO_NET_F_MAC);
137 if ((v3_lock_init(&(virtio->rx_lock)) == -1) ||
138 (v3_lock_init(&(virtio->tx_lock)) == -1)){
139 PrintError("Virtio NIC: Failure to init locks for net_state\n");
145 static int tx_one_pkt(struct guest_info * core,
146 struct virtio_net_state * virtio,
147 struct vring_desc * buf_desc)
149 uint8_t * buf = NULL;
150 uint32_t len = buf_desc->length;
152 if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
153 PrintDebug("Could not translate buffer address\n");
157 if(virtio->net_ops->send(buf, len, virtio->backend_data) >= 0){
158 virtio->statistics.tx_pkts ++;
159 virtio->statistics.tx_bytes += len;
164 virtio->statistics.tx_dropped ++;
171 copy_data_to_desc(struct guest_info * core,
172 struct virtio_net_state * virtio_state,
173 struct vring_desc * desc,
179 uint8_t * desc_buf = NULL;
181 if (v3_gpa_to_hva(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
182 PrintError("Could not translate buffer address\n");
185 len = (desc->length < buf_len)?(desc->length - offset):buf_len;
186 memcpy(desc_buf+offset, buf, len);
192 static int get_desc_count(struct virtio_queue * q, int index) {
193 struct vring_desc * tmp_desc = &(q->desc[index]);
196 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
197 tmp_desc = &(q->desc[tmp_desc->next]);
204 static inline void enable_cb(struct virtio_queue *queue){
205 queue->used->flags &= ~ VRING_NO_NOTIFY_FLAG;
208 static inline void disable_cb(struct virtio_queue *queue) {
209 queue->used->flags |= VRING_NO_NOTIFY_FLAG;
213 /* interrupt the guest, so the guest core get EXIT to Palacios */
214 static inline void notify_guest(struct virtio_net_state * virtio){
215 v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].cpu_id, 0);
219 /* guest free some pkts for rx queue */
220 static int handle_rx_queue_kick(struct guest_info * core,
221 struct virtio_net_state * virtio)
227 static int handle_ctrl(struct guest_info * core,
228 struct virtio_net_state * virtio) {
233 static int handle_pkt_tx(struct guest_info * core,
234 struct virtio_net_state * virtio_state)
236 struct virtio_queue *q = &(virtio_state->tx_vq);
237 struct virtio_net_hdr *hdr = NULL;
241 if (!q->ring_avail_addr) {
245 flags = v3_lock_irqsave(virtio_state->tx_lock);
246 while (q->cur_avail_idx != q->avail->index) {
247 struct vring_desc * hdr_desc = NULL;
249 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
250 int desc_cnt = get_desc_count(q, desc_idx);
251 uint32_t req_len = 0;
254 hdr_desc = &(q->desc[desc_idx]);
255 if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
256 PrintError("Could not translate block header address\n");
260 hdr = (struct virtio_net_hdr *)hdr_addr;
261 desc_idx = hdr_desc->next;
264 PrintError("VNIC: merged rx buffer not supported\n");
268 /* here we assumed that one ethernet pkt is not splitted into multiple virtio buffer */
269 for (i = 0; i < desc_cnt - 1; i++) {
270 struct vring_desc * buf_desc = &(q->desc[desc_idx]);
271 if (tx_one_pkt(core, virtio_state, buf_desc) == -1) {
272 PrintError("Error handling nic operation\n");
276 req_len += buf_desc->length;
277 desc_idx = buf_desc->next;
280 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
281 q->used->ring[q->used->index % q->queue_size].length = req_len; /* What do we set this to???? */
289 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
291 if (txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
292 v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev);
293 virtio_state->virtio_cfg.pci_isr = 0x1;
295 virtio_state->statistics.interrupts ++;
302 v3_unlock_irqrestore(virtio_state->tx_lock, flags);
307 static int virtio_setup_queue(struct guest_info *core,
308 struct virtio_net_state * virtio_state,
309 struct virtio_queue * queue,
310 addr_t pfn, addr_t page_addr) {
313 queue->ring_desc_addr = page_addr;
314 queue->ring_avail_addr = page_addr + (queue->queue_size * sizeof(struct vring_desc));
315 queue->ring_used_addr = ((queue->ring_avail_addr) +
316 (sizeof(struct vring_avail)) +
317 (queue->queue_size * sizeof(uint16_t)));
319 // round up to next page boundary.
320 queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
321 if (v3_gpa_to_hva(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
322 PrintError("Could not translate ring descriptor address\n");
326 if (v3_gpa_to_hva(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
327 PrintError("Could not translate ring available address\n");
331 if (v3_gpa_to_hva(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
332 PrintError("Could not translate ring used address\n");
336 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
337 (void *)(queue->ring_desc_addr),
338 (void *)(queue->ring_avail_addr),
339 (void *)(queue->ring_used_addr));
341 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
342 queue->desc, queue->avail, queue->used);
347 static int virtio_io_write(struct guest_info *core,
348 uint16_t port, void * src,
349 uint_t length, void * private_data)
351 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
352 int port_idx = port % virtio->io_range_size;
354 PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n", private_data,
355 port, port_idx, length, *(uint32_t *)src);
358 case GUEST_FEATURES_PORT:
360 PrintError("Illegal write length for guest features\n");
363 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
366 case VRING_PG_NUM_PORT:
368 PrintError("Illegal write length for page frame number\n");
371 addr_t pfn = *(uint32_t *)src;
372 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
373 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
376 virtio_setup_queue(core, virtio, &virtio->rx_vq, pfn, page_addr);
379 virtio_setup_queue(core, virtio, &virtio->tx_vq, pfn, page_addr);
382 virtio_setup_queue(core, virtio, &virtio->ctrl_vq, pfn, page_addr);
389 case VRING_Q_SEL_PORT:
390 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
391 if (virtio->virtio_cfg.vring_queue_selector > 2) {
392 PrintError("Virtio NIC: wrong queue idx: %d\n",
393 virtio->virtio_cfg.vring_queue_selector);
398 case VRING_Q_NOTIFY_PORT:
400 uint16_t queue_idx = *(uint16_t *)src;
402 if(handle_rx_queue_kick(core, virtio) == -1){
403 PrintError("Could not handle Virtio NIC rx kick\n");
406 } else if (queue_idx == 1){
407 if (handle_pkt_tx(core, virtio) == -1) {
408 PrintError("Could not handle Virtio NIC tx kick\n");
411 } else if (queue_idx == 2){
412 if (handle_ctrl(core, virtio) == -1) {
413 PrintError("Could not handle Virtio NIC ctrl kick\n");
417 PrintError("Wrong queue index %d\n", queue_idx);
422 case VIRTIO_STATUS_PORT:
423 virtio->virtio_cfg.status = *(uint8_t *)src;
424 if (virtio->virtio_cfg.status == 0) {
425 virtio_init_state(virtio);
429 case VIRTIO_ISR_PORT:
430 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
441 static int virtio_io_read(struct guest_info *core,
442 uint16_t port, void * dst,
443 uint_t length, void * private_data)
445 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
446 int port_idx = port % virtio->io_range_size;
447 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
449 PrintDebug("Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n", private_data,
450 port, port_idx, length);
453 case HOST_FEATURES_PORT:
455 PrintError("Illegal read length for host features\n");
458 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
461 case VRING_PG_NUM_PORT:
463 PrintError("Illegal read length for page frame number\n");
468 *(uint32_t *)dst = virtio->rx_vq.pfn;
471 *(uint32_t *)dst = virtio->tx_vq.pfn;
474 *(uint32_t *)dst = virtio->ctrl_vq.pfn;
481 case VRING_SIZE_PORT:
483 PrintError("Illegal read length for vring size\n");
488 *(uint16_t *)dst = virtio->rx_vq.queue_size;
491 *(uint16_t *)dst = virtio->tx_vq.queue_size;
494 *(uint16_t *)dst = virtio->ctrl_vq.queue_size;
501 case VIRTIO_STATUS_PORT:
503 PrintError("Illegal read length for status\n");
506 *(uint8_t *)dst = virtio->virtio_cfg.status;
509 case VIRTIO_ISR_PORT:
510 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
511 virtio->virtio_cfg.pci_isr = 0;
512 v3_pci_lower_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
515 case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN:
516 *(uint8_t *)dst = virtio->net_cfg.mac[port_idx-VIRTIO_NET_CONFIG];
520 PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n", port_idx);
528 /* receiving raw ethernet pkt from backend */
529 static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) {
530 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
531 struct virtio_queue * q = &(virtio->rx_vq);
532 struct virtio_net_hdr_mrg_rxbuf hdr;
533 uint32_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
538 #ifndef CONFIG_DEBUG_VIRTIO_NET
539 PrintDebug("Virtio-NIC: virtio_rx: size: %d\n", size);
540 v3_hexdump(buf, size, NULL, 0);
543 flags = v3_lock_irqsave(virtio->rx_lock);
546 memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
548 if (q->ring_avail_addr == 0) {
549 PrintDebug("Queue is not set\n");
553 if (q->cur_avail_idx != q->avail->index){
555 uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
556 uint16_t buf_idx = 0;
557 struct vring_desc * hdr_desc = NULL;
559 hdr_desc = &(q->desc[hdr_idx]);
560 if (v3_gpa_to_hva(&(virtio->virtio_dev->vm->cores[0]), hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
561 PrintDebug("Could not translate receive buffer address\n");
565 memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr_mrg_rxbuf));
566 if (offset >= data_len) {
567 hdr_desc->flags &= ~VIRTIO_NEXT_FLAG;
570 struct vring_desc * buf_desc = NULL;
571 for (buf_idx = hdr_desc->next; offset < data_len; buf_idx = q->desc[hdr_idx].next) {
573 buf_desc = &(q->desc[buf_idx]);
575 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf + offset, data_len - offset, 0);
577 if (offset < data_len) {
578 buf_desc->flags = VIRTIO_NEXT_FLAG;
580 buf_desc->length = len;
582 buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
584 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
585 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) */
589 virtio->statistics.rx_pkts ++;
590 virtio->statistics.rx_bytes += size;
592 virtio->statistics.rx_dropped ++;
597 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
598 PrintDebug("Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line);
600 v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
601 virtio->virtio_cfg.pci_isr = 0x1;
602 virtio->statistics.interrupts ++;
605 v3_unlock_irqrestore(virtio->rx_lock, flags);
611 v3_unlock_irqrestore(virtio->rx_lock, flags);
616 static int virtio_free(struct virtio_dev_state * virtio) {
617 struct virtio_net_state * backend = NULL;
618 struct virtio_net_state * tmp = NULL;
621 list_for_each_entry_safe(backend, tmp, &(virtio->dev_list), dev_link) {
623 // unregister from PCI
625 list_del(&(backend->dev_link));
634 static struct v3_device_ops dev_ops = {
635 .free = (int (*)(void *))virtio_free,
639 static void virtio_nic_poll(struct v3_vm_info * vm, void * data){
640 struct virtio_net_state * virtio = (struct virtio_net_state *)data;
642 handle_pkt_tx(&(vm->cores[0]), virtio);
645 static int register_dev(struct virtio_dev_state * virtio,
646 struct virtio_net_state * net_state)
648 struct pci_device * pci_dev = NULL;
649 struct v3_pci_bar bars[6];
650 int num_ports = sizeof(struct virtio_config);
651 int tmp_ports = num_ports;
654 // This gets the number of ports, rounded up to a power of 2
655 net_state->io_range_size = 1; // must be a power of 2
656 while (tmp_ports > 0) {
658 net_state->io_range_size <<= 1;
661 /* this is to account for any low order bits being set in num_ports
662 * if there are none, then num_ports was already a power of 2 so we shift right to reset it
664 if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
665 net_state->io_range_size >>= 1;
668 for (i = 0; i < 6; i++) {
669 bars[i].type = PCI_BAR_NONE;
672 PrintDebug("Virtio-NIC io_range_size = %d\n", net_state->io_range_size);
674 bars[0].type = PCI_BAR_IO;
675 bars[0].default_base_port = -1;
676 bars[0].num_ports = net_state->io_range_size;
677 bars[0].io_read = virtio_io_read;
678 bars[0].io_write = virtio_io_write;
679 bars[0].private_data = net_state;
681 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
682 0, 4/*PCI_AUTO_DEV_NUM*/, 0,
683 "LNX_VIRTIO_NIC", bars,
684 NULL, NULL, NULL, net_state);
687 PrintError("Virtio NIC: Could not register PCI Device\n");
691 PrintDebug("Virtio NIC: registered to PCI bus\n");
693 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
694 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
697 pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
698 pci_dev->config_header.class = PCI_CLASS_NETWORK;
699 pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
700 pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;
701 pci_dev->config_header.intr_pin = 1;
702 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
704 net_state->pci_dev = pci_dev;
705 net_state->virtio_dev = virtio;
707 memcpy(net_state->net_cfg.mac, virtio->mac, 6);
709 virtio_init_state(net_state);
711 /* Add backend to list of devices */
712 list_add(&(net_state->dev_link), &(virtio->dev_list));
717 static int connect_fn(struct v3_vm_info * info,
718 void * frontend_data,
719 struct v3_dev_net_ops * ops,
721 void * private_data) {
722 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
723 struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
725 memset(net_state, 0, sizeof(struct virtio_net_state));
726 register_dev(virtio, net_state);
728 net_state->net_ops = ops;
729 net_state->backend_data = private_data;
730 net_state->virtio_dev = virtio;
733 ops->recv = virtio_rx;
734 ops->poll = virtio_nic_poll;
735 ops->frontend_data = net_state;
736 memcpy(ops->fnt_mac, virtio->mac, ETH_ALEN);
741 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
742 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
743 struct virtio_dev_state * virtio_state = NULL;
744 char * dev_id = v3_cfg_val(cfg, "ID");
745 char * macstr = v3_cfg_val(cfg, "mac");
747 if (pci_bus == NULL) {
748 PrintError("Virtio NIC: VirtIO devices require a PCI Bus");
752 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
753 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
755 INIT_LIST_HEAD(&(virtio_state->dev_list));
756 virtio_state->pci_bus = pci_bus;
757 virtio_state->vm = vm;
759 if (macstr != NULL && !str2mac(macstr, virtio_state->mac)) {
760 PrintDebug("Virtio NIC: Mac specified %s\n", macstr);
762 PrintDebug("Virtio NIC: MAC not specified\n");
763 random_ethaddr(virtio_state->mac);
766 struct vm_device * dev = v3_add_device(vm, dev_id, &dev_ops, virtio_state);
769 PrintError("Virtio NIC: Could not attach device %s\n", dev_id);
770 V3_Free(virtio_state);
774 if (v3_dev_add_net_frontend(vm, dev_id, connect_fn, (void *)virtio_state) == -1) {
775 PrintError("Virtio NIC: Could not register %s as net frontend\n", dev_id);
776 v3_remove_device(dev);
783 device_register("LNX_VIRTIO_NIC", virtio_init)