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) 2008, Lei Xia <lxia@northwestern.edu>
11 * Copyright (c) 2008, Cui Zheng <cuizheng@cs.unm.edu>
12 * Copyright (c) 2008, 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>
31 #include <devices/pci.h>
34 #ifndef CONFIG_DEBUG_VIRTIO_NET
36 #define PrintDebug(fmt, args...)
39 #define VIRTIO_NIC_PROFILE
41 #define VIRTIO_NET_S_LINK_UP 1 /* Link is up */
42 #define VIRTIO_NET_MAX_BUFSIZE (sizeof(struct virtio_net_hdr) + (64 << 10))
44 struct virtio_net_hdr {
47 #define VIRTIO_NET_HDR_GSO_NONE 0 /* Not a GSO frame */
49 uint16_t hdr_len; /* Ethernet + IP + tcp/udp hdrs */
50 uint16_t gso_size; /* Bytes to append to hdr_len per frame */
51 uint16_t csum_start; /* Position to start checksumming from */
52 uint16_t csum_offset; /* Offset after that to place checksum */
53 }__attribute__((packed));
56 #define QUEUE_SIZE 1024
57 #define CTRL_QUEUE_SIZE 64
60 struct virtio_net_config
62 uint8_t mac[ETH_ALEN]; //VIRTIO_NET_F_MAC
64 } __attribute__((packed));
66 struct virtio_dev_state {
67 struct vm_device * pci_bus;
68 struct list_head dev_list;
69 struct v3_vm_info *vm;
72 struct virtio_net_state {
73 struct virtio_net_config net_cfg;
74 struct virtio_config virtio_cfg;
76 struct vm_device * dev;
77 struct pci_device * pci_dev;
80 struct virtio_queue rx_vq; //index 0, rvq in Linux virtio driver, handle packet to guest
81 struct virtio_queue tx_vq; //index 1, svq in Linux virtio driver, handle packet from guest
82 struct virtio_queue ctrl_vq; //index 2, ctrol info from guest
84 ulong_t pkt_sent, pkt_recv, pkt_drop;
86 struct v3_dev_net_ops * net_ops;
91 struct virtio_dev_state * virtio_dev;
92 struct list_head dev_link;
95 static int virtio_free(struct vm_device * dev)
101 static int virtio_init_state(struct virtio_net_state * virtio)
103 virtio->rx_vq.ring_desc_addr = 0;
104 virtio->rx_vq.ring_avail_addr = 0;
105 virtio->rx_vq.ring_used_addr = 0;
106 virtio->rx_vq.pfn = 0;
107 virtio->rx_vq.cur_avail_idx = 0;
109 virtio->tx_vq.ring_desc_addr = 0;
110 virtio->tx_vq.ring_avail_addr = 0;
111 virtio->tx_vq.ring_used_addr = 0;
112 virtio->tx_vq.pfn = 0;
113 virtio->tx_vq.cur_avail_idx = 0;
115 virtio->ctrl_vq.ring_desc_addr = 0;
116 virtio->ctrl_vq.ring_avail_addr = 0;
117 virtio->ctrl_vq.ring_used_addr = 0;
118 virtio->ctrl_vq.pfn = 0;
119 virtio->ctrl_vq.cur_avail_idx = 0;
121 virtio->virtio_cfg.host_features = 0;
122 //virtio->virtio_cfg.status = VIRTIO_NET_S_LINK_UP;
123 virtio->virtio_cfg.pci_isr = 0;
125 if (v3_lock_init(&(virtio->lock)) == -1){
126 PrintError("Virtio NIC: Failure to init lock for net_state\n");
129 virtio->pkt_sent = virtio->pkt_recv = virtio->pkt_drop = 0;
134 static int pkt_tx(struct guest_info *core, struct virtio_net_state * virtio, struct vring_desc * buf_desc)
136 uint8_t * buf = NULL;
137 uint32_t len = buf_desc->length;
139 PrintDebug("Virtio NIC: Virtio Pkt Sending, net_state: %p, pkt size: %d\n", virtio, len);
141 if (guest_pa_to_host_va(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
142 PrintError("Could not translate buffer address\n");
146 if (virtio->net_ops->send(buf, len, (void *)virtio, NULL) == -1) {
153 static int build_receive_header(struct virtio_net_hdr * hdr, const void * buf, int raw) {
157 memcpy(hdr, buf, sizeof(struct virtio_net_hdr));
159 memset(hdr, 0, sizeof(struct virtio_net_hdr));
165 static int copy_data_to_desc(struct guest_info *core,
166 struct virtio_net_state * virtio_state,
167 struct vring_desc * desc,
172 uint8_t * desc_buf = NULL;
174 if (guest_pa_to_host_va(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
175 PrintError("Could not translate buffer address\n");
178 len = (desc->length < buf_len)?desc->length:buf_len;
179 memcpy(desc_buf, buf, len);
187 static int get_desc_count(struct virtio_queue * q, int index) {
188 struct vring_desc * tmp_desc = &(q->desc[index]);
191 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
192 tmp_desc = &(q->desc[tmp_desc->next]);
199 static int handle_ctrl(struct guest_info *core, struct virtio_net_state * dev) {
203 static int handle_pkt_tx(struct guest_info *core, struct virtio_net_state * virtio_state)
205 struct virtio_queue * q = &(virtio_state->tx_vq);
206 struct virtio_net_hdr * hdr = NULL;
208 if (q->avail->index < q->last_avail_idx) {
209 q->idx_overflow = true;
212 q->last_avail_idx = q->avail->index;
214 while (q->cur_avail_idx < q->avail->index ||
215 (q->idx_overflow && q->cur_avail_idx < (q->avail->index + 65536))) {
216 struct vring_desc * hdr_desc = NULL;
218 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
219 int desc_cnt = get_desc_count(q, desc_idx);
220 uint32_t req_len = 0;
223 hdr_desc = &(q->desc[desc_idx]);
224 if (guest_pa_to_host_va(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
225 PrintError("Could not translate block header address\n");
229 hdr = (struct virtio_net_hdr*)hdr_addr;
230 desc_idx = hdr_desc->next;
232 for (i = 0; i < desc_cnt - 1; i++) {
233 struct vring_desc * buf_desc = &(q->desc[desc_idx]);
234 if (pkt_tx(core, virtio_state, buf_desc) == -1) {
235 PrintError("Error handling nic operation\n");
239 req_len += buf_desc->length;
240 desc_idx = buf_desc->next;
242 virtio_state->pkt_sent ++;
244 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
245 q->used->ring[q->used->index % q->queue_size].length = req_len; // What do we set this to????
248 int last_idx = q->cur_avail_idx;
252 if (q->cur_avail_idx < last_idx) {
253 q->idx_overflow = false;
257 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
258 v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev);
259 virtio_state->virtio_cfg.pci_isr = 0x1;
262 #ifdef VIRTIO_NIC_PROFILE
263 if(virtio_state->pkt_sent % 10000 == 0)
264 PrintError("Virtio NIC: %p, pkt_sent: %ld\n", virtio_state, virtio_state->pkt_sent);
271 static int virtio_setup_queue(struct guest_info *core,
272 struct virtio_net_state * virtio_state,
273 struct virtio_queue * queue,
278 queue->ring_desc_addr = page_addr;
279 queue->ring_avail_addr = page_addr + (queue->queue_size * sizeof(struct vring_desc));
280 queue->ring_used_addr = ((queue->ring_avail_addr) +
281 (sizeof(struct vring_avail)) +
282 (queue->queue_size * sizeof(uint16_t)));
284 PrintDebug("Virtio NIC: In setup queue: queue %p\n", (void *)queue);
286 // round up to next page boundary.
287 queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
288 if (guest_pa_to_host_va(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
289 PrintError("Could not translate ring descriptor address\n");
293 PrintDebug("Virtio NIC: In setup queue111: queue %p\n", (void *)queue);
295 if (guest_pa_to_host_va(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
296 PrintError("Could not translate ring available address\n");
300 PrintDebug("Virtio NIC: In setup queue222: queue %p\n", (void *)queue);
302 if (guest_pa_to_host_va(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
303 PrintError("Could not translate ring used address\n");
307 PrintDebug("Virtio NIC: In setup queue333: queue %p\n", (void *)queue);
309 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
310 (void *)(queue->ring_desc_addr),
311 (void *)(queue->ring_avail_addr),
312 (void *)(queue->ring_used_addr));
314 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
315 queue->desc, queue->avail, queue->used);
320 static int virtio_io_write(struct guest_info *core, uint16_t port, void * src, uint_t length, void * private_data)
322 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
323 int port_idx = port % virtio->io_range_size;
325 PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n", private_data,
326 port, port_idx, length, *(uint32_t *)src);
329 case GUEST_FEATURES_PORT:
331 PrintError("Illegal write length for guest features\n");
334 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
335 PrintDebug("Setting Guest Features to %x\n", virtio->virtio_cfg.guest_features);
338 case VRING_PG_NUM_PORT:
340 PrintError("Illegal write length for page frame number\n");
343 addr_t pfn = *(uint32_t *)src;
344 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
345 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
347 PrintDebug("Virtio Write: pfn: %p, page_addr %p, queue_idx %d\n", (void *)pfn, (void *)page_addr, queue_idx);
351 virtio_setup_queue(core, virtio, &virtio->rx_vq, pfn, page_addr);
354 virtio_setup_queue(core, virtio, &virtio->tx_vq, pfn, page_addr);
357 virtio_setup_queue(core, virtio, &virtio->ctrl_vq, pfn, page_addr);
364 case VRING_Q_SEL_PORT:
365 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
366 if (virtio->virtio_cfg.vring_queue_selector > 2) {
367 PrintError("Virtio NIC device only uses 3 queue, selected %d\n",
368 virtio->virtio_cfg.vring_queue_selector);
373 case VRING_Q_NOTIFY_PORT:
375 uint16_t queue_idx = *(uint16_t *)src;
377 PrintDebug("receive queue notification 0, packet get by Guest\n");
378 } else if (queue_idx == 1){
379 if (handle_pkt_tx(core, virtio) == -1) {
380 PrintError("Could not handle NIC Notification\n");
383 } else if (queue_idx == 2){
384 if (handle_ctrl(core, virtio) == -1) {
385 PrintError("Could not handle NIC Notification\n");
389 PrintError("Virtio NIC device only uses 3 queue, selected %d\n",
395 case VIRTIO_STATUS_PORT:
396 virtio->virtio_cfg.status = *(uint8_t *)src;
397 if (virtio->virtio_cfg.status == 0) {
398 PrintDebug("Resetting device\n");
399 virtio_init_state(virtio);
403 case VIRTIO_ISR_PORT:
404 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
414 static int virtio_io_read(struct guest_info *core, uint16_t port, void * dst, uint_t length, void * private_data)
416 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
417 int port_idx = port % virtio->io_range_size;
418 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
420 PrintDebug("Virtio NIC %p: Read for port %d (index =%d), length=%d\n", private_data,
421 port, port_idx, length);
424 case HOST_FEATURES_PORT:
426 PrintError("Illegal read length for host features\n");
429 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
432 case VRING_PG_NUM_PORT:
434 PrintError("Illegal read length for page frame number\n");
439 *(uint32_t *)dst = virtio->rx_vq.pfn;
442 *(uint32_t *)dst = virtio->tx_vq.pfn;
445 *(uint32_t *)dst = virtio->ctrl_vq.pfn;
452 case VRING_SIZE_PORT:
454 PrintError("Illegal read length for vring size\n");
459 *(uint16_t *)dst = virtio->rx_vq.queue_size;
462 *(uint16_t *)dst = virtio->tx_vq.queue_size;
465 *(uint16_t *)dst = virtio->ctrl_vq.queue_size;
470 PrintDebug("queue index: %d, value=0x%x\n", (int)queue_idx, *(uint16_t *)dst);
473 case VIRTIO_STATUS_PORT:
475 PrintError("Illegal read length for status\n");
478 *(uint8_t *)dst = virtio->virtio_cfg.status;
481 case VIRTIO_ISR_PORT:
482 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
483 virtio->virtio_cfg.pci_isr = 0;
484 v3_pci_lower_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
488 PrintError("Virtio NIC: Read of Unhandled Virtio Read\n");
498 static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) {
499 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
500 struct virtio_queue * q = &(virtio->rx_vq);
501 struct virtio_net_hdr hdr;
502 uint32_t hdr_len = sizeof(struct virtio_net_hdr);
503 uint32_t data_len = size;
508 flags = v3_lock_irqsave(virtio->lock);
510 PrintDebug("VIRTIO NIC: receiving packet to virtio nic %p, size:%d\n", virtio, size);
515 build_receive_header(&hdr, buf, 1);
517 if (q->ring_avail_addr == 0) {
518 PrintError("Queue is not set\n");
523 if (q->last_avail_idx > q->avail->index)
524 q->idx_overflow = true;
525 q->last_avail_idx = q->avail->index;
527 if (q->cur_avail_idx < q->avail->index || (q->idx_overflow && q->cur_avail_idx < q->avail->index+65536)){
529 uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
530 uint16_t buf_idx = 0;
531 struct vring_desc * hdr_desc = NULL;
533 hdr_desc = &(q->desc[hdr_idx]);
534 if (guest_pa_to_host_va(&(virtio->virtio_dev->vm->cores[0]), hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
535 PrintError("Could not translate receive buffer address\n");
540 memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr));
541 if (offset >= data_len) {
542 hdr_desc->flags &= ~VIRTIO_NEXT_FLAG;
545 for (buf_idx = hdr_desc->next; offset < data_len; buf_idx = q->desc[hdr_idx].next) {
546 struct vring_desc * buf_desc = &(q->desc[buf_idx]);
549 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf + offset, data_len - offset);
551 if (offset < data_len) {
552 buf_desc->flags = VIRTIO_NEXT_FLAG;
554 buf_desc->length = len;
557 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
558 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)
561 int last_idx = q->cur_avail_idx;
563 if (q->cur_avail_idx < last_idx)
564 q->idx_overflow = false;
568 #ifdef VIRTIO_NIC_PROFILE
569 PrintError("Virtio NIC: %p, one pkt dropped receieved: %ld, dropped: %ld, sent: %ld curidx: %d, avaiIdx: %d\n",
570 virtio, virtio->pkt_recv, virtio->pkt_drop, virtio->pkt_sent, q->cur_avail_idx, q->avail->index);
574 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
575 PrintDebug("Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line);
576 v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
577 virtio->virtio_cfg.pci_isr = 0x1;
580 #ifdef VIRTIO_NIC_PROFILE
581 if ((virtio->pkt_recv % 10000) == 0){
582 PrintError("Virtio NIC: %p, receieved: %ld, dropped: %ld, sent: %ld\n",
583 virtio, virtio->pkt_recv, virtio->pkt_drop, virtio->pkt_sent);
591 v3_unlock_irqrestore(virtio->lock, flags);
597 static struct v3_device_ops dev_ops = {
604 static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_state * net_state)
606 struct pci_device * pci_dev = NULL;
607 struct v3_pci_bar bars[6];
608 int num_ports = sizeof(struct virtio_config);
609 int tmp_ports = num_ports;
612 // This gets the number of ports, rounded up to a power of 2
613 net_state->io_range_size = 1; // must be a power of 2
614 while (tmp_ports > 0) {
616 net_state->io_range_size <<= 1;
619 // this is to account for any low order bits being set in num_ports
620 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
621 if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
622 net_state->io_range_size >>= 1;
625 for (i = 0; i < 6; i++) {
626 bars[i].type = PCI_BAR_NONE;
629 PrintDebug("Virtio-NIC io_range_size = %d\n", net_state->io_range_size);
631 bars[0].type = PCI_BAR_IO;
632 bars[0].default_base_port = -1;
633 bars[0].num_ports = net_state->io_range_size;
634 bars[0].io_read = virtio_io_read;
635 bars[0].io_write = virtio_io_write;
636 bars[0].private_data = net_state;
638 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
639 0, PCI_AUTO_DEV_NUM, 0,
640 "LNX_VIRTIO_NIC", bars,
641 NULL, NULL, NULL, net_state);
644 PrintError("Virtio NIC: Could not register PCI Device\n");
648 PrintDebug("Virtio NIC: registered to PCI bus\n");
650 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
651 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
654 pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
655 pci_dev->config_header.class = PCI_CLASS_NETWORK;
656 pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
657 pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;
658 pci_dev->config_header.intr_pin = 1;
659 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
661 net_state->pci_dev = pci_dev;
662 net_state->virtio_cfg.host_features = 0; //no features support now
663 net_state->rx_vq.queue_size = QUEUE_SIZE;
664 net_state->tx_vq.queue_size = QUEUE_SIZE;
665 net_state->ctrl_vq.queue_size = CTRL_QUEUE_SIZE;
666 net_state->virtio_dev = virtio;
668 virtio_init_state(net_state);
673 static int connect_fn(struct v3_vm_info * info,
674 void * frontend_data,
675 struct v3_dev_net_ops * ops,
677 void * private_data) {
678 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
679 struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
681 memset(net_state, 0, sizeof(struct virtio_net_state));
682 register_dev(virtio, net_state);
684 net_state->net_ops = ops;
685 net_state->backend_data = private_data;
687 ops->recv = virtio_rx;
688 ops->frontend_data = net_state;
693 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
694 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
695 struct virtio_dev_state * virtio_state = NULL;
696 char * name = v3_cfg_val(cfg, "name");
698 PrintDebug("Virtio NIC: Initializing VIRTIO Network device: %s\n", name);
700 if (pci_bus == NULL) {
701 PrintError("Virtio NIC: VirtIO devices require a PCI Bus");
705 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
706 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
708 INIT_LIST_HEAD(&(virtio_state->dev_list));
709 virtio_state->pci_bus = pci_bus;
710 virtio_state->vm = vm;
712 struct vm_device * dev = v3_allocate_device(name, &dev_ops, virtio_state);
713 if (v3_attach_device(vm, dev) == -1) {
714 PrintError("Virtio NIC: Could not attach device %s\n", name);
718 if (v3_dev_add_net_frontend(vm, name, connect_fn, (void *)virtio_state) == -1) {
719 PrintError("Virtio NIC: Could not register %s as net frontend\n", name);
726 device_register("LNX_VIRTIO_NIC", virtio_init)