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 guest_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 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: Handling Virtio Write, net_state: %p\n", virtio);
141 if (guest_pa_to_host_va(virtio->virtio_dev->vm, 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 virtio_net_state * virtio_state, struct vring_desc * desc, uchar_t * buf, uint_t buf_len)
168 uint8_t * desc_buf = NULL;
170 if (guest_pa_to_host_va(virtio_state->virtio_dev->vm, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
171 PrintError("Could not translate buffer address\n");
174 len = (desc->length < buf_len)?desc->length:buf_len;
175 memcpy(desc_buf, buf, len);
183 static int get_desc_count(struct virtio_queue * q, int index) {
184 struct vring_desc * tmp_desc = &(q->desc[index]);
187 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
188 tmp_desc = &(q->desc[tmp_desc->next]);
195 static int handle_ctrl(struct virtio_net_state * dev) {
199 static int handle_pkt_tx(struct virtio_net_state * virtio_state)
201 struct virtio_queue * q = &(virtio_state->tx_vq);
202 struct virtio_net_hdr * hdr = NULL;
204 if (q->avail->index < q->last_avail_idx) {
205 q->idx_overflow = true;
208 q->last_avail_idx = q->avail->index;
210 while (q->cur_avail_idx < q->avail->index ||
211 (q->idx_overflow && q->cur_avail_idx < (q->avail->index + 65536))) {
212 struct vring_desc * hdr_desc = NULL;
214 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
215 int desc_cnt = get_desc_count(q, desc_idx);
216 uint32_t req_len = 0;
219 hdr_desc = &(q->desc[desc_idx]);
220 if (guest_pa_to_host_va(virtio_state->virtio_dev->vm, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
221 PrintError("Could not translate block header address\n");
225 hdr = (struct virtio_net_hdr*)hdr_addr;
226 desc_idx = hdr_desc->next;
228 for (i = 0; i < desc_cnt - 1; i++) {
229 struct vring_desc * buf_desc = &(q->desc[desc_idx]);
230 if (pkt_tx(virtio_state, buf_desc) == -1) {
231 PrintError("Error handling nic operation\n");
235 req_len += buf_desc->length;
236 desc_idx = buf_desc->next;
238 virtio_state->pkt_sent ++;
240 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
241 q->used->ring[q->used->index % q->queue_size].length = req_len; // What do we set this to????
244 int last_idx = q->cur_avail_idx;
248 if (q->cur_avail_idx < last_idx) {
249 q->idx_overflow = false;
253 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
254 v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev);
255 virtio_state->virtio_cfg.pci_isr = 0x1;
258 #ifdef VIRTIO_NIC_PROFILE
259 if(virtio_state->pkt_sent % 10000 == 0)
260 PrintError("Virtio NIC: %p, pkt_sent: %ld\n", virtio_state, virtio_state->pkt_sent);
266 static int virtio_setup_queue(struct virtio_net_state * virtio_state, struct virtio_queue * queue, addr_t pfn, addr_t page_addr) {
269 queue->ring_desc_addr = page_addr;
270 queue->ring_avail_addr = page_addr + (queue->queue_size * sizeof(struct vring_desc));
271 queue->ring_used_addr = ((queue->ring_avail_addr) +
272 (sizeof(struct vring_avail)) +
273 (queue->queue_size * sizeof(uint16_t)));
275 // round up to next page boundary.
276 queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
277 if (guest_pa_to_host_va(virtio_state->virtio_dev->vm, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
278 PrintError("Could not translate ring descriptor address\n");
282 if (guest_pa_to_host_va(virtio_state->virtio_dev->vm, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
283 PrintError("Could not translate ring available address\n");
287 if (guest_pa_to_host_va(virtio_state->virtio_dev->vm, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
288 PrintError("Could not translate ring used address\n");
292 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
293 (void *)(queue->ring_desc_addr),
294 (void *)(queue->ring_avail_addr),
295 (void *)(queue->ring_used_addr));
297 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
298 queue->desc, queue->avail, queue->used);
303 static int virtio_io_write(struct guest_info *core, uint16_t port, void * src, uint_t length, void * private_data)
305 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
306 int port_idx = port % virtio->io_range_size;
308 PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n", private_data,
309 port, port_idx, length, *(uint32_t *)src);
312 case GUEST_FEATURES_PORT:
314 PrintError("Illegal write length for guest features\n");
317 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
318 PrintDebug("Setting Guest Features to %x\n", virtio->virtio_cfg.guest_features);
321 case VRING_PG_NUM_PORT:
323 PrintError("Illegal write length for page frame number\n");
326 addr_t pfn = *(uint32_t *)src;
327 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
328 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
331 virtio_setup_queue(virtio, &virtio->rx_vq, pfn, page_addr);
334 virtio_setup_queue(virtio, &virtio->tx_vq, pfn, page_addr);
337 virtio_setup_queue(virtio, &virtio->ctrl_vq, pfn, page_addr);
344 case VRING_Q_SEL_PORT:
345 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
346 if (virtio->virtio_cfg.vring_queue_selector > 2) {
347 PrintError("Virtio NIC device only uses 3 queue, selected %d\n",
348 virtio->virtio_cfg.vring_queue_selector);
353 case VRING_Q_NOTIFY_PORT:
355 uint16_t queue_idx = *(uint16_t *)src;
357 PrintDebug("receive queue notification 0, packet get by Guest\n");
358 } else if (queue_idx == 1){
359 if (handle_pkt_tx(virtio) == -1) {
360 PrintError("Could not handle NIC Notification\n");
363 } else if (queue_idx == 2){
364 if (handle_ctrl(virtio) == -1) {
365 PrintError("Could not handle NIC Notification\n");
369 PrintError("Virtio NIC device only uses 3 queue, selected %d\n",
375 case VIRTIO_STATUS_PORT:
376 virtio->virtio_cfg.status = *(uint8_t *)src;
377 if (virtio->virtio_cfg.status == 0) {
378 PrintDebug("Resetting device\n");
379 virtio_init_state(virtio);
383 case VIRTIO_ISR_PORT:
384 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
394 static int virtio_io_read(struct guest_info *core, uint16_t port, void * dst, uint_t length, void * private_data)
396 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
397 int port_idx = port % virtio->io_range_size;
398 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
400 PrintDebug("Virtio NIC %p: Read for port %d (index =%d), length=%d", private_data,
401 port, port_idx, length);
404 case HOST_FEATURES_PORT:
406 PrintError("Illegal read length for host features\n");
409 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
412 case VRING_PG_NUM_PORT:
414 PrintError("Illegal read length for page frame number\n");
419 *(uint32_t *)dst = virtio->rx_vq.pfn;
422 *(uint32_t *)dst = virtio->tx_vq.pfn;
425 *(uint32_t *)dst = virtio->ctrl_vq.pfn;
432 case VRING_SIZE_PORT:
434 PrintError("Illegal read length for vring size\n");
439 *(uint16_t *)dst = virtio->rx_vq.queue_size;
442 *(uint16_t *)dst = virtio->tx_vq.queue_size;
445 *(uint16_t *)dst = virtio->ctrl_vq.queue_size;
450 PrintDebug("queue index: %d, value=0x%x\n", (int)queue_idx, *(uint16_t *)dst);
453 case VIRTIO_STATUS_PORT:
455 PrintError("Illegal read length for status\n");
458 *(uint8_t *)dst = virtio->virtio_cfg.status;
461 case VIRTIO_ISR_PORT:
462 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
463 virtio->virtio_cfg.pci_isr = 0;
464 v3_pci_lower_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
468 PrintError("Virtio NIC: Read of Unhandled Virtio Read\n");
478 static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) {
479 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
480 struct virtio_queue * q = &(virtio->rx_vq);
481 struct virtio_net_hdr hdr;
482 uint32_t hdr_len = sizeof(struct virtio_net_hdr);
483 uint32_t data_len = size;
487 flags = v3_lock_irqsave(virtio->lock);
489 PrintDebug("VIRTIO NIC: sending packet to virtio nic %p, size:%d", virtio, size);
494 build_receive_header(&hdr, buf, 1);
496 if (q->ring_avail_addr == 0) {
497 PrintError("Queue is not set\n");
501 if (q->last_avail_idx > q->avail->index)
502 q->idx_overflow = true;
503 q->last_avail_idx = q->avail->index;
505 if (q->cur_avail_idx < q->avail->index || (q->idx_overflow && q->cur_avail_idx < q->avail->index+65536)){
507 uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
508 uint16_t buf_idx = 0;
509 struct vring_desc * hdr_desc = NULL;
511 hdr_desc = &(q->desc[hdr_idx]);
512 if (guest_pa_to_host_va(virtio->virtio_dev->vm, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
513 PrintError("Could not translate receive buffer address\n");
517 memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr));
518 if (offset >= data_len) {
519 hdr_desc->flags &= ~VIRTIO_NEXT_FLAG;
522 for (buf_idx = hdr_desc->next; offset < data_len; buf_idx = q->desc[hdr_idx].next) {
523 struct vring_desc * buf_desc = &(q->desc[buf_idx]);
526 len = copy_data_to_desc(virtio, buf_desc, buf + offset, data_len - offset);
528 if (offset < data_len) {
529 buf_desc->flags = VIRTIO_NEXT_FLAG;
531 buf_desc->length = len;
534 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
535 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)
538 int last_idx = q->cur_avail_idx;
540 if (q->cur_avail_idx < last_idx)
541 q->idx_overflow = false;
545 #ifdef VIRTIO_NIC_PROFILE
546 PrintError("Virtio NIC: %p, one pkt dropped receieved: %ld, dropped: %ld, sent: %ld curidx: %d, avaiIdx: %d\n",
547 virtio, virtio->pkt_recv, virtio->pkt_drop, virtio->pkt_sent, q->cur_avail_idx, q->avail->index);
551 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
552 PrintDebug("Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line);
553 v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
554 virtio->virtio_cfg.pci_isr = 0x1;
557 #ifdef VIRTIO_NIC_PROFILE
558 if ((virtio->pkt_recv % 10000) == 0){
559 PrintError("Virtio NIC: %p, receieved: %ld, dropped: %ld, sent: %ld\n",
560 virtio, virtio->pkt_recv, virtio->pkt_drop, virtio->pkt_sent);
564 v3_unlock_irqrestore(virtio->lock, flags);
570 static struct v3_device_ops dev_ops = {
577 static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_state * net_state)
579 struct pci_device * pci_dev = NULL;
580 struct v3_pci_bar bars[6];
581 int num_ports = sizeof(struct virtio_config);
582 int tmp_ports = num_ports;
585 // This gets the number of ports, rounded up to a power of 2
586 net_state->io_range_size = 1; // must be a power of 2
587 while (tmp_ports > 0) {
589 net_state->io_range_size <<= 1;
592 // this is to account for any low order bits being set in num_ports
593 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
594 if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
595 net_state->io_range_size >>= 1;
598 for (i = 0; i < 6; i++) {
599 bars[i].type = PCI_BAR_NONE;
602 PrintDebug("Virtio-NIC io_range_size = %d\n", net_state->io_range_size);
604 bars[0].type = PCI_BAR_IO;
605 bars[0].default_base_port = -1;
606 bars[0].num_ports = net_state->io_range_size;
607 bars[0].io_read = virtio_io_read;
608 bars[0].io_write = virtio_io_write;
609 bars[0].private_data = net_state;
611 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
612 0, PCI_AUTO_DEV_NUM, 0,
613 "LNX_VIRTIO_NIC", bars,
614 NULL, NULL, NULL, net_state);
617 PrintError("Virtio NIC: Could not register PCI Device\n");
621 PrintDebug("Virtio NIC: registered to PCI bus\n");
623 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
624 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
627 pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
628 pci_dev->config_header.class = PCI_CLASS_NETWORK;
629 pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
630 pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;
631 pci_dev->config_header.intr_pin = 1;
632 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
634 net_state->pci_dev = pci_dev;
635 net_state->virtio_cfg.host_features = 0; //no features support now
636 net_state->rx_vq.queue_size = QUEUE_SIZE;
637 net_state->tx_vq.queue_size = QUEUE_SIZE;
638 net_state->ctrl_vq.queue_size = CTRL_QUEUE_SIZE;
639 net_state->virtio_dev = virtio;
641 virtio_init_state(net_state);
646 static int connect_fn(struct v3_vm_info * info,
647 void * frontend_data,
648 struct v3_dev_net_ops * ops,
650 void * private_data) {
651 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
652 struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
654 memset(net_state, 0, sizeof(struct virtio_net_state));
655 register_dev(virtio, net_state);
657 net_state->net_ops = ops;
658 net_state->backend_data = private_data;
660 ops->recv = virtio_rx;
661 ops->frontend_data = net_state;
666 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
667 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
668 struct virtio_dev_state * virtio_state = NULL;
669 char * name = v3_cfg_val(cfg, "name");
671 PrintDebug("Virtio NIC: Initializing VIRTIO Network device: %s\n", name);
673 if (pci_bus == NULL) {
674 PrintError("Virtio NIC: VirtIO devices require a PCI Bus");
678 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
679 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
681 INIT_LIST_HEAD(&(virtio_state->dev_list));
682 virtio_state->pci_bus = pci_bus;
684 struct vm_device * dev = v3_allocate_device(name, &dev_ops, virtio_state);
685 if (v3_attach_device(vm, dev) == -1) {
686 PrintError("Virtio NIC: Could not attach device %s\n", name);
690 if (v3_dev_add_net_frontend(vm, name, connect_fn, (void *)virtio_state) == -1) {
691 PrintError("Virtio NIC: Could not register %s as net frontend\n", name);
698 device_register("LNX_VIRTIO_NIC", virtio_init)