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>
30 #include <palacios/vmm_util.h>
32 #include <devices/pci.h>
35 #ifndef CONFIG_DEBUG_VIRTIO_NET
37 #define PrintDebug(fmt, args...)
40 #define VIRTIO_NET_S_LINK_UP 1 /* Link is up */
41 #define VIRTIO_NET_MAX_BUFSIZE (sizeof(struct virtio_net_hdr) + (64 << 10))
43 struct virtio_net_hdr {
46 #define VIRTIO_NET_HDR_GSO_NONE 0 /* Not a GSO frame */
48 uint16_t hdr_len; /* Ethernet + IP + tcp/udp hdrs */
49 uint16_t gso_size; /* Bytes to append to hdr_len per frame */
50 uint16_t csum_start; /* Position to start checksumming from */
51 uint16_t csum_offset; /* Offset after that to place checksum */
52 }__attribute__((packed));
55 #define QUEUE_SIZE 1024
56 #define CTRL_QUEUE_SIZE 64
59 struct virtio_net_config
61 uint8_t mac[ETH_ALEN]; //VIRTIO_NET_F_MAC
63 } __attribute__((packed));
65 struct virtio_dev_state {
66 struct vm_device * pci_bus;
67 struct list_head dev_list;
68 struct v3_vm_info *vm;
71 struct virtio_net_state {
72 struct virtio_net_config net_cfg;
73 struct virtio_config virtio_cfg;
75 struct vm_device * dev;
76 struct pci_device * pci_dev;
79 struct virtio_queue rx_vq; //index 0, rvq in Linux virtio driver, handle packet to guest
80 struct virtio_queue tx_vq; //index 1, svq in Linux virtio driver, handle packet from guest
81 struct virtio_queue ctrl_vq; //index 2, ctrol info from guest
83 ulong_t pkt_sent, pkt_recv, pkt_drop;
85 #if 1 //for temporary performance testing purpose
86 long last_sent_time, last_recv_time;
89 struct v3_dev_net_ops * net_ops;
94 struct virtio_dev_state * virtio_dev;
95 struct list_head dev_link;
98 static int virtio_free(struct vm_device * dev)
104 static int virtio_init_state(struct virtio_net_state * virtio)
106 virtio->rx_vq.ring_desc_addr = 0;
107 virtio->rx_vq.ring_avail_addr = 0;
108 virtio->rx_vq.ring_used_addr = 0;
109 virtio->rx_vq.pfn = 0;
110 virtio->rx_vq.cur_avail_idx = 0;
112 virtio->tx_vq.ring_desc_addr = 0;
113 virtio->tx_vq.ring_avail_addr = 0;
114 virtio->tx_vq.ring_used_addr = 0;
115 virtio->tx_vq.pfn = 0;
116 virtio->tx_vq.cur_avail_idx = 0;
118 virtio->ctrl_vq.ring_desc_addr = 0;
119 virtio->ctrl_vq.ring_avail_addr = 0;
120 virtio->ctrl_vq.ring_used_addr = 0;
121 virtio->ctrl_vq.pfn = 0;
122 virtio->ctrl_vq.cur_avail_idx = 0;
124 virtio->virtio_cfg.host_features = 0;
125 //virtio->virtio_cfg.status = VIRTIO_NET_S_LINK_UP;
126 virtio->virtio_cfg.pci_isr = 0;
128 if (v3_lock_init(&(virtio->lock)) == -1){
129 PrintError("Virtio NIC: Failure to init lock for net_state\n");
132 virtio->pkt_sent = virtio->pkt_recv = virtio->pkt_drop = 0;
137 static int pkt_tx(struct guest_info *core, struct virtio_net_state * virtio, struct vring_desc * buf_desc)
139 uint8_t * buf = NULL;
140 uint32_t len = buf_desc->length;
142 PrintDebug("Virtio NIC: Virtio Pkt Sending, net_state: %p, pkt size: %d\n", virtio, len);
144 if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
145 PrintError("Could not translate buffer address\n");
149 if (virtio->net_ops->send(buf, len, virtio->backend_data, NULL) == -1) {
156 static int build_receive_header(struct virtio_net_hdr * hdr, const void * buf, int raw) {
160 memcpy(hdr, buf, sizeof(struct virtio_net_hdr));
162 memset(hdr, 0, sizeof(struct virtio_net_hdr));
168 static int copy_data_to_desc(struct guest_info *core,
169 struct virtio_net_state * virtio_state,
170 struct vring_desc * desc,
175 uint8_t * desc_buf = NULL;
177 if (v3_gpa_to_hva(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
178 PrintError("Could not translate buffer address\n");
181 len = (desc->length < buf_len)?desc->length:buf_len;
182 memcpy(desc_buf, buf, len);
190 static int get_desc_count(struct virtio_queue * q, int index) {
191 struct vring_desc * tmp_desc = &(q->desc[index]);
194 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
195 tmp_desc = &(q->desc[tmp_desc->next]);
202 static int handle_ctrl(struct guest_info *core, struct virtio_net_state * dev) {
206 static int handle_pkt_tx(struct guest_info *core, struct virtio_net_state * virtio_state)
208 struct virtio_queue * q = &(virtio_state->tx_vq);
209 struct virtio_net_hdr * hdr = NULL;
211 PrintDebug("Virtio NIC: TX: running on cpu: %d\n", V3_Get_CPU());
213 while (q->cur_avail_idx != q->avail->index) {
214 struct vring_desc * hdr_desc = NULL;
216 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
217 int desc_cnt = get_desc_count(q, desc_idx);
218 uint32_t req_len = 0;
221 hdr_desc = &(q->desc[desc_idx]);
222 if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
223 PrintError("Could not translate block header address\n");
227 hdr = (struct virtio_net_hdr*)hdr_addr;
228 desc_idx = hdr_desc->next;
230 for (i = 0; i < desc_cnt - 1; i++) {
231 struct vring_desc * buf_desc = &(q->desc[desc_idx]);
232 if (pkt_tx(core, virtio_state, buf_desc) == -1) {
233 PrintError("Error handling nic operation\n");
237 req_len += buf_desc->length;
238 desc_idx = buf_desc->next;
240 virtio_state->pkt_sent ++;
242 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
243 q->used->ring[q->used->index % q->queue_size].length = req_len; // What do we set this to????
249 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
250 v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev);
251 virtio_state->virtio_cfg.pci_isr = 0x1;
254 #ifdef CONFIG_VNET_PROFILE
255 if (virtio_state->pkt_sent % 50000 == 0){
258 time = cur_time - virtio_state->last_sent_time;
259 PrintError("Virtio NIC: last sent 50000 cycles: %ld\n",time);
260 //PrintError("Virtio NIC: sent: %ld, rxed: %ld, dropped: %ld\n",
261 // virtio_state->pkt_sent,
262 // virtio_state->pkt_recv,
263 // virtio_state->pkt_drop);
264 rdtscll(virtio_state->last_sent_time);
272 static int virtio_setup_queue(struct guest_info *core,
273 struct virtio_net_state * virtio_state,
274 struct virtio_queue * queue,
279 queue->ring_desc_addr = page_addr;
280 queue->ring_avail_addr = page_addr + (queue->queue_size * sizeof(struct vring_desc));
281 queue->ring_used_addr = ((queue->ring_avail_addr) +
282 (sizeof(struct vring_avail)) +
283 (queue->queue_size * sizeof(uint16_t)));
285 // round up to next page boundary.
286 queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
287 if (v3_gpa_to_hva(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
288 PrintError("Could not translate ring descriptor address\n");
292 if (v3_gpa_to_hva(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
293 PrintError("Could not translate ring available address\n");
297 if (v3_gpa_to_hva(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
298 PrintError("Could not translate ring used address\n");
302 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
303 (void *)(queue->ring_desc_addr),
304 (void *)(queue->ring_avail_addr),
305 (void *)(queue->ring_used_addr));
307 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
308 queue->desc, queue->avail, queue->used);
313 static int virtio_io_write(struct guest_info *core, uint16_t port, void * src, uint_t length, void * private_data)
315 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
316 int port_idx = port % virtio->io_range_size;
319 PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n", private_data,
320 port, port_idx, length, *(uint32_t *)src);
323 case GUEST_FEATURES_PORT:
325 PrintError("Illegal write length for guest features\n");
328 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
329 PrintDebug("Setting Guest Features to %x\n", virtio->virtio_cfg.guest_features);
332 case VRING_PG_NUM_PORT:
334 PrintError("Illegal write length for page frame number\n");
337 addr_t pfn = *(uint32_t *)src;
338 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
339 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
341 PrintDebug("Virtio Write: pfn: %p, page_addr %p, queue_idx %d\n", (void *)pfn, (void *)page_addr, queue_idx);
345 virtio_setup_queue(core, virtio, &virtio->rx_vq, pfn, page_addr);
348 virtio_setup_queue(core, virtio, &virtio->tx_vq, pfn, page_addr);
351 virtio_setup_queue(core, virtio, &virtio->ctrl_vq, pfn, page_addr);
358 case VRING_Q_SEL_PORT:
359 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
360 if (virtio->virtio_cfg.vring_queue_selector > 2) {
361 PrintError("Virtio NIC device only uses 3 queue, selected %d\n",
362 virtio->virtio_cfg.vring_queue_selector);
367 case VRING_Q_NOTIFY_PORT:
369 uint16_t queue_idx = *(uint16_t *)src;
371 PrintDebug("receive queue notification 0, packet get by Guest\n");
372 } else if (queue_idx == 1){
373 if (handle_pkt_tx(core, virtio) == -1) {
374 PrintError("Could not handle NIC Notification\n");
377 } else if (queue_idx == 2){
378 if (handle_ctrl(core, virtio) == -1) {
379 PrintError("Could not handle NIC Notification\n");
383 PrintError("Virtio NIC device only uses 3 queue, selected %d\n",
389 case VIRTIO_STATUS_PORT:
390 virtio->virtio_cfg.status = *(uint8_t *)src;
391 if (virtio->virtio_cfg.status == 0) {
392 PrintDebug("Resetting device\n");
393 virtio_init_state(virtio);
397 case VIRTIO_ISR_PORT:
398 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
408 static int virtio_io_read(struct guest_info *core, uint16_t port, void * dst, uint_t length, void * private_data)
410 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
411 int port_idx = port % virtio->io_range_size;
412 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
414 PrintDebug("Virtio NIC %p: Read for port %d (index =%d), length=%d\n", private_data,
415 port, port_idx, length);
418 case HOST_FEATURES_PORT:
420 PrintError("Illegal read length for host features\n");
423 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
426 case VRING_PG_NUM_PORT:
428 PrintError("Illegal read length for page frame number\n");
433 *(uint32_t *)dst = virtio->rx_vq.pfn;
436 *(uint32_t *)dst = virtio->tx_vq.pfn;
439 *(uint32_t *)dst = virtio->ctrl_vq.pfn;
446 case VRING_SIZE_PORT:
448 PrintError("Illegal read length for vring size\n");
453 *(uint16_t *)dst = virtio->rx_vq.queue_size;
456 *(uint16_t *)dst = virtio->tx_vq.queue_size;
459 *(uint16_t *)dst = virtio->ctrl_vq.queue_size;
464 PrintDebug("queue index: %d, value=0x%x\n", (int)queue_idx, *(uint16_t *)dst);
467 case VIRTIO_STATUS_PORT:
469 PrintError("Illegal read length for status\n");
472 *(uint8_t *)dst = virtio->virtio_cfg.status;
475 case VIRTIO_ISR_PORT:
476 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
477 virtio->virtio_cfg.pci_isr = 0;
478 v3_pci_lower_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
482 PrintError("Virtio NIC: Read of Unhandled Virtio Read\n");
492 static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) {
493 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
494 struct virtio_queue * q = &(virtio->rx_vq);
495 struct virtio_net_hdr hdr;
496 uint32_t hdr_len = sizeof(struct virtio_net_hdr);
497 uint32_t data_len = size;
503 flags = v3_lock_irqsave(virtio->lock);
505 PrintDebug("VIRTIO NIC: RX on cpu %d to virtio nic %p, size:%d\n", V3_Get_CPU(), virtio, size);
513 build_receive_header(&hdr, buf, raw);
515 if (q->ring_avail_addr == 0) {
516 PrintError("Queue is not set\n");
520 if (q->cur_avail_idx != q->avail->index){
522 uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
523 uint16_t buf_idx = 0;
524 struct vring_desc * hdr_desc = NULL;
526 hdr_desc = &(q->desc[hdr_idx]);
527 if (v3_gpa_to_hva(&(virtio->virtio_dev->vm->cores[0]), hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
528 PrintError("Could not translate receive buffer address\n");
533 memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr));
534 if (offset >= data_len) {
535 hdr_desc->flags &= ~VIRTIO_NEXT_FLAG;
538 for (buf_idx = hdr_desc->next; offset < data_len; buf_idx = q->desc[hdr_idx].next) {
539 struct vring_desc * buf_desc = &(q->desc[buf_idx]);
542 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf + offset, data_len - offset);
544 if (offset < data_len) {
545 buf_desc->flags = VIRTIO_NEXT_FLAG;
547 buf_desc->length = len;
550 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
551 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)
560 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
561 PrintDebug("Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line);
562 v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
563 virtio->virtio_cfg.pci_isr = 0x1;
570 #ifdef CONFIG_VNET_PROFILE
571 if (virtio->pkt_recv % 50000 == 0){
574 time = cur_time - virtio->last_recv_time;
575 PrintError("Virtio NIC: last recv 50000 cycles: %ld\n",time);
576 //PrintError("Virtio NIC: sent: %ld, rxed: %ld, dropped: %ld\n",
580 rdtscll(virtio->last_recv_time);
584 v3_unlock_irqrestore(virtio->lock, flags);
590 static struct v3_device_ops dev_ops = {
597 static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_state * net_state)
599 struct pci_device * pci_dev = NULL;
600 struct v3_pci_bar bars[6];
601 int num_ports = sizeof(struct virtio_config);
602 int tmp_ports = num_ports;
605 // This gets the number of ports, rounded up to a power of 2
606 net_state->io_range_size = 1; // must be a power of 2
607 while (tmp_ports > 0) {
609 net_state->io_range_size <<= 1;
612 // this is to account for any low order bits being set in num_ports
613 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
614 if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
615 net_state->io_range_size >>= 1;
618 for (i = 0; i < 6; i++) {
619 bars[i].type = PCI_BAR_NONE;
622 PrintDebug("Virtio-NIC io_range_size = %d\n", net_state->io_range_size);
624 bars[0].type = PCI_BAR_IO;
625 bars[0].default_base_port = -1;
626 bars[0].num_ports = net_state->io_range_size;
627 bars[0].io_read = virtio_io_read;
628 bars[0].io_write = virtio_io_write;
629 bars[0].private_data = net_state;
631 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
632 0, PCI_AUTO_DEV_NUM, 0,
633 "LNX_VIRTIO_NIC", bars,
634 NULL, NULL, NULL, net_state);
637 PrintError("Virtio NIC: Could not register PCI Device\n");
641 PrintDebug("Virtio NIC: registered to PCI bus\n");
643 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
644 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
647 pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
648 pci_dev->config_header.class = PCI_CLASS_NETWORK;
649 pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
650 pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;
651 pci_dev->config_header.intr_pin = 1;
652 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
654 net_state->pci_dev = pci_dev;
655 net_state->virtio_cfg.host_features = 0; //no features support now
656 net_state->rx_vq.queue_size = QUEUE_SIZE;
657 net_state->tx_vq.queue_size = QUEUE_SIZE;
658 net_state->ctrl_vq.queue_size = CTRL_QUEUE_SIZE;
659 net_state->virtio_dev = virtio;
661 virtio_init_state(net_state);
666 static int connect_fn(struct v3_vm_info * info,
667 void * frontend_data,
668 struct v3_dev_net_ops * ops,
670 void * private_data) {
671 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
672 struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
674 memset(net_state, 0, sizeof(struct virtio_net_state));
675 register_dev(virtio, net_state);
677 net_state->net_ops = ops;
678 net_state->backend_data = private_data;
680 ops->recv = virtio_rx;
681 ops->frontend_data = net_state;
686 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
687 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
688 struct virtio_dev_state * virtio_state = NULL;
689 char * name = v3_cfg_val(cfg, "name");
691 PrintDebug("Virtio NIC: Initializing VIRTIO Network device: %s\n", name);
693 if (pci_bus == NULL) {
694 PrintError("Virtio NIC: VirtIO devices require a PCI Bus");
698 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
699 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
701 INIT_LIST_HEAD(&(virtio_state->dev_list));
702 virtio_state->pci_bus = pci_bus;
703 virtio_state->vm = vm;
705 struct vm_device * dev = v3_allocate_device(name, &dev_ops, virtio_state);
706 if (v3_attach_device(vm, dev) == -1) {
707 PrintError("Virtio NIC: Could not attach device %s\n", name);
711 if (v3_dev_add_net_frontend(vm, name, connect_fn, (void *)virtio_state) == -1) {
712 PrintError("Virtio NIC: Could not register %s as net frontend\n", name);
719 device_register("LNX_VIRTIO_NIC", virtio_init)