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 struct v3_dev_net_ops * net_ops;
90 struct virtio_dev_state * virtio_dev;
91 struct list_head dev_link;
94 static int virtio_free(struct vm_device * dev)
100 static int virtio_init_state(struct virtio_net_state * virtio)
102 virtio->rx_vq.ring_desc_addr = 0;
103 virtio->rx_vq.ring_avail_addr = 0;
104 virtio->rx_vq.ring_used_addr = 0;
105 virtio->rx_vq.pfn = 0;
106 virtio->rx_vq.cur_avail_idx = 0;
108 virtio->tx_vq.ring_desc_addr = 0;
109 virtio->tx_vq.ring_avail_addr = 0;
110 virtio->tx_vq.ring_used_addr = 0;
111 virtio->tx_vq.pfn = 0;
112 virtio->tx_vq.cur_avail_idx = 0;
114 virtio->ctrl_vq.ring_desc_addr = 0;
115 virtio->ctrl_vq.ring_avail_addr = 0;
116 virtio->ctrl_vq.ring_used_addr = 0;
117 virtio->ctrl_vq.pfn = 0;
118 virtio->ctrl_vq.cur_avail_idx = 0;
120 virtio->virtio_cfg.host_features = 0;
121 //virtio->virtio_cfg.status = VIRTIO_NET_S_LINK_UP;
122 virtio->virtio_cfg.pci_isr = 0;
124 if (v3_lock_init(&(virtio->lock)) == -1){
125 PrintError("Virtio NIC: Failure to init lock for net_state\n");
128 virtio->pkt_sent = virtio->pkt_recv = virtio->pkt_drop = 0;
133 static int pkt_tx(struct guest_info *core, struct virtio_net_state * virtio, struct vring_desc * buf_desc)
135 uint8_t * buf = NULL;
136 uint32_t len = buf_desc->length;
138 PrintDebug("Virtio NIC: Virtio Pkt Sending, net_state: %p, pkt size: %d\n", virtio, len);
140 if (guest_pa_to_host_va(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
141 PrintError("Could not translate buffer address\n");
145 if (virtio->net_ops->send(buf, len, virtio->backend_data, NULL) == -1) {
152 static int build_receive_header(struct virtio_net_hdr * hdr, const void * buf, int raw) {
156 memcpy(hdr, buf, sizeof(struct virtio_net_hdr));
158 memset(hdr, 0, sizeof(struct virtio_net_hdr));
164 static int copy_data_to_desc(struct guest_info *core,
165 struct virtio_net_state * virtio_state,
166 struct vring_desc * desc,
171 uint8_t * desc_buf = NULL;
173 if (guest_pa_to_host_va(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
174 PrintError("Could not translate buffer address\n");
177 len = (desc->length < buf_len)?desc->length:buf_len;
178 memcpy(desc_buf, buf, len);
186 static int get_desc_count(struct virtio_queue * q, int index) {
187 struct vring_desc * tmp_desc = &(q->desc[index]);
190 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
191 tmp_desc = &(q->desc[tmp_desc->next]);
198 static int handle_ctrl(struct guest_info *core, struct virtio_net_state * dev) {
202 static int handle_pkt_tx(struct guest_info *core, struct virtio_net_state * virtio_state)
204 struct virtio_queue * q = &(virtio_state->tx_vq);
205 struct virtio_net_hdr * hdr = NULL;
207 PrintDebug("Virtio NIC: TX: running on cpu: %d\n", V3_Get_CPU());
209 while (q->cur_avail_idx != q->avail->index) {
210 struct vring_desc * hdr_desc = NULL;
212 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
213 int desc_cnt = get_desc_count(q, desc_idx);
214 uint32_t req_len = 0;
217 hdr_desc = &(q->desc[desc_idx]);
218 if (guest_pa_to_host_va(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
219 PrintError("Could not translate block header address\n");
223 hdr = (struct virtio_net_hdr*)hdr_addr;
224 desc_idx = hdr_desc->next;
226 for (i = 0; i < desc_cnt - 1; i++) {
227 struct vring_desc * buf_desc = &(q->desc[desc_idx]);
228 if (pkt_tx(core, virtio_state, buf_desc) == -1) {
229 PrintError("Error handling nic operation\n");
233 req_len += buf_desc->length;
234 desc_idx = buf_desc->next;
236 virtio_state->pkt_sent ++;
238 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
239 q->used->ring[q->used->index % q->queue_size].length = req_len; // What do we set this to????
245 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
246 v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev);
247 virtio_state->virtio_cfg.pci_isr = 0x1;
250 #ifdef CONFIG_VNET_PROFILE
251 if (virtio_state->pkt_sent % 10000 == 0){
252 PrintError("Virtio NIC: sent: %ld, rxed: %ld, dropped: %ld\n",
253 virtio_state->pkt_sent,
254 virtio_state->pkt_recv,
255 virtio_state->pkt_drop);
263 static int virtio_setup_queue(struct guest_info *core,
264 struct virtio_net_state * virtio_state,
265 struct virtio_queue * queue,
270 queue->ring_desc_addr = page_addr;
271 queue->ring_avail_addr = page_addr + (queue->queue_size * sizeof(struct vring_desc));
272 queue->ring_used_addr = ((queue->ring_avail_addr) +
273 (sizeof(struct vring_avail)) +
274 (queue->queue_size * sizeof(uint16_t)));
276 // round up to next page boundary.
277 queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
278 if (guest_pa_to_host_va(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
279 PrintError("Could not translate ring descriptor address\n");
283 if (guest_pa_to_host_va(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
284 PrintError("Could not translate ring available address\n");
288 if (guest_pa_to_host_va(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
289 PrintError("Could not translate ring used address\n");
293 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
294 (void *)(queue->ring_desc_addr),
295 (void *)(queue->ring_avail_addr),
296 (void *)(queue->ring_used_addr));
298 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
299 queue->desc, queue->avail, queue->used);
304 static int virtio_io_write(struct guest_info *core, uint16_t port, void * src, uint_t length, void * private_data)
306 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
307 int port_idx = port % virtio->io_range_size;
310 PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n", private_data,
311 port, port_idx, length, *(uint32_t *)src);
314 case GUEST_FEATURES_PORT:
316 PrintError("Illegal write length for guest features\n");
319 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
320 PrintDebug("Setting Guest Features to %x\n", virtio->virtio_cfg.guest_features);
323 case VRING_PG_NUM_PORT:
325 PrintError("Illegal write length for page frame number\n");
328 addr_t pfn = *(uint32_t *)src;
329 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
330 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
332 PrintDebug("Virtio Write: pfn: %p, page_addr %p, queue_idx %d\n", (void *)pfn, (void *)page_addr, queue_idx);
336 virtio_setup_queue(core, virtio, &virtio->rx_vq, pfn, page_addr);
339 virtio_setup_queue(core, virtio, &virtio->tx_vq, pfn, page_addr);
342 virtio_setup_queue(core, virtio, &virtio->ctrl_vq, pfn, page_addr);
349 case VRING_Q_SEL_PORT:
350 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
351 if (virtio->virtio_cfg.vring_queue_selector > 2) {
352 PrintError("Virtio NIC device only uses 3 queue, selected %d\n",
353 virtio->virtio_cfg.vring_queue_selector);
358 case VRING_Q_NOTIFY_PORT:
360 uint16_t queue_idx = *(uint16_t *)src;
362 PrintDebug("receive queue notification 0, packet get by Guest\n");
363 } else if (queue_idx == 1){
364 if (handle_pkt_tx(core, virtio) == -1) {
365 PrintError("Could not handle NIC Notification\n");
368 } else if (queue_idx == 2){
369 if (handle_ctrl(core, virtio) == -1) {
370 PrintError("Could not handle NIC Notification\n");
374 PrintError("Virtio NIC device only uses 3 queue, selected %d\n",
380 case VIRTIO_STATUS_PORT:
381 virtio->virtio_cfg.status = *(uint8_t *)src;
382 if (virtio->virtio_cfg.status == 0) {
383 PrintDebug("Resetting device\n");
384 virtio_init_state(virtio);
388 case VIRTIO_ISR_PORT:
389 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
399 static int virtio_io_read(struct guest_info *core, uint16_t port, void * dst, uint_t length, void * private_data)
401 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
402 int port_idx = port % virtio->io_range_size;
403 uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
405 PrintDebug("Virtio NIC %p: Read for port %d (index =%d), length=%d\n", private_data,
406 port, port_idx, length);
409 case HOST_FEATURES_PORT:
411 PrintError("Illegal read length for host features\n");
414 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
417 case VRING_PG_NUM_PORT:
419 PrintError("Illegal read length for page frame number\n");
424 *(uint32_t *)dst = virtio->rx_vq.pfn;
427 *(uint32_t *)dst = virtio->tx_vq.pfn;
430 *(uint32_t *)dst = virtio->ctrl_vq.pfn;
437 case VRING_SIZE_PORT:
439 PrintError("Illegal read length for vring size\n");
444 *(uint16_t *)dst = virtio->rx_vq.queue_size;
447 *(uint16_t *)dst = virtio->tx_vq.queue_size;
450 *(uint16_t *)dst = virtio->ctrl_vq.queue_size;
455 PrintDebug("queue index: %d, value=0x%x\n", (int)queue_idx, *(uint16_t *)dst);
458 case VIRTIO_STATUS_PORT:
460 PrintError("Illegal read length for status\n");
463 *(uint8_t *)dst = virtio->virtio_cfg.status;
466 case VIRTIO_ISR_PORT:
467 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
468 virtio->virtio_cfg.pci_isr = 0;
469 v3_pci_lower_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
473 PrintError("Virtio NIC: Read of Unhandled Virtio Read\n");
483 static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) {
484 struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
485 struct virtio_queue * q = &(virtio->rx_vq);
486 struct virtio_net_hdr hdr;
487 uint32_t hdr_len = sizeof(struct virtio_net_hdr);
488 uint32_t data_len = size;
494 flags = v3_lock_irqsave(virtio->lock);
496 PrintDebug("VIRTIO NIC: RX on cpu %d to virtio nic %p, size:%d\n", V3_Get_CPU(), virtio, size);
504 build_receive_header(&hdr, buf, raw);
506 if (q->ring_avail_addr == 0) {
507 PrintError("Queue is not set\n");
511 if (q->cur_avail_idx != q->avail->index){
513 uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
514 uint16_t buf_idx = 0;
515 struct vring_desc * hdr_desc = NULL;
517 hdr_desc = &(q->desc[hdr_idx]);
518 if (guest_pa_to_host_va(&(virtio->virtio_dev->vm->cores[0]), hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
519 PrintError("Could not translate receive buffer address\n");
524 memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr));
525 if (offset >= data_len) {
526 hdr_desc->flags &= ~VIRTIO_NEXT_FLAG;
529 for (buf_idx = hdr_desc->next; offset < data_len; buf_idx = q->desc[hdr_idx].next) {
530 struct vring_desc * buf_desc = &(q->desc[buf_idx]);
533 len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf + offset, data_len - offset);
535 if (offset < data_len) {
536 buf_desc->flags = VIRTIO_NEXT_FLAG;
538 buf_desc->length = len;
541 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
542 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)
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;
561 #ifdef CONFIG_VNET_PROFILE
562 if (virtio->pkt_recv % 100000 == 0){
563 PrintError("Virtio NIC: sent: %ld, rxed: %ld, dropped: %ld\n",
570 v3_unlock_irqrestore(virtio->lock, flags);
576 static struct v3_device_ops dev_ops = {
583 static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_state * net_state)
585 struct pci_device * pci_dev = NULL;
586 struct v3_pci_bar bars[6];
587 int num_ports = sizeof(struct virtio_config);
588 int tmp_ports = num_ports;
591 // This gets the number of ports, rounded up to a power of 2
592 net_state->io_range_size = 1; // must be a power of 2
593 while (tmp_ports > 0) {
595 net_state->io_range_size <<= 1;
598 // this is to account for any low order bits being set in num_ports
599 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
600 if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
601 net_state->io_range_size >>= 1;
604 for (i = 0; i < 6; i++) {
605 bars[i].type = PCI_BAR_NONE;
608 PrintDebug("Virtio-NIC io_range_size = %d\n", net_state->io_range_size);
610 bars[0].type = PCI_BAR_IO;
611 bars[0].default_base_port = -1;
612 bars[0].num_ports = net_state->io_range_size;
613 bars[0].io_read = virtio_io_read;
614 bars[0].io_write = virtio_io_write;
615 bars[0].private_data = net_state;
617 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
618 0, PCI_AUTO_DEV_NUM, 0,
619 "LNX_VIRTIO_NIC", bars,
620 NULL, NULL, NULL, net_state);
623 PrintError("Virtio NIC: Could not register PCI Device\n");
627 PrintDebug("Virtio NIC: registered to PCI bus\n");
629 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
630 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
633 pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
634 pci_dev->config_header.class = PCI_CLASS_NETWORK;
635 pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
636 pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;
637 pci_dev->config_header.intr_pin = 1;
638 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
640 net_state->pci_dev = pci_dev;
641 net_state->virtio_cfg.host_features = 0; //no features support now
642 net_state->rx_vq.queue_size = QUEUE_SIZE;
643 net_state->tx_vq.queue_size = QUEUE_SIZE;
644 net_state->ctrl_vq.queue_size = CTRL_QUEUE_SIZE;
645 net_state->virtio_dev = virtio;
647 virtio_init_state(net_state);
652 static int connect_fn(struct v3_vm_info * info,
653 void * frontend_data,
654 struct v3_dev_net_ops * ops,
656 void * private_data) {
657 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
658 struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
660 memset(net_state, 0, sizeof(struct virtio_net_state));
661 register_dev(virtio, net_state);
663 net_state->net_ops = ops;
664 net_state->backend_data = private_data;
666 ops->recv = virtio_rx;
667 ops->frontend_data = net_state;
672 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
673 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
674 struct virtio_dev_state * virtio_state = NULL;
675 char * name = v3_cfg_val(cfg, "name");
677 PrintDebug("Virtio NIC: Initializing VIRTIO Network device: %s\n", name);
679 if (pci_bus == NULL) {
680 PrintError("Virtio NIC: VirtIO devices require a PCI Bus");
684 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
685 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
687 INIT_LIST_HEAD(&(virtio_state->dev_list));
688 virtio_state->pci_bus = pci_bus;
689 virtio_state->vm = vm;
691 struct vm_device * dev = v3_allocate_device(name, &dev_ops, virtio_state);
692 if (v3_attach_device(vm, dev) == -1) {
693 PrintError("Virtio NIC: Could not attach device %s\n", name);
697 if (v3_dev_add_net_frontend(vm, name, connect_fn, (void *)virtio_state) == -1) {
698 PrintError("Virtio NIC: Could not register %s as net frontend\n", name);
705 device_register("LNX_VIRTIO_NIC", virtio_init)