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, Jack Lange <jarusl@cs.northwestern.edu>
11 * Copyright (c) 2008, Lei Xia <lxia@cs.northwestern.edu>
12 * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org>
13 * All rights reserved.
15 * Author: Jack Lange <jarusl@cs.northwestern.edu>
16 * Lei Xia <lxia@cs.northwestern.edu>
18 * This is free software. You are permitted to use,
19 * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
22 #include <palacios/vmm.h>
23 #include <palacios/vmm_dev_mgr.h>
24 #include <palacios/vm_guest_mem.h>
25 #include <devices/lnx_virtio_pci.h>
26 #include <vnet/vnet.h>
27 #include <palacios/vmm_sprintf.h>
28 #include <devices/pci.h>
31 #ifndef V3_CONFIG_DEBUG_LINUX_VIRTIO_VNET
33 #define PrintDebug(fmt, args...)
37 #define QUEUE_SIZE 4096
38 #define CMD_QUEUE_SIZE 128
44 } __attribute__((packed));
51 struct virtio_vnet_state {
52 struct v3_vm_info * vm;
53 struct vnet_config vnet_cfg;
54 struct virtio_config virtio_cfg;
56 struct vm_device * pci_bus;
57 struct pci_device * pci_dev;
59 struct virtio_queue queue[NUM_QUEUES];
61 struct virtio_queue * cur_queue;
66 ulong_t pkt_sent, pkt_recv, pkt_drop, tx_exit, rx_exit, total_exit;
70 #define VNET_GET_ROUTES 10
71 #define VNET_ADD_ROUTE 11
72 #define VNET_DEL_ROUTE 12
74 #define VNET_GET_LINKS 20
75 #define VNET_ADD_LINK 21
76 #define VNET_DEL_LINK 22
78 struct vnet_ctrl_hdr {
81 } __attribute__((packed));
84 struct vnet_bridge_pkt {
87 uint8_t pkt[ETHERNET_PACKET_LEN];
88 }__attribute__((packed));
91 static int virtio_reset(struct virtio_vnet_state * vnet_state) {
93 memset(vnet_state->queue, 0, sizeof(struct virtio_queue) * NUM_QUEUES);
95 vnet_state->cur_queue = &(vnet_state->queue[0]);
97 vnet_state->virtio_cfg.status = 0;
98 vnet_state->virtio_cfg.pci_isr = 0;
100 vnet_state->queue[0].queue_size = CMD_QUEUE_SIZE;
101 vnet_state->queue[1].queue_size = QUEUE_SIZE;
102 vnet_state->queue[2].queue_size = QUEUE_SIZE;
104 memset(&(vnet_state->vnet_cfg), 0, sizeof(struct vnet_config));
105 v3_lock_init(&(vnet_state->lock));
112 static int get_desc_count(struct virtio_queue * q, int index) {
113 struct vring_desc * tmp_desc = &(q->desc[index]);
116 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
117 tmp_desc = &(q->desc[tmp_desc->next]);
125 static int handle_cmd_kick(struct guest_info * core,
126 struct virtio_vnet_state * vnet_state) {
127 struct virtio_queue * q = &(vnet_state->queue[0]);
129 PrintDebug("VNET Bridge: Handling command queue\n");
131 while (q->cur_avail_idx != q->avail->index) {
132 struct vring_desc * hdr_desc = NULL;
133 struct vring_desc * buf_desc = NULL;
134 struct vring_desc * status_desc = NULL;
135 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
136 uint16_t desc_cnt = get_desc_count(q, desc_idx);
137 struct vnet_ctrl_hdr * hdr = NULL;
140 uint8_t * status_ptr = NULL;
143 PrintDebug("VNET Bridge: CMD: Descriptor Count=%d, index=%d, desc_idx=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE, desc_idx);
146 PrintError("VNET Bridge cmd must include at least 3 descriptors (cnt=%d)\n", desc_cnt);
150 hdr_desc = &(q->desc[desc_idx]);
152 if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, (addr_t *)&hdr) == -1) {
153 PrintError("Could not translate VirtioVNET header address\n");
157 desc_idx = hdr_desc->next;
159 if (hdr->cmd_type == VNET_ADD_ROUTE) {
160 for (i = 0; i < hdr->num_cmds; i++) {
161 uint8_t tmp_status = 0;
162 struct v3_vnet_route * route = NULL;
164 buf_desc = &(q->desc[desc_idx]);
166 if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(route)) == -1) {
167 PrintError("Could not translate route address\n");
171 tmp_status = v3_vnet_add_route(*route);
172 if (tmp_status != 0) {
173 PrintError("Error adding VNET ROUTE\n");
178 PrintDebug("VNET Route Added\n");
180 xfer_len += buf_desc->length;
181 desc_idx = buf_desc->next;
186 status_desc = &(q->desc[desc_idx]);
188 if (v3_gpa_to_hva(core, status_desc->addr_gpa, (addr_t *)&status_ptr) == -1) {
189 PrintError("VirtioVNET Error could not translate status address\n");
193 xfer_len += status_desc->length;
194 *status_ptr = status;
196 PrintDebug("Transferred %d bytes (xfer_len)\n", xfer_len);
197 q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
198 q->used->ring[q->used->index % QUEUE_SIZE].length = xfer_len; // set to total inbound xfer length
205 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
206 PrintDebug("Raising IRQ %d\n", vnet_state->pci_dev->config_header.intr_line);
207 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
208 vnet_state->virtio_cfg.pci_isr = 1;
216 static int vnet_pkt_input_cb(struct v3_vm_info * vm,
217 struct v3_vnet_pkt * pkt,
218 void * private_data){
219 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
220 struct virtio_queue * q = &(vnet_state->queue[RECV_QUEUE]);
224 flags = v3_lock_irqsave(vnet_state->lock);
226 if (q->ring_avail_addr == 0) {
227 PrintError("Queue is not set\n");
232 if (q->cur_avail_idx != q->avail->index) {
233 uint16_t pkt_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
234 struct vring_desc * pkt_desc = NULL;
235 struct vnet_bridge_pkt * virtio_pkt = NULL;
237 pkt_desc = &(q->desc[pkt_idx]);
239 if (v3_gpa_to_hva(&(vm->cores[0]), pkt_desc->addr_gpa, (addr_t *)&(virtio_pkt)) == -1) {
240 PrintError("Could not translate buffer address\n");
244 PrintDebug("VNET Bridge: RX: pkt sent to guest pkt size: %d, dst link: %d\n", pkt->size, pkt->dst_id);
246 // Fill in dst packet buffer
247 virtio_pkt->link_id = pkt->dst_id;
248 virtio_pkt->pkt_size = pkt->size;
249 memcpy(virtio_pkt->pkt, pkt->data, pkt->size);
251 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
252 q->used->ring[q->used->index % q->queue_size].length = sizeof(struct vnet_bridge_pkt);
257 vnet_state->pkt_drop ++;
260 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
261 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
262 vnet_state->virtio_cfg.pci_isr = 0x1;
263 PrintDebug("Raising IRQ %d\n", vnet_state->pci_dev->config_header.intr_line);
270 v3_unlock_irqrestore(vnet_state->lock, flags);
275 static int do_tx_pkts(struct guest_info * core,
276 struct virtio_vnet_state * vnet_state)
278 struct virtio_queue * q = &(vnet_state->queue[XMIT_QUEUE]);
281 if (q->ring_avail_addr == 0) {
285 while (q->cur_avail_idx != q->avail->index) {
286 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
287 struct vring_desc * pkt_desc = NULL;
288 struct vnet_bridge_pkt * virtio_pkt = NULL;
290 pkt_desc = &(q->desc[desc_idx]);
292 if (v3_gpa_to_hva(core, pkt_desc->addr_gpa, (addr_t *)&(virtio_pkt)) == -1) {
293 PrintError("Could not translate buffer address\n");
297 //PrintDebug("VNET Bridge: TX: on cpu %d pkt size: %d, dst link: %d\n", cpu, virtio_pkt->pkt_size, virtio_pkt->link_id);
299 struct v3_vnet_pkt pkt;
300 pkt.size = virtio_pkt->pkt_size;
301 pkt.src_type = LINK_EDGE;
303 memcpy(pkt.header, virtio_pkt->pkt, ETHERNET_HEADER_LEN);
304 pkt.data = virtio_pkt->pkt;
306 v3_vnet_send_pkt(&pkt, NULL, 1);
308 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
309 q->used->ring[q->used->index % q->queue_size].length = pkt_desc->length; // What do we set this to????
312 vnet_state->pkt_sent ++;
322 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
323 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
324 vnet_state->virtio_cfg.pci_isr = 0x1;
330 static void vnet_virtio_poll(struct v3_vm_info * vm, void * private_data){
331 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
333 if(vm == vnet_state->vm){
334 do_tx_pkts(&(vm->cores[0]), vnet_state);
338 static int handle_rx_queue_kick(struct guest_info *core,
339 struct virtio_vnet_state * vnet_state)
344 static int vnet_virtio_io_write(struct guest_info * core,
345 uint16_t port, void * src,
346 uint_t length, void * private_data) {
347 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
348 int port_idx = port % vnet_state->io_range_size;
350 PrintDebug("VNET Bridge: VIRTIO VNET Write for port %d len=%d, value=%x\n",
351 port, length, *(uint32_t *)src);
353 vnet_state->total_exit ++;
355 case GUEST_FEATURES_PORT:
357 PrintError("Illegal write length for guest features\n");
360 vnet_state->virtio_cfg.guest_features = *(uint32_t *)src;
363 case VRING_PG_NUM_PORT:
365 addr_t pfn = *(uint32_t *)src;
366 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
368 vnet_state->cur_queue->pfn = pfn;
370 vnet_state->cur_queue->ring_desc_addr = page_addr ;
371 vnet_state->cur_queue->ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
372 vnet_state->cur_queue->ring_used_addr = ( vnet_state->cur_queue->ring_avail_addr + \
373 sizeof(struct vring_avail) + \
374 (QUEUE_SIZE * sizeof(uint16_t)));
376 // round up to next page boundary.
377 vnet_state->cur_queue->ring_used_addr = (vnet_state->cur_queue->ring_used_addr + 0xfff) & ~0xfff;
379 if (v3_gpa_to_hva(core, vnet_state->cur_queue->ring_desc_addr, (addr_t *)&(vnet_state->cur_queue->desc)) == -1) {
380 PrintError("Could not translate ring descriptor address\n");
384 if (v3_gpa_to_hva(core, vnet_state->cur_queue->ring_avail_addr, (addr_t *)&(vnet_state->cur_queue->avail)) == -1) {
385 PrintError("Could not translate ring available address\n");
389 if (v3_gpa_to_hva(core, vnet_state->cur_queue->ring_used_addr, (addr_t *)&(vnet_state->cur_queue->used)) == -1) {
390 PrintError("Could not translate ring used address\n");
394 PrintDebug("VNET Bridge: RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
395 (void *)(vnet_state->cur_queue->ring_desc_addr),
396 (void *)(vnet_state->cur_queue->ring_avail_addr),
397 (void *)(vnet_state->cur_queue->ring_used_addr));
399 PrintDebug("VNET Bridge: RingDesc=%p, Avail=%p, Used=%p\n",
400 vnet_state->cur_queue->desc, vnet_state->cur_queue->avail, vnet_state->cur_queue->used);
402 if(vnet_state->queue[RECV_QUEUE].avail != NULL){
403 vnet_state->ready = 1;
404 vnet_state->queue[RECV_QUEUE].used->flags |= VRING_NO_NOTIFY_FLAG;
407 //No notify when there is pkt tx from guest
408 //palacios will do the polling
409 if(vnet_state->queue[XMIT_QUEUE].used != NULL){
410 vnet_state->queue[XMIT_QUEUE].used->flags |= VRING_NO_NOTIFY_FLAG;
413 PrintError("Illegal write length for page frame number\n");
417 case VRING_Q_SEL_PORT:
418 vnet_state->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
420 if (vnet_state->virtio_cfg.vring_queue_selector > NUM_QUEUES) {
421 PrintError("VNET Bridge device has no qeueues. Selected %d\n",
422 vnet_state->virtio_cfg.vring_queue_selector);
426 vnet_state->cur_queue = &(vnet_state->queue[vnet_state->virtio_cfg.vring_queue_selector]);
429 case VRING_Q_NOTIFY_PORT: {
430 uint16_t queue_idx = *(uint16_t *)src;
432 PrintDebug("VNET Bridge: Handling Kick\n");
434 if (queue_idx == 0) {
435 if (handle_cmd_kick(core, vnet_state) == -1) {
436 PrintError("Could not handle Virtio VNET Control command\n");
439 } else if (queue_idx == 1) {
440 if (do_tx_pkts(core, vnet_state) == -1){
441 PrintError("Could not handle Virtio VNET TX\n");
444 PrintError("Notify on TX\n");
445 } else if (queue_idx == 2) {
446 if (handle_rx_queue_kick(core, vnet_state) == -1){
447 PrintError("Could not handle Virtio RX buffer refills Kick\n");
450 vnet_state->rx_exit ++;
452 PrintError("VNET Bridge: Kick on invalid queue (%d)\n", queue_idx);
458 case VIRTIO_STATUS_PORT:
459 vnet_state->virtio_cfg.status = *(uint8_t *)src;
461 if (vnet_state->virtio_cfg.status == 0) {
462 PrintDebug("VNET Bridge: Resetting device\n");
463 virtio_reset(vnet_state);
468 case VIRTIO_ISR_PORT:
469 vnet_state->virtio_cfg.pci_isr = *(uint8_t *)src;
480 static int vnet_virtio_io_read(struct guest_info * core,
481 uint16_t port, void * dst,
482 uint_t length, void * private_data) {
484 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
485 int port_idx = port % vnet_state->io_range_size;
488 case HOST_FEATURES_PORT:
490 PrintError("Illegal read length for host features\n");
494 *(uint32_t *)dst = vnet_state->virtio_cfg.host_features;
497 case VRING_PG_NUM_PORT:
499 PrintError("Illegal read length for page frame number\n");
503 *(uint32_t *)dst = vnet_state->cur_queue->pfn;
506 case VRING_SIZE_PORT:
508 PrintError("Illegal read length for vring size\n");
512 *(uint16_t *)dst = vnet_state->cur_queue->queue_size;
516 case VIRTIO_STATUS_PORT:
518 PrintError("Illegal read length for status\n");
522 *(uint8_t *)dst = vnet_state->virtio_cfg.status;
525 case VIRTIO_ISR_PORT:
526 *(uint8_t *)dst = vnet_state->virtio_cfg.pci_isr;
527 vnet_state->virtio_cfg.pci_isr = 0;
528 v3_pci_lower_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
532 if ( (port_idx >= sizeof(struct virtio_config)) &&
533 (port_idx < (sizeof(struct virtio_config) + sizeof(struct vnet_config))) ) {
534 int cfg_offset = port_idx - sizeof(struct virtio_config);
535 uint8_t * cfg_ptr = (uint8_t *)&(vnet_state->vnet_cfg);
537 memcpy(dst, cfg_ptr + cfg_offset, length);
540 PrintError("Read of Unhandled Virtio Read\n");
551 static int virtio_free(struct virtio_vnet_state * vnet_state) {
553 // unregister from PCI
560 static struct v3_device_ops dev_ops = {
561 .free = (int (*)(void *))virtio_free,
564 static int dev_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
565 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
566 struct virtio_vnet_state * vnet_state = NULL;
567 struct pci_device * pci_dev = NULL;
568 char * name = v3_cfg_val(cfg, "name");
570 PrintDebug("VNET Bridge: Initializing VNET Bridge Control device: %s\n", name);
572 if (pci_bus == NULL) {
573 PrintError("VNET Bridge device require a PCI Bus");
577 vnet_state = (struct virtio_vnet_state *)V3_Malloc(sizeof(struct virtio_vnet_state));
578 memset(vnet_state, 0, sizeof(struct virtio_vnet_state));
582 struct vm_device * dev = v3_add_device(vm, name, &dev_ops, vnet_state);
585 PrintError("Could not attach device %s\n", name);
591 // PCI initialization
593 struct v3_pci_bar bars[6];
594 int num_ports = sizeof(struct virtio_config) + sizeof(struct vnet_config);
595 int tmp_ports = num_ports;
598 // This gets the number of ports, rounded up to a power of 2
599 vnet_state->io_range_size = 1; // must be a power of 2
601 while (tmp_ports > 0) {
603 vnet_state->io_range_size <<= 1;
606 // this is to account for any low order bits being set in num_ports
607 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
608 if ((num_ports & ((vnet_state->io_range_size >> 1) - 1)) == 0) {
609 vnet_state->io_range_size >>= 1;
612 for (i = 0; i < 6; i++) {
613 bars[i].type = PCI_BAR_NONE;
616 bars[0].type = PCI_BAR_IO;
617 bars[0].default_base_port = -1;
618 bars[0].num_ports = vnet_state->io_range_size;
619 bars[0].io_read = vnet_virtio_io_read;
620 bars[0].io_write = vnet_virtio_io_write;
621 bars[0].private_data = vnet_state;
623 pci_dev = v3_pci_register_device(pci_bus, PCI_STD_DEVICE,
624 0, 5 /*PCI_AUTO_DEV_NUM*/, 0,
625 "LNX_VIRTIO_VNET", bars,
626 NULL, NULL, NULL, vnet_state);
629 PrintError("Could not register PCI Device\n");
630 v3_remove_device(dev);
634 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
635 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
636 pci_dev->config_header.device_id = VIRTIO_VNET_DEV_ID;
637 pci_dev->config_header.class = PCI_CLASS_MEMORY;
638 pci_dev->config_header.subclass = PCI_MEM_SUBCLASS_RAM;
639 pci_dev->config_header.subsystem_id = VIRTIO_VNET_SUBDEVICE_ID;
640 pci_dev->config_header.intr_pin = 1;
641 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
644 vnet_state->pci_dev = pci_dev;
645 vnet_state->pci_bus = pci_bus;
648 virtio_reset(vnet_state);
650 struct v3_vnet_bridge_ops brg_ops;
651 brg_ops.input = vnet_pkt_input_cb;
652 brg_ops.poll = vnet_virtio_poll;
654 V3_Print("Registering Virtio device as vnet bridge\n");
656 v3_vnet_add_bridge(vm, &brg_ops, CTL_VM_BRIDGE, (void *)vnet_state);
662 device_register("LNX_VIRTIO_VNET", dev_init)