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 <palacios/vmm_vnet.h>
27 #include <palacios/vmm_sprintf.h>
28 #include <devices/pci.h>
31 #ifndef 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 // structure of the vnet command header
79 struct vnet_ctrl_hdr {
82 } __attribute__((packed));
85 struct vnet_bridge_pkt {
88 uint8_t pkt[ETHERNET_PACKET_LEN];
89 }__attribute__((packed));
92 static int virtio_reset(struct virtio_vnet_state * vnet_state) {
94 memset(vnet_state->queue, 0, sizeof(struct virtio_queue) * NUM_QUEUES);
96 vnet_state->cur_queue = &(vnet_state->queue[0]);
98 vnet_state->virtio_cfg.status = 0;
99 vnet_state->virtio_cfg.pci_isr = 0;
101 vnet_state->queue[0].queue_size = CMD_QUEUE_SIZE;
102 vnet_state->queue[1].queue_size = QUEUE_SIZE;
103 vnet_state->queue[2].queue_size = QUEUE_SIZE;
105 memset(&(vnet_state->vnet_cfg), 0, sizeof(struct vnet_config));
106 v3_lock_init(&(vnet_state->lock));
113 static int get_desc_count(struct virtio_queue * q, int index) {
114 struct vring_desc * tmp_desc = &(q->desc[index]);
117 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
118 tmp_desc = &(q->desc[tmp_desc->next]);
128 static int handle_cmd_kick(struct guest_info * core, struct virtio_vnet_state * vnet_state) {
129 struct virtio_queue * q = &(vnet_state->queue[0]);
131 PrintDebug("VNET Bridge: Handling command queue\n");
133 while (q->cur_avail_idx != q->avail->index) {
134 struct vring_desc * hdr_desc = NULL;
135 struct vring_desc * buf_desc = NULL;
136 struct vring_desc * status_desc = NULL;
137 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
138 uint16_t desc_cnt = get_desc_count(q, desc_idx);
139 struct vnet_ctrl_hdr * hdr = NULL;
142 uint8_t * status_ptr = NULL;
146 PrintDebug("VNET Bridge: CMD: Descriptor Count=%d, index=%d, desc_idx=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE, desc_idx);
149 PrintError("VNET Bridge cmd must include at least 3 descriptors (cnt=%d)\n", desc_cnt);
153 hdr_desc = &(q->desc[desc_idx]);
155 if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, (addr_t *)&hdr) == -1) {
156 PrintError("Could not translate VirtioVNET header address\n");
160 desc_idx = hdr_desc->next;
162 if (hdr->cmd_type == VNET_ADD_ROUTE) {
164 for (i = 0; i < hdr->num_cmds; i++) {
165 uint8_t tmp_status = 0;
166 struct v3_vnet_route * route = NULL;
168 buf_desc = &(q->desc[desc_idx]);
170 if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(route)) == -1) {
171 PrintError("Could not translate route address\n");
176 PrintDebug("VNET Bridge: Adding VNET Route\n");
178 tmp_status = v3_vnet_add_route(*route);
180 PrintDebug("VNET Route Added\n");
182 if (tmp_status != 0) {
183 PrintError("Error adding VNET ROUTE\n");
187 xfer_len += buf_desc->length;
188 desc_idx = buf_desc->next;
195 status_desc = &(q->desc[desc_idx]);
197 if (v3_gpa_to_hva(core, status_desc->addr_gpa, (addr_t *)&status_ptr) == -1) {
198 PrintError("VirtioVNET Error could not translate status address\n");
202 xfer_len += status_desc->length;
203 *status_ptr = status;
205 PrintDebug("Transferred %d bytes (xfer_len)\n", xfer_len);
206 q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
207 q->used->ring[q->used->index % QUEUE_SIZE].length = xfer_len; // set to total inbound xfer length
214 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
215 PrintDebug("Raising IRQ %d\n", vnet_state->pci_dev->config_header.intr_line);
216 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
217 vnet_state->virtio_cfg.pci_isr = 1;
225 static int vnet_pkt_input_cb(struct v3_vm_info * vm, struct v3_vnet_pkt *pkt, void * private_data){
226 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
227 struct virtio_queue * q = &(vnet_state->queue[RECV_QUEUE]);
231 flags = v3_lock_irqsave(vnet_state->lock);
233 if (q->ring_avail_addr == 0) {
234 PrintError("Queue is not set\n");
238 if (q->cur_avail_idx != q->avail->index) {
239 uint16_t pkt_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
240 struct vring_desc * pkt_desc = NULL;
241 struct vnet_bridge_pkt * virtio_pkt = NULL;
243 pkt_desc = &(q->desc[pkt_idx]);
244 PrintDebug("VNET Bridge RX: buffer desc len: %d\n", pkt_desc->length);
246 if (v3_gpa_to_hva(&(vm->cores[0]), pkt_desc->addr_gpa, (addr_t *)&(virtio_pkt)) == -1) {
247 PrintError("Could not translate buffer address\n");
251 PrintDebug("VNET Bridge: RX: pkt sent to guest pkt size: %d, dst link: %d\n", pkt->size, pkt->dst_id);
253 // Fill in dst packet buffer
254 virtio_pkt->link_id = pkt->dst_id;
255 virtio_pkt->pkt_size = pkt->size;
256 memcpy(virtio_pkt->pkt, pkt->data, pkt->size);
258 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
259 q->used->ring[q->used->index % q->queue_size].length = sizeof(struct vnet_bridge_pkt);
264 vnet_state->pkt_drop ++;
267 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
268 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
269 vnet_state->virtio_cfg.pci_isr = 0x1;
270 PrintDebug("Raising IRQ %d\n", vnet_state->pci_dev->config_header.intr_line);
277 v3_unlock_irqrestore(vnet_state->lock, flags);
282 static int vnet_pkt_input_xcall(void *data){
283 struct v3_vnet_bridge_xcall_args *args = (struct v3_vnet_bridge_xcall_args *)data;
286 for(i = 0; i < args->pkt_num; i++) {
287 vnet_pkt_input_cb(args->vm, &(args->vnet_pkts[i]), args->private_data);
293 static int handle_pkt_kick(struct guest_info *core, struct virtio_vnet_state * vnet_state)
295 struct virtio_queue * q = &(vnet_state->queue[XMIT_QUEUE]);
298 if (q->ring_avail_addr == 0) {
302 while (q->cur_avail_idx != q->avail->index) {
303 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
304 struct vring_desc * pkt_desc = NULL;
305 struct vnet_bridge_pkt * virtio_pkt = NULL;
307 pkt_desc = &(q->desc[desc_idx]);
309 PrintDebug("VNET Bridge: Handle TX desc buf_len: %d\n", pkt_desc->length);
311 if (v3_gpa_to_hva(core, pkt_desc->addr_gpa, (addr_t *)&(virtio_pkt)) == -1) {
312 PrintError("Could not translate buffer address\n");
316 //PrintDebug("VNET Bridge: TX: on cpu %d pkt size: %d, dst link: %d\n", cpu, virtio_pkt->pkt_size, virtio_pkt->link_id);
318 struct v3_vnet_pkt pkt;
319 pkt.size = virtio_pkt->pkt_size;
320 pkt.src_type = LINK_EDGE;
322 memcpy(pkt.header, virtio_pkt->pkt, ETHERNET_HEADER_LEN);
323 pkt.data = virtio_pkt->pkt;
325 v3_vnet_send_pkt(&pkt, NULL);
327 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
328 q->used->ring[q->used->index % q->queue_size].length = pkt_desc->length; // What do we set this to????
331 vnet_state->pkt_sent ++;
341 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
342 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
343 vnet_state->virtio_cfg.pci_isr = 0x1;
349 static void vnet_virtio_poll(struct v3_vm_info * vm, void *private_data){
350 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
352 if(vm == vnet_state->vm){
353 handle_pkt_kick(&(vm->cores[0]), vnet_state);
357 static int handle_rx_kick(struct guest_info *core, struct virtio_vnet_state * vnet_state)
359 //v3_vnet_enable_bridge();
364 static int vnet_virtio_io_write(struct guest_info * core, uint16_t port, void * src, uint_t length, void * private_data) {
365 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
366 int port_idx = port % vnet_state->io_range_size;
368 PrintDebug("VNET Bridge: VIRTIO VNET Write for port %d len=%d, value=%x\n",
369 port, length, *(uint32_t *)src);
371 vnet_state->total_exit ++;
373 case GUEST_FEATURES_PORT:
375 PrintError("Illegal write length for guest features\n");
378 vnet_state->virtio_cfg.guest_features = *(uint32_t *)src;
381 case VRING_PG_NUM_PORT:
383 addr_t pfn = *(uint32_t *)src;
384 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
386 vnet_state->cur_queue->pfn = pfn;
388 vnet_state->cur_queue->ring_desc_addr = page_addr ;
389 vnet_state->cur_queue->ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
390 vnet_state->cur_queue->ring_used_addr = ( vnet_state->cur_queue->ring_avail_addr + \
391 sizeof(struct vring_avail) + \
392 (QUEUE_SIZE * sizeof(uint16_t)));
394 // round up to next page boundary.
395 vnet_state->cur_queue->ring_used_addr = (vnet_state->cur_queue->ring_used_addr + 0xfff) & ~0xfff;
397 if (v3_gpa_to_hva(core, vnet_state->cur_queue->ring_desc_addr, (addr_t *)&(vnet_state->cur_queue->desc)) == -1) {
398 PrintError("Could not translate ring descriptor address\n");
402 if (v3_gpa_to_hva(core, vnet_state->cur_queue->ring_avail_addr, (addr_t *)&(vnet_state->cur_queue->avail)) == -1) {
403 PrintError("Could not translate ring available address\n");
407 if (v3_gpa_to_hva(core, vnet_state->cur_queue->ring_used_addr, (addr_t *)&(vnet_state->cur_queue->used)) == -1) {
408 PrintError("Could not translate ring used address\n");
412 PrintDebug("VNET Bridge: RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
413 (void *)(vnet_state->cur_queue->ring_desc_addr),
414 (void *)(vnet_state->cur_queue->ring_avail_addr),
415 (void *)(vnet_state->cur_queue->ring_used_addr));
417 PrintDebug("VNET Bridge: RingDesc=%p, Avail=%p, Used=%p\n",
418 vnet_state->cur_queue->desc, vnet_state->cur_queue->avail, vnet_state->cur_queue->used);
420 if(vnet_state->queue[RECV_QUEUE].avail != NULL){
421 vnet_state->ready = 1;
422 vnet_state->queue[RECV_QUEUE].used->flags |= VRING_NO_NOTIFY_FLAG;
425 //No notify when there is pkt tx from guest
426 //palacios will do the polling
427 if(vnet_state->queue[XMIT_QUEUE].used != NULL){
428 vnet_state->queue[XMIT_QUEUE].used->flags |= VRING_NO_NOTIFY_FLAG;
431 PrintError("Illegal write length for page frame number\n");
435 case VRING_Q_SEL_PORT:
436 vnet_state->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
438 if (vnet_state->virtio_cfg.vring_queue_selector > NUM_QUEUES) {
439 PrintError("VNET Bridge device has no qeueues. Selected %d\n",
440 vnet_state->virtio_cfg.vring_queue_selector);
444 vnet_state->cur_queue = &(vnet_state->queue[vnet_state->virtio_cfg.vring_queue_selector]);
447 case VRING_Q_NOTIFY_PORT: {
448 uint16_t queue_idx = *(uint16_t *)src;
450 PrintDebug("VNET Bridge: Handling Kick\n");
452 if (queue_idx == 0) {
453 if (handle_cmd_kick(core, vnet_state) == -1) {
454 PrintError("Could not handle Virtio VNET Control command\n");
457 } else if (queue_idx == 1) {
458 if (handle_pkt_kick(core, vnet_state) == -1){
459 PrintError("Could not handle Virtio VNET TX\n");
462 PrintError("Notify on TX\n");
463 } else if (queue_idx == 2) {
464 if (handle_rx_kick(core, vnet_state) == -1){
465 PrintError("Could not handle Virtio RX buffer refills Kick\n");
468 vnet_state->rx_exit ++;
470 PrintError("VNET Bridge: Kick on invalid queue (%d)\n", queue_idx);
476 case VIRTIO_STATUS_PORT:
477 vnet_state->virtio_cfg.status = *(uint8_t *)src;
479 if (vnet_state->virtio_cfg.status == 0) {
480 PrintDebug("VNET Bridge: Resetting device\n");
481 virtio_reset(vnet_state);
486 case VIRTIO_ISR_PORT:
487 vnet_state->virtio_cfg.pci_isr = *(uint8_t *)src;
498 static int vnet_virtio_io_read(struct guest_info * core, uint16_t port, void * dst, uint_t length, void * private_data) {
500 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
501 int port_idx = port % vnet_state->io_range_size;
504 case HOST_FEATURES_PORT:
506 PrintError("Illegal read length for host features\n");
510 *(uint32_t *)dst = vnet_state->virtio_cfg.host_features;
513 case VRING_PG_NUM_PORT:
515 PrintError("Illegal read length for page frame number\n");
519 *(uint32_t *)dst = vnet_state->cur_queue->pfn;
522 case VRING_SIZE_PORT:
524 PrintError("Illegal read length for vring size\n");
528 *(uint16_t *)dst = vnet_state->cur_queue->queue_size;
532 case VIRTIO_STATUS_PORT:
534 PrintError("Illegal read length for status\n");
538 *(uint8_t *)dst = vnet_state->virtio_cfg.status;
541 case VIRTIO_ISR_PORT:
542 *(uint8_t *)dst = vnet_state->virtio_cfg.pci_isr;
543 vnet_state->virtio_cfg.pci_isr = 0;
544 v3_pci_lower_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
548 if ( (port_idx >= sizeof(struct virtio_config)) &&
549 (port_idx < (sizeof(struct virtio_config) + sizeof(struct vnet_config))) ) {
550 int cfg_offset = port_idx - sizeof(struct virtio_config);
551 uint8_t * cfg_ptr = (uint8_t *)&(vnet_state->vnet_cfg);
553 memcpy(dst, cfg_ptr + cfg_offset, length);
556 PrintError("Read of Unhandled Virtio Read\n");
568 static struct v3_device_ops dev_ops = {
575 static int dev_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
576 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
577 struct virtio_vnet_state * vnet_state = NULL;
578 struct pci_device * pci_dev = NULL;
579 char * name = v3_cfg_val(cfg, "name");
581 PrintDebug("VNET Bridge: Initializing VNET Bridge Control device: %s\n", name);
583 if (pci_bus == NULL) {
584 PrintError("VNET Bridge device require a PCI Bus");
588 vnet_state = (struct virtio_vnet_state *)V3_Malloc(sizeof(struct virtio_vnet_state));
589 memset(vnet_state, 0, sizeof(struct virtio_vnet_state));
593 struct vm_device * dev = v3_allocate_device(name, &dev_ops, vnet_state);
595 if (v3_attach_device(vm, dev) == -1) {
596 PrintError("Could not attach device %s\n", name);
601 // PCI initialization
603 struct v3_pci_bar bars[6];
604 int num_ports = sizeof(struct virtio_config) + sizeof(struct vnet_config);
605 int tmp_ports = num_ports;
608 // This gets the number of ports, rounded up to a power of 2
609 vnet_state->io_range_size = 1; // must be a power of 2
611 while (tmp_ports > 0) {
613 vnet_state->io_range_size <<= 1;
616 // this is to account for any low order bits being set in num_ports
617 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
618 if ((num_ports & ((vnet_state->io_range_size >> 1) - 1)) == 0) {
619 vnet_state->io_range_size >>= 1;
622 for (i = 0; i < 6; i++) {
623 bars[i].type = PCI_BAR_NONE;
626 bars[0].type = PCI_BAR_IO;
627 bars[0].default_base_port = -1;
628 bars[0].num_ports = vnet_state->io_range_size;
629 bars[0].io_read = vnet_virtio_io_read;
630 bars[0].io_write = vnet_virtio_io_write;
631 bars[0].private_data = vnet_state;
633 pci_dev = v3_pci_register_device(pci_bus, PCI_STD_DEVICE,
634 0, 5 /*PCI_AUTO_DEV_NUM*/, 0,
635 "LNX_VIRTIO_VNET", bars,
636 NULL, NULL, NULL, vnet_state);
639 PrintError("Could not register PCI Device\n");
643 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
644 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
645 pci_dev->config_header.device_id = VIRTIO_VNET_DEV_ID;
646 pci_dev->config_header.class = PCI_CLASS_MEMORY;
647 pci_dev->config_header.subclass = PCI_MEM_SUBCLASS_RAM;
648 pci_dev->config_header.subsystem_id = VIRTIO_VNET_SUBDEVICE_ID;
649 pci_dev->config_header.intr_pin = 1;
650 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
653 vnet_state->pci_dev = pci_dev;
654 vnet_state->pci_bus = pci_bus;
657 virtio_reset(vnet_state);
659 struct v3_vnet_bridge_ops brg_ops;
660 brg_ops.input = vnet_pkt_input_cb;
661 brg_ops.polling_pkt = vnet_virtio_poll;
662 brg_ops.xcall_input = vnet_pkt_input_xcall;
664 V3_Print("Registering Virtio device as vnet bridge\n");
665 v3_vnet_add_bridge(vm, &brg_ops, (void *)vnet_state);
671 device_register("LNX_VIRTIO_VNET", dev_init)