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, The V3VEE Project <http://www.v3vee.org>
12 * All rights reserved.
14 * Author: Jack Lange <jarusl@cs.northwestern.edu>
16 * This is free software. You are permitted to use,
17 * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
20 #include <palacios/vmm.h>
21 #include <palacios/vmm_dev_mgr.h>
22 #include <palacios/vm_guest_mem.h>
23 #include <devices/lnx_virtio_pci.h>
24 #include <palacios/vmm_vnet.h>
25 #include <palacios/vmm_sprintf.h>
26 #include <devices/pci.h>
29 #ifndef CONFIG_DEBUG_LINUX_VIRTIO_VNET
31 #define PrintDebug(fmt, args...)
35 #define QUEUE_SIZE 128
41 } __attribute__((packed));
48 struct virtio_vnet_state {
49 struct v3_vm_info * vm;
50 struct vnet_config vnet_cfg;
51 struct virtio_config virtio_cfg;
53 struct vm_device * pci_bus;
54 struct pci_device * pci_dev;
56 struct virtio_queue queue[NUM_QUEUES];
58 struct virtio_queue * cur_queue;
64 #define VNET_GET_ROUTES 10
65 #define VNET_ADD_ROUTE 11
66 #define VNET_DEL_ROUTE 12
68 #define VNET_GET_LINKS 20
69 #define VNET_ADD_LINK 21
70 #define VNET_DEL_LINK 22
72 // structure of the vnet command header
73 struct vnet_ctrl_hdr {
76 } __attribute__((packed));
78 struct vnet_virtio_pkt {
81 uint8_t pkt[ETHERNET_PACKET_LEN];
82 }__attribute__((packed));
85 static int virtio_reset(struct virtio_vnet_state * vnet_state) {
87 memset(vnet_state->queue, 0, sizeof(struct virtio_queue) * 2);
89 vnet_state->cur_queue = &(vnet_state->queue[0]);
91 vnet_state->virtio_cfg.status = 0;
92 vnet_state->virtio_cfg.pci_isr = 0;
94 vnet_state->queue[0].queue_size = QUEUE_SIZE;
95 vnet_state->queue[1].queue_size = QUEUE_SIZE;
96 vnet_state->queue[2].queue_size = QUEUE_SIZE;
98 memset(&(vnet_state->vnet_cfg), 0, sizeof(struct vnet_config));
99 v3_lock_init(&(vnet_state->lock));
106 static int get_desc_count(struct virtio_queue * q, int index) {
107 struct vring_desc * tmp_desc = &(q->desc[index]);
110 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
111 tmp_desc = &(q->desc[tmp_desc->next]);
121 static int handle_cmd_kick(struct guest_info * core, struct virtio_vnet_state * vnet_state) {
122 struct virtio_queue * q = &(vnet_state->queue[0]);
124 PrintDebug("VNET Bridge: Handling command queue\n");
126 while (q->cur_avail_idx != q->avail->index) {
127 struct vring_desc * hdr_desc = NULL;
128 struct vring_desc * buf_desc = NULL;
129 struct vring_desc * status_desc = NULL;
130 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
131 uint16_t desc_cnt = get_desc_count(q, desc_idx);
132 struct vnet_ctrl_hdr * hdr = NULL;
135 uint8_t * status_ptr = NULL;
139 PrintDebug("VNET Bridge: CMD: Descriptor Count=%d, index=%d, desc_idx=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE, desc_idx);
142 PrintError("VNET Bridge cmd must include at least 3 descriptors (cnt=%d)\n", desc_cnt);
146 hdr_desc = &(q->desc[desc_idx]);
148 if (guest_pa_to_host_va(core, hdr_desc->addr_gpa, (addr_t *)&hdr) == -1) {
149 PrintError("Could not translate VirtioVNET header address\n");
153 desc_idx = hdr_desc->next;
155 if (hdr->cmd_type == VNET_ADD_ROUTE) {
157 for (i = 0; i < hdr->num_cmds; i++) {
158 uint8_t tmp_status = 0;
159 struct v3_vnet_route * route = NULL;
161 buf_desc = &(q->desc[desc_idx]);
163 if (guest_pa_to_host_va(core, buf_desc->addr_gpa, (addr_t *)&(route)) == -1) {
164 PrintError("Could not translate route address\n");
169 PrintDebug("VNET Bridge: Adding VNET Route\n");
171 tmp_status = v3_vnet_add_route(*route);
173 PrintDebug("VNET Route Added\n");
175 if (tmp_status != 0) {
176 PrintError("Error adding VNET ROUTE\n");
180 xfer_len += buf_desc->length;
181 desc_idx = buf_desc->next;
188 status_desc = &(q->desc[desc_idx]);
190 if (guest_pa_to_host_va(core, status_desc->addr_gpa, (addr_t *)&status_ptr) == -1) {
191 PrintError("VirtioVNET Error could not translate status address\n");
195 xfer_len += status_desc->length;
196 *status_ptr = status;
198 PrintDebug("Transferred %d bytes (xfer_len)\n", xfer_len);
199 q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
200 q->used->ring[q->used->index % QUEUE_SIZE].length = xfer_len; // set to total inbound xfer length
207 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
208 PrintDebug("Raising IRQ %d\n", vnet_state->pci_dev->config_header.intr_line);
209 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
210 vnet_state->virtio_cfg.pci_isr = 1;
218 static int vnet_pkt_input_cb(struct v3_vm_info * vm, struct v3_vnet_pkt * pkt, 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_virtio_pkt * virtio_pkt = NULL;
238 pkt_desc = &(q->desc[pkt_idx]);
239 PrintDebug("VNET Bridge RX: buffer desc len: %d\n", pkt_desc->length);
241 if (guest_pa_to_host_va(&(vm->cores[0]), pkt_desc->addr_gpa, (addr_t *)&(virtio_pkt)) == -1) {
242 PrintError("Could not translate buffer address\n");
246 PrintDebug("VNET Bridge: RX: pkt sent to guest pkt size: %d, dst link: %d\n", pkt->size, pkt->dst_id);
248 // Fill in dst packet buffer
249 virtio_pkt->link_id = pkt->dst_id;
250 virtio_pkt->pkt_size = pkt->size;
251 memcpy(virtio_pkt->pkt, pkt->data, pkt->size);
253 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
254 q->used->ring[q->used->index % q->queue_size].length = sizeof(struct vnet_virtio_pkt); // This should be the total length of data sent to guest (header+pkt_data)
259 PrintError("Packet buffer overflow in the guest: cur_avai_idx %d, idx: %d\n", q->cur_avail_idx, q->avail->index);
262 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
263 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
264 vnet_state->virtio_cfg.pci_isr = 0x1;
265 PrintDebug("Raising IRQ %d\n", vnet_state->pci_dev->config_header.intr_line);
271 v3_unlock_irqrestore(vnet_state->lock, flags);
277 static int handle_pkt_kick(struct guest_info *core, struct virtio_vnet_state * vnet_state)
279 struct virtio_queue * q = &(vnet_state->queue[XMIT_QUEUE]);
280 struct v3_vnet_pkt pkt;
282 while (q->cur_avail_idx != q->avail->index) {
283 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
284 struct vring_desc * pkt_desc = NULL;
285 struct vnet_virtio_pkt * virtio_pkt = NULL;
287 pkt_desc = &(q->desc[desc_idx]);
289 PrintDebug("VNET Bridge: Handle TX desc buf_len: %d\n", pkt_desc->length);
291 //PrintDebug("q: %p, %p, %p, %d\n", q, vnet_state->queue, &vnet_state->queue[XMIT_QUEUE], XMIT_QUEUE);
292 //PrintDebug("q->used: %p\n", q->used);
293 //PrintDebug("q->used->ring %p\n", q->used->ring);
295 if (guest_pa_to_host_va(core, pkt_desc->addr_gpa, (addr_t *)&(virtio_pkt)) == -1) {
296 PrintError("Could not translate buffer address\n");
300 PrintDebug("VNET Bridge: TX: pkt size: %d, dst link: %d\n", virtio_pkt->pkt_size, virtio_pkt->link_id);
302 pkt.size = virtio_pkt->pkt_size;
303 pkt.src_id = virtio_pkt->link_id;
304 pkt.src_type = LINK_EDGE;
305 memcpy(pkt.header, virtio_pkt->pkt, ETHERNET_HEADER_LEN);
306 pkt.data = virtio_pkt->pkt;
308 v3_vnet_send_pkt(&pkt, NULL);
310 q = (struct virtio_queue *)((addr_t)vnet_state->queue + XMIT_QUEUE*sizeof(struct virtio_queue));
312 PrintDebug("After q: %p, , %p, %p, %d\n", q, vnet_state->queue, &(vnet_state->queue[XMIT_QUEUE]), XMIT_QUEUE);
313 //PrintDebug("After q->used: %p\n", q->used);
314 //PrintDebug("After q->used->ring %p\n", q->used->ring);
316 q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
317 q->used->ring[q->used->index % q->queue_size].length = pkt_desc->length; // What do we set this to????
323 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
324 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
325 vnet_state->virtio_cfg.pci_isr = 0x1;
331 static int virtio_io_write(struct guest_info * core, uint16_t port, void * src, uint_t length, void * private_data) {
332 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
333 int port_idx = port % vnet_state->io_range_size;
335 PrintDebug("VNET Bridge: VIRTIO VNET Write for port %d len=%d, value=%x\n",
336 port, length, *(uint32_t *)src);
337 PrintDebug("VNET Bridge: port idx=%d\n", port_idx);
341 case GUEST_FEATURES_PORT:
343 PrintError("Illegal write length for guest features\n");
346 vnet_state->virtio_cfg.guest_features = *(uint32_t *)src;
349 case VRING_PG_NUM_PORT:
351 addr_t pfn = *(uint32_t *)src;
352 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
354 vnet_state->cur_queue->pfn = pfn;
356 vnet_state->cur_queue->ring_desc_addr = page_addr ;
357 vnet_state->cur_queue->ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
358 vnet_state->cur_queue->ring_used_addr = ( vnet_state->cur_queue->ring_avail_addr + \
359 sizeof(struct vring_avail) + \
360 (QUEUE_SIZE * sizeof(uint16_t)));
362 // round up to next page boundary.
363 vnet_state->cur_queue->ring_used_addr = (vnet_state->cur_queue->ring_used_addr + 0xfff) & ~0xfff;
365 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_desc_addr, (addr_t *)&(vnet_state->cur_queue->desc)) == -1) {
366 PrintError("Could not translate ring descriptor address\n");
370 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_avail_addr, (addr_t *)&(vnet_state->cur_queue->avail)) == -1) {
371 PrintError("Could not translate ring available address\n");
375 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_used_addr, (addr_t *)&(vnet_state->cur_queue->used)) == -1) {
376 PrintError("Could not translate ring used address\n");
380 PrintDebug("VNET Bridge: RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
381 (void *)(vnet_state->cur_queue->ring_desc_addr),
382 (void *)(vnet_state->cur_queue->ring_avail_addr),
383 (void *)(vnet_state->cur_queue->ring_used_addr));
385 PrintDebug("VNET Bridge: RingDesc=%p, Avail=%p, Used=%p\n",
386 vnet_state->cur_queue->desc, vnet_state->cur_queue->avail, vnet_state->cur_queue->used);
389 PrintError("Illegal write length for page frame number\n");
393 case VRING_Q_SEL_PORT:
394 vnet_state->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
396 if (vnet_state->virtio_cfg.vring_queue_selector > NUM_QUEUES) {
397 PrintError("VNET Bridge device has no qeueues. Selected %d\n",
398 vnet_state->virtio_cfg.vring_queue_selector);
402 vnet_state->cur_queue = &(vnet_state->queue[vnet_state->virtio_cfg.vring_queue_selector]);
405 case VRING_Q_NOTIFY_PORT: {
406 uint16_t queue_idx = *(uint16_t *)src;
408 PrintDebug("VNET Bridge: Handling Kick\n");
410 if (queue_idx == 0) {
411 if (handle_cmd_kick(core, vnet_state) == -1) {
412 PrintError("Could not handle Virtio VNET Control command\n");
415 } else if (queue_idx == 1) {
416 if (handle_pkt_kick(core, vnet_state) == -1){
417 PrintError("Could not handle Virtio VNET TX\n");
420 } else if (queue_idx == 2) {
421 PrintDebug("VNET Bridge: receive kick on RX Queue\n");
423 PrintError("VNET Bridge: Kick on invalid queue (%d)\n", queue_idx);
429 case VIRTIO_STATUS_PORT:
430 vnet_state->virtio_cfg.status = *(uint8_t *)src;
432 if (vnet_state->virtio_cfg.status == 0) {
433 PrintDebug("VNET Bridge: Resetting device\n");
434 virtio_reset(vnet_state);
439 case VIRTIO_ISR_PORT:
440 vnet_state->virtio_cfg.pci_isr = *(uint8_t *)src;
451 static int virtio_io_read(struct guest_info * core, uint16_t port, void * dst, uint_t length, void * private_data) {
453 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
454 int port_idx = port % vnet_state->io_range_size;
457 PrintDebug("VirtioVNET: VIRTIO SYMBIOTIC Read for port %d (index =%d), length=%d\n",
458 port, port_idx, length);
461 case HOST_FEATURES_PORT:
463 PrintError("Illegal read length for host features\n");
467 *(uint32_t *)dst = vnet_state->virtio_cfg.host_features;
470 case VRING_PG_NUM_PORT:
472 PrintError("Illegal read length for page frame number\n");
476 *(uint32_t *)dst = vnet_state->cur_queue->pfn;
479 case VRING_SIZE_PORT:
481 PrintError("Illegal read length for vring size\n");
485 *(uint16_t *)dst = vnet_state->cur_queue->queue_size;
489 case VIRTIO_STATUS_PORT:
491 PrintError("Illegal read length for status\n");
495 *(uint8_t *)dst = vnet_state->virtio_cfg.status;
498 case VIRTIO_ISR_PORT:
499 *(uint8_t *)dst = vnet_state->virtio_cfg.pci_isr;
500 vnet_state->virtio_cfg.pci_isr = 0;
501 v3_pci_lower_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
505 if ( (port_idx >= sizeof(struct virtio_config)) &&
506 (port_idx < (sizeof(struct virtio_config) + sizeof(struct vnet_config))) ) {
507 int cfg_offset = port_idx - sizeof(struct virtio_config);
508 uint8_t * cfg_ptr = (uint8_t *)&(vnet_state->vnet_cfg);
510 memcpy(dst, cfg_ptr + cfg_offset, length);
513 PrintError("Read of Unhandled Virtio Read\n");
525 static struct v3_device_ops dev_ops = {
533 static int dev_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
534 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
535 struct virtio_vnet_state * vnet_state = NULL;
536 struct pci_device * pci_dev = NULL;
537 char * name = v3_cfg_val(cfg, "name");
539 PrintDebug("VNET Bridge: Initializing VNET Bridge Control device: %s\n", name);
541 if (pci_bus == NULL) {
542 PrintError("VNET Bridge device require a PCI Bus");
546 vnet_state = (struct virtio_vnet_state *)V3_Malloc(sizeof(struct virtio_vnet_state));
547 memset(vnet_state, 0, sizeof(struct virtio_vnet_state));
551 struct vm_device * dev = v3_allocate_device(name, &dev_ops, vnet_state);
553 if (v3_attach_device(vm, dev) == -1) {
554 PrintError("Could not attach device %s\n", name);
559 // PCI initialization
561 struct v3_pci_bar bars[6];
562 int num_ports = sizeof(struct virtio_config) + sizeof(struct vnet_config);
563 int tmp_ports = num_ports;
566 // This gets the number of ports, rounded up to a power of 2
567 vnet_state->io_range_size = 1; // must be a power of 2
569 while (tmp_ports > 0) {
571 vnet_state->io_range_size <<= 1;
574 // this is to account for any low order bits being set in num_ports
575 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
576 if ((num_ports & ((vnet_state->io_range_size >> 1) - 1)) == 0) {
577 vnet_state->io_range_size >>= 1;
580 for (i = 0; i < 6; i++) {
581 bars[i].type = PCI_BAR_NONE;
584 bars[0].type = PCI_BAR_IO;
585 bars[0].default_base_port = -1;
586 bars[0].num_ports = vnet_state->io_range_size;
587 bars[0].io_read = virtio_io_read;
588 bars[0].io_write = virtio_io_write;
589 bars[0].private_data = vnet_state;
591 pci_dev = v3_pci_register_device(pci_bus, PCI_STD_DEVICE,
592 0, PCI_AUTO_DEV_NUM, 0,
593 "LNX_VIRTIO_VNET", bars,
594 NULL, NULL, NULL, vnet_state);
597 PrintError("Could not register PCI Device\n");
601 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
602 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
603 pci_dev->config_header.device_id = VIRTIO_VNET_DEV_ID;
604 pci_dev->config_header.class = PCI_CLASS_MEMORY;
605 pci_dev->config_header.subclass = PCI_MEM_SUBCLASS_RAM;
606 pci_dev->config_header.subsystem_id = VIRTIO_VNET_SUBDEVICE_ID;
607 pci_dev->config_header.intr_pin = 1;
608 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
611 vnet_state->pci_dev = pci_dev;
612 vnet_state->pci_bus = pci_bus;
615 virtio_reset(vnet_state);
617 V3_Print("Registering Virtio device as vnet bridge\n");
618 v3_vnet_add_bridge(vm, vnet_pkt_input_cb, (void *)vnet_state);
624 device_register("LNX_VIRTIO_VNET", dev_init)