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 <devices/pci.h>
28 #define QUEUE_SIZE 128
34 } __attribute__((packed));
38 struct virtio_vnet_state {
39 struct vnet_config vnet_cfg;
40 struct virtio_config virtio_cfg;
43 struct vm_device * pci_bus;
44 struct pci_device * pci_dev;
50 struct virtio_queue queue[NUM_QUEUES];
52 struct virtio_queue * cur_queue;
58 #define VNET_GET_ROUTES 10
59 #define VNET_ADD_ROUTE 11
60 #define VNET_DEL_ROUTE 12
62 #define VNET_GET_LINKS 20
63 #define VNET_ADD_LINK 21
64 #define VNET_DEL_LINK 22
66 // structure of the vnet command header
67 struct vnet_ctrl_hdr {
70 } __attribute__((packed));
73 static int virtio_reset(struct virtio_vnet_state * virtio) {
75 memset(virtio->queue, 0, sizeof(struct virtio_queue) * 2);
77 virtio->cur_queue = &(virtio->queue[0]);
79 virtio->virtio_cfg.status = 0;
80 virtio->virtio_cfg.pci_isr = 0;
82 virtio->queue[0].queue_size = QUEUE_SIZE;
83 virtio->queue[1].queue_size = QUEUE_SIZE;
84 virtio->queue[2].queue_size = QUEUE_SIZE;
86 memset(&(virtio->vnet_cfg), 0, sizeof(struct vnet_config));
93 static int get_desc_count(struct virtio_queue * q, int index) {
94 struct vring_desc * tmp_desc = &(q->desc[index]);
97 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
98 tmp_desc = &(q->desc[tmp_desc->next]);
108 static int handle_cmd_kick(struct guest_info * core, struct virtio_vnet_state * vnet_state) {
109 struct virtio_queue * q = &(vnet_state->queue[0]);
111 PrintDebug("VirtioVNET: Virtio Kick on command queue\n");
113 while (q->cur_avail_idx < q->avail->index) {
114 struct vring_desc * hdr_desc = NULL;
115 struct vring_desc * buf_desc = NULL;
116 struct vring_desc * status_desc = NULL;
117 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
118 uint16_t desc_cnt = get_desc_count(q, desc_idx);
119 struct vnet_ctrl_hdr * hdr = NULL;
122 uint8_t * status_ptr = NULL;
126 PrintDebug("Descriptor Count=%d, index=%d, desc_idx=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE, desc_idx);
129 PrintError("VirtioVNET cmd must include at least 3 descriptors (cnt=%d)\n", desc_cnt);
133 hdr_desc = &(q->desc[desc_idx]);
135 if (guest_pa_to_host_va(core, hdr_desc->addr_gpa, (addr_t *)&hdr) == -1) {
136 PrintError("Could not translate VirtioVNET header address\n");
140 desc_idx = hdr_desc->next;
142 if (hdr->cmd_type == VNET_ADD_ROUTE) {
144 for (i = 0; i < hdr->num_cmds; i++) {
145 uint8_t tmp_status = 0;
146 struct v3_vnet_route * route = NULL;
148 buf_desc = &(q->desc[desc_idx]);
150 if (guest_pa_to_host_va(core, buf_desc->addr_gpa, (addr_t *)&(route)) == -1) {
151 PrintError("Could not translate route address\n");
156 PrintDebug("Adding VNET Route\n");
158 tmp_status = v3_vnet_add_route(route);
160 PrintDebug("VNET Route Added\n");
162 if (tmp_status != 0) {
163 PrintError("Error adding VNET ROUTE\n");
167 xfer_len += buf_desc->length;
168 desc_idx = buf_desc->next;
171 } else if (hdr->cmd_type == VNET_ADD_LINK) {
178 status_desc = &(q->desc[desc_idx]);
180 if (guest_pa_to_host_va(core, status_desc->addr_gpa, (addr_t *)&status_ptr) == -1) {
181 PrintError("VirtioVNET Error could not translate status address\n");
185 xfer_len += status_desc->length;
186 *status_ptr = status;
188 PrintDebug("Transferred %d bytes (xfer_len)\n", xfer_len);
189 q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
190 q->used->ring[q->used->index % QUEUE_SIZE].length = xfer_len; // set to total inbound xfer length
197 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
198 PrintDebug("Raising IRQ %d\n", vnet_state->pci_dev->config_header.intr_line);
199 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
200 vnet_state->virtio_cfg.pci_isr = 1;
208 static int virtio_io_write(struct guest_info * core, uint16_t port, void * src, uint_t length, void * private_data) {
209 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
210 int port_idx = port % vnet_state->io_range_size;
213 PrintDebug("VirtioVNET: VIRTIO VNET Write for port %d len=%d, value=%x\n",
214 port, length, *(uint32_t *)src);
215 PrintDebug("VirtioVNET: port idx=%d\n", port_idx);
219 case GUEST_FEATURES_PORT:
221 PrintError("Illegal write length for guest features\n");
225 vnet_state->virtio_cfg.guest_features = *(uint32_t *)src;
228 case VRING_PG_NUM_PORT:
230 addr_t pfn = *(uint32_t *)src;
231 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
233 vnet_state->cur_queue->pfn = pfn;
235 vnet_state->cur_queue->ring_desc_addr = page_addr ;
236 vnet_state->cur_queue->ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
237 vnet_state->cur_queue->ring_used_addr = ( vnet_state->cur_queue->ring_avail_addr + \
238 sizeof(struct vring_avail) + \
239 (QUEUE_SIZE * sizeof(uint16_t)));
241 // round up to next page boundary.
242 vnet_state->cur_queue->ring_used_addr = (vnet_state->cur_queue->ring_used_addr + 0xfff) & ~0xfff;
244 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_desc_addr, (addr_t *)&(vnet_state->cur_queue->desc)) == -1) {
245 PrintError("Could not translate ring descriptor address\n");
250 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_avail_addr, (addr_t *)&(vnet_state->cur_queue->avail)) == -1) {
251 PrintError("Could not translate ring available address\n");
256 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_used_addr, (addr_t *)&(vnet_state->cur_queue->used)) == -1) {
257 PrintError("Could not translate ring used address\n");
261 PrintDebug("VirtioVNET: RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
262 (void *)(vnet_state->cur_queue->ring_desc_addr),
263 (void *)(vnet_state->cur_queue->ring_avail_addr),
264 (void *)(vnet_state->cur_queue->ring_used_addr));
266 PrintDebug("VirtioVNET: RingDesc=%p, Avail=%p, Used=%p\n",
267 vnet_state->cur_queue->desc, vnet_state->cur_queue->avail, vnet_state->cur_queue->used);
270 PrintError("Illegal write length for page frame number\n");
274 case VRING_Q_SEL_PORT:
275 vnet_state->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
277 if (vnet_state->virtio_cfg.vring_queue_selector > NUM_QUEUES) {
278 PrintError("Virtio Symbiotic device has no qeueues. Selected %d\n",
279 vnet_state->virtio_cfg.vring_queue_selector);
283 vnet_state->cur_queue = &(vnet_state->queue[vnet_state->virtio_cfg.vring_queue_selector]);
286 case VRING_Q_NOTIFY_PORT: {
287 uint16_t queue_idx = *(uint16_t *)src;
289 PrintDebug("VirtioVNET: Handling Kick\n");
291 if (queue_idx == 0) {
292 if (handle_cmd_kick(core, vnet_state) == -1) {
293 PrintError("Could not handle VNET Control command\n");
296 } else if (queue_idx == 1) {
299 } else if (queue_idx == 2) {
302 PrintError("Kick on invalid queue (%d)\n", queue_idx);
308 case VIRTIO_STATUS_PORT:
309 vnet_state->virtio_cfg.status = *(uint8_t *)src;
311 if (vnet_state->virtio_cfg.status == 0) {
312 PrintDebug("VirtioVNET: Resetting device\n");
313 virtio_reset(vnet_state);
318 case VIRTIO_ISR_PORT:
319 vnet_state->virtio_cfg.pci_isr = *(uint8_t *)src;
330 static int virtio_io_read(struct guest_info * core, uint16_t port, void * dst, 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;
336 PrintDebug("VirtioVNET: VIRTIO SYMBIOTIC Read for port %d (index =%d), length=%d\n",
337 port, port_idx, length);
340 case HOST_FEATURES_PORT:
342 PrintError("Illegal read length for host features\n");
346 *(uint32_t *)dst = vnet_state->virtio_cfg.host_features;
349 case VRING_PG_NUM_PORT:
351 PrintError("Illegal read length for page frame number\n");
355 *(uint32_t *)dst = vnet_state->cur_queue->pfn;
358 case VRING_SIZE_PORT:
360 PrintError("Illegal read length for vring size\n");
364 *(uint16_t *)dst = vnet_state->cur_queue->queue_size;
368 case VIRTIO_STATUS_PORT:
370 PrintError("Illegal read length for status\n");
374 *(uint8_t *)dst = vnet_state->virtio_cfg.status;
377 case VIRTIO_ISR_PORT:
378 *(uint8_t *)dst = vnet_state->virtio_cfg.pci_isr;
379 vnet_state->virtio_cfg.pci_isr = 0;
380 v3_pci_lower_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
384 if ( (port_idx >= sizeof(struct virtio_config)) &&
385 (port_idx < (sizeof(struct virtio_config) + sizeof(struct vnet_config))) ) {
386 int cfg_offset = port_idx - sizeof(struct virtio_config);
387 uint8_t * cfg_ptr = (uint8_t *)&(vnet_state->vnet_cfg);
389 memcpy(dst, cfg_ptr + cfg_offset, length);
392 PrintError("Read of Unhandled Virtio Read\n");
404 static struct v3_device_ops dev_ops = {
412 static int vnet_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
413 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
414 struct virtio_vnet_state * virtio_state = NULL;
415 struct pci_device * pci_dev = NULL;
416 char * name = v3_cfg_val(cfg, "name");
418 PrintDebug("VirtioVNET: Initializing VIRTIO VNET Control device\n");
420 if (pci_bus == NULL) {
421 PrintError("VirtIO devices require a PCI Bus");
425 virtio_state = (struct virtio_vnet_state *)V3_Malloc(sizeof(struct virtio_vnet_state));
426 memset(virtio_state, 0, sizeof(struct virtio_vnet_state));
428 struct vm_device * dev = v3_allocate_device(name, &dev_ops, virtio_state);
430 if (v3_attach_device(vm, dev) == -1) {
431 PrintError("Could not attach device %s\n", name);
436 // PCI initialization
438 struct v3_pci_bar bars[6];
439 int num_ports = sizeof(struct virtio_config) + sizeof(struct vnet_config);
440 int tmp_ports = num_ports;
444 // This gets the number of ports, rounded up to a power of 2
445 virtio_state->io_range_size = 1; // must be a power of 2
447 while (tmp_ports > 0) {
449 virtio_state->io_range_size <<= 1;
452 // this is to account for any low order bits being set in num_ports
453 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
454 if ((num_ports & ((virtio_state->io_range_size >> 1) - 1)) == 0) {
455 virtio_state->io_range_size >>= 1;
459 for (i = 0; i < 6; i++) {
460 bars[i].type = PCI_BAR_NONE;
463 bars[0].type = PCI_BAR_IO;
464 bars[0].default_base_port = -1;
465 bars[0].num_ports = virtio_state->io_range_size;
467 bars[0].io_read = virtio_io_read;
468 bars[0].io_write = virtio_io_write;
469 bars[0].private_data = virtio_state;
471 pci_dev = v3_pci_register_device(pci_bus, PCI_STD_DEVICE,
472 0, PCI_AUTO_DEV_NUM, 0,
473 "LNX_VIRTIO_VNET", bars,
474 NULL, NULL, NULL, virtio_state);
477 PrintError("Could not register PCI Device\n");
481 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
482 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
485 pci_dev->config_header.device_id = VIRTIO_VNET_DEV_ID;
486 pci_dev->config_header.class = PCI_CLASS_MEMORY;
487 pci_dev->config_header.subclass = PCI_MEM_SUBCLASS_RAM;
489 pci_dev->config_header.subsystem_id = VIRTIO_VNET_SUBDEVICE_ID;
492 pci_dev->config_header.intr_pin = 1;
494 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
497 virtio_state->pci_dev = pci_dev;
498 virtio_state->pci_bus = pci_bus;
501 virtio_reset(virtio_state);
508 device_register("LNX_VIRTIO_VNET", vnet_init)