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;
84 memset(&(virtio->vnet_cfg), 0, sizeof(struct vnet_config));
91 static int get_desc_count(struct virtio_queue * q, int index) {
92 struct vring_desc * tmp_desc = &(q->desc[index]);
95 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
96 tmp_desc = &(q->desc[tmp_desc->next]);
106 static int handle_cmd_kick(struct guest_info * core, struct virtio_vnet_state * vnet_state) {
107 struct virtio_queue * q = vnet_state->cur_queue;
109 PrintDebug("VirtioVNET: Virtio Kick on command queue\n");
111 while (q->cur_avail_idx < q->avail->index) {
112 struct vring_desc * hdr_desc = NULL;
113 struct vring_desc * buf_desc = NULL;
114 struct vring_desc * status_desc = NULL;
115 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
116 uint16_t desc_cnt = get_desc_count(q, desc_idx);
117 struct vnet_ctrl_hdr * hdr = NULL;
120 uint8_t * status_ptr = NULL;
124 PrintDebug("Descriptor Count=%d, index=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE);
127 PrintError("VirtioVNET cmd must include at least 3 descriptors (cnt=%d)\n", desc_cnt);
131 hdr_desc = &(q->desc[desc_idx]);
133 if (guest_pa_to_host_va(core, hdr_desc->addr_gpa, (addr_t *)&hdr) == -1) {
134 PrintError("Could not translate VirtioVNET header address\n");
138 desc_idx = hdr_desc->next;
140 if (hdr->cmd_type == VNET_ADD_ROUTE) {
142 for (i = 0; i < hdr->num_cmds; i++) {
143 uint8_t tmp_status = 0;
144 struct v3_vnet_route * route = NULL;
146 buf_desc = &(q->desc[desc_idx]);
148 if (guest_pa_to_host_va(core, buf_desc->addr_gpa, (addr_t *)&(route)) == -1) {
149 PrintError("Could not translate route address\n");
154 PrintDebug("Adding VNET Route\n");
156 tmp_status = v3_vnet_add_route(route);
158 PrintDebug("VNET Route Added\n");
160 if (tmp_status != 0) {
161 PrintError("Error adding VNET ROUTE\n");
165 xfer_len += buf_desc->length;
166 desc_idx = buf_desc->next;
169 } else if (hdr->cmd_type == VNET_ADD_LINK) {
176 status_desc = &(q->desc[desc_idx]);
178 if (guest_pa_to_host_va(core, status_desc->addr_gpa, (addr_t *)&status_ptr) == -1) {
179 PrintError("VirtioVNET Error could not translate status address\n");
183 xfer_len += status_desc->length;
184 *status_ptr = status;
186 PrintDebug("Transferred %d bytes (xfer_len)\n", xfer_len);
187 q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
188 q->used->ring[q->used->index % QUEUE_SIZE].length = xfer_len; // set to total inbound xfer length
195 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
196 PrintDebug("Raising IRQ %d\n", vnet_state->pci_dev->config_header.intr_line);
197 v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
198 vnet_state->virtio_cfg.pci_isr = 1;
206 static int virtio_io_write(struct guest_info * core, uint16_t port, void * src, uint_t length, void * private_data) {
207 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
208 int port_idx = port % vnet_state->io_range_size;
211 PrintDebug("VirtioVNET: VIRTIO VNET Write for port %d len=%d, value=%x\n",
212 port, length, *(uint32_t *)src);
213 PrintDebug("VirtioVNET: port idx=%d\n", port_idx);
217 case GUEST_FEATURES_PORT:
219 PrintError("Illegal write length for guest features\n");
223 vnet_state->virtio_cfg.guest_features = *(uint32_t *)src;
226 case VRING_PG_NUM_PORT:
228 addr_t pfn = *(uint32_t *)src;
229 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
231 vnet_state->cur_queue->pfn = pfn;
233 vnet_state->cur_queue->ring_desc_addr = page_addr ;
234 vnet_state->cur_queue->ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
235 vnet_state->cur_queue->ring_used_addr = ( vnet_state->cur_queue->ring_avail_addr + \
236 sizeof(struct vring_avail) + \
237 (QUEUE_SIZE * sizeof(uint16_t)));
239 // round up to next page boundary.
240 vnet_state->cur_queue->ring_used_addr = (vnet_state->cur_queue->ring_used_addr + 0xfff) & ~0xfff;
242 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_desc_addr, (addr_t *)&(vnet_state->cur_queue->desc)) == -1) {
243 PrintError("Could not translate ring descriptor address\n");
248 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_avail_addr, (addr_t *)&(vnet_state->cur_queue->avail)) == -1) {
249 PrintError("Could not translate ring available address\n");
254 if (guest_pa_to_host_va(core, vnet_state->cur_queue->ring_used_addr, (addr_t *)&(vnet_state->cur_queue->used)) == -1) {
255 PrintError("Could not translate ring used address\n");
259 PrintDebug("VirtioVNET: RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
260 (void *)(vnet_state->cur_queue->ring_desc_addr),
261 (void *)(vnet_state->cur_queue->ring_avail_addr),
262 (void *)(vnet_state->cur_queue->ring_used_addr));
264 PrintDebug("VirtioVNET: RingDesc=%p, Avail=%p, Used=%p\n",
265 vnet_state->cur_queue->desc, vnet_state->cur_queue->avail, vnet_state->cur_queue->used);
268 PrintError("Illegal write length for page frame number\n");
272 case VRING_Q_SEL_PORT:
273 vnet_state->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
275 if (vnet_state->virtio_cfg.vring_queue_selector > NUM_QUEUES) {
276 PrintError("Virtio Symbiotic device has no qeueues. Selected %d\n",
277 vnet_state->virtio_cfg.vring_queue_selector);
281 vnet_state->cur_queue = &(vnet_state->queue[vnet_state->virtio_cfg.vring_queue_selector]);
284 case VRING_Q_NOTIFY_PORT: {
285 uint16_t queue_idx = *(uint16_t *)src;
287 PrintDebug("VirtioVNET: Handling Kick\n");
289 if (queue_idx == 0) {
290 if (handle_cmd_kick(core, vnet_state) == -1) {
291 PrintError("Could not handle VNET Control command\n");
295 PrintError("Kick on invalid queue (%d)\n", queue_idx);
301 case VIRTIO_STATUS_PORT:
302 vnet_state->virtio_cfg.status = *(uint8_t *)src;
304 if (vnet_state->virtio_cfg.status == 0) {
305 PrintDebug("VirtioVNET: Resetting device\n");
306 virtio_reset(vnet_state);
311 case VIRTIO_ISR_PORT:
312 vnet_state->virtio_cfg.pci_isr = *(uint8_t *)src;
323 static int virtio_io_read(struct guest_info * core, uint16_t port, void * dst, uint_t length, void * private_data) {
325 struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data;
326 int port_idx = port % vnet_state->io_range_size;
329 PrintDebug("VirtioVNET: VIRTIO SYMBIOTIC Read for port %d (index =%d), length=%d\n",
330 port, port_idx, length);
333 case HOST_FEATURES_PORT:
335 PrintError("Illegal read length for host features\n");
339 *(uint32_t *)dst = vnet_state->virtio_cfg.host_features;
342 case VRING_PG_NUM_PORT:
344 PrintError("Illegal read length for page frame number\n");
348 *(uint32_t *)dst = vnet_state->cur_queue->pfn;
351 case VRING_SIZE_PORT:
353 PrintError("Illegal read length for vring size\n");
357 *(uint16_t *)dst = vnet_state->cur_queue->queue_size;
361 case VIRTIO_STATUS_PORT:
363 PrintError("Illegal read length for status\n");
367 *(uint8_t *)dst = vnet_state->virtio_cfg.status;
370 case VIRTIO_ISR_PORT:
371 *(uint8_t *)dst = vnet_state->virtio_cfg.pci_isr;
372 vnet_state->virtio_cfg.pci_isr = 0;
373 v3_pci_lower_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev);
377 if ( (port_idx >= sizeof(struct virtio_config)) &&
378 (port_idx < (sizeof(struct virtio_config) + sizeof(struct vnet_config))) ) {
379 int cfg_offset = port_idx - sizeof(struct virtio_config);
380 uint8_t * cfg_ptr = (uint8_t *)&(vnet_state->vnet_cfg);
382 memcpy(dst, cfg_ptr + cfg_offset, length);
385 PrintError("Read of Unhandled Virtio Read\n");
397 static struct v3_device_ops dev_ops = {
405 static int vnet_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
406 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
407 struct virtio_vnet_state * virtio_state = NULL;
408 struct pci_device * pci_dev = NULL;
409 char * name = v3_cfg_val(cfg, "name");
411 PrintDebug("VirtioVNET: Initializing VIRTIO VNET Control device\n");
413 if (pci_bus == NULL) {
414 PrintError("VirtIO devices require a PCI Bus");
418 virtio_state = (struct virtio_vnet_state *)V3_Malloc(sizeof(struct virtio_vnet_state));
419 memset(virtio_state, 0, sizeof(struct virtio_vnet_state));
421 struct vm_device * dev = v3_allocate_device(name, &dev_ops, virtio_state);
423 if (v3_attach_device(vm, dev) == -1) {
424 PrintError("Could not attach device %s\n", name);
429 // PCI initialization
431 struct v3_pci_bar bars[6];
432 int num_ports = sizeof(struct virtio_config) + sizeof(struct vnet_config);
433 int tmp_ports = num_ports;
437 // This gets the number of ports, rounded up to a power of 2
438 virtio_state->io_range_size = 1; // must be a power of 2
440 while (tmp_ports > 0) {
442 virtio_state->io_range_size <<= 1;
445 // this is to account for any low order bits being set in num_ports
446 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
447 if ((num_ports & ((virtio_state->io_range_size >> 1) - 1)) == 0) {
448 virtio_state->io_range_size >>= 1;
452 for (i = 0; i < 6; i++) {
453 bars[i].type = PCI_BAR_NONE;
456 bars[0].type = PCI_BAR_IO;
457 bars[0].default_base_port = -1;
458 bars[0].num_ports = virtio_state->io_range_size;
460 bars[0].io_read = virtio_io_read;
461 bars[0].io_write = virtio_io_write;
462 bars[0].private_data = virtio_state;
464 pci_dev = v3_pci_register_device(pci_bus, PCI_STD_DEVICE,
465 0, PCI_AUTO_DEV_NUM, 0,
466 "LNX_VIRTIO_VNET", bars,
467 NULL, NULL, NULL, virtio_state);
470 PrintError("Could not register PCI Device\n");
474 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
475 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
478 pci_dev->config_header.device_id = VIRTIO_VNET_DEV_ID;
479 pci_dev->config_header.class = PCI_CLASS_MEMORY;
480 pci_dev->config_header.subclass = PCI_MEM_SUBCLASS_RAM;
482 pci_dev->config_header.subsystem_id = VIRTIO_VNET_SUBDEVICE_ID;
485 pci_dev->config_header.intr_pin = 1;
487 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
490 virtio_state->pci_dev = pci_dev;
491 virtio_state->pci_bus = pci_bus;
494 virtio_reset(virtio_state);
501 device_register("LNX_VIRTIO_VNET", vnet_init)