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 <devices/lnx_virtio_pci.h>
23 #include <palacios/vm_guest_mem.h>
25 #include <devices/pci.h>
29 #ifndef CONFIG_DEBUG_VIRTIO_BLK
31 #define PrintDebug(fmt, args...)
35 #define BLK_CAPACITY_PORT 20
36 #define BLK_MAX_SIZE_PORT 28
37 #define BLK_MAX_SEG_PORT 32
38 #define BLK_CYLINDERS_PORT 36
39 #define BLK_HEADS_PORT 38
40 #define BLK_SECTS_PORT 39
44 #define BLK_SCSI_CMD 2
46 #define BLK_BARRIER_FLAG 0x80000000
48 #define BLK_STATUS_OK 0
49 #define BLK_STATUS_ERR 1
50 #define BLK_STATUS_NOT_SUPPORTED 2
60 } __attribute__((packed));
68 } __attribute__((packed));
70 #define QUEUE_SIZE 128
72 /* Host Feature flags */
73 #define VIRTIO_BARRIER 0x01 /* Does host support barriers? */
74 #define VIRTIO_SIZE_MAX 0x02 /* Indicates maximum segment size */
75 #define VIRTIO_SEG_MAX 0x04 /* Indicates maximum # of segments */
76 #define VIRTIO_LEGACY_GEOM 0x10 /* Indicates support of legacy geometry */
79 struct virtio_dev_state {
80 struct vm_device * pci_bus;
81 struct list_head dev_list;
82 struct guest_info * vm;
85 struct virtio_blk_state {
87 struct pci_device * pci_dev;
88 struct blk_config block_cfg;
89 struct virtio_config virtio_cfg;
92 struct virtio_queue queue;
94 struct v3_dev_blk_ops * ops;
100 struct virtio_dev_state * virtio_dev;
102 struct list_head dev_link;
106 static int virtio_free(struct vm_device * dev) {
110 static int blk_reset(struct virtio_blk_state * virtio) {
112 virtio->queue.ring_desc_addr = 0;
113 virtio->queue.ring_avail_addr = 0;
114 virtio->queue.ring_used_addr = 0;
115 virtio->queue.pfn = 0;
116 virtio->queue.cur_avail_idx = 0;
118 virtio->virtio_cfg.status = 0;
119 virtio->virtio_cfg.pci_isr = 0;
124 static int virtio_reset(struct vm_device * dev) {
125 struct virtio_dev_state * dev_state = (struct virtio_dev_state *)(dev->private_data);
126 struct virtio_blk_state * blk_state = NULL;
128 list_for_each_entry(blk_state, &(dev_state->dev_list), dev_link) {
129 blk_reset(blk_state);
135 static int handle_read_op(struct virtio_blk_state * blk_state, uint8_t * buf, uint64_t * sector, uint64_t len) {
138 PrintDebug("Reading Disk\n");
139 ret = blk_state->ops->read(buf, *sector, len, (void *)(blk_state->backend_data));
146 static int handle_write_op(struct virtio_blk_state * blk_state, uint8_t * buf, uint64_t * sector, uint64_t len) {
149 PrintDebug("Writing Disk\n");
150 ret = blk_state->ops->write(buf, *sector, len, (void *)(blk_state->backend_data));
158 // multiple block operations need to increment the sector
160 static int handle_block_op(struct virtio_blk_state * blk_state, struct blk_op_hdr * hdr,
161 struct vring_desc * buf_desc, uint8_t * status) {
162 uint8_t * buf = NULL;
164 PrintDebug("Handling Block op\n");
165 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
166 PrintError("Could not translate buffer address\n");
170 PrintDebug("Sector=%p Length=%d\n", (void *)(addr_t)(hdr->sector), buf_desc->length);
172 if (hdr->type == BLK_IN_REQ) {
173 if (handle_read_op(blk_state, buf, &(hdr->sector), buf_desc->length) == -1) {
174 *status = BLK_STATUS_ERR;
177 *status = BLK_STATUS_OK;
179 } else if (hdr->type == BLK_OUT_REQ) {
180 if (handle_write_op(blk_state, buf, &(hdr->sector), buf_desc->length) == -1) {
181 *status = BLK_STATUS_ERR;
184 *status = BLK_STATUS_OK;
186 } else if (hdr->type == BLK_SCSI_CMD) {
187 PrintError("VIRTIO: SCSI Command Not supported!!!\n");
188 *status = BLK_STATUS_NOT_SUPPORTED;
192 PrintDebug("Returning Status: %d\n", *status);
197 static int get_desc_count(struct virtio_queue * q, int index) {
198 struct vring_desc * tmp_desc = &(q->desc[index]);
201 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
202 tmp_desc = &(q->desc[tmp_desc->next]);
211 static int handle_kick(struct virtio_blk_state * blk_state) {
212 struct virtio_queue * q = &(blk_state->queue);
214 PrintDebug("VIRTIO KICK: cur_index=%d (mod=%d), avail_index=%d\n",
215 q->cur_avail_idx, q->cur_avail_idx % QUEUE_SIZE, q->avail->index);
217 while (q->cur_avail_idx < q->avail->index) {
218 struct vring_desc * hdr_desc = NULL;
219 struct vring_desc * buf_desc = NULL;
220 struct vring_desc * status_desc = NULL;
221 struct blk_op_hdr hdr;
223 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
224 int desc_cnt = get_desc_count(q, desc_idx);
226 uint8_t * status_ptr = NULL;
227 uint8_t status = BLK_STATUS_OK;
228 uint32_t req_len = 0;
230 PrintDebug("Descriptor Count=%d, index=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE);
233 PrintError("Block operations must include at least 3 descriptors\n");
237 hdr_desc = &(q->desc[desc_idx]);
240 PrintDebug("Header Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", hdr_desc,
241 (void *)(hdr_desc->addr_gpa), hdr_desc->length, hdr_desc->flags, hdr_desc->next);
243 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
244 PrintError("Could not translate block header address\n");
248 // We copy the block op header out because we are going to modify its contents
249 memcpy(&hdr, (void *)hdr_addr, sizeof(struct blk_op_hdr));
251 PrintDebug("Blk Op Hdr (ptr=%p) type=%d, sector=%p\n", (void *)hdr_addr, hdr.type, (void *)hdr.sector);
253 desc_idx = hdr_desc->next;
255 for (i = 0; i < desc_cnt - 2; i++) {
256 uint8_t tmp_status = BLK_STATUS_OK;
258 buf_desc = &(q->desc[desc_idx]);
260 PrintDebug("Buffer Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", buf_desc,
261 (void *)(buf_desc->addr_gpa), buf_desc->length, buf_desc->flags, buf_desc->next);
263 if (handle_block_op(blk_state, &hdr, buf_desc, &tmp_status) == -1) {
264 PrintError("Error handling block operation\n");
268 if (tmp_status != BLK_STATUS_OK) {
272 req_len += buf_desc->length;
273 desc_idx = buf_desc->next;
276 status_desc = &(q->desc[desc_idx]);
278 PrintDebug("Status Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", status_desc,
279 (void *)(status_desc->addr_gpa), status_desc->length, status_desc->flags, status_desc->next);
281 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, status_desc->addr_gpa, (addr_t *)&(status_ptr)) == -1) {
282 PrintError("Could not translate status address\n");
286 req_len += status_desc->length;
287 *status_ptr = status;
289 q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
290 q->used->ring[q->used->index % QUEUE_SIZE].length = req_len; // What do we set this to????
296 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
297 PrintDebug("Raising IRQ %d\n", blk_state->pci_dev->config_header.intr_line);
298 v3_pci_raise_irq(blk_state->virtio_dev->pci_bus, 0, blk_state->pci_dev);
299 blk_state->virtio_cfg.pci_isr = 1;
305 static int virtio_io_write(uint16_t port, void * src, uint_t length, void * private_data) {
306 struct virtio_blk_state * blk_state = (struct virtio_blk_state *)private_data;
307 int port_idx = port % blk_state->io_range_size;
310 PrintDebug("VIRTIO BLOCK Write for port %d (index=%d) len=%d, value=%x\n",
311 port, port_idx, length, *(uint32_t *)src);
316 case GUEST_FEATURES_PORT:
318 PrintError("Illegal write length for guest features\n");
322 blk_state->virtio_cfg.guest_features = *(uint32_t *)src;
323 PrintDebug("Setting Guest Features to %x\n", blk_state->virtio_cfg.guest_features);
326 case VRING_PG_NUM_PORT:
328 addr_t pfn = *(uint32_t *)src;
329 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
332 blk_state->queue.pfn = pfn;
334 blk_state->queue.ring_desc_addr = page_addr ;
335 blk_state->queue.ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
336 blk_state->queue.ring_used_addr = ( blk_state->queue.ring_avail_addr + \
337 sizeof(struct vring_avail) + \
338 (QUEUE_SIZE * sizeof(uint16_t)));
340 // round up to next page boundary.
341 blk_state->queue.ring_used_addr = (blk_state->queue.ring_used_addr + 0xfff) & ~0xfff;
343 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, blk_state->queue.ring_desc_addr, (addr_t *)&(blk_state->queue.desc)) == -1) {
344 PrintError("Could not translate ring descriptor address\n");
349 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, blk_state->queue.ring_avail_addr, (addr_t *)&(blk_state->queue.avail)) == -1) {
350 PrintError("Could not translate ring available address\n");
355 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, blk_state->queue.ring_used_addr, (addr_t *)&(blk_state->queue.used)) == -1) {
356 PrintError("Could not translate ring used address\n");
360 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
361 (void *)(blk_state->queue.ring_desc_addr),
362 (void *)(blk_state->queue.ring_avail_addr),
363 (void *)(blk_state->queue.ring_used_addr));
365 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
366 blk_state->queue.desc, blk_state->queue.avail, blk_state->queue.used);
369 PrintError("Illegal write length for page frame number\n");
373 case VRING_Q_SEL_PORT:
374 blk_state->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
376 if (blk_state->virtio_cfg.vring_queue_selector != 0) {
377 PrintError("Virtio Block device only uses 1 queue, selected %d\n",
378 blk_state->virtio_cfg.vring_queue_selector);
383 case VRING_Q_NOTIFY_PORT:
384 PrintDebug("Handling Kick\n");
385 if (handle_kick(blk_state) == -1) {
386 PrintError("Could not handle Block Notification\n");
390 case VIRTIO_STATUS_PORT:
391 blk_state->virtio_cfg.status = *(uint8_t *)src;
393 if (blk_state->virtio_cfg.status == 0) {
394 PrintDebug("Resetting device\n");
395 blk_reset(blk_state);
400 case VIRTIO_ISR_PORT:
401 blk_state->virtio_cfg.pci_isr = *(uint8_t *)src;
412 static int virtio_io_read(uint16_t port, void * dst, uint_t length, void * private_data) {
413 struct virtio_blk_state * blk_state = (struct virtio_blk_state *)private_data;
414 int port_idx = port % blk_state->io_range_size;
417 PrintDebug("VIRTIO BLOCK Read for port %d (index =%d), length=%d\n",
418 port, port_idx, length);
421 case HOST_FEATURES_PORT:
423 PrintError("Illegal read length for host features\n");
427 *(uint32_t *)dst = blk_state->virtio_cfg.host_features;
430 case VRING_PG_NUM_PORT:
432 PrintError("Illegal read length for page frame number\n");
436 *(uint32_t *)dst = blk_state->queue.pfn;
439 case VRING_SIZE_PORT:
441 PrintError("Illegal read length for vring size\n");
445 *(uint16_t *)dst = blk_state->queue.queue_size;
449 case VIRTIO_STATUS_PORT:
451 PrintError("Illegal read length for status\n");
455 *(uint8_t *)dst = blk_state->virtio_cfg.status;
458 case VIRTIO_ISR_PORT:
459 *(uint8_t *)dst = blk_state->virtio_cfg.pci_isr;
460 blk_state->virtio_cfg.pci_isr = 0;
461 v3_pci_lower_irq(blk_state->virtio_dev->pci_bus, 0, blk_state->pci_dev);
465 if ( (port_idx >= sizeof(struct virtio_config)) &&
466 (port_idx < (sizeof(struct virtio_config) + sizeof(struct blk_config))) ) {
467 int cfg_offset = port_idx - sizeof(struct virtio_config);
468 uint8_t * cfg_ptr = (uint8_t *)&(blk_state->block_cfg);
470 memcpy(dst, cfg_ptr + cfg_offset, length);
473 PrintError("Read of Unhandled Virtio Read\n");
486 static struct v3_device_ops dev_ops = {
488 .reset = virtio_reset,
497 static int register_dev(struct virtio_dev_state * virtio, struct virtio_blk_state * blk_state) {
499 struct pci_device * pci_dev = NULL;
500 struct v3_pci_bar bars[6];
501 int num_ports = sizeof(struct virtio_config) + sizeof(struct blk_config);
502 int tmp_ports = num_ports;
507 // This gets the number of ports, rounded up to a power of 2
508 blk_state->io_range_size = 1; // must be a power of 2
510 while (tmp_ports > 0) {
512 blk_state->io_range_size <<= 1;
515 // this is to account for any low order bits being set in num_ports
516 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
517 if ((num_ports & ((blk_state->io_range_size >> 1) - 1)) == 0) {
518 blk_state->io_range_size >>= 1;
522 for (i = 0; i < 6; i++) {
523 bars[i].type = PCI_BAR_NONE;
526 PrintDebug("Virtio-BLK io_range_size = %d\n", blk_state->io_range_size);
528 bars[0].type = PCI_BAR_IO;
529 bars[0].default_base_port = -1;
530 bars[0].num_ports = blk_state->io_range_size;
532 bars[0].io_read = virtio_io_read;
533 bars[0].io_write = virtio_io_write;
534 bars[0].private_data = blk_state;
536 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
537 0, PCI_AUTO_DEV_NUM, 0,
538 "LNX_VIRTIO_BLK", bars,
539 NULL, NULL, NULL, blk_state);
542 PrintError("Could not register PCI Device\n");
546 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
547 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
550 pci_dev->config_header.device_id = VIRTIO_BLOCK_DEV_ID;
551 pci_dev->config_header.class = PCI_CLASS_STORAGE;
552 pci_dev->config_header.subclass = PCI_STORAGE_SUBCLASS_OTHER;
554 pci_dev->config_header.subsystem_id = VIRTIO_BLOCK_SUBDEVICE_ID;
557 pci_dev->config_header.intr_pin = 1;
559 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
562 blk_state->pci_dev = pci_dev;
564 /* Block configuration */
565 blk_state->virtio_cfg.host_features = VIRTIO_SEG_MAX;
566 blk_state->block_cfg.max_seg = QUEUE_SIZE - 2;
568 // Virtio Block only uses one queue
569 blk_state->queue.queue_size = QUEUE_SIZE;
571 blk_state->virtio_dev = virtio;
573 blk_reset(blk_state);
580 static int connect_fn(struct guest_info * info,
581 void * frontend_data,
582 struct v3_dev_blk_ops * ops,
584 void * private_data) {
586 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
588 struct virtio_blk_state * blk_state = (struct virtio_blk_state *)V3_Malloc(sizeof(struct virtio_blk_state));
589 memset(blk_state, 0, sizeof(struct virtio_blk_state));
591 register_dev(virtio, blk_state);
593 blk_state->ops = ops;
594 blk_state->backend_data = private_data;
596 blk_state->block_cfg.capacity = ops->get_capacity(private_data);
598 PrintDebug("Virtio Capacity = %d -- 0x%p\n", (int)(virtio->block_cfg.capacity),
599 (void *)(addr_t)(virtio->block_cfg.capacity));
605 static int virtio_init(struct guest_info * vm, v3_cfg_tree_t * cfg) {
606 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
607 struct virtio_dev_state * virtio_state = NULL;
608 char * name = v3_cfg_val(cfg, "name");
610 PrintDebug("Initializing VIRTIO Block device\n");
612 if (pci_bus == NULL) {
613 PrintError("VirtIO devices require a PCI Bus");
618 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
619 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
621 INIT_LIST_HEAD(&(virtio_state->dev_list));
622 virtio_state->pci_bus = pci_bus;
623 virtio_state->vm = vm;
625 struct vm_device * dev = v3_allocate_device(name, &dev_ops, virtio_state);
626 if (v3_attach_device(vm, dev) == -1) {
627 PrintError("Could not attach device %s\n", name);
631 if (v3_dev_add_blk_frontend(vm, name, connect_fn, (void *)virtio_state) == -1) {
632 PrintError("Could not register %s as block frontend\n", name);
640 device_register("LNX_VIRTIO_BLK", virtio_init)