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 V3_CONFIG_DEBUG_VIRTIO_BLK
31 #define PrintDebug(fmt, args...)
35 #define SECTOR_SIZE 512
37 #define BLK_CAPACITY_PORT 20
38 #define BLK_MAX_SIZE_PORT 28
39 #define BLK_MAX_SEG_PORT 32
40 #define BLK_CYLINDERS_PORT 36
41 #define BLK_HEADS_PORT 38
42 #define BLK_SECTS_PORT 39
46 #define BLK_SCSI_CMD 2
48 #define BLK_BARRIER_FLAG 0x80000000
50 #define BLK_STATUS_OK 0
51 #define BLK_STATUS_ERR 1
52 #define BLK_STATUS_NOT_SUPPORTED 2
64 uint8_t phys_block_exp;
65 uint8_t alignment_offset;
69 } __attribute__((packed));
77 } __attribute__((packed));
79 #define QUEUE_SIZE 128
81 /* Host Feature flags */
82 #define VIRTIO_BARRIER 0x01 /* Does host support barriers? */
83 #define VIRTIO_SIZE_MAX 0x02 /* Indicates maximum segment size */
84 #define VIRTIO_SEG_MAX 0x04 /* Indicates maximum # of segments */
85 #define VIRTIO_LEGACY_GEOM 0x10 /* Indicates support of legacy geometry */
88 struct virtio_dev_state {
89 struct vm_device * pci_bus;
90 struct list_head dev_list;
93 struct virtio_blk_state {
95 struct pci_device * pci_dev;
96 struct blk_config block_cfg;
97 struct virtio_config virtio_cfg;
100 struct virtio_queue queue;
102 struct v3_dev_blk_ops * ops;
108 struct virtio_dev_state * virtio_dev;
110 struct list_head dev_link;
116 static int blk_reset(struct virtio_blk_state * virtio) {
118 virtio->queue.ring_desc_addr = 0;
119 virtio->queue.ring_avail_addr = 0;
120 virtio->queue.ring_used_addr = 0;
121 virtio->queue.pfn = 0;
122 virtio->queue.cur_avail_idx = 0;
124 virtio->virtio_cfg.status = 0;
125 virtio->virtio_cfg.pci_isr = 0;
132 static int handle_read_op(struct virtio_blk_state * blk_state, uint8_t * buf, uint64_t * sector, uint64_t len) {
135 PrintDebug(info->vm_info, info, "Reading Disk\n");
136 ret = blk_state->ops->read(buf, (*sector) * SECTOR_SIZE, len, (void *)(blk_state->backend_data));
137 *sector += (len / SECTOR_SIZE);
143 static int handle_write_op(struct virtio_blk_state * blk_state, uint8_t * buf, uint64_t * sector, uint64_t len) {
146 PrintDebug(info->vm_info, info, "Writing Disk\n");
147 ret = blk_state->ops->write(buf, (*sector) * SECTOR_SIZE, len, (void *)(blk_state->backend_data));
148 *sector += (len / SECTOR_SIZE);
155 // multiple block operations need to increment the sector
157 static int handle_block_op(struct guest_info * core, struct virtio_blk_state * blk_state, struct blk_op_hdr * hdr,
158 struct vring_desc * buf_desc, uint8_t * status) {
159 uint8_t * buf = NULL;
161 PrintDebug(core->vm_info, core, "Handling Block op\n");
162 if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
163 PrintError(core->vm_info, core, "Could not translate buffer address\n");
167 PrintDebug(core->vm_info, core, "Sector=%p Length=%d\n", (void *)(addr_t)(hdr->sector), buf_desc->length);
169 if (hdr->type == BLK_IN_REQ) {
170 if (handle_read_op(blk_state, buf, &(hdr->sector), buf_desc->length) == -1) {
171 *status = BLK_STATUS_ERR;
174 *status = BLK_STATUS_OK;
176 } else if (hdr->type == BLK_OUT_REQ) {
177 if (handle_write_op(blk_state, buf, &(hdr->sector), buf_desc->length) == -1) {
178 *status = BLK_STATUS_ERR;
181 *status = BLK_STATUS_OK;
183 } else if (hdr->type == BLK_SCSI_CMD) {
184 PrintError(core->vm_info, core, "VIRTIO: SCSI Command Not supported!!!\n");
185 *status = BLK_STATUS_NOT_SUPPORTED;
189 PrintDebug(core->vm_info, core, "Returning Status: %d\n", *status);
194 static int get_desc_count(struct virtio_queue * q, int index) {
195 struct vring_desc * tmp_desc = &(q->desc[index]);
198 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
199 tmp_desc = &(q->desc[tmp_desc->next]);
208 static int handle_kick(struct guest_info * core, struct virtio_blk_state * blk_state) {
209 struct virtio_queue * q = &(blk_state->queue);
211 PrintDebug(core->vm_info, core, "VIRTIO KICK: cur_index=%d (mod=%d), avail_index=%d\n",
212 q->cur_avail_idx, q->cur_avail_idx % QUEUE_SIZE, q->avail->index);
214 while (q->cur_avail_idx != q->avail->index) {
215 struct vring_desc * hdr_desc = NULL;
216 struct vring_desc * buf_desc = NULL;
217 struct vring_desc * status_desc = NULL;
218 struct blk_op_hdr hdr;
220 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
221 int desc_cnt = get_desc_count(q, desc_idx);
223 uint8_t * status_ptr = NULL;
224 uint8_t status = BLK_STATUS_OK;
225 uint32_t req_len = 0;
227 PrintDebug(core->vm_info, core, "Descriptor Count=%d, index=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE);
230 PrintError(core->vm_info, core, "Block operations must include at least 3 descriptors\n");
234 hdr_desc = &(q->desc[desc_idx]);
237 PrintDebug(core->vm_info, core, "Header Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", hdr_desc,
238 (void *)(hdr_desc->addr_gpa), hdr_desc->length, hdr_desc->flags, hdr_desc->next);
240 if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
241 PrintError(core->vm_info, core, "Could not translate block header address\n");
245 // We copy the block op header out because we are going to modify its contents
246 memcpy(&hdr, (void *)hdr_addr, sizeof(struct blk_op_hdr));
248 PrintDebug(core->vm_info, core, "Blk Op Hdr (ptr=%p) type=%d, sector=%p\n", (void *)hdr_addr, hdr.type, (void *)hdr.sector);
250 desc_idx = hdr_desc->next;
252 for (i = 0; i < desc_cnt - 2; i++) {
253 uint8_t tmp_status = BLK_STATUS_OK;
255 buf_desc = &(q->desc[desc_idx]);
257 PrintDebug(core->vm_info, core, "Buffer Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", buf_desc,
258 (void *)(buf_desc->addr_gpa), buf_desc->length, buf_desc->flags, buf_desc->next);
260 if (handle_block_op(core, blk_state, &hdr, buf_desc, &tmp_status) == -1) {
261 PrintError(core->vm_info, core, "Error handling block operation\n");
265 if (tmp_status != BLK_STATUS_OK) {
269 req_len += buf_desc->length;
270 desc_idx = buf_desc->next;
273 status_desc = &(q->desc[desc_idx]);
275 PrintDebug(core->vm_info, core, "Status Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", status_desc,
276 (void *)(status_desc->addr_gpa), status_desc->length, status_desc->flags, status_desc->next);
278 if (v3_gpa_to_hva(core, status_desc->addr_gpa, (addr_t *)&(status_ptr)) == -1) {
279 PrintError(core->vm_info, core, "Could not translate status address\n");
283 req_len += status_desc->length;
284 *status_ptr = status;
286 q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
287 q->used->ring[q->used->index % QUEUE_SIZE].length = req_len; // What do we set this to????
293 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
294 PrintDebug(core->vm_info, core, "Raising IRQ %d\n", blk_state->pci_dev->config_header.intr_line);
295 v3_pci_raise_irq(blk_state->virtio_dev->pci_bus, blk_state->pci_dev, 0);
296 blk_state->virtio_cfg.pci_isr = 1;
302 static int virtio_io_write(struct guest_info * core, uint16_t port, void * src, uint_t length, void * private_data) {
303 struct virtio_blk_state * blk_state = (struct virtio_blk_state *)private_data;
304 int port_idx = port % blk_state->io_range_size;
307 PrintDebug(core->vm_info, core, "VIRTIO BLOCK Write for port %d (index=%d) len=%d, value=%x\n",
308 port, port_idx, length, *(uint32_t *)src);
313 case GUEST_FEATURES_PORT:
315 PrintError(core->vm_info, core, "Illegal write length for guest features\n");
319 blk_state->virtio_cfg.guest_features = *(uint32_t *)src;
320 PrintDebug(core->vm_info, core, "Setting Guest Features to %x\n", blk_state->virtio_cfg.guest_features);
323 case VRING_PG_NUM_PORT:
325 addr_t pfn = *(uint32_t *)src;
326 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
329 blk_state->queue.pfn = pfn;
331 blk_state->queue.ring_desc_addr = page_addr ;
332 blk_state->queue.ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
333 blk_state->queue.ring_used_addr = ( blk_state->queue.ring_avail_addr + \
334 sizeof(struct vring_avail) + \
335 (QUEUE_SIZE * sizeof(uint16_t)));
337 // round up to next page boundary.
338 blk_state->queue.ring_used_addr = (blk_state->queue.ring_used_addr + 0xfff) & ~0xfff;
340 if (v3_gpa_to_hva(core, blk_state->queue.ring_desc_addr, (addr_t *)&(blk_state->queue.desc)) == -1) {
341 PrintError(core->vm_info, core, "Could not translate ring descriptor address\n");
346 if (v3_gpa_to_hva(core, blk_state->queue.ring_avail_addr, (addr_t *)&(blk_state->queue.avail)) == -1) {
347 PrintError(core->vm_info, core, "Could not translate ring available address\n");
352 if (v3_gpa_to_hva(core, blk_state->queue.ring_used_addr, (addr_t *)&(blk_state->queue.used)) == -1) {
353 PrintError(core->vm_info, core, "Could not translate ring used address\n");
357 PrintDebug(core->vm_info, core, "RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
358 (void *)(blk_state->queue.ring_desc_addr),
359 (void *)(blk_state->queue.ring_avail_addr),
360 (void *)(blk_state->queue.ring_used_addr));
362 PrintDebug(core->vm_info, core, "RingDesc=%p, Avail=%p, Used=%p\n",
363 blk_state->queue.desc, blk_state->queue.avail, blk_state->queue.used);
366 PrintError(core->vm_info, core, "Illegal write length for page frame number\n");
370 case VRING_Q_SEL_PORT:
371 blk_state->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
373 if (blk_state->virtio_cfg.vring_queue_selector != 0) {
374 PrintError(core->vm_info, core, "Virtio Block device only uses 1 queue, selected %d\n",
375 blk_state->virtio_cfg.vring_queue_selector);
380 case VRING_Q_NOTIFY_PORT:
381 PrintDebug(core->vm_info, core, "Handling Kick\n");
382 if (handle_kick(core, blk_state) == -1) {
383 PrintError(core->vm_info, core, "Could not handle Block Notification\n");
387 case VIRTIO_STATUS_PORT:
388 blk_state->virtio_cfg.status = *(uint8_t *)src;
390 if (blk_state->virtio_cfg.status == 0) {
391 PrintDebug(core->vm_info, core, "Resetting device\n");
392 blk_reset(blk_state);
397 case VIRTIO_ISR_PORT:
398 blk_state->virtio_cfg.pci_isr = *(uint8_t *)src;
409 static int virtio_io_read(struct guest_info * core, uint16_t port, void * dst, uint_t length, void * private_data) {
410 struct virtio_blk_state * blk_state = (struct virtio_blk_state *)private_data;
411 int port_idx = port % blk_state->io_range_size;
414 PrintDebug(core->vm_info, core, "VIRTIO BLOCK Read for port %d (index =%d), length=%d\n",
415 port, port_idx, length);
419 case HOST_FEATURES_PORT:
420 case HOST_FEATURES_PORT + 1:
421 case HOST_FEATURES_PORT + 2:
422 case HOST_FEATURES_PORT + 3:
423 if (port_idx + length > HOST_FEATURES_PORT + 4) {
424 PrintError(core->vm_info, core, "Illegal read length for host features (len=%d)\n", length);
428 memcpy(dst, &(blk_state->virtio_cfg.host_features), length);
430 case VRING_PG_NUM_PORT:
431 case VRING_PG_NUM_PORT + 1:
432 case VRING_PG_NUM_PORT + 2:
433 case VRING_PG_NUM_PORT + 3:
434 if (port_idx + length > VRING_PG_NUM_PORT + 4) {
435 PrintError(core->vm_info, core, "Illegal read length for vring pg num (len=%d)\n", length);
439 memcpy(dst, &(blk_state->queue.pfn), length);
441 case VRING_SIZE_PORT:
442 case VRING_SIZE_PORT + 1:
444 PrintError(core->vm_info, core, "Illegal read length for vring size (len=%d)\n", length);
448 memcpy(dst, &(blk_state->queue.queue_size), length);
452 case VIRTIO_STATUS_PORT:
454 PrintError(core->vm_info, core, "Illegal read length for status (len=%d)\n", length);
458 *(uint8_t *)dst = blk_state->virtio_cfg.status;
461 case VIRTIO_ISR_PORT:
462 *(uint8_t *)dst = blk_state->virtio_cfg.pci_isr;
463 blk_state->virtio_cfg.pci_isr = 0;
464 v3_pci_lower_irq(blk_state->virtio_dev->pci_bus, blk_state->pci_dev, 0);
468 if ( (port_idx >= sizeof(struct virtio_config)) &&
469 (port_idx < (sizeof(struct virtio_config) + sizeof(struct blk_config))) ) {
470 int cfg_offset = port_idx - sizeof(struct virtio_config);
471 uint8_t * cfg_ptr = (uint8_t *)&(blk_state->block_cfg);
473 memcpy(dst, cfg_ptr + cfg_offset, length);
477 PrintError(core->vm_info,core,"Read of Unhandled Virtio Read. Returning 0\n");
480 } else if (length == 2) {
481 *(uint16_t *)dst = 0;
482 } else if (length == 4) {
483 *(uint32_t *)dst = 0;
495 static int virtio_free(struct virtio_dev_state * virtio) {
496 struct virtio_blk_state * blk_state = NULL;
497 struct virtio_blk_state * tmp = NULL;
499 list_for_each_entry_safe(blk_state, tmp, &(virtio->dev_list), dev_link) {
501 // unregister from PCI
503 list_del(&(blk_state->dev_link));
515 static struct v3_device_ops dev_ops = {
516 .free = (int (*)(void *))virtio_free,
524 static int register_dev(struct virtio_dev_state * virtio, struct virtio_blk_state * blk_state) {
526 struct pci_device * pci_dev = NULL;
527 struct v3_pci_bar bars[6];
528 int num_ports = sizeof(struct virtio_config) + sizeof(struct blk_config);
529 int tmp_ports = num_ports;
534 // This gets the number of ports, rounded up to a power of 2
535 blk_state->io_range_size = 1; // must be a power of 2
537 while (tmp_ports > 0) {
539 blk_state->io_range_size <<= 1;
542 // this is to account for any low order bits being set in num_ports
543 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
544 if ((num_ports & ((blk_state->io_range_size >> 1) - 1)) == 0) {
545 blk_state->io_range_size >>= 1;
549 for (i = 0; i < 6; i++) {
550 bars[i].type = PCI_BAR_NONE;
553 PrintDebug(VM_NONE, VCORE_NONE, "Virtio-BLK io_range_size = %d\n", blk_state->io_range_size);
555 bars[0].type = PCI_BAR_IO;
556 bars[0].default_base_port = -1;
557 bars[0].num_ports = blk_state->io_range_size;
559 bars[0].io_read = virtio_io_read;
560 bars[0].io_write = virtio_io_write;
561 bars[0].private_data = blk_state;
563 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
564 0, PCI_AUTO_DEV_NUM, 0,
565 "LNX_VIRTIO_BLK", bars,
566 NULL, NULL, NULL, NULL, blk_state);
569 PrintError(VM_NONE, VCORE_NONE, "Could not register PCI Device\n");
573 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
574 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
577 pci_dev->config_header.device_id = VIRTIO_BLOCK_DEV_ID;
578 pci_dev->config_header.class = PCI_CLASS_STORAGE;
579 pci_dev->config_header.subclass = PCI_STORAGE_SUBCLASS_OTHER;
581 pci_dev->config_header.subsystem_id = VIRTIO_BLOCK_SUBDEVICE_ID;
584 pci_dev->config_header.intr_pin = 1;
586 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
589 blk_state->pci_dev = pci_dev;
592 /* Add backend to list of devices */
593 list_add(&(blk_state->dev_link), &(virtio->dev_list));
595 /* Block configuration */
596 blk_state->virtio_cfg.host_features = VIRTIO_SEG_MAX;
597 blk_state->block_cfg.max_seg = QUEUE_SIZE - 2;
600 // Virtio Block only uses one queue
601 blk_state->queue.queue_size = QUEUE_SIZE;
603 blk_state->virtio_dev = virtio;
605 blk_reset(blk_state);
612 static int connect_fn(struct v3_vm_info * vm,
613 void * frontend_data,
614 struct v3_dev_blk_ops * ops,
616 void * private_data) {
618 struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
620 struct virtio_blk_state * blk_state = (struct virtio_blk_state *)V3_Malloc(sizeof(struct virtio_blk_state));
623 PrintError(vm, VCORE_NONE, "Cannot allocate in connect\n");
628 memset(blk_state, 0, sizeof(struct virtio_blk_state));
630 register_dev(virtio, blk_state);
632 blk_state->ops = ops;
633 blk_state->backend_data = private_data;
635 blk_state->block_cfg.capacity = ops->get_capacity(private_data) / SECTOR_SIZE;
637 PrintDebug(vm, VCORE_NONE, "Virtio Capacity = %d -- 0x%p\n", (int)(blk_state->block_cfg.capacity),
638 (void *)(addr_t)(blk_state->block_cfg.capacity));
644 static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) {
645 struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus"));
646 struct virtio_dev_state * virtio_state = NULL;
647 char * dev_id = v3_cfg_val(cfg, "ID");
649 PrintDebug(vm, VCORE_NONE, "Initializing VIRTIO Block device\n");
651 if (pci_bus == NULL) {
652 PrintError(vm, VCORE_NONE, "VirtIO devices require a PCI Bus");
657 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
660 PrintError(vm, VCORE_NONE, "Cannot allocate in init\n");
664 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
666 INIT_LIST_HEAD(&(virtio_state->dev_list));
667 virtio_state->pci_bus = pci_bus;
670 struct vm_device * dev = v3_add_device(vm, dev_id, &dev_ops, virtio_state);
673 PrintError(vm, VCORE_NONE, "Could not attach device %s\n", dev_id);
674 V3_Free(virtio_state);
678 if (v3_dev_add_blk_frontend(vm, dev_id, connect_fn, (void *)virtio_state) == -1) {
679 PrintError(vm, VCORE_NONE, "Could not register %s as block frontend\n", dev_id);
680 v3_remove_device(dev);
688 device_register("LNX_VIRTIO_BLK", virtio_init)