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 <devices/lnx_virtio_blk.h>
24 #include <devices/block_dev.h>
25 #include <palacios/vm_guest_mem.h>
27 #include <devices/pci.h>
30 #ifndef DEBUG_VIRTIO_BLK
32 #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 */
81 struct virtio_blk_state {
82 struct blk_config block_cfg;
83 struct virtio_config virtio_cfg;
85 struct vm_device * pci_bus;
86 struct pci_device * pci_dev;
88 struct virtio_queue queue;
91 struct v3_cd_ops * cd_ops;
92 struct v3_hd_ops * hd_ops;
95 v3_block_type_t block_type;
102 static int virtio_free(struct vm_device * dev) {
106 static int virtio_reset(struct vm_device * dev) {
107 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
109 virtio->queue.ring_desc_addr = 0;
110 virtio->queue.ring_avail_addr = 0;
111 virtio->queue.ring_used_addr = 0;
112 virtio->queue.pfn = 0;
113 virtio->queue.cur_avail_idx = 0;
115 virtio->virtio_cfg.status = 0;
116 virtio->virtio_cfg.pci_isr = 0;
121 static int handle_read_op(struct vm_device * dev, uint8_t * buf, uint64_t * sector, uint32_t len) {
122 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
125 if (virtio->block_type == BLOCK_DISK) {
126 if (len % HD_SECTOR_SIZE) {
127 PrintError("Write of something that is not a sector len %d, mod=%d\n", len, len % HD_SECTOR_SIZE);
132 PrintDebug("Reading Disk\n");
134 ret = virtio->hd_ops->read(buf, len / HD_SECTOR_SIZE, *sector, virtio->backend_data);
136 *sector += len / HD_SECTOR_SIZE;
138 } else if (virtio->block_type == BLOCK_CDROM) {
139 if (len % ATAPI_BLOCK_SIZE) {
140 PrintError("Write of something that is not an ATAPI block len %d, mod=%d\n", len, len % ATAPI_BLOCK_SIZE);
144 ret = virtio->cd_ops->read(buf, len / ATAPI_BLOCK_SIZE, *sector , virtio->backend_data);
146 *sector += len / ATAPI_BLOCK_SIZE;
153 static int handle_write_op(struct vm_device * dev, uint8_t * buf, uint64_t * sector, uint32_t len) {
154 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
157 if (virtio->block_type == BLOCK_DISK) {
158 if (len % HD_SECTOR_SIZE) {
159 PrintError("Write of something that is not a sector len %d, mod=%d\n", len, len % HD_SECTOR_SIZE);
163 PrintDebug("Writing Disk\n");
165 ret = virtio->hd_ops->write(buf, len / HD_SECTOR_SIZE, *sector, virtio->backend_data);
167 *sector += len / HD_SECTOR_SIZE;
175 // multiple block operations need to increment the sector
177 static int handle_block_op(struct vm_device * dev, struct blk_op_hdr * hdr,
178 struct vring_desc * buf_desc, uint8_t * status) {
179 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
180 uint8_t * buf = NULL;
182 PrintDebug("Handling Block op\n");
186 if (guest_pa_to_host_va(dev->vm, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
187 PrintError("Could not translate buffer address\n");
192 PrintDebug("Sector=%p Length=%d\n", (void *)(addr_t)(hdr->sector), buf_desc->length);
194 if (hdr->type == BLK_IN_REQ) {
195 if (virtio->block_type != BLOCK_NONE) {
196 if (handle_read_op(dev, buf, &(hdr->sector), buf_desc->length) == -1) {
197 *status = BLK_STATUS_ERR;
200 *status = BLK_STATUS_OK;
203 *status = BLK_STATUS_NOT_SUPPORTED;
206 } else if (hdr->type == BLK_OUT_REQ) {
207 if (virtio->block_type == BLOCK_DISK) {
208 if (handle_write_op(dev, buf, &(hdr->sector), buf_desc->length) == -1) {
209 *status = BLK_STATUS_ERR;
212 *status = BLK_STATUS_OK;
216 *status = BLK_STATUS_NOT_SUPPORTED;
219 } else if (hdr->type == BLK_SCSI_CMD) {
220 PrintError("VIRTIO: SCSI Command Not supported!!!\n");
221 *status = BLK_STATUS_NOT_SUPPORTED;
227 PrintDebug("Returning Status: %d\n", *status);
232 static int get_desc_count(struct virtio_queue * q, int index) {
233 struct vring_desc * tmp_desc = &(q->desc[index]);
236 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
237 tmp_desc = &(q->desc[tmp_desc->next]);
246 static int handle_kick(struct vm_device * dev) {
247 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
248 struct virtio_queue * q = &(virtio->queue);
250 PrintDebug("VIRTIO KICK: cur_index=%d (mod=%d), avail_index=%d\n",
251 q->cur_avail_idx, q->cur_avail_idx % QUEUE_SIZE, q->avail->index);
253 while (q->cur_avail_idx < q->avail->index) {
254 struct vring_desc * hdr_desc = NULL;
255 struct vring_desc * buf_desc = NULL;
256 struct vring_desc * status_desc = NULL;
257 struct blk_op_hdr hdr;
259 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
260 int desc_cnt = get_desc_count(q, desc_idx);
262 uint8_t * status_ptr = NULL;
263 uint8_t status = BLK_STATUS_OK;
264 uint32_t req_len = 0;
266 PrintDebug("Descriptor Count=%d, index=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE);
269 PrintError("Block operations must include at least 3 descriptors\n");
273 hdr_desc = &(q->desc[desc_idx]);
276 PrintDebug("Header Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", hdr_desc,
277 (void *)(hdr_desc->addr_gpa), hdr_desc->length, hdr_desc->flags, hdr_desc->next);
279 if (guest_pa_to_host_va(dev->vm, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
280 PrintError("Could not translate block header address\n");
284 // We copy the block op header out because we are going to modify its contents
285 memcpy(&hdr, (void *)hdr_addr, sizeof(struct blk_op_hdr));
287 PrintDebug("Blk Op Hdr (ptr=%p) type=%d, sector=%p\n", (void *)hdr_addr, hdr.type, (void *)hdr.sector);
289 desc_idx = hdr_desc->next;
291 for (i = 0; i < desc_cnt - 2; i++) {
292 uint8_t tmp_status = BLK_STATUS_OK;
294 buf_desc = &(q->desc[desc_idx]);
296 PrintDebug("Buffer Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", buf_desc,
297 (void *)(buf_desc->addr_gpa), buf_desc->length, buf_desc->flags, buf_desc->next);
299 if (handle_block_op(dev, &hdr, buf_desc, &tmp_status) == -1) {
300 PrintError("Error handling block operation\n");
304 if (tmp_status != BLK_STATUS_OK) {
308 req_len += buf_desc->length;
309 desc_idx = buf_desc->next;
312 status_desc = &(q->desc[desc_idx]);
314 PrintDebug("Status Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", status_desc,
315 (void *)(status_desc->addr_gpa), status_desc->length, status_desc->flags, status_desc->next);
317 if (guest_pa_to_host_va(dev->vm, status_desc->addr_gpa, (addr_t *)&(status_ptr)) == -1) {
318 PrintError("Could not translate status address\n");
322 req_len += status_desc->length;
323 *status_ptr = status;
325 q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
326 q->used->ring[q->used->index % QUEUE_SIZE].length = req_len; // What do we set this to????
332 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
333 PrintDebug("Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line);
334 v3_pci_raise_irq(virtio->pci_bus, 0, virtio->pci_dev);
335 virtio->virtio_cfg.pci_isr = 1;
341 static int virtio_io_write(uint16_t port, void * src, uint_t length, struct vm_device * dev) {
342 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
343 int port_idx = port % virtio->io_range_size;
346 PrintDebug("VIRTIO BLOCK Write for port %d (index=%d) len=%d, value=%x\n",
347 port, port_idx, length, *(uint32_t *)src);
352 case GUEST_FEATURES_PORT:
354 PrintError("Illegal write length for guest features\n");
358 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
359 PrintDebug("Setting Guest Features to %x\n", virtio->virtio_cfg.guest_features);
362 case VRING_PG_NUM_PORT:
364 addr_t pfn = *(uint32_t *)src;
365 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
368 virtio->queue.pfn = pfn;
370 virtio->queue.ring_desc_addr = page_addr ;
371 virtio->queue.ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
372 virtio->queue.ring_used_addr = ( virtio->queue.ring_avail_addr + \
373 sizeof(struct vring_avail) + \
374 (QUEUE_SIZE * sizeof(uint16_t)));
376 // round up to next page boundary.
377 virtio->queue.ring_used_addr = (virtio->queue.ring_used_addr + 0xfff) & ~0xfff;
379 if (guest_pa_to_host_va(dev->vm, virtio->queue.ring_desc_addr, (addr_t *)&(virtio->queue.desc)) == -1) {
380 PrintError("Could not translate ring descriptor address\n");
385 if (guest_pa_to_host_va(dev->vm, virtio->queue.ring_avail_addr, (addr_t *)&(virtio->queue.avail)) == -1) {
386 PrintError("Could not translate ring available address\n");
391 if (guest_pa_to_host_va(dev->vm, virtio->queue.ring_used_addr, (addr_t *)&(virtio->queue.used)) == -1) {
392 PrintError("Could not translate ring used address\n");
396 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
397 (void *)(virtio->queue.ring_desc_addr),
398 (void *)(virtio->queue.ring_avail_addr),
399 (void *)(virtio->queue.ring_used_addr));
401 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
402 virtio->queue.desc, virtio->queue.avail, virtio->queue.used);
405 PrintError("Illegal write length for page frame number\n");
409 case VRING_Q_SEL_PORT:
410 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
412 if (virtio->virtio_cfg.vring_queue_selector != 0) {
413 PrintError("Virtio Block device only uses 1 queue, selected %d\n",
414 virtio->virtio_cfg.vring_queue_selector);
419 case VRING_Q_NOTIFY_PORT:
420 PrintDebug("Handling Kick\n");
421 if (handle_kick(dev) == -1) {
422 PrintError("Could not handle Block Notification\n");
426 case VIRTIO_STATUS_PORT:
427 virtio->virtio_cfg.status = *(uint8_t *)src;
429 if (virtio->virtio_cfg.status == 0) {
430 PrintDebug("Resetting device\n");
436 case VIRTIO_ISR_PORT:
437 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
448 static int virtio_io_read(uint16_t port, void * dst, uint_t length, struct vm_device * dev) {
449 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
450 int port_idx = port % virtio->io_range_size;
453 PrintDebug("VIRTIO BLOCK Read for port %d (index =%d), length=%d\n",
454 port, port_idx, length);
457 case HOST_FEATURES_PORT:
459 PrintError("Illegal read length for host features\n");
463 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
466 case VRING_PG_NUM_PORT:
468 PrintError("Illegal read length for page frame number\n");
472 *(uint32_t *)dst = virtio->queue.pfn;
475 case VRING_SIZE_PORT:
477 PrintError("Illegal read length for vring size\n");
481 *(uint16_t *)dst = virtio->queue.queue_size;
485 case VIRTIO_STATUS_PORT:
487 PrintError("Illegal read length for status\n");
491 *(uint8_t *)dst = virtio->virtio_cfg.status;
494 case VIRTIO_ISR_PORT:
495 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
496 virtio->virtio_cfg.pci_isr = 0;
497 v3_pci_lower_irq(virtio->pci_bus, 0, virtio->pci_dev);
501 if ( (port_idx >= sizeof(struct virtio_config)) &&
502 (port_idx < (sizeof(struct virtio_config) + sizeof(struct blk_config))) ) {
503 int cfg_offset = port_idx - sizeof(struct virtio_config);
504 uint8_t * cfg_ptr = (uint8_t *)&(virtio->block_cfg);
506 memcpy(dst, cfg_ptr + cfg_offset, length);
509 PrintError("Read of Unhandled Virtio Read\n");
522 static struct v3_device_ops dev_ops = {
532 int v3_virtio_register_cdrom(struct vm_device * dev, struct v3_cd_ops * ops, void * private_data) {
533 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
535 virtio->block_type = BLOCK_CDROM;
536 virtio->cd_ops = ops;
537 virtio->backend_data = private_data;
539 virtio->block_cfg.capacity = ops->get_capacity(private_data);
545 int v3_virtio_register_harddisk(struct vm_device * dev, struct v3_hd_ops * ops, void * private_data) {
546 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
548 virtio->block_type = BLOCK_DISK;
549 virtio->hd_ops = ops;
550 virtio->backend_data = private_data;
552 virtio->block_cfg.capacity = ops->get_capacity(private_data);
554 PrintDebug("Virtio Capacity = %d -- 0x%p\n", (int)(virtio->block_cfg.capacity),
555 (void *)(addr_t)(virtio->block_cfg.capacity));
562 static int virtio_init(struct guest_info * vm, void * cfg_data) {
563 struct vm_device * pci_bus = v3_find_dev(vm, (char *)cfg_data);
564 struct virtio_blk_state * virtio_state = NULL;
565 struct pci_device * pci_dev = NULL;
567 PrintDebug("Initializing VIRTIO Block device\n");
569 if (pci_bus == NULL) {
570 PrintError("VirtIO devices require a PCI Bus");
575 virtio_state = (struct virtio_blk_state *)V3_Malloc(sizeof(struct virtio_blk_state));
576 memset(virtio_state, 0, sizeof(struct virtio_blk_state));
579 struct vm_device * dev = v3_allocate_device("LNX_VIRTIO_BLK", &dev_ops, virtio_state);
580 if (v3_attach_device(vm, dev) == -1) {
581 PrintError("Could not attach device %s\n", "LNX_VIRTIO_BLK");
586 // PCI initialization
588 struct v3_pci_bar bars[6];
589 int num_ports = sizeof(struct virtio_config) + sizeof(struct blk_config);
590 int tmp_ports = num_ports;
595 // This gets the number of ports, rounded up to a power of 2
596 virtio_state->io_range_size = 1; // must be a power of 2
598 while (tmp_ports > 0) {
600 virtio_state->io_range_size <<= 1;
603 // this is to account for any low order bits being set in num_ports
604 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
605 if ((num_ports & ((virtio_state->io_range_size >> 1) - 1)) == 0) {
606 virtio_state->io_range_size >>= 1;
610 for (i = 0; i < 6; i++) {
611 bars[i].type = PCI_BAR_NONE;
614 bars[0].type = PCI_BAR_IO;
615 bars[0].default_base_port = -1;
616 bars[0].num_ports = virtio_state->io_range_size;
618 bars[0].io_read = virtio_io_read;
619 bars[0].io_write = virtio_io_write;
621 pci_dev = v3_pci_register_device(pci_bus, PCI_STD_DEVICE,
622 0, PCI_AUTO_DEV_NUM, 0,
623 "LNX_VIRTIO_BLK", bars,
624 NULL, NULL, NULL, dev);
627 PrintError("Could not register PCI Device\n");
631 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
632 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
635 pci_dev->config_header.device_id = VIRTIO_BLOCK_DEV_ID;
636 pci_dev->config_header.class = PCI_CLASS_STORAGE;
637 pci_dev->config_header.subclass = PCI_STORAGE_SUBCLASS_OTHER;
639 pci_dev->config_header.subsystem_id = VIRTIO_BLOCK_SUBDEVICE_ID;
642 pci_dev->config_header.intr_pin = 1;
644 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
647 virtio_state->pci_dev = pci_dev;
648 virtio_state->pci_bus = pci_bus;
651 /* Block configuration */
652 virtio_state->virtio_cfg.host_features = VIRTIO_SEG_MAX;
653 virtio_state->block_cfg.max_seg = QUEUE_SIZE - 2;
655 // Virtio Block only uses one queue
656 virtio_state->queue.queue_size = QUEUE_SIZE;
662 virtio_state->backend_data = NULL;
663 virtio_state->block_type = BLOCK_NONE;
664 virtio_state->hd_ops = NULL;
670 device_register("LNX_VIRTIO_BLK", virtio_init)