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>
31 #ifndef CONFIG_DEBUG_VIRTIO_BLK
33 #define PrintDebug(fmt, args...)
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
62 } __attribute__((packed));
70 } __attribute__((packed));
72 #define QUEUE_SIZE 128
74 /* Host Feature flags */
75 #define VIRTIO_BARRIER 0x01 /* Does host support barriers? */
76 #define VIRTIO_SIZE_MAX 0x02 /* Indicates maximum segment size */
77 #define VIRTIO_SEG_MAX 0x04 /* Indicates maximum # of segments */
78 #define VIRTIO_LEGACY_GEOM 0x10 /* Indicates support of legacy geometry */
81 struct virtio_dev_state {
82 struct vm_device * pci_bus;
83 struct list_head dev_list;
84 struct guest_info * vm;
87 struct virtio_blk_state {
89 struct pci_device * pci_dev;
90 struct blk_config block_cfg;
91 struct virtio_config virtio_cfg;
94 struct virtio_queue queue;
97 struct v3_cd_ops * cd_ops;
98 struct v3_hd_ops * hd_ops;
101 v3_block_type_t block_type;
106 struct virtio_dev_state * virtio_dev;
108 struct list_head dev_link;
112 static int virtio_free(struct vm_device * dev) {
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;
130 static int virtio_reset(struct vm_device * dev) {
131 struct virtio_dev_state * dev_state = (struct virtio_dev_state *)(dev->private_data);
132 struct virtio_blk_state * blk_state = NULL;
134 list_for_each_entry(blk_state, &(dev_state->dev_list), dev_link) {
135 blk_reset(blk_state);
141 static int handle_read_op(struct virtio_blk_state * blk_state, uint8_t * buf, uint64_t * sector, uint32_t len) {
144 if (blk_state->block_type == BLOCK_DISK) {
145 if (len % HD_SECTOR_SIZE) {
146 PrintError("Write of something that is not a sector len %d, mod=%d\n", len, len % HD_SECTOR_SIZE);
151 PrintDebug("Reading Disk\n");
153 ret = blk_state->hd_ops->read(buf, len / HD_SECTOR_SIZE, *sector, blk_state->backend_data);
155 *sector += len / HD_SECTOR_SIZE;
157 } else if (blk_state->block_type == BLOCK_CDROM) {
158 if (len % ATAPI_BLOCK_SIZE) {
159 PrintError("Write of something that is not an ATAPI block len %d, mod=%d\n", len, len % ATAPI_BLOCK_SIZE);
163 ret = blk_state->cd_ops->read(buf, len / ATAPI_BLOCK_SIZE, *sector , blk_state->backend_data);
165 *sector += len / ATAPI_BLOCK_SIZE;
172 static int handle_write_op(struct virtio_blk_state * blk_state, uint8_t * buf, uint64_t * sector, uint32_t len) {
175 if (blk_state->block_type == BLOCK_DISK) {
176 if (len % HD_SECTOR_SIZE) {
177 PrintError("Write of something that is not a sector len %d, mod=%d\n", len, len % HD_SECTOR_SIZE);
181 PrintDebug("Writing Disk\n");
183 ret = blk_state->hd_ops->write(buf, len / HD_SECTOR_SIZE, *sector, blk_state->backend_data);
185 *sector += len / HD_SECTOR_SIZE;
193 // multiple block operations need to increment the sector
195 static int handle_block_op(struct virtio_blk_state * blk_state, struct blk_op_hdr * hdr,
196 struct vring_desc * buf_desc, uint8_t * status) {
197 uint8_t * buf = NULL;
199 PrintDebug("Handling Block op\n");
203 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
204 PrintError("Could not translate buffer address\n");
209 PrintDebug("Sector=%p Length=%d\n", (void *)(addr_t)(hdr->sector), buf_desc->length);
211 if (hdr->type == BLK_IN_REQ) {
212 if (blk_state->block_type != BLOCK_NONE) {
213 if (handle_read_op(blk_state, buf, &(hdr->sector), buf_desc->length) == -1) {
214 *status = BLK_STATUS_ERR;
217 *status = BLK_STATUS_OK;
220 *status = BLK_STATUS_NOT_SUPPORTED;
223 } else if (hdr->type == BLK_OUT_REQ) {
224 if (blk_state->block_type == BLOCK_DISK) {
225 if (handle_write_op(blk_state, buf, &(hdr->sector), buf_desc->length) == -1) {
226 *status = BLK_STATUS_ERR;
229 *status = BLK_STATUS_OK;
233 *status = BLK_STATUS_NOT_SUPPORTED;
236 } else if (hdr->type == BLK_SCSI_CMD) {
237 PrintError("VIRTIO: SCSI Command Not supported!!!\n");
238 *status = BLK_STATUS_NOT_SUPPORTED;
244 PrintDebug("Returning Status: %d\n", *status);
249 static int get_desc_count(struct virtio_queue * q, int index) {
250 struct vring_desc * tmp_desc = &(q->desc[index]);
253 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
254 tmp_desc = &(q->desc[tmp_desc->next]);
263 static int handle_kick(struct virtio_blk_state * blk_state) {
264 struct virtio_queue * q = &(blk_state->queue);
266 PrintDebug("VIRTIO KICK: cur_index=%d (mod=%d), avail_index=%d\n",
267 q->cur_avail_idx, q->cur_avail_idx % QUEUE_SIZE, q->avail->index);
269 while (q->cur_avail_idx < q->avail->index) {
270 struct vring_desc * hdr_desc = NULL;
271 struct vring_desc * buf_desc = NULL;
272 struct vring_desc * status_desc = NULL;
273 struct blk_op_hdr hdr;
275 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
276 int desc_cnt = get_desc_count(q, desc_idx);
278 uint8_t * status_ptr = NULL;
279 uint8_t status = BLK_STATUS_OK;
280 uint32_t req_len = 0;
282 PrintDebug("Descriptor Count=%d, index=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE);
285 PrintError("Block operations must include at least 3 descriptors\n");
289 hdr_desc = &(q->desc[desc_idx]);
292 PrintDebug("Header Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", hdr_desc,
293 (void *)(hdr_desc->addr_gpa), hdr_desc->length, hdr_desc->flags, hdr_desc->next);
295 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
296 PrintError("Could not translate block header address\n");
300 // We copy the block op header out because we are going to modify its contents
301 memcpy(&hdr, (void *)hdr_addr, sizeof(struct blk_op_hdr));
303 PrintDebug("Blk Op Hdr (ptr=%p) type=%d, sector=%p\n", (void *)hdr_addr, hdr.type, (void *)hdr.sector);
305 desc_idx = hdr_desc->next;
307 for (i = 0; i < desc_cnt - 2; i++) {
308 uint8_t tmp_status = BLK_STATUS_OK;
310 buf_desc = &(q->desc[desc_idx]);
312 PrintDebug("Buffer Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", buf_desc,
313 (void *)(buf_desc->addr_gpa), buf_desc->length, buf_desc->flags, buf_desc->next);
315 if (handle_block_op(blk_state, &hdr, buf_desc, &tmp_status) == -1) {
316 PrintError("Error handling block operation\n");
320 if (tmp_status != BLK_STATUS_OK) {
324 req_len += buf_desc->length;
325 desc_idx = buf_desc->next;
328 status_desc = &(q->desc[desc_idx]);
330 PrintDebug("Status Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", status_desc,
331 (void *)(status_desc->addr_gpa), status_desc->length, status_desc->flags, status_desc->next);
333 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, status_desc->addr_gpa, (addr_t *)&(status_ptr)) == -1) {
334 PrintError("Could not translate status address\n");
338 req_len += status_desc->length;
339 *status_ptr = status;
341 q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
342 q->used->ring[q->used->index % QUEUE_SIZE].length = req_len; // What do we set this to????
348 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
349 PrintDebug("Raising IRQ %d\n", blk_state->pci_dev->config_header.intr_line);
350 v3_pci_raise_irq(blk_state->virtio_dev->pci_bus, 0, blk_state->pci_dev);
351 blk_state->virtio_cfg.pci_isr = 1;
357 static int virtio_io_write(uint16_t port, void * src, uint_t length, void * private_data) {
358 struct virtio_blk_state * blk_state = (struct virtio_blk_state *)private_data;
359 int port_idx = port % blk_state->io_range_size;
362 PrintDebug("VIRTIO BLOCK Write for port %d (index=%d) len=%d, value=%x\n",
363 port, port_idx, length, *(uint32_t *)src);
368 case GUEST_FEATURES_PORT:
370 PrintError("Illegal write length for guest features\n");
374 blk_state->virtio_cfg.guest_features = *(uint32_t *)src;
375 PrintDebug("Setting Guest Features to %x\n", blk_state->virtio_cfg.guest_features);
378 case VRING_PG_NUM_PORT:
380 addr_t pfn = *(uint32_t *)src;
381 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
384 blk_state->queue.pfn = pfn;
386 blk_state->queue.ring_desc_addr = page_addr ;
387 blk_state->queue.ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
388 blk_state->queue.ring_used_addr = ( blk_state->queue.ring_avail_addr + \
389 sizeof(struct vring_avail) + \
390 (QUEUE_SIZE * sizeof(uint16_t)));
392 // round up to next page boundary.
393 blk_state->queue.ring_used_addr = (blk_state->queue.ring_used_addr + 0xfff) & ~0xfff;
395 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, blk_state->queue.ring_desc_addr, (addr_t *)&(blk_state->queue.desc)) == -1) {
396 PrintError("Could not translate ring descriptor address\n");
401 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, blk_state->queue.ring_avail_addr, (addr_t *)&(blk_state->queue.avail)) == -1) {
402 PrintError("Could not translate ring available address\n");
407 if (guest_pa_to_host_va(blk_state->virtio_dev->vm, blk_state->queue.ring_used_addr, (addr_t *)&(blk_state->queue.used)) == -1) {
408 PrintError("Could not translate ring used address\n");
412 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
413 (void *)(blk_state->queue.ring_desc_addr),
414 (void *)(blk_state->queue.ring_avail_addr),
415 (void *)(blk_state->queue.ring_used_addr));
417 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
418 blk_state->queue.desc, blk_state->queue.avail, blk_state->queue.used);
421 PrintError("Illegal write length for page frame number\n");
425 case VRING_Q_SEL_PORT:
426 blk_state->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
428 if (blk_state->virtio_cfg.vring_queue_selector != 0) {
429 PrintError("Virtio Block device only uses 1 queue, selected %d\n",
430 blk_state->virtio_cfg.vring_queue_selector);
435 case VRING_Q_NOTIFY_PORT:
436 PrintDebug("Handling Kick\n");
437 if (handle_kick(blk_state) == -1) {
438 PrintError("Could not handle Block Notification\n");
442 case VIRTIO_STATUS_PORT:
443 blk_state->virtio_cfg.status = *(uint8_t *)src;
445 if (blk_state->virtio_cfg.status == 0) {
446 PrintDebug("Resetting device\n");
447 blk_reset(blk_state);
452 case VIRTIO_ISR_PORT:
453 blk_state->virtio_cfg.pci_isr = *(uint8_t *)src;
464 static int virtio_io_read(uint16_t port, void * dst, uint_t length, void * private_data) {
465 struct virtio_blk_state * blk_state = (struct virtio_blk_state *)private_data;
466 int port_idx = port % blk_state->io_range_size;
469 PrintDebug("VIRTIO BLOCK Read for port %d (index =%d), length=%d\n",
470 port, port_idx, length);
473 case HOST_FEATURES_PORT:
475 PrintError("Illegal read length for host features\n");
479 *(uint32_t *)dst = blk_state->virtio_cfg.host_features;
482 case VRING_PG_NUM_PORT:
484 PrintError("Illegal read length for page frame number\n");
488 *(uint32_t *)dst = blk_state->queue.pfn;
491 case VRING_SIZE_PORT:
493 PrintError("Illegal read length for vring size\n");
497 *(uint16_t *)dst = blk_state->queue.queue_size;
501 case VIRTIO_STATUS_PORT:
503 PrintError("Illegal read length for status\n");
507 *(uint8_t *)dst = blk_state->virtio_cfg.status;
510 case VIRTIO_ISR_PORT:
511 *(uint8_t *)dst = blk_state->virtio_cfg.pci_isr;
512 blk_state->virtio_cfg.pci_isr = 0;
513 v3_pci_lower_irq(blk_state->virtio_dev->pci_bus, 0, blk_state->pci_dev);
517 if ( (port_idx >= sizeof(struct virtio_config)) &&
518 (port_idx < (sizeof(struct virtio_config) + sizeof(struct blk_config))) ) {
519 int cfg_offset = port_idx - sizeof(struct virtio_config);
520 uint8_t * cfg_ptr = (uint8_t *)&(blk_state->block_cfg);
522 memcpy(dst, cfg_ptr + cfg_offset, length);
525 PrintError("Read of Unhandled Virtio Read\n");
538 static struct v3_device_ops dev_ops = {
540 .reset = virtio_reset,
549 static int register_dev(struct virtio_dev_state * virtio, struct virtio_blk_state * blk_state) {
551 struct pci_device * pci_dev = NULL;
552 struct v3_pci_bar bars[6];
553 int num_ports = sizeof(struct virtio_config) + sizeof(struct blk_config);
554 int tmp_ports = num_ports;
559 // This gets the number of ports, rounded up to a power of 2
560 blk_state->io_range_size = 1; // must be a power of 2
562 while (tmp_ports > 0) {
564 blk_state->io_range_size <<= 1;
567 // this is to account for any low order bits being set in num_ports
568 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
569 if ((num_ports & ((blk_state->io_range_size >> 1) - 1)) == 0) {
570 blk_state->io_range_size >>= 1;
574 for (i = 0; i < 6; i++) {
575 bars[i].type = PCI_BAR_NONE;
578 PrintDebug("Virtio-BLK io_range_size = %d\n", blk_state->io_range_size);
580 bars[0].type = PCI_BAR_IO;
581 bars[0].default_base_port = -1;
582 bars[0].num_ports = blk_state->io_range_size;
584 bars[0].io_read = virtio_io_read;
585 bars[0].io_write = virtio_io_write;
586 bars[0].private_data = blk_state;
588 pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE,
589 0, PCI_AUTO_DEV_NUM, 0,
590 "LNX_VIRTIO_BLK", bars,
591 NULL, NULL, NULL, blk_state);
594 PrintError("Could not register PCI Device\n");
598 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
599 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
602 pci_dev->config_header.device_id = VIRTIO_BLOCK_DEV_ID;
603 pci_dev->config_header.class = PCI_CLASS_STORAGE;
604 pci_dev->config_header.subclass = PCI_STORAGE_SUBCLASS_OTHER;
606 pci_dev->config_header.subsystem_id = VIRTIO_BLOCK_SUBDEVICE_ID;
609 pci_dev->config_header.intr_pin = 1;
611 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
614 blk_state->pci_dev = pci_dev;
616 /* Block configuration */
617 blk_state->virtio_cfg.host_features = VIRTIO_SEG_MAX;
618 blk_state->block_cfg.max_seg = QUEUE_SIZE - 2;
620 // Virtio Block only uses one queue
621 blk_state->queue.queue_size = QUEUE_SIZE;
623 blk_state->virtio_dev = virtio;
625 blk_reset(blk_state);
632 int v3_virtio_register_cdrom(struct vm_device * dev, struct v3_cd_ops * ops, void * private_data) {
633 struct virtio_dev_state * virtio = (struct virtio_dev_state *)dev->private_data;
634 struct virtio_blk_state * blk_state = (struct virtio_blk_state *)V3_Malloc(sizeof(struct virtio_blk_state));
635 memset(blk_state, 0, sizeof(struct virtio_blk_state));
637 register_dev(virtio, blk_state);
639 blk_state->block_type = BLOCK_CDROM;
640 blk_state->cd_ops = ops;
641 blk_state->backend_data = private_data;
643 blk_state->block_cfg.capacity = ops->get_capacity(private_data);
649 int v3_virtio_register_harddisk(struct vm_device * dev, struct v3_hd_ops * ops, void * private_data) {
650 struct virtio_dev_state * virtio = (struct virtio_dev_state *)dev->private_data;
651 struct virtio_blk_state * blk_state = (struct virtio_blk_state *)V3_Malloc(sizeof(struct virtio_blk_state));
652 memset(blk_state, 0, sizeof(struct virtio_blk_state));
654 register_dev(virtio, blk_state);
656 blk_state->block_type = BLOCK_DISK;
657 blk_state->hd_ops = ops;
658 blk_state->backend_data = private_data;
660 blk_state->block_cfg.capacity = ops->get_capacity(private_data);
662 PrintDebug("Virtio Capacity = %d -- 0x%p\n", (int)(virtio->block_cfg.capacity),
663 (void *)(addr_t)(virtio->block_cfg.capacity));
671 static int virtio_init(struct guest_info * vm, void * cfg_data) {
672 struct vm_device * pci_bus = v3_find_dev(vm, (char *)cfg_data);
673 struct virtio_dev_state * virtio_state = NULL;
676 PrintDebug("Initializing VIRTIO Block device\n");
678 if (pci_bus == NULL) {
679 PrintError("VirtIO devices require a PCI Bus");
684 virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
685 memset(virtio_state, 0, sizeof(struct virtio_dev_state));
687 INIT_LIST_HEAD(&(virtio_state->dev_list));
688 virtio_state->pci_bus = pci_bus;
689 virtio_state->vm = vm;
691 struct vm_device * dev = v3_allocate_device("LNX_VIRTIO_BLK", &dev_ops, virtio_state);
692 if (v3_attach_device(vm, dev) == -1) {
693 PrintError("Could not attach device %s\n", "LNX_VIRTIO_BLK");
703 device_register("LNX_VIRTIO_BLK", virtio_init)