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 #define BLK_CAPACITY_PORT 20
31 #define BLK_MAX_SIZE_PORT 28
32 #define BLK_MAX_SEG_PORT 32
33 #define BLK_CYLINDERS_PORT 36
34 #define BLK_HEADS_PORT 38
35 #define BLK_SECTS_PORT 39
39 #define BLK_SCSI_CMD 2
41 #define BLK_BARRIER_FLAG 0x80000000
43 #define BLK_STATUS_OK 0
44 #define BLK_STATUS_ERR 1
45 #define BLK_STATUS_NOT_SUPPORTED 2
55 } __attribute__((packed));
63 } __attribute__((packed));
65 #define QUEUE_SIZE 128
67 /* Host Feature flags */
68 #define VIRTIO_BARRIER 0x01 /* Does host support barriers? */
69 #define VIRTIO_SIZE_MAX 0x02 /* Indicates maximum segment size */
70 #define VIRTIO_SEG_MAX 0x04 /* Indicates maximum # of segments */
71 #define VIRTIO_LEGACY_GEOM 0x10 /* Indicates support of legacy geometry */
76 struct virtio_blk_state {
77 struct blk_config block_cfg;
78 struct virtio_config virtio_cfg;
80 struct vm_device * pci_bus;
81 struct pci_device * pci_dev;
83 struct virtio_queue queue;
86 struct v3_cd_ops * cd_ops;
87 struct v3_hd_ops * hd_ops;
90 v3_block_type_t block_type;
97 static int virtio_free(struct vm_device * dev) {
101 static int virtio_reset(struct vm_device * dev) {
102 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
104 virtio->queue.ring_desc_addr = 0;
105 virtio->queue.ring_avail_addr = 0;
106 virtio->queue.ring_used_addr = 0;
107 virtio->queue.pfn = 0;
108 virtio->queue.cur_avail_idx = 0;
110 virtio->virtio_cfg.status = 0;
111 virtio->virtio_cfg.pci_isr = 0;
116 static int handle_read_op(struct vm_device * dev, uint8_t * buf, uint64_t sector, uint32_t len) {
117 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
119 if (virtio->block_type == BLOCK_DISK) {
120 if (len % HD_SECTOR_SIZE) {
121 PrintError("Write of something that is not a sector len %d, mod=%d\n", len, len % HD_SECTOR_SIZE);
126 PrintDebug("Reading Disk\n");
128 return virtio->hd_ops->read(buf, len / HD_SECTOR_SIZE, sector / HD_SECTOR_SIZE, virtio->backend_data);
129 } else if (virtio->block_type == BLOCK_CDROM) {
130 if (len % ATAPI_BLOCK_SIZE) {
131 PrintError("Write of something that is not an ATAPI block len %d, mod=%d\n", len, len % ATAPI_BLOCK_SIZE);
135 return virtio->cd_ops->read(buf, len / ATAPI_BLOCK_SIZE, sector / ATAPI_BLOCK_SIZE, virtio->backend_data);
143 static int handle_write_op(struct vm_device * dev, uint8_t * buf, uint64_t sector, uint32_t len) {
144 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
147 if (virtio->block_type == BLOCK_DISK) {
148 if (len % HD_SECTOR_SIZE) {
149 PrintError("Write of something that is not a sector len %d, mod=%d\n", len, len % HD_SECTOR_SIZE);
153 PrintDebug("Writing Disk\n");
155 return virtio->hd_ops->write(buf, len / HD_SECTOR_SIZE, sector / HD_SECTOR_SIZE, virtio->backend_data);
162 static int handle_block_op(struct vm_device * dev, struct blk_op_hdr * hdr,
163 struct vring_desc * buf_desc, uint8_t * status) {
164 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
165 uint8_t * buf = NULL;
167 PrintDebug("Handling Block op\n");
171 if (guest_pa_to_host_va(dev->vm, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
172 PrintError("Could not translate buffer address\n");
177 PrintDebug("Sector=%p Length=%d\n", (void *)(addr_t)(hdr->sector), buf_desc->length);
179 if (hdr->type == BLK_IN_REQ) {
180 if (virtio->block_type != BLOCK_NONE) {
181 if (handle_read_op(dev, buf, hdr->sector, buf_desc->length) == -1) {
182 *status = BLK_STATUS_ERR;
184 *status = BLK_STATUS_OK;
187 *status = BLK_STATUS_NOT_SUPPORTED;
190 } else if (hdr->type == BLK_OUT_REQ) {
191 if (virtio->block_type == BLOCK_DISK) {
192 if (handle_write_op(dev, buf, hdr->sector, buf_desc->length) == -1) {
193 *status = BLK_STATUS_ERR;
195 *status = BLK_STATUS_OK;
198 *status = BLK_STATUS_NOT_SUPPORTED;
200 } else if (hdr->type == BLK_SCSI_CMD) {
201 PrintDebug("VIRTIO: SCSI Command Not supported!!!\n");
202 *status = BLK_STATUS_NOT_SUPPORTED;
205 PrintDebug("Returning Status: %d\n", *status);
210 static int get_desc_count(struct virtio_queue * q, int index) {
211 struct vring_desc * tmp_desc = &(q->desc[index]);
214 while (tmp_desc->flags & VIRTIO_NEXT_FLAG) {
215 tmp_desc = &(q->desc[tmp_desc->next]);
224 static int handle_kick(struct vm_device * dev) {
225 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
226 struct virtio_queue * q = &(virtio->queue);
228 PrintDebug("VIRTIO KICK: cur_index=%d (mod=%d), avail_index=%d\n",
229 q->cur_avail_idx, q->cur_avail_idx % QUEUE_SIZE, q->avail->index);
231 while (q->cur_avail_idx < q->avail->index) {
232 struct vring_desc * hdr_desc = NULL;
233 struct vring_desc * buf_desc = NULL;
234 struct vring_desc * status_desc = NULL;
235 struct blk_op_hdr * hdr = NULL;
236 uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
237 int desc_cnt = get_desc_count(q, desc_idx);
239 uint8_t * status_ptr = NULL;
240 uint8_t status = BLK_STATUS_OK;
241 uint32_t req_len = 0;
243 PrintDebug("Descriptor Count=%d, index=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE);
246 PrintError("Block operations must include at least 3 descriptors\n");
250 hdr_desc = &(q->desc[desc_idx]);
253 PrintDebug("Header Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", hdr_desc,
254 (void *)(hdr_desc->addr_gpa), hdr_desc->length, hdr_desc->flags, hdr_desc->next);
256 if (guest_pa_to_host_va(dev->vm, hdr_desc->addr_gpa, (addr_t *)&(hdr)) == -1) {
257 PrintError("Could not translate block header address\n");
261 desc_idx = hdr_desc->next;
263 for (i = 0; i < desc_cnt - 2; i++) {
264 uint8_t tmp_status = BLK_STATUS_OK;
266 buf_desc = &(q->desc[desc_idx]);
268 PrintDebug("Buffer Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", buf_desc,
269 (void *)(buf_desc->addr_gpa), buf_desc->length, buf_desc->flags, buf_desc->next);
271 if (handle_block_op(dev, hdr, buf_desc, &tmp_status) == -1) {
272 PrintError("Error handling block operation\n");
276 if (tmp_status != BLK_STATUS_OK) {
280 req_len += buf_desc->length;
281 desc_idx = buf_desc->next;
284 status_desc = &(q->desc[desc_idx]);
286 PrintDebug("Status Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", status_desc,
287 (void *)(status_desc->addr_gpa), status_desc->length, status_desc->flags, status_desc->next);
289 if (guest_pa_to_host_va(dev->vm, status_desc->addr_gpa, (addr_t *)&(status_ptr)) == -1) {
290 PrintError("Could not translate status address\n");
294 req_len += status_desc->length;
295 *status_ptr = status;
297 q->used->ring[q->used->index].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
298 q->used->ring[q->used->index].length = req_len; // What do we set this to????
304 if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
305 PrintDebug("Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line);
306 v3_pci_raise_irq(virtio->pci_bus, 0, virtio->pci_dev);
307 virtio->virtio_cfg.pci_isr = 1;
313 static int virtio_io_write(uint16_t port, void * src, uint_t length, struct vm_device * dev) {
314 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
315 int port_idx = port % virtio->io_range_size;
318 PrintDebug("VIRTIO BLOCK Write for port %d (index=%d) len=%d, value=%x\n",
319 port, port_idx, length, *(uint32_t *)src);
324 case GUEST_FEATURES_PORT:
326 PrintError("Illegal write length for guest features\n");
330 virtio->virtio_cfg.guest_features = *(uint32_t *)src;
331 PrintDebug("Setting Guest Features to %x\n", virtio->virtio_cfg.guest_features);
334 case VRING_PG_NUM_PORT:
336 addr_t pfn = *(uint32_t *)src;
337 addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
340 virtio->queue.pfn = pfn;
342 virtio->queue.ring_desc_addr = page_addr ;
343 virtio->queue.ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
344 virtio->queue.ring_used_addr = ( virtio->queue.ring_avail_addr + \
345 sizeof(struct vring_avail) + \
346 (QUEUE_SIZE * sizeof(uint16_t)));
348 // round up to next page boundary.
349 virtio->queue.ring_used_addr = (virtio->queue.ring_used_addr + 0xfff) & ~0xfff;
351 if (guest_pa_to_host_va(dev->vm, virtio->queue.ring_desc_addr, (addr_t *)&(virtio->queue.desc)) == -1) {
352 PrintError("Could not translate ring descriptor address\n");
357 if (guest_pa_to_host_va(dev->vm, virtio->queue.ring_avail_addr, (addr_t *)&(virtio->queue.avail)) == -1) {
358 PrintError("Could not translate ring available address\n");
363 if (guest_pa_to_host_va(dev->vm, virtio->queue.ring_used_addr, (addr_t *)&(virtio->queue.used)) == -1) {
364 PrintError("Could not translate ring used address\n");
368 PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
369 (void *)(virtio->queue.ring_desc_addr),
370 (void *)(virtio->queue.ring_avail_addr),
371 (void *)(virtio->queue.ring_used_addr));
373 PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
374 virtio->queue.desc, virtio->queue.avail, virtio->queue.used);
377 PrintError("Illegal write length for page frame number\n");
381 case VRING_Q_SEL_PORT:
382 virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
384 if (virtio->virtio_cfg.vring_queue_selector != 0) {
385 PrintError("Virtio Block device only uses 1 queue, selected %d\n",
386 virtio->virtio_cfg.vring_queue_selector);
391 case VRING_Q_NOTIFY_PORT:
392 PrintDebug("Handling Kick\n");
393 if (handle_kick(dev) == -1) {
394 PrintError("Could not handle Block Notification\n");
398 case VIRTIO_STATUS_PORT:
399 virtio->virtio_cfg.status = *(uint8_t *)src;
401 if (virtio->virtio_cfg.status == 0) {
402 PrintDebug("Resetting device\n");
408 case VIRTIO_ISR_PORT:
409 virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
420 static int virtio_io_read(uint16_t port, void * dst, uint_t length, struct vm_device * dev) {
421 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
422 int port_idx = port % virtio->io_range_size;
425 PrintDebug("VIRTIO BLOCK Read for port %d (index =%d), length=%d\n",
426 port, port_idx, length);
429 case HOST_FEATURES_PORT:
431 PrintError("Illegal read length for host features\n");
435 *(uint32_t *)dst = virtio->virtio_cfg.host_features;
438 case VRING_PG_NUM_PORT:
440 PrintError("Illegal read length for page frame number\n");
444 *(uint32_t *)dst = virtio->queue.pfn;
447 case VRING_SIZE_PORT:
449 PrintError("Illegal read length for vring size\n");
453 *(uint16_t *)dst = virtio->queue.queue_size;
457 case VIRTIO_STATUS_PORT:
459 PrintError("Illegal read length for status\n");
463 *(uint8_t *)dst = virtio->virtio_cfg.status;
466 case VIRTIO_ISR_PORT:
467 *(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
468 virtio->virtio_cfg.pci_isr = 0;
469 v3_pci_lower_irq(virtio->pci_bus, 0, virtio->pci_dev);
473 if ( (port_idx >= sizeof(struct virtio_config)) &&
474 (port_idx < (sizeof(struct virtio_config) + sizeof(struct blk_config))) ) {
475 int cfg_offset = port_idx - sizeof(struct virtio_config);
476 uint8_t * cfg_ptr = (uint8_t *)&(virtio->block_cfg);
478 memcpy(dst, cfg_ptr + cfg_offset, length);
481 PrintError("Read of Unhandled Virtio Read\n");
494 static struct v3_device_ops dev_ops = {
504 int v3_virtio_register_cdrom(struct vm_device * dev, struct v3_cd_ops * ops, void * private_data) {
505 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
507 virtio->block_type = BLOCK_CDROM;
508 virtio->cd_ops = ops;
509 virtio->backend_data = private_data;
511 virtio->block_cfg.capacity = ops->get_capacity(private_data);
517 int v3_virtio_register_harddisk(struct vm_device * dev, struct v3_hd_ops * ops, void * private_data) {
518 struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
520 virtio->block_type = BLOCK_DISK;
521 virtio->hd_ops = ops;
522 virtio->backend_data = private_data;
524 virtio->block_cfg.capacity = ops->get_capacity(private_data);
526 PrintDebug("Virtio Capacity = %d -- 0x%p\n", (int)(virtio->block_cfg.capacity),
527 (void *)(addr_t)(virtio->block_cfg.capacity));
534 static int virtio_init(struct guest_info * vm, void * cfg_data) {
535 struct vm_device * pci_bus = v3_find_dev(vm, (char *)cfg_data);
536 struct virtio_blk_state * virtio_state = NULL;
537 struct pci_device * pci_dev = NULL;
539 PrintDebug("Initializing VIRTIO Block device\n");
541 if (pci_bus == NULL) {
542 PrintError("VirtIO devices require a PCI Bus");
547 virtio_state = (struct virtio_blk_state *)V3_Malloc(sizeof(struct virtio_blk_state));
548 memset(virtio_state, 0, sizeof(struct virtio_blk_state));
551 struct vm_device * dev = v3_allocate_device("LNX_VIRTIO_BLK", &dev_ops, virtio_state);
552 if (v3_attach_device(vm, dev) == -1) {
553 PrintError("Could not attach device %s\n", "LNX_VIRTIO_BLK");
558 // PCI initialization
560 struct v3_pci_bar bars[6];
561 int num_ports = sizeof(struct virtio_config) + sizeof(struct blk_config);
562 int tmp_ports = num_ports;
567 // This gets the number of ports, rounded up to a power of 2
568 virtio_state->io_range_size = 1; // must be a power of 2
570 while (tmp_ports > 0) {
572 virtio_state->io_range_size <<= 1;
575 // this is to account for any low order bits being set in num_ports
576 // if there are none, then num_ports was already a power of 2 so we shift right to reset it
577 if ((num_ports & ((virtio_state->io_range_size >> 1) - 1)) == 0) {
578 virtio_state->io_range_size >>= 1;
582 for (i = 0; i < 6; i++) {
583 bars[i].type = PCI_BAR_NONE;
586 bars[0].type = PCI_BAR_IO;
587 bars[0].default_base_port = -1;
588 bars[0].num_ports = virtio_state->io_range_size;
590 bars[0].io_read = virtio_io_read;
591 bars[0].io_write = virtio_io_write;
593 pci_dev = v3_pci_register_device(pci_bus, PCI_STD_DEVICE,
594 0, PCI_AUTO_DEV_NUM, 0,
595 "LNX_VIRTIO_BLK", bars,
596 NULL, NULL, NULL, dev);
599 PrintError("Could not register PCI Device\n");
603 pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
604 pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
607 pci_dev->config_header.device_id = VIRTIO_BLOCK_DEV_ID;
608 pci_dev->config_header.class = PCI_CLASS_STORAGE;
609 pci_dev->config_header.subclass = PCI_STORAGE_SUBCLASS_OTHER;
611 pci_dev->config_header.subsystem_id = VIRTIO_BLOCK_SUBDEVICE_ID;
614 pci_dev->config_header.intr_pin = 1;
616 pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
619 virtio_state->pci_dev = pci_dev;
620 virtio_state->pci_bus = pci_bus;
623 /* Block configuration */
624 virtio_state->virtio_cfg.host_features = VIRTIO_SEG_MAX;
625 virtio_state->block_cfg.max_seg = QUEUE_SIZE - 2;
627 // Virtio Block only uses one queue
628 virtio_state->queue.queue_size = QUEUE_SIZE;
634 virtio_state->backend_data = NULL;
635 virtio_state->block_type = BLOCK_NONE;
636 virtio_state->hd_ops = NULL;
642 device_register("LNX_VIRTIO_BLK", virtio_init)