Palacios Public Git Repository

To checkout Palacios execute

  git clone http://v3vee.org/palacios/palacios.web/palacios.git
This will give you the master branch. You probably want the devel branch or one of the release branches. To switch to the devel branch, simply execute
  cd palacios
  git checkout --track -b devel origin/devel
The other branches are similar.


updated the block device interfaces to be more generic, and have the virtio block...
[palacios.git] / palacios / src / devices / lnx_virtio_blk.c
index cdbad38..07870b2 100644 (file)
 #include <palacios/vmm.h>
 #include <palacios/vmm_dev_mgr.h>
 #include <devices/lnx_virtio_pci.h>
-
+#include <devices/lnx_virtio_blk.h>
+#include <devices/block_dev.h>
+#include <palacios/vm_guest_mem.h>
 
 #include <devices/pci.h>
 
 
+#define BLK_CAPACITY_PORT     20
+#define BLK_MAX_SIZE_PORT     28
+#define BLK_MAX_SEG_PORT      32
+#define BLK_CYLINDERS_PORT    36
+#define BLK_HEADS_PORT        38
+#define BLK_SECTS_PORT        39
+
+#define BLK_IN_REQ            0
+#define BLK_OUT_REQ           1
+#define BLK_SCSI_CMD          2
+
+#define BLK_BARRIER_FLAG     0x80000000
+
+#define BLK_STATUS_OK             0
+#define BLK_STATUS_ERR            1
+#define BLK_STATUS_NOT_SUPPORTED  2
+
+
 struct blk_config {
     uint64_t capacity;
     uint32_t max_size;
@@ -36,6 +56,13 @@ struct blk_config {
 
 
 
+struct blk_op_hdr {
+    uint32_t type;
+    uint32_t prior;
+    uint64_t sector;
+} __attribute__((packed));
+
+#define QUEUE_SIZE 128
 
 /* Host Feature flags */
 #define VIRTIO_BARRIER       0x01       /* Does host support barriers? */
@@ -45,22 +72,233 @@ struct blk_config {
 
 
 
+
 struct virtio_blk_state {
     struct blk_config block_cfg;
     struct virtio_config virtio_cfg;
 
     struct vm_device * pci_bus;
     struct pci_device * pci_dev;
-
-    struct virtio_device * virtio_dev; // the virtio device struction for _this_ device
     
+    struct virtio_queue queue;
+
+    union {
+       struct v3_cd_ops * cd_ops;
+       struct v3_hd_ops * hd_ops;
+    };
+
+    v3_block_type_t block_type;
+    void * backend_data;
 
     int io_range_size;
 };
 
 
+static int virtio_free(struct vm_device * dev) {
+    return -1;
+}
+
+static int virtio_reset(struct vm_device * dev) {
+    struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
+
+    virtio->queue.ring_desc_addr = 0;
+    virtio->queue.ring_avail_addr = 0;
+    virtio->queue.ring_used_addr = 0;
+    virtio->queue.pfn = 0;
+    virtio->queue.cur_avail_idx = 0;
+
+    virtio->virtio_cfg.status = 0;
+    virtio->virtio_cfg.pci_isr = 0;
+
+    return 0;
+}
+
+static int handle_read_op(struct vm_device * dev, uint8_t * buf, uint64_t sector, uint32_t len) {
+    struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data; 
+
+    if (virtio->block_type == BLOCK_DISK) {
+       if (len % HD_SECTOR_SIZE) {
+           PrintError("Write of something that is not a sector len %d, mod=%d\n", len, len % HD_SECTOR_SIZE);
+           return -1;
+       }
+
+
+       PrintDebug("Reading Disk\n");
+
+       return virtio->hd_ops->read(buf, len / HD_SECTOR_SIZE, sector * HD_SECTOR_SIZE, virtio->backend_data);
+    } else if (virtio->block_type == BLOCK_CDROM) {
+       if (len % ATAPI_BLOCK_SIZE) {
+           PrintError("Write of something that is not an ATAPI block len %d, mod=%d\n", len, len % ATAPI_BLOCK_SIZE);
+           return -1;
+       }
+
+       return virtio->cd_ops->read(buf, len / ATAPI_BLOCK_SIZE, sector * ATAPI_BLOCK_SIZE, virtio->backend_data);
+
+    }
+
+    return -1;
+}
+
+
+static int handle_write_op(struct vm_device * dev, uint8_t * buf, uint64_t sector, uint32_t len) {
+    struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data; 
+
+
+    if (virtio->block_type == BLOCK_DISK) {
+       if (len % HD_SECTOR_SIZE) {
+           PrintError("Write of something that is not a sector len %d, mod=%d\n", len, len % HD_SECTOR_SIZE);
+           return -1;
+       }
+
+       PrintDebug("Writing Disk\n");
+
+       return virtio->hd_ops->write(buf, len / HD_SECTOR_SIZE, sector * HD_SECTOR_SIZE, virtio->backend_data);
+    }
+
+    return -1;
+}
+
+
+static int handle_block_op(struct vm_device * dev, struct vring_desc * hdr_desc, 
+                          struct vring_desc * buf_desc, struct vring_desc * status_desc) {
+    struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;    
+    struct blk_op_hdr * hdr = NULL;
+    uint8_t * buf = NULL;
+    uint8_t * status = NULL;
+
+
+    PrintDebug("Handling Block op\n");
+
+    if (guest_pa_to_host_va(dev->vm, hdr_desc->addr_gpa, (addr_t *)&(hdr)) == -1) {
+       PrintError("Could not translate block header address\n");
+       return -1;
+    }
+
+
+    if (guest_pa_to_host_va(dev->vm, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
+       PrintError("Could not translate buffer address\n");
+       return -1;
+    }
+
+    if (guest_pa_to_host_va(dev->vm, status_desc->addr_gpa, (addr_t *)&(status)) == -1) {
+       PrintError("Could not translate status address\n");
+       return -1;
+    }
+
+    if (hdr->type == BLK_IN_REQ) {
+       if (virtio->block_type != BLOCK_NONE) {
+           if (handle_read_op(dev, buf, hdr->sector, buf_desc->length) == -1) {
+               *status = BLK_STATUS_ERR;
+           } else {
+               *status = BLK_STATUS_OK;
+           }
+       } else {
+           *status = BLK_STATUS_NOT_SUPPORTED;
+       } 
+
+    } else if (hdr->type == BLK_OUT_REQ) {
+       if (virtio->block_type == BLOCK_DISK) {
+           if (handle_write_op(dev, buf, hdr->sector, buf_desc->length) == -1) {
+               *status = BLK_STATUS_ERR;
+           } else {
+               *status = BLK_STATUS_OK;
+           }
+       } else {
+           *status = BLK_STATUS_NOT_SUPPORTED;
+       }
+    } else if (hdr->type == BLK_SCSI_CMD) {
+       *status = BLK_STATUS_NOT_SUPPORTED;
+    }
+
+
+
+
+    return 0;
+}
+
+
+
+static int handle_kick(struct vm_device * dev) {
+    struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;    
+    struct virtio_queue * q = &(virtio->queue);
+
+
+    PrintDebug("VIRTIO KICK: cur_index=%d, avail_index=%d\n", q->cur_avail_idx, q->avail->index);
+
+    while (q->cur_avail_idx < q->avail->index) {
+       struct vring_desc * hdr_desc = NULL;
+       struct vring_desc * buf_desc = NULL;
+       struct vring_desc * status_desc = NULL;
+       uint16_t chain_idx = q->avail->ring[q->cur_avail_idx];
+       uint32_t req_len = 0;
+       int chained = 1;
+
+       PrintDebug("chained=%d, Chain Index=%d\n", chained, chain_idx);
+
+       while (chained) {
+           hdr_desc = &(q->desc[chain_idx]);
+           
+           PrintDebug("Header Descriptor gpa=%p, len=%d, flags=%x, next=%d\n", 
+                      (void *)(hdr_desc->addr_gpa), hdr_desc->length, hdr_desc->flags, hdr_desc->next);
+
+       
+           if (!(hdr_desc->flags & VIRTIO_NEXT_FLAG)) {
+               PrintError("Block operations must chain a buffer descriptor\n");
+               return -1;
+           }
+
+           buf_desc = &(q->desc[hdr_desc->next]);
+           
+
+           PrintDebug("Buffer  Descriptor gpa=%p, len=%d, flags=%x, next=%d\n", 
+                      (void *)(buf_desc->addr_gpa), buf_desc->length, buf_desc->flags, buf_desc->next);
+           
+
+           if (!(buf_desc->flags & VIRTIO_NEXT_FLAG)) {
+               PrintError("Block operatoins must chain a status descriptor\n");
+               return -1;
+           }
+           
+           status_desc = &(q->desc[buf_desc->next]);
+
+           // We detect whether we are chained here...
+           if (status_desc->flags & VIRTIO_NEXT_FLAG) {
+               chained = 1;
+               chain_idx = status_desc->next; 
+           } else {
+               chained = 0;
+           }
+
+           PrintDebug("Status  Descriptor gpa=%p, len=%d, flags=%x, next=%d\n", 
+                      (void *)(status_desc->addr_gpa), status_desc->length, status_desc->flags, status_desc->next);
+           
+
+           if (handle_block_op(dev, hdr_desc, buf_desc, status_desc) == -1) {
+               PrintError("Error handling block operation\n");
+               return -1;
+           }
+
+           req_len += (buf_desc->length + status_desc->length);
+
+       }
+
+       q->used->ring[q->used->index].id = q->cur_avail_idx;
+       q->used->ring[q->used->index].length = req_len; // What do we set this to????
+
+       q->used->index = (q->used->index + 1) % (QUEUE_SIZE * sizeof(struct vring_desc));;
 
 
+       q->cur_avail_idx = (q->cur_avail_idx + 1) % (QUEUE_SIZE * sizeof(struct vring_desc));
+    }
+
+    if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
+       PrintDebug("Raising IRQ %d\n",  virtio->pci_dev->config_header.intr_line);
+       v3_pci_raise_irq(virtio->pci_bus, 0, virtio->pci_dev);
+    }
+
+    return 0;
+}
+
 static int virtio_io_write(uint16_t port, void * src, uint_t length, struct vm_device * dev) {
     struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
     int port_idx = port % virtio->io_range_size;
@@ -72,17 +310,84 @@ static int virtio_io_write(uint16_t port, void * src, uint_t length, struct vm_d
 
 
     switch (port_idx) {
+       case GUEST_FEATURES_PORT:
+           if (length != 4) {
+               PrintError("Illegal write length for guest features\n");
+               return -1;
+           }
+           
+           virtio->virtio_cfg.guest_features = *(uint32_t *)src;
+
+           break;
+       case VRING_PG_NUM_PORT:
+           if (length == 4) {
+               addr_t pfn = *(uint32_t *)src;
+               addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
+
+
+               virtio->queue.pfn = pfn;
+               
+               virtio->queue.ring_desc_addr = page_addr ;
+               virtio->queue.ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
+               virtio->queue.ring_used_addr = ( virtio->queue.ring_avail_addr + \
+                                                sizeof(struct vring_avail)    + \
+                                                (QUEUE_SIZE * sizeof(uint16_t)));
+               
+               // round up to next page boundary.
+               virtio->queue.ring_used_addr = (virtio->queue.ring_used_addr + 0xfff) & ~0xfff;
+
+               if (guest_pa_to_host_va(dev->vm, virtio->queue.ring_desc_addr, (addr_t *)&(virtio->queue.desc)) == -1) {
+                   PrintError("Could not translate ring descriptor address\n");
+                   return -1;
+               }
+
+
+               if (guest_pa_to_host_va(dev->vm, virtio->queue.ring_avail_addr, (addr_t *)&(virtio->queue.avail)) == -1) {
+                   PrintError("Could not translate ring available address\n");
+                   return -1;
+               }
+
+
+               if (guest_pa_to_host_va(dev->vm, virtio->queue.ring_used_addr, (addr_t *)&(virtio->queue.used)) == -1) {
+                   PrintError("Could not translate ring used address\n");
+                   return -1;
+               }
+
+               PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
+                          (void *)(virtio->queue.ring_desc_addr),
+                          (void *)(virtio->queue.ring_avail_addr),
+                          (void *)(virtio->queue.ring_used_addr));
+
+           } else {
+               PrintError("Illegal write length for page frame number\n");
+               return -1;
+           }
+           break;
+       case VRING_Q_SEL_PORT:
+           virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
+
+           if (virtio->virtio_cfg.vring_queue_selector != 0) {
+               PrintError("Virtio Block device only uses 1 queue, selected %d\n", 
+                          virtio->virtio_cfg.vring_queue_selector);
+               return -1;
+           }
+
+           break;
        case VRING_Q_NOTIFY_PORT:
-           // handle output
-           PrintError("Notification\n");
-           return -1;
+           PrintDebug("Handling Kick\n");
+           if (handle_kick(dev) == -1) {
+               PrintError("Could not handle Block Notification\n");
+               return -1;
+           }
            break;
        case VIRTIO_STATUS_PORT:
+           virtio->virtio_cfg.status = *(uint8_t *)src;
+
            if (virtio->virtio_cfg.status == 0) {
                PrintDebug("Resetting device\n");
-               return -1;
-               //reset
+               virtio_reset(dev);
            }
+
            break;
        default:
            return -1;
@@ -105,19 +410,62 @@ static int virtio_io_read(uint16_t port, void * dst, uint_t length, struct vm_de
               port, port_idx, length);
 
     switch (port_idx) {
-       // search for device....
-       // call and return dev config read
+       case HOST_FEATURES_PORT:
+           if (length != 4) {
+               PrintError("Illegal read length for host features\n");
+               return -1;
+           }
+
+           *(uint32_t *)dst = virtio->virtio_cfg.host_features;
+       
+           break;
+       case VRING_PG_NUM_PORT:
+           if (length != 4) {
+               PrintError("Illegal read length for page frame number\n");
+               return -1;
+           }
+
+           *(uint32_t *)dst = virtio->queue.pfn;
+
+           break;
+       case VRING_SIZE_PORT:
+           if (length != 2) {
+               PrintError("Illegal read length for vring size\n");
+               return -1;
+           }
+               
+           *(uint16_t *)dst = virtio->queue.queue_size;
+
+           break;
+
+       case VIRTIO_STATUS_PORT:
+           if (length != 1) {
+               PrintError("Illegal read length for status\n");
+               return -1;
+           }
+
+           *(uint8_t *)dst = virtio->virtio_cfg.status;
+           break;
+
        default:
-       return -1;
+           if ( (port_idx >= sizeof(struct virtio_config)) && 
+                (port_idx < (sizeof(struct virtio_config) + sizeof(struct blk_config))) ) {
+
+               uint8_t * cfg_ptr = (uint8_t *)&(virtio->block_cfg);
+               memcpy(dst, cfg_ptr, length);
+               
+           } else {
+               PrintError("Read of Unhandled Virtio Read\n");
+               return -1;
+           }
+         
+           break;
     }
 
 
-    return length;
-}
 
 
-static int virtio_free(struct vm_device * dev) {
-    return -1;
+    return length;
 }
 
 
@@ -133,6 +481,31 @@ static struct v3_device_ops dev_ops = {
 
 
 
+int v3_virtio_register_cdrom(struct vm_device * dev, struct v3_cd_ops * ops, void * private_data) {
+    struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
+    
+    virtio->block_type = BLOCK_CDROM;
+    virtio->cd_ops = ops;
+    virtio->backend_data = private_data;
+
+    virtio->block_cfg.capacity = ops->get_capacity(private_data);
+
+    return 0;
+}
+
+
+int v3_virtio_register_harddisk(struct vm_device * dev, struct v3_hd_ops * ops, void * private_data) {
+    struct virtio_blk_state * virtio = (struct virtio_blk_state *)dev->private_data;
+
+    virtio->block_type = BLOCK_DISK;
+    virtio->hd_ops = ops;
+    virtio->backend_data = private_data;
+
+    virtio->block_cfg.capacity = ops->get_capacity(private_data);
+
+    return 0;
+}
+
 
 
 static int virtio_init(struct guest_info * vm, void * cfg_data) {
@@ -222,10 +595,20 @@ static int virtio_init(struct guest_info * vm, void * cfg_data) {
 
        virtio_state->pci_dev = pci_dev;
        virtio_state->pci_bus = pci_bus;
-
-       /* Block configuration */
     }
-    
+
+    /* Block configuration */
+    virtio_state->virtio_cfg.host_features = VIRTIO_SEG_MAX;
+
+    // Virtio Block only uses one queue
+    virtio_state->queue.queue_size = QUEUE_SIZE;
+
+    virtio_reset(dev);
+
+    virtio_state->backend_data = NULL;
+    virtio_state->block_type = BLOCK_NONE;
+    virtio_state->hd_ops = NULL;
+
     return 0;
 }