[Qemu-devel] [PATCH 1/1] Virtio block device support

[Hollis Blanchard]

From: Anthony Liguori <address@hidden>

This has been posted before but I believe it now has addressed all outstanding
concerns.  I'd like to apply it if there are no objections.

This patch adds virtio-blk support to QEMU.  virtio-blk is a paravirtual disk
controller that can achieve good performance when using KVM.

Since virtio is based on a scatter/gather API, we don't have a linear buffer
for each request.  This forces us to allocate a linear buffer since the current
block driver API does not have a scatter/gather operation.  This allocation
can never exceed the maximum data limit on the ring queue so it isn't
unbounded.

posix-aio cannot support a scatter/gather asynchronous operation so we'll need
to introduce our own thread pool to eliminate this limitation.  There is work
underway to do this.

Signed-off-by: Anthony Liguori <address@hidden>
---
 Makefile.target |    2 +-
 hw/pc.c         |   12 ++
 hw/pc.h         |    4 +
 hw/virtio-blk.c |  303 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 sysemu.h        |    2 +-
 vl.c            |    4 +
 6 files changed, 325 insertions(+), 2 deletions(-)
 create mode 100644 hw/virtio-blk.c

diff --git a/Makefile.target b/Makefile.target
index f7fc3a2..773b615 100644
--- a/Makefile.target
+++ b/Makefile.target
@@ -588,7 +588,7 @@ endif #CONFIG_BSD_USER
 ifndef CONFIG_USER_ONLY
 
 OBJS=vl.o osdep.o monitor.o pci.o loader.o isa_mmio.o machine.o
-OBJS+=fw_cfg.o virtio.o
+OBJS+=fw_cfg.o virtio.o virtio-blk.o
 ifdef CONFIG_KVM
 OBJS+=kvm.o kvm-all.o
 endif
diff --git a/hw/pc.c b/hw/pc.c
index 1486b68..8651e49 100644
--- a/hw/pc.c
+++ b/hw/pc.c
@@ -1092,6 +1092,18 @@ static void pc_init1(ram_addr_t ram_size, int 
vga_ram_size,
            }
         }
     }
+
+    /* Add virtio block devices */
+    if (pci_enabled) {
+        int index;
+        int unit_id = 0;
+
+        while ((index = drive_get_index(IF_VIRTIO, 0, unit_id)) != -1) {
+            virtio_blk_init(pci_bus, 0x1AF4, 0x1001,
+                            drives_table[index].bdrv);
+            unit_id++;
+        }
+    }
 }
 
 static void pc_init_pci(ram_addr_t ram_size, int vga_ram_size,
diff --git a/hw/pc.h b/hw/pc.h
index f156b9e..edbc5e8 100644
--- a/hw/pc.h
+++ b/hw/pc.h
@@ -152,4 +152,8 @@ void pci_piix4_ide_init(PCIBus *bus, BlockDriverState 
**hd_table, int devfn,
 
 void isa_ne2000_init(int base, qemu_irq irq, NICInfo *nd);
 
+/* virtio-blk.c */
+void *virtio_blk_init(PCIBus *bus, uint16_t vendor, uint16_t device,
+                      BlockDriverState *bs);
+
 #endif
diff --git a/hw/virtio-blk.c b/hw/virtio-blk.c
new file mode 100644
index 0000000..da87b66
--- /dev/null
+++ b/hw/virtio-blk.c
@@ -0,0 +1,303 @@
+/*
+ * Virtio Block Device
+ *
+ * Copyright IBM, Corp. 2007
+ *
+ * Authors:
+ *  Anthony Liguori   <address@hidden>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2.  See
+ * the COPYING file in the top-level directory.
+ *
+ */
+
+#include "virtio.h"
+#include "block.h"
+#include "block_int.h"
+#include "pc.h"
+
+/* from Linux's linux/virtio_blk.h */
+
+/* The ID for virtio_block */
+#define VIRTIO_ID_BLOCK        2
+
+/* Feature bits */
+#define VIRTIO_BLK_F_BARRIER   0      /* Does host support barriers? */
+#define VIRTIO_BLK_F_SIZE_MAX  1     /* Indicates maximum segment size */
+#define VIRTIO_BLK_F_SEG_MAX   2     /* Indicates maximum # of segments */
+#define VIRTIO_BLK_F_GEOMETRY  4     /* Indicates support of legacy geometry */
+
+struct virtio_blk_config
+{
+    uint64_t capacity;
+    uint32_t size_max;
+    uint32_t seg_max;
+    uint16_t cylinders;
+    uint8_t heads;
+    uint8_t sectors;
+} __attribute__((packed));
+
+/* These two define direction. */
+#define VIRTIO_BLK_T_IN        0
+#define VIRTIO_BLK_T_OUT       1
+
+/* This bit says it's a scsi command, not an actual read or write. */
+#define VIRTIO_BLK_T_SCSI_CMD  2
+
+/* Barrier before this op. */
+#define VIRTIO_BLK_T_BARRIER   0x80000000
+
+/* This is the first element of the read scatter-gather list. */
+struct virtio_blk_outhdr
+{
+    /* VIRTIO_BLK_T* */
+    uint32_t type;
+    /* io priority. */
+    uint32_t ioprio;
+    /* Sector (ie. 512 byte offset) */
+    uint64_t sector;
+};
+
+#define VIRTIO_BLK_S_OK        0
+#define VIRTIO_BLK_S_IOERR     1
+#define VIRTIO_BLK_S_UNSUPP    2
+
+/* This is the first element of the write scatter-gather list */
+struct virtio_blk_inhdr
+{
+    unsigned char status;
+};
+
+typedef struct VirtIOBlock
+{
+    VirtIODevice vdev;
+    BlockDriverState *bs;
+    VirtQueue *vq;
+} VirtIOBlock;
+
+static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
+{
+    return (VirtIOBlock *)vdev;
+}
+
+typedef struct VirtIOBlockReq
+{
+    VirtIOBlock *dev;
+    VirtQueueElement elem;
+    struct virtio_blk_inhdr *in;
+    struct virtio_blk_outhdr *out;
+    size_t size;
+    uint8_t *buffer;
+} VirtIOBlockReq;
+
+static void virtio_blk_rw_complete(void *opaque, int ret)
+{
+    VirtIOBlockReq *req = opaque;
+    VirtIOBlock *s = req->dev;
+
+    /* Copy read data to the guest */
+    if (!ret && !(req->out->type & VIRTIO_BLK_T_OUT)) {
+        size_t offset = 0;
+        int i;
+
+        for (i = 0; i < req->elem.in_num - 1; i++) {
+            size_t len;
+
+            /* Be pretty defensive wrt malicious guests */
+            len = MIN(req->elem.in_sg[i].iov_len,
+                      req->size - offset);
+
+            memcpy(req->elem.in_sg[i].iov_base,
+                   req->buffer + offset,
+                   len);
+            offset += len;
+        }
+    }
+
+    req->in->status = ret ? VIRTIO_BLK_S_IOERR : VIRTIO_BLK_S_OK;
+    virtqueue_push(s->vq, &req->elem, req->size + sizeof(*req->in));
+    virtio_notify(&s->vdev, s->vq);
+
+    qemu_free(req->buffer);
+    qemu_free(req);
+}
+
+static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
+{
+    VirtIOBlockReq *req;
+
+    req = qemu_mallocz(sizeof(*req));
+    if (req == NULL)
+        return NULL;
+
+    req->dev = s;
+    if (!virtqueue_pop(s->vq, &req->elem)) {
+        qemu_free(req);
+        return NULL;
+    }
+
+    return req;
+}
+
+static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
+{
+    VirtIOBlock *s = to_virtio_blk(vdev);
+    VirtIOBlockReq *req;
+
+    while ((req = virtio_blk_get_request(s))) {
+        int i;
+
+        if (req->elem.out_num < 1 || req->elem.in_num < 1) {
+            fprintf(stderr, "virtio-blk missing headers\n");
+            exit(1);
+        }
+
+        if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
+            req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
+            fprintf(stderr, "virtio-blk header not in correct element\n");
+            exit(1);
+        }
+
+        req->out = (void *)req->elem.out_sg[0].iov_base;
+        req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
+
+        if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
+            unsigned int len = sizeof(*req->in);
+
+            req->in->status = VIRTIO_BLK_S_UNSUPP;
+            virtqueue_push(vq, &req->elem, len);
+            virtio_notify(vdev, vq);
+            qemu_free(req);
+        } else if (req->out->type & VIRTIO_BLK_T_OUT) {
+            size_t offset;
+
+            for (i = 1; i < req->elem.out_num; i++)
+                req->size += req->elem.out_sg[i].iov_len;
+
+            req->buffer = qemu_memalign(512, req->size);
+            if (req->buffer == NULL) {
+                qemu_free(req);
+                break;
+            }
+
+            /* We copy the data from the SG list to avoid splitting up the 
request.  This helps
+               performance a lot until we can pass full sg lists as AIO 
operations */
+            offset = 0;
+            for (i = 1; i < req->elem.out_num; i++) {
+                size_t len;
+
+                len = MIN(req->elem.out_sg[i].iov_len,
+                          req->size - offset);
+                memcpy(req->buffer + offset,
+                       req->elem.out_sg[i].iov_base,
+                       len);
+                offset += len;
+            }
+
+            bdrv_aio_write(s->bs, req->out->sector,
+                           req->buffer,
+                           req->size / 512,
+                           virtio_blk_rw_complete,
+                           req);
+        } else {
+            for (i = 0; i < req->elem.in_num - 1; i++)
+                req->size += req->elem.in_sg[i].iov_len;
+
+            req->buffer = qemu_memalign(512, req->size);
+            if (req->buffer == NULL) {
+                qemu_free(req);
+                break;
+            }
+
+            bdrv_aio_read(s->bs, req->out->sector,
+                          req->buffer,
+                          req->size / 512,
+                          virtio_blk_rw_complete,
+                          req);
+        }
+    }
+    /*
+     * FIXME: Want to check for completions before returning to guest mode,
+     * so cached reads and writes are reported as quickly as possible. But
+     * that should be done in the generic block layer.
+     */
+}
+
+static void virtio_blk_reset(VirtIODevice *vdev)
+{
+    /*
+     * This should cancel pending requests, but can't do nicely until there
+     * are per-device request lists.
+     */
+    qemu_aio_flush();
+}
+
+static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
+{
+    VirtIOBlock *s = to_virtio_blk(vdev);
+    struct virtio_blk_config blkcfg;
+    uint64_t capacity;
+    int cylinders, heads, secs;
+
+    bdrv_get_geometry(s->bs, &capacity);
+    bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
+    stq_raw(&blkcfg.capacity, capacity);
+    stl_raw(&blkcfg.seg_max, 128 - 2);
+    stw_raw(&blkcfg.cylinders, cylinders);
+    blkcfg.heads = heads;
+    blkcfg.sectors = secs;
+    memcpy(config, &blkcfg, sizeof(blkcfg));
+}
+
+static uint32_t virtio_blk_get_features(VirtIODevice *vdev)
+{
+    return (1 << VIRTIO_BLK_F_SEG_MAX | 1 << VIRTIO_BLK_F_GEOMETRY);
+}
+
+static void virtio_blk_save(QEMUFile *f, void *opaque)
+{
+    VirtIOBlock *s = opaque;
+    virtio_save(&s->vdev, f);
+}
+
+static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
+{
+    VirtIOBlock *s = opaque;
+
+    if (version_id != 1)
+        return -EINVAL;
+
+    virtio_load(&s->vdev, f);
+
+    return 0;
+}
+
+void *virtio_blk_init(PCIBus *bus, uint16_t vendor, uint16_t device,
+                      BlockDriverState *bs)
+{
+    VirtIOBlock *s;
+    int cylinders, heads, secs;
+    static int virtio_blk_id;
+
+    s = (VirtIOBlock *)virtio_init_pci(bus, "virtio-blk", vendor, device,
+                                       0, VIRTIO_ID_BLOCK,
+                                       0x01, 0x80, 0x00,
+                                       sizeof(struct virtio_blk_config), 
sizeof(VirtIOBlock));
+    if (!s)
+        return NULL;
+
+    s->vdev.get_config = virtio_blk_update_config;
+    s->vdev.get_features = virtio_blk_get_features;
+    s->vdev.reset = virtio_blk_reset;
+    s->bs = bs;
+
+    bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
+    bdrv_set_geometry_hint(s->bs, cylinders, heads, secs);
+
+    s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
+
+    register_savevm("virtio-blk", virtio_blk_id++, 1,
+                    virtio_blk_save, virtio_blk_load, s);
+
+    return s;
+}
diff --git a/sysemu.h b/sysemu.h
index f72ec96..94cffaf 100644
--- a/sysemu.h
+++ b/sysemu.h
@@ -123,7 +123,7 @@ extern unsigned int nb_prom_envs;
 #endif
 
 typedef enum {
-    IF_IDE, IF_SCSI, IF_FLOPPY, IF_PFLASH, IF_MTD, IF_SD
+    IF_IDE, IF_SCSI, IF_FLOPPY, IF_PFLASH, IF_MTD, IF_SD, IF_VIRTIO
 } BlockInterfaceType;
 
 typedef struct DriveInfo {
diff --git a/vl.c b/vl.c
index e4edf20..33a6fec 100644
--- a/vl.c
+++ b/vl.c
@@ -2267,6 +2267,9 @@ static int drive_init(struct drive_opt *arg, int snapshot,
        } else if (!strcmp(buf, "sd")) {
            type = IF_SD;
             max_devs = 0;
+        } else if (!strcmp(buf, "virtio")) {
+            type = IF_VIRTIO;
+            max_devs = 0;
        } else {
             fprintf(stderr, "qemu: '%s' unsupported bus type '%s'\n", str, 
buf);
             return -1;
@@ -2474,6 +2477,7 @@ static int drive_init(struct drive_opt *arg, int snapshot,
         break;
     case IF_PFLASH:
     case IF_MTD:
+    case IF_VIRTIO:
         break;
     }
     if (!file[0])
-- 
1.5.6.5