lguest: the block driver
authorRusty Russell <rusty@rustcorp.com.au>
Thu, 19 Jul 2007 08:49:29 +0000 (01:49 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Thu, 19 Jul 2007 17:04:53 +0000 (10:04 -0700)
Lguest block driver

A simple block driver for lguest.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Andi Kleen <ak@suse.de>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
drivers/block/Makefile
drivers/block/lguest_blk.c [new file with mode: 0644]

index 3e31532df0eda9bfa6266ad5db9058d238220b68..819c829125fb9ee549c5cbf9e416db6383de67ae 100644 (file)
@@ -30,3 +30,4 @@ obj-$(CONFIG_BLK_DEV_SX8)     += sx8.o
 obj-$(CONFIG_BLK_DEV_UB)       += ub.o
 
 obj-$(CONFIG_XEN_BLKDEV_FRONTEND)      += xen-blkfront.o
+obj-$(CONFIG_LGUEST_GUEST)     += lguest_blk.o
diff --git a/drivers/block/lguest_blk.c b/drivers/block/lguest_blk.c
new file mode 100644 (file)
index 0000000..1634c2d
--- /dev/null
@@ -0,0 +1,275 @@
+/* A simple block driver for lguest.
+ *
+ * Copyright 2006 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+//#define DEBUG
+#include <linux/init.h>
+#include <linux/types.h>
+#include <linux/blkdev.h>
+#include <linux/interrupt.h>
+#include <linux/lguest_bus.h>
+
+static char next_block_index = 'a';
+
+struct blockdev
+{
+       spinlock_t lock;
+
+       /* The disk structure for the kernel. */
+       struct gendisk *disk;
+
+       /* The major number for this disk. */
+       int major;
+       int irq;
+
+       unsigned long phys_addr;
+       /* The mapped block page. */
+       struct lguest_block_page *lb_page;
+
+       /* We only have a single request outstanding at a time. */
+       struct lguest_dma dma;
+       struct request *req;
+};
+
+/* Jens gave me this nice helper to end all chunks of a request. */
+static void end_entire_request(struct request *req, int uptodate)
+{
+       if (end_that_request_first(req, uptodate, req->hard_nr_sectors))
+               BUG();
+       add_disk_randomness(req->rq_disk);
+       blkdev_dequeue_request(req);
+       end_that_request_last(req, uptodate);
+}
+
+static irqreturn_t lgb_irq(int irq, void *_bd)
+{
+       struct blockdev *bd = _bd;
+       unsigned long flags;
+
+       if (!bd->req) {
+               pr_debug("No work!\n");
+               return IRQ_NONE;
+       }
+
+       if (!bd->lb_page->result) {
+               pr_debug("No result!\n");
+               return IRQ_NONE;
+       }
+
+       spin_lock_irqsave(&bd->lock, flags);
+       end_entire_request(bd->req, bd->lb_page->result == 1);
+       bd->req = NULL;
+       bd->dma.used_len = 0;
+       blk_start_queue(bd->disk->queue);
+       spin_unlock_irqrestore(&bd->lock, flags);
+       return IRQ_HANDLED;
+}
+
+static unsigned int req_to_dma(struct request *req, struct lguest_dma *dma)
+{
+       unsigned int i = 0, idx, len = 0;
+       struct bio *bio;
+
+       rq_for_each_bio(bio, req) {
+               struct bio_vec *bvec;
+               bio_for_each_segment(bvec, bio, idx) {
+                       BUG_ON(i == LGUEST_MAX_DMA_SECTIONS);
+                       BUG_ON(!bvec->bv_len);
+                       dma->addr[i] = page_to_phys(bvec->bv_page)
+                               + bvec->bv_offset;
+                       dma->len[i] = bvec->bv_len;
+                       len += bvec->bv_len;
+                       i++;
+               }
+       }
+       if (i < LGUEST_MAX_DMA_SECTIONS)
+               dma->len[i] = 0;
+       return len;
+}
+
+static void empty_dma(struct lguest_dma *dma)
+{
+       dma->len[0] = 0;
+}
+
+static void setup_req(struct blockdev *bd,
+                     int type, struct request *req, struct lguest_dma *dma)
+{
+       bd->lb_page->type = type;
+       bd->lb_page->sector = req->sector;
+       bd->lb_page->result = 0;
+       bd->req = req;
+       bd->lb_page->bytes = req_to_dma(req, dma);
+}
+
+static void do_write(struct blockdev *bd, struct request *req)
+{
+       struct lguest_dma send;
+
+       pr_debug("lgb: WRITE sector %li\n", (long)req->sector);
+       setup_req(bd, 1, req, &send);
+
+       lguest_send_dma(bd->phys_addr, &send);
+}
+
+static void do_read(struct blockdev *bd, struct request *req)
+{
+       struct lguest_dma ping;
+
+       pr_debug("lgb: READ sector %li\n", (long)req->sector);
+       setup_req(bd, 0, req, &bd->dma);
+
+       empty_dma(&ping);
+       lguest_send_dma(bd->phys_addr, &ping);
+}
+
+static void do_lgb_request(request_queue_t *q)
+{
+       struct blockdev *bd;
+       struct request *req;
+
+again:
+       req = elv_next_request(q);
+       if (!req)
+               return;
+
+       bd = req->rq_disk->private_data;
+       /* Sometimes we get repeated requests after blk_stop_queue. */
+       if (bd->req)
+               return;
+
+       if (!blk_fs_request(req)) {
+               pr_debug("Got non-command 0x%08x\n", req->cmd_type);
+               req->errors++;
+               end_entire_request(req, 0);
+               goto again;
+       }
+
+       if (rq_data_dir(req) == WRITE)
+               do_write(bd, req);
+       else
+               do_read(bd, req);
+
+       /* Wait for interrupt to tell us it's done. */
+       blk_stop_queue(q);
+}
+
+static struct block_device_operations lguestblk_fops = {
+       .owner = THIS_MODULE,
+};
+
+static int lguestblk_probe(struct lguest_device *lgdev)
+{
+       struct blockdev *bd;
+       int err;
+       int irqflags = IRQF_SHARED;
+
+       bd = kmalloc(sizeof(*bd), GFP_KERNEL);
+       if (!bd)
+               return -ENOMEM;
+
+       spin_lock_init(&bd->lock);
+       bd->irq = lgdev_irq(lgdev);
+       bd->req = NULL;
+       bd->dma.used_len = 0;
+       bd->dma.len[0] = 0;
+       bd->phys_addr = (lguest_devices[lgdev->index].pfn << PAGE_SHIFT);
+
+       bd->lb_page = lguest_map(bd->phys_addr, 1);
+       if (!bd->lb_page) {
+               err = -ENOMEM;
+               goto out_free_bd;
+       }
+
+       bd->major = register_blkdev(0, "lguestblk");
+       if (bd->major < 0) {
+               err = bd->major;
+               goto out_unmap;
+       }
+
+       bd->disk = alloc_disk(1);
+       if (!bd->disk) {
+               err = -ENOMEM;
+               goto out_unregister_blkdev;
+       }
+
+       bd->disk->queue = blk_init_queue(do_lgb_request, &bd->lock);
+       if (!bd->disk->queue) {
+               err = -ENOMEM;
+               goto out_put_disk;
+       }
+
+       /* We can only handle a certain number of sg entries */
+       blk_queue_max_hw_segments(bd->disk->queue, LGUEST_MAX_DMA_SECTIONS);
+       /* Buffers must not cross page boundaries */
+       blk_queue_segment_boundary(bd->disk->queue, PAGE_SIZE-1);
+
+       sprintf(bd->disk->disk_name, "lgb%c", next_block_index++);
+       if (lguest_devices[lgdev->index].features & LGUEST_DEVICE_F_RANDOMNESS)
+               irqflags |= IRQF_SAMPLE_RANDOM;
+       err = request_irq(bd->irq, lgb_irq, irqflags, bd->disk->disk_name, bd);
+       if (err)
+               goto out_cleanup_queue;
+
+       err = lguest_bind_dma(bd->phys_addr, &bd->dma, 1, bd->irq);
+       if (err)
+               goto out_free_irq;
+
+       bd->disk->major = bd->major;
+       bd->disk->first_minor = 0;
+       bd->disk->private_data = bd;
+       bd->disk->fops = &lguestblk_fops;
+       /* This is initialized to the disk size by the other end. */
+       set_capacity(bd->disk, bd->lb_page->num_sectors);
+       add_disk(bd->disk);
+
+       printk(KERN_INFO "%s: device %i at major %d\n",
+              bd->disk->disk_name, lgdev->index, bd->major);
+
+       lgdev->private = bd;
+       return 0;
+
+out_free_irq:
+       free_irq(bd->irq, bd);
+out_cleanup_queue:
+       blk_cleanup_queue(bd->disk->queue);
+out_put_disk:
+       put_disk(bd->disk);
+out_unregister_blkdev:
+       unregister_blkdev(bd->major, "lguestblk");
+out_unmap:
+       lguest_unmap(bd->lb_page);
+out_free_bd:
+       kfree(bd);
+       return err;
+}
+
+static struct lguest_driver lguestblk_drv = {
+       .name = "lguestblk",
+       .owner = THIS_MODULE,
+       .device_type = LGUEST_DEVICE_T_BLOCK,
+       .probe = lguestblk_probe,
+};
+
+static __init int lguestblk_init(void)
+{
+       return register_lguest_driver(&lguestblk_drv);
+}
+module_init(lguestblk_init);
+
+MODULE_DESCRIPTION("Lguest block driver");
+MODULE_LICENSE("GPL");