viodasd.c

Linux块设备驱动源码

	return -1;
}

/*
 * This is the external request processing routine
 */
static void do_viodasd_request(request_queue_t *q)
{
	struct request *req;

	/*
	 * If we already have the maximum number of requests
	 * outstanding to OS/400 just bail out. We'll come
	 * back later.
	 */
	while (num_req_outstanding < VIOMAXREQ) {
		req = elv_next_request(q);
		if (req == NULL)
			return;
		/* dequeue the current request from the queue */
		blkdev_dequeue_request(req);
		/* check that request contains a valid command */
		if (!blk_fs_request(req)) {
			viodasd_end_request(req, 0, req->hard_nr_sectors);
			continue;
		}
		/* Try sending the request */
		if (send_request(req) != 0)
			viodasd_end_request(req, 0, req->hard_nr_sectors);
	}
}

/*
 * Probe a single disk and fill in the viodasd_device structure
 * for it.
 */
static void probe_disk(struct viodasd_device *d)
{
	HvLpEvent_Rc hvrc;
	struct viodasd_waitevent we;
	int dev_no = DEVICE_NO(d);
	struct gendisk *g;
	struct request_queue *q;
	u16 flags = 0;

retry:
	init_completion(&we.com);

	/* Send the open event to OS/400 */
	hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
			HvLpEvent_Type_VirtualIo,
			viomajorsubtype_blockio | vioblockopen,
			HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
			viopath_sourceinst(viopath_hostLp),
			viopath_targetinst(viopath_hostLp),
			(u64)(unsigned long)&we, VIOVERSION << 16,
			((u64)dev_no << 48) | ((u64)flags<< 32),
			0, 0, 0);
	if (hvrc != 0) {
		printk(VIOD_KERN_WARNING "bad rc on HV open %d\n", (int)hvrc);
		return;
	}

	wait_for_completion(&we.com);

	if (we.rc != 0) {
		if (flags != 0)
			return;
		/* try again with read only flag set */
		flags = vioblockflags_ro;
		goto retry;
	}
	if (we.max_disk > (MAX_DISKNO - 1)) {
		static int warned;

		if (warned == 0) {
			warned++;
			printk(VIOD_KERN_INFO
				"Only examining the first %d "
				"of %d disks connected\n",
				MAX_DISKNO, we.max_disk + 1);
		}
	}

	/* Send the close event to OS/400.  We DON'T expect a response */
	hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
			HvLpEvent_Type_VirtualIo,
			viomajorsubtype_blockio | vioblockclose,
			HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
			viopath_sourceinst(viopath_hostLp),
			viopath_targetinst(viopath_hostLp),
			0, VIOVERSION << 16,
			((u64)dev_no << 48) | ((u64)flags << 32),
			0, 0, 0);
	if (hvrc != 0) {
		printk(VIOD_KERN_WARNING
		       "bad rc sending event to OS/400 %d\n", (int)hvrc);
		return;
	}
	/* create the request queue for the disk */
	spin_lock_init(&d->q_lock);
	q = blk_init_queue(do_viodasd_request, &d->q_lock);
	if (q == NULL) {
		printk(VIOD_KERN_WARNING "cannot allocate queue for disk %d\n",
				dev_no);
		return;
	}
	g = alloc_disk(1 << PARTITION_SHIFT);
	if (g == NULL) {
		printk(VIOD_KERN_WARNING
				"cannot allocate disk structure for disk %d\n",
				dev_no);
		blk_cleanup_queue(q);
		return;
	}

	d->disk = g;
	blk_queue_max_hw_segments(q, VIOMAXBLOCKDMA);
	blk_queue_max_phys_segments(q, VIOMAXBLOCKDMA);
	blk_queue_max_sectors(q, VIODASD_MAXSECTORS);
	g->major = VIODASD_MAJOR;
	g->first_minor = dev_no << PARTITION_SHIFT;
	if (dev_no >= 26)
		snprintf(g->disk_name, sizeof(g->disk_name),
				VIOD_GENHD_NAME "%c%c",
				'a' + (dev_no / 26) - 1, 'a' + (dev_no % 26));
	else
		snprintf(g->disk_name, sizeof(g->disk_name),
				VIOD_GENHD_NAME "%c", 'a' + (dev_no % 26));
	snprintf(g->devfs_name, sizeof(g->devfs_name),
			"%s%d", VIOD_GENHD_DEVFS_NAME, dev_no);
	g->fops = &viodasd_fops;
	g->queue = q;
	g->private_data = d;
	g->driverfs_dev = d->dev;
	set_capacity(g, d->size >> 9);

	printk(VIOD_KERN_INFO "disk %d: %lu sectors (%lu MB) "
			"CHS=%d/%d/%d sector size %d%s\n",
			dev_no, (unsigned long)(d->size >> 9),
			(unsigned long)(d->size >> 20),
			(int)d->cylinders, (int)d->tracks,
			(int)d->sectors, (int)d->bytes_per_sector,
			d->read_only ? " (RO)" : "");

	/* register us in the global list */
	add_disk(g);
}

/* returns the total number of scatterlist elements converted */
static int block_event_to_scatterlist(const struct vioblocklpevent *bevent,
		struct scatterlist *sg, int *total_len)
{
	int i, numsg;
	const struct rw_data *rw_data = &bevent->u.rw_data;
	static const int offset =
		offsetof(struct vioblocklpevent, u.rw_data.dma_info);
	static const int element_size = sizeof(rw_data->dma_info[0]);

	numsg = ((bevent->event.xSizeMinus1 + 1) - offset) / element_size;
	if (numsg > VIOMAXBLOCKDMA)
		numsg = VIOMAXBLOCKDMA;

	*total_len = 0;
	memset(sg, 0, sizeof(sg[0]) * VIOMAXBLOCKDMA);

	for (i = 0; (i < numsg) && (rw_data->dma_info[i].len > 0); ++i) {
		sg_dma_address(&sg[i]) = rw_data->dma_info[i].token;
		sg_dma_len(&sg[i]) = rw_data->dma_info[i].len;
		*total_len += rw_data->dma_info[i].len;
	}
	return i;
}

/*
 * Restart all queues, starting with the one _after_ the disk given,
 * thus reducing the chance of starvation of higher numbered disks.
 */
static void viodasd_restart_all_queues_starting_from(int first_index)
{
	int i;

	for (i = first_index + 1; i < MAX_DISKNO; ++i)
		if (viodasd_devices[i].disk)
			blk_run_queue(viodasd_devices[i].disk->queue);
	for (i = 0; i <= first_index; ++i)
		if (viodasd_devices[i].disk)
			blk_run_queue(viodasd_devices[i].disk->queue);
}

/*
 * For read and write requests, decrement the number of outstanding requests,
 * Free the DMA buffers we allocated.
 */
static int viodasd_handle_read_write(struct vioblocklpevent *bevent)
{
	int num_sg, num_sect, pci_direction, total_len;
	struct request *req;
	struct scatterlist sg[VIOMAXBLOCKDMA];
	struct HvLpEvent *event = &bevent->event;
	unsigned long irq_flags;
	struct viodasd_device *d;
	int error;
	spinlock_t *qlock;

	num_sg = block_event_to_scatterlist(bevent, sg, &total_len);
	num_sect = total_len >> 9;
	if (event->xSubtype == (viomajorsubtype_blockio | vioblockread))
		pci_direction = DMA_FROM_DEVICE;
	else
		pci_direction = DMA_TO_DEVICE;
	req = (struct request *)bevent->event.xCorrelationToken;
	d = req->rq_disk->private_data;

	dma_unmap_sg(d->dev, sg, num_sg, pci_direction);

	/*
	 * Since this is running in interrupt mode, we need to make sure
	 * we're not stepping on any global I/O operations
	 */
	spin_lock_irqsave(&viodasd_spinlock, irq_flags);
	num_req_outstanding--;
	spin_unlock_irqrestore(&viodasd_spinlock, irq_flags);

	error = event->xRc != HvLpEvent_Rc_Good;
	if (error) {
		const struct vio_error_entry *err;
		err = vio_lookup_rc(viodasd_err_table, bevent->sub_result);
		printk(VIOD_KERN_WARNING "read/write error %d:0x%04x (%s)\n",
				event->xRc, bevent->sub_result, err->msg);
		num_sect = req->hard_nr_sectors;
	}
	qlock = req->q->queue_lock;
	spin_lock_irqsave(qlock, irq_flags);
	viodasd_end_request(req, !error, num_sect);
	spin_unlock_irqrestore(qlock, irq_flags);

	/* Finally, try to get more requests off of this device's queue */
	viodasd_restart_all_queues_starting_from(DEVICE_NO(d));

	return 0;
}

/* This routine handles incoming block LP events */
static void handle_block_event(struct HvLpEvent *event)
{
	struct vioblocklpevent *bevent = (struct vioblocklpevent *)event;
	struct viodasd_waitevent *pwe;

	if (event == NULL)
		/* Notification that a partition went away! */
		return;
	/* First, we should NEVER get an int here...only acks */
	if (event->xFlags.xFunction == HvLpEvent_Function_Int) {
		printk(VIOD_KERN_WARNING
		       "Yikes! got an int in viodasd event handler!\n");
		if (event->xFlags.xAckInd == HvLpEvent_AckInd_DoAck) {
			event->xRc = HvLpEvent_Rc_InvalidSubtype;
			HvCallEvent_ackLpEvent(event);
		}
	}

	switch (event->xSubtype & VIOMINOR_SUBTYPE_MASK) {
	case vioblockopen:
		/*
		 * Handle a response to an open request.  We get all the
		 * disk information in the response, so update it.  The
		 * correlation token contains a pointer to a waitevent
		 * structure that has a completion in it.  update the
		 * return code in the waitevent structure and post the
		 * completion to wake up the guy who sent the request
		 */
		pwe = (struct viodasd_waitevent *)event->xCorrelationToken;
		pwe->rc = event->xRc;
		pwe->sub_result = bevent->sub_result;
		if (event->xRc == HvLpEvent_Rc_Good) {
			const struct open_data *data = &bevent->u.open_data;
			struct viodasd_device *device =
				&viodasd_devices[bevent->disk];
			device->read_only =
				bevent->flags & vioblockflags_ro;
			device->size = data->disk_size;
			device->cylinders = data->cylinders;
			device->tracks = data->tracks;
			device->sectors = data->sectors;
			device->bytes_per_sector = data->bytes_per_sector;
			pwe->max_disk = data->max_disk;
		}
		complete(&pwe->com);
		break;
	case vioblockclose:
		break;
	case vioblockread:
	case vioblockwrite:
		viodasd_handle_read_write(bevent);
		break;

	default:
		printk(VIOD_KERN_WARNING "invalid subtype!");
		if (event->xFlags.xAckInd == HvLpEvent_AckInd_DoAck) {
			event->xRc = HvLpEvent_Rc_InvalidSubtype;
			HvCallEvent_ackLpEvent(event);
		}
	}
}

/*
 * Get the driver to reprobe for more disks.
 */
static ssize_t probe_disks(struct device_driver *drv, const char *buf,
		size_t count)
{
	struct viodasd_device *d;

	for (d = viodasd_devices; d < &viodasd_devices[MAX_DISKNO]; d++) {
		if (d->disk == NULL)
			probe_disk(d);
	}
	return count;
}
static DRIVER_ATTR(probe, S_IWUSR, NULL, probe_disks);

static int viodasd_probe(struct vio_dev *vdev, const struct vio_device_id *id)
{
	struct viodasd_device *d = &viodasd_devices[vdev->unit_address];

	d->dev = &vdev->dev;
	probe_disk(d);
	if (d->disk == NULL)
		return -ENODEV;
	return 0;
}

static int viodasd_remove(struct vio_dev *vdev)
{
	struct viodasd_device *d;

	d = &viodasd_devices[vdev->unit_address];
	if (d->disk) {
		del_gendisk(d->disk);
		blk_cleanup_queue(d->disk->queue);
		put_disk(d->disk);
		d->disk = NULL;
	}
	d->dev = NULL;
	return 0;
}

/**
 * viodasd_device_table: Used by vio.c to match devices that we
 * support.
 */
static struct vio_device_id viodasd_device_table[] __devinitdata = {
	{ "viodasd", "" },
	{ "", "" }
};

MODULE_DEVICE_TABLE(vio, viodasd_device_table);
static struct vio_driver viodasd_driver = {
	.name = "viodasd",
	.id_table = viodasd_device_table,
	.probe = viodasd_probe,
	.remove = viodasd_remove
};

/*
 * Initialize the whole device driver.  Handle module and non-module
 * versions
 */
static int __init viodasd_init(void)
{
	int rc;

	/* Try to open to our host lp */
	if (viopath_hostLp == HvLpIndexInvalid)
		vio_set_hostlp();

	if (viopath_hostLp == HvLpIndexInvalid) {
		printk(VIOD_KERN_WARNING "invalid hosting partition\n");
		return -EIO;
	}

	printk(VIOD_KERN_INFO "vers " VIOD_VERS ", hosting partition %d\n",
			viopath_hostLp);

        /* register the block device */
	if (register_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME)) {
		printk(VIOD_KERN_WARNING
				"Unable to get major number %d for %s\n",
				VIODASD_MAJOR, VIOD_GENHD_NAME);
		return -EIO;
	}
	/* Actually open the path to the hosting partition */
	if (viopath_open(viopath_hostLp, viomajorsubtype_blockio,
				VIOMAXREQ + 2)) {
		printk(VIOD_KERN_WARNING
		       "error opening path to host partition %d\n",
		       viopath_hostLp);
		unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
		return -EIO;
	}

	/* Initialize our request handler */
	vio_setHandler(viomajorsubtype_blockio, handle_block_event);

	rc = vio_register_driver(&viodasd_driver);
	if (rc == 0)
		driver_create_file(&viodasd_driver.driver, &driver_attr_probe);
	return rc;
}
module_init(viodasd_init);

void viodasd_exit(void)
{
	driver_remove_file(&viodasd_driver.driver, &driver_attr_probe);
	vio_unregister_driver(&viodasd_driver);
	vio_clearHandler(viomajorsubtype_blockio);
	unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
	viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
}

module_exit(viodasd_exit);