mtip32xx.c

/*
 * Executes a taskfile
 * See ide_taskfile_ioctl() for derivation
 */
static int exec_drive_taskfile(struct driver_data *dd,
			       void __user *buf,
			       ide_task_request_t *req_task,
			       int outtotal)
{
	struct host_to_dev_fis	fis;
	struct host_to_dev_fis *reply;
	u8 *outbuf = NULL;
	u8 *inbuf = NULL;
	dma_addr_t outbuf_dma = 0;
	dma_addr_t inbuf_dma = 0;
	dma_addr_t dma_buffer = 0;
	int err = 0;
	unsigned int taskin = 0;
	unsigned int taskout = 0;
	u8 nsect = 0;
	unsigned int timeout = MTIP_IOCTL_COMMAND_TIMEOUT_MS;
	unsigned int force_single_sector;
	unsigned int transfer_size;
	unsigned long task_file_data;
	int intotal = outtotal + req_task->out_size;

	taskout = req_task->out_size;
	taskin = req_task->in_size;
	/* 130560 = 512 * 0xFF*/
	if (taskin > 130560 || taskout > 130560) {
		err = -EINVAL;
		goto abort;
	}

	if (taskout) {
		outbuf = kzalloc(taskout, GFP_KERNEL);
		if (outbuf == NULL) {
			err = -ENOMEM;
			goto abort;
		}
		if (copy_from_user(outbuf, buf + outtotal, taskout)) {
			err = -EFAULT;
			goto abort;
		}
		outbuf_dma = pci_map_single(dd->pdev,
					 outbuf,
					 taskout,
					 DMA_TO_DEVICE);
		if (outbuf_dma == 0) {
			err = -ENOMEM;
			goto abort;
		}
		dma_buffer = outbuf_dma;
	}

	if (taskin) {
		inbuf = kzalloc(taskin, GFP_KERNEL);
		if (inbuf == NULL) {
			err = -ENOMEM;
			goto abort;
		}

		if (copy_from_user(inbuf, buf + intotal, taskin)) {
			err = -EFAULT;
			goto abort;
		}
		inbuf_dma = pci_map_single(dd->pdev,
					 inbuf,
					 taskin, DMA_FROM_DEVICE);
		if (inbuf_dma == 0) {
			err = -ENOMEM;
			goto abort;
		}
		dma_buffer = inbuf_dma;
	}

	/* only supports PIO and non-data commands from this ioctl. */
	switch (req_task->data_phase) {
	case TASKFILE_OUT:
		nsect = taskout / ATA_SECT_SIZE;
		reply = (dd->port->rxfis + RX_FIS_PIO_SETUP);
		break;
	case TASKFILE_IN:
		reply = (dd->port->rxfis + RX_FIS_PIO_SETUP);
		break;
	case TASKFILE_NO_DATA:
		reply = (dd->port->rxfis + RX_FIS_D2H_REG);
		break;
	default:
		err = -EINVAL;
		goto abort;
	}

	/* Build the FIS. */
	memset(&fis, 0, sizeof(struct host_to_dev_fis));

	fis.type	= 0x27;
	fis.opts	= 1 << 7;
	fis.command	= req_task->io_ports[7];
	fis.features	= req_task->io_ports[1];
	fis.sect_count	= req_task->io_ports[2];
	fis.lba_low	= req_task->io_ports[3];
	fis.lba_mid	= req_task->io_ports[4];
	fis.lba_hi	= req_task->io_ports[5];
	 /* Clear the dev bit*/
	fis.device	= req_task->io_ports[6] & ~0x10;

	if ((req_task->in_flags.all == 0) && (req_task->out_flags.all & 1)) {
		req_task->in_flags.all	=
			IDE_TASKFILE_STD_IN_FLAGS |
			(IDE_HOB_STD_IN_FLAGS << 8);
		fis.lba_low_ex		= req_task->hob_ports[3];
		fis.lba_mid_ex		= req_task->hob_ports[4];
		fis.lba_hi_ex		= req_task->hob_ports[5];
		fis.features_ex		= req_task->hob_ports[1];
		fis.sect_cnt_ex		= req_task->hob_ports[2];

	} else {
		req_task->in_flags.all = IDE_TASKFILE_STD_IN_FLAGS;
	}

	force_single_sector = implicit_sector(fis.command, fis.features);

	if ((taskin || taskout) && (!fis.sect_count)) {
		if (nsect)
			fis.sect_count = nsect;
		else {
			if (!force_single_sector) {
				dev_warn(&dd->pdev->dev,
					"data movement but "
					"sect_count is 0\n");
					err = -EINVAL;
					goto abort;
			}
		}
	}

	dbg_printk(MTIP_DRV_NAME
		" %s: cmd %x, feat %x, nsect %x,"
		" sect/lbal %x, lcyl/lbam %x, hcyl/lbah %x,"
		" head/dev %x\n",
		__func__,
		fis.command,
		fis.features,
		fis.sect_count,
		fis.lba_low,
		fis.lba_mid,
		fis.lba_hi,
		fis.device);

	switch (fis.command) {
	case ATA_CMD_DOWNLOAD_MICRO:
		/* Change timeout for Download Microcode to 2 minutes */
		timeout = 120000;
		break;
	case ATA_CMD_SEC_ERASE_UNIT:
		/* Change timeout for Security Erase Unit to 4 minutes.*/
		timeout = 240000;
		break;
	case ATA_CMD_STANDBYNOW1:
		/* Change timeout for standby immediate to 10 seconds.*/
		timeout = 10000;
		break;
	case 0xF7:
	case 0xFA:
		/* Change timeout for vendor unique command to 10 secs */
		timeout = 10000;
		break;
	case ATA_CMD_SMART:
		/* Change timeout for vendor unique command to 15 secs */
		timeout = 15000;
		break;
	default:
		timeout = MTIP_IOCTL_COMMAND_TIMEOUT_MS;
		break;
	}

	/* Determine the correct transfer size.*/
	if (force_single_sector)
		transfer_size = ATA_SECT_SIZE;
	else
		transfer_size = ATA_SECT_SIZE * fis.sect_count;

	/* Execute the command.*/
	if (mtip_exec_internal_command(dd->port,
				 &fis,
				 5,
				 dma_buffer,
				 transfer_size,
				 0,
				 GFP_KERNEL,
				 timeout) < 0) {
		err = -EIO;
		goto abort;
	}

	task_file_data = readl(dd->port->mmio+PORT_TFDATA);

	if ((req_task->data_phase == TASKFILE_IN) && !(task_file_data & 1)) {
		reply = dd->port->rxfis + RX_FIS_PIO_SETUP;
		req_task->io_ports[7] = reply->control;
	} else {
		reply = dd->port->rxfis + RX_FIS_D2H_REG;
		req_task->io_ports[7] = reply->command;
	}

	/* reclaim the DMA buffers.*/
	if (inbuf_dma)
		pci_unmap_single(dd->pdev, inbuf_dma,
			taskin, DMA_FROM_DEVICE);
	if (outbuf_dma)
		pci_unmap_single(dd->pdev, outbuf_dma,
			taskout, DMA_TO_DEVICE);
	inbuf_dma  = 0;
	outbuf_dma = 0;

	/* return the ATA registers to the caller.*/
	req_task->io_ports[1] = reply->features;
	req_task->io_ports[2] = reply->sect_count;
	req_task->io_ports[3] = reply->lba_low;
	req_task->io_ports[4] = reply->lba_mid;
	req_task->io_ports[5] = reply->lba_hi;
	req_task->io_ports[6] = reply->device;

	if (req_task->out_flags.all & 1)  {

		req_task->hob_ports[3] = reply->lba_low_ex;
		req_task->hob_ports[4] = reply->lba_mid_ex;
		req_task->hob_ports[5] = reply->lba_hi_ex;
		req_task->hob_ports[1] = reply->features_ex;
		req_task->hob_ports[2] = reply->sect_cnt_ex;
	}
	dbg_printk(MTIP_DRV_NAME
		" %s: Completion: stat %x,"
		"err %x, sect_cnt %x, lbalo %x,"
		"lbamid %x, lbahi %x, dev %x\n",
		__func__,
		req_task->io_ports[7],
		req_task->io_ports[1],
		req_task->io_ports[2],
		req_task->io_ports[3],
		req_task->io_ports[4],
		req_task->io_ports[5],
		req_task->io_ports[6]);

	if (taskout) {
		if (copy_to_user(buf + outtotal, outbuf, taskout)) {
			err = -EFAULT;
			goto abort;
		}
	}
	if (taskin) {
		if (copy_to_user(buf + intotal, inbuf, taskin)) {
			err = -EFAULT;
			goto abort;
		}
	}
abort:
	if (inbuf_dma)
		pci_unmap_single(dd->pdev, inbuf_dma,
					taskin, DMA_FROM_DEVICE);
	if (outbuf_dma)
		pci_unmap_single(dd->pdev, outbuf_dma,
					taskout, DMA_TO_DEVICE);
	kfree(outbuf);
	kfree(inbuf);

	return err;
}

/*
 * Handle IOCTL calls from the Block Layer.
 *
 * This function is called by the Block Layer when it receives an IOCTL
 * command that it does not understand. If the IOCTL command is not supported
 * this function returns -ENOTTY.
 *
 * @dd  Pointer to the driver data structure.
 * @cmd IOCTL command passed from the Block Layer.
 * @arg IOCTL argument passed from the Block Layer.
 *
 * return value
 *	0	The IOCTL completed successfully.
 *	-ENOTTY The specified command is not supported.
 *	-EFAULT An error occurred copying data to a user space buffer.
 *	-EIO	An error occurred while executing the command.
 */
static int mtip_hw_ioctl(struct driver_data *dd, unsigned int cmd,
			 unsigned long arg)
{
	switch (cmd) {
	case HDIO_GET_IDENTITY:
		if (mtip_get_identify(dd->port, (void __user *) arg) < 0) {
			dev_warn(&dd->pdev->dev,
				"Unable to read identity\n");
			return -EIO;
		}

		break;
	case HDIO_DRIVE_CMD:
	{
		u8 drive_command[4];

		/* Copy the user command info to our buffer. */
		if (copy_from_user(drive_command,
					 (void __user *) arg,
					 sizeof(drive_command)))
			return -EFAULT;

		/* Execute the drive command. */
		if (exec_drive_command(dd->port,
					 drive_command,
					 (void __user *) (arg+4)))
			return -EIO;

		/* Copy the status back to the users buffer. */
		if (copy_to_user((void __user *) arg,
					 drive_command,
					 sizeof(drive_command)))
			return -EFAULT;

		break;
	}
	case HDIO_DRIVE_TASK:
	{
		u8 drive_command[7];

		/* Copy the user command info to our buffer. */
		if (copy_from_user(drive_command,
					 (void __user *) arg,
					 sizeof(drive_command)))
			return -EFAULT;

		/* Execute the drive command. */
		if (exec_drive_task(dd->port, drive_command))
			return -EIO;

		/* Copy the status back to the users buffer. */
		if (copy_to_user((void __user *) arg,
					 drive_command,
					 sizeof(drive_command)))
			return -EFAULT;

		break;
	}
	case HDIO_DRIVE_TASKFILE: {
		ide_task_request_t req_task;
		int ret, outtotal;

		if (copy_from_user(&req_task, (void __user *) arg,
					sizeof(req_task)))
			return -EFAULT;

		outtotal = sizeof(req_task);

		ret = exec_drive_taskfile(dd, (void __user *) arg,
						&req_task, outtotal);

		if (copy_to_user((void __user *) arg, &req_task,
							sizeof(req_task)))
			return -EFAULT;

		return ret;
	}

	default:
		return -EINVAL;
	}
	return 0;
}

/*
 * Submit an IO to the hw
 *
 * This function is called by the block layer to issue an io
 * to the device. Upon completion, the callback function will
 * be called with the data parameter passed as the callback data.
 *
 * @dd       Pointer to the driver data structure.
 * @start    First sector to read.
 * @nsect    Number of sectors to read.
 * @nents    Number of entries in scatter list for the read command.
 * @tag      The tag of this read command.
 * @callback Pointer to the function that should be called
 *	     when the read completes.
 * @data     Callback data passed to the callback function
 *	     when the read completes.
 * @dir      Direction (read or write)
 *
 * return value
 *	None
 */
static void mtip_hw_submit_io(struct driver_data *dd, sector_t start,
			      int nsect, int nents, int tag, void *callback,
			      void *data, int dir)
{
	struct host_to_dev_fis	*fis;
	struct mtip_port *port = dd->port;
	struct mtip_cmd *command = &port->commands[tag];
	int dma_dir = (dir == READ) ? DMA_FROM_DEVICE : DMA_TO_DEVICE;

	/* Map the scatter list for DMA access */
	nents = dma_map_sg(&dd->pdev->dev, command->sg, nents, dma_dir);

	command->scatter_ents = nents;

	/*
	 * The number of retries for this command before it is
	 * reported as a failure to the upper layers.
	 */
	command->retries = MTIP_MAX_RETRIES;

	/* Fill out fis */
	fis = command->command;
	fis->type        = 0x27;
	fis->opts        = 1 << 7;
	fis->command     =
		(dir == READ ? ATA_CMD_FPDMA_READ : ATA_CMD_FPDMA_WRITE);
	*((unsigned int *) &fis->lba_low) = (start & 0xFFFFFF);
	*((unsigned int *) &fis->lba_low_ex) = ((start >> 24) & 0xFFFFFF);
	fis->device	 = 1 << 6;
	fis->features    = nsect & 0xFF;
	fis->features_ex = (nsect >> 8) & 0xFF;
	fis->sect_count  = ((tag << 3) | (tag >> 5));
	fis->sect_cnt_ex = 0;
	fis->control     = 0;
	fis->res2        = 0;
	fis->res3        = 0;
	fill_command_sg(dd, command, nents);

	/* Populate the command header */
	command->command_header->opts =
			__force_bit2int cpu_to_le32(
				(nents << 16) | 5 | AHCI_CMD_PREFETCH);
	command->command_header->byte_count = 0;

	/*
	 * Set the completion function and data for the command
	 * within this layer.
	 */
	command->comp_data = dd;
	command->comp_func = mtip_async_complete;
	command->direction = dma_dir;

	/*
	 * Set the completion function and data for the command passed
	 * from the upper layer.
	 */
	command->async_data = data;
	command->async_callback = callback;

	/*
	 * To prevent this command from being issued
	 * if an internal command is in progress or error handling is active.
	 */
	if (port->flags & MTIP_PF_PAUSE_IO) {
		set_bit(tag, port->cmds_to_issue);
		set_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags);
		return;
	}

	/* Issue the command to the hardware */
	mtip_issue_ncq_command(port, tag);

	return;
}

/*
 * Release a command slot.
 *
 * @dd  Pointer to the driver data structure.
 * @tag Slot tag
 *
 * return value
 *      None
 */
static void mtip_hw_release_scatterlist(struct driver_data *dd, int tag)
{
	release_slot(dd->port, tag);
}

/*
 * Obtain a command slot and return its associated scatter list.
 *
 * @dd  Pointer to the driver data structure.
 * @tag Pointer to an int that will receive the allocated command
 *            slot tag.
 *
 * return value
 *	Pointer to the scatter list for the allocated command slot
 *	or NULL if no command slots are available.
 */
static struct scatterlist *mtip_hw_get_scatterlist(struct driver_data *dd,
						   int *tag)
{
	/*
	 * It is possible that, even with this semaphore, a thread
	 * may think that no command slots are available. Therefore, we
	 * need to make an attempt to get_slot().
	 */
	down(&dd->port->cmd_slot);
	*tag = get_slot(dd->port);

	if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag))) {
		up(&dd->port->cmd_slot);
		return NULL;
	}
	if (unlikely(*tag < 0))
		return NULL;

	return dd->port->commands[*tag].sg;
}

/*
 * Sysfs register/status dump.
 *
 * @dev  Pointer to the device structure, passed by the kernrel.
 * @attr Pointer to the device_attribute structure passed by the kernel.
 * @buf  Pointer to the char buffer that will receive the stats info.
 *
 * return value
 *	The size, in bytes, of the data copied into buf.
 */
static ssize_t mtip_hw_show_registers(struct device *dev,
				struct device_attribute *attr,
				char *buf)
{
	u32 group_allocated;
	struct driver_data *dd = dev_to_disk(dev)->private_data;
	int size = 0;
	int n;

	size += sprintf(&buf[size], "S ACTive:\n");

	for (n = 0; n < dd->slot_groups; n++)
		size += sprintf(&buf[size], "0x%08x\n",
					 readl(dd->port->s_active[n]));

	size += sprintf(&buf[size], "Command Issue:\n");

	for (n = 0; n < dd->slot_groups; n++)
		size += sprintf(&buf[size], "0x%08x\n",
					readl(dd->port->cmd_issue[n]));

	size += sprintf(&buf[size], "Allocated:\n");

	for (n = 0; n < dd->slot_groups; n++) {
		if (sizeof(long) > sizeof(u32))
			group_allocated =
				dd->port->allocated[n/2] >> (32*(n&1));
		else
			group_allocated = dd->port->allocated[n];
		size += sprintf(&buf[size], "0x%08x\n",
				 group_allocated);
	}

	size += sprintf(&buf[size], "Completed:\n");

	for (n = 0; n < dd->slot_groups; n++)
		size += sprintf(&buf[size], "0x%08x\n",
				readl(dd->port->completed[n]));

	size += sprintf(&buf[size], "PORT IRQ STAT : 0x%08x\n",
				readl(dd->port->mmio + PORT_IRQ_STAT));
	size += sprintf(&buf[size], "HOST IRQ STAT : 0x%08x\n",
				readl(dd->mmio + HOST_IRQ_STAT));

	return size;
}

static ssize_t mtip_hw_show_status(struct device *dev,
				struct device_attribute *attr,
				char *buf)
{
	struct driver_data *dd = dev_to_disk(dev)->private_data;
	int size = 0;

	if (test_bit(MTIP_DDF_OVER_TEMP_BIT, &dd->dd_flag))
		size += sprintf(buf, "%s", "thermal_shutdown\n");
	else if (test_bit(MTIP_DDF_WRITE_PROTECT_BIT, &dd->dd_flag))
		size += sprintf(buf, "%s", "write_protect\n");
	else
		size += sprintf(buf, "%s", "online\n");

	return size;
}

static DEVICE_ATTR(registers, S_IRUGO, mtip_hw_show_registers, NULL);
static DEVICE_ATTR(status, S_IRUGO, mtip_hw_show_status, NULL);

/*
 * Create the sysfs related attributes.
 *
 * @dd   Pointer to the driver data structure.
 * @kobj Pointer to the kobj for the block device.
 *
 * return value
 *	0	Operation completed successfully.
 *	-EINVAL Invalid parameter.
 */
static int mtip_hw_sysfs_init(struct driver_data *dd, struct kobject *kobj)
{
	if (!kobj || !dd)
		return -EINVAL;

	if (sysfs_create_file(kobj, &dev_attr_registers.attr))
		dev_warn(&dd->pdev->dev,
			"Error creating 'registers' sysfs entry\n");
	if (sysfs_create_file(kobj, &dev_attr_status.attr))
		dev_warn(&dd->pdev->dev,
			"Error creating 'status' sysfs entry\n");
	return 0;
}

/*
 * Remove the sysfs related attributes.
 *
 * @dd   Pointer to the driver data structure.
 * @kobj Pointer to the kobj for the block device.
 *
 * return value
 *	0	Operation completed successfully.
 *	-EINVAL Invalid parameter.
 */
static int mtip_hw_sysfs_exit(struct driver_data *dd, struct kobject *kobj)
{
	if (!kobj || !dd)
		return -EINVAL;

	sysfs_remove_file(kobj, &dev_attr_registers.attr);
	sysfs_remove_file(kobj, &dev_attr_status.attr);

	return 0;
}

/*
 * Perform any init/resume time hardware setup
 *
 * @dd Pointer to the driver data structure.
 *
 * return value
 *	None
 */
static inline void hba_setup(struct driver_data *dd)
{
	u32 hwdata;
	hwdata = readl(dd->mmio + HOST_HSORG);

	/* interrupt bug workaround: use only 1 IS bit.*/
	writel(hwdata |
		HSORG_DISABLE_SLOTGRP_INTR |
		HSORG_DISABLE_SLOTGRP_PXIS,
		dd->mmio + HOST_HSORG);
}

/*
 * Detect the details of the product, and store anything needed
 * into the driver data structure.  This includes product type and
 * version and number of slot groups.
 *
 * @dd Pointer to the driver data structure.
 *
 * return value
 *	None
 */
static void mtip_detect_product(struct driver_data *dd)
{
	u32 hwdata;
	unsigned int rev, slotgroups;

	/*
	 * HBA base + 0xFC [15:0] - vendor-specific hardware interface
	 * info register:
	 * [15:8] hardware/software interface rev#
	 * [   3] asic-style interface
	 * [ 2:0] number of slot groups, minus 1 (only valid for asic-style).
	 */
	hwdata = readl(dd->mmio + HOST_HSORG);

	dd->product_type = MTIP_PRODUCT_UNKNOWN;
	dd->slot_groups = 1;

	if (hwdata & 0x8) {
		dd->product_type = MTIP_PRODUCT_ASICFPGA;
		rev = (hwdata & HSORG_HWREV) >> 8;
		slotgroups = (hwdata & HSORG_SLOTGROUPS) + 1;
		dev_info(&dd->pdev->dev,
			"ASIC-FPGA design, HS rev 0x%x, "
			"%i slot groups [%i slots]\n",
			 rev,
			 slotgroups,
			 slotgroups * 32);

		if (slotgroups > MTIP_MAX_SLOT_GROUPS) {
			dev_warn(&dd->pdev->dev,
				"Warning: driver only supports "
				"%i slot groups.\n", MTIP_MAX_SLOT_GROUPS);
			slotgroups = MTIP_MAX_SLOT_GROUPS;
		}
		dd->slot_groups = slotgroups;
		return;
	}

	dev_warn(&dd->pdev->dev, "Unrecognized product id\n");
}

/*
 * Blocking wait for FTL rebuild to complete
 *
 * @dd Pointer to the DRIVER_DATA structure.
 *
 * return value
 *	0	FTL rebuild completed successfully
 *	-EFAULT FTL rebuild error/timeout/interruption
 */
static int mtip_ftl_rebuild_poll(struct driver_data *dd)
{
	unsigned long timeout, cnt = 0, start;

	dev_warn(&dd->pdev->dev,
		"FTL rebuild in progress. Polling for completion.\n");

	start = jiffies;
	timeout = jiffies + msecs_to_jiffies(MTIP_FTL_REBUILD_TIMEOUT_MS);

	do {
		if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
				&dd->dd_flag)))
			return -EFAULT;
		if (mtip_check_surprise_removal(dd->pdev))
			return -EFAULT;

		if (mtip_get_identify(dd->port, NULL) < 0)
			return -EFAULT;

		if (*(dd->port->identify + MTIP_FTL_REBUILD_OFFSET) ==
			MTIP_FTL_REBUILD_MAGIC) {
			ssleep(1);
			/* Print message every 3 minutes */
			if (cnt++ >= 180) {
				dev_warn(&dd->pdev->dev,
				"FTL rebuild in progress (%d secs).\n",
				jiffies_to_msecs(jiffies - start) / 1000);
				cnt = 0;
			}
		} else {
			dev_warn(&dd->pdev->dev,
				"FTL rebuild complete (%d secs).\n",
			jiffies_to_msecs(jiffies - start) / 1000);
			mtip_block_initialize(dd);
			return 0;
		}
		ssleep(10);
	} while (time_before(jiffies, timeout));

	/* Check for timeout */
	dev_err(&dd->pdev->dev,
		"Timed out waiting for FTL rebuild to complete (%d secs).\n",
		jiffies_to_msecs(jiffies - start) / 1000);
	return -EFAULT;
}

/*
 * service thread to issue queued commands
 *
 * @data Pointer to the driver data structure.
 *
 * return value
 *	0
 */

static int mtip_service_thread(void *data)
{
	struct driver_data *dd = (struct driver_data *)data;
	unsigned long slot, slot_start, slot_wrap;
	unsigned int num_cmd_slots = dd->slot_groups * 32;
	struct mtip_port *port = dd->port;

	while (1) {
		/*
		 * the condition is to check neither an internal command is
		 * is in progress nor error handling is active
		 */
		wait_event_interruptible(port->svc_wait, (port->flags) &&
			!(port->flags & MTIP_PF_PAUSE_IO));

		if (kthread_should_stop())
			break;

		if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT,
				&dd->dd_flag)))
			break;

		set_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags);
		if (test_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags)) {
			slot = 1;
			/* used to restrict the loop to one iteration */
			slot_start = num_cmd_slots;
			slot_wrap = 0;
			while (1) {
				slot = find_next_bit(port->cmds_to_issue,
						num_cmd_slots, slot);
				if (slot_wrap == 1) {
					if ((slot_start >= slot) ||
						(slot >= num_cmd_slots))
						break;
				}
				if (unlikely(slot_start == num_cmd_slots))
					slot_start = slot;

				if (unlikely(slot == num_cmd_slots)) {
					slot = 1;
					slot_wrap = 1;
					continue;
				}

				/* Issue the command to the hardware */
				mtip_issue_ncq_command(port, slot);

				clear_bit(slot, port->cmds_to_issue);
			}

			clear_bit(MTIP_PF_ISSUE_CMDS_BIT, &port->flags);
		} else if (test_bit(MTIP_PF_REBUILD_BIT, &port->flags)) {
			if (!mtip_ftl_rebuild_poll(dd))
				set_bit(MTIP_DDF_REBUILD_FAILED_BIT,
							&dd->dd_flag);
			clear_bit(MTIP_PF_REBUILD_BIT, &port->flags);
		}
		clear_bit(MTIP_PF_SVC_THD_ACTIVE_BIT, &port->flags);

		if (test_bit(MTIP_PF_SVC_THD_STOP_BIT, &port->flags))
			break;
	}
	return 0;
}

/*
 * Called once for each card.
 *
 * @dd Pointer to the driver data structure.
 *
 * return value
 *	0 on success, else an error code.
 */
static int mtip_hw_init(struct driver_data *dd)
{
	int i;
	int rv;
	unsigned int num_command_slots;
	unsigned long timeout, timetaken;
	unsigned char *buf;
	struct smart_attr attr242;

	dd->mmio = pcim_iomap_table(dd->pdev)[MTIP_ABAR];

	mtip_detect_product(dd);
	if (dd->product_type == MTIP_PRODUCT_UNKNOWN) {
		rv = -EIO;
		goto out1;
	}
	num_command_slots = dd->slot_groups * 32;

	hba_setup(dd);

	tasklet_init(&dd->tasklet, mtip_tasklet, (unsigned long)dd);

	dd->port = kzalloc(sizeof(struct mtip_port), GFP_KERNEL);
	if (!dd->port) {
		dev_err(&dd->pdev->dev,
			"Memory allocation: port structure\n");
		return -ENOMEM;
	}

	/* Counting semaphore to track command slot usage */
	sema_init(&dd->port->cmd_slot, num_command_slots - 1);

	/* Spinlock to prevent concurrent issue */
	spin_lock_init(&dd->port->cmd_issue_lock);

	/* Set the port mmio base address. */
	dd->port->mmio	= dd->mmio + PORT_OFFSET;
	dd->port->dd	= dd;

	/* Allocate memory for the command list. */
	dd->port->command_list =
		dmam_alloc_coherent(&dd->pdev->dev,
			HW_PORT_PRIV_DMA_SZ + (ATA_SECT_SIZE * 4),
			&dd->port->command_list_dma,
			GFP_KERNEL);
	if (!dd->port->command_list) {
		dev_err(&dd->pdev->dev,
			"Memory allocation: command list\n");
		rv = -ENOMEM;
		goto out1;
	}

	/* Clear the memory we have allocated. */
	memset(dd->port->command_list,
		0,
		HW_PORT_PRIV_DMA_SZ + (ATA_SECT_SIZE * 4));

	/* Setup the addresse of the RX FIS. */
	dd->port->rxfis	    = dd->port->command_list + HW_CMD_SLOT_SZ;
	dd->port->rxfis_dma = dd->port->command_list_dma + HW_CMD_SLOT_SZ;

	/* Setup the address of the command tables. */
	dd->port->command_table	  = dd->port->rxfis + AHCI_RX_FIS_SZ;
	dd->port->command_tbl_dma = dd->port->rxfis_dma + AHCI_RX_FIS_SZ;

	/* Setup the address of the identify data. */
	dd->port->identify     = dd->port->command_table +
					HW_CMD_TBL_AR_SZ;
	dd->port->identify_dma = dd->port->command_tbl_dma +
					HW_CMD_TBL_AR_SZ;

	/* Setup the address of the sector buffer - for some non-ncq cmds */
	dd->port->sector_buffer	= (void *) dd->port->identify + ATA_SECT_SIZE;
	dd->port->sector_buffer_dma = dd->port->identify_dma + ATA_SECT_SIZE;

	/* Setup the address of the log buf - for read log command */
	dd->port->log_buf = (void *)dd->port->sector_buffer  + ATA_SECT_SIZE;
	dd->port->log_buf_dma = dd->port->sector_buffer_dma + ATA_SECT_SIZE;

	/* Setup the address of the smart buf - for smart read data command */
	dd->port->smart_buf = (void *)dd->port->log_buf  + ATA_SECT_SIZE;
	dd->port->smart_buf_dma = dd->port->log_buf_dma + ATA_SECT_SIZE;


	/* Point the command headers at the command tables. */
	for (i = 0; i < num_command_slots; i++) {
		dd->port->commands[i].command_header =
					dd->port->command_list +
					(sizeof(struct mtip_cmd_hdr) * i);
		dd->port->commands[i].command_header_dma =
					dd->port->command_list_dma +
					(sizeof(struct mtip_cmd_hdr) * i);

		dd->port->commands[i].command =
			dd->port->command_table + (HW_CMD_TBL_SZ * i);
		dd->port->commands[i].command_dma =
			dd->port->command_tbl_dma + (HW_CMD_TBL_SZ * i);

		if (readl(dd->mmio + HOST_CAP) & HOST_CAP_64)
			dd->port->commands[i].command_header->ctbau =
			__force_bit2int cpu_to_le32(
			(dd->port->commands[i].command_dma >> 16) >> 16);
		dd->port->commands[i].command_header->ctba =
			__force_bit2int cpu_to_le32(
			dd->port->commands[i].command_dma & 0xFFFFFFFF);

		/*
		 * If this is not done, a bug is reported by the stock
		 * FC11 i386. Due to the fact that it has lots of kernel
		 * debugging enabled.
		 */
		sg_init_table(dd->port->commands[i].sg, MTIP_MAX_SG);

		/* Mark all commands as currently inactive.*/
		atomic_set(&dd->port->commands[i].active, 0);
	}

	/* Setup the pointers to the extended s_active and CI registers. */
	for (i = 0; i < dd->slot_groups; i++) {
		dd->port->s_active[i] =
			dd->port->mmio + i*0x80 + PORT_SCR_ACT;
		dd->port->cmd_issue[i] =
			dd->port->mmio + i*0x80 + PORT_COMMAND_ISSUE;
		dd->port->completed[i] =
			dd->port->mmio + i*0x80 + PORT_SDBV;
	}

	timetaken = jiffies;
	timeout = jiffies + msecs_to_jiffies(30000);
	while (((readl(dd->port->mmio + PORT_SCR_STAT) & 0x0F) != 0x03) &&
		 time_before(jiffies, timeout)) {
		mdelay(100);
	}
	if (unlikely(mtip_check_surprise_removal(dd->pdev))) {
		timetaken = jiffies - timetaken;
		dev_warn(&dd->pdev->dev,
			"Surprise removal detected at %u ms\n",
			jiffies_to_msecs(timetaken));
		rv = -ENODEV;
		goto out2 ;
	}
	if (unlikely(test_bit(MTIP_DDF_REMOVE_PENDING_BIT, &dd->dd_flag))) {
		timetaken = jiffies - timetaken;
		dev_warn(&dd->pdev->dev,
			"Removal detected at %u ms\n",
			jiffies_to_msecs(timetaken));
		rv = -EFAULT;
		goto out2;
	}

	/* Conditionally reset the HBA. */
	if (!(readl(dd->mmio + HOST_CAP) & HOST_CAP_NZDMA)) {
		if (mtip_hba_reset(dd) < 0) {
			dev_err(&dd->pdev->dev,