[PATCH v2 28/48] hpsa: honor queue depth of physical devices

[Don Brace <don.brace@pmcs.com>]

When using the ioaccel submission methods, requests destined for RAID volumes
are sometimes diverted to physical devices.  The OS has no or limited
knowledge of these physical devices, so it is up to the driver to avoid
pushing the device too hard.  It is better to honor the physical device queue
limit rather than making the device spew zillions of TASK SET FULL responses.

This is so that hpsa based devices support /sys/block/sdNN/device/queue_type
of simple, which lets the SCSI midlayer automatically adjust the queue_depth
based on TASK SET FULL and GOOD status.

Adjust the queue depth for a new device after it is created based on the
maximum queue depths of the physical devices that constitute the
device. This drops the maximum queue depth from .can_queue of 1024 to
something like 174 for single-drive RAID-0, 348 for two-drive RAID-1, etc.
It also adjusts for the ratio of data to parity drives.

Reviewed-by: Scott Teel <scott.teel@pmcs.com>
Signed-off-by: Webb Scales <webbnh@hp.com>
Signed-off-by: Don Brace <don.brace@pmcs.com>
---
 drivers/scsi/hpsa.c     |  318 +++++++++++++++++++++++++++++++++++++----------
 drivers/scsi/hpsa.h     |   14 ++
 drivers/scsi/hpsa_cmd.h |  148 ++++++++++++++++++++++
 3 files changed, 412 insertions(+), 68 deletions(-)

diff --git a/drivers/scsi/hpsa.c b/drivers/scsi/hpsa.c
index dcacb29..60f5734 100644
--- a/drivers/scsi/hpsa.c
+++ b/drivers/scsi/hpsa.c
@@ -247,7 +247,7 @@ static void hpsa_drain_accel_commands(struct ctlr_info *h);
 static void hpsa_flush_cache(struct ctlr_info *h);
 static int hpsa_scsi_ioaccel_queue_command(struct ctlr_info *h,
 	struct CommandList *c, u32 ioaccel_handle, u8 *cdb, int cdb_len,
-	u8 *scsi3addr);
+	u8 *scsi3addr, struct hpsa_scsi_dev_t *phys_disk);
 static void hpsa_command_resubmit_worker(struct work_struct *work);
 
 static inline struct ctlr_info *sdev_to_hba(struct scsi_device *sdev)
@@ -965,12 +965,24 @@ static void hpsa_scsi_update_entry(struct ctlr_info *h, int hostno,
 	/* Raid level changed. */
 	h->dev[entry]->raid_level = new_entry->raid_level;
 
-	/* Raid offload parameters changed. */
+	/* Raid offload parameters changed.  Careful about the ordering. */
+	if (new_entry->offload_config && new_entry->offload_enabled) {
+		/*
+		 * if drive is newly offload_enabled, we want to copy the
+		 * raid map data first.  If previously offload_enabled and
+		 * offload_config were set, raid map data had better be
+		 * the same as it was before.  if raid map data is changed
+		 * then it had better be the case that
+		 * h->dev[entry]->offload_enabled is currently 0.
+		 */
+		h->dev[entry]->raid_map = new_entry->raid_map;
+		h->dev[entry]->ioaccel_handle = new_entry->ioaccel_handle;
+		wmb(); /* ensure raid map updated prior to ->offload_enabled */
+	}
 	h->dev[entry]->offload_config = new_entry->offload_config;
-	h->dev[entry]->offload_enabled = new_entry->offload_enabled;
-	h->dev[entry]->ioaccel_handle = new_entry->ioaccel_handle;
 	h->dev[entry]->offload_to_mirror = new_entry->offload_to_mirror;
-	h->dev[entry]->raid_map = new_entry->raid_map;
+	h->dev[entry]->offload_enabled = new_entry->offload_enabled;
+	h->dev[entry]->queue_depth = new_entry->queue_depth;
 
 	dev_info(&h->pdev->dev, "%s device c%db%dt%dl%d updated.\n",
 		scsi_device_type(new_entry->devtype), hostno, new_entry->bus,
@@ -1096,6 +1108,8 @@ static inline int device_updated(struct hpsa_scsi_dev_t *dev1,
 		return 1;
 	if (dev1->offload_enabled != dev2->offload_enabled)
 		return 1;
+	if (dev1->queue_depth != dev2->queue_depth)
+		return 1;
 	return 0;
 }
 
@@ -1241,6 +1255,85 @@ static void hpsa_show_volume_status(struct ctlr_info *h,
 	}
 }
 
+/*
+ * Figure the list of physical drive pointers for a logical drive with
+ * raid offload configured.
+ */
+static void hpsa_figure_phys_disk_ptrs(struct ctlr_info *h,
+				struct hpsa_scsi_dev_t *dev[], int ndevices,
+				struct hpsa_scsi_dev_t *logical_drive)
+{
+	struct raid_map_data *map = &logical_drive->raid_map;
+	struct raid_map_disk_data *dd = &map->data[0];
+	int i, j;
+	int total_disks_per_row = le16_to_cpu(map->data_disks_per_row) +
+				le16_to_cpu(map->metadata_disks_per_row);
+	int nraid_map_entries = le16_to_cpu(map->row_cnt) *
+				le16_to_cpu(map->layout_map_count) *
+				total_disks_per_row;
+	int nphys_disk = le16_to_cpu(map->layout_map_count) *
+				total_disks_per_row;
+	int qdepth;
+
+	if (nraid_map_entries > RAID_MAP_MAX_ENTRIES)
+		nraid_map_entries = RAID_MAP_MAX_ENTRIES;
+
+	qdepth = 0;
+	for (i = 0; i < nraid_map_entries; i++) {
+		logical_drive->phys_disk[i] = NULL;
+		if (!logical_drive->offload_config)
+			continue;
+		for (j = 0; j < ndevices; j++) {
+			if (dev[j]->devtype != TYPE_DISK)
+				continue;
+			if (is_logical_dev_addr_mode(dev[j]->scsi3addr))
+				continue;
+			if (dev[j]->ioaccel_handle != dd[i].ioaccel_handle)
+				continue;
+
+			logical_drive->phys_disk[i] = dev[j];
+			if (i < nphys_disk)
+				qdepth = min(h->nr_cmds, qdepth +
+				    logical_drive->phys_disk[i]->queue_depth);
+			break;
+		}
+
+		/*
+		 * This can happen if a physical drive is removed and
+		 * the logical drive is degraded.  In that case, the RAID
+		 * map data will refer to a physical disk which isn't actually
+		 * present.  And in that case offload_enabled should already
+		 * be 0, but we'll turn it off here just in case
+		 */
+		if (!logical_drive->phys_disk[i]) {
+			logical_drive->offload_enabled = 0;
+			logical_drive->queue_depth = h->nr_cmds;
+		}
+	}
+	if (nraid_map_entries)
+		/*
+		 * This is correct for reads, too high for full stripe writes,
+		 * way too high for partial stripe writes
+		 */
+		logical_drive->queue_depth = qdepth;
+	else
+		logical_drive->queue_depth = h->nr_cmds;
+}
+
+static void hpsa_update_log_drive_phys_drive_ptrs(struct ctlr_info *h,
+				struct hpsa_scsi_dev_t *dev[], int ndevices)
+{
+	int i;
+
+	for (i = 0; i < ndevices; i++) {
+		if (dev[i]->devtype != TYPE_DISK)
+			continue;
+		if (!is_logical_dev_addr_mode(dev[i]->scsi3addr))
+			continue;
+		hpsa_figure_phys_disk_ptrs(h, dev, ndevices, dev[i]);
+	}
+}
+
 static void adjust_hpsa_scsi_table(struct ctlr_info *h, int hostno,
 	struct hpsa_scsi_dev_t *sd[], int nsds)
 {
@@ -1425,8 +1518,12 @@ static int hpsa_slave_alloc(struct scsi_device *sdev)
 	spin_lock_irqsave(&h->devlock, flags);
 	sd = lookup_hpsa_scsi_dev(h, sdev_channel(sdev),
 		sdev_id(sdev), sdev->lun);
-	if (sd != NULL)
+	if (sd != NULL) {
 		sdev->hostdata = sd;
+		if (sd->queue_depth)
+			scsi_change_queue_depth(sdev, sd->queue_depth);
+		atomic_set(&sd->ioaccel_cmds_out, 0);
+	}
 	spin_unlock_irqrestore(&h->devlock, flags);
 	return 0;
 }
@@ -1679,6 +1776,9 @@ static void complete_scsi_command(struct CommandList *cp)
 	cmd->result = (DID_OK << 16); 		/* host byte */
 	cmd->result |= (COMMAND_COMPLETE << 8);	/* msg byte */
 
+	if (cp->cmd_type == CMD_IOACCEL2 || cp->cmd_type == CMD_IOACCEL1)
+		atomic_dec(&cp->phys_disk->ioaccel_cmds_out);
+
 	if (cp->cmd_type == CMD_IOACCEL2)
 		return process_ioaccel2_completion(h, cp, cmd, dev);
 
@@ -1686,6 +1786,8 @@ static void complete_scsi_command(struct CommandList *cp)
 
 	scsi_set_resid(cmd, ei->ResidualCnt);
 	if (ei->CommandStatus == 0) {
+		if (cp->cmd_type == CMD_IOACCEL1)
+			atomic_dec(&cp->phys_disk->ioaccel_cmds_out);
 		cmd_free(h, cp);
 		cmd->scsi_done(cmd);
 		return;
@@ -2248,6 +2350,34 @@ static int hpsa_get_raid_map(struct ctlr_info *h,
 	return rc;
 }
 
+static int hpsa_bmic_id_physical_device(struct ctlr_info *h,
+		unsigned char scsi3addr[], u16 bmic_device_index,
+		struct bmic_identify_physical_device *buf, size_t bufsize)
+{
+	int rc = IO_OK;
+	struct CommandList *c;
+	struct ErrorInfo *ei;
+
+	c = cmd_alloc(h);
+	rc = fill_cmd(c, BMIC_IDENTIFY_PHYSICAL_DEVICE, h, buf, bufsize,
+		0, RAID_CTLR_LUNID, TYPE_CMD);
+	if (rc)
+		goto out;
+
+	c->Request.CDB[2] = bmic_device_index & 0xff;
+	c->Request.CDB[9] = (bmic_device_index >> 8) & 0xff;
+
+	hpsa_scsi_do_simple_cmd_with_retry(h, c, PCI_DMA_FROMDEVICE);
+	ei = c->err_info;
+	if (ei->CommandStatus != 0 && ei->CommandStatus != CMD_DATA_UNDERRUN) {
+		hpsa_scsi_interpret_error(h, c);
+		rc = -1;
+	}
+out:
+	cmd_free(h, c);
+	return rc;
+}
+
 static int hpsa_vpd_page_supported(struct ctlr_info *h,
 	unsigned char scsi3addr[], u8 page)
 {
@@ -2348,7 +2478,7 @@ static int hpsa_get_device_id(struct ctlr_info *h, unsigned char *scsi3addr,
 }
 
 static int hpsa_scsi_do_report_luns(struct ctlr_info *h, int logical,
-		struct ReportLUNdata *buf, int bufsize,
+		void *buf, int bufsize,
 		int extended_response)
 {
 	int rc = IO_OK;
@@ -2377,11 +2507,13 @@ static int hpsa_scsi_do_report_luns(struct ctlr_info *h, int logical,
 		hpsa_scsi_interpret_error(h, c);
 		rc = -1;
 	} else {
-		if (buf->extended_response_flag != extended_response) {
+		struct ReportLUNdata *rld = buf;
+
+		if (rld->extended_response_flag != extended_response) {
 			dev_err(&h->pdev->dev,
 				"report luns requested format %u, got %u\n",
 				extended_response,
-				buf->extended_response_flag);
+				rld->extended_response_flag);
 			rc = -1;
 		}
 	}
@@ -2391,10 +2523,10 @@ out:
 }
 
 static inline int hpsa_scsi_do_report_phys_luns(struct ctlr_info *h,
-		struct ReportLUNdata *buf,
-		int bufsize, int extended_response)
+		struct ReportExtendedLUNdata *buf, int bufsize)
 {
-	return hpsa_scsi_do_report_luns(h, 0, buf, bufsize, extended_response);
+	return hpsa_scsi_do_report_luns(h, 0, buf, bufsize,
+						HPSA_REPORT_PHYS_EXTENDED);
 }
 
 static inline int hpsa_scsi_do_report_log_luns(struct ctlr_info *h,
@@ -2569,6 +2701,7 @@ static int hpsa_update_device_info(struct ctlr_info *h,
 		this_device->offload_config = 0;
 		this_device->offload_enabled = 0;
 		this_device->volume_offline = 0;
+		this_device->queue_depth = h->nr_cmds;
 	}
 
 	if (is_OBDR_device) {
@@ -2711,7 +2844,6 @@ static int hpsa_get_pdisk_of_ioaccel2(struct ctlr_info *h,
 {
 	struct ReportExtendedLUNdata *physicals = NULL;
 	int responsesize = 24;	/* size of physical extended response */
-	int extended = 2;	/* flag forces reporting 'other dev info'. */
 	int reportsize = sizeof(*physicals) + HPSA_MAX_PHYS_LUN * responsesize;
 	u32 nphysicals = 0;	/* number of reported physical devs */
 	int found = 0;		/* found match (1) or not (0) */
@@ -2758,8 +2890,7 @@ static int hpsa_get_pdisk_of_ioaccel2(struct ctlr_info *h,
 	physicals = kzalloc(reportsize, GFP_KERNEL);
 	if (physicals == NULL)
 		return 0;
-	if (hpsa_scsi_do_report_phys_luns(h, (struct ReportLUNdata *) physicals,
-		reportsize, extended)) {
+	if (hpsa_scsi_do_report_phys_luns(h, physicals, reportsize)) {
 		dev_err(&h->pdev->dev,
 			"Can't lookup %s device handle: report physical LUNs failed.\n",
 			"HP SSD Smart Path");
@@ -2800,34 +2931,20 @@ static int hpsa_get_pdisk_of_ioaccel2(struct ctlr_info *h,
  * Returns 0 on success, -1 otherwise.
  */
 static int hpsa_gather_lun_info(struct ctlr_info *h,
-	int reportphyslunsize, int reportloglunsize,
-	struct ReportLUNdata *physdev, u32 *nphysicals, int *physical_mode,
+	struct ReportExtendedLUNdata *physdev, u32 *nphysicals,
 	struct ReportLUNdata *logdev, u32 *nlogicals)
 {
-	int physical_entry_size = 8;
-
-	*physical_mode = 0;
-
-	/* For I/O accelerator mode we need to read physical device handles */
-	if (h->transMethod & CFGTBL_Trans_io_accel1 ||
-		h->transMethod & CFGTBL_Trans_io_accel2) {
-		*physical_mode = HPSA_REPORT_PHYS_EXTENDED;
-		physical_entry_size = 24;
-	}
-	if (hpsa_scsi_do_report_phys_luns(h, physdev, reportphyslunsize,
-							*physical_mode)) {
+	if (hpsa_scsi_do_report_phys_luns(h, physdev, sizeof(*physdev))) {
 		dev_err(&h->pdev->dev, "report physical LUNs failed.\n");
 		return -1;
 	}
-	*nphysicals = be32_to_cpu(*((__be32 *)physdev->LUNListLength)) /
-							physical_entry_size;
+	*nphysicals = be32_to_cpu(*((__be32 *)physdev->LUNListLength)) / 24;
 	if (*nphysicals > HPSA_MAX_PHYS_LUN) {
-		dev_warn(&h->pdev->dev, "maximum physical LUNs (%d) exceeded."
-			"  %d LUNs ignored.\n", HPSA_MAX_PHYS_LUN,
-			*nphysicals - HPSA_MAX_PHYS_LUN);
+		dev_warn(&h->pdev->dev, "maximum physical LUNs (%d) exceeded. %d LUNs ignored.\n",
+			HPSA_MAX_PHYS_LUN, *nphysicals - HPSA_MAX_PHYS_LUN);
 		*nphysicals = HPSA_MAX_PHYS_LUN;
 	}
-	if (hpsa_scsi_do_report_log_luns(h, logdev, reportloglunsize)) {
+	if (hpsa_scsi_do_report_log_luns(h, logdev, sizeof(*logdev))) {
 		dev_err(&h->pdev->dev, "report logical LUNs failed.\n");
 		return -1;
 	}
@@ -2900,6 +3017,33 @@ static int hpsa_hba_mode_enabled(struct ctlr_info *h)
 	return hba_mode_enabled;
 }
 
+/* get physical drive ioaccel handle and queue depth */
+static void hpsa_get_ioaccel_drive_info(struct ctlr_info *h,
+		struct hpsa_scsi_dev_t *dev,
+		u8 *lunaddrbytes,
+		struct bmic_identify_physical_device *id_phys)
+{
+	int rc;
+	struct ext_report_lun_entry *rle =
+		(struct ext_report_lun_entry *) lunaddrbytes;
+
+	dev->ioaccel_handle = rle->ioaccel_handle;
+	memset(id_phys, 0, sizeof(*id_phys));
+	rc = hpsa_bmic_id_physical_device(h, lunaddrbytes,
+			GET_BMIC_DRIVE_NUMBER(lunaddrbytes), id_phys,
+			sizeof(*id_phys));
+	if (!rc)
+		/* Reserve space for FW operations */
+#define DRIVE_CMDS_RESERVED_FOR_FW 2
+#define DRIVE_QUEUE_DEPTH 7
+		dev->queue_depth =
+			le16_to_cpu(id_phys->current_queue_depth_limit) -
+				DRIVE_CMDS_RESERVED_FOR_FW;
+	else
+		dev->queue_depth = DRIVE_QUEUE_DEPTH; /* conservative */
+	atomic_set(&dev->ioaccel_cmds_out, 0);
+}
+
 static void hpsa_update_scsi_devices(struct ctlr_info *h, int hostno)
 {
 	/* the idea here is we could get notified
@@ -2914,9 +3058,9 @@ static void hpsa_update_scsi_devices(struct ctlr_info *h, int hostno)
 	 */
 	struct ReportExtendedLUNdata *physdev_list = NULL;
 	struct ReportLUNdata *logdev_list = NULL;
+	struct bmic_identify_physical_device *id_phys = NULL;
 	u32 nphysicals = 0;
 	u32 nlogicals = 0;
-	int physical_mode = 0;
 	u32 ndev_allocated = 0;
 	struct hpsa_scsi_dev_t **currentsd, *this_device, *tmpdevice;
 	int ncurrent = 0;
@@ -2929,8 +3073,10 @@ static void hpsa_update_scsi_devices(struct ctlr_info *h, int hostno)
 	physdev_list = kzalloc(sizeof(*physdev_list), GFP_KERNEL);
 	logdev_list = kzalloc(sizeof(*logdev_list), GFP_KERNEL);
 	tmpdevice = kzalloc(sizeof(*tmpdevice), GFP_KERNEL);
+	id_phys = kzalloc(sizeof(*id_phys), GFP_KERNEL);
 
-	if (!currentsd || !physdev_list || !logdev_list || !tmpdevice) {
+	if (!currentsd || !physdev_list || !logdev_list ||
+		!tmpdevice || !id_phys) {
 		dev_err(&h->pdev->dev, "out of memory\n");
 		goto out;
 	}
@@ -2947,10 +3093,8 @@ static void hpsa_update_scsi_devices(struct ctlr_info *h, int hostno)
 
 	h->hba_mode_enabled = rescan_hba_mode;
 
-	if (hpsa_gather_lun_info(h,
-			sizeof(*physdev_list), sizeof(*logdev_list),
-			(struct ReportLUNdata *) physdev_list, &nphysicals,
-			&physical_mode, logdev_list, &nlogicals))
+	if (hpsa_gather_lun_info(h, physdev_list, &nphysicals,
+			logdev_list, &nlogicals))
 		goto out;
 
 	/* We might see up to the maximum number of logical and physical disks
@@ -3047,10 +3191,11 @@ static void hpsa_update_scsi_devices(struct ctlr_info *h, int hostno)
 				ncurrent++;
 				break;
 			}
-			if (physical_mode == HPSA_REPORT_PHYS_EXTENDED) {
-				memcpy(&this_device->ioaccel_handle,
-					&lunaddrbytes[20],
-					sizeof(this_device->ioaccel_handle));
+			if (h->transMethod & CFGTBL_Trans_io_accel1 ||
+				h->transMethod & CFGTBL_Trans_io_accel2) {
+				hpsa_get_ioaccel_drive_info(h, this_device,
+							lunaddrbytes, id_phys);
+				atomic_set(&this_device->ioaccel_cmds_out, 0);
 				ncurrent++;
 			}
 			break;
@@ -3074,6 +3219,7 @@ static void hpsa_update_scsi_devices(struct ctlr_info *h, int hostno)
 		if (ncurrent >= HPSA_MAX_DEVICES)
 			break;
 	}
+	hpsa_update_log_drive_phys_drive_ptrs(h, currentsd, ncurrent);
 	adjust_hpsa_scsi_table(h, hostno, currentsd, ncurrent);
 out:
 	kfree(tmpdevice);
@@ -3082,6 +3228,7 @@ out:
 	kfree(currentsd);
 	kfree(physdev_list);
 	kfree(logdev_list);
+	kfree(id_phys);
 }
 
 /*
@@ -3197,7 +3344,7 @@ static int fixup_ioaccel_cdb(u8 *cdb, int *cdb_len)
 
 static int hpsa_scsi_ioaccel1_queue_command(struct ctlr_info *h,
 	struct CommandList *c, u32 ioaccel_handle, u8 *cdb, int cdb_len,
-	u8 *scsi3addr)
+	u8 *scsi3addr, struct hpsa_scsi_dev_t *phys_disk)
 {
 	struct scsi_cmnd *cmd = c->scsi_cmd;
 	struct io_accel1_cmd *cp = &h->ioaccel_cmd_pool[c->cmdindex];
@@ -3210,13 +3357,17 @@ static int hpsa_scsi_ioaccel1_queue_command(struct ctlr_info *h,
 	u32 control = IOACCEL1_CONTROL_SIMPLEQUEUE;
 
 	/* TODO: implement chaining support */
-	if (scsi_sg_count(cmd) > h->ioaccel_maxsg)
+	if (scsi_sg_count(cmd) > h->ioaccel_maxsg) {
+		atomic_dec(&phys_disk->ioaccel_cmds_out);
 		return IO_ACCEL_INELIGIBLE;
+	}
 
 	BUG_ON(cmd->cmd_len > IOACCEL1_IOFLAGS_CDBLEN_MAX);
 
-	if (fixup_ioaccel_cdb(cdb, &cdb_len))
+	if (fixup_ioaccel_cdb(cdb, &cdb_len)) {
+		atomic_dec(&phys_disk->ioaccel_cmds_out);
 		return IO_ACCEL_INELIGIBLE;
+	}
 
 	c->cmd_type = CMD_IOACCEL1;
 
@@ -3226,8 +3377,10 @@ static int hpsa_scsi_ioaccel1_queue_command(struct ctlr_info *h,
 	BUG_ON(c->busaddr & 0x0000007F);
 
 	use_sg = scsi_dma_map(cmd);
-	if (use_sg < 0)
+	if (use_sg < 0) {
+		atomic_dec(&phys_disk->ioaccel_cmds_out);
 		return use_sg;
+	}
 
 	if (use_sg) {
 		curr_sg = cp->SG;
@@ -3286,8 +3439,10 @@ static int hpsa_scsi_ioaccel_direct_map(struct ctlr_info *h,
 	struct scsi_cmnd *cmd = c->scsi_cmd;
 	struct hpsa_scsi_dev_t *dev = cmd->device->hostdata;
 
+	c->phys_disk = dev;
+
 	return hpsa_scsi_ioaccel_queue_command(h, c, dev->ioaccel_handle,
-		cmd->cmnd, cmd->cmd_len, dev->scsi3addr);
+		cmd->cmnd, cmd->cmd_len, dev->scsi3addr, dev);
 }
 
 /*
@@ -3351,7 +3506,7 @@ static void set_encrypt_ioaccel2(struct ctlr_info *h,
 
 static int hpsa_scsi_ioaccel2_queue_command(struct ctlr_info *h,
 	struct CommandList *c, u32 ioaccel_handle, u8 *cdb, int cdb_len,
-	u8 *scsi3addr)
+	u8 *scsi3addr, struct hpsa_scsi_dev_t *phys_disk)
 {
 	struct scsi_cmnd *cmd = c->scsi_cmd;
 	struct io_accel2_cmd *cp = &h->ioaccel2_cmd_pool[c->cmdindex];
@@ -3362,11 +3517,16 @@ static int hpsa_scsi_ioaccel2_queue_command(struct ctlr_info *h,
 	u32 len;
 	u32 total_len = 0;
 
-	if (scsi_sg_count(cmd) > h->ioaccel_maxsg)
+	if (scsi_sg_count(cmd) > h->ioaccel_maxsg) {
+		atomic_dec(&phys_disk->ioaccel_cmds_out);
 		return IO_ACCEL_INELIGIBLE;
+	}
 
-	if (fixup_ioaccel_cdb(cdb, &cdb_len))
+	if (fixup_ioaccel_cdb(cdb, &cdb_len)) {
+		atomic_dec(&phys_disk->ioaccel_cmds_out);
 		return IO_ACCEL_INELIGIBLE;
+	}
+
 	c->cmd_type = CMD_IOACCEL2;
 	/* Adjust the DMA address to point to the accelerated command buffer */
 	c->busaddr = (u32) h->ioaccel2_cmd_pool_dhandle +
@@ -3377,8 +3537,10 @@ static int hpsa_scsi_ioaccel2_queue_command(struct ctlr_info *h,
 	cp->IU_type = IOACCEL2_IU_TYPE;
 
 	use_sg = scsi_dma_map(cmd);
-	if (use_sg < 0)
+	if (use_sg < 0) {
+		atomic_dec(&phys_disk->ioaccel_cmds_out);
 		return use_sg;
+	}
 
 	if (use_sg) {
 		BUG_ON(use_sg > IOACCEL2_MAXSGENTRIES);
@@ -3444,14 +3606,22 @@ static int hpsa_scsi_ioaccel2_queue_command(struct ctlr_info *h,
  */
 static int hpsa_scsi_ioaccel_queue_command(struct ctlr_info *h,
 	struct CommandList *c, u32 ioaccel_handle, u8 *cdb, int cdb_len,
-	u8 *scsi3addr)
+	u8 *scsi3addr, struct hpsa_scsi_dev_t *phys_disk)
 {
+	/* Try to honor the device's queue depth */
+	if (atomic_inc_return(&phys_disk->ioaccel_cmds_out) >
+					phys_disk->queue_depth) {
+		atomic_dec(&phys_disk->ioaccel_cmds_out);
+		return IO_ACCEL_INELIGIBLE;
+	}
 	if (h->transMethod & CFGTBL_Trans_io_accel1)
 		return hpsa_scsi_ioaccel1_queue_command(h, c, ioaccel_handle,
-						cdb, cdb_len, scsi3addr);
+						cdb, cdb_len, scsi3addr,
+						phys_disk);
 	else
 		return hpsa_scsi_ioaccel2_queue_command(h, c, ioaccel_handle,
-						cdb, cdb_len, scsi3addr);
+						cdb, cdb_len, scsi3addr,
+						phys_disk);
 }
 
 static void raid_map_helper(struct raid_map_data *map,
@@ -3755,6 +3925,8 @@ static int hpsa_scsi_ioaccel_raid_map(struct ctlr_info *h,
 		return IO_ACCEL_INELIGIBLE;
 	}
 
+	c->phys_disk = dev->phys_disk[map_index];
+
 	disk_handle = dd[map_index].ioaccel_handle;
 	disk_block = le64_to_cpu(map->disk_starting_blk) +
 			first_row * le16_to_cpu(map->strip_size) +
@@ -3802,7 +3974,8 @@ static int hpsa_scsi_ioaccel_raid_map(struct ctlr_info *h,
 		cdb_len = 10;
 	}
 	return hpsa_scsi_ioaccel_queue_command(h, c, disk_handle, cdb, cdb_len,
-						dev->scsi3addr);
+						dev->scsi3addr,
+						dev->phys_disk[map_index]);
 }
 
 /* Submit commands down the "normal" RAID stack path */
@@ -4016,15 +4189,17 @@ static void hpsa_scan_start(struct Scsi_Host *sh)
 
 static int hpsa_change_queue_depth(struct scsi_device *sdev, int qdepth)
 {
-	struct ctlr_info *h = sdev_to_hba(sdev);
+	struct hpsa_scsi_dev_t *logical_drive = sdev->hostdata;
+
+	if (!logical_drive)
+		return -ENODEV;
 
 	if (qdepth < 1)
 		qdepth = 1;
-	else
-		if (qdepth > h->nr_cmds)
-			qdepth = h->nr_cmds;
-	scsi_change_queue_depth(sdev, qdepth);
-	return sdev->queue_depth;
+	else if (qdepth > logical_drive->queue_depth)
+		qdepth = logical_drive->queue_depth;
+
+	return scsi_change_queue_depth(sdev, qdepth);
 }
 
 static int hpsa_scan_finished(struct Scsi_Host *sh,
@@ -4068,10 +4243,7 @@ static int hpsa_register_scsi(struct ctlr_info *h)
 			HPSA_CMDS_RESERVED_FOR_ABORTS -
 			HPSA_CMDS_RESERVED_FOR_DRIVER -
 			HPSA_MAX_CONCURRENT_PASSTHRUS;
-	if (h->hba_mode_enabled)
-		sh->cmd_per_lun = 7;
-	else
-		sh->cmd_per_lun = sh->can_queue;
+	sh->cmd_per_lun = sh->can_queue;
 	sh->sg_tablesize = h->maxsgentries;
 	h->scsi_host = sh;
 	sh->hostdata[0] = (unsigned long) h;
@@ -5090,6 +5262,16 @@ static int fill_cmd(struct CommandList *c, u8 cmd, struct ctlr_info *h,
 			c->Request.CDB[7] = (size >> 16) & 0xFF;
 			c->Request.CDB[8] = (size >> 8) & 0xFF;
 			break;
+		case BMIC_IDENTIFY_PHYSICAL_DEVICE:
+			c->Request.CDBLen = 10;
+			c->Request.type_attr_dir =
+				TYPE_ATTR_DIR(cmd_type, ATTR_SIMPLE, XFER_READ);
+			c->Request.Timeout = 0;
+			c->Request.CDB[0] = BMIC_READ;
+			c->Request.CDB[6] = BMIC_IDENTIFY_PHYSICAL_DEVICE;
+			c->Request.CDB[7] = (size >> 16) & 0xFF;
+			c->Request.CDB[8] = (size >> 8) & 0XFF;
+			break;
 		default:
 			dev_warn(&h->pdev->dev, "unknown command 0x%c\n", cmd);
 			BUG();
diff --git a/drivers/scsi/hpsa.h b/drivers/scsi/hpsa.h
index a0f4268..d0fb854 100644
--- a/drivers/scsi/hpsa.h
+++ b/drivers/scsi/hpsa.h
@@ -46,6 +46,11 @@ struct hpsa_scsi_dev_t {
 	unsigned char model[16];        /* bytes 16-31 of inquiry data */
 	unsigned char raid_level;	/* from inquiry page 0xC1 */
 	unsigned char volume_offline;	/* discovered via TUR or VPD */
+	u16 queue_depth;		/* max queue_depth for this device */
+	atomic_t ioaccel_cmds_out;	/* Only used for physical devices
+					 * counts commands sent to physical
+					 * device via "ioaccel" path.
+					 */
 	u32 ioaccel_handle;
 	int offload_config;		/* I/O accel RAID offload configured */
 	int offload_enabled;		/* I/O accel RAID offload enabled */
@@ -54,6 +59,15 @@ struct hpsa_scsi_dev_t {
 					 */
 	struct raid_map_data raid_map;	/* I/O accelerator RAID map */
 
+	/*
+	 * Pointers from logical drive map indices to the phys drives that
+	 * make those logical drives.  Note, multiple logical drives may
+	 * share physical drives.  You can have for instance 5 physical
+	 * drives with 3 logical drives each using those same 5 physical
+	 * disks. We need these pointers for counting i/o's out to physical
+	 * devices in order to honor physical device queue depth limits.
+	 */
+	struct hpsa_scsi_dev_t *phys_disk[RAID_MAP_MAX_ENTRIES];
 };
 
 struct reply_queue_buffer {
diff --git a/drivers/scsi/hpsa_cmd.h b/drivers/scsi/hpsa_cmd.h
index 3f2f0af..4726dbb 100644
--- a/drivers/scsi/hpsa_cmd.h
+++ b/drivers/scsi/hpsa_cmd.h
@@ -240,6 +240,10 @@ struct ReportLUNdata {
 
 struct ext_report_lun_entry {
 	u8 lunid[8];
+#define GET_BMIC_BUS(lunid) ((lunid)[7] & 0x3F)
+#define GET_BMIC_LEVEL_TWO_TARGET(lunid) ((lunid)[6])
+#define GET_BMIC_DRIVE_NUMBER(lunid) (((GET_BMIC_BUS((lunid)) - 1) << 8) + \
+			GET_BMIC_LEVEL_TWO_TARGET((lunid)))
 	u8 wwid[8];
 	u8 device_type;
 	u8 device_flags;
@@ -268,6 +272,7 @@ struct SenseSubsystem_info {
 #define HPSA_CACHE_FLUSH 0x01	/* C2 was already being used by HPSA */
 #define BMIC_FLASH_FIRMWARE 0xF7
 #define BMIC_SENSE_CONTROLLER_PARAMETERS 0x64
+#define BMIC_IDENTIFY_PHYSICAL_DEVICE 0x15
 
 /* Command List Structure */
 union SCSI3Addr {
@@ -405,6 +410,17 @@ struct CommandList {
 	struct completion *waiting;
 	void   *scsi_cmd;
 	struct work_struct work;
+
+	/*
+	 * For commands using either of the two "ioaccel" paths to
+	 * bypass the RAID stack and go directly to the physical disk
+	 * phys_disk is a pointer to the hpsa_scsi_dev_t to which the
+	 * i/o is destined.  We need to store that here because the command
+	 * may potentially encounter TASK SET FULL and need to be resubmitted
+	 * For "normal" i/o's not using the "ioaccel" paths, phys_disk is
+	 * not used.
+	 */
+	struct hpsa_scsi_dev_t *phys_disk;
 } __aligned(COMMANDLIST_ALIGNMENT);
 
 /* Max S/G elements in I/O accelerator command */
@@ -641,5 +657,137 @@ struct hpsa_pci_info {
 	u32		board_id;
 };
 
+struct bmic_identify_physical_device {
+	u8    scsi_bus;          /* SCSI Bus number on controller */
+	u8    scsi_id;           /* SCSI ID on this bus */
+	__le16 block_size;	     /* sector size in bytes */
+	__le32 total_blocks;	     /* number for sectors on drive */
+	__le32 reserved_blocks;   /* controller reserved (RIS) */
+	u8    model[40];         /* Physical Drive Model */
+	u8    serial_number[40]; /* Drive Serial Number */
+	u8    firmware_revision[8]; /* drive firmware revision */
+	u8    scsi_inquiry_bits; /* inquiry byte 7 bits */
+	u8    compaq_drive_stamp; /* 0 means drive not stamped */
+	u8    last_failure_reason;
+#define BMIC_LAST_FAILURE_TOO_SMALL_IN_LOAD_CONFIG		0x01
+#define BMIC_LAST_FAILURE_ERROR_ERASING_RIS			0x02
+#define BMIC_LAST_FAILURE_ERROR_SAVING_RIS			0x03
+#define BMIC_LAST_FAILURE_FAIL_DRIVE_COMMAND			0x04
+#define BMIC_LAST_FAILURE_MARK_BAD_FAILED			0x05
+#define BMIC_LAST_FAILURE_MARK_BAD_FAILED_IN_FINISH_REMAP	0x06
+#define BMIC_LAST_FAILURE_TIMEOUT				0x07
+#define BMIC_LAST_FAILURE_AUTOSENSE_FAILED			0x08
+#define BMIC_LAST_FAILURE_MEDIUM_ERROR_1			0x09
+#define BMIC_LAST_FAILURE_MEDIUM_ERROR_2			0x0a
+#define BMIC_LAST_FAILURE_NOT_READY_BAD_SENSE			0x0b
+#define BMIC_LAST_FAILURE_NOT_READY				0x0c
+#define BMIC_LAST_FAILURE_HARDWARE_ERROR			0x0d
+#define BMIC_LAST_FAILURE_ABORTED_COMMAND			0x0e
+#define BMIC_LAST_FAILURE_WRITE_PROTECTED			0x0f
+#define BMIC_LAST_FAILURE_SPIN_UP_FAILURE_IN_RECOVER		0x10
+#define BMIC_LAST_FAILURE_REBUILD_WRITE_ERROR			0x11
+#define BMIC_LAST_FAILURE_TOO_SMALL_IN_HOT_PLUG			0x12
+#define BMIC_LAST_FAILURE_BUS_RESET_RECOVERY_ABORTED		0x13
+#define BMIC_LAST_FAILURE_REMOVED_IN_HOT_PLUG			0x14
+#define BMIC_LAST_FAILURE_INIT_REQUEST_SENSE_FAILED		0x15
+#define BMIC_LAST_FAILURE_INIT_START_UNIT_FAILED		0x16
+#define BMIC_LAST_FAILURE_INQUIRY_FAILED			0x17
+#define BMIC_LAST_FAILURE_NON_DISK_DEVICE			0x18
+#define BMIC_LAST_FAILURE_READ_CAPACITY_FAILED			0x19
+#define BMIC_LAST_FAILURE_INVALID_BLOCK_SIZE			0x1a
+#define BMIC_LAST_FAILURE_HOT_PLUG_REQUEST_SENSE_FAILED		0x1b
+#define BMIC_LAST_FAILURE_HOT_PLUG_START_UNIT_FAILED		0x1c
+#define BMIC_LAST_FAILURE_WRITE_ERROR_AFTER_REMAP		0x1d
+#define BMIC_LAST_FAILURE_INIT_RESET_RECOVERY_ABORTED		0x1e
+#define BMIC_LAST_FAILURE_DEFERRED_WRITE_ERROR			0x1f
+#define BMIC_LAST_FAILURE_MISSING_IN_SAVE_RIS			0x20
+#define BMIC_LAST_FAILURE_WRONG_REPLACE				0x21
+#define BMIC_LAST_FAILURE_GDP_VPD_INQUIRY_FAILED		0x22
+#define BMIC_LAST_FAILURE_GDP_MODE_SENSE_FAILED			0x23
+#define BMIC_LAST_FAILURE_DRIVE_NOT_IN_48BIT_MODE		0x24
+#define BMIC_LAST_FAILURE_DRIVE_TYPE_MIX_IN_HOT_PLUG		0x25
+#define BMIC_LAST_FAILURE_DRIVE_TYPE_MIX_IN_LOAD_CFG		0x26
+#define BMIC_LAST_FAILURE_PROTOCOL_ADAPTER_FAILED		0x27
+#define BMIC_LAST_FAILURE_FAULTY_ID_BAY_EMPTY			0x28
+#define BMIC_LAST_FAILURE_FAULTY_ID_BAY_OCCUPIED		0x29
+#define BMIC_LAST_FAILURE_FAULTY_ID_INVALID_BAY			0x2a
+#define BMIC_LAST_FAILURE_WRITE_RETRIES_FAILED			0x2b
+
+#define BMIC_LAST_FAILURE_SMART_ERROR_REPORTED			0x37
+#define BMIC_LAST_FAILURE_PHY_RESET_FAILED			0x38
+#define BMIC_LAST_FAILURE_ONLY_ONE_CTLR_CAN_SEE_DRIVE		0x40
+#define BMIC_LAST_FAILURE_KC_VOLUME_FAILED			0x41
+#define BMIC_LAST_FAILURE_UNEXPECTED_REPLACEMENT		0x42
+#define BMIC_LAST_FAILURE_OFFLINE_ERASE				0x80
+#define BMIC_LAST_FAILURE_OFFLINE_TOO_SMALL			0x81
+#define BMIC_LAST_FAILURE_OFFLINE_DRIVE_TYPE_MIX		0x82
+#define BMIC_LAST_FAILURE_OFFLINE_ERASE_COMPLETE		0x83
+
+	u8     flags;
+	u8     more_flags;
+	u8     scsi_lun;          /* SCSI LUN for phys drive */
+	u8     yet_more_flags;
+	u8     even_more_flags;
+	__le32 spi_speed_rules;/* SPI Speed data:Ultra disable diagnose */
+	u8     phys_connector[2];         /* connector number on controller */
+	u8     phys_box_on_bus;  /* phys enclosure this drive resides */
+	u8     phys_bay_in_box;  /* phys drv bay this drive resides */
+	__le32 rpm;              /* Drive rotational speed in rpm */
+	u8     device_type;       /* type of drive */
+	u8     sata_version;     /* only valid when drive_type is SATA */
+	__le64 big_total_block_count;
+	__le64 ris_starting_lba;
+	__le32 ris_size;
+	u8     wwid[20];
+	u8     controller_phy_map[32];
+	__le16 phy_count;
+	u8     phy_connected_dev_type[256];
+	u8     phy_to_drive_bay_num[256];
+	__le16 phy_to_attached_dev_index[256];
+	u8     box_index;
+	u8     reserved;
+	__le16 extra_physical_drive_flags;
+#define BMIC_PHYS_DRIVE_SUPPORTS_GAS_GAUGE(idphydrv) \
+	(idphydrv->extra_physical_drive_flags & (1 << 10))
+	u8     negotiated_link_rate[256];
+	u8     phy_to_phy_map[256];
+	u8     redundant_path_present_map;
+	u8     redundant_path_failure_map;
+	u8     active_path_number;
+	__le16 alternate_paths_phys_connector[8];
+	u8     alternate_paths_phys_box_on_port[8];
+	u8     multi_lun_device_lun_count;
+	u8     minimum_good_fw_revision[8];
+	u8     unique_inquiry_bytes[20];
+	u8     current_temperature_degreesC;
+	u8     temperature_threshold_degreesC;
+	u8     max_temperature_degreesC;
+	u8     logical_blocks_per_phys_block_exp; /* phyblocksize = 512*2^exp */
+	__le16 current_queue_depth_limit;
+	u8     switch_name[10];
+	__le16 switch_port;
+	u8     alternate_paths_switch_name[40];
+	u8     alternate_paths_switch_port[8];
+	__le16 power_on_hours; /* valid only if gas gauge supported */
+	__le16 percent_endurance_used; /* valid only if gas gauge supported. */
+#define BMIC_PHYS_DRIVE_SSD_WEAROUT(idphydrv) \
+	((idphydrv->percent_endurance_used & 0x80) || \
+	 (idphydrv->percent_endurance_used > 10000))
+	u8     drive_authentication;
+#define BMIC_PHYS_DRIVE_AUTHENTICATED(idphydrv) \
+	(idphydrv->drive_authentication == 0x80)
+	u8     smart_carrier_authentication;
+#define BMIC_SMART_CARRIER_AUTHENTICATION_SUPPORTED(idphydrv) \
+	(idphydrv->smart_carrier_authentication != 0x0)
+#define BMIC_SMART_CARRIER_AUTHENTICATED(idphydrv) \
+	(idphydrv->smart_carrier_authentication == 0x01)
+	u8     smart_carrier_app_fw_version;
+	u8     smart_carrier_bootloader_fw_version;
+	u8     encryption_key_name[64];
+	__le32 misc_drive_flags;
+	__le16 dek_index;
+	u8     padding[112];
+};
+
 #pragma pack()
 #endif /* HPSA_CMD_H */