/*
 *    driver for Microsemi PQI-based storage controllers
 *    Copyright (c) 2016-2017 Microsemi Corporation
 *    Copyright (c) 2016 PMC-Sierra, Inc.
 *
 *    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; version 2 of the License.
 *
 *    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, GOOD TITLE or
 *    NON INFRINGEMENT.  See the GNU General Public License for more details.
 *
 *    Questions/Comments/Bugfixes to esc.storagedev@microsemi.com
 *
 */

#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/pci.h>
#include <linux/delay.h>
#include <linux/interrupt.h>
#include <linux/sched.h>
#include <linux/rtc.h>
#include <linux/bcd.h>
#include <linux/reboot.h>
#include <linux/cciss_ioctl.h>
#include <linux/blk-mq-pci.h>
#include <scsi/scsi_host.h>
#include <scsi/scsi_cmnd.h>
#include <scsi/scsi_device.h>
#include <scsi/scsi_eh.h>
#include <scsi/scsi_transport_sas.h>
#include <asm/unaligned.h>
#include "smartpqi.h"
#include "smartpqi_sis.h"

#if !defined(BUILD_TIMESTAMP)
#define BUILD_TIMESTAMP
#endif

#define DRIVER_VERSION          "1.0.4-100"
#define DRIVER_MAJOR            1
#define DRIVER_MINOR            0
#define DRIVER_RELEASE          4
#define DRIVER_REVISION         100

#define DRIVER_NAME             "Microsemi PQI Driver (v" \
                                DRIVER_VERSION BUILD_TIMESTAMP ")"
#define DRIVER_NAME_SHORT       "smartpqi"

#define PQI_EXTRA_SGL_MEMORY    (12 * sizeof(struct pqi_sg_descriptor))

MODULE_AUTHOR("Microsemi");
MODULE_DESCRIPTION("Driver for Microsemi Smart Family Controller version "
        DRIVER_VERSION);
MODULE_SUPPORTED_DEVICE("Microsemi Smart Family Controllers");
MODULE_VERSION(DRIVER_VERSION);
MODULE_LICENSE("GPL");

static void pqi_take_ctrl_offline(struct pqi_ctrl_info *ctrl_info);
static void pqi_ctrl_offline_worker(struct work_struct *work);
static void pqi_retry_raid_bypass_requests(struct pqi_ctrl_info *ctrl_info);
static int pqi_scan_scsi_devices(struct pqi_ctrl_info *ctrl_info);
static void pqi_scan_start(struct Scsi_Host *shost);
static void pqi_start_io(struct pqi_ctrl_info *ctrl_info,
        struct pqi_queue_group *queue_group, enum pqi_io_path path,
        struct pqi_io_request *io_request);
static int pqi_submit_raid_request_synchronous(struct pqi_ctrl_info *ctrl_info,
        struct pqi_iu_header *request, unsigned int flags,
        struct pqi_raid_error_info *error_info, unsigned long timeout_msecs);
static int pqi_aio_submit_io(struct pqi_ctrl_info *ctrl_info,
        struct scsi_cmnd *scmd, u32 aio_handle, u8 *cdb,
        unsigned int cdb_length, struct pqi_queue_group *queue_group,
        struct pqi_encryption_info *encryption_info, bool raid_bypass);

/* for flags argument to pqi_submit_raid_request_synchronous() */
#define PQI_SYNC_FLAGS_INTERRUPTABLE    0x1

static struct scsi_transport_template *pqi_sas_transport_template;

static atomic_t pqi_controller_count = ATOMIC_INIT(0);

enum pqi_lockup_action {
        NONE,
        REBOOT,
        PANIC
};

static enum pqi_lockup_action pqi_lockup_action = NONE;

static struct {
        enum pqi_lockup_action  action;
        char                    *name;
} pqi_lockup_actions[] = {
        {
                .action = NONE,
                .name = "none",
        },
        {
                .action = REBOOT,
                .name = "reboot",
        },
        {
                .action = PANIC,
                .name = "panic",
        },
};

static unsigned int pqi_supported_event_types[] = {
        PQI_EVENT_TYPE_HOTPLUG,
        PQI_EVENT_TYPE_HARDWARE,
        PQI_EVENT_TYPE_PHYSICAL_DEVICE,
        PQI_EVENT_TYPE_LOGICAL_DEVICE,
        PQI_EVENT_TYPE_AIO_STATE_CHANGE,
        PQI_EVENT_TYPE_AIO_CONFIG_CHANGE,
};

static int pqi_disable_device_id_wildcards;
module_param_named(disable_device_id_wildcards,
        pqi_disable_device_id_wildcards, int, 0644);
MODULE_PARM_DESC(disable_device_id_wildcards,
        "Disable device ID wildcards.");

static int pqi_disable_heartbeat;
module_param_named(disable_heartbeat,
        pqi_disable_heartbeat, int, 0644);
MODULE_PARM_DESC(disable_heartbeat,
        "Disable heartbeat.");

static int pqi_disable_ctrl_shutdown;
module_param_named(disable_ctrl_shutdown,
        pqi_disable_ctrl_shutdown, int, 0644);
MODULE_PARM_DESC(disable_ctrl_shutdown,
        "Disable controller shutdown when controller locked up.");

static char *pqi_lockup_action_param;
module_param_named(lockup_action,
        pqi_lockup_action_param, charp, 0644);
MODULE_PARM_DESC(lockup_action, "Action to take when controller locked up.\n"
        "\t\tSupported: none, reboot, panic\n"
        "\t\tDefault: none");

static char *raid_levels[] = {
        "RAID-0",
        "RAID-4",
        "RAID-1(1+0)",
        "RAID-5",
        "RAID-5+1",
        "RAID-ADG",
        "RAID-1(ADM)",
};

static char *pqi_raid_level_to_string(u8 raid_level)
{
        if (raid_level < ARRAY_SIZE(raid_levels))
                return raid_levels[raid_level];

        return "RAID UNKNOWN";
}

#define SA_RAID_0               0
#define SA_RAID_4               1
#define SA_RAID_1               2       /* also used for RAID 10 */
#define SA_RAID_5               3       /* also used for RAID 50 */
#define SA_RAID_51              4
#define SA_RAID_6               5       /* also used for RAID 60 */
#define SA_RAID_ADM             6       /* also used for RAID 1+0 ADM */
#define SA_RAID_MAX             SA_RAID_ADM
#define SA_RAID_UNKNOWN         0xff

static inline void pqi_scsi_done(struct scsi_cmnd *scmd)
{
        pqi_prep_for_scsi_done(scmd);
        scmd->scsi_done(scmd);
}

static inline bool pqi_scsi3addr_equal(u8 *scsi3addr1, u8 *scsi3addr2)
{
        return memcmp(scsi3addr1, scsi3addr2, 8) == 0;
}

static inline struct pqi_ctrl_info *shost_to_hba(struct Scsi_Host *shost)
{
        void *hostdata = shost_priv(shost);

        return *((struct pqi_ctrl_info **)hostdata);
}

static inline bool pqi_is_logical_device(struct pqi_scsi_dev *device)
{
        return !device->is_physical_device;
}

static inline bool pqi_is_external_raid_addr(u8 *scsi3addr)
{
        return scsi3addr[2] != 0;
}

static inline bool pqi_ctrl_offline(struct pqi_ctrl_info *ctrl_info)
{
        return !ctrl_info->controller_online;
}

static inline void pqi_check_ctrl_health(struct pqi_ctrl_info *ctrl_info)
{
        if (ctrl_info->controller_online)
                if (!sis_is_firmware_running(ctrl_info))
                        pqi_take_ctrl_offline(ctrl_info);
}

static inline bool pqi_is_hba_lunid(u8 *scsi3addr)
{
        return pqi_scsi3addr_equal(scsi3addr, RAID_CTLR_LUNID);
}

static inline enum pqi_ctrl_mode pqi_get_ctrl_mode(
        struct pqi_ctrl_info *ctrl_info)
{
        return sis_read_driver_scratch(ctrl_info);
}

static inline void pqi_save_ctrl_mode(struct pqi_ctrl_info *ctrl_info,
        enum pqi_ctrl_mode mode)
{
        sis_write_driver_scratch(ctrl_info, mode);
}

static inline void pqi_ctrl_block_requests(struct pqi_ctrl_info *ctrl_info)
{
        ctrl_info->block_requests = true;
        scsi_block_requests(ctrl_info->scsi_host);
}

static inline void pqi_ctrl_unblock_requests(struct pqi_ctrl_info *ctrl_info)
{
        ctrl_info->block_requests = false;
        wake_up_all(&ctrl_info->block_requests_wait);
        pqi_retry_raid_bypass_requests(ctrl_info);
        scsi_unblock_requests(ctrl_info->scsi_host);
}

static inline bool pqi_ctrl_blocked(struct pqi_ctrl_info *ctrl_info)
{
        return ctrl_info->block_requests;
}

static unsigned long pqi_wait_if_ctrl_blocked(struct pqi_ctrl_info *ctrl_info,
        unsigned long timeout_msecs)
{
        unsigned long remaining_msecs;

        if (!pqi_ctrl_blocked(ctrl_info))
                return timeout_msecs;

        atomic_inc(&ctrl_info->num_blocked_threads);

        if (timeout_msecs == NO_TIMEOUT) {
                wait_event(ctrl_info->block_requests_wait,
                        !pqi_ctrl_blocked(ctrl_info));
                remaining_msecs = timeout_msecs;
        } else {
                unsigned long remaining_jiffies;

                remaining_jiffies =
                        wait_event_timeout(ctrl_info->block_requests_wait,
                                !pqi_ctrl_blocked(ctrl_info),
                                msecs_to_jiffies(timeout_msecs));
                remaining_msecs = jiffies_to_msecs(remaining_jiffies);
        }

        atomic_dec(&ctrl_info->num_blocked_threads);

        return remaining_msecs;
}

static inline void pqi_ctrl_busy(struct pqi_ctrl_info *ctrl_info)
{
        atomic_inc(&ctrl_info->num_busy_threads);
}

static inline void pqi_ctrl_unbusy(struct pqi_ctrl_info *ctrl_info)
{
        atomic_dec(&ctrl_info->num_busy_threads);
}

static inline void pqi_ctrl_wait_until_quiesced(struct pqi_ctrl_info *ctrl_info)
{
        while (atomic_read(&ctrl_info->num_busy_threads) >
                atomic_read(&ctrl_info->num_blocked_threads))
                usleep_range(1000, 2000);
}

static inline bool pqi_device_offline(struct pqi_scsi_dev *device)
{
        return device->device_offline;
}

static inline void pqi_device_reset_start(struct pqi_scsi_dev *device)
{
        device->in_reset = true;
}

static inline void pqi_device_reset_done(struct pqi_scsi_dev *device)
{
        device->in_reset = false;
}

static inline bool pqi_device_in_reset(struct pqi_scsi_dev *device)
{
        return device->in_reset;
}

static inline void pqi_schedule_rescan_worker_with_delay(
        struct pqi_ctrl_info *ctrl_info, unsigned long delay)
{
        if (pqi_ctrl_offline(ctrl_info))
                return;

        schedule_delayed_work(&ctrl_info->rescan_work, delay);
}

static inline void pqi_schedule_rescan_worker(struct pqi_ctrl_info *ctrl_info)
{
        pqi_schedule_rescan_worker_with_delay(ctrl_info, 0);
}

#define PQI_RESCAN_WORK_DELAY  (10 * HZ)

static inline void pqi_schedule_rescan_worker_delayed(
        struct pqi_ctrl_info *ctrl_info)
{
        pqi_schedule_rescan_worker_with_delay(ctrl_info, PQI_RESCAN_WORK_DELAY);
}

static inline void pqi_cancel_rescan_worker(struct pqi_ctrl_info *ctrl_info)
{
        cancel_delayed_work_sync(&ctrl_info->rescan_work);
}

static inline u32 pqi_read_heartbeat_counter(struct pqi_ctrl_info *ctrl_info)
{
        if (!ctrl_info->heartbeat_counter)
                return 0;

        return readl(ctrl_info->heartbeat_counter);
}

static int pqi_map_single(struct pci_dev *pci_dev,
        struct pqi_sg_descriptor *sg_descriptor, void *buffer,
        size_t buffer_length, int data_direction)
{
        dma_addr_t bus_address;

        if (!buffer || buffer_length == 0 || data_direction == PCI_DMA_NONE)
                return 0;

        bus_address = pci_map_single(pci_dev, buffer, buffer_length,
                data_direction);
        if (pci_dma_mapping_error(pci_dev, bus_address))
                return -ENOMEM;

        put_unaligned_le64((u64)bus_address, &sg_descriptor->address);
        put_unaligned_le32(buffer_length, &sg_descriptor->length);
        put_unaligned_le32(CISS_SG_LAST, &sg_descriptor->flags);

        return 0;
}

static void pqi_pci_unmap(struct pci_dev *pci_dev,
        struct pqi_sg_descriptor *descriptors, int num_descriptors,
        int data_direction)
{
        int i;

        if (data_direction == PCI_DMA_NONE)
                return;

        for (i = 0; i < num_descriptors; i++)
                pci_unmap_single(pci_dev,
                        (dma_addr_t)get_unaligned_le64(&descriptors[i].address),
                        get_unaligned_le32(&descriptors[i].length),
                        data_direction);
}

static int pqi_build_raid_path_request(struct pqi_ctrl_info *ctrl_info,
        struct pqi_raid_path_request *request, u8 cmd,
        u8 *scsi3addr, void *buffer, size_t buffer_length,
        u16 vpd_page, int *pci_direction)
{
        u8 *cdb;
        int pci_dir;

        memset(request, 0, sizeof(*request));

        request->header.iu_type = PQI_REQUEST_IU_RAID_PATH_IO;
        put_unaligned_le16(offsetof(struct pqi_raid_path_request,
                sg_descriptors[1]) - PQI_REQUEST_HEADER_LENGTH,
                &request->header.iu_length);
        put_unaligned_le32(buffer_length, &request->buffer_length);
        memcpy(request->lun_number, scsi3addr, sizeof(request->lun_number));
        request->task_attribute = SOP_TASK_ATTRIBUTE_SIMPLE;
        request->additional_cdb_bytes_usage = SOP_ADDITIONAL_CDB_BYTES_0;

        cdb = request->cdb;

        switch (cmd) {
        case INQUIRY:
                request->data_direction = SOP_READ_FLAG;
                cdb[0] = INQUIRY;
                if (vpd_page & VPD_PAGE) {
                        cdb[1] = 0x1;
                        cdb[2] = (u8)vpd_page;
                }
                cdb[4] = (u8)buffer_length;
                break;
        case CISS_REPORT_LOG:
        case CISS_REPORT_PHYS:
                request->data_direction = SOP_READ_FLAG;
                cdb[0] = cmd;
                if (cmd == CISS_REPORT_PHYS)
                        cdb[1] = CISS_REPORT_PHYS_EXTENDED;
                else
                        cdb[1] = CISS_REPORT_LOG_EXTENDED;
                put_unaligned_be32(buffer_length, &cdb[6]);
                break;
        case CISS_GET_RAID_MAP:
                request->data_direction = SOP_READ_FLAG;
                cdb[0] = CISS_READ;
                cdb[1] = CISS_GET_RAID_MAP;
                put_unaligned_be32(buffer_length, &cdb[6]);
                break;
        case SA_CACHE_FLUSH:
                request->data_direction = SOP_WRITE_FLAG;
                cdb[0] = BMIC_WRITE;
                cdb[6] = BMIC_CACHE_FLUSH;
                put_unaligned_be16(buffer_length, &cdb[7]);
                break;
        case BMIC_IDENTIFY_CONTROLLER:
        case BMIC_IDENTIFY_PHYSICAL_DEVICE:
                request->data_direction = SOP_READ_FLAG;
                cdb[0] = BMIC_READ;
                cdb[6] = cmd;
                put_unaligned_be16(buffer_length, &cdb[7]);
                break;
        case BMIC_WRITE_HOST_WELLNESS:
                request->data_direction = SOP_WRITE_FLAG;
                cdb[0] = BMIC_WRITE;
                cdb[6] = cmd;
                put_unaligned_be16(buffer_length, &cdb[7]);
                break;
        default:
                dev_err(&ctrl_info->pci_dev->dev, "unknown command 0x%c\n",
                        cmd);
                break;
        }

        switch (request->data_direction) {
        case SOP_READ_FLAG:
                pci_dir = PCI_DMA_FROMDEVICE;
                break;
        case SOP_WRITE_FLAG:
                pci_dir = PCI_DMA_TODEVICE;
                break;
        case SOP_NO_DIRECTION_FLAG:
                pci_dir = PCI_DMA_NONE;
                break;
        default:
                pci_dir = PCI_DMA_BIDIRECTIONAL;
                break;
        }

        *pci_direction = pci_dir;

        return pqi_map_single(ctrl_info->pci_dev, &request->sg_descriptors[0],
                buffer, buffer_length, pci_dir);
}

static inline void pqi_reinit_io_request(struct pqi_io_request *io_request)
{
        io_request->scmd = NULL;
        io_request->status = 0;
        io_request->error_info = NULL;
        io_request->raid_bypass = false;
}

static struct pqi_io_request *pqi_alloc_io_request(
        struct pqi_ctrl_info *ctrl_info)
{
        struct pqi_io_request *io_request;
        u16 i = ctrl_info->next_io_request_slot;        /* benignly racy */

        while (1) {
                io_request = &ctrl_info->io_request_pool[i];
                if (atomic_inc_return(&io_request->refcount) == 1)
                        break;
                atomic_dec(&io_request->refcount);
                i = (i + 1) % ctrl_info->max_io_slots;
        }

        /* benignly racy */
        ctrl_info->next_io_request_slot = (i + 1) % ctrl_info->max_io_slots;

        pqi_reinit_io_request(io_request);

        return io_request;
}

static void pqi_free_io_request(struct pqi_io_request *io_request)
{
        atomic_dec(&io_request->refcount);
}

static int pqi_identify_controller(struct pqi_ctrl_info *ctrl_info,
        struct bmic_identify_controller *buffer)
{
        int rc;
        int pci_direction;
        struct pqi_raid_path_request request;

        rc = pqi_build_raid_path_request(ctrl_info, &request,
                BMIC_IDENTIFY_CONTROLLER, RAID_CTLR_LUNID, buffer,
                sizeof(*buffer), 0, &pci_direction);
        if (rc)
                return rc;

        rc = pqi_submit_raid_request_synchronous(ctrl_info, &request.header, 0,
                NULL, NO_TIMEOUT);

        pqi_pci_unmap(ctrl_info->pci_dev, request.sg_descriptors, 1,
                pci_direction);

        return rc;
}

static int pqi_scsi_inquiry(struct pqi_ctrl_info *ctrl_info,
        u8 *scsi3addr, u16 vpd_page, void *buffer, size_t buffer_length)
{
        int rc;
        int pci_direction;
        struct pqi_raid_path_request request;

        rc = pqi_build_raid_path_request(ctrl_info, &request,
                INQUIRY, scsi3addr, buffer, buffer_length, vpd_page,
                &pci_direction);
        if (rc)
                return rc;

        rc = pqi_submit_raid_request_synchronous(ctrl_info, &request.header, 0,
                NULL, NO_TIMEOUT);

        pqi_pci_unmap(ctrl_info->pci_dev, request.sg_descriptors, 1,
                pci_direction);

        return rc;
}

static int pqi_identify_physical_device(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device,
        struct bmic_identify_physical_device *buffer,
        size_t buffer_length)
{
        int rc;
        int pci_direction;
        u16 bmic_device_index;
        struct pqi_raid_path_request request;

        rc = pqi_build_raid_path_request(ctrl_info, &request,
                BMIC_IDENTIFY_PHYSICAL_DEVICE, RAID_CTLR_LUNID, buffer,
                buffer_length, 0, &pci_direction);
        if (rc)
                return rc;

        bmic_device_index = CISS_GET_DRIVE_NUMBER(device->scsi3addr);
        request.cdb[2] = (u8)bmic_device_index;
        request.cdb[9] = (u8)(bmic_device_index >> 8);

        rc = pqi_submit_raid_request_synchronous(ctrl_info, &request.header,
                0, NULL, NO_TIMEOUT);

        pqi_pci_unmap(ctrl_info->pci_dev, request.sg_descriptors, 1,
                pci_direction);

        return rc;
}

#define SA_CACHE_FLUSH_BUFFER_LENGTH    4

static int pqi_flush_cache(struct pqi_ctrl_info *ctrl_info)
{
        int rc;
        struct pqi_raid_path_request request;
        int pci_direction;
        u8 *buffer;

        /*
         * Don't bother trying to flush the cache if the controller is
         * locked up.
         */
        if (pqi_ctrl_offline(ctrl_info))
                return -ENXIO;

        buffer = kzalloc(SA_CACHE_FLUSH_BUFFER_LENGTH, GFP_KERNEL);
        if (!buffer)
                return -ENOMEM;

        rc = pqi_build_raid_path_request(ctrl_info, &request,
                SA_CACHE_FLUSH, RAID_CTLR_LUNID, buffer,
                SA_CACHE_FLUSH_BUFFER_LENGTH, 0, &pci_direction);
        if (rc)
                goto out;

        rc = pqi_submit_raid_request_synchronous(ctrl_info, &request.header,
                0, NULL, NO_TIMEOUT);

        pqi_pci_unmap(ctrl_info->pci_dev, request.sg_descriptors, 1,
                pci_direction);

out:
        kfree(buffer);

        return rc;
}

static int pqi_write_host_wellness(struct pqi_ctrl_info *ctrl_info,
        void *buffer, size_t buffer_length)
{
        int rc;
        struct pqi_raid_path_request request;
        int pci_direction;

        rc = pqi_build_raid_path_request(ctrl_info, &request,
                BMIC_WRITE_HOST_WELLNESS, RAID_CTLR_LUNID, buffer,
                buffer_length, 0, &pci_direction);
        if (rc)
                return rc;

        rc = pqi_submit_raid_request_synchronous(ctrl_info, &request.header,
                0, NULL, NO_TIMEOUT);

        pqi_pci_unmap(ctrl_info->pci_dev, request.sg_descriptors, 1,
                pci_direction);

        return rc;
}

#pragma pack(1)

struct bmic_host_wellness_driver_version {
        u8      start_tag[4];
        u8      driver_version_tag[2];
        __le16  driver_version_length;
        char    driver_version[32];
        u8      end_tag[2];
};

#pragma pack()

static int pqi_write_driver_version_to_host_wellness(
        struct pqi_ctrl_info *ctrl_info)
{
        int rc;
        struct bmic_host_wellness_driver_version *buffer;
        size_t buffer_length;

        buffer_length = sizeof(*buffer);

        buffer = kmalloc(buffer_length, GFP_KERNEL);
        if (!buffer)
                return -ENOMEM;

        buffer->start_tag[0] = '<';
        buffer->start_tag[1] = 'H';
        buffer->start_tag[2] = 'W';
        buffer->start_tag[3] = '>';
        buffer->driver_version_tag[0] = 'D';
        buffer->driver_version_tag[1] = 'V';
        put_unaligned_le16(sizeof(buffer->driver_version),
                &buffer->driver_version_length);
        strncpy(buffer->driver_version, "Linux " DRIVER_VERSION,
                sizeof(buffer->driver_version) - 1);
        buffer->driver_version[sizeof(buffer->driver_version) - 1] = '\0';
        buffer->end_tag[0] = 'Z';
        buffer->end_tag[1] = 'Z';

        rc = pqi_write_host_wellness(ctrl_info, buffer, buffer_length);

        kfree(buffer);

        return rc;
}

#pragma pack(1)

struct bmic_host_wellness_time {
        u8      start_tag[4];
        u8      time_tag[2];
        __le16  time_length;
        u8      time[8];
        u8      dont_write_tag[2];
        u8      end_tag[2];
};

#pragma pack()

static int pqi_write_current_time_to_host_wellness(
        struct pqi_ctrl_info *ctrl_info)
{
        int rc;
        struct bmic_host_wellness_time *buffer;
        size_t buffer_length;
        time64_t local_time;
        unsigned int year;
        struct tm tm;

        buffer_length = sizeof(*buffer);

        buffer = kmalloc(buffer_length, GFP_KERNEL);
        if (!buffer)
                return -ENOMEM;

        buffer->start_tag[0] = '<';
        buffer->start_tag[1] = 'H';
        buffer->start_tag[2] = 'W';
        buffer->start_tag[3] = '>';
        buffer->time_tag[0] = 'T';
        buffer->time_tag[1] = 'D';
        put_unaligned_le16(sizeof(buffer->time),
                &buffer->time_length);

        local_time = ktime_get_real_seconds();
        time64_to_tm(local_time, -sys_tz.tz_minuteswest * 60, &tm);
        year = tm.tm_year + 1900;

        buffer->time[0] = bin2bcd(tm.tm_hour);
        buffer->time[1] = bin2bcd(tm.tm_min);
        buffer->time[2] = bin2bcd(tm.tm_sec);
        buffer->time[3] = 0;
        buffer->time[4] = bin2bcd(tm.tm_mon + 1);
        buffer->time[5] = bin2bcd(tm.tm_mday);
        buffer->time[6] = bin2bcd(year / 100);
        buffer->time[7] = bin2bcd(year % 100);

        buffer->dont_write_tag[0] = 'D';
        buffer->dont_write_tag[1] = 'W';
        buffer->end_tag[0] = 'Z';
        buffer->end_tag[1] = 'Z';

        rc = pqi_write_host_wellness(ctrl_info, buffer, buffer_length);

        kfree(buffer);

        return rc;
}

#define PQI_UPDATE_TIME_WORK_INTERVAL   (24UL * 60 * 60 * HZ)

static void pqi_update_time_worker(struct work_struct *work)
{
        int rc;
        struct pqi_ctrl_info *ctrl_info;

        ctrl_info = container_of(to_delayed_work(work), struct pqi_ctrl_info,
                update_time_work);

        if (pqi_ctrl_offline(ctrl_info))
                return;

        rc = pqi_write_current_time_to_host_wellness(ctrl_info);
        if (rc)
                dev_warn(&ctrl_info->pci_dev->dev,
                        "error updating time on controller\n");

        schedule_delayed_work(&ctrl_info->update_time_work,
                PQI_UPDATE_TIME_WORK_INTERVAL);
}

static inline void pqi_schedule_update_time_worker(
        struct pqi_ctrl_info *ctrl_info)
{
        schedule_delayed_work(&ctrl_info->update_time_work, 0);
}

static inline void pqi_cancel_update_time_worker(
        struct pqi_ctrl_info *ctrl_info)
{
        cancel_delayed_work_sync(&ctrl_info->update_time_work);
}

static int pqi_report_luns(struct pqi_ctrl_info *ctrl_info, u8 cmd,
        void *buffer, size_t buffer_length)
{
        int rc;
        int pci_direction;
        struct pqi_raid_path_request request;

        rc = pqi_build_raid_path_request(ctrl_info, &request,
                cmd, RAID_CTLR_LUNID, buffer, buffer_length, 0, &pci_direction);
        if (rc)
                return rc;

        rc = pqi_submit_raid_request_synchronous(ctrl_info, &request.header, 0,
                NULL, NO_TIMEOUT);

        pqi_pci_unmap(ctrl_info->pci_dev, request.sg_descriptors, 1,
                pci_direction);

        return rc;
}

static int pqi_report_phys_logical_luns(struct pqi_ctrl_info *ctrl_info, u8 cmd,
        void **buffer)
{
        int rc;
        size_t lun_list_length;
        size_t lun_data_length;
        size_t new_lun_list_length;
        void *lun_data = NULL;
        struct report_lun_header *report_lun_header;

        report_lun_header = kmalloc(sizeof(*report_lun_header), GFP_KERNEL);
        if (!report_lun_header) {
                rc = -ENOMEM;
                goto out;
        }

        rc = pqi_report_luns(ctrl_info, cmd, report_lun_header,
                sizeof(*report_lun_header));
        if (rc)
                goto out;

        lun_list_length = get_unaligned_be32(&report_lun_header->list_length);

again:
        lun_data_length = sizeof(struct report_lun_header) + lun_list_length;

        lun_data = kmalloc(lun_data_length, GFP_KERNEL);
        if (!lun_data) {
                rc = -ENOMEM;
                goto out;
        }

        if (lun_list_length == 0) {
                memcpy(lun_data, report_lun_header, sizeof(*report_lun_header));
                goto out;
        }

        rc = pqi_report_luns(ctrl_info, cmd, lun_data, lun_data_length);
        if (rc)
                goto out;

        new_lun_list_length = get_unaligned_be32(
                &((struct report_lun_header *)lun_data)->list_length);

        if (new_lun_list_length > lun_list_length) {
                lun_list_length = new_lun_list_length;
                kfree(lun_data);
                goto again;
        }

out:
        kfree(report_lun_header);

        if (rc) {
                kfree(lun_data);
                lun_data = NULL;
        }

        *buffer = lun_data;

        return rc;
}

static inline int pqi_report_phys_luns(struct pqi_ctrl_info *ctrl_info,
        void **buffer)
{
        return pqi_report_phys_logical_luns(ctrl_info, CISS_REPORT_PHYS,
                buffer);
}

static inline int pqi_report_logical_luns(struct pqi_ctrl_info *ctrl_info,
        void **buffer)
{
        return pqi_report_phys_logical_luns(ctrl_info, CISS_REPORT_LOG, buffer);
}

static int pqi_get_device_lists(struct pqi_ctrl_info *ctrl_info,
        struct report_phys_lun_extended **physdev_list,
        struct report_log_lun_extended **logdev_list)
{
        int rc;
        size_t logdev_list_length;
        size_t logdev_data_length;
        struct report_log_lun_extended *internal_logdev_list;
        struct report_log_lun_extended *logdev_data;
        struct report_lun_header report_lun_header;

        rc = pqi_report_phys_luns(ctrl_info, (void **)physdev_list);
        if (rc)
                dev_err(&ctrl_info->pci_dev->dev,
                        "report physical LUNs failed\n");

        rc = pqi_report_logical_luns(ctrl_info, (void **)logdev_list);
        if (rc)
                dev_err(&ctrl_info->pci_dev->dev,
                        "report logical LUNs failed\n");

        /*
         * Tack the controller itself onto the end of the logical device list.
         */

        logdev_data = *logdev_list;

        if (logdev_data) {
                logdev_list_length =
                        get_unaligned_be32(&logdev_data->header.list_length);
        } else {
                memset(&report_lun_header, 0, sizeof(report_lun_header));
                logdev_data =
                        (struct report_log_lun_extended *)&report_lun_header;
                logdev_list_length = 0;
        }

        logdev_data_length = sizeof(struct report_lun_header) +
                logdev_list_length;

        internal_logdev_list = kmalloc(logdev_data_length +
                sizeof(struct report_log_lun_extended), GFP_KERNEL);
        if (!internal_logdev_list) {
                kfree(*logdev_list);
                *logdev_list = NULL;
                return -ENOMEM;
        }

        memcpy(internal_logdev_list, logdev_data, logdev_data_length);
        memset((u8 *)internal_logdev_list + logdev_data_length, 0,
                sizeof(struct report_log_lun_extended_entry));
        put_unaligned_be32(logdev_list_length +
                sizeof(struct report_log_lun_extended_entry),
                &internal_logdev_list->header.list_length);

        kfree(*logdev_list);
        *logdev_list = internal_logdev_list;

        return 0;
}

static inline void pqi_set_bus_target_lun(struct pqi_scsi_dev *device,
        int bus, int target, int lun)
{
        device->bus = bus;
        device->target = target;
        device->lun = lun;
}

static void pqi_assign_bus_target_lun(struct pqi_scsi_dev *device)
{
        u8 *scsi3addr;
        u32 lunid;
        int bus;
        int target;
        int lun;

        scsi3addr = device->scsi3addr;
        lunid = get_unaligned_le32(scsi3addr);

        if (pqi_is_hba_lunid(scsi3addr)) {
                /* The specified device is the controller. */
                pqi_set_bus_target_lun(device, PQI_HBA_BUS, 0, lunid & 0x3fff);
                device->target_lun_valid = true;
                return;
        }

        if (pqi_is_logical_device(device)) {
                if (device->is_external_raid_device) {
                        bus = PQI_EXTERNAL_RAID_VOLUME_BUS;
                        target = (lunid >> 16) & 0x3fff;
                        lun = lunid & 0xff;
                } else {
                        bus = PQI_RAID_VOLUME_BUS;
                        target = 0;
                        lun = lunid & 0x3fff;
                }
                pqi_set_bus_target_lun(device, bus, target, lun);
                device->target_lun_valid = true;
                return;
        }

        /*
         * Defer target and LUN assignment for non-controller physical devices
         * because the SAS transport layer will make these assignments later.
         */
        pqi_set_bus_target_lun(device, PQI_PHYSICAL_DEVICE_BUS, 0, 0);
}

static void pqi_get_raid_level(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device)
{
        int rc;

        u8 raid_level;
        u8 *buffer;

        raid_level = SA_RAID_UNKNOWN;

        buffer = kmalloc(64, GFP_KERNEL);
        if (buffer) {
                rc = pqi_scsi_inquiry(ctrl_info, device->scsi3addr,
                        VPD_PAGE | CISS_VPD_LV_DEVICE_GEOMETRY, buffer, 64);
                if (rc == 0) {
                        raid_level = buffer[8];
                        if (raid_level > SA_RAID_MAX)
                                raid_level = SA_RAID_UNKNOWN;
                }
                kfree(buffer);
        }

        device->raid_level = raid_level;
}

static int pqi_validate_raid_map(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device, struct raid_map *raid_map)
{
        char *err_msg;
        u32 raid_map_size;
        u32 r5or6_blocks_per_row;
        unsigned int num_phys_disks;
        unsigned int num_raid_map_entries;

        raid_map_size = get_unaligned_le32(&raid_map->structure_size);

        if (raid_map_size < offsetof(struct raid_map, disk_data)) {
                err_msg = "RAID map too small";
                goto bad_raid_map;
        }

        if (raid_map_size > sizeof(*raid_map)) {
                err_msg = "RAID map too large";
                goto bad_raid_map;
        }

        num_phys_disks = get_unaligned_le16(&raid_map->layout_map_count) *
                (get_unaligned_le16(&raid_map->data_disks_per_row) +
                get_unaligned_le16(&raid_map->metadata_disks_per_row));
        num_raid_map_entries = num_phys_disks *
                get_unaligned_le16(&raid_map->row_cnt);

        if (num_raid_map_entries > RAID_MAP_MAX_ENTRIES) {
                err_msg = "invalid number of map entries in RAID map";
                goto bad_raid_map;
        }

        if (device->raid_level == SA_RAID_1) {
                if (get_unaligned_le16(&raid_map->layout_map_count) != 2) {
                        err_msg = "invalid RAID-1 map";
                        goto bad_raid_map;
                }
        } else if (device->raid_level == SA_RAID_ADM) {
                if (get_unaligned_le16(&raid_map->layout_map_count) != 3) {
                        err_msg = "invalid RAID-1(ADM) map";
                        goto bad_raid_map;
                }
        } else if ((device->raid_level == SA_RAID_5 ||
                device->raid_level == SA_RAID_6) &&
                get_unaligned_le16(&raid_map->layout_map_count) > 1) {
                /* RAID 50/60 */
                r5or6_blocks_per_row =
                        get_unaligned_le16(&raid_map->strip_size) *
                        get_unaligned_le16(&raid_map->data_disks_per_row);
                if (r5or6_blocks_per_row == 0) {
                        err_msg = "invalid RAID-5 or RAID-6 map";
                        goto bad_raid_map;
                }
        }

        return 0;

bad_raid_map:
        dev_warn(&ctrl_info->pci_dev->dev,
                "scsi %d:%d:%d:%d %s\n",
                ctrl_info->scsi_host->host_no,
                device->bus, device->target, device->lun, err_msg);

        return -EINVAL;
}

static int pqi_get_raid_map(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device)
{
        int rc;
        int pci_direction;
        struct pqi_raid_path_request request;
        struct raid_map *raid_map;

        raid_map = kmalloc(sizeof(*raid_map), GFP_KERNEL);
        if (!raid_map)
                return -ENOMEM;

        rc = pqi_build_raid_path_request(ctrl_info, &request,
                CISS_GET_RAID_MAP, device->scsi3addr, raid_map,
                sizeof(*raid_map), 0, &pci_direction);
        if (rc)
                goto error;

        rc = pqi_submit_raid_request_synchronous(ctrl_info, &request.header, 0,
                NULL, NO_TIMEOUT);

        pqi_pci_unmap(ctrl_info->pci_dev, request.sg_descriptors, 1,
                pci_direction);

        if (rc)
                goto error;

        rc = pqi_validate_raid_map(ctrl_info, device, raid_map);
        if (rc)
                goto error;

        device->raid_map = raid_map;

        return 0;

error:
        kfree(raid_map);

        return rc;
}

static void pqi_get_raid_bypass_status(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device)
{
        int rc;
        u8 *buffer;
        u8 bypass_status;

        buffer = kmalloc(64, GFP_KERNEL);
        if (!buffer)
                return;

        rc = pqi_scsi_inquiry(ctrl_info, device->scsi3addr,
                VPD_PAGE | CISS_VPD_LV_BYPASS_STATUS, buffer, 64);
        if (rc)
                goto out;

#define RAID_BYPASS_STATUS      4
#define RAID_BYPASS_CONFIGURED  0x1
#define RAID_BYPASS_ENABLED     0x2

        bypass_status = buffer[RAID_BYPASS_STATUS];
        device->raid_bypass_configured =
                (bypass_status & RAID_BYPASS_CONFIGURED) != 0;
        if (device->raid_bypass_configured &&
                (bypass_status & RAID_BYPASS_ENABLED) &&
                pqi_get_raid_map(ctrl_info, device) == 0)
                device->raid_bypass_enabled = true;

out:
        kfree(buffer);
}

/*
 * Use vendor-specific VPD to determine online/offline status of a volume.
 */

static void pqi_get_volume_status(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device)
{
        int rc;
        size_t page_length;
        u8 volume_status = CISS_LV_STATUS_UNAVAILABLE;
        bool volume_offline = true;
        u32 volume_flags;
        struct ciss_vpd_logical_volume_status *vpd;

        vpd = kmalloc(sizeof(*vpd), GFP_KERNEL);
        if (!vpd)
                goto no_buffer;

        rc = pqi_scsi_inquiry(ctrl_info, device->scsi3addr,
                VPD_PAGE | CISS_VPD_LV_STATUS, vpd, sizeof(*vpd));
        if (rc)
                goto out;

        page_length = offsetof(struct ciss_vpd_logical_volume_status,
                volume_status) + vpd->page_length;
        if (page_length < sizeof(*vpd))
                goto out;

        volume_status = vpd->volume_status;
        volume_flags = get_unaligned_be32(&vpd->flags);
        volume_offline = (volume_flags & CISS_LV_FLAGS_NO_HOST_IO) != 0;

out:
        kfree(vpd);
no_buffer:
        device->volume_status = volume_status;
        device->volume_offline = volume_offline;
}

static int pqi_get_device_info(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device)
{
        int rc;
        u8 *buffer;

        buffer = kmalloc(64, GFP_KERNEL);
        if (!buffer)
                return -ENOMEM;

        /* Send an inquiry to the device to see what it is. */
        rc = pqi_scsi_inquiry(ctrl_info, device->scsi3addr, 0, buffer, 64);
        if (rc)
                goto out;

        scsi_sanitize_inquiry_string(&buffer[8], 8);
        scsi_sanitize_inquiry_string(&buffer[16], 16);

        device->devtype = buffer[0] & 0x1f;
        memcpy(device->vendor, &buffer[8], sizeof(device->vendor));
        memcpy(device->model, &buffer[16], sizeof(device->model));

        if (pqi_is_logical_device(device) && device->devtype == TYPE_DISK) {
                if (device->is_external_raid_device) {
                        device->raid_level = SA_RAID_UNKNOWN;
                        device->volume_status = CISS_LV_OK;
                        device->volume_offline = false;
                } else {
                        pqi_get_raid_level(ctrl_info, device);
                        pqi_get_raid_bypass_status(ctrl_info, device);
                        pqi_get_volume_status(ctrl_info, device);
                }
        }

out:
        kfree(buffer);

        return rc;
}

static void pqi_get_physical_disk_info(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device,
        struct bmic_identify_physical_device *id_phys)
{
        int rc;

        memset(id_phys, 0, sizeof(*id_phys));

        rc = pqi_identify_physical_device(ctrl_info, device,
                id_phys, sizeof(*id_phys));
        if (rc) {
                device->queue_depth = PQI_PHYSICAL_DISK_DEFAULT_MAX_QUEUE_DEPTH;
                return;
        }

        device->queue_depth =
                get_unaligned_le16(&id_phys->current_queue_depth_limit);
        device->device_type = id_phys->device_type;
        device->active_path_index = id_phys->active_path_number;
        device->path_map = id_phys->redundant_path_present_map;
        memcpy(&device->box,
                &id_phys->alternate_paths_phys_box_on_port,
                sizeof(device->box));
        memcpy(&device->phys_connector,
                &id_phys->alternate_paths_phys_connector,
                sizeof(device->phys_connector));
        device->bay = id_phys->phys_bay_in_box;
}

static void pqi_show_volume_status(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device)
{
        char *status;
        static const char unknown_state_str[] =
                "Volume is in an unknown state (%u)";
        char unknown_state_buffer[sizeof(unknown_state_str) + 10];

        switch (device->volume_status) {
        case CISS_LV_OK:
                status = "Volume online";
                break;
        case CISS_LV_FAILED:
                status = "Volume failed";
                break;
        case CISS_LV_NOT_CONFIGURED:
                status = "Volume not configured";
                break;
        case CISS_LV_DEGRADED:
                status = "Volume degraded";
                break;
        case CISS_LV_READY_FOR_RECOVERY:
                status = "Volume ready for recovery operation";
                break;
        case CISS_LV_UNDERGOING_RECOVERY:
                status = "Volume undergoing recovery";
                break;
        case CISS_LV_WRONG_PHYSICAL_DRIVE_REPLACED:
                status = "Wrong physical drive was replaced";
                break;
        case CISS_LV_PHYSICAL_DRIVE_CONNECTION_PROBLEM:
                status = "A physical drive not properly connected";
                break;
        case CISS_LV_HARDWARE_OVERHEATING:
                status = "Hardware is overheating";
                break;
        case CISS_LV_HARDWARE_HAS_OVERHEATED:
                status = "Hardware has overheated";
                break;
        case CISS_LV_UNDERGOING_EXPANSION:
                status = "Volume undergoing expansion";
                break;
        case CISS_LV_NOT_AVAILABLE:
                status = "Volume waiting for transforming volume";
                break;
        case CISS_LV_QUEUED_FOR_EXPANSION:
                status = "Volume queued for expansion";
                break;
        case CISS_LV_DISABLED_SCSI_ID_CONFLICT:
                status = "Volume disabled due to SCSI ID conflict";
                break;
        case CISS_LV_EJECTED:
                status = "Volume has been ejected";
                break;
        case CISS_LV_UNDERGOING_ERASE:
                status = "Volume undergoing background erase";
                break;
        case CISS_LV_READY_FOR_PREDICTIVE_SPARE_REBUILD:
                status = "Volume ready for predictive spare rebuild";
                break;
        case CISS_LV_UNDERGOING_RPI:
                status = "Volume undergoing rapid parity initialization";
                break;
        case CISS_LV_PENDING_RPI:
                status = "Volume queued for rapid parity initialization";
                break;
        case CISS_LV_ENCRYPTED_NO_KEY:
                status = "Encrypted volume inaccessible - key not present";
                break;
        case CISS_LV_UNDERGOING_ENCRYPTION:
                status = "Volume undergoing encryption process";
                break;
        case CISS_LV_UNDERGOING_ENCRYPTION_REKEYING:
                status = "Volume undergoing encryption re-keying process";
                break;
        case CISS_LV_ENCRYPTED_IN_NON_ENCRYPTED_CONTROLLER:
                status = "Volume encrypted but encryption is disabled";
                break;
        case CISS_LV_PENDING_ENCRYPTION:
                status = "Volume pending migration to encrypted state";
                break;
        case CISS_LV_PENDING_ENCRYPTION_REKEYING:
                status = "Volume pending encryption rekeying";
                break;
        case CISS_LV_NOT_SUPPORTED:
                status = "Volume not supported on this controller";
                break;
        case CISS_LV_STATUS_UNAVAILABLE:
                status = "Volume status not available";
                break;
        default:
                snprintf(unknown_state_buffer, sizeof(unknown_state_buffer),
                        unknown_state_str, device->volume_status);
                status = unknown_state_buffer;
                break;
        }

        dev_info(&ctrl_info->pci_dev->dev,
                "scsi %d:%d:%d:%d %s\n",
                ctrl_info->scsi_host->host_no,
                device->bus, device->target, device->lun, status);
}

static void pqi_rescan_worker(struct work_struct *work)
{
        struct pqi_ctrl_info *ctrl_info;

        ctrl_info = container_of(to_delayed_work(work), struct pqi_ctrl_info,
                rescan_work);

        pqi_scan_scsi_devices(ctrl_info);
}

static int pqi_add_device(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device)
{
        int rc;

        if (pqi_is_logical_device(device))
                rc = scsi_add_device(ctrl_info->scsi_host, device->bus,
                        device->target, device->lun);
        else
                rc = pqi_add_sas_device(ctrl_info->sas_host, device);

        return rc;
}

static inline void pqi_remove_device(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device)
{
        if (pqi_is_logical_device(device))
                scsi_remove_device(device->sdev);
        else
                pqi_remove_sas_device(device);
}

/* Assumes the SCSI device list lock is held. */

static struct pqi_scsi_dev *pqi_find_scsi_dev(struct pqi_ctrl_info *ctrl_info,
        int bus, int target, int lun)
{
        struct pqi_scsi_dev *device;

        list_for_each_entry(device, &ctrl_info->scsi_device_list,
                scsi_device_list_entry)
                if (device->bus == bus && device->target == target &&
                        device->lun == lun)
                        return device;

        return NULL;
}

static inline bool pqi_device_equal(struct pqi_scsi_dev *dev1,
        struct pqi_scsi_dev *dev2)
{
        if (dev1->is_physical_device != dev2->is_physical_device)
                return false;

        if (dev1->is_physical_device)
                return dev1->wwid == dev2->wwid;

        return memcmp(dev1->volume_id, dev2->volume_id,
                sizeof(dev1->volume_id)) == 0;
}

enum pqi_find_result {
        DEVICE_NOT_FOUND,
        DEVICE_CHANGED,
        DEVICE_SAME,
};

static enum pqi_find_result pqi_scsi_find_entry(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device_to_find,
        struct pqi_scsi_dev **matching_device)
{
        struct pqi_scsi_dev *device;

        list_for_each_entry(device, &ctrl_info->scsi_device_list,
                scsi_device_list_entry) {
                if (pqi_scsi3addr_equal(device_to_find->scsi3addr,
                        device->scsi3addr)) {
                        *matching_device = device;
                        if (pqi_device_equal(device_to_find, device)) {
                                if (device_to_find->volume_offline)
                                        return DEVICE_CHANGED;
                                return DEVICE_SAME;
                        }
                        return DEVICE_CHANGED;
                }
        }

        return DEVICE_NOT_FOUND;
}

#define PQI_DEV_INFO_BUFFER_LENGTH      128

static void pqi_dev_info(struct pqi_ctrl_info *ctrl_info,
        char *action, struct pqi_scsi_dev *device)
{
        ssize_t count;
        char buffer[PQI_DEV_INFO_BUFFER_LENGTH];

        count = snprintf(buffer, PQI_DEV_INFO_BUFFER_LENGTH,
                "%d:%d:", ctrl_info->scsi_host->host_no, device->bus);

        if (device->target_lun_valid)
                count += snprintf(buffer + count,
                        PQI_DEV_INFO_BUFFER_LENGTH - count,
                        "%d:%d",
                        device->target,
                        device->lun);
        else
                count += snprintf(buffer + count,
                        PQI_DEV_INFO_BUFFER_LENGTH - count,
                        "-:-");

        if (pqi_is_logical_device(device))
                count += snprintf(buffer + count,
                        PQI_DEV_INFO_BUFFER_LENGTH - count,
                        " %08x%08x",
                        *((u32 *)&device->scsi3addr),
                        *((u32 *)&device->scsi3addr[4]));
        else
                count += snprintf(buffer + count,
                        PQI_DEV_INFO_BUFFER_LENGTH - count,
                        " %016llx", device->sas_address);

        count += snprintf(buffer + count, PQI_DEV_INFO_BUFFER_LENGTH - count,
                " %s %.8s %.16s ",
                scsi_device_type(device->devtype),
                device->vendor,
                device->model);

        if (pqi_is_logical_device(device)) {
                if (device->devtype == TYPE_DISK)
                        count += snprintf(buffer + count,
                                PQI_DEV_INFO_BUFFER_LENGTH - count,
                                "SSDSmartPathCap%c En%c %-12s",
                                device->raid_bypass_configured ? '+' : '-',
                                device->raid_bypass_enabled ? '+' : '-',
                                pqi_raid_level_to_string(device->raid_level));
        } else {
                count += snprintf(buffer + count,
                        PQI_DEV_INFO_BUFFER_LENGTH - count,
                        "AIO%c", device->aio_enabled ? '+' : '-');
                if (device->devtype == TYPE_DISK ||
                        device->devtype == TYPE_ZBC)
                        count += snprintf(buffer + count,
                                PQI_DEV_INFO_BUFFER_LENGTH - count,
                                " qd=%-6d", device->queue_depth);
        }

        dev_info(&ctrl_info->pci_dev->dev, "%s %s\n", action, buffer);
}

/* Assumes the SCSI device list lock is held. */

static void pqi_scsi_update_device(struct pqi_scsi_dev *existing_device,
        struct pqi_scsi_dev *new_device)
{
        existing_device->devtype = new_device->devtype;
        existing_device->device_type = new_device->device_type;
        existing_device->bus = new_device->bus;
        if (new_device->target_lun_valid) {
                existing_device->target = new_device->target;
                existing_device->lun = new_device->lun;
                existing_device->target_lun_valid = true;
        }

        /* By definition, the scsi3addr and wwid fields are already the same. */

        existing_device->is_physical_device = new_device->is_physical_device;
        existing_device->is_external_raid_device =
                new_device->is_external_raid_device;
        existing_device->aio_enabled = new_device->aio_enabled;
        memcpy(existing_device->vendor, new_device->vendor,
                sizeof(existing_device->vendor));
        memcpy(existing_device->model, new_device->model,
                sizeof(existing_device->model));
        existing_device->sas_address = new_device->sas_address;
        existing_device->raid_level = new_device->raid_level;
        existing_device->queue_depth = new_device->queue_depth;
        existing_device->aio_handle = new_device->aio_handle;
        existing_device->volume_status = new_device->volume_status;
        existing_device->active_path_index = new_device->active_path_index;
        existing_device->path_map = new_device->path_map;
        existing_device->bay = new_device->bay;
        memcpy(existing_device->box, new_device->box,
                sizeof(existing_device->box));
        memcpy(existing_device->phys_connector, new_device->phys_connector,
                sizeof(existing_device->phys_connector));
        existing_device->offload_to_mirror = 0;
        kfree(existing_device->raid_map);
        existing_device->raid_map = new_device->raid_map;
        existing_device->raid_bypass_configured =
                new_device->raid_bypass_configured;
        existing_device->raid_bypass_enabled =
                new_device->raid_bypass_enabled;

        /* To prevent this from being freed later. */
        new_device->raid_map = NULL;
}

static inline void pqi_free_device(struct pqi_scsi_dev *device)
{
        if (device) {
                kfree(device->raid_map);
                kfree(device);
        }
}

/*
 * Called when exposing a new device to the OS fails in order to re-adjust
 * our internal SCSI device list to match the SCSI ML's view.
 */

static inline void pqi_fixup_botched_add(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device)
{
        unsigned long flags;

        spin_lock_irqsave(&ctrl_info->scsi_device_list_lock, flags);
        list_del(&device->scsi_device_list_entry);
        spin_unlock_irqrestore(&ctrl_info->scsi_device_list_lock, flags);

        /* Allow the device structure to be freed later. */
        device->keep_device = false;
}

static void pqi_update_device_list(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *new_device_list[], unsigned int num_new_devices)
{
        int rc;
        unsigned int i;
        unsigned long flags;
        enum pqi_find_result find_result;
        struct pqi_scsi_dev *device;
        struct pqi_scsi_dev *next;
        struct pqi_scsi_dev *matching_device;
        LIST_HEAD(add_list);
        LIST_HEAD(delete_list);

        /*
         * The idea here is to do as little work as possible while holding the
         * spinlock.  That's why we go to great pains to defer anything other
         * than updating the internal device list until after we release the
         * spinlock.
         */

        spin_lock_irqsave(&ctrl_info->scsi_device_list_lock, flags);

        /* Assume that all devices in the existing list have gone away. */
        list_for_each_entry(device, &ctrl_info->scsi_device_list,
                scsi_device_list_entry)
                device->device_gone = true;

        for (i = 0; i < num_new_devices; i++) {
                device = new_device_list[i];

                find_result = pqi_scsi_find_entry(ctrl_info, device,
                                                &matching_device);

                switch (find_result) {
                case DEVICE_SAME:
                        /*
                         * The newly found device is already in the existing
                         * device list.
                         */
                        device->new_device = false;
                        matching_device->device_gone = false;
                        pqi_scsi_update_device(matching_device, device);
                        break;
                case DEVICE_NOT_FOUND:
                        /*
                         * The newly found device is NOT in the existing device
                         * list.
                         */
                        device->new_device = true;
                        break;
                case DEVICE_CHANGED:
                        /*
                         * The original device has gone away and we need to add
                         * the new device.
                         */
                        device->new_device = true;
                        break;
                }
        }

        /* Process all devices that have gone away. */
        list_for_each_entry_safe(device, next, &ctrl_info->scsi_device_list,
                scsi_device_list_entry) {
                if (device->device_gone) {
                        list_del(&device->scsi_device_list_entry);
                        list_add_tail(&device->delete_list_entry, &delete_list);
                }
        }

        /* Process all new devices. */
        for (i = 0; i < num_new_devices; i++) {
                device = new_device_list[i];
                if (!device->new_device)
                        continue;
                if (device->volume_offline)
                        continue;
                list_add_tail(&device->scsi_device_list_entry,
                        &ctrl_info->scsi_device_list);
                list_add_tail(&device->add_list_entry, &add_list);
                /* To prevent this device structure from being freed later. */
                device->keep_device = true;
        }

        spin_unlock_irqrestore(&ctrl_info->scsi_device_list_lock, flags);

        /* Remove all devices that have gone away. */
        list_for_each_entry_safe(device, next, &delete_list,
                delete_list_entry) {
                if (device->volume_offline) {
                        pqi_dev_info(ctrl_info, "offline", device);
                        pqi_show_volume_status(ctrl_info, device);
                } else {
                        pqi_dev_info(ctrl_info, "removed", device);
                }
                if (device->sdev)
                        pqi_remove_device(ctrl_info, device);
                list_del(&device->delete_list_entry);
                pqi_free_device(device);
        }

        /*
         * Notify the SCSI ML if the queue depth of any existing device has
         * changed.
         */
        list_for_each_entry(device, &ctrl_info->scsi_device_list,
                scsi_device_list_entry) {
                if (device->sdev && device->queue_depth !=
                        device->advertised_queue_depth) {
                        device->advertised_queue_depth = device->queue_depth;
                        scsi_change_queue_depth(device->sdev,
                                device->advertised_queue_depth);
                }
        }

        /* Expose any new devices. */
        list_for_each_entry_safe(device, next, &add_list, add_list_entry) {
                if (!device->sdev) {
                        pqi_dev_info(ctrl_info, "added", device);
                        rc = pqi_add_device(ctrl_info, device);
                        if (rc) {
                                dev_warn(&ctrl_info->pci_dev->dev,
                                        "scsi %d:%d:%d:%d addition failed, device not added\n",
                                        ctrl_info->scsi_host->host_no,
                                        device->bus, device->target,
                                        device->lun);
                                pqi_fixup_botched_add(ctrl_info, device);
                        }
                }
        }
}

static bool pqi_is_supported_device(struct pqi_scsi_dev *device)
{
        bool is_supported = false;

        switch (device->devtype) {
        case TYPE_DISK:
        case TYPE_ZBC:
        case TYPE_TAPE:
        case TYPE_MEDIUM_CHANGER:
        case TYPE_ENCLOSURE:
                is_supported = true;
                break;
        case TYPE_RAID:
                /*
                 * Only support the HBA controller itself as a RAID
                 * controller.  If it's a RAID controller other than
                 * the HBA itself (an external RAID controller, for
                 * example), we don't support it.
                 */
                if (pqi_is_hba_lunid(device->scsi3addr))
                        is_supported = true;
                break;
        }

        return is_supported;
}

static inline bool pqi_skip_device(u8 *scsi3addr)
{
        /* Ignore all masked devices. */
        if (MASKED_DEVICE(scsi3addr))
                return true;

        return false;
}

static int pqi_update_scsi_devices(struct pqi_ctrl_info *ctrl_info)
{
        int i;
        int rc;
        LIST_HEAD(new_device_list_head);
        struct report_phys_lun_extended *physdev_list = NULL;
        struct report_log_lun_extended *logdev_list = NULL;
        struct report_phys_lun_extended_entry *phys_lun_ext_entry;
        struct report_log_lun_extended_entry *log_lun_ext_entry;
        struct bmic_identify_physical_device *id_phys = NULL;
        u32 num_physicals;
        u32 num_logicals;
        struct pqi_scsi_dev **new_device_list = NULL;
        struct pqi_scsi_dev *device;
        struct pqi_scsi_dev *next;
        unsigned int num_new_devices;
        unsigned int num_valid_devices;
        bool is_physical_device;
        u8 *scsi3addr;
        static char *out_of_memory_msg =
                "failed to allocate memory, device discovery stopped";

        rc = pqi_get_device_lists(ctrl_info, &physdev_list, &logdev_list);
        if (rc)
                goto out;

        if (physdev_list)
                num_physicals =
                        get_unaligned_be32(&physdev_list->header.list_length)
                                / sizeof(physdev_list->lun_entries[0]);
        else
                num_physicals = 0;

        if (logdev_list)
                num_logicals =
                        get_unaligned_be32(&logdev_list->header.list_length)
                                / sizeof(logdev_list->lun_entries[0]);
        else
                num_logicals = 0;

        if (num_physicals) {
                /*
                 * We need this buffer for calls to pqi_get_physical_disk_info()
                 * below.  We allocate it here instead of inside
                 * pqi_get_physical_disk_info() because it's a fairly large
                 * buffer.
                 */
                id_phys = kmalloc(sizeof(*id_phys), GFP_KERNEL);
                if (!id_phys) {
                        dev_warn(&ctrl_info->pci_dev->dev, "%s\n",
                                out_of_memory_msg);
                        rc = -ENOMEM;
                        goto out;
                }
        }

        num_new_devices = num_physicals + num_logicals;

        new_device_list = kmalloc(sizeof(*new_device_list) *
                num_new_devices, GFP_KERNEL);
        if (!new_device_list) {
                dev_warn(&ctrl_info->pci_dev->dev, "%s\n", out_of_memory_msg);
                rc = -ENOMEM;
                goto out;
        }

        for (i = 0; i < num_new_devices; i++) {
                device = kzalloc(sizeof(*device), GFP_KERNEL);
                if (!device) {
                        dev_warn(&ctrl_info->pci_dev->dev, "%s\n",
                                out_of_memory_msg);
                        rc = -ENOMEM;
                        goto out;
                }
                list_add_tail(&device->new_device_list_entry,
                        &new_device_list_head);
        }

        device = NULL;
        num_valid_devices = 0;

        for (i = 0; i < num_new_devices; i++) {

                if (i < num_physicals) {
                        is_physical_device = true;
                        phys_lun_ext_entry = &physdev_list->lun_entries[i];
                        log_lun_ext_entry = NULL;
                        scsi3addr = phys_lun_ext_entry->lunid;
                } else {
                        is_physical_device = false;
                        phys_lun_ext_entry = NULL;
                        log_lun_ext_entry =
                                &logdev_list->lun_entries[i - num_physicals];
                        scsi3addr = log_lun_ext_entry->lunid;
                }

                if (is_physical_device && pqi_skip_device(scsi3addr))
                        continue;

                if (device)
                        device = list_next_entry(device, new_device_list_entry);
                else
                        device = list_first_entry(&new_device_list_head,
                                struct pqi_scsi_dev, new_device_list_entry);

                memcpy(device->scsi3addr, scsi3addr, sizeof(device->scsi3addr));
                device->is_physical_device = is_physical_device;
                if (!is_physical_device)
                        device->is_external_raid_device =
                                pqi_is_external_raid_addr(scsi3addr);

                /* Gather information about the device. */
                rc = pqi_get_device_info(ctrl_info, device);
                if (rc == -ENOMEM) {
                        dev_warn(&ctrl_info->pci_dev->dev, "%s\n",
                                out_of_memory_msg);
                        goto out;
                }
                if (rc) {
                        if (device->is_physical_device)
                                dev_warn(&ctrl_info->pci_dev->dev,
                                        "obtaining device info failed, skipping physical device %016llx\n",
                                        get_unaligned_be64(
                                                &phys_lun_ext_entry->wwid));
                        else
                                dev_warn(&ctrl_info->pci_dev->dev,
                                        "obtaining device info failed, skipping logical device %08x%08x\n",
                                        *((u32 *)&device->scsi3addr),
                                        *((u32 *)&device->scsi3addr[4]));
                        rc = 0;
                        continue;
                }

                if (!pqi_is_supported_device(device))
                        continue;

                pqi_assign_bus_target_lun(device);

                if (device->is_physical_device) {
                        device->wwid = phys_lun_ext_entry->wwid;
                        if ((phys_lun_ext_entry->device_flags &
                                REPORT_PHYS_LUN_DEV_FLAG_AIO_ENABLED) &&
                                phys_lun_ext_entry->aio_handle)
                                device->aio_enabled = true;
                } else {
                        memcpy(device->volume_id, log_lun_ext_entry->volume_id,
                                sizeof(device->volume_id));
                }

                switch (device->devtype) {
                case TYPE_DISK:
                case TYPE_ZBC:
                case TYPE_ENCLOSURE:
                        if (device->is_physical_device) {
                                device->sas_address =
                                        get_unaligned_be64(&device->wwid);
                                if (device->devtype == TYPE_DISK ||
                                        device->devtype == TYPE_ZBC) {
                                        device->aio_handle =
                                                phys_lun_ext_entry->aio_handle;
                                        pqi_get_physical_disk_info(ctrl_info,
                                                device, id_phys);
                                }
                        }
                        break;
                }

                new_device_list[num_valid_devices++] = device;
        }

        pqi_update_device_list(ctrl_info, new_device_list, num_valid_devices);

out:
        list_for_each_entry_safe(device, next, &new_device_list_head,
                new_device_list_entry) {
                if (device->keep_device)
                        continue;
                list_del(&device->new_device_list_entry);
                pqi_free_device(device);
        }

        kfree(new_device_list);
        kfree(physdev_list);
        kfree(logdev_list);
        kfree(id_phys);

        return rc;
}

static void pqi_remove_all_scsi_devices(struct pqi_ctrl_info *ctrl_info)
{
        unsigned long flags;
        struct pqi_scsi_dev *device;

        while (1) {
                spin_lock_irqsave(&ctrl_info->scsi_device_list_lock, flags);

                device = list_first_entry_or_null(&ctrl_info->scsi_device_list,
                        struct pqi_scsi_dev, scsi_device_list_entry);
                if (device)
                        list_del(&device->scsi_device_list_entry);

                spin_unlock_irqrestore(&ctrl_info->scsi_device_list_lock,
                        flags);

                if (!device)
                        break;

                if (device->sdev)
                        pqi_remove_device(ctrl_info, device);
                pqi_free_device(device);
        }
}

static int pqi_scan_scsi_devices(struct pqi_ctrl_info *ctrl_info)
{
        int rc;

        if (pqi_ctrl_offline(ctrl_info))
                return -ENXIO;

        mutex_lock(&ctrl_info->scan_mutex);

        rc = pqi_update_scsi_devices(ctrl_info);
        if (rc)
                pqi_schedule_rescan_worker_delayed(ctrl_info);

        mutex_unlock(&ctrl_info->scan_mutex);

        return rc;
}

static void pqi_scan_start(struct Scsi_Host *shost)
{
        pqi_scan_scsi_devices(shost_to_hba(shost));
}

/* Returns TRUE if scan is finished. */

static int pqi_scan_finished(struct Scsi_Host *shost,
        unsigned long elapsed_time)
{
        struct pqi_ctrl_info *ctrl_info;

        ctrl_info = shost_priv(shost);

        return !mutex_is_locked(&ctrl_info->scan_mutex);
}

static void pqi_wait_until_scan_finished(struct pqi_ctrl_info *ctrl_info)
{
        mutex_lock(&ctrl_info->scan_mutex);
        mutex_unlock(&ctrl_info->scan_mutex);
}

static void pqi_wait_until_lun_reset_finished(struct pqi_ctrl_info *ctrl_info)
{
        mutex_lock(&ctrl_info->lun_reset_mutex);
        mutex_unlock(&ctrl_info->lun_reset_mutex);
}

static inline void pqi_set_encryption_info(
        struct pqi_encryption_info *encryption_info, struct raid_map *raid_map,
        u64 first_block)
{
        u32 volume_blk_size;

        /*
         * Set the encryption tweak values based on logical block address.
         * If the block size is 512, the tweak value is equal to the LBA.
         * For other block sizes, tweak value is (LBA * block size) / 512.
         */
        volume_blk_size = get_unaligned_le32(&raid_map->volume_blk_size);
        if (volume_blk_size != 512)
                first_block = (first_block * volume_blk_size) / 512;

        encryption_info->data_encryption_key_index =
                get_unaligned_le16(&raid_map->data_encryption_key_index);
        encryption_info->encrypt_tweak_lower = lower_32_bits(first_block);
        encryption_info->encrypt_tweak_upper = upper_32_bits(first_block);
}

/*
 * Attempt to perform RAID bypass mapping for a logical volume I/O.
 */

#define PQI_RAID_BYPASS_INELIGIBLE      1

static int pqi_raid_bypass_submit_scsi_cmd(struct pqi_ctrl_info *ctrl_info,
        struct pqi_scsi_dev *device, struct scsi_cmnd *scmd,
        struct pqi_queue_group *queue_group)
{
        struct raid_map *raid_map;
        bool is_write = false;
        u32 map_index;
        u64 first_block;
        u64 last_block;
        u32 block_cnt;
        u32 blocks_per_row;
        u64 first_row;
        u64 last_row;
        u32 first_row_offset;
        u32 last_row_offset;
        u32 first_column;
        u32 last_column;
        u64 r0_first_row;
        u64 r0_last_row;
        u32 r5or6_blocks_per_row;
        u64 r5or6_first_row;
        u64 r5or6_last_row;
        u32 r5or6_first_row_offset;
        u32 r5or6_last_row_offset;
        u32 r5or6_first_column;
        u32 r5or6_last_column;
        u16 data_disks_per_row;
        u32 total_disks_per_row;
        u16 layout_map_count;
        u32 stripesize;
        u16 strip_size;
        u32 first_group;
        u32 last_group;
        u32 current_group;
        u32 map_row;
        u32 aio_handle;
        u64 disk_block;
        u32 disk_block_cnt;
        u8 cdb[16];
        u8 cdb_length;
        int offload_to_mirror;
        struct pqi_encryption_info *encryption_info_ptr;
        struct pqi_encryption_info encryption_info;
#if BITS_PER_LONG == 32
        u64 tmpdiv;
#endif

        /* Check for valid opcode, get LBA and block count. */
        switch (scmd->cmnd[0]) {
        case WRITE_6:
                is_write = true;
                /* fall through */
        case READ_6:
                first_block = (u64)(((scmd->cmnd[1] & 0x1f) << 16) |
                        (scmd->cmnd[2] << 8) | scmd->cmnd[3]);
                block_cnt = (u32)scmd->cmnd[4];
                if (block_cnt == 0)
                        block_cnt = 256;
                break;
        case WRITE_10:
                is_write = true;
                /* fall through */
        case READ_10:
                first_block = (u64)get_unaligned_be32(&scmd->cmnd[2]);
                block_cnt = (u32)get_unaligned_be16(&scmd->cmnd[7]);
                break;
        case WRITE_12:
                is_write = true;
                /* fall through */
        case READ_12:
                first_block = (u64)get_unaligned_be32(&scmd->cmnd[2]);
                block_cnt = get_unaligned_be32(&scmd->cmnd[6]);
                break;
        case WRITE_16:
                is_write = true;
                /* fall through */
        case READ_16:
                first_block = get_unaligned_be64(&scmd->cmnd[2]);
                block_cnt = get_unaligned_be32(&scmd->cmnd[10]);
                break;
        default:
                /* Process via normal I/O path. */
                return PQI_RAID_BYPASS_INELIGIBLE;
        }

        /* Check for write to non-RAID-0. */
        if (is_write && device->raid_level != SA_RAID_0)
                return PQI_RAID_BYPASS_INELIGIBLE;

        if (unlikely(block_cnt == 0))
                return PQI_RAID_BYPASS_INELIGIBLE;

        last_block = first_block + block_cnt - 1;
        raid_map = device->raid_map;

        /* Check for invalid block or wraparound. */
        if (last_block >= get_unaligned_le64(&raid_map->volume_blk_cnt) ||
                last_block < first_block)
                return PQI_RAID_BYPASS_INELIGIBLE;

        data_disks_per_row = get_unaligned_le16(&raid_map->data_disks_per_row);
        strip_size = get_unaligned_le16(&raid_map->strip_size);
        layout_map_count = get_unaligned_le16(&raid_map->layout_map_count);

        /* Calculate stripe information for the request. */
        blocks_per_row = data_disks_per_row * strip_size;
#if BITS_PER_LONG == 32
        tmpdiv = first_block;
        do_div(tmpdiv, blocks_per_row);
        first_row = tmpdiv;
        tmpdiv = last_block;
        do_div(tmpdiv, blocks_per_row);
        last_row = tmpdiv;
        first_row_offset = (u32)(first_block - (first_row * blocks_per_row));
        last_row_offset = (u32)(last_block - (last_row * blocks_per_row));
        tmpdiv = first_row_offset;
        do_div(tmpdiv, strip_size);
        first_column = tmpdiv;
        tmpdiv = last_row_offset;
        do_div(tmpdiv, strip_size);
        last_column = tmpdiv;
#else
        first_row = first_block / blocks_per_row;
        last_row = last_block / blocks_per_row;
        first_row_offset = (u32)(first_block - (first_row * blocks_per_row));
        last_row_offset = (u32)(last_block - (last_row * blocks_per_row));
        first_column = first_row_offset / strip_size;
        last_column = last_row_offset / strip_size;
#endif

        /* If this isn't a single row/column then give to the controller. */
        if (first_row != last_row || first_column != last_column)
                return PQI_RAID_BYPASS_INELIGIBLE;

        /* Proceeding with driver mapping. */
        total_disks_per_row = data_disks_per_row +
                get_unaligned_le16(&raid_map->metadata_disks_per_row);
        map_row = ((u32)(first_row >> raid_map->parity_rotation_shift)) %
                get_unaligned_le16(&raid_map->row_cnt);
        map_index = (map_row * total_disks_per_row) + first_column;

        /* RAID 1 */
        if (device->raid_level == SA_RAID_1) {
                if (device->offload_to_mirror)
                        map_index += data_disks_per_row;
                device->offload_to_mirror = !device->offload_to_mirror;
        } else if (device->raid_level == SA_RAID_ADM) {
                /* RAID ADM */
                /*
                 * Handles N-way mirrors  (R1-ADM) and R10 with # of drives
                 * divisible by 3.
                 */
                offload_to_mirror = device->offload_to_mirror;
                if (offload_to_mirror == 0)  {
                        /* use physical disk in the first mirrored group. */
                        map_index %= data_disks_per_row;
                } else {
                        do {
                                /*
                                 * Determine mirror group that map_index
                                 * indicates.
                                 */
                                current_group = map_index / data_disks_per_row;

                                if (offload_to_mirror != current_group) {
                                        if (current_group <
                                                layout_map_count - 1) {
                                                /*
                                                 * Select raid index from
                                                 * next group.
                                                 */
                                                map_index += data_disks_per_row;
                                                current_group++;
                                        } else {
                                                /*
                                                 * Select raid index from first
                                                 * group.
                                                 */
                                                map_index %= data_disks_per_row;
                                                current_group = 0;
                                        }
                                }
                        } while (offload_to_mirror != current_group);
                }

                /* Set mirror group to use next time. */
                offload_to_mirror =
                        (offload_to_mirror >= layout_map_count - 1) ?
                                0 : offload_to_mirror + 1;
                WARN_ON(offload_to_mirror >= layout_map_count);
                device->offload_to_mirror = offload_to_mirror;
                /*
                 * Avoid direct use of device->offload_to_mirror within this
                 * function since multiple threads might simultaneously
                 * increment it beyond the range of device->layout_map_count -1.
                 */
        } else if ((device->raid_level == SA_RAID_5 ||
                device->raid_level == SA_RAID_6) && layout_map_count > 1) {
                /* RAID 50/60 */
                /* Verify first and last block are in same RAID group */
                r5or6_blocks_per_row = strip_size * data_disks_per_row;
                stripesize = r5or6_blocks_per_row * layout_map_count;
#if BITS_PER_LONG == 32
                tmpdiv = first_block;
                first_group = do_div(tmpdiv, stripesize);
                tmpdiv = first_group;
                do_div(tmpdiv, r5or6_blocks_per_row);
                first_group = tmpdiv;
                tmpdiv = last_block;
                last_group = do_div(tmpdiv, stripesize);
                tmpdiv = last_group;
                do_div(tmpdiv, r5or6_blocks_per_row);
                last_group = tmpdiv;
#else
                first_group = (first_block % stripesize) / r5or6_blocks_per_row;
                last_group = (last_block % stripesize) / r5or6_blocks_per_row;
#endif
                if (first_group != last_group)
                        return PQI_RAID_BYPASS_INELIGIBLE;

                /* Verify request is in a single row of RAID 5/6 */
#if BITS_PER_LONG == 32
                tmpdiv = first_block;
                do_div(tmpdiv, stripesize);
                first_row = r5or6_first_row = r0_first_row = tmpdiv;
                tmpdiv = last_block;
                do_div(tmpdiv, stripesize);
                r5or6_last_row = r0_last_row = tmpdiv;
#else
                first_row = r5or6_first_row = r0_first_row =
                        first_block / stripesize;
                r5or6_last_row = r0_last_row = last_block / stripesize;
#endif
                if (r5or6_first_row != r5or6_last_row)
                        return PQI_RAID_BYPASS_INELIGIBLE;

                /* Verify request is in a single column */
#if BITS_PER_LONG == 32
                tmpdiv = first_block;
                first_row_offset = do_div(tmpdiv, stripesize);
                tmpdiv = first_row_offset;
                first_row_offset = (u32)do_div(tmpdiv, r5or6_blocks_per_row);
                r5or6_first_row_offset = first_row_offset;
                tmpdiv = last_block;
                r5or6_last_row_offset = do_div(tmpdiv, stripesize);
                tmpdiv = r5or6_last_row_offset;
                r5or6_last_row_offset = do_div(tmpdiv, r5or6_blocks_per_row);
                tmpdiv = r5or6_first_row_offset;
                do_div(tmpdiv, strip_size);
                first_column = r5or6_first_column = tmpdiv;
                tmpdiv = r5or6_last_row_offset;
                do_div(tmpdiv, strip_size);
                r5or6_last_column = tmpdiv;
#else
                first_row_offset = r5or6_first_row_offset =
                        (u32)((first_block % stripesize) %
                        r5or6_blocks_per_row);

                r5or6_last_row_offset =
                        (u32)((last_block % stripesize) %
                        r5or6_blocks_per_row);

                first_column = r5or6_first_row_offset / strip_size;
                r5or6_first_column = first_column;
                r5or6_last_column = r5or6_last_row_offset / strip_size;
#endif
                if (r5or6_first_column != r5or6_last_column)
                        return PQI_RAID_BYPASS_INELIGIBLE;

                /* Request is eligible */
                map_row =
                        ((u32)(first_row >> raid_map->parity_rotation_shift)) %
                        get_unaligned_le16(&raid_map->row_cnt);

                map_index = (first_group *
                        (get_unaligned_le16(&raid_map->row_cnt) *
                        total_disks_per_row)) +
                        (map_row * total_disks_per_row) + first_column;
        }

        if (unlikely(map_index >= RAID_MAP_MAX_ENTRIES))
                return PQI_RAID_BYPASS_INELIGIBLE;

        aio_handle = raid_map->disk_data[map_index].aio_handle;
        disk_block = get_unaligned_le64(&raid_map->disk_starting_blk) +
                first_row * strip_size +
                (first_row_offset - first_column * strip_size);
        disk_block_cnt = block_cnt;

        /* Handle differing logical/physical block sizes. */
        if (raid_map->phys_blk_shift) {
                disk_block <<= raid_map->phys_blk_shift;
                disk_block_cnt <<= raid_map->phys_blk_shift;
        }

        if (unlikely(disk_block_cnt > 0xffff))
                return PQI_RAID_BYPASS_INELIGIBLE;

        /* Build the new CDB for the physical disk I/O. */
        if (disk_block > 0xffffffff) {
                cdb[0] = is_write ? WRITE_16 : READ_16;
                cdb[1] = 0;
                put_unaligned_be64(disk_block, &cdb[2]);
                put_unaligned_be32(disk_block_cnt, &cdb[10]);
                cdb[14] = 0;
                cdb[15] = 0;
                cdb_length = 16;
        } else {
                cdb[0] = is_write ? WRITE_10 : READ_10;
                cdb[1] = 0;
                put_unaligned_be32((u32)disk_block, &cdb[2]);
                cdb[6] = 0;
                put_unaligned_be16((u16)disk_block_cnt, &cdb[7]);
                cdb[9] = 0;
                cdb_length = 10;
        }

        if (get_unaligned_le16(&raid_map->flags) &
                RAID_MAP_ENCRYPTION_ENABLED) {
                pqi_set_encryption_info(&encryption_info, raid_map,
                        first_block);
                encryption_info_ptr = &encryption_info;
        } else {
                encryption_info_ptr = NULL;
        }

        return pqi_aio_submit_io(ctrl_info, scmd, aio_handle,
                cdb, cdb_length, queue_group, encryption_info_ptr, true);
}

#define PQI_STATUS_IDLE         0x0

#define PQI_CREATE_ADMIN_QUEUE_PAIR     1
#define PQI_DELETE_ADMIN_QUEUE_PAIR     2

#define PQI_DEVICE_STATE_POWER_ON_AND_RESET             0x0
#define PQI_DEVICE_STATE_STATUS_AVAILABLE               0x1
#define PQI_DEVICE_STATE_ALL_REGISTERS_READY            0x2
#define PQI_DEVICE_STATE_ADMIN_QUEUE_PAIR_READY         0x3
#define PQI_DEVICE_STATE_ERROR                          0x4

#define PQI_MODE_READY_TIMEOUT_SECS             30
#define PQI_MODE_READY_POLL_INTERVAL_MSECS      1

static int pqi_wait_for_pqi_mode_ready(struct pqi_ctrl_info *ctrl_info)
{
        struct pqi_device_registers __iomem *pqi_registers;
        unsigned long timeout;
        u64 signature;
        u8 status;

        pqi_registers = ctrl_info->pqi_registers;
        timeout = (PQI_MODE_READY_TIMEOUT_SECS * HZ) + jiffies;

        while (1) {
                signature = readq(&pqi_registers->signature);
                if (memcmp(&signature, PQI_DEVICE_SIGNATURE,
                        sizeof(signature)) == 0)
                        break;
                if (time_after(jiffies, timeout)) {
                        dev_err(&ctrl_info->pci_dev->dev,
                                "timed out waiting for PQI signature\n");
                        return -ETIMEDOUT;
                }
                msleep(PQI_MODE_READY_POLL_INTERVAL_MSECS);
        }

        while (1) {
                status = readb(&pqi_registers->function_and_status_code);
                if (status == PQI_STATUS_IDLE)
                        break;
                if (time_after(jiffies, timeout)) {
                        dev_err(&ctrl_info->pci_dev->dev,
                                "timed out waiting for PQI IDLE\n");
                        return -ETIMEDOUT;
                }
                msleep(PQI_MODE_READY_POLL_INTERVAL_MSECS);
        }

        while (1) {
                if (readl(&pqi_registers->device_status) ==
                        PQI_DEVICE_STATE_ALL_REGISTERS_READY)
                        break;
                if (time_after(jiffies, timeout)) {
                        dev_err(&ctrl_info->pci_dev->dev,
                                "timed out waiting for PQI all registers ready\n");
                        return -ETIMEDOUT;
                }
                msleep(PQI_MODE_READY_POLL_INTERVAL_MSECS);
        }

        return 0;
}

static inline void pqi_aio_path_disabled(struct pqi_io_request *io_request)
{
        struct pqi_scsi_dev *device;

        device = io_request->scmd->device->hostdata;
        device->raid_bypass_enabled = false;
        device->aio_enabled = false;
}

static inline void pqi_take_device_offline(struct scsi_device *sdev, char *path)
{
        struct pqi_ctrl_info *ctrl_info;
        struct pqi_scsi_dev *device;

        device = sdev->hostdata;
        if (device->device_offline)
                return;

        device->device_offline = true;
        scsi_device_set_state(sdev, SDEV_OFFLINE);
        ctrl_info = shost_to_hba(sdev->host);
        pqi_schedule_rescan_worker(ctrl_info);
        dev_err(&ctrl_info->pci_dev->dev, "offlined %s scsi %d:%d:%d:%d\n",
                path, ctrl_info->scsi_host->host_no, device->bus,
                device->target, device->lun);
}

static void pqi_process_raid_io_error(struct pqi_io_request *io_request)
{
        u8 scsi_status;
        u8 host_byte;
        struct scsi_cmnd *scmd;
        struct pqi_raid_error_info *error_info;
        size_t sense_data_length;
        int residual_count;
        int xfer_count;
        struct scsi_sense_hdr sshdr;

        scmd = io_request->scmd;
        if (!scmd)
                return;

        error_info = io_request->error_info;
        scsi_status = error_info->status;
        host_byte = DID_OK;

        switch (error_info->data_out_result) {
        case PQI_DATA_IN_OUT_GOOD:
                break;
        case PQI_DATA_IN_OUT_UNDERFLOW:
                xfer_count =
                        get_unaligned_le32(&error_info->data_out_transferred);
                residual_count = scsi_bufflen(scmd) - xfer_count;
                scsi_set_resid(scmd, residual_count);
                if (xfer_count < scmd->underflow)
                        host_byte = DID_SOFT_ERROR;
                break;
        case PQI_DATA_IN_OUT_UNSOLICITED_ABORT:
        case PQI_DATA_IN_OUT_ABORTED:
                host_byte = DID_ABORT;
                break;
        case PQI_DATA_IN_OUT_TIMEOUT:
                host_byte = DID_TIME_OUT;
                break;
        case PQI_DATA_IN_OUT_BUFFER_OVERFLOW:
        case PQI_DATA_IN_OUT_PROTOCOL_ERROR:
        case PQI_DATA_IN_OUT_BUFFER_ERROR:
        case PQI_DATA_IN_OUT_BUFFER_OVERFLOW_DESCRIPTOR_AREA:
        case PQI_DATA_IN_OUT_BUFFER_OVERFLOW_BRIDGE:
        case PQI_DATA_IN_OUT_ERROR:
        case PQI_DATA_IN_OUT_HARDWARE_ERROR:
        case PQI_DATA_IN_OUT_PCIE_FABRIC_ERROR:
        case PQI_DATA_IN_OUT_PCIE_COMPLETION_TIMEOUT:
        case PQI_DATA_IN_OUT_PCIE_COMPLETER_ABORT_RECEIVED:
        case PQI_DATA_IN_OUT_PCIE_UNSUPPORTED_REQUEST_RECEIVED:
        case PQI_DATA_IN_OUT_PCIE_ECRC_CHECK_FAILED:
        case PQI_DATA_IN_OUT_PCIE_UNSUPPORTED_REQUEST:
        case PQI_DATA_IN_OUT_PCIE_ACS_VIOLATION:
        case PQI_DATA_IN_OUT_PCIE_TLP_PREFIX_BLOCKED:
        case PQI_DATA_IN_OUT_PCIE_POISONED_MEMORY_READ:
        default:
                host_byte = DID_ERROR;
                break;
        }

        sense_data_length = get_unaligned_le16(&error_info->sense_data_length);
        if (sense_data_length == 0)
                sense_data_length =
                        get_unaligned_le16(&error_info->response_data_length);
        if (sense_data_length) {
                if (sense_data_length > sizeof(error_info->data))
                        sense_data_length = sizeof(error_info->data);

                if (scsi_status == SAM_STAT_CHECK_CONDITION &&
                        scsi_normalize_sense(error_info->data,
                                sense_data_length, &sshdr) &&
                                sshdr.sense_key == HARDWARE_ERROR &&
                                sshdr.asc == 0x3e &&
                                sshdr.ascq == 0x1) {
                        pqi_take_device_offline(scmd->device, "RAID");
                        host_byte = DID_NO_CONNECT;
                }

                if (sense_data_length > SCSI_SENSE_BUFFERSIZE)
                        sense_data_length = SCSI_SENSE_BUFFERSIZE;
                memcpy(scmd->sense_buffer, error_info->data,
                        sense_data_length);
        }

        scmd->result = scsi_status;
        set_host_byte(scmd, host_byte);
}

static void pqi_process_aio_io_error(struct pqi_io_request *io_request)
{
        u8 scsi_status;
        u8 host_byte;
        struct scsi_cmnd *scmd;
        struct pqi_aio_error_info *error_info;
        size_t sense_data_length;
        int residual_count;
        int xfer_count;
        bool device_offline;

        scmd = io_request->scmd;
        error_info = io_request->error_info;
        host_byte = DID_OK;
        sense_data_length = 0;
        device_offline = false;

        switch (error_info->service_response) {
        case PQI_AIO_SERV_RESPONSE_COMPLETE:
                scsi_status = error_info->status;
                break;
        case PQI_AIO_SERV_RESPONSE_FAILURE:
                switch (error_info->status) {
                case PQI_AIO_STATUS_IO_ABORTED:
                        scsi_status = SAM_STAT_TASK_ABORTED;
                        break;
                case PQI_AIO_STATUS_UNDERRUN:
                        scsi_status = SAM_STAT_GOOD;
                        residual_count = get_unaligned_le32(
                                                &error_info->residual_count);
                        scsi_set_resid(scmd, residual_count);
                        xfer_count = scsi_bufflen(scmd) - residual_count;
                        if (xfer_count < scmd->underflow)
                                host_byte = DID_SOFT_ERROR;
                        break;
                case PQI_AIO_STATUS_OVERRUN:
                        scsi_status = SAM_STAT_GOOD;
                        break;
                case PQI_AIO_STATUS_AIO_PATH_DISABLED:
                        pqi_aio_path_disabled(io_request);
                        scsi_status = SAM_STAT_GOOD;
                        io_request->status = -EAGAIN;
                        break;
                case PQI_AIO_STATUS_NO_PATH_TO_DEVICE:
                case PQI_AIO_STATUS_INVALID_DEVICE:
                        if (!io_request->raid_bypass) {
                                device_offline = true;
                                pqi_take_device_offline(scmd->device, "AIO");
                                host_byte = DID_NO_CONNECT;
                        }
                        scsi_status = SAM_STAT_CHECK_CONDITION;
                        break;
                case PQI_AIO_STATUS_IO_ERROR:
                default:
                        scsi_status = SAM_STAT_CHECK_CONDITION;
                        break;
                }
                break;
        case PQI_AIO_SERV_RESPONSE_TMF_COMPLETE:
        case PQI_AIO_SERV_RESPONSE_TMF_SUCCEEDED:
                scsi_status = SAM_STAT_GOOD;
                break;
        case PQI_AIO_SERV_RESPONSE_TMF_REJECTED:
        case PQI_AIO_SERV_RESPONSE_TMF_INCORRECT_LUN:
        default:
                scsi_status = SAM_STAT_CHECK_CONDITION;
                break;
        }

        if (error_info->data_present) {
                sense_data_length =
                        get_unaligned_le16(&error_info->data_length);
                if (sense_data_length) {
                        if (sense_data_length > sizeof(error_info->data))
                                sense_data_length = sizeof(error_info->data);
                        if (sense_data_length > SCSI_SENSE_BUFFERSIZE)
                                sense_data_length = SCSI_SENSE_BUFFERSIZE;
                        memcpy(scmd->sense_buffer, error_info->data,
                                sense_data_length);
                }
        }

        if (device_offline && sense_data_length == 0)
                scsi_build_sense_buffer(0, scmd->sense_buffer, HARDWARE_ERROR,
                        0x3e, 0x1);

        scmd->result = scsi_status;
        set_host_byte(scmd, host_byte);
}

static void pqi_process_io_error(unsigned int iu_type,
        struct pqi_io_request *io_request)
{
        switch (iu_type) {
        case PQI_RESPONSE_IU_RAID_PATH_IO_ERROR:
                pqi_process_raid_io_error(io_request);
                break;
        case PQI_RESPONSE_IU_AIO_PATH_IO_ERROR:
                pqi_process_aio_io_error(io_request);
                break;
        }
}

static int pqi_interpret_task_management_response(
        struct pqi_task_management_response *response)
{
        int rc;

        switch (response->response_code) {
        case SOP_TMF_COMPLETE:
        case SOP_TMF_FUNCTION_SUCCEEDED:
                rc = 0;
                break;
        default:
                rc = -EIO;
                break;
        }

        return rc;
}

static unsigned int pqi_process_io_intr(struct pqi_ctrl_info *ctrl_info,
        struct pqi_queue_group *queue_group)
{
        unsigned int num_responses;
        pqi_index_t oq_pi;
        pqi_index_t oq_ci;
        struct pqi_io_request *io_request;
        struct pqi_io_response *response;
        u16 request_id;

        num_responses = 0;
        oq_ci = queue_group->oq_ci_copy;

        while (1) {
                oq_pi = *queue_group->oq_pi;
                if (oq_pi == oq_ci)
                        break;

                num_responses++;
                response = queue_group->oq_element_array +
                        (oq_ci * PQI_OPERATIONAL_OQ_ELEMENT_LENGTH);

                request_id = get_unaligned_le16(&response->request_id);
                WARN_ON(request_id >= ctrl_info->max_io_slots);

                io_request = &ctrl_info->io_request_pool[request_id];
                WARN_ON(atomic_read(&io_request->refcount) == 0);

                switch (response->header.iu_type) {
                case PQI_RESPONSE_IU_RAID_PATH_IO_SUCCESS:
                case PQI_RESPONSE_IU_AIO_PATH_IO_SUCCESS:
                case PQI_RESPONSE_IU_GENERAL_MANAGEMENT:
                        break;
                case PQI_RESPONSE_IU_TASK_MANAGEMENT:
                        io_request->status =
                                pqi_interpret_task_management_response(
                                        (void *)response);
                        break;
                case PQI_RESPONSE_IU_AIO_PATH_DISABLED:
                        pqi_aio_path_disabled(io_request);
                        io_request->status = -EAGAIN;
                        break;
                case PQI_RESPONSE_IU_RAID_PATH_IO_ERROR:
                case PQI_RESPONSE_IU_AIO_PATH_IO_ERROR:
                        io_request->error_info = ctrl_info->error_buffer +
                                (get_unaligned_le16(&response->error_index) *
                                PQI_ERROR_BUFFER_ELEMENT_LENGTH);
                        pqi_process_io_error(response->header.iu_type,
                                io_request);
                        break;
                default:
                        dev_err(&ctrl_info->pci_dev->dev,
                                "unexpected IU type: 0x%x\n",
                                response->header.iu_type);
                        break;
                }

                io_request->io_complete_callback(io_request,
                        io_request->context);

                /*
                 * Note that the I/O request structure CANNOT BE TOUCHED after
                 * returning from the I/O completion callback!
                 */

                oq_ci = (oq_ci + 1) % ctrl_info->num_elements_per_oq;
        }

        if (num_responses) {
                queue_group->oq_ci_copy = oq_ci;
                writel(oq_ci, queue_group->oq_ci);
        }

        return num_responses;
}

static inline unsigned int pqi_num_elements_free(unsigned int pi,
        unsigned int ci, unsigned int elements_in_queue)
{
        unsigned int num_elements_used;

        if (pi >= ci)
                num_elements_used = pi - ci;
        else
                num_elements_used = elements_in_queue - ci + pi;

        return elements_in_queue - num_elements_used - 1;
}

static void pqi_send_event_ack(struct pqi_ctrl_info *ctrl_info,
        struct pqi_event_acknowledge_request *iu, size_t iu_length)
{
        pqi_index_t iq_pi;
        pqi_index_t iq_ci;
        unsigned long flags;
        void *next_element;
        struct pqi_queue_group *queue_group;

        queue_group = &ctrl_info->queue_groups[PQI_DEFAULT_QUEUE_GROUP];
        put_unaligned_le16(queue_group->oq_id, &iu->header.response_queue_id);

        while (1) {
                spin_lock_irqsave(&queue_group->submit_lock[RAID_PATH], flags);

                iq_pi = queue_group->iq_pi_copy[RAID_PATH];
                iq_ci = *queue_group->iq_ci[RAID_PATH];

                if (pqi_num_elements_free(iq_pi, iq_ci,
                        ctrl_info->num_elements_per_iq))
                        break;

                spin_unlock_irqrestore(
                        &queue_group->submit_lock[RAID_PATH], flags);

                if (pqi_ctrl_offline(ctrl_info))
                        return;
        }

        next_element = queue_group->iq_element_array[RAID_PATH] +
                (iq_pi * PQI_OPERATIONAL_IQ_ELEMENT_LENGTH);

        memcpy(next_element, iu, iu_length);

        iq_pi = (iq_pi + 1) % ctrl_info->num_elements_per_iq;
        queue_group->iq_pi_copy[RAID_PATH] = iq_pi;

        /*
         * This write notifies the controller that an IU is available to be
         * processed.
         */
        writel(iq_pi, queue_group->iq_pi[RAID_PATH]);

        spin_unlock_irqrestore(&queue_group->submit_lock[RAID_PATH], flags);
}

static void pqi_acknowledge_event(struct pqi_ctrl_info *ctrl_info,
        struct pqi_event *event)
{
        struct pqi_event_acknowledge_request request;

        memset(&request, 0, sizeof(request));

        request.header.iu_type = PQI_REQUEST_IU_ACKNOWLEDGE_VENDOR_EVENT;
        put_unaligned_le16(sizeof(request) - PQI_REQUEST_HEADER_LENGTH,
                &request.header.iu_length);
        request.event_type = event->event_type;
        request.event_id = event->event_id;
        request.additional_event_id = event->additional_event_id;

        pqi_send_event_ack(ctrl_info, &request, sizeof(request));
}

static void pqi_event_worker(struct work_struct *work)
{
        unsigned int i;
        struct pqi_ctrl_info *ctrl_info;
        struct pqi_event *event;

        ctrl_info = container_of(work, struct pqi_ctrl_info, event_work);

        pqi_ctrl_busy(ctrl_info);
        pqi_wait_if_ctrl_blocked(ctrl_info, NO_TIMEOUT);
        if (pqi_ctrl_offline(ctrl_info))
                goto out;

        pqi_schedule_rescan_worker_delayed(ctrl_info);

        event = ctrl_info->events;
        for (i = 0; i < PQI_NUM_SUPPORTED_EVENTS; i++) {
                if (event->pending) {
                        event->pending = false;
                        pqi_acknowledge_event(ctrl_info, event);
                }
                event++;
        }

out:
        pqi_ctrl_unbusy(ctrl_info);
}

#define PQI_HEARTBEAT_TIMER_INTERVAL    (10 * HZ)

static void pqi_heartbeat_timer_handler(unsigned long data)
{
        int num_interrupts;
        u32 heartbeat_count;
        struct pqi_ctrl_info *ctrl_info = (struct pqi_ctrl_info *)data;

        pqi_check_ctrl_health(ctrl_info);
        if (pqi_ctrl_offline(ctrl_info))
                return;

        num_interrupts = atomic_read(&ctrl_info->num_interrupts);
        heartbeat_count = pqi_read_heartbeat_counter(ctrl_info);

        if (num_interrupts == ctrl_info->previous_num_interrupts) {
                if (heartbeat_count == ctrl_info->previous_heartbeat_count) {
                        dev_err(&ctrl_info->pci_dev->dev,
                                "no heartbeat detected - last heartbeat count: %u\n",
                                heartbeat_count);
                        pqi_take_ctrl_offline(ctrl_info);
                        return;
                }
        } else {
                ctrl_info->previous_num_interrupts = num_interrupts;
        }

        ctrl_info->previous_heartbeat_count = heartbeat_count;
        mod_timer(&ctrl_info->heartbeat_timer,
                jiffies + PQI_HEARTBEAT_TIMER_INTERVAL);
}

static void pqi_start_heartbeat_timer(struct pqi_ctrl_info *ctrl_info)
{
        if (!ctrl_info->heartbeat_counter)
                return;

        ctrl_info->previous_num_interrupts =
                atomic_read(&ctrl_info->num_interrupts);
        ctrl_info->previous_heartbeat_count =
                pqi_read_heartbeat_counter(ctrl_info);

        ctrl_info->heartbeat_timer.expires =
                jiffies + PQI_HEARTBEAT_TIMER_INTERVAL;
        ctrl_info->heartbeat_timer.data = (unsigned long)ctrl_info;
        ctrl_info->heartbeat_timer.function = pqi_heartbeat_timer_handler;
        add_timer(&ctrl_info->heartbeat_timer);
}

static inline void pqi_stop_heartbeat_timer(struct pqi_ctrl_info *ctrl_info)
{
        del_timer_sync(&ctrl_info->heartbeat_timer);
}

static inline int pqi_event_type_to_event_index(unsigned int event_type)
{
        int index;

        for (index = 0; index < ARRAY_SIZE(pqi_supported_event_types); index++)
                if (event_type == pqi_supported_event_types[index])
                        return index;

        return -1;
}

static inline bool pqi_is_supported_event(unsigned int event_type)
{
        return pqi_event_type_to_event_index(event_type) != -1;
}

static unsigned int pqi_process_event_intr(struct pqi_ctrl_info *ctrl_info)
{
        unsigned int num_events;
        pqi_index_t oq_pi;
        pqi_index_t oq_ci;
        struct pqi_event_queue *event_queue;
        struct pqi_event_response *response;
        struct pqi_event *event;
        int event_index;

        event_queue = &ctrl_info->event_queue;
        num_events = 0;
        oq_ci = event_queue->oq_ci_copy;

        while (1) {
                oq_pi = *event_queue->oq_pi;
                if (oq_pi == oq_ci)
                        break;

                num_events++;
                response = event_queue->oq_element_array +
                        (oq_ci * PQI_EVENT_OQ_ELEMENT_LENGTH);

                event_index =
                        pqi_event_type_to_event_index(response->event_type);

                if (event_index >= 0) {
                        if (response->request_acknowlege) {
                                event = &ctrl_info->events[event_index];
                                event->pending = true;
                                event->event_type = response->event_type;
                                event->event_id = response->event_id;
                                event->additional_event_id =
                                        response->additional_event_id;
                        }
                }

                oq_ci = (oq_ci + 1) % PQI_NUM_EVENT_QUEUE_ELEMENTS;
        }

        if (num_events) {
                event_queue->oq_ci_copy = oq_ci;
                writel(oq_ci, event_queue->oq_ci);
                schedule_work(&ctrl_info->event_work);
        }

        return num_events;
}

#define PQI_LEGACY_INTX_MASK    0x1

static inline void pqi_configure_legacy_intx(struct pqi_ctrl_info *ctrl_info,
                                                bool enable_intx)
{
        u32 intx_mask;
        struct pqi_device_registers __iomem *pqi_registers;
        volatile void __iomem *register_addr;

        pqi_registers = ctrl_info->pqi_registers;

        if (enable_intx)
                register_addr = &pqi_registers->legacy_intx_mask_clear;
        else
                register_addr = &pqi_registers->legacy_intx_mask_set;

        intx_mask = readl(register_addr);
        intx_mask |= PQI_LEGACY_INTX_MASK;
        writel(intx_mask, register_addr);
}

static void pqi_change_irq_mode(struct pqi_ctrl_info *ctrl_info,

        enum pqi_irq_mode new_mode)
{
        switch (ctrl_info->irq_mode) {
        case IRQ_MODE_MSIX:
                switch (new_mode) {
                case IRQ_MODE_MSIX:
                        break;
                case IRQ_MODE_INTX:
                        pqi_configure_legacy_intx(ctrl_info, true);
                        sis_disable_msix(ctrl_info);
                        sis_enable_intx(ctrl_info);
                        break;
                case IRQ_MODE_NONE:
                        sis_disable_msix(ctrl_info);
                        break;
                }
                break;
        case IRQ_MODE_INTX:
                switch (new_mode) {
                case IRQ_MODE_MSIX:
                        pqi_configure_legacy_intx(ctrl_info, false);
                        sis_disable_intx(ctrl_info);
                        sis_enable_msix(ctrl_info);
                        break;
                case IRQ_MODE_INTX:
                        break;
                case IRQ_MODE_NONE:
                        pqi_configure_legacy_intx(ctrl_info, false);
                        sis_disable_intx(ctrl_info);
                        break;
                }
                break;
        case IRQ_MODE_NONE:
                switch (new_mode) {
                case IRQ_MODE_MSIX:
                        sis_enable_msix(ctrl_info);
                        break;
                case IRQ_MODE_INTX:
                        pqi_configure_legacy_intx(ctrl_info, true);
                        sis_enable_intx(ctrl_info);
                        break;
                case IRQ_MODE_NONE:
                        break;
                }
                break;
        }

        ctrl_info->irq_mode = new_mode;
}

#define PQI_LEGACY_INTX_PENDING         0x1

static inline bool pqi_is_valid_irq(struct pqi_ctrl_info *ctrl_info)
{
        bool valid_irq;
        u32 intx_status;

        switch (ctrl_info->irq_mode) {
        case IRQ_MODE_MSIX:
                valid_irq = true;
                break;
        case IRQ_MODE_INTX:
                intx_status =
                        readl(&ctrl_info->pqi_registers->legacy_intx_status);
                if (intx_status & PQI_LEGACY_INTX_PENDING)
                        valid_irq = true;
                else
                        valid_irq = false;
                break;
        case IRQ_MODE_NONE:
        default:
                valid_irq = false;
                break;
        }

        return valid_irq;
}

static irqreturn_t pqi_irq_handler(int irq, void *data)
{
        struct pqi_ctrl_info *ctrl_info;
        struct pqi_queue_group *queue_group;
        unsigned int num_responses_handled;

        queue_group = data;
        ctrl_info = queue_group->ctrl_info;

        if (!pqi_is_valid_irq(ctrl_info))
                return IRQ_NONE;

        num_responses_handled = pqi_process_io_intr(ctrl_info, queue_group);

        if (irq == ctrl_info->event_irq)
                num_responses_handled += pqi_process_event_intr(ctrl_info);

        if (num_responses_handled)
                atomic_inc(&ctrl_info->num_interrupts);

        pqi_start_io(ctrl_info, queue_group, RAID_PATH, NULL);
        pqi_start_io(ctrl_info, queue_group, AIO_PATH, NULL);

        return IRQ_HANDLED;
}

static int pqi_request_irqs(struct pqi_ctrl_info *ctrl_info)
{
        struct pci_dev *pci_dev = ctrl_info->pci_dev;
        int i;
        int rc;

        ctrl_info->event_irq = pci_irq_vector(pci_dev, 0);

        for (i = 0; i < ctrl_info->num_msix_vectors_enabled; i++) {
                rc = request_irq(pci_irq_vector(pci_dev, i), pqi_irq_handler, 0,
                        DRIVER_NAME_SHORT, &ctrl_info->queue_groups[i]);
                if (rc) {
                        dev_err(&pci_dev->dev,
                                "irq %u init failed with error %d\n",
                                pci_irq_vector(pci_dev, i), rc);
                        return rc;
                }
                ctrl_info->num_msix_vectors_initialized++;
        }

        return 0;
}

static void pqi_free_irqs(struct pqi_ctrl_info *ctrl_info)
{
        int i;

        for (i = 0; i < ctrl_info->num_msix_vectors_initialized; i++)
                free_irq(pci_irq_vector(ctrl_info->pci_dev, i),
                        &ctrl_info->queue_groups[i]);

        ctrl_info->num_msix_vectors_initialized = 0;
}

static int pqi_enable_msix_interrupts(struct pqi_ctrl_info *ctrl_info)
{
        int num_vectors_enabled;

        num_vectors_enabled = pci_alloc_irq_vectors(ctrl_info->pci_dev,
                        PQI_MIN_MSIX_VECTORS, ctrl_info->num_queue_groups,
                        PCI_IRQ_MSIX | PCI_IRQ_AFFINITY);
        if (num_vectors_enabled < 0) {
                dev_err(&ctrl_info->pci_dev->dev,
                        "MSI-X init failed with error %d\n",
                        num_vectors_enabled);
                return num_vectors_enabled;
        }

        ctrl_info->num_msix_vectors_enabled = num_vectors_enabled;
        ctrl_info->irq_mode = IRQ_MODE_MSIX;
        return 0;
}

static void pqi_disable_msix_interrupts(struct pqi_ctrl_info *ctrl_info)
{
        if (ctrl_info->num_msix_vectors_enabled) {
                pci_free_irq_vectors(ctrl_info->pci_dev);
                ctrl_info->num_msix_vectors_enabled = 0;
        }
}

static int pqi_alloc_operational_queues(struct pqi_ctrl_info *ctrl_info)
{
        unsigned int i;
        size_t alloc_length;
        size_t element_array_length_per_iq;
        size_t element_array_length_per_oq;
        void *element_array;
        void *next_queue_index;
        void *aligned_pointer;
        unsigned int num_inbound_queues;
        unsigned int num_outbound_queues;
        unsigned int num_queue_indexes;
        struct pqi_queue_group *queue_group;

        element_array_length_per_iq =
                PQI_OPERATIONAL_IQ_ELEMENT_LENGTH *
                ctrl_info->num_elements_per_iq;
        element_array_length_per_oq =
                PQI_OPERATIONAL_OQ_ELEMENT_LENGTH *
                ctrl_info->num_elements_per_oq;
        num_inbound_queues = ctrl_info->num_queue_groups * 2;
        num_outbound_queues = ctrl_info->num_queue_groups;
        num_queue_indexes = (ctrl_info->num_queue_groups * 3) + 1;

        aligned_pointer = NULL;

        for (i = 0; i < num_inbound_queues; i++) {
                aligned_pointer = PTR_ALIGN(aligned_pointer,
                        PQI_QUEUE_ELEMENT_ARRAY_ALIGNMENT);
                aligned_pointer += element_array_length_per_iq;
        }

        for (i = 0; i < num_outbound_queues; i++) {
                aligned_pointer = PTR_ALIGN(aligned_pointer,
                        PQI_QUEUE_ELEMENT_ARRAY_ALIGNMENT);
                aligned_pointer += element_array_length_per_oq;
        }

        aligned_pointer = PTR_ALIGN(aligned_pointer,
                PQI_QUEUE_ELEMENT_ARRAY_ALIGNMENT);
        aligned_pointer += PQI_NUM_EVENT_QUEUE_ELEMENTS *
                PQI_EVENT_OQ_ELEMENT_LENGTH;

        for (i = 0; i < num_queue_indexes; i++) {
                aligned_pointer = PTR_ALIGN(aligned_pointer,
                        PQI_OPERATIONAL_INDEX_ALIGNMENT);
                aligned_pointer += sizeof(pqi_index_t);
        }

        alloc_length = (size_t)aligned_pointer +
                PQI_QUEUE_ELEMENT_ARRAY_ALIGNMENT;

        alloc_length += PQI_EXTRA_SGL_MEMORY;

        ctrl_info->queue_memory_base =
                dma_zalloc_coherent(&ctrl_info->pci_dev->dev,
                        alloc_length,
                        &ctrl_info->queue_memory_base_dma_handle, GFP_KERNEL);

        if (!ctrl_info->queue_memory_base)
                return -ENOMEM;

        ctrl_info->queue_memory_length = alloc_length;

        element_array = PTR_ALIGN(ctrl_info->queue_memory_base,
                PQI_QUEUE_ELEMENT_ARRAY_ALIGNMENT);

        for (i = 0; i < ctrl_info->num_queue_groups; i++) {
                queue_group = &ctrl_info->queue_groups[i];
                queue_group->iq_element_array[RAID_PATH] = element_array;
                queue_group->iq_element_array_bus_addr[RAID_PATH] =
                        ctrl_info->queue_memory_base_dma_handle +
                                (element_array - ctrl_info->queue_memory_base);
                element_array += element_array_length_per_iq;
                element_array = PTR_ALIGN(element_array,
                        PQI_QUEUE_ELEMENT_ARRAY_ALIGNMENT);
                queue_group->iq_element_array[AIO_PATH] = element_array;
                queue_group->iq_element_array_bus_addr[AIO_PATH] =
                        ctrl_info->queue_memory_base_dma_handle +
                        (element_array - ctrl_info->queue_memory_base);
                element_array += element_array_length_per_iq;
                element_array = PTR_ALIGN(element_array,
                        PQI_QUEUE_ELEMENT_ARRAY_ALIGNMENT);
        }

        for (i = 0; i < ctrl_info->num_queue_groups; i++) {
                queue_group = &ctrl_info->queue_groups[i];
                queue_group->oq_element_array = element_array;
                queue_group->oq_element_array_bus_addr =
                        ctrl_info->queue_memory_base_dma_handle +
                        (element_array - ctrl_info->queue_memory_base);
                element_array += element_array_length_per_oq;
                element_array = PTR_ALIGN(element_array,
                        PQI_QUEUE_ELEMENT_ARRAY_ALIGNMENT);
        }

        ctrl_info->event_queue.oq_element_array = element_array;
        ctrl_info->event_queue.oq_element_array_bus_addr =
                ctrl_info->queue_memory_base_dma_handle +
                (element_array - ctrl_info->queue_memory_base);
        element_array += PQI_NUM_EVENT_QUEUE_ELEMENTS *
                PQI_EVENT_OQ_ELEMENT_LENGTH;

        next_queue_index = PTR_ALIGN(element_array,
                PQI_OPERATIONAL_INDEX_ALIGNMENT);

        for (i = 0; i < ctrl_info->num_queue_groups; i++) {
                queue_group = &ctrl_info->queue_groups[i];
                queue_group->iq_ci[RAID_PATH] = next_queue_index;
                queue_group->iq_ci_bus_addr[RAID_PATH] =
                        ctrl_info->queue_memory_base_dma_handle +
                        (next_queue_index - ctrl_info->queue_memory_base);
                next_queue_index += sizeof(pqi_index_t);
                next_queue_index = PTR_ALIGN(next_queue_index,
                        PQI_OPERATIONAL_INDEX_ALIGNMENT);
                queue_group->iq_ci[AIO_PATH] = next_queue_index;
                queue_group->iq_ci_bus_addr[AIO_PATH] =
                        ctrl_info->queue_memory_base_dma_handle +
                        (next_queue_index - ctrl_info->queue_memory_base);
                next_queue_index += sizeof(pqi_index_t);
                next_queue_index = PTR_ALIGN(next_queue_index,
                        PQI_OPERATIONAL_INDEX_ALIGNMENT);
                queue_group->oq_pi = next_queue_index;
                queue_group->oq_pi_bus_addr =
                        ctrl_info->queue_memory_base_dma_handle +
                        (next_queue_index - ctrl_info->queue_memory_base);
                next_queue_index += sizeof(pqi_index_t);
                next_queue_index = PTR_ALIGN(next_queue_index,
                        PQI_OPERATIONAL_INDEX_ALIGNMENT);
        }

        ctrl_info->event_queue.oq_pi = next_queue_index;
        ctrl_info->event_queue.oq_pi_bus_addr =
                ctrl_info->queue_memory_base_dma_handle +
                (next_queue_index - ctrl_info->queue_memory_base);

        return 0;
}

static void pqi_init_operational_queues(struct pqi_ctrl_info *ctrl_info)
{
        unsigned int i;
        u16 next_iq_id = PQI_MIN_OPERATIONAL_QUEUE_ID;
        u16 next_oq_id = PQI_MIN_OPERATIONAL_QUEUE_ID;

        /*
         * Initialize the backpointers to the controller structure in
         * each operational queue group structure.
         */
        for (i = 0; i < ctrl_info->num_queue_groups; i++)
                ctrl_info->queue_groups[i].ctrl_info = ctrl_info;

        /*
         * Assign IDs to all operational queues.  Note that the IDs
         * assigned to operational IQs are independent of the IDs
         * assigned to operational OQs.
         */
        ctrl_info->event_queue.oq_id = next_oq_id++;
        for (i = 0; i < ctrl_info->num_queue_groups; i++) {
                ctrl_info->queue_groups[i].iq_id[RAID_PATH] = next_iq_id++;
                ctrl_info->queue_groups[i].iq_id[AIO_PATH] = next_iq_id++;
                ctrl_info->queue_groups[i].oq_id = next_oq_id++;
        }

        /*
         * Assign MSI-X table entry indexes to all queues.  Note that the
         * interrupt for the event queue is shared with the first queue group.
         */
        ctrl_info->event_queue.int_msg_num = 0;
        for (i = 0; i < ctrl_info->num_queue_groups; i++)
                ctrl_info->queue_groups[i].int_msg_num = i;

        for (i = 0; i < ctrl_info->num_queue_groups; i++) {
                spin_lock_init(&ctrl_info->queue_groups[i].submit_lock[0]);
                spin_lock_init(&ctrl_info->queue_groups[i].submit_lock[1]);
                INIT_LIST_HEAD(&ctrl_info->queue_groups[i].request_list[0]);
                INIT_LIST_HEAD(&ctrl_info->queue_groups[i].request_list[1]);
        }
}

static int pqi_alloc_admin_queues(struct pqi_ctrl_info *ctrl_info)
{
        size_t alloc_length;
        struct pqi_admin_queues_aligned *admin_queues_aligned;
        struct pqi_admin_queues *admin_queues;

        alloc_length = sizeof(struct pqi_admin_queues_aligned) +
                PQI_QUEUE_ELEMENT_ARRAY_ALIGNMENT;

        ctrl_info->admin_queue_memory_base =
                dma_zalloc_coherent(&ctrl_info->pci_dev->dev,
                        alloc_length,
                        &ctrl_info->admin_queue_memory_base_dma_handle,
                        GFP_KERNEL);

        if (!ctrl_info->admin_queue_memory_base)
                return -ENOMEM;

        ctrl_info->admin_queue_memory_length = alloc_length;

        admin_queues = &ctrl_info->admin_queues;
        admin_queues_aligned = PTR_ALIGN(ctrl_info->admin_queue_memory_base,
                PQI_QUEUE_ELEMENT_ARRAY_ALIGNMENT);
        admin_queues->iq_element_array =
                &admin_queues_aligned->iq_element_array;
        admin_queues->oq_element_array =
                &admin_queues_aligned->oq_element_array;
        admin_queues->iq_ci = &admin_queues_aligned->iq_ci;
        admin_queues->oq_pi = &admin_queues_aligned->oq_pi;

        admin_queues->iq_element_array_bus_addr =
                ctrl_info->admin_queue_memory_base_dma_handle +
                (admin_queues->iq_element_array -
                ctrl_info->admin_queue_memory_base);
        admin_queues->oq_element_array_bus_addr =
                ctrl_info->admin_queue_memory_base_dma_handle +
                (admin_queues->oq_element_array -
                ctrl_info->admin_queue_memory_base);
        admin_queues->iq_ci_bus_addr =
                ctrl_info->admin_queue_memory_base_dma_handle +
                ((void *)admin_queues->iq_ci -
                ctrl_info->admin_queue_memory_base);
        admin_queues->oq_pi_bus_addr =
                ctrl_info->admin_queue_memory_base_dma_handle +
                ((void *)admin_queues->oq_pi -
                ctrl_info->admin_queue_memory_base);

        return 0;
}

#define PQI_ADMIN_QUEUE_CREATE_TIMEOUT_JIFFIES          HZ
#define PQI_ADMIN_QUEUE_CREATE_POLL_INTERVAL_MSECS      1

static int pqi_create_admin_queues(struct pqi_ctrl_info *ctrl_info)
{
        struct pqi_device_registers __iomem *pqi_registers;
        struct pqi_admin_queues *admin_queues;
        unsigned long timeout;
        u8 status;
        u32 reg;

        pqi_registers = ctrl_info->pqi_registers;
        admin_queues = &ctrl_info->admin_queues;

        writeq((u64)admin_queues->iq_element_array_bus_addr,
                &pqi_registers->admin_iq_element_array_addr);
        writeq((u64)admin_queues->oq_element_array_bus_addr,
                &pqi_registers->admin_oq_element_array_addr);
        writeq((u64)admin_queues->iq_ci_bus_addr,
                &pqi_registers->admin_iq_ci_addr);
        writeq((u64)admin_queues->oq_pi_bus_addr,
                &pqi_registers->admin_oq_pi_addr);

        reg = PQI_ADMIN_IQ_NUM_ELEMENTS |
                (PQI_ADMIN_OQ_NUM_ELEMENTS) << 8 |
                (admin_queues->int_msg_num << 16);
        writel(reg, &pqi_registers->admin_iq_num_elements);
        writel(PQI_CREATE_ADMIN_QUEUE_PAIR,
                &pqi_registers->function_and_status_code);

        timeout = PQI_ADMIN_QUEUE_CREATE_TIMEOUT_JIFFIES + jiffies;
        while (1) {
                status = readb(&pqi_registers->function_and_status_code);
                if (status == PQI_STATUS_IDLE)
                        break;
                if (time_after(jiffies, timeout))
                        return -ETIMEDOUT;
                msleep(PQI_ADMIN_QUEUE_CREATE_POLL_INTERVAL_MSECS);
        }

        /*
         * The offset registers are not initialized to the correct
         * offsets until *after* the create admin queue pair command
         * completes successfully.
         */
        admin_queues->iq_pi = ctrl_info->iomem_base +
                PQI_DEVICE_REGISTERS_OFFSET +
                readq(&pqi_registers->admin_iq_pi_offset);
        admin_queues->oq_ci = ctrl_info->iomem_base +
                PQI_DEVICE_REGISTERS_OFFSET +
                readq(&pqi_registers->admin_oq_ci_offset);

        return 0;
}

static void pqi_submit_admin_request(struct pqi_ctrl_info *ctrl_info,
        struct pqi_general_admin_request *request)
{
        struct pqi_admin_queues *admin_queues;
        void *next_element;
        pqi_index_t iq_pi;

        admin_queues = &ctrl_info->admin_queues;
        iq_pi = admin_queues->iq_pi_copy;

        next_element = admin_queues->iq_element_array +
                (iq_pi * PQI_ADMIN_IQ_ELEMENT_LENGTH);

        memcpy(next_element, request, sizeof(*request));

        iq_pi = (iq_pi + 1) % PQI_ADMIN_IQ_NUM_ELEMENTS;
        admin_queues->iq_pi_copy = iq_pi;

        /*
         * This write notifies the controller that an IU is available to be
         * processed.
         */
        writel(iq_pi, admin_queues->iq_pi);
}

#define PQI_ADMIN_REQUEST_TIMEOUT_SECS  60

static int pqi_poll_for_admin_response(struct pqi_ctrl_info *ctrl_info,
        struct pqi_general_admin_response *response)
{
        struct pqi_admin_queues *admin_queues;
        pqi_index_t oq_pi;
        pqi_index_t oq_ci;
        unsigned long timeout;

        admin_queues = &ctrl_info->admin_queues;
        oq_ci = admin_queues->oq_ci_copy;

        timeout = (PQI_ADMIN_REQUEST_TIMEOUT_SECS * HZ) + jiffies;

        while (1) {
                oq_pi = *admin_queues->oq_pi;
                if (oq_pi != oq_ci)
                        break;
                if (time_after(jiffies, timeout)) {
                        dev_err(&ctrl_info->pci_dev->dev,
                                "timed out waiting for admin response\n");
                        return -ETIMEDOUT;
                }
                if (!sis_is_firmware_running(ctrl_info))
                        return -ENXIO;
                usleep_range(1000, 2000);
        }

        memcpy(response, admin_queues->oq_element_array +
                (oq_ci * PQI_ADMIN_OQ_ELEMENT_LENGTH), sizeof(*response));

        oq_ci = (oq_ci + 1) % PQI_ADMIN_OQ_NUM_ELEMENTS;
        admin_queues->oq_ci_copy = oq_ci;
        writel(oq_ci, admin_queues->oq_ci);

        return 0;
}

static void pqi_start_io(struct pqi_ctrl_info *ctrl_info,
        struct pqi_queue_group *queue_group, enum pqi_io_path path,
        struct pqi_io_request *io_request)
{
        struct pqi_io_request *next;
        void *next_element;
        pqi_index_t iq_pi;
        pqi_index_t iq_ci;
        size_t iu_length;
        unsigned long flags;
        unsigned int num_elements_needed;
        unsigned int num_elements_to_end_of_queue;
        size_t copy_count;
        struct pqi_iu_header *request;

        spin_lock_irqsave(&queue_group->submit_lock[path], flags);

        if (io_request) {
                io_request->queue_group = queue_group;
                list_add_tail(&io_request->request_list_entry,
                        &queue_group->request_list[path]);
        }

        iq_pi = queue_group->iq_pi_copy[path];

        list_for_each_entry_safe(io_request, next,
                &queue_group->request_list[path], request_list_entry) {

                request = io_request->iu;

                iu_length = get_unaligned_le16(&request->iu_length) +
                        PQI_REQUEST_HEADER_LENGTH;
                num_elements_needed =
                        DIV_ROUND_UP(iu_length,
                                PQI_OPERATIONAL_IQ_ELEMENT_LENGTH);

                iq_ci = *queue_group->iq_ci[path];

                if (num_elements_needed > pqi_num_elements_free(iq_pi, iq_ci,
                        ctrl_info->num_elements_per_iq))
                        break;

                put_unaligned_le16(queue_group->oq_id,
                        &request->response_queue_id);

                next_element = queue_group->iq_element_array[path] +
                        (iq_pi * PQI_OPERATIONAL_IQ_ELEMENT_LENGTH);

                num_elements_to_end_of_queue =
                        ctrl_info->num_elements_per_iq - iq_pi;

                if (num_elements_needed <= num_elements_to_end_of_queue) {
                        memcpy(next_element, request, iu_length);
                } else {
                        copy_count = num_elements_to_end_of_queue *
                                PQI_OPERATIONAL_IQ_ELEMENT_LENGTH;
                        memcpy(next_element, request, copy_count);
                        memcpy(queue_group->iq_element_array[path],
                                (u8 *)request + copy_count,
                                iu_length - copy_count);
                }

                iq_pi = (iq_pi + num_elements_needed) %
                        ctrl_info->num_elements_per_iq;

                list_del(&io_request->request_list_entry);
        }

        if (iq_pi != queue_group->iq_pi_copy[path]) {
                queue_group->iq_pi_copy[path] = iq_pi;
                /*
                 * This write notifies the controller that one or more IUs are
                 * available to be processed.
                 */
                writel(iq_pi, queue_group->iq_pi[path]);
        }

        spin_unlock_irqrestore(&queue_group->submit_lock[path], flags);
}

#define PQI_WAIT_FOR_COMPLETION_IO_TIMEOUT_SECS         10

static int pqi_wait_for_completion_io(struct pqi_ctrl_info *ctrl_info,
        struct completion *wait)
{
        int rc;

        while (1) {
                if (wait_for_completion_io_timeout(wait,
                        PQI_WAIT_FOR_COMPLETION_IO_TIMEOUT_SECS * HZ)) {
                        rc = 0;
                        break;
                }

                pqi_check_ctrl_health(ctrl_info);
                if (pqi_ctrl_offline(ctrl_info)) {
                        rc = -ENXIO;
                        break;
                }
        }

        return rc;
}

static void pqi_raid_synchronous_complete(struct pqi_io_request *io_request,
        void *context)
{
        struct completion *waiting = context;

        complete(waiting);
}

static int pqi_submit_raid_request_synchronous_with_io_request(
        struct pqi_ctrl_info *ctrl_info, struct pqi_io_request *io_request,
        unsigned long timeout_msecs)
{
        int rc = 0;
        DECLARE_COMPLETION_ONSTACK(wait);

        io_request->io_complete_callback = pqi_raid_synchronous_complete;
        io_request->context = &wait;

        pqi_start_io(ctrl_info,
                &ctrl_info->queue_groups[PQI_DEFAULT_QUEUE_GROUP], RAID_PATH,
                io_request);

        if (timeout_msecs == NO_TIMEOUT) {
                pqi_wait_for_completion_io(ctrl_info, &wait);
        } else {
                if (!wait_for_completion_io_timeout(&wait,
                        msecs_to_jiffies(timeout_msecs))) {
                        dev_warn(&ctrl_info->pci_dev->dev,
                                "command timed out\n");
                        rc = -ETIMEDOUT;
                }
        }

        return rc;
}

static int pqi_submit_raid_request_synchronous(struct pqi_ctrl_info *ctrl_info,
        struct pqi_iu_header *request, unsigned int flags,
        struct pqi_raid_error_info *error_info, unsigned long timeout_msecs)
{
        int rc;
        struct pqi_io_request *io_request;
        unsigned long start_jiffies;
        unsigned long msecs_blocked;
        size_t iu_length;

        /*
         * Note that specifying PQI_SYNC_FLAGS_INTERRUPTABLE and a timeout value
         * are mutually exclusive.
         */

        if (flags & PQI_SYNC_FLAGS_INTERRUPTABLE) {
                if (down_interruptible(&ctrl_info->sync_request_sem))
                        return -ERESTARTSYS;
        } else {
                if (timeout_msecs == NO_TIMEOUT) {
                        down(&ctrl_info->sync_request_sem);
                } else {
                        start_jiffies = jiffies;
                        if (down_timeout(&ctrl_info->sync_request_sem,
                                msecs_to_jiffies(timeout_msecs)))
                                return -ETIMEDOUT;
                        msecs_blocked =
                                jiffies_to_msecs(jiffies - start_jiffies);
                        if (msecs_blocked >= timeout_msecs)
                                return -ETIMEDOUT;
                        timeout_msecs -= msecs_blocked;
                }
        }

        pqi_ctrl_busy(ctrl_info);
        timeout_msecs = pqi_wait_if_ctrl_blocked(ctrl_info, timeout_msecs);
        if (timeout_msecs == 0) {
                rc = -ETIMEDOUT;
                goto out;
        }

        if (pqi_ctrl_offline(ctrl_info)) {
                rc = -ENXIO;
                goto out;
        }

        io_request = pqi_alloc_io_request(ctrl_info);

        put_unaligned_le16(io_request->index,
                &(((struct pqi_raid_path_request *)request)->request_id));

        if (request->iu_type == PQI_REQUEST_IU_RAID_PATH_IO)
                ((struct pqi_raid_path_request *)request)->error_index =
                        ((struct pqi_raid_path_request *)request)->request_id;

        iu_length = get_unaligned_le16(&request->iu_length) +
                PQI_REQUEST_HEADER_LENGTH;
        memcpy(io_request->iu, request, iu_length);

        rc = pqi_submit_raid_request_synchronous_with_io_request(ctrl_info,
                io_request, timeout_msecs);

        if (error_info) {
                if (io_request->error_info)
                        memcpy(error_info, io_request->error_info,
                                sizeof(*error_info));
                else
                        memset(error_info, 0, sizeof(*error_info));
        } else if (rc == 0 && io_request->error_info) {
                u8 scsi_status;
                struct pqi_raid_error_info *raid_error_info;

                raid_error_info = io_request->error_info;
                scsi_status = raid_error_info->status;

                if (scsi_status == SAM_STAT_CHECK_CONDITION &&
                        raid_error_info->data_out_result ==
                        PQI_DATA_IN_OUT_UNDERFLOW)
                        scsi_status = SAM_STAT_GOOD;

                if (scsi_status != SAM_STAT_GOOD)
                        rc = -EIO;
        }

        pqi_free_io_request(io_request);

out:
        pqi_ctrl_unbusy(ctrl_info);
        up(&ctrl_info->sync_request_sem);

        return rc;
}

static int pqi_validate_admin_response(
        struct pqi_general_admin_response *response, u8 expected_function_code)
{
        if (response->header.iu_type != PQI_RESPONSE_IU_GENERAL_ADMIN)
                return -EINVAL;

        if (get_unaligned_le16(&response->header.iu_length) !=
                PQI_GENERAL_ADMIN_IU_LENGTH)
                return -EINVAL;

        if (response->function_code != expected_function_code)
                return -EINVAL;

        if (response->status != PQI_GENERAL_ADMIN_STATUS_SUCCESS)
                return -EINVAL;

        return 0;
}

static int pqi_submit_admin_request_synchronous(
        struct pqi_ctrl_info *ctrl_info,
        struct pqi_general_admin_request *request,
        struct pqi_general_admin_response *response)
{
        int rc;

        pqi_submit_admin_request(ctrl_info, request);

        rc = pqi_poll_for_admin_response(ctrl_info, response);

        if (rc == 0)
                rc = pqi_validate_admin_response(response,
                        request->function_code);

        return rc;
}

static int pqi_report_device_capability(struct pqi_ctrl_info *ctrl_info)
{
        int rc;
        struct pqi_general_admin_request request;
        struct pqi_general_admin_response response;
        struct pqi_device_capability *capability;
        struct pqi_iu_layer_descriptor *sop_iu_layer_descriptor;

        capability = kmalloc(sizeof(*capability), GFP_KERNEL);
        if (!capability)
                return -ENOMEM;

        memset(&request, 0, sizeof(request));

        request.header.iu_type = PQI_REQUEST_IU_GENERAL_ADMIN;
        put_unaligned_le16(PQI_GENERAL_ADMIN_IU_LENGTH,
                &request.header.iu_length);
        request.function_code =
                PQI_GENERAL_ADMIN_FUNCTION_REPORT_DEVICE_CAPABILITY;
        put_unaligned_le32(sizeof(*capability),
                &request.data.report_device_capability.buffer_length);

        rc = pqi_map_single(ctrl_info->pci_dev,
                &request.data.report_device_capability.sg_descriptor,
                capability, sizeof(*capability),
                PCI_DMA_FROMDEVICE);
        if (rc)
                goto out;

        rc = pqi_submit_admin_request_synchronous(ctrl_info, &request,
                &response);

        pqi_pci_unmap(ctrl_info->pci_dev,
                &request.data.report_device_capability.sg_descriptor, 1,
                PCI_DMA_FROMDEVICE);

        if (rc)
                goto out;

        if (response.status != PQI_GENERAL_ADMIN_STATUS_SUCCESS) {
                rc = -EIO;
                goto out;
        }

        ctrl_info->max_inbound_queues =
                get_unaligned_le16(&capability->max_inbound_queues);
        ctrl_info->max_elements_per_iq =
                get_unaligned_le16(&capability->max_elements_per_iq);
        ctrl_info->max_iq_element_length =
                get_unaligned_le16(&capability->max_iq_element_length)
                * 16;
        ctrl_info->max_outbound_queues =
                get_unaligned_le16(&capability->max_outbound_queues);
        ctrl_info->max_elements_per_oq =
                get_unaligned_le16(&capability->max_elements_per_oq);
        ctrl_info->max_oq_element_length =
                get_unaligned_le16(&capability->max_oq_element_length)
                * 16;

        sop_iu_layer_descriptor =
                &capability->iu_layer_descriptors[PQI_PROTOCOL_SOP];

        ctrl_info->max_inbound_iu_length_per_firmware =
                get_unaligned_le16(
                        &sop_iu_layer_descriptor->max_inbound_iu_length);
        ctrl_info->inbound_spanning_supported =
                sop_iu_layer_descriptor->inbound_spanning_supported;
        ctrl_info->outbound_spanning_supported =
                sop_iu_layer_descriptor->outbound_spanning_supported;

out:
        kfree(capability);

        return rc;
}

static int pqi_validate_device_capability(struct pqi_ctrl_info *ctrl_info)
{
        if (ctrl_info->max_iq_element_length <
                PQI_OPERATIONAL_IQ_ELEMENT_LENGTH) {
                dev_err(&ctrl_info->pci_dev->dev,
                        "max. inbound queue element length of %d is less than the required length of %d\n",
                        ctrl_info->max_iq_element_length,
                        PQI_OPERATIONAL_IQ_ELEMENT_LENGTH);
                return -EINVAL;
        }

        if (ctrl_info->max_oq_element_length <
                PQI_OPERATIONAL_OQ_ELEMENT_LENGTH) {
                dev_err(&ctrl_info->pci_dev->dev,
                        "max. outbound queue element length of %d is less than the required length of %d\n",
                        ctrl_info->max_oq_element_length,
                        PQI_OPERATIONAL_OQ_ELEMENT_LENGTH);
                return -EINVAL;
        }

        if (ctrl_info->max_inbound_iu_length_per_firmware <
                PQI_OPERATIONAL_IQ_ELEMENT_LENGTH) {
                dev_err(&ctrl_info->pci_dev->dev,
                        "max. inbound IU length of %u is less than the min. required length of %d\n",
                        ctrl_info->max_inbound_iu_length_per_firmware,
                        PQI_OPERATIONAL_IQ_ELEMENT_LENGTH);
                return -EINVAL;
        }

        if (!ctrl_info->inbound_spanning_supported) {
                dev_err(&ctrl_info->pci_dev->dev,
                        "the controller does not support inbound spanning\n");
                return -EINVAL;
        }

        if (ctrl_info->outbound_spanning_supported) {
                dev_err(&ctrl_info->pci_dev->dev,
                        "the controller supports outbound spanning but this driver does not\n");
                return -EINVAL;
        }

        return 0;
}

static int pqi_delete_operational_queue(struct pqi_ctrl_info *ctrl_info,
        bool inbound_queue, u16 queue_id)
{
        struct pqi_general_admin_request request;
        struct pqi_general_admin_response response;

        memset(&request, 0, sizeof(request));
        request.header.iu_type = PQI_REQUEST_IU_GENERAL_ADMIN;
        put_unaligned_le16(PQI_GENERAL_ADMIN_IU_LENGTH,
                &request.header.iu_length);
        if (inbound_queue)
                request.function_code =
                        PQI_GENERAL_ADMIN_FUNCTION_DELETE_IQ;
        else
                request.function_code =
                        PQI_GENERAL_ADMIN_FUNCTION_DELETE_OQ;
        put_unaligned_le16(queue_id,
                &request.data.delete_operational_queue.queue_id);

        return pqi_submit_admin_request_synchronous(ctrl_info, &request,
                &response);
}

static int pqi_create_event_queue(struct pqi_ctrl_info *ctrl_info)
{
        int rc;
        struct pqi_event_queue *event_queue;
        struct pqi_general_admin_request request;
        struct pqi_general_admin_response response;

        event_queue = &ctrl_info->event_queue;

        /*
         * Create OQ (Outbound Queue - device to host queue) to dedicate
         * to events.
         */
        memset(&request, 0, sizeof(request));
        request.header.iu_type = PQI_REQUEST_IU_GENERAL_ADMIN;
        put_unaligned_le16(PQI_GENERAL_ADMIN_IU_LENGTH,
                &request.header.iu_length);
        request.function_code = PQI_GENERAL_ADMIN_FUNCTION_CREATE_OQ;
        put_unaligned_le16(event_queue->oq_id,
                &request.data.create_operational_oq.queue_id);
        put_unaligned_le64((u64)event_queue->oq_element_array_bus_addr,
                &request.data.create_operational_oq.element_array_addr);
        put_unaligned_le64((u64)event_queue->oq_pi_bus_addr,
                &request.data.create_operational_oq.pi_addr);
        put_unaligned_le16(PQI_NUM_EVENT_QUEUE_ELEMENTS,
                &request.data.create_operational_oq.num_elements);
        put_unaligned_le16(PQI_EVENT_OQ_ELEMENT_LENGTH / 16,
                &request.data.create_operational_oq.element_length);
        request.data.create_operational_oq.queue_protocol = PQI_PROTOCOL_SOP;
        put_unaligned_le16(event_queue->int_msg_num,
                &request.data.create_operational_oq.int_msg_num);

        rc = pqi_submit_admin_request_synchronous(ctrl_info, &request,
                &response);
        if (rc)
                return rc;

        event_queue->oq_ci = ctrl_info->iomem_base +
                PQI_DEVICE_REGISTERS_OFFSET +
                get_unaligned_le64(
                        &response.data.create_operational_oq.oq_ci_offset);

        return 0;
}

static int pqi_create_queue_group(struct pqi_ctrl_info *ctrl_info,
        unsigned int group_number)
{
        int rc;
        struct pqi_queue_group *queue_group;
        struct pqi_general_admin_request request;
        struct pqi_general_admin_response response;

        queue_group = &ctrl_info->queue_groups[group_number];

        /*
         * Create IQ (Inbound Queue - host to device queue) for
         * RAID path.
         */
        memset(&request, 0, sizeof(request));
        request.header.iu_type = PQI_REQUEST_IU_GENERAL_ADMIN;
        put_unaligned_le16(PQI_GENERAL_ADMIN_IU_LENGTH,
                &request.header.iu_length);
        request.function_code = PQI_GENERAL_ADMIN_FUNCTION_CREATE_IQ;
        put_unaligned_le16(queue_group->iq_id[RAID_PATH],
                &request.data.create_operational_iq.queue_id);
        put_unaligned_le64(
                (u64)queue_group->iq_element_array_bus_addr[RAID_PATH],
                &request.data.create_operational_iq.element_array_addr);
        put_unaligned_le64((u64)queue_group->iq_ci_bus_addr[RAID_PATH],
                &request.data.create_operational_iq.ci_addr);
        put_unaligned_le16(ctrl_info->num_elements_per_iq,
                &request.data.create_operational_iq.num_elements);