drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/pci-aspm.h>
  28 #include <linux/kernel.h>
  29 #include <linux/slab.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/compat.h>
  43 #include <linux/mutex.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/io.h>
  46
  47 #include <linux/dma-mapping.h>
  48 #include <linux/blkdev.h>
  49 #include <linux/genhd.h>
  50 #include <linux/completion.h>
  51 #include <scsi/scsi.h>
  52 #include <scsi/sg.h>
  53 #include <scsi/scsi_ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/scatterlist.h>
  56 #include <linux/kthread.h>
  57
  58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  59 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  61
  62 /* Embedded module documentation macros - see modules.h */
  63 MODULE_AUTHOR("Hewlett-Packard Company");
  64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  65 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  66 MODULE_VERSION("3.6.26");
  67 MODULE_LICENSE("GPL");
  68 static int cciss_tape_cmds = 6;
  69 module_param(cciss_tape_cmds, int, 0644);
  70 MODULE_PARM_DESC(cciss_tape_cmds,
  71         "number of commands to allocate for tape devices (default: 6)");
  72 static int cciss_simple_mode;
  73 module_param(cciss_simple_mode, int, S_IRUGO|S_IWUSR);
  74 MODULE_PARM_DESC(cciss_simple_mode,
  75         "Use 'simple mode' rather than 'performant mode'");
  76
  77 static DEFINE_MUTEX(cciss_mutex);
  78 static struct proc_dir_entry *proc_cciss;
  79
  80 #include "cciss_cmd.h"
  81 #include "cciss.h"
  82 #include <linux/cciss_ioctl.h>
  83
  84 /* define the PCI info for the cards we can control */
  85 static const struct pci_device_id cciss_pci_device_id[] = {
  86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  90         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  91         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  92         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  93         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  94         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 106         {0,}
 107 };
 108
 109 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 110
 111 /*  board_id = Subsystem Device ID & Vendor ID
 112  *  product = Marketing Name for the board
 113  *  access = Address of the struct of function pointers
 114  */
 115 static struct board_type products[] = {
 116         {0x40700E11, "Smart Array 5300", &SA5_access},
 117         {0x40800E11, "Smart Array 5i", &SA5B_access},
 118         {0x40820E11, "Smart Array 532", &SA5B_access},
 119         {0x40830E11, "Smart Array 5312", &SA5B_access},
 120         {0x409A0E11, "Smart Array 641", &SA5_access},
 121         {0x409B0E11, "Smart Array 642", &SA5_access},
 122         {0x409C0E11, "Smart Array 6400", &SA5_access},
 123         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 124         {0x40910E11, "Smart Array 6i", &SA5_access},
 125         {0x3225103C, "Smart Array P600", &SA5_access},
 126         {0x3223103C, "Smart Array P800", &SA5_access},
 127         {0x3234103C, "Smart Array P400", &SA5_access},
 128         {0x3235103C, "Smart Array P400i", &SA5_access},
 129         {0x3211103C, "Smart Array E200i", &SA5_access},
 130         {0x3212103C, "Smart Array E200", &SA5_access},
 131         {0x3213103C, "Smart Array E200i", &SA5_access},
 132         {0x3214103C, "Smart Array E200i", &SA5_access},
 133         {0x3215103C, "Smart Array E200i", &SA5_access},
 134         {0x3237103C, "Smart Array E500", &SA5_access},
 135         {0x3223103C, "Smart Array P800", &SA5_access},
 136         {0x3234103C, "Smart Array P400", &SA5_access},
 137         {0x323D103C, "Smart Array P700m", &SA5_access},
 138 };
 139
 140 /* How long to wait (in milliseconds) for board to go into simple mode */
 141 #define MAX_CONFIG_WAIT 30000
 142 #define MAX_IOCTL_CONFIG_WAIT 1000
 143
 144 /*define how many times we will try a command because of bus resets */
 145 #define MAX_CMD_RETRIES 3
 146
 147 #define MAX_CTLR        32
 148
 149 /* Originally cciss driver only supports 8 major numbers */
 150 #define MAX_CTLR_ORIG   8
 151
 152 static ctlr_info_t *hba[MAX_CTLR];
 153
 154 static struct task_struct *cciss_scan_thread;
 155 static DEFINE_MUTEX(scan_mutex);
 156 static LIST_HEAD(scan_q);
 157
 158 static void do_cciss_request(struct request_queue *q);
 159 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 160 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 161 static int cciss_open(struct block_device *bdev, fmode_t mode);
 162 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 163 static int cciss_release(struct gendisk *disk, fmode_t mode);
 164 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 165                        unsigned int cmd, unsigned long arg);
 166 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 167
 168 static int cciss_revalidate(struct gendisk *disk);
 169 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 170 static int deregister_disk(ctlr_info_t *h, int drv_index,
 171                            int clear_all, int via_ioctl);
 172
 173 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 174                         sector_t *total_size, unsigned int *block_size);
 175 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 176                         sector_t *total_size, unsigned int *block_size);
 177 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 178                         sector_t total_size,
 179                         unsigned int block_size, InquiryData_struct *inq_buff,
 180                                    drive_info_struct *drv);
 181 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
 182 static int __devinit cciss_enter_simple_mode(struct ctlr_info *h);
 183 static void start_io(ctlr_info_t *h);
 184 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 185                         __u8 page_code, unsigned char scsi3addr[],
 186                         int cmd_type);
 187 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 188         int attempt_retry);
 189 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 190
 191 static int add_to_scan_list(struct ctlr_info *h);
 192 static int scan_thread(void *data);
 193 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 194 static void cciss_hba_release(struct device *dev);
 195 static void cciss_device_release(struct device *dev);
 196 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 197 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 198 static inline u32 next_command(ctlr_info_t *h);
 199 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
 200         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 201         u64 *cfg_offset);
 202 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 203         unsigned long *memory_bar);
 204 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag);
 205 static __devinit int write_driver_ver_to_cfgtable(
 206         CfgTable_struct __iomem *cfgtable);
 207
 208 /* performant mode helper functions */
 209 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 210                                 int *bucket_map);
 211 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 212
 213 #ifdef CONFIG_PROC_FS
 214 static void cciss_procinit(ctlr_info_t *h);
 215 #else
 216 static void cciss_procinit(ctlr_info_t *h)
 217 {
 218 }
 219 #endif                          /* CONFIG_PROC_FS */
 220
 221 #ifdef CONFIG_COMPAT
 222 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 223                               unsigned, unsigned long);
 224 #endif
 225
 226 static const struct block_device_operations cciss_fops = {
 227         .owner = THIS_MODULE,
 228         .open = cciss_unlocked_open,
 229         .release = cciss_release,
 230         .ioctl = cciss_ioctl,
 231         .getgeo = cciss_getgeo,
 232 #ifdef CONFIG_COMPAT
 233         .compat_ioctl = cciss_compat_ioctl,
 234 #endif
 235         .revalidate_disk = cciss_revalidate,
 236 };
 237
 238 /* set_performant_mode: Modify the tag for cciss performant
 239  * set bit 0 for pull model, bits 3-1 for block fetch
 240  * register number
 241  */
 242 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 243 {
 244         if (likely(h->transMethod & CFGTBL_Trans_Performant))
 245                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 246 }
 247
 248 /*
 249  * Enqueuing and dequeuing functions for cmdlists.
 250  */
 251 static inline void addQ(struct list_head *list, CommandList_struct *c)
 252 {
 253         list_add_tail(&c->list, list);
 254 }
 255
 256 static inline void removeQ(CommandList_struct *c)
 257 {
 258         /*
 259          * After kexec/dump some commands might still
 260          * be in flight, which the firmware will try
 261          * to complete. Resetting the firmware doesn't work
 262          * with old fw revisions, so we have to mark
 263          * them off as 'stale' to prevent the driver from
 264          * falling over.
 265          */
 266         if (WARN_ON(list_empty(&c->list))) {
 267                 c->cmd_type = CMD_MSG_STALE;
 268                 return;
 269         }
 270
 271         list_del_init(&c->list);
 272 }
 273
 274 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 275         CommandList_struct *c)
 276 {
 277         unsigned long flags;
 278         set_performant_mode(h, c);
 279         spin_lock_irqsave(&h->lock, flags);
 280         addQ(&h->reqQ, c);
 281         h->Qdepth++;
 282         if (h->Qdepth > h->maxQsinceinit)
 283                 h->maxQsinceinit = h->Qdepth;
 284         start_io(h);
 285         spin_unlock_irqrestore(&h->lock, flags);
 286 }
 287
 288 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 289         int nr_cmds)
 290 {
 291         int i;
 292
 293         if (!cmd_sg_list)
 294                 return;
 295         for (i = 0; i < nr_cmds; i++) {
 296                 kfree(cmd_sg_list[i]);
 297                 cmd_sg_list[i] = NULL;
 298         }
 299         kfree(cmd_sg_list);
 300 }
 301
 302 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 303         ctlr_info_t *h, int chainsize, int nr_cmds)
 304 {
 305         int j;
 306         SGDescriptor_struct **cmd_sg_list;
 307
 308         if (chainsize <= 0)
 309                 return NULL;
 310
 311         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 312         if (!cmd_sg_list)
 313                 return NULL;
 314
 315         /* Build up chain blocks for each command */
 316         for (j = 0; j < nr_cmds; j++) {
 317                 /* Need a block of chainsized s/g elements. */
 318                 cmd_sg_list[j] = kmalloc((chainsize *
 319                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 320                 if (!cmd_sg_list[j]) {
 321                         dev_err(&h->pdev->dev, "Cannot get memory "
 322                                 "for s/g chains.\n");
 323                         goto clean;
 324                 }
 325         }
 326         return cmd_sg_list;
 327 clean:
 328         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 329         return NULL;
 330 }
 331
 332 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 333 {
 334         SGDescriptor_struct *chain_sg;
 335         u64bit temp64;
 336
 337         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 338                 return;
 339
 340         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 341         temp64.val32.lower = chain_sg->Addr.lower;
 342         temp64.val32.upper = chain_sg->Addr.upper;
 343         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 344 }
 345
 346 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 347         SGDescriptor_struct *chain_block, int len)
 348 {
 349         SGDescriptor_struct *chain_sg;
 350         u64bit temp64;
 351
 352         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 353         chain_sg->Ext = CCISS_SG_CHAIN;
 354         chain_sg->Len = len;
 355         temp64.val = pci_map_single(h->pdev, chain_block, len,
 356                                 PCI_DMA_TODEVICE);
 357         chain_sg->Addr.lower = temp64.val32.lower;
 358         chain_sg->Addr.upper = temp64.val32.upper;
 359 }
 360
 361 #include "cciss_scsi.c"         /* For SCSI tape support */
 362
 363 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 364         "UNKNOWN"
 365 };
 366 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 367
 368 #ifdef CONFIG_PROC_FS
 369
 370 /*
 371  * Report information about this controller.
 372  */
 373 #define ENG_GIG 1000000000
 374 #define ENG_GIG_FACTOR (ENG_GIG/512)
 375 #define ENGAGE_SCSI     "engage scsi"
 376
 377 static void cciss_seq_show_header(struct seq_file *seq)
 378 {
 379         ctlr_info_t *h = seq->private;
 380
 381         seq_printf(seq, "%s: HP %s Controller\n"
 382                 "Board ID: 0x%08lx\n"
 383                 "Firmware Version: %c%c%c%c\n"
 384                 "IRQ: %d\n"
 385                 "Logical drives: %d\n"
 386                 "Current Q depth: %d\n"
 387                 "Current # commands on controller: %d\n"
 388                 "Max Q depth since init: %d\n"
 389                 "Max # commands on controller since init: %d\n"
 390                 "Max SG entries since init: %d\n",
 391                 h->devname,
 392                 h->product_name,
 393                 (unsigned long)h->board_id,
 394                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 395                 h->firm_ver[3], (unsigned int)h->intr[h->intr_mode],
 396                 h->num_luns,
 397                 h->Qdepth, h->commands_outstanding,
 398                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 399
 400 #ifdef CONFIG_CISS_SCSI_TAPE
 401         cciss_seq_tape_report(seq, h);
 402 #endif /* CONFIG_CISS_SCSI_TAPE */
 403 }
 404
 405 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 406 {
 407         ctlr_info_t *h = seq->private;
 408         unsigned long flags;
 409
 410         /* prevent displaying bogus info during configuration
 411          * or deconfiguration of a logical volume
 412          */
 413         spin_lock_irqsave(&h->lock, flags);
 414         if (h->busy_configuring) {
 415                 spin_unlock_irqrestore(&h->lock, flags);
 416                 return ERR_PTR(-EBUSY);
 417         }
 418         h->busy_configuring = 1;
 419         spin_unlock_irqrestore(&h->lock, flags);
 420
 421         if (*pos == 0)
 422                 cciss_seq_show_header(seq);
 423
 424         return pos;
 425 }
 426
 427 static int cciss_seq_show(struct seq_file *seq, void *v)
 428 {
 429         sector_t vol_sz, vol_sz_frac;
 430         ctlr_info_t *h = seq->private;
 431         unsigned ctlr = h->ctlr;
 432         loff_t *pos = v;
 433         drive_info_struct *drv = h->drv[*pos];
 434
 435         if (*pos > h->highest_lun)
 436                 return 0;
 437
 438         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 439                 return 0;
 440
 441         if (drv->heads == 0)
 442                 return 0;
 443
 444         vol_sz = drv->nr_blocks;
 445         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 446         vol_sz_frac *= 100;
 447         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 448
 449         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 450                 drv->raid_level = RAID_UNKNOWN;
 451         seq_printf(seq, "cciss/c%dd%d:"
 452                         "\t%4u.%02uGB\tRAID %s\n",
 453                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 454                         raid_label[drv->raid_level]);
 455         return 0;
 456 }
 457
 458 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 459 {
 460         ctlr_info_t *h = seq->private;
 461
 462         if (*pos > h->highest_lun)
 463                 return NULL;
 464         *pos += 1;
 465
 466         return pos;
 467 }
 468
 469 static void cciss_seq_stop(struct seq_file *seq, void *v)
 470 {
 471         ctlr_info_t *h = seq->private;
 472
 473         /* Only reset h->busy_configuring if we succeeded in setting
 474          * it during cciss_seq_start. */
 475         if (v == ERR_PTR(-EBUSY))
 476                 return;
 477
 478         h->busy_configuring = 0;
 479 }
 480
 481 static const struct seq_operations cciss_seq_ops = {
 482         .start = cciss_seq_start,
 483         .show  = cciss_seq_show,
 484         .next  = cciss_seq_next,
 485         .stop  = cciss_seq_stop,
 486 };
 487
 488 static int cciss_seq_open(struct inode *inode, struct file *file)
 489 {
 490         int ret = seq_open(file, &cciss_seq_ops);
 491         struct seq_file *seq = file->private_data;
 492
 493         if (!ret)
 494                 seq->private = PDE(inode)->data;
 495
 496         return ret;
 497 }
 498
 499 static ssize_t
 500 cciss_proc_write(struct file *file, const char __user *buf,
 501                  size_t length, loff_t *ppos)
 502 {
 503         int err;
 504         char *buffer;
 505
 506 #ifndef CONFIG_CISS_SCSI_TAPE
 507         return -EINVAL;
 508 #endif
 509
 510         if (!buf || length > PAGE_SIZE - 1)
 511                 return -EINVAL;
 512
 513         buffer = (char *)__get_free_page(GFP_KERNEL);
 514         if (!buffer)
 515                 return -ENOMEM;
 516
 517         err = -EFAULT;
 518         if (copy_from_user(buffer, buf, length))
 519                 goto out;
 520         buffer[length] = '\0';
 521
 522 #ifdef CONFIG_CISS_SCSI_TAPE
 523         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 524                 struct seq_file *seq = file->private_data;
 525                 ctlr_info_t *h = seq->private;
 526
 527                 err = cciss_engage_scsi(h);
 528                 if (err == 0)
 529                         err = length;
 530         } else
 531 #endif /* CONFIG_CISS_SCSI_TAPE */
 532                 err = -EINVAL;
 533         /* might be nice to have "disengage" too, but it's not
 534            safely possible. (only 1 module use count, lock issues.) */
 535
 536 out:
 537         free_page((unsigned long)buffer);
 538         return err;
 539 }
 540
 541 static const struct file_operations cciss_proc_fops = {
 542         .owner   = THIS_MODULE,
 543         .open    = cciss_seq_open,
 544         .read    = seq_read,
 545         .llseek  = seq_lseek,
 546         .release = seq_release,
 547         .write   = cciss_proc_write,
 548 };
 549
 550 static void __devinit cciss_procinit(ctlr_info_t *h)
 551 {
 552         struct proc_dir_entry *pde;
 553
 554         if (proc_cciss == NULL)
 555                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 556         if (!proc_cciss)
 557                 return;
 558         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 559                                         S_IROTH, proc_cciss,
 560                                         &cciss_proc_fops, h);
 561 }
 562 #endif                          /* CONFIG_PROC_FS */
 563
 564 #define MAX_PRODUCT_NAME_LEN 19
 565
 566 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 567 #define to_drv(n) container_of(n, drive_info_struct, dev)
 568
 569 /* List of controllers which cannot be hard reset on kexec with reset_devices */
 570 static u32 unresettable_controller[] = {
 571         0x324a103C, /* Smart Array P712m */
 572         0x324b103C, /* SmartArray P711m */
 573         0x3223103C, /* Smart Array P800 */
 574         0x3234103C, /* Smart Array P400 */
 575         0x3235103C, /* Smart Array P400i */
 576         0x3211103C, /* Smart Array E200i */
 577         0x3212103C, /* Smart Array E200 */
 578         0x3213103C, /* Smart Array E200i */
 579         0x3214103C, /* Smart Array E200i */
 580         0x3215103C, /* Smart Array E200i */
 581         0x3237103C, /* Smart Array E500 */
 582         0x323D103C, /* Smart Array P700m */
 583         0x409C0E11, /* Smart Array 6400 */
 584         0x409D0E11, /* Smart Array 6400 EM */
 585 };
 586
 587 /* List of controllers which cannot even be soft reset */
 588 static u32 soft_unresettable_controller[] = {
 589         0x409C0E11, /* Smart Array 6400 */
 590         0x409D0E11, /* Smart Array 6400 EM */
 591 };
 592
 593 static int ctlr_is_hard_resettable(u32 board_id)
 594 {
 595         int i;
 596
 597         for (i = 0; i < ARRAY_SIZE(unresettable_controller); i++)
 598                 if (unresettable_controller[i] == board_id)
 599                         return 0;
 600         return 1;
 601 }
 602
 603 static int ctlr_is_soft_resettable(u32 board_id)
 604 {
 605         int i;
 606
 607         for (i = 0; i < ARRAY_SIZE(soft_unresettable_controller); i++)
 608                 if (soft_unresettable_controller[i] == board_id)
 609                         return 0;
 610         return 1;
 611 }
 612
 613 static int ctlr_is_resettable(u32 board_id)
 614 {
 615         return ctlr_is_hard_resettable(board_id) ||
 616                 ctlr_is_soft_resettable(board_id);
 617 }
 618
 619 static ssize_t host_show_resettable(struct device *dev,
 620                                     struct device_attribute *attr,
 621                                     char *buf)
 622 {
 623         struct ctlr_info *h = to_hba(dev);
 624
 625         return snprintf(buf, 20, "%d\n", ctlr_is_resettable(h->board_id));
 626 }
 627 static DEVICE_ATTR(resettable, S_IRUGO, host_show_resettable, NULL);
 628
 629 static ssize_t host_store_rescan(struct device *dev,
 630                                  struct device_attribute *attr,
 631                                  const char *buf, size_t count)
 632 {
 633         struct ctlr_info *h = to_hba(dev);
 634
 635         add_to_scan_list(h);
 636         wake_up_process(cciss_scan_thread);
 637         wait_for_completion_interruptible(&h->scan_wait);
 638
 639         return count;
 640 }
 641 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 642
 643 static ssize_t host_show_transport_mode(struct device *dev,
 644                                  struct device_attribute *attr,
 645                                  char *buf)
 646 {
 647         struct ctlr_info *h = to_hba(dev);
 648
 649         return snprintf(buf, 20, "%s\n",
 650                 h->transMethod & CFGTBL_Trans_Performant ?
 651                         "performant" : "simple");
 652 }
 653 static DEVICE_ATTR(transport_mode, S_IRUGO, host_show_transport_mode, NULL);
 654
 655 static ssize_t dev_show_unique_id(struct device *dev,
 656                                  struct device_attribute *attr,
 657                                  char *buf)
 658 {
 659         drive_info_struct *drv = to_drv(dev);
 660         struct ctlr_info *h = to_hba(drv->dev.parent);
 661         __u8 sn[16];
 662         unsigned long flags;
 663         int ret = 0;
 664
 665         spin_lock_irqsave(&h->lock, flags);
 666         if (h->busy_configuring)
 667                 ret = -EBUSY;
 668         else
 669                 memcpy(sn, drv->serial_no, sizeof(sn));
 670         spin_unlock_irqrestore(&h->lock, flags);
 671
 672         if (ret)
 673                 return ret;
 674         else
 675                 return snprintf(buf, 16 * 2 + 2,
 676                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 677                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 678                                 sn[0], sn[1], sn[2], sn[3],
 679                                 sn[4], sn[5], sn[6], sn[7],
 680                                 sn[8], sn[9], sn[10], sn[11],
 681                                 sn[12], sn[13], sn[14], sn[15]);
 682 }
 683 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 684
 685 static ssize_t dev_show_vendor(struct device *dev,
 686                                struct device_attribute *attr,
 687                                char *buf)
 688 {
 689         drive_info_struct *drv = to_drv(dev);
 690         struct ctlr_info *h = to_hba(drv->dev.parent);
 691         char vendor[VENDOR_LEN + 1];
 692         unsigned long flags;
 693         int ret = 0;
 694
 695         spin_lock_irqsave(&h->lock, flags);
 696         if (h->busy_configuring)
 697                 ret = -EBUSY;
 698         else
 699                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 700         spin_unlock_irqrestore(&h->lock, flags);
 701
 702         if (ret)
 703                 return ret;
 704         else
 705                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 706 }
 707 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 708
 709 static ssize_t dev_show_model(struct device *dev,
 710                               struct device_attribute *attr,
 711                               char *buf)
 712 {
 713         drive_info_struct *drv = to_drv(dev);
 714         struct ctlr_info *h = to_hba(drv->dev.parent);
 715         char model[MODEL_LEN + 1];
 716         unsigned long flags;
 717         int ret = 0;
 718
 719         spin_lock_irqsave(&h->lock, flags);
 720         if (h->busy_configuring)
 721                 ret = -EBUSY;
 722         else
 723                 memcpy(model, drv->model, MODEL_LEN + 1);
 724         spin_unlock_irqrestore(&h->lock, flags);
 725
 726         if (ret)
 727                 return ret;
 728         else
 729                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 730 }
 731 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 732
 733 static ssize_t dev_show_rev(struct device *dev,
 734                             struct device_attribute *attr,
 735                             char *buf)
 736 {
 737         drive_info_struct *drv = to_drv(dev);
 738         struct ctlr_info *h = to_hba(drv->dev.parent);
 739         char rev[REV_LEN + 1];
 740         unsigned long flags;
 741         int ret = 0;
 742
 743         spin_lock_irqsave(&h->lock, flags);
 744         if (h->busy_configuring)
 745                 ret = -EBUSY;
 746         else
 747                 memcpy(rev, drv->rev, REV_LEN + 1);
 748         spin_unlock_irqrestore(&h->lock, flags);
 749
 750         if (ret)
 751                 return ret;
 752         else
 753                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 754 }
 755 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 756
 757 static ssize_t cciss_show_lunid(struct device *dev,
 758                                 struct device_attribute *attr, char *buf)
 759 {
 760         drive_info_struct *drv = to_drv(dev);
 761         struct ctlr_info *h = to_hba(drv->dev.parent);
 762         unsigned long flags;
 763         unsigned char lunid[8];
 764
 765         spin_lock_irqsave(&h->lock, flags);
 766         if (h->busy_configuring) {
 767                 spin_unlock_irqrestore(&h->lock, flags);
 768                 return -EBUSY;
 769         }
 770         if (!drv->heads) {
 771                 spin_unlock_irqrestore(&h->lock, flags);
 772                 return -ENOTTY;
 773         }
 774         memcpy(lunid, drv->LunID, sizeof(lunid));
 775         spin_unlock_irqrestore(&h->lock, flags);
 776         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 777                 lunid[0], lunid[1], lunid[2], lunid[3],
 778                 lunid[4], lunid[5], lunid[6], lunid[7]);
 779 }
 780 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 781
 782 static ssize_t cciss_show_raid_level(struct device *dev,
 783                                      struct device_attribute *attr, char *buf)
 784 {
 785         drive_info_struct *drv = to_drv(dev);
 786         struct ctlr_info *h = to_hba(drv->dev.parent);
 787         int raid;
 788         unsigned long flags;
 789
 790         spin_lock_irqsave(&h->lock, flags);
 791         if (h->busy_configuring) {
 792                 spin_unlock_irqrestore(&h->lock, flags);
 793                 return -EBUSY;
 794         }
 795         raid = drv->raid_level;
 796         spin_unlock_irqrestore(&h->lock, flags);
 797         if (raid < 0 || raid > RAID_UNKNOWN)
 798                 raid = RAID_UNKNOWN;
 799
 800         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 801                         raid_label[raid]);
 802 }
 803 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 804
 805 static ssize_t cciss_show_usage_count(struct device *dev,
 806                                       struct device_attribute *attr, char *buf)
 807 {
 808         drive_info_struct *drv = to_drv(dev);
 809         struct ctlr_info *h = to_hba(drv->dev.parent);
 810         unsigned long flags;
 811         int count;
 812
 813         spin_lock_irqsave(&h->lock, flags);
 814         if (h->busy_configuring) {
 815                 spin_unlock_irqrestore(&h->lock, flags);
 816                 return -EBUSY;
 817         }
 818         count = drv->usage_count;
 819         spin_unlock_irqrestore(&h->lock, flags);
 820         return snprintf(buf, 20, "%d\n", count);
 821 }
 822 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 823
 824 static struct attribute *cciss_host_attrs[] = {
 825         &dev_attr_rescan.attr,
 826         &dev_attr_resettable.attr,
 827         &dev_attr_transport_mode.attr,
 828         NULL
 829 };
 830
 831 static struct attribute_group cciss_host_attr_group = {
 832         .attrs = cciss_host_attrs,
 833 };
 834
 835 static const struct attribute_group *cciss_host_attr_groups[] = {
 836         &cciss_host_attr_group,
 837         NULL
 838 };
 839
 840 static struct device_type cciss_host_type = {
 841         .name           = "cciss_host",
 842         .groups         = cciss_host_attr_groups,
 843         .release        = cciss_hba_release,
 844 };
 845
 846 static struct attribute *cciss_dev_attrs[] = {
 847         &dev_attr_unique_id.attr,
 848         &dev_attr_model.attr,
 849         &dev_attr_vendor.attr,
 850         &dev_attr_rev.attr,
 851         &dev_attr_lunid.attr,
 852         &dev_attr_raid_level.attr,
 853         &dev_attr_usage_count.attr,
 854         NULL
 855 };
 856
 857 static struct attribute_group cciss_dev_attr_group = {
 858         .attrs = cciss_dev_attrs,
 859 };
 860
 861 static const struct attribute_group *cciss_dev_attr_groups[] = {
 862         &cciss_dev_attr_group,
 863         NULL
 864 };
 865
 866 static struct device_type cciss_dev_type = {
 867         .name           = "cciss_device",
 868         .groups         = cciss_dev_attr_groups,
 869         .release        = cciss_device_release,
 870 };
 871
 872 static struct bus_type cciss_bus_type = {
 873         .name           = "cciss",
 874 };
 875
 876 /*
 877  * cciss_hba_release is called when the reference count
 878  * of h->dev goes to zero.
 879  */
 880 static void cciss_hba_release(struct device *dev)
 881 {
 882         /*
 883          * nothing to do, but need this to avoid a warning
 884          * about not having a release handler from lib/kref.c.
 885          */
 886 }
 887
 888 /*
 889  * Initialize sysfs entry for each controller.  This sets up and registers
 890  * the 'cciss#' directory for each individual controller under
 891  * /sys/bus/pci/devices/<dev>/.
 892  */
 893 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 894 {
 895         device_initialize(&h->dev);
 896         h->dev.type = &cciss_host_type;
 897         h->dev.bus = &cciss_bus_type;
 898         dev_set_name(&h->dev, "%s", h->devname);
 899         h->dev.parent = &h->pdev->dev;
 900
 901         return device_add(&h->dev);
 902 }
 903
 904 /*
 905  * Remove sysfs entries for an hba.
 906  */
 907 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 908 {
 909         device_del(&h->dev);
 910         put_device(&h->dev); /* final put. */
 911 }
 912
 913 /* cciss_device_release is called when the reference count
 914  * of h->drv[x]dev goes to zero.
 915  */
 916 static void cciss_device_release(struct device *dev)
 917 {
 918         drive_info_struct *drv = to_drv(dev);
 919         kfree(drv);
 920 }
 921
 922 /*
 923  * Initialize sysfs for each logical drive.  This sets up and registers
 924  * the 'c#d#' directory for each individual logical drive under
 925  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 926  * /sys/block/cciss!c#d# to this entry.
 927  */
 928 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 929                                        int drv_index)
 930 {
 931         struct device *dev;
 932
 933         if (h->drv[drv_index]->device_initialized)
 934                 return 0;
 935
 936         dev = &h->drv[drv_index]->dev;
 937         device_initialize(dev);
 938         dev->type = &cciss_dev_type;
 939         dev->bus = &cciss_bus_type;
 940         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 941         dev->parent = &h->dev;
 942         h->drv[drv_index]->device_initialized = 1;
 943         return device_add(dev);
 944 }
 945
 946 /*
 947  * Remove sysfs entries for a logical drive.
 948  */
 949 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 950         int ctlr_exiting)
 951 {
 952         struct device *dev = &h->drv[drv_index]->dev;
 953
 954         /* special case for c*d0, we only destroy it on controller exit */
 955         if (drv_index == 0 && !ctlr_exiting)
 956                 return;
 957
 958         device_del(dev);
 959         put_device(dev); /* the "final" put. */
 960         h->drv[drv_index] = NULL;
 961 }
 962
 963 /*
 964  * For operations that cannot sleep, a command block is allocated at init,
 965  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 966  * which ones are free or in use.
 967  */
 968 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 969 {
 970         CommandList_struct *c;
 971         int i;
 972         u64bit temp64;
 973         dma_addr_t cmd_dma_handle, err_dma_handle;
 974
 975         do {
 976                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 977                 if (i == h->nr_cmds)
 978                         return NULL;
 979         } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
 980                   h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 981         c = h->cmd_pool + i;
 982         memset(c, 0, sizeof(CommandList_struct));
 983         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
 984         c->err_info = h->errinfo_pool + i;
 985         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 986         err_dma_handle = h->errinfo_pool_dhandle
 987             + i * sizeof(ErrorInfo_struct);
 988         h->nr_allocs++;
 989
 990         c->cmdindex = i;
 991
 992         INIT_LIST_HEAD(&c->list);
 993         c->busaddr = (__u32) cmd_dma_handle;
 994         temp64.val = (__u64) err_dma_handle;
 995         c->ErrDesc.Addr.lower = temp64.val32.lower;
 996         c->ErrDesc.Addr.upper = temp64.val32.upper;
 997         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 998
 999         c->ctlr = h->ctlr;
1000         return c;
1001 }
1002
1003 /* allocate a command using pci_alloc_consistent, used for ioctls,
1004  * etc., not for the main i/o path.
1005  */
1006 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
1007 {
1008         CommandList_struct *c;
1009         u64bit temp64;
1010         dma_addr_t cmd_dma_handle, err_dma_handle;
1011
1012         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
1013                 sizeof(CommandList_struct), &cmd_dma_handle);
1014         if (c == NULL)
1015                 return NULL;
1016         memset(c, 0, sizeof(CommandList_struct));
1017
1018         c->cmdindex = -1;
1019
1020         c->err_info = (ErrorInfo_struct *)
1021             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
1022                     &err_dma_handle);
1023
1024         if (c->err_info == NULL) {
1025                 pci_free_consistent(h->pdev,
1026                         sizeof(CommandList_struct), c, cmd_dma_handle);
1027                 return NULL;
1028         }
1029         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
1030
1031         INIT_LIST_HEAD(&c->list);
1032         c->busaddr = (__u32) cmd_dma_handle;
1033         temp64.val = (__u64) err_dma_handle;
1034         c->ErrDesc.Addr.lower = temp64.val32.lower;
1035         c->ErrDesc.Addr.upper = temp64.val32.upper;
1036         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
1037
1038         c->ctlr = h->ctlr;
1039         return c;
1040 }
1041
1042 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
1043 {
1044         int i;
1045
1046         i = c - h->cmd_pool;
1047         clear_bit(i & (BITS_PER_LONG - 1),
1048                   h->cmd_pool_bits + (i / BITS_PER_LONG));
1049         h->nr_frees++;
1050 }
1051
1052 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
1053 {
1054         u64bit temp64;
1055
1056         temp64.val32.lower = c->ErrDesc.Addr.lower;
1057         temp64.val32.upper = c->ErrDesc.Addr.upper;
1058         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1059                             c->err_info, (dma_addr_t) temp64.val);
1060         pci_free_consistent(h->pdev, sizeof(CommandList_struct), c,
1061                 (dma_addr_t) cciss_tag_discard_error_bits(h, (u32) c->busaddr));
1062 }
1063
1064 static inline ctlr_info_t *get_host(struct gendisk *disk)
1065 {
1066         return disk->queue->queuedata;
1067 }
1068
1069 static inline drive_info_struct *get_drv(struct gendisk *disk)
1070 {
1071         return disk->private_data;
1072 }
1073
1074 /*
1075  * Open.  Make sure the device is really there.
1076  */
1077 static int cciss_open(struct block_device *bdev, fmode_t mode)
1078 {
1079         ctlr_info_t *h = get_host(bdev->bd_disk);
1080         drive_info_struct *drv = get_drv(bdev->bd_disk);
1081
1082         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1083         if (drv->busy_configuring)
1084                 return -EBUSY;
1085         /*
1086          * Root is allowed to open raw volume zero even if it's not configured
1087          * so array config can still work. Root is also allowed to open any
1088          * volume that has a LUN ID, so it can issue IOCTL to reread the
1089          * disk information.  I don't think I really like this
1090          * but I'm already using way to many device nodes to claim another one
1091          * for "raw controller".
1092          */
1093         if (drv->heads == 0) {
1094                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1095                         /* if not node 0 make sure it is a partition = 0 */
1096                         if (MINOR(bdev->bd_dev) & 0x0f) {
1097                                 return -ENXIO;
1098                                 /* if it is, make sure we have a LUN ID */
1099                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1100                                 sizeof(drv->LunID))) {
1101                                 return -ENXIO;
1102                         }
1103                 }
1104                 if (!capable(CAP_SYS_ADMIN))
1105                         return -EPERM;
1106         }
1107         drv->usage_count++;
1108         h->usage_count++;
1109         return 0;
1110 }
1111
1112 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1113 {
1114         int ret;
1115
1116         mutex_lock(&cciss_mutex);
1117         ret = cciss_open(bdev, mode);
1118         mutex_unlock(&cciss_mutex);
1119
1120         return ret;
1121 }
1122
1123 /*
1124  * Close.  Sync first.
1125  */
1126 static int cciss_release(struct gendisk *disk, fmode_t mode)
1127 {
1128         ctlr_info_t *h;
1129         drive_info_struct *drv;
1130
1131         mutex_lock(&cciss_mutex);
1132         h = get_host(disk);
1133         drv = get_drv(disk);
1134         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1135         drv->usage_count--;
1136         h->usage_count--;
1137         mutex_unlock(&cciss_mutex);
1138         return 0;
1139 }
1140
1141 #ifdef CONFIG_COMPAT
1142
1143 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1144                                   unsigned cmd, unsigned long arg);
1145 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1146                                       unsigned cmd, unsigned long arg);
1147
1148 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1149                               unsigned cmd, unsigned long arg)
1150 {
1151         switch (cmd) {
1152         case CCISS_GETPCIINFO:
1153         case CCISS_GETINTINFO:
1154         case CCISS_SETINTINFO:
1155         case CCISS_GETNODENAME:
1156         case CCISS_SETNODENAME:
1157         case CCISS_GETHEARTBEAT:
1158         case CCISS_GETBUSTYPES:
1159         case CCISS_GETFIRMVER:
1160         case CCISS_GETDRIVVER:
1161         case CCISS_REVALIDVOLS:
1162         case CCISS_DEREGDISK:
1163         case CCISS_REGNEWDISK:
1164         case CCISS_REGNEWD:
1165         case CCISS_RESCANDISK:
1166         case CCISS_GETLUNINFO:
1167                 return cciss_ioctl(bdev, mode, cmd, arg);
1168
1169         case CCISS_PASSTHRU32:
1170                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1171         case CCISS_BIG_PASSTHRU32:
1172                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1173
1174         default:
1175                 return -ENOIOCTLCMD;
1176         }
1177 }
1178
1179 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1180                                   unsigned cmd, unsigned long arg)
1181 {
1182         IOCTL32_Command_struct __user *arg32 =
1183             (IOCTL32_Command_struct __user *) arg;
1184         IOCTL_Command_struct arg64;
1185         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1186         int err;
1187         u32 cp;
1188
1189         memset(&arg64, 0, sizeof(arg64));
1190         err = 0;
1191         err |=
1192             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1193                            sizeof(arg64.LUN_info));
1194         err |=
1195             copy_from_user(&arg64.Request, &arg32->Request,
1196                            sizeof(arg64.Request));
1197         err |=
1198             copy_from_user(&arg64.error_info, &arg32->error_info,
1199                            sizeof(arg64.error_info));
1200         err |= get_user(arg64.buf_size, &arg32->buf_size);
1201         err |= get_user(cp, &arg32->buf);
1202         arg64.buf = compat_ptr(cp);
1203         err |= copy_to_user(p, &arg64, sizeof(arg64));
1204
1205         if (err)
1206                 return -EFAULT;
1207
1208         err = cciss_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1209         if (err)
1210                 return err;
1211         err |=
1212             copy_in_user(&arg32->error_info, &p->error_info,
1213                          sizeof(arg32->error_info));
1214         if (err)
1215                 return -EFAULT;
1216         return err;
1217 }
1218
1219 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1220                                       unsigned cmd, unsigned long arg)
1221 {
1222         BIG_IOCTL32_Command_struct __user *arg32 =
1223             (BIG_IOCTL32_Command_struct __user *) arg;
1224         BIG_IOCTL_Command_struct arg64;
1225         BIG_IOCTL_Command_struct __user *p =
1226             compat_alloc_user_space(sizeof(arg64));
1227         int err;
1228         u32 cp;
1229
1230         memset(&arg64, 0, sizeof(arg64));
1231         err = 0;
1232         err |=
1233             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1234                            sizeof(arg64.LUN_info));
1235         err |=
1236             copy_from_user(&arg64.Request, &arg32->Request,
1237                            sizeof(arg64.Request));
1238         err |=
1239             copy_from_user(&arg64.error_info, &arg32->error_info,
1240                            sizeof(arg64.error_info));
1241         err |= get_user(arg64.buf_size, &arg32->buf_size);
1242         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1243         err |= get_user(cp, &arg32->buf);
1244         arg64.buf = compat_ptr(cp);
1245         err |= copy_to_user(p, &arg64, sizeof(arg64));
1246
1247         if (err)
1248                 return -EFAULT;
1249
1250         err = cciss_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1251         if (err)
1252                 return err;
1253         err |=
1254             copy_in_user(&arg32->error_info, &p->error_info,
1255                          sizeof(arg32->error_info));
1256         if (err)
1257                 return -EFAULT;
1258         return err;
1259 }
1260 #endif
1261
1262 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1263 {
1264         drive_info_struct *drv = get_drv(bdev->bd_disk);
1265
1266         if (!drv->cylinders)
1267                 return -ENXIO;
1268
1269         geo->heads = drv->heads;
1270         geo->sectors = drv->sectors;
1271         geo->cylinders = drv->cylinders;
1272         return 0;
1273 }
1274
1275 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1276 {
1277         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1278                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1279                 (void)check_for_unit_attention(h, c);
1280 }
1281
1282 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1283 {
1284         cciss_pci_info_struct pciinfo;
1285
1286         if (!argp)
1287                 return -EINVAL;
1288         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1289         pciinfo.bus = h->pdev->bus->number;
1290         pciinfo.dev_fn = h->pdev->devfn;
1291         pciinfo.board_id = h->board_id;
1292         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1293                 return -EFAULT;
1294         return 0;
1295 }
1296
1297 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1298 {
1299         cciss_coalint_struct intinfo;
1300         unsigned long flags;
1301
1302         if (!argp)
1303                 return -EINVAL;
1304         spin_lock_irqsave(&h->lock, flags);
1305         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1306         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1307         spin_unlock_irqrestore(&h->lock, flags);
1308         if (copy_to_user
1309             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1310                 return -EFAULT;
1311         return 0;
1312 }
1313
1314 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1315 {
1316         cciss_coalint_struct intinfo;
1317         unsigned long flags;
1318         int i;
1319
1320         if (!argp)
1321                 return -EINVAL;
1322         if (!capable(CAP_SYS_ADMIN))
1323                 return -EPERM;
1324         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1325                 return -EFAULT;
1326         if ((intinfo.delay == 0) && (intinfo.count == 0))
1327                 return -EINVAL;
1328         spin_lock_irqsave(&h->lock, flags);
1329         /* Update the field, and then ring the doorbell */
1330         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1331         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1332         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1333
1334         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1335                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1336                         break;
1337                 udelay(1000); /* delay and try again */
1338         }
1339         spin_unlock_irqrestore(&h->lock, flags);
1340         if (i >= MAX_IOCTL_CONFIG_WAIT)
1341                 return -EAGAIN;
1342         return 0;
1343 }
1344
1345 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1346 {
1347         NodeName_type NodeName;
1348         unsigned long flags;
1349         int i;
1350
1351         if (!argp)
1352                 return -EINVAL;
1353         spin_lock_irqsave(&h->lock, flags);
1354         for (i = 0; i < 16; i++)
1355                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1356         spin_unlock_irqrestore(&h->lock, flags);
1357         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1358                 return -EFAULT;
1359         return 0;
1360 }
1361
1362 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1363 {
1364         NodeName_type NodeName;
1365         unsigned long flags;
1366         int i;
1367
1368         if (!argp)
1369                 return -EINVAL;
1370         if (!capable(CAP_SYS_ADMIN))
1371                 return -EPERM;
1372         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1373                 return -EFAULT;
1374         spin_lock_irqsave(&h->lock, flags);
1375         /* Update the field, and then ring the doorbell */
1376         for (i = 0; i < 16; i++)
1377                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1378         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1379         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1380                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1381                         break;
1382                 udelay(1000); /* delay and try again */
1383         }
1384         spin_unlock_irqrestore(&h->lock, flags);
1385         if (i >= MAX_IOCTL_CONFIG_WAIT)
1386                 return -EAGAIN;
1387         return 0;
1388 }
1389
1390 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1391 {
1392         Heartbeat_type heartbeat;
1393         unsigned long flags;
1394
1395         if (!argp)
1396                 return -EINVAL;
1397         spin_lock_irqsave(&h->lock, flags);
1398         heartbeat = readl(&h->cfgtable->HeartBeat);
1399         spin_unlock_irqrestore(&h->lock, flags);
1400         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1401                 return -EFAULT;
1402         return 0;
1403 }
1404
1405 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1406 {
1407         BusTypes_type BusTypes;
1408         unsigned long flags;
1409
1410         if (!argp)
1411                 return -EINVAL;
1412         spin_lock_irqsave(&h->lock, flags);
1413         BusTypes = readl(&h->cfgtable->BusTypes);
1414         spin_unlock_irqrestore(&h->lock, flags);
1415         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1416                 return -EFAULT;
1417         return 0;
1418 }
1419
1420 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1421 {
1422         FirmwareVer_type firmware;
1423
1424         if (!argp)
1425                 return -EINVAL;
1426         memcpy(firmware, h->firm_ver, 4);
1427
1428         if (copy_to_user
1429             (argp, firmware, sizeof(FirmwareVer_type)))
1430                 return -EFAULT;
1431         return 0;
1432 }
1433
1434 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1435 {
1436         DriverVer_type DriverVer = DRIVER_VERSION;
1437
1438         if (!argp)
1439                 return -EINVAL;
1440         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1441                 return -EFAULT;
1442         return 0;
1443 }
1444
1445 static int cciss_getluninfo(ctlr_info_t *h,
1446         struct gendisk *disk, void __user *argp)
1447 {
1448         LogvolInfo_struct luninfo;
1449         drive_info_struct *drv = get_drv(disk);
1450
1451         if (!argp)
1452                 return -EINVAL;
1453         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1454         luninfo.num_opens = drv->usage_count;
1455         luninfo.num_parts = 0;
1456         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1457                 return -EFAULT;
1458         return 0;
1459 }
1460
1461 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1462 {
1463         IOCTL_Command_struct iocommand;
1464         CommandList_struct *c;
1465         char *buff = NULL;
1466         u64bit temp64;
1467         DECLARE_COMPLETION_ONSTACK(wait);
1468
1469         if (!argp)
1470                 return -EINVAL;
1471
1472         if (!capable(CAP_SYS_RAWIO))
1473                 return -EPERM;
1474
1475         if (copy_from_user
1476             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1477                 return -EFAULT;
1478         if ((iocommand.buf_size < 1) &&
1479             (iocommand.Request.Type.Direction != XFER_NONE)) {
1480                 return -EINVAL;
1481         }
1482         if (iocommand.buf_size > 0) {
1483                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1484                 if (buff == NULL)
1485                         return -EFAULT;
1486         }
1487         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1488                 /* Copy the data into the buffer we created */
1489                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1490                         kfree(buff);
1491                         return -EFAULT;
1492                 }
1493         } else {
1494                 memset(buff, 0, iocommand.buf_size);
1495         }
1496         c = cmd_special_alloc(h);
1497         if (!c) {
1498                 kfree(buff);
1499                 return -ENOMEM;
1500         }
1501         /* Fill in the command type */
1502         c->cmd_type = CMD_IOCTL_PEND;
1503         /* Fill in Command Header */
1504         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1505         if (iocommand.buf_size > 0) { /* buffer to fill */
1506                 c->Header.SGList = 1;
1507                 c->Header.SGTotal = 1;
1508         } else { /* no buffers to fill */
1509                 c->Header.SGList = 0;
1510                 c->Header.SGTotal = 0;
1511         }
1512         c->Header.LUN = iocommand.LUN_info;
1513         /* use the kernel address the cmd block for tag */
1514         c->Header.Tag.lower = c->busaddr;
1515
1516         /* Fill in Request block */
1517         c->Request = iocommand.Request;
1518
1519         /* Fill in the scatter gather information */
1520         if (iocommand.buf_size > 0) {
1521                 temp64.val = pci_map_single(h->pdev, buff,
1522                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1523                 c->SG[0].Addr.lower = temp64.val32.lower;
1524                 c->SG[0].Addr.upper = temp64.val32.upper;
1525                 c->SG[0].Len = iocommand.buf_size;
1526                 c->SG[0].Ext = 0;  /* we are not chaining */
1527         }
1528         c->waiting = &wait;
1529
1530         enqueue_cmd_and_start_io(h, c);
1531         wait_for_completion(&wait);
1532
1533         /* unlock the buffers from DMA */
1534         temp64.val32.lower = c->SG[0].Addr.lower;
1535         temp64.val32.upper = c->SG[0].Addr.upper;
1536         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1537                          PCI_DMA_BIDIRECTIONAL);
1538         check_ioctl_unit_attention(h, c);
1539
1540         /* Copy the error information out */
1541         iocommand.error_info = *(c->err_info);
1542         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1543                 kfree(buff);
1544                 cmd_special_free(h, c);
1545                 return -EFAULT;
1546         }
1547
1548         if (iocommand.Request.Type.Direction == XFER_READ) {
1549                 /* Copy the data out of the buffer we created */
1550                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1551                         kfree(buff);
1552                         cmd_special_free(h, c);
1553                         return -EFAULT;
1554                 }
1555         }
1556         kfree(buff);
1557         cmd_special_free(h, c);
1558         return 0;
1559 }
1560
1561 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1562 {
1563         BIG_IOCTL_Command_struct *ioc;
1564         CommandList_struct *c;
1565         unsigned char **buff = NULL;
1566         int *buff_size = NULL;
1567         u64bit temp64;
1568         BYTE sg_used = 0;
1569         int status = 0;
1570         int i;
1571         DECLARE_COMPLETION_ONSTACK(wait);
1572         __u32 left;
1573         __u32 sz;
1574         BYTE __user *data_ptr;
1575
1576         if (!argp)
1577                 return -EINVAL;
1578         if (!capable(CAP_SYS_RAWIO))
1579                 return -EPERM;
1580         ioc = kmalloc(sizeof(*ioc), GFP_KERNEL);
1581         if (!ioc) {
1582                 status = -ENOMEM;
1583                 goto cleanup1;
1584         }
1585         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1586                 status = -EFAULT;
1587                 goto cleanup1;
1588         }
1589         if ((ioc->buf_size < 1) &&
1590             (ioc->Request.Type.Direction != XFER_NONE)) {
1591                 status = -EINVAL;
1592                 goto cleanup1;
1593         }
1594         /* Check kmalloc limits  using all SGs */
1595         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1596                 status = -EINVAL;
1597                 goto cleanup1;
1598         }
1599         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1600                 status = -EINVAL;
1601                 goto cleanup1;
1602         }
1603         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1604         if (!buff) {
1605                 status = -ENOMEM;
1606                 goto cleanup1;
1607         }
1608         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1609         if (!buff_size) {
1610                 status = -ENOMEM;
1611                 goto cleanup1;
1612         }
1613         left = ioc->buf_size;
1614         data_ptr = ioc->buf;
1615         while (left) {
1616                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1617                 buff_size[sg_used] = sz;
1618                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1619                 if (buff[sg_used] == NULL) {
1620                         status = -ENOMEM;
1621                         goto cleanup1;
1622                 }
1623                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1624                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1625                                 status = -EFAULT;
1626                                 goto cleanup1;
1627                         }
1628                 } else {
1629                         memset(buff[sg_used], 0, sz);
1630                 }
1631                 left -= sz;
1632                 data_ptr += sz;
1633                 sg_used++;
1634         }
1635         c = cmd_special_alloc(h);
1636         if (!c) {
1637                 status = -ENOMEM;
1638                 goto cleanup1;
1639         }
1640         c->cmd_type = CMD_IOCTL_PEND;
1641         c->Header.ReplyQueue = 0;
1642         c->Header.SGList = sg_used;
1643         c->Header.SGTotal = sg_used;
1644         c->Header.LUN = ioc->LUN_info;
1645         c->Header.Tag.lower = c->busaddr;
1646
1647         c->Request = ioc->Request;
1648         for (i = 0; i < sg_used; i++) {
1649                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1650                                     PCI_DMA_BIDIRECTIONAL);
1651                 c->SG[i].Addr.lower = temp64.val32.lower;
1652                 c->SG[i].Addr.upper = temp64.val32.upper;
1653                 c->SG[i].Len = buff_size[i];
1654                 c->SG[i].Ext = 0;       /* we are not chaining */
1655         }
1656         c->waiting = &wait;
1657         enqueue_cmd_and_start_io(h, c);
1658         wait_for_completion(&wait);
1659         /* unlock the buffers from DMA */
1660         for (i = 0; i < sg_used; i++) {
1661                 temp64.val32.lower = c->SG[i].Addr.lower;
1662                 temp64.val32.upper = c->SG[i].Addr.upper;
1663                 pci_unmap_single(h->pdev,
1664                         (dma_addr_t) temp64.val, buff_size[i],
1665                         PCI_DMA_BIDIRECTIONAL);
1666         }
1667         check_ioctl_unit_attention(h, c);
1668         /* Copy the error information out */
1669         ioc->error_info = *(c->err_info);
1670         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1671                 cmd_special_free(h, c);
1672                 status = -EFAULT;
1673                 goto cleanup1;
1674         }
1675         if (ioc->Request.Type.Direction == XFER_READ) {
1676                 /* Copy the data out of the buffer we created */
1677                 BYTE __user *ptr = ioc->buf;
1678                 for (i = 0; i < sg_used; i++) {
1679                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1680                                 cmd_special_free(h, c);
1681                                 status = -EFAULT;
1682                                 goto cleanup1;
1683                         }
1684                         ptr += buff_size[i];
1685                 }
1686         }
1687         cmd_special_free(h, c);
1688         status = 0;
1689 cleanup1:
1690         if (buff) {
1691                 for (i = 0; i < sg_used; i++)
1692                         kfree(buff[i]);
1693                 kfree(buff);
1694         }
1695         kfree(buff_size);
1696         kfree(ioc);
1697         return status;
1698 }
1699
1700 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1701         unsigned int cmd, unsigned long arg)
1702 {
1703         struct gendisk *disk = bdev->bd_disk;
1704         ctlr_info_t *h = get_host(disk);
1705         void __user *argp = (void __user *)arg;
1706
1707         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1708                 cmd, arg);
1709         switch (cmd) {
1710         case CCISS_GETPCIINFO:
1711                 return cciss_getpciinfo(h, argp);
1712         case CCISS_GETINTINFO:
1713                 return cciss_getintinfo(h, argp);
1714         case CCISS_SETINTINFO:
1715                 return cciss_setintinfo(h, argp);
1716         case CCISS_GETNODENAME:
1717                 return cciss_getnodename(h, argp);
1718         case CCISS_SETNODENAME:
1719                 return cciss_setnodename(h, argp);
1720         case CCISS_GETHEARTBEAT:
1721                 return cciss_getheartbeat(h, argp);
1722         case CCISS_GETBUSTYPES:
1723                 return cciss_getbustypes(h, argp);
1724         case CCISS_GETFIRMVER:
1725                 return cciss_getfirmver(h, argp);
1726         case CCISS_GETDRIVVER:
1727                 return cciss_getdrivver(h, argp);
1728         case CCISS_DEREGDISK:
1729         case CCISS_REGNEWD:
1730         case CCISS_REVALIDVOLS:
1731                 return rebuild_lun_table(h, 0, 1);
1732         case CCISS_GETLUNINFO:
1733                 return cciss_getluninfo(h, disk, argp);
1734         case CCISS_PASSTHRU:
1735                 return cciss_passthru(h, argp);
1736         case CCISS_BIG_PASSTHRU:
1737                 return cciss_bigpassthru(h, argp);
1738
1739         /* scsi_cmd_blk_ioctl handles these, below, though some are not */
1740         /* very meaningful for cciss.  SG_IO is the main one people want. */
1741
1742         case SG_GET_VERSION_NUM:
1743         case SG_SET_TIMEOUT:
1744         case SG_GET_TIMEOUT:
1745         case SG_GET_RESERVED_SIZE:
1746         case SG_SET_RESERVED_SIZE:
1747         case SG_EMULATED_HOST:
1748         case SG_IO:
1749         case SCSI_IOCTL_SEND_COMMAND:
1750                 return scsi_cmd_blk_ioctl(bdev, mode, cmd, argp);
1751
1752         /* scsi_cmd_blk_ioctl would normally handle these, below, but */
1753         /* they aren't a good fit for cciss, as CD-ROMs are */
1754         /* not supported, and we don't have any bus/target/lun */
1755         /* which we present to the kernel. */
1756
1757         case CDROM_SEND_PACKET:
1758         case CDROMCLOSETRAY:
1759         case CDROMEJECT:
1760         case SCSI_IOCTL_GET_IDLUN:
1761         case SCSI_IOCTL_GET_BUS_NUMBER:
1762         default:
1763                 return -ENOTTY;
1764         }
1765 }
1766
1767 static void cciss_check_queues(ctlr_info_t *h)
1768 {
1769         int start_queue = h->next_to_run;
1770         int i;
1771
1772         /* check to see if we have maxed out the number of commands that can
1773          * be placed on the queue.  If so then exit.  We do this check here
1774          * in case the interrupt we serviced was from an ioctl and did not
1775          * free any new commands.
1776          */
1777         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1778                 return;
1779
1780         /* We have room on the queue for more commands.  Now we need to queue
1781          * them up.  We will also keep track of the next queue to run so
1782          * that every queue gets a chance to be started first.
1783          */
1784         for (i = 0; i < h->highest_lun + 1; i++) {
1785                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1786                 /* make sure the disk has been added and the drive is real
1787                  * because this can be called from the middle of init_one.
1788                  */
1789                 if (!h->drv[curr_queue])
1790                         continue;
1791                 if (!(h->drv[curr_queue]->queue) ||
1792                         !(h->drv[curr_queue]->heads))
1793                         continue;
1794                 blk_start_queue(h->gendisk[curr_queue]->queue);
1795
1796                 /* check to see if we have maxed out the number of commands
1797                  * that can be placed on the queue.
1798                  */
1799                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1800                         if (curr_queue == start_queue) {
1801                                 h->next_to_run =
1802                                     (start_queue + 1) % (h->highest_lun + 1);
1803                                 break;
1804                         } else {
1805                                 h->next_to_run = curr_queue;
1806                                 break;
1807                         }
1808                 }
1809         }
1810 }
1811
1812 static void cciss_softirq_done(struct request *rq)
1813 {
1814         CommandList_struct *c = rq->completion_data;
1815         ctlr_info_t *h = hba[c->ctlr];
1816         SGDescriptor_struct *curr_sg = c->SG;
1817         u64bit temp64;
1818         unsigned long flags;
1819         int i, ddir;
1820         int sg_index = 0;
1821
1822         if (c->Request.Type.Direction == XFER_READ)
1823                 ddir = PCI_DMA_FROMDEVICE;
1824         else
1825                 ddir = PCI_DMA_TODEVICE;
1826
1827         /* command did not need to be retried */
1828         /* unmap the DMA mapping for all the scatter gather elements */
1829         for (i = 0; i < c->Header.SGList; i++) {
1830                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1831                         cciss_unmap_sg_chain_block(h, c);
1832                         /* Point to the next block */
1833                         curr_sg = h->cmd_sg_list[c->cmdindex];
1834                         sg_index = 0;
1835                 }
1836                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1837                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1838                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1839                                 ddir);
1840                 ++sg_index;
1841         }
1842
1843         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1844
1845         /* set the residual count for pc requests */
1846         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1847                 rq->resid_len = c->err_info->ResidualCnt;
1848
1849         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1850
1851         spin_lock_irqsave(&h->lock, flags);
1852         cmd_free(h, c);
1853         cciss_check_queues(h);
1854         spin_unlock_irqrestore(&h->lock, flags);
1855 }
1856
1857 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1858         unsigned char scsi3addr[], uint32_t log_unit)
1859 {
1860         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1861                 sizeof(h->drv[log_unit]->LunID));
1862 }
1863
1864 /* This function gets the SCSI vendor, model, and revision of a logical drive
1865  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1866  * they cannot be read.
1867  */
1868 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1869                                    char *vendor, char *model, char *rev)
1870 {
1871         int rc;
1872         InquiryData_struct *inq_buf;
1873         unsigned char scsi3addr[8];
1874
1875         *vendor = '\0';
1876         *model = '\0';
1877         *rev = '\0';
1878
1879         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1880         if (!inq_buf)
1881                 return;
1882
1883         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1884         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1885                         scsi3addr, TYPE_CMD);
1886         if (rc == IO_OK) {
1887                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1888                 vendor[VENDOR_LEN] = '\0';
1889                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1890                 model[MODEL_LEN] = '\0';
1891                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1892                 rev[REV_LEN] = '\0';
1893         }
1894
1895         kfree(inq_buf);
1896         return;
1897 }
1898
1899 /* This function gets the serial number of a logical drive via
1900  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1901  * number cannot be had, for whatever reason, 16 bytes of 0xff
1902  * are returned instead.
1903  */
1904 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1905                                 unsigned char *serial_no, int buflen)
1906 {
1907 #define PAGE_83_INQ_BYTES 64
1908         int rc;
1909         unsigned char *buf;
1910         unsigned char scsi3addr[8];
1911
1912         if (buflen > 16)
1913                 buflen = 16;
1914         memset(serial_no, 0xff, buflen);
1915         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1916         if (!buf)
1917                 return;
1918         memset(serial_no, 0, buflen);
1919         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1920         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1921                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1922         if (rc == IO_OK)
1923                 memcpy(serial_no, &buf[8], buflen);
1924         kfree(buf);
1925         return;
1926 }
1927
1928 /*
1929  * cciss_add_disk sets up the block device queue for a logical drive
1930  */
1931 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1932                                 int drv_index)
1933 {
1934         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1935         if (!disk->queue)
1936                 goto init_queue_failure;
1937         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1938         disk->major = h->major;
1939         disk->first_minor = drv_index << NWD_SHIFT;
1940         disk->fops = &cciss_fops;
1941         if (cciss_create_ld_sysfs_entry(h, drv_index))
1942                 goto cleanup_queue;
1943         disk->private_data = h->drv[drv_index];
1944         disk->driverfs_dev = &h->drv[drv_index]->dev;
1945
1946         /* Set up queue information */
1947         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1948
1949         /* This is a hardware imposed limit. */
1950         blk_queue_max_segments(disk->queue, h->maxsgentries);
1951
1952         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1953
1954         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1955
1956         disk->queue->queuedata = h;
1957
1958         blk_queue_logical_block_size(disk->queue,
1959                                      h->drv[drv_index]->block_size);
1960
1961         /* Make sure all queue data is written out before */
1962         /* setting h->drv[drv_index]->queue, as setting this */
1963         /* allows the interrupt handler to start the queue */
1964         wmb();
1965         h->drv[drv_index]->queue = disk->queue;
1966         add_disk(disk);
1967         return 0;
1968
1969 cleanup_queue:
1970         blk_cleanup_queue(disk->queue);
1971         disk->queue = NULL;
1972 init_queue_failure:
1973         return -1;
1974 }
1975
1976 /* This function will check the usage_count of the drive to be updated/added.
1977  * If the usage_count is zero and it is a heretofore unknown drive, or,
1978  * the drive's capacity, geometry, or serial number has changed,
1979  * then the drive information will be updated and the disk will be
1980  * re-registered with the kernel.  If these conditions don't hold,
1981  * then it will be left alone for the next reboot.  The exception to this
1982  * is disk 0 which will always be left registered with the kernel since it
1983  * is also the controller node.  Any changes to disk 0 will show up on
1984  * the next reboot.
1985  */
1986 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1987         int first_time, int via_ioctl)
1988 {
1989         struct gendisk *disk;
1990         InquiryData_struct *inq_buff = NULL;
1991         unsigned int block_size;
1992         sector_t total_size;
1993         unsigned long flags = 0;
1994         int ret = 0;
1995         drive_info_struct *drvinfo;
1996
1997         /* Get information about the disk and modify the driver structure */
1998         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1999         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
2000         if (inq_buff == NULL || drvinfo == NULL)
2001                 goto mem_msg;
2002
2003         /* testing to see if 16-byte CDBs are already being used */
2004         if (h->cciss_read == CCISS_READ_16) {
2005                 cciss_read_capacity_16(h, drv_index,
2006                         &total_size, &block_size);
2007
2008         } else {
2009                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
2010                 /* if read_capacity returns all F's this volume is >2TB */
2011                 /* in size so we switch to 16-byte CDB's for all */
2012                 /* read/write ops */
2013                 if (total_size == 0xFFFFFFFFULL) {
2014                         cciss_read_capacity_16(h, drv_index,
2015                         &total_size, &block_size);
2016                         h->cciss_read = CCISS_READ_16;
2017                         h->cciss_write = CCISS_WRITE_16;
2018                 } else {
2019                         h->cciss_read = CCISS_READ_10;
2020                         h->cciss_write = CCISS_WRITE_10;
2021                 }
2022         }
2023
2024         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
2025                                inq_buff, drvinfo);
2026         drvinfo->block_size = block_size;
2027         drvinfo->nr_blocks = total_size + 1;
2028
2029         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
2030                                 drvinfo->model, drvinfo->rev);
2031         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
2032                         sizeof(drvinfo->serial_no));
2033         /* Save the lunid in case we deregister the disk, below. */
2034         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
2035                 sizeof(drvinfo->LunID));
2036
2037         /* Is it the same disk we already know, and nothing's changed? */
2038         if (h->drv[drv_index]->raid_level != -1 &&
2039                 ((memcmp(drvinfo->serial_no,
2040                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2041                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2042                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2043                 drvinfo->heads == h->drv[drv_index]->heads &&
2044                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2045                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2046                         /* The disk is unchanged, nothing to update */
2047                         goto freeret;
2048
2049         /* If we get here it's not the same disk, or something's changed,
2050          * so we need to * deregister it, and re-register it, if it's not
2051          * in use.
2052          * If the disk already exists then deregister it before proceeding
2053          * (unless it's the first disk (for the controller node).
2054          */
2055         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2056                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2057                 spin_lock_irqsave(&h->lock, flags);
2058                 h->drv[drv_index]->busy_configuring = 1;
2059                 spin_unlock_irqrestore(&h->lock, flags);
2060
2061                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2062                  * which keeps the interrupt handler from starting
2063                  * the queue.
2064                  */
2065                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2066         }
2067
2068         /* If the disk is in use return */
2069         if (ret)
2070                 goto freeret;
2071
2072         /* Save the new information from cciss_geometry_inquiry
2073          * and serial number inquiry.  If the disk was deregistered
2074          * above, then h->drv[drv_index] will be NULL.
2075          */
2076         if (h->drv[drv_index] == NULL) {
2077                 drvinfo->device_initialized = 0;
2078                 h->drv[drv_index] = drvinfo;
2079                 drvinfo = NULL; /* so it won't be freed below. */
2080         } else {
2081                 /* special case for cxd0 */
2082                 h->drv[drv_index]->block_size = drvinfo->block_size;
2083                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2084                 h->drv[drv_index]->heads = drvinfo->heads;
2085                 h->drv[drv_index]->sectors = drvinfo->sectors;
2086                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2087                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2088                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2089                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2090                         VENDOR_LEN + 1);
2091                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2092                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2093         }
2094
2095         ++h->num_luns;
2096         disk = h->gendisk[drv_index];
2097         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2098
2099         /* If it's not disk 0 (drv_index != 0)
2100          * or if it was disk 0, but there was previously
2101          * no actual corresponding configured logical drive
2102          * (raid_leve == -1) then we want to update the
2103          * logical drive's information.
2104          */
2105         if (drv_index || first_time) {
2106                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2107                         cciss_free_gendisk(h, drv_index);
2108                         cciss_free_drive_info(h, drv_index);
2109                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2110                                 drv_index);
2111                         --h->num_luns;
2112                 }
2113         }
2114
2115 freeret:
2116         kfree(inq_buff);
2117         kfree(drvinfo);
2118         return;
2119 mem_msg:
2120         dev_err(&h->pdev->dev, "out of memory\n");
2121         goto freeret;
2122 }
2123
2124 /* This function will find the first index of the controllers drive array
2125  * that has a null drv pointer and allocate the drive info struct and
2126  * will return that index   This is where new drives will be added.
2127  * If the index to be returned is greater than the highest_lun index for
2128  * the controller then highest_lun is set * to this new index.
2129  * If there are no available indexes or if tha allocation fails, then -1
2130  * is returned.  * "controller_node" is used to know if this is a real
2131  * logical drive, or just the controller node, which determines if this
2132  * counts towards highest_lun.
2133  */
2134 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2135 {
2136         int i;
2137         drive_info_struct *drv;
2138
2139         /* Search for an empty slot for our drive info */
2140         for (i = 0; i < CISS_MAX_LUN; i++) {
2141
2142                 /* if not cxd0 case, and it's occupied, skip it. */
2143                 if (h->drv[i] && i != 0)
2144                         continue;
2145                 /*
2146                  * If it's cxd0 case, and drv is alloc'ed already, and a
2147                  * disk is configured there, skip it.
2148                  */
2149                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2150                         continue;
2151
2152                 /*
2153                  * We've found an empty slot.  Update highest_lun
2154                  * provided this isn't just the fake cxd0 controller node.
2155                  */
2156                 if (i > h->highest_lun && !controller_node)
2157                         h->highest_lun = i;
2158
2159                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2160                 if (i == 0 && h->drv[i] != NULL)
2161                         return i;
2162
2163                 /*
2164                  * Found an empty slot, not already alloc'ed.  Allocate it.
2165                  * Mark it with raid_level == -1, so we know it's new later on.
2166                  */
2167                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2168                 if (!drv)
2169                         return -1;
2170                 drv->raid_level = -1; /* so we know it's new */
2171                 h->drv[i] = drv;
2172                 return i;
2173         }
2174         return -1;
2175 }
2176
2177 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2178 {
2179         kfree(h->drv[drv_index]);
2180         h->drv[drv_index] = NULL;
2181 }
2182
2183 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2184 {
2185         put_disk(h->gendisk[drv_index]);
2186         h->gendisk[drv_index] = NULL;
2187 }
2188
2189 /* cciss_add_gendisk finds a free hba[]->drv structure
2190  * and allocates a gendisk if needed, and sets the lunid
2191  * in the drvinfo structure.   It returns the index into
2192  * the ->drv[] array, or -1 if none are free.
2193  * is_controller_node indicates whether highest_lun should
2194  * count this disk, or if it's only being added to provide
2195  * a means to talk to the controller in case no logical
2196  * drives have yet been configured.
2197  */
2198 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2199         int controller_node)
2200 {
2201         int drv_index;
2202
2203         drv_index = cciss_alloc_drive_info(h, controller_node);
2204         if (drv_index == -1)
2205                 return -1;
2206
2207         /*Check if the gendisk needs to be allocated */
2208         if (!h->gendisk[drv_index]) {
2209                 h->gendisk[drv_index] =
2210                         alloc_disk(1 << NWD_SHIFT);
2211                 if (!h->gendisk[drv_index]) {
2212                         dev_err(&h->pdev->dev,
2213                                 "could not allocate a new disk %d\n",
2214                                 drv_index);
2215                         goto err_free_drive_info;
2216                 }
2217         }
2218         memcpy(h->drv[drv_index]->LunID, lunid,
2219                 sizeof(h->drv[drv_index]->LunID));
2220         if (cciss_create_ld_sysfs_entry(h, drv_index))
2221                 goto err_free_disk;
2222         /* Don't need to mark this busy because nobody */
2223         /* else knows about this disk yet to contend */
2224         /* for access to it. */
2225         h->drv[drv_index]->busy_configuring = 0;
2226         wmb();
2227         return drv_index;
2228
2229 err_free_disk:
2230         cciss_free_gendisk(h, drv_index);
2231 err_free_drive_info:
2232         cciss_free_drive_info(h, drv_index);
2233         return -1;
2234 }
2235
2236 /* This is for the special case of a controller which
2237  * has no logical drives.  In this case, we still need
2238  * to register a disk so the controller can be accessed
2239  * by the Array Config Utility.
2240  */
2241 static void cciss_add_controller_node(ctlr_info_t *h)
2242 {
2243         struct gendisk *disk;
2244         int drv_index;
2245
2246         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2247                 return;
2248
2249         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2250         if (drv_index == -1)
2251                 goto error;
2252         h->drv[drv_index]->block_size = 512;
2253         h->drv[drv_index]->nr_blocks = 0;
2254         h->drv[drv_index]->heads = 0;
2255         h->drv[drv_index]->sectors = 0;
2256         h->drv[drv_index]->cylinders = 0;
2257         h->drv[drv_index]->raid_level = -1;
2258         memset(h->drv[drv_index]->serial_no, 0, 16);
2259         disk = h->gendisk[drv_index];
2260         if (cciss_add_disk(h, disk, drv_index) == 0)
2261                 return;
2262         cciss_free_gendisk(h, drv_index);
2263         cciss_free_drive_info(h, drv_index);
2264 error:
2265         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2266         return;
2267 }
2268
2269 /* This function will add and remove logical drives from the Logical
2270  * drive array of the controller and maintain persistency of ordering
2271  * so that mount points are preserved until the next reboot.  This allows
2272  * for the removal of logical drives in the middle of the drive array
2273  * without a re-ordering of those drives.
2274  * INPUT
2275  * h            = The controller to perform the operations on
2276  */
2277 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2278         int via_ioctl)
2279 {
2280         int num_luns;
2281         ReportLunData_struct *ld_buff = NULL;
2282         int return_code;
2283         int listlength = 0;
2284         int i;
2285         int drv_found;
2286         int drv_index = 0;
2287         unsigned char lunid[8] = CTLR_LUNID;
2288         unsigned long flags;
2289
2290         if (!capable(CAP_SYS_RAWIO))
2291                 return -EPERM;
2292
2293         /* Set busy_configuring flag for this operation */
2294         spin_lock_irqsave(&h->lock, flags);
2295         if (h->busy_configuring) {
2296                 spin_unlock_irqrestore(&h->lock, flags);
2297                 return -EBUSY;
2298         }
2299         h->busy_configuring = 1;
2300         spin_unlock_irqrestore(&h->lock, flags);
2301
2302         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2303         if (ld_buff == NULL)
2304                 goto mem_msg;
2305
2306         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2307                                       sizeof(ReportLunData_struct),
2308                                       0, CTLR_LUNID, TYPE_CMD);
2309
2310         if (return_code == IO_OK)
2311                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2312         else {  /* reading number of logical volumes failed */
2313                 dev_warn(&h->pdev->dev,
2314                         "report logical volume command failed\n");
2315                 listlength = 0;
2316                 goto freeret;
2317         }
2318
2319         num_luns = listlength / 8;      /* 8 bytes per entry */
2320         if (num_luns > CISS_MAX_LUN) {
2321                 num_luns = CISS_MAX_LUN;
2322                 dev_warn(&h->pdev->dev, "more luns configured"
2323                        " on controller than can be handled by"
2324                        " this driver.\n");
2325         }
2326
2327         if (num_luns == 0)
2328                 cciss_add_controller_node(h);
2329
2330         /* Compare controller drive array to driver's drive array
2331          * to see if any drives are missing on the controller due
2332          * to action of Array Config Utility (user deletes drive)
2333          * and deregister logical drives which have disappeared.
2334          */
2335         for (i = 0; i <= h->highest_lun; i++) {
2336                 int j;
2337                 drv_found = 0;
2338
2339                 /* skip holes in the array from already deleted drives */
2340                 if (h->drv[i] == NULL)
2341                         continue;
2342
2343                 for (j = 0; j < num_luns; j++) {
2344                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2345                         if (memcmp(h->drv[i]->LunID, lunid,
2346                                 sizeof(lunid)) == 0) {
2347                                 drv_found = 1;
2348                                 break;
2349                         }
2350                 }
2351                 if (!drv_found) {
2352                         /* Deregister it from the OS, it's gone. */
2353                         spin_lock_irqsave(&h->lock, flags);
2354                         h->drv[i]->busy_configuring = 1;
2355                         spin_unlock_irqrestore(&h->lock, flags);
2356                         return_code = deregister_disk(h, i, 1, via_ioctl);
2357                         if (h->drv[i] != NULL)
2358                                 h->drv[i]->busy_configuring = 0;
2359                 }
2360         }
2361
2362         /* Compare controller drive array to driver's drive array.
2363          * Check for updates in the drive information and any new drives
2364          * on the controller due to ACU adding logical drives, or changing
2365          * a logical drive's size, etc.  Reregister any new/changed drives
2366          */
2367         for (i = 0; i < num_luns; i++) {
2368                 int j;
2369
2370                 drv_found = 0;
2371
2372                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2373                 /* Find if the LUN is already in the drive array
2374                  * of the driver.  If so then update its info
2375                  * if not in use.  If it does not exist then find
2376                  * the first free index and add it.
2377                  */
2378                 for (j = 0; j <= h->highest_lun; j++) {
2379                         if (h->drv[j] != NULL &&
2380                                 memcmp(h->drv[j]->LunID, lunid,
2381                                         sizeof(h->drv[j]->LunID)) == 0) {
2382                                 drv_index = j;
2383                                 drv_found = 1;
2384                                 break;
2385                         }
2386                 }
2387
2388                 /* check if the drive was found already in the array */
2389                 if (!drv_found) {
2390                         drv_index = cciss_add_gendisk(h, lunid, 0);
2391                         if (drv_index == -1)
2392                                 goto freeret;
2393                 }
2394                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2395         }               /* end for */
2396
2397 freeret:
2398         kfree(ld_buff);
2399         h->busy_configuring = 0;
2400         /* We return -1 here to tell the ACU that we have registered/updated
2401          * all of the drives that we can and to keep it from calling us
2402          * additional times.
2403          */
2404         return -1;
2405 mem_msg:
2406         dev_err(&h->pdev->dev, "out of memory\n");
2407         h->busy_configuring = 0;
2408         goto freeret;
2409 }
2410
2411 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2412 {
2413         /* zero out the disk size info */
2414         drive_info->nr_blocks = 0;
2415         drive_info->block_size = 0;
2416         drive_info->heads = 0;
2417         drive_info->sectors = 0;
2418         drive_info->cylinders = 0;
2419         drive_info->raid_level = -1;
2420         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2421         memset(drive_info->model, 0, sizeof(drive_info->model));
2422         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2423         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2424         /*
2425          * don't clear the LUNID though, we need to remember which
2426          * one this one is.
2427          */
2428 }
2429
2430 /* This function will deregister the disk and it's queue from the
2431  * kernel.  It must be called with the controller lock held and the
2432  * drv structures busy_configuring flag set.  It's parameters are:
2433  *
2434  * disk = This is the disk to be deregistered
2435  * drv  = This is the drive_info_struct associated with the disk to be
2436  *        deregistered.  It contains information about the disk used
2437  *        by the driver.
2438  * clear_all = This flag determines whether or not the disk information
2439  *             is going to be completely cleared out and the highest_lun
2440  *             reset.  Sometimes we want to clear out information about
2441  *             the disk in preparation for re-adding it.  In this case
2442  *             the highest_lun should be left unchanged and the LunID
2443  *             should not be cleared.
2444  * via_ioctl
2445  *    This indicates whether we've reached this path via ioctl.
2446  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2447  *    If this path is reached via ioctl(), then the max_usage_count will
2448  *    be 1, as the process calling ioctl() has got to have the device open.
2449  *    If we get here via sysfs, then the max usage count will be zero.
2450 */
2451 static int deregister_disk(ctlr_info_t *h, int drv_index,
2452                            int clear_all, int via_ioctl)
2453 {
2454         int i;
2455         struct gendisk *disk;
2456         drive_info_struct *drv;
2457         int recalculate_highest_lun;
2458
2459         if (!capable(CAP_SYS_RAWIO))
2460                 return -EPERM;
2461
2462         drv = h->drv[drv_index];
2463         disk = h->gendisk[drv_index];
2464
2465         /* make sure logical volume is NOT is use */
2466         if (clear_all || (h->gendisk[0] == disk)) {
2467                 if (drv->usage_count > via_ioctl)
2468                         return -EBUSY;
2469         } else if (drv->usage_count > 0)
2470                 return -EBUSY;
2471
2472         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2473
2474         /* invalidate the devices and deregister the disk.  If it is disk
2475          * zero do not deregister it but just zero out it's values.  This
2476          * allows us to delete disk zero but keep the controller registered.
2477          */
2478         if (h->gendisk[0] != disk) {
2479                 struct request_queue *q = disk->queue;
2480                 if (disk->flags & GENHD_FL_UP) {
2481                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2482                         del_gendisk(disk);
2483                 }
2484                 if (q)
2485                         blk_cleanup_queue(q);
2486                 /* If clear_all is set then we are deleting the logical
2487                  * drive, not just refreshing its info.  For drives
2488                  * other than disk 0 we will call put_disk.  We do not
2489                  * do this for disk 0 as we need it to be able to
2490                  * configure the controller.
2491                  */
2492                 if (clear_all){
2493                         /* This isn't pretty, but we need to find the
2494                          * disk in our array and NULL our the pointer.
2495                          * This is so that we will call alloc_disk if
2496                          * this index is used again later.
2497                          */
2498                         for (i=0; i < CISS_MAX_LUN; i++){
2499                                 if (h->gendisk[i] == disk) {
2500                                         h->gendisk[i] = NULL;
2501                                         break;
2502                                 }
2503                         }
2504                         put_disk(disk);
2505                 }
2506         } else {
2507                 set_capacity(disk, 0);
2508                 cciss_clear_drive_info(drv);
2509         }
2510
2511         --h->num_luns;
2512
2513         /* if it was the last disk, find the new hightest lun */
2514         if (clear_all && recalculate_highest_lun) {
2515                 int newhighest = -1;
2516                 for (i = 0; i <= h->highest_lun; i++) {
2517                         /* if the disk has size > 0, it is available */
2518                         if (h->drv[i] && h->drv[i]->heads)
2519                                 newhighest = i;
2520                 }
2521                 h->highest_lun = newhighest;
2522         }
2523         return 0;
2524 }
2525
2526 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2527                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2528                 int cmd_type)
2529 {
2530         u64bit buff_dma_handle;
2531         int status = IO_OK;
2532
2533         c->cmd_type = CMD_IOCTL_PEND;
2534         c->Header.ReplyQueue = 0;
2535         if (buff != NULL) {
2536                 c->Header.SGList = 1;
2537                 c->Header.SGTotal = 1;
2538         } else {
2539                 c->Header.SGList = 0;
2540                 c->Header.SGTotal = 0;
2541         }
2542         c->Header.Tag.lower = c->busaddr;
2543         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2544
2545         c->Request.Type.Type = cmd_type;
2546         if (cmd_type == TYPE_CMD) {
2547                 switch (cmd) {
2548                 case CISS_INQUIRY:
2549                         /* are we trying to read a vital product page */
2550                         if (page_code != 0) {
2551                                 c->Request.CDB[1] = 0x01;
2552                                 c->Request.CDB[2] = page_code;
2553                         }
2554                         c->Request.CDBLen = 6;
2555                         c->Request.Type.Attribute = ATTR_SIMPLE;
2556                         c->Request.Type.Direction = XFER_READ;
2557                         c->Request.Timeout = 0;
2558                         c->Request.CDB[0] = CISS_INQUIRY;
2559                         c->Request.CDB[4] = size & 0xFF;
2560                         break;
2561                 case CISS_REPORT_LOG:
2562                 case CISS_REPORT_PHYS:
2563                         /* Talking to controller so It's a physical command
2564                            mode = 00 target = 0.  Nothing to write.
2565                          */
2566                         c->Request.CDBLen = 12;
2567                         c->Request.Type.Attribute = ATTR_SIMPLE;
2568                         c->Request.Type.Direction = XFER_READ;
2569                         c->Request.Timeout = 0;
2570                         c->Request.CDB[0] = cmd;
2571                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2572                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2573                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2574                         c->Request.CDB[9] = size & 0xFF;
2575                         break;
2576
2577                 case CCISS_READ_CAPACITY:
2578                         c->Request.CDBLen = 10;
2579                         c->Request.Type.Attribute = ATTR_SIMPLE;
2580                         c->Request.Type.Direction = XFER_READ;
2581                         c->Request.Timeout = 0;
2582                         c->Request.CDB[0] = cmd;
2583                         break;
2584                 case CCISS_READ_CAPACITY_16:
2585                         c->Request.CDBLen = 16;
2586                         c->Request.Type.Attribute = ATTR_SIMPLE;
2587                         c->Request.Type.Direction = XFER_READ;
2588                         c->Request.Timeout = 0;
2589                         c->Request.CDB[0] = cmd;
2590                         c->Request.CDB[1] = 0x10;
2591                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2592                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2593                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2594                         c->Request.CDB[13] = size & 0xFF;
2595                         c->Request.Timeout = 0;
2596                         c->Request.CDB[0] = cmd;
2597                         break;
2598                 case CCISS_CACHE_FLUSH:
2599                         c->Request.CDBLen = 12;
2600                         c->Request.Type.Attribute = ATTR_SIMPLE;
2601                         c->Request.Type.Direction = XFER_WRITE;
2602                         c->Request.Timeout = 0;
2603                         c->Request.CDB[0] = BMIC_WRITE;
2604                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2605                         c->Request.CDB[7] = (size >> 8) & 0xFF;
2606                         c->Request.CDB[8] = size & 0xFF;
2607                         break;
2608                 case TEST_UNIT_READY:
2609                         c->Request.CDBLen = 6;
2610                         c->Request.Type.Attribute = ATTR_SIMPLE;
2611                         c->Request.Type.Direction = XFER_NONE;
2612                         c->Request.Timeout = 0;
2613                         break;
2614                 default:
2615                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2616                         return IO_ERROR;
2617                 }
2618         } else if (cmd_type == TYPE_MSG) {
2619                 switch (cmd) {
2620                 case CCISS_ABORT_MSG:
2621                         c->Request.CDBLen = 12;
2622                         c->Request.Type.Attribute = ATTR_SIMPLE;
2623                         c->Request.Type.Direction = XFER_WRITE;
2624                         c->Request.Timeout = 0;
2625                         c->Request.CDB[0] = cmd;        /* abort */
2626                         c->Request.CDB[1] = 0;  /* abort a command */
2627                         /* buff contains the tag of the command to abort */
2628                         memcpy(&c->Request.CDB[4], buff, 8);
2629                         break;
2630                 case CCISS_RESET_MSG:
2631                         c->Request.CDBLen = 16;
2632                         c->Request.Type.Attribute = ATTR_SIMPLE;
2633                         c->Request.Type.Direction = XFER_NONE;
2634                         c->Request.Timeout = 0;
2635                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2636                         c->Request.CDB[0] = cmd;        /* reset */
2637                         c->Request.CDB[1] = CCISS_RESET_TYPE_TARGET;
2638                         break;
2639                 case CCISS_NOOP_MSG:
2640                         c->Request.CDBLen = 1;
2641                         c->Request.Type.Attribute = ATTR_SIMPLE;
2642                         c->Request.Type.Direction = XFER_WRITE;
2643                         c->Request.Timeout = 0;
2644                         c->Request.CDB[0] = cmd;
2645                         break;
2646                 default:
2647                         dev_warn(&h->pdev->dev,
2648                                 "unknown message type %d\n", cmd);
2649                         return IO_ERROR;
2650                 }
2651         } else {
2652                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2653                 return IO_ERROR;
2654         }
2655         /* Fill in the scatter gather information */
2656         if (size > 0) {
2657                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2658                                                              buff, size,
2659                                                              PCI_DMA_BIDIRECTIONAL);
2660                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2661                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2662                 c->SG[0].Len = size;
2663                 c->SG[0].Ext = 0;       /* we are not chaining */
2664         }
2665         return status;
2666 }
2667
2668 static int __devinit cciss_send_reset(ctlr_info_t *h, unsigned char *scsi3addr,
2669         u8 reset_type)
2670 {
2671         CommandList_struct *c;
2672         int return_status;
2673
2674         c = cmd_alloc(h);
2675         if (!c)
2676                 return -ENOMEM;
2677         return_status = fill_cmd(h, c, CCISS_RESET_MSG, NULL, 0, 0,
2678                 CTLR_LUNID, TYPE_MSG);
2679         c->Request.CDB[1] = reset_type; /* fill_cmd defaults to target reset */
2680         if (return_status != IO_OK) {
2681                 cmd_special_free(h, c);
2682                 return return_status;
2683         }
2684         c->waiting = NULL;
2685         enqueue_cmd_and_start_io(h, c);
2686         /* Don't wait for completion, the reset won't complete.  Don't free
2687          * the command either.  This is the last command we will send before
2688          * re-initializing everything, so it doesn't matter and won't leak.
2689          */
2690         return 0;
2691 }
2692
2693 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2694 {
2695         switch (c->err_info->ScsiStatus) {
2696         case SAM_STAT_GOOD:
2697                 return IO_OK;
2698         case SAM_STAT_CHECK_CONDITION:
2699                 switch (0xf & c->err_info->SenseInfo[2]) {
2700                 case 0: return IO_OK; /* no sense */
2701                 case 1: return IO_OK; /* recovered error */
2702                 default:
2703                         if (check_for_unit_attention(h, c))
2704                                 return IO_NEEDS_RETRY;
2705                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2706                                 "check condition, sense key = 0x%02x\n",
2707                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2708                 }
2709                 break;
2710         default:
2711                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2712                         "scsi status = 0x%02x\n",
2713                         c->Request.CDB[0], c->err_info->ScsiStatus);
2714                 break;
2715         }
2716         return IO_ERROR;
2717 }
2718
2719 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2720 {
2721         int return_status = IO_OK;
2722
2723         if (c->err_info->CommandStatus == CMD_SUCCESS)
2724                 return IO_OK;
2725
2726         switch (c->err_info->CommandStatus) {
2727         case CMD_TARGET_STATUS:
2728                 return_status = check_target_status(h, c);
2729                 break;
2730         case CMD_DATA_UNDERRUN:
2731         case CMD_DATA_OVERRUN:
2732                 /* expected for inquiry and report lun commands */
2733                 break;
2734         case CMD_INVALID:
2735                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2736                        "reported invalid\n", c->Request.CDB[0]);
2737                 return_status = IO_ERROR;
2738                 break;
2739         case CMD_PROTOCOL_ERR:
2740                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2741                        "protocol error\n", c->Request.CDB[0]);
2742                 return_status = IO_ERROR;
2743                 break;
2744         case CMD_HARDWARE_ERR:
2745                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2746                        " hardware error\n", c->Request.CDB[0]);
2747                 return_status = IO_ERROR;
2748                 break;
2749         case CMD_CONNECTION_LOST:
2750                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2751                        "connection lost\n", c->Request.CDB[0]);
2752                 return_status = IO_ERROR;
2753                 break;
2754         case CMD_ABORTED:
2755                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2756                        "aborted\n", c->Request.CDB[0]);
2757                 return_status = IO_ERROR;
2758                 break;
2759         case CMD_ABORT_FAILED:
2760                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2761                        "abort failed\n", c->Request.CDB[0]);
2762                 return_status = IO_ERROR;
2763                 break;
2764         case CMD_UNSOLICITED_ABORT:
2765                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2766                         c->Request.CDB[0]);
2767                 return_status = IO_NEEDS_RETRY;
2768                 break;
2769         case CMD_UNABORTABLE:
2770                 dev_warn(&h->pdev->dev, "cmd unabortable\n");
2771                 return_status = IO_ERROR;
2772                 break;
2773         default:
2774                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2775                        "unknown status %x\n", c->Request.CDB[0],
2776                        c->err_info->CommandStatus);
2777                 return_status = IO_ERROR;
2778         }
2779         return return_status;
2780 }
2781
2782 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2783         int attempt_retry)
2784 {
2785         DECLARE_COMPLETION_ONSTACK(wait);
2786         u64bit buff_dma_handle;
2787         int return_status = IO_OK;
2788
2789 resend_cmd2:
2790         c->waiting = &wait;
2791         enqueue_cmd_and_start_io(h, c);
2792
2793         wait_for_completion(&wait);
2794
2795         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2796                 goto command_done;
2797
2798         return_status = process_sendcmd_error(h, c);
2799
2800         if (return_status == IO_NEEDS_RETRY &&
2801                 c->retry_count < MAX_CMD_RETRIES) {
2802                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2803                         c->Request.CDB[0]);
2804                 c->retry_count++;
2805                 /* erase the old error information */
2806                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2807                 return_status = IO_OK;
2808                 INIT_COMPLETION(wait);
2809                 goto resend_cmd2;
2810         }
2811
2812 command_done:
2813         /* unlock the buffers from DMA */
2814         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2815         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2816         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2817                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2818         return return_status;
2819 }
2820
2821 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2822                            __u8 page_code, unsigned char scsi3addr[],
2823                         int cmd_type)
2824 {
2825         CommandList_struct *c;
2826         int return_status;
2827
2828         c = cmd_special_alloc(h);
2829         if (!c)
2830                 return -ENOMEM;
2831         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2832                 scsi3addr, cmd_type);
2833         if (return_status == IO_OK)
2834                 return_status = sendcmd_withirq_core(h, c, 1);
2835
2836         cmd_special_free(h, c);
2837         return return_status;
2838 }
2839
2840 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2841                                    sector_t total_size,
2842                                    unsigned int block_size,
2843                                    InquiryData_struct *inq_buff,
2844                                    drive_info_struct *drv)
2845 {
2846         int return_code;
2847         unsigned long t;
2848         unsigned char scsi3addr[8];
2849
2850         memset(inq_buff, 0, sizeof(InquiryData_struct));
2851         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2852         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2853                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2854         if (return_code == IO_OK) {
2855                 if (inq_buff->data_byte[8] == 0xFF) {
2856                         dev_warn(&h->pdev->dev,
2857                                "reading geometry failed, volume "
2858                                "does not support reading geometry\n");
2859                         drv->heads = 255;
2860                         drv->sectors = 32;      /* Sectors per track */
2861                         drv->cylinders = total_size + 1;
2862                         drv->raid_level = RAID_UNKNOWN;
2863                 } else {
2864                         drv->heads = inq_buff->data_byte[6];
2865                         drv->sectors = inq_buff->data_byte[7];
2866                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2867                         drv->cylinders += inq_buff->data_byte[5];
2868                         drv->raid_level = inq_buff->data_byte[8];
2869                 }
2870                 drv->block_size = block_size;
2871                 drv->nr_blocks = total_size + 1;
2872                 t = drv->heads * drv->sectors;
2873                 if (t > 1) {
2874                         sector_t real_size = total_size + 1;
2875                         unsigned long rem = sector_div(real_size, t);
2876                         if (rem)
2877                                 real_size++;
2878                         drv->cylinders = real_size;
2879                 }
2880         } else {                /* Get geometry failed */
2881                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2882         }
2883 }
2884
2885 static void
2886 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2887                     unsigned int *block_size)
2888 {
2889         ReadCapdata_struct *buf;
2890         int return_code;
2891         unsigned char scsi3addr[8];
2892
2893         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2894         if (!buf) {
2895                 dev_warn(&h->pdev->dev, "out of memory\n");
2896                 return;
2897         }
2898
2899         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2900         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2901                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2902         if (return_code == IO_OK) {
2903                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2904                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2905         } else {                /* read capacity command failed */
2906                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2907                 *total_size = 0;
2908                 *block_size = BLOCK_SIZE;
2909         }
2910         kfree(buf);
2911 }
2912
2913 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2914         sector_t *total_size, unsigned int *block_size)
2915 {
2916         ReadCapdata_struct_16 *buf;
2917         int return_code;
2918         unsigned char scsi3addr[8];
2919
2920         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2921         if (!buf) {
2922                 dev_warn(&h->pdev->dev, "out of memory\n");
2923                 return;
2924         }
2925
2926         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2927         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2928                 buf, sizeof(ReadCapdata_struct_16),
2929                         0, scsi3addr, TYPE_CMD);
2930         if (return_code == IO_OK) {
2931                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2932                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2933         } else {                /* read capacity command failed */
2934                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2935                 *total_size = 0;
2936                 *block_size = BLOCK_SIZE;
2937         }
2938         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2939                (unsigned long long)*total_size+1, *block_size);
2940         kfree(buf);
2941 }
2942
2943 static int cciss_revalidate(struct gendisk *disk)
2944 {
2945         ctlr_info_t *h = get_host(disk);
2946         drive_info_struct *drv = get_drv(disk);
2947         int logvol;
2948         int FOUND = 0;
2949         unsigned int block_size;
2950         sector_t total_size;
2951         InquiryData_struct *inq_buff = NULL;
2952
2953         for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2954                 if (!h->drv[logvol])
2955                         continue;
2956                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2957                         sizeof(drv->LunID)) == 0) {
2958                         FOUND = 1;
2959                         break;
2960                 }
2961         }
2962
2963         if (!FOUND)
2964                 return 1;
2965
2966         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2967         if (inq_buff == NULL) {
2968                 dev_warn(&h->pdev->dev, "out of memory\n");
2969                 return 1;
2970         }
2971         if (h->cciss_read == CCISS_READ_10) {
2972                 cciss_read_capacity(h, logvol,
2973                                         &total_size, &block_size);
2974         } else {
2975                 cciss_read_capacity_16(h, logvol,
2976                                         &total_size, &block_size);
2977         }
2978         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2979                                inq_buff, drv);
2980
2981         blk_queue_logical_block_size(drv->queue, drv->block_size);
2982         set_capacity(disk, drv->nr_blocks);
2983
2984         kfree(inq_buff);
2985         return 0;
2986 }
2987
2988 /*
2989  * Map (physical) PCI mem into (virtual) kernel space
2990  */
2991 static void __iomem *remap_pci_mem(ulong base, ulong size)
2992 {
2993         ulong page_base = ((ulong) base) & PAGE_MASK;
2994         ulong page_offs = ((ulong) base) - page_base;
2995         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2996
2997         return page_remapped ? (page_remapped + page_offs) : NULL;
2998 }
2999
3000 /*
3001  * Takes jobs of the Q and sends them to the hardware, then puts it on
3002  * the Q to wait for completion.
3003  */
3004 static void start_io(ctlr_info_t *h)
3005 {
3006         CommandList_struct *c;
3007
3008         while (!list_empty(&h->reqQ)) {
3009                 c = list_entry(h->reqQ.next, CommandList_struct, list);
3010                 /* can't do anything if fifo is full */
3011                 if ((h->access.fifo_full(h))) {
3012                         dev_warn(&h->pdev->dev, "fifo full\n");
3013                         break;
3014                 }
3015
3016                 /* Get the first entry from the Request Q */
3017                 removeQ(c);
3018                 h->Qdepth--;
3019
3020                 /* Tell the controller execute command */
3021                 h->access.submit_command(h, c);
3022
3023                 /* Put job onto the completed Q */
3024                 addQ(&h->cmpQ, c);
3025         }
3026 }
3027
3028 /* Assumes that h->lock is held. */
3029 /* Zeros out the error record and then resends the command back */
3030 /* to the controller */
3031 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
3032 {
3033         /* erase the old error information */
3034         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
3035
3036         /* add it to software queue and then send it to the controller */
3037         addQ(&h->reqQ, c);
3038         h->Qdepth++;
3039         if (h->Qdepth > h->maxQsinceinit)
3040                 h->maxQsinceinit = h->Qdepth;
3041
3042         start_io(h);
3043 }
3044
3045 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
3046         unsigned int msg_byte, unsigned int host_byte,
3047         unsigned int driver_byte)
3048 {
3049         /* inverse of macros in scsi.h */
3050         return (scsi_status_byte & 0xff) |
3051                 ((msg_byte & 0xff) << 8) |
3052                 ((host_byte & 0xff) << 16) |
3053                 ((driver_byte & 0xff) << 24);
3054 }
3055
3056 static inline int evaluate_target_status(ctlr_info_t *h,
3057                         CommandList_struct *cmd, int *retry_cmd)
3058 {
3059         unsigned char sense_key;
3060         unsigned char status_byte, msg_byte, host_byte, driver_byte;
3061         int error_value;
3062
3063         *retry_cmd = 0;
3064         /* If we get in here, it means we got "target status", that is, scsi status */
3065         status_byte = cmd->err_info->ScsiStatus;
3066         driver_byte = DRIVER_OK;
3067         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3068
3069         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3070                 host_byte = DID_PASSTHROUGH;
3071         else
3072                 host_byte = DID_OK;
3073
3074         error_value = make_status_bytes(status_byte, msg_byte,
3075                 host_byte, driver_byte);
3076
3077         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3078                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3079                         dev_warn(&h->pdev->dev, "cmd %p "
3080                                "has SCSI Status 0x%x\n",
3081                                cmd, cmd->err_info->ScsiStatus);
3082                 return error_value;
3083         }
3084
3085         /* check the sense key */
3086         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3087         /* no status or recovered error */
3088         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3089             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3090                 error_value = 0;
3091
3092         if (check_for_unit_attention(h, cmd)) {
3093                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3094                 return 0;
3095         }
3096
3097         /* Not SG_IO or similar? */
3098         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3099                 if (error_value != 0)
3100                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3101                                " sense key = 0x%x\n", cmd, sense_key);
3102                 return error_value;
3103         }
3104
3105         /* SG_IO or similar, copy sense data back */
3106         if (cmd->rq->sense) {
3107                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3108                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3109                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3110                         cmd->rq->sense_len);
3111         } else
3112                 cmd->rq->sense_len = 0;
3113
3114         return error_value;
3115 }
3116
3117 /* checks the status of the job and calls complete buffers to mark all
3118  * buffers for the completed job. Note that this function does not need
3119  * to hold the hba/queue lock.
3120  */
3121 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3122                                     int timeout)
3123 {
3124         int retry_cmd = 0;
3125         struct request *rq = cmd->rq;
3126
3127         rq->errors = 0;
3128
3129         if (timeout)
3130                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3131
3132         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3133                 goto after_error_processing;
3134
3135         switch (cmd->err_info->CommandStatus) {
3136         case CMD_TARGET_STATUS:
3137                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3138                 break;
3139         case CMD_DATA_UNDERRUN:
3140                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3141                         dev_warn(&h->pdev->dev, "cmd %p has"
3142                                " completed with data underrun "
3143                                "reported\n", cmd);
3144                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3145                 }
3146                 break;
3147         case CMD_DATA_OVERRUN:
3148                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3149                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3150                                " completed with data overrun "
3151                                "reported\n", cmd);
3152                 break;
3153         case CMD_INVALID:
3154                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3155                        "reported invalid\n", cmd);
3156                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3157                         cmd->err_info->CommandStatus, DRIVER_OK,
3158                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3159                                 DID_PASSTHROUGH : DID_ERROR);
3160                 break;
3161         case CMD_PROTOCOL_ERR:
3162                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3163                        "protocol error\n", cmd);
3164                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3165                         cmd->err_info->CommandStatus, DRIVER_OK,
3166                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3167                                 DID_PASSTHROUGH : DID_ERROR);
3168                 break;
3169         case CMD_HARDWARE_ERR:
3170                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3171                        " hardware error\n", cmd);
3172                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3173                         cmd->err_info->CommandStatus, DRIVER_OK,
3174                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3175                                 DID_PASSTHROUGH : DID_ERROR);
3176                 break;
3177         case CMD_CONNECTION_LOST:
3178                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3179                        "connection lost\n", cmd);
3180                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3181                         cmd->err_info->CommandStatus, DRIVER_OK,
3182                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3183                                 DID_PASSTHROUGH : DID_ERROR);
3184                 break;
3185         case CMD_ABORTED:
3186                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3187                        "aborted\n", cmd);
3188                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3189                         cmd->err_info->CommandStatus, DRIVER_OK,
3190                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3191                                 DID_PASSTHROUGH : DID_ABORT);
3192                 break;
3193         case CMD_ABORT_FAILED:
3194                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3195                        "abort failed\n", cmd);
3196                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3197                         cmd->err_info->CommandStatus, DRIVER_OK,
3198                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3199                                 DID_PASSTHROUGH : DID_ERROR);
3200                 break;
3201         case CMD_UNSOLICITED_ABORT:
3202                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3203                        "abort %p\n", h->ctlr, cmd);
3204                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3205                         retry_cmd = 1;
3206                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3207                         cmd->retry_count++;
3208                 } else
3209                         dev_warn(&h->pdev->dev,
3210                                 "%p retried too many times\n", cmd);
3211                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3212                         cmd->err_info->CommandStatus, DRIVER_OK,
3213                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3214                                 DID_PASSTHROUGH : DID_ABORT);
3215                 break;
3216         case CMD_TIMEOUT:
3217                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3218                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3219                         cmd->err_info->CommandStatus, DRIVER_OK,
3220                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3221                                 DID_PASSTHROUGH : DID_ERROR);
3222                 break;
3223         case CMD_UNABORTABLE:
3224                 dev_warn(&h->pdev->dev, "cmd %p unabortable\n", cmd);
3225                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3226                         cmd->err_info->CommandStatus, DRIVER_OK,
3227                         cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC ?
3228                                 DID_PASSTHROUGH : DID_ERROR);
3229                 break;
3230         default:
3231                 dev_warn(&h->pdev->dev, "cmd %p returned "
3232                        "unknown status %x\n", cmd,
3233                        cmd->err_info->CommandStatus);
3234                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3235                         cmd->err_info->CommandStatus, DRIVER_OK,
3236                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3237                                 DID_PASSTHROUGH : DID_ERROR);
3238         }
3239
3240 after_error_processing:
3241
3242         /* We need to return this command */
3243         if (retry_cmd) {
3244                 resend_cciss_cmd(h, cmd);
3245                 return;
3246         }
3247         cmd->rq->completion_data = cmd;
3248         blk_complete_request(cmd->rq);
3249 }
3250
3251 static inline u32 cciss_tag_contains_index(u32 tag)
3252 {
3253 #define DIRECT_LOOKUP_BIT 0x10
3254         return tag & DIRECT_LOOKUP_BIT;
3255 }
3256
3257 static inline u32 cciss_tag_to_index(u32 tag)
3258 {
3259 #define DIRECT_LOOKUP_SHIFT 5
3260         return tag >> DIRECT_LOOKUP_SHIFT;
3261 }
3262
3263 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag)
3264 {
3265 #define CCISS_PERF_ERROR_BITS ((1 << DIRECT_LOOKUP_SHIFT) - 1)
3266 #define CCISS_SIMPLE_ERROR_BITS 0x03
3267         if (likely(h->transMethod & CFGTBL_Trans_Performant))
3268                 return tag & ~CCISS_PERF_ERROR_BITS;
3269         return tag & ~CCISS_SIMPLE_ERROR_BITS;
3270 }
3271
3272 static inline void cciss_mark_tag_indexed(u32 *tag)
3273 {
3274         *tag |= DIRECT_LOOKUP_BIT;
3275 }
3276
3277 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3278 {
3279         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3280 }
3281
3282 /*
3283  * Get a request and submit it to the controller.
3284  */
3285 static void do_cciss_request(struct request_queue *q)
3286 {
3287         ctlr_info_t *h = q->queuedata;
3288         CommandList_struct *c;
3289         sector_t start_blk;
3290         int seg;
3291         struct request *creq;
3292         u64bit temp64;
3293         struct scatterlist *tmp_sg;
3294         SGDescriptor_struct *curr_sg;
3295         drive_info_struct *drv;
3296         int i, dir;
3297         int sg_index = 0;
3298         int chained = 0;
3299
3300       queue:
3301         creq = blk_peek_request(q);
3302         if (!creq)
3303                 goto startio;
3304
3305         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3306
3307         c = cmd_alloc(h);
3308         if (!c)
3309                 goto full;
3310
3311         blk_start_request(creq);
3312
3313         tmp_sg = h->scatter_list[c->cmdindex];
3314         spin_unlock_irq(q->queue_lock);
3315
3316         c->cmd_type = CMD_RWREQ;
3317         c->rq = creq;
3318
3319         /* fill in the request */
3320         drv = creq->rq_disk->private_data;
3321         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3322         /* got command from pool, so use the command block index instead */
3323         /* for direct lookups. */
3324         /* The first 2 bits are reserved for controller error reporting. */
3325         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3326         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3327         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3328         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3329         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3330         c->Request.Type.Attribute = ATTR_SIMPLE;
3331         c->Request.Type.Direction =
3332             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3333         c->Request.Timeout = 0; /* Don't time out */
3334         c->Request.CDB[0] =
3335             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3336         start_blk = blk_rq_pos(creq);
3337         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3338                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3339         sg_init_table(tmp_sg, h->maxsgentries);
3340         seg = blk_rq_map_sg(q, creq, tmp_sg);
3341
3342         /* get the DMA records for the setup */
3343         if (c->Request.Type.Direction == XFER_READ)
3344                 dir = PCI_DMA_FROMDEVICE;
3345         else
3346                 dir = PCI_DMA_TODEVICE;
3347
3348         curr_sg = c->SG;
3349         sg_index = 0;
3350         chained = 0;
3351
3352         for (i = 0; i < seg; i++) {
3353                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3354                         !chained && ((seg - i) > 1)) {
3355                         /* Point to next chain block. */
3356                         curr_sg = h->cmd_sg_list[c->cmdindex];
3357                         sg_index = 0;
3358                         chained = 1;
3359                 }
3360                 curr_sg[sg_index].Len = tmp_sg[i].length;
3361                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3362                                                 tmp_sg[i].offset,
3363                                                 tmp_sg[i].length, dir);
3364                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3365                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3366                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3367                 ++sg_index;
3368         }
3369         if (chained)
3370                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3371                         (seg - (h->max_cmd_sgentries - 1)) *
3372                                 sizeof(SGDescriptor_struct));
3373
3374         /* track how many SG entries we are using */
3375         if (seg > h->maxSG)
3376                 h->maxSG = seg;
3377
3378         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3379                         "chained[%d]\n",
3380                         blk_rq_sectors(creq), seg, chained);
3381
3382         c->Header.SGTotal = seg + chained;
3383         if (seg <= h->max_cmd_sgentries)
3384                 c->Header.SGList = c->Header.SGTotal;
3385         else
3386                 c->Header.SGList = h->max_cmd_sgentries;
3387         set_performant_mode(h, c);
3388
3389         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3390                 if(h->cciss_read == CCISS_READ_10) {
3391                         c->Request.CDB[1] = 0;
3392                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3393                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3394                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3395                         c->Request.CDB[5] = start_blk & 0xff;
3396                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3397                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3398                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3399                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3400                 } else {
3401                         u32 upper32 = upper_32_bits(start_blk);
3402
3403                         c->Request.CDBLen = 16;
3404                         c->Request.CDB[1]= 0;
3405                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3406                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3407                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3408                         c->Request.CDB[5]= upper32 & 0xff;
3409                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3410                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3411                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3412                         c->Request.CDB[9]= start_blk & 0xff;
3413                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3414                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3415                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3416                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3417                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3418                 }
3419         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3420                 c->Request.CDBLen = creq->cmd_len;
3421                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3422         } else {
3423                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3424                         creq->cmd_type);
3425                 BUG();
3426         }
3427
3428         spin_lock_irq(q->queue_lock);
3429
3430         addQ(&h->reqQ, c);
3431         h->Qdepth++;
3432         if (h->Qdepth > h->maxQsinceinit)
3433                 h->maxQsinceinit = h->Qdepth;
3434
3435         goto queue;
3436 full:
3437         blk_stop_queue(q);
3438 startio:
3439         /* We will already have the driver lock here so not need
3440          * to lock it.
3441          */
3442         start_io(h);
3443 }
3444
3445 static inline unsigned long get_next_completion(ctlr_info_t *h)
3446 {
3447         return h->access.command_completed(h);
3448 }
3449
3450 static inline int interrupt_pending(ctlr_info_t *h)
3451 {
3452         return h->access.intr_pending(h);
3453 }
3454
3455 static inline long interrupt_not_for_us(ctlr_info_t *h)
3456 {
3457         return ((h->access.intr_pending(h) == 0) ||
3458                 (h->interrupts_enabled == 0));
3459 }
3460
3461 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3462                         u32 raw_tag)
3463 {
3464         if (unlikely(tag_index >= h->nr_cmds)) {
3465                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3466                 return 1;
3467         }
3468         return 0;
3469 }
3470
3471 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3472                                 u32 raw_tag)
3473 {
3474         removeQ(c);
3475         if (likely(c->cmd_type == CMD_RWREQ))
3476                 complete_command(h, c, 0);
3477         else if (c->cmd_type == CMD_IOCTL_PEND)
3478                 complete(c->waiting);
3479 #ifdef CONFIG_CISS_SCSI_TAPE
3480         else if (c->cmd_type == CMD_SCSI)
3481                 complete_scsi_command(c, 0, raw_tag);
3482 #endif
3483 }
3484
3485 static inline u32 next_command(ctlr_info_t *h)
3486 {
3487         u32 a;
3488
3489         if (unlikely(!(h->transMethod & CFGTBL_Trans_Performant)))
3490                 return h->access.command_completed(h);
3491
3492         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3493                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3494                 (h->reply_pool_head)++;
3495                 h->commands_outstanding--;
3496         } else {
3497                 a = FIFO_EMPTY;
3498         }
3499         /* Check for wraparound */
3500         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3501                 h->reply_pool_head = h->reply_pool;
3502                 h->reply_pool_wraparound ^= 1;
3503         }
3504         return a;
3505 }
3506
3507 /* process completion of an indexed ("direct lookup") command */
3508 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3509 {
3510         u32 tag_index;
3511         CommandList_struct *c;
3512
3513         tag_index = cciss_tag_to_index(raw_tag);
3514         if (bad_tag(h, tag_index, raw_tag))
3515                 return next_command(h);
3516         c = h->cmd_pool + tag_index;
3517         finish_cmd(h, c, raw_tag);
3518         return next_command(h);
3519 }
3520
3521 /* process completion of a non-indexed command */
3522 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3523 {
3524         CommandList_struct *c = NULL;
3525         __u32 busaddr_masked, tag_masked;
3526
3527         tag_masked = cciss_tag_discard_error_bits(h, raw_tag);
3528         list_for_each_entry(c, &h->cmpQ, list) {
3529                 busaddr_masked = cciss_tag_discard_error_bits(h, c->busaddr);
3530                 if (busaddr_masked == tag_masked) {
3531                         finish_cmd(h, c, raw_tag);
3532                         return next_command(h);
3533                 }
3534         }
3535         bad_tag(h, h->nr_cmds + 1, raw_tag);
3536         return next_command(h);
3537 }
3538
3539 /* Some controllers, like p400, will give us one interrupt
3540  * after a soft reset, even if we turned interrupts off.
3541  * Only need to check for this in the cciss_xxx_discard_completions
3542  * functions.
3543  */
3544 static int ignore_bogus_interrupt(ctlr_info_t *h)
3545 {
3546         if (likely(!reset_devices))
3547                 return 0;
3548
3549         if (likely(h->interrupts_enabled))
3550                 return 0;
3551
3552         dev_info(&h->pdev->dev, "Received interrupt while interrupts disabled "
3553                 "(known firmware bug.)  Ignoring.\n");
3554
3555         return 1;
3556 }
3557
3558 static irqreturn_t cciss_intx_discard_completions(int irq, void *dev_id)
3559 {
3560         ctlr_info_t *h = dev_id;
3561         unsigned long flags;
3562         u32 raw_tag;
3563
3564         if (ignore_bogus_interrupt(h))
3565                 return IRQ_NONE;
3566
3567         if (interrupt_not_for_us(h))
3568                 return IRQ_NONE;
3569         spin_lock_irqsave(&h->lock, flags);
3570         while (interrupt_pending(h)) {
3571                 raw_tag = get_next_completion(h);
3572                 while (raw_tag != FIFO_EMPTY)
3573                         raw_tag = next_command(h);
3574         }
3575         spin_unlock_irqrestore(&h->lock, flags);
3576         return IRQ_HANDLED;
3577 }
3578
3579 static irqreturn_t cciss_msix_discard_completions(int irq, void *dev_id)
3580 {
3581         ctlr_info_t *h = dev_id;
3582         unsigned long flags;
3583         u32 raw_tag;
3584
3585         if (ignore_bogus_interrupt(h))
3586                 return IRQ_NONE;
3587
3588         spin_lock_irqsave(&h->lock, flags);
3589         raw_tag = get_next_completion(h);
3590         while (raw_tag != FIFO_EMPTY)
3591                 raw_tag = next_command(h);
3592         spin_unlock_irqrestore(&h->lock, flags);
3593         return IRQ_HANDLED;
3594 }
3595
3596 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3597 {
3598         ctlr_info_t *h = dev_id;
3599         unsigned long flags;
3600         u32 raw_tag;
3601
3602         if (interrupt_not_for_us(h))
3603                 return IRQ_NONE;
3604         spin_lock_irqsave(&h->lock, flags);
3605         while (interrupt_pending(h)) {
3606                 raw_tag = get_next_completion(h);
3607                 while (raw_tag != FIFO_EMPTY) {
3608                         if (cciss_tag_contains_index(raw_tag))
3609                                 raw_tag = process_indexed_cmd(h, raw_tag);
3610                         else
3611                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3612                 }
3613         }
3614         spin_unlock_irqrestore(&h->lock, flags);
3615         return IRQ_HANDLED;
3616 }
3617
3618 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3619  * check the interrupt pending register because it is not set.
3620  */
3621 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3622 {
3623         ctlr_info_t *h = dev_id;
3624         unsigned long flags;
3625         u32 raw_tag;
3626
3627         spin_lock_irqsave(&h->lock, flags);
3628         raw_tag = get_next_completion(h);
3629         while (raw_tag != FIFO_EMPTY) {
3630                 if (cciss_tag_contains_index(raw_tag))
3631                         raw_tag = process_indexed_cmd(h, raw_tag);
3632                 else
3633                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3634         }
3635         spin_unlock_irqrestore(&h->lock, flags);
3636         return IRQ_HANDLED;
3637 }
3638
3639 /**
3640  * add_to_scan_list() - add controller to rescan queue
3641  * @h:                Pointer to the controller.
3642  *
3643  * Adds the controller to the rescan queue if not already on the queue.
3644  *
3645  * returns 1 if added to the queue, 0 if skipped (could be on the
3646  * queue already, or the controller could be initializing or shutting
3647  * down).
3648  **/
3649 static int add_to_scan_list(struct ctlr_info *h)
3650 {
3651         struct ctlr_info *test_h;
3652         int found = 0;
3653         int ret = 0;
3654
3655         if (h->busy_initializing)
3656                 return 0;
3657
3658         if (!mutex_trylock(&h->busy_shutting_down))
3659                 return 0;
3660
3661         mutex_lock(&scan_mutex);
3662         list_for_each_entry(test_h, &scan_q, scan_list) {
3663                 if (test_h == h) {
3664                         found = 1;
3665                         break;
3666                 }
3667         }
3668         if (!found && !h->busy_scanning) {
3669                 INIT_COMPLETION(h->scan_wait);
3670                 list_add_tail(&h->scan_list, &scan_q);
3671                 ret = 1;
3672         }
3673         mutex_unlock(&scan_mutex);
3674         mutex_unlock(&h->busy_shutting_down);
3675
3676         return ret;
3677 }
3678
3679 /**
3680  * remove_from_scan_list() - remove controller from rescan queue
3681  * @h:                     Pointer to the controller.
3682  *
3683  * Removes the controller from the rescan queue if present. Blocks if
3684  * the controller is currently conducting a rescan.  The controller
3685  * can be in one of three states:
3686  * 1. Doesn't need a scan
3687  * 2. On the scan list, but not scanning yet (we remove it)
3688  * 3. Busy scanning (and not on the list). In this case we want to wait for
3689  *    the scan to complete to make sure the scanning thread for this
3690  *    controller is completely idle.
3691  **/
3692 static void remove_from_scan_list(struct ctlr_info *h)
3693 {
3694         struct ctlr_info *test_h, *tmp_h;
3695
3696         mutex_lock(&scan_mutex);
3697         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3698                 if (test_h == h) { /* state 2. */
3699                         list_del(&h->scan_list);
3700                         complete_all(&h->scan_wait);
3701                         mutex_unlock(&scan_mutex);
3702                         return;
3703                 }
3704         }
3705         if (h->busy_scanning) { /* state 3. */
3706                 mutex_unlock(&scan_mutex);
3707                 wait_for_completion(&h->scan_wait);
3708         } else { /* state 1, nothing to do. */
3709                 mutex_unlock(&scan_mutex);
3710         }
3711 }
3712
3713 /**
3714  * scan_thread() - kernel thread used to rescan controllers
3715  * @data:        Ignored.
3716  *
3717  * A kernel thread used scan for drive topology changes on
3718  * controllers. The thread processes only one controller at a time
3719  * using a queue.  Controllers are added to the queue using
3720  * add_to_scan_list() and removed from the queue either after done
3721  * processing or using remove_from_scan_list().
3722  *
3723  * returns 0.
3724  **/
3725 static int scan_thread(void *data)
3726 {
3727         struct ctlr_info *h;
3728
3729         while (1) {
3730                 set_current_state(TASK_INTERRUPTIBLE);
3731                 schedule();
3732                 if (kthread_should_stop())
3733                         break;
3734
3735                 while (1) {
3736                         mutex_lock(&scan_mutex);
3737                         if (list_empty(&scan_q)) {
3738                                 mutex_unlock(&scan_mutex);
3739                                 break;
3740                         }
3741
3742                         h = list_entry(scan_q.next,
3743                                        struct ctlr_info,
3744                                        scan_list);
3745                         list_del(&h->scan_list);
3746                         h->busy_scanning = 1;
3747                         mutex_unlock(&scan_mutex);
3748
3749                         rebuild_lun_table(h, 0, 0);
3750                         complete_all(&h->scan_wait);
3751                         mutex_lock(&scan_mutex);
3752                         h->busy_scanning = 0;
3753                         mutex_unlock(&scan_mutex);
3754                 }
3755         }
3756
3757         return 0;
3758 }
3759
3760 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3761 {
3762         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3763                 return 0;
3764
3765         switch (c->err_info->SenseInfo[12]) {
3766         case STATE_CHANGED:
3767                 dev_warn(&h->pdev->dev, "a state change "
3768                         "detected, command retried\n");
3769                 return 1;
3770         break;
3771         case LUN_FAILED:
3772                 dev_warn(&h->pdev->dev, "LUN failure "
3773                         "detected, action required\n");
3774                 return 1;
3775         break;
3776         case REPORT_LUNS_CHANGED:
3777                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3778         /*
3779          * Here, we could call add_to_scan_list and wake up the scan thread,
3780          * except that it's quite likely that we will get more than one
3781          * REPORT_LUNS_CHANGED condition in quick succession, which means
3782          * that those which occur after the first one will likely happen
3783          * *during* the scan_thread's rescan.  And the rescan code is not
3784          * robust enough to restart in the middle, undoing what it has already
3785          * done, and it's not clear that it's even possible to do this, since
3786          * part of what it does is notify the block layer, which starts
3787          * doing it's own i/o to read partition tables and so on, and the
3788          * driver doesn't have visibility to know what might need undoing.
3789          * In any event, if possible, it is horribly complicated to get right
3790          * so we just don't do it for now.
3791          *
3792          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3793          */
3794                 return 1;
3795         break;
3796         case POWER_OR_RESET:
3797                 dev_warn(&h->pdev->dev,
3798                         "a power on or device reset detected\n");
3799                 return 1;
3800         break;
3801         case UNIT_ATTENTION_CLEARED:
3802                 dev_warn(&h->pdev->dev,
3803                         "unit attention cleared by another initiator\n");
3804                 return 1;
3805         break;
3806         default:
3807                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3808                 return 1;
3809         }
3810 }
3811
3812 /*
3813  *  We cannot read the structure directly, for portability we must use
3814  *   the io functions.
3815  *   This is for debug only.
3816  */
3817 static void print_cfg_table(ctlr_info_t *h)
3818 {
3819         int i;
3820         char temp_name[17];
3821         CfgTable_struct *tb = h->cfgtable;
3822
3823         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3824         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3825         for (i = 0; i < 4; i++)
3826                 temp_name[i] = readb(&(tb->Signature[i]));
3827         temp_name[4] = '\0';
3828         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3829         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3830                 readl(&(tb->SpecValence)));
3831         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3832                readl(&(tb->TransportSupport)));
3833         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3834                readl(&(tb->TransportActive)));
3835         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3836                readl(&(tb->HostWrite.TransportRequest)));
3837         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3838                readl(&(tb->HostWrite.CoalIntDelay)));
3839         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3840                readl(&(tb->HostWrite.CoalIntCount)));
3841         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3842                readl(&(tb->CmdsOutMax)));
3843         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3844                 readl(&(tb->BusTypes)));
3845         for (i = 0; i < 16; i++)
3846                 temp_name[i] = readb(&(tb->ServerName[i]));
3847         temp_name[16] = '\0';
3848         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3849         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3850                 readl(&(tb->HeartBeat)));
3851 }
3852
3853 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3854 {
3855         int i, offset, mem_type, bar_type;
3856         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3857                 return 0;
3858         offset = 0;
3859         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3860                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3861                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3862                         offset += 4;
3863                 else {
3864                         mem_type = pci_resource_flags(pdev, i) &
3865                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3866                         switch (mem_type) {
3867                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3868                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3869                                 offset += 4;    /* 32 bit */
3870                                 break;
3871                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3872                                 offset += 8;
3873                                 break;
3874                         default:        /* reserved in PCI 2.2 */
3875                                 dev_warn(&pdev->dev,
3876                                        "Base address is invalid\n");
3877                                 return -1;
3878                                 break;
3879                         }
3880                 }
3881                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3882                         return i + 1;
3883         }
3884         return -1;
3885 }
3886
3887 /* Fill in bucket_map[], given nsgs (the max number of
3888  * scatter gather elements supported) and bucket[],
3889  * which is an array of 8 integers.  The bucket[] array
3890  * contains 8 different DMA transfer sizes (in 16
3891  * byte increments) which the controller uses to fetch
3892  * commands.  This function fills in bucket_map[], which
3893  * maps a given number of scatter gather elements to one of
3894  * the 8 DMA transfer sizes.  The point of it is to allow the
3895  * controller to only do as much DMA as needed to fetch the
3896  * command, with the DMA transfer size encoded in the lower
3897  * bits of the command address.
3898  */
3899 static void  calc_bucket_map(int bucket[], int num_buckets,
3900         int nsgs, int *bucket_map)
3901 {
3902         int i, j, b, size;
3903
3904         /* even a command with 0 SGs requires 4 blocks */
3905 #define MINIMUM_TRANSFER_BLOCKS 4
3906 #define NUM_BUCKETS 8
3907         /* Note, bucket_map must have nsgs+1 entries. */
3908         for (i = 0; i <= nsgs; i++) {
3909                 /* Compute size of a command with i SG entries */
3910                 size = i + MINIMUM_TRANSFER_BLOCKS;
3911                 b = num_buckets; /* Assume the biggest bucket */
3912                 /* Find the bucket that is just big enough */
3913                 for (j = 0; j < 8; j++) {
3914                         if (bucket[j] >= size) {
3915                                 b = j;
3916                                 break;
3917                         }
3918                 }
3919                 /* for a command with i SG entries, use bucket b. */
3920                 bucket_map[i] = b;
3921         }
3922 }
3923
3924 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3925 {
3926         int i;
3927
3928         /* under certain very rare conditions, this can take awhile.
3929          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3930          * as we enter this code.) */
3931         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3932                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3933                         break;
3934                 usleep_range(10000, 20000);
3935         }
3936 }
3937
3938 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h,
3939         u32 use_short_tags)
3940 {
3941         /* This is a bit complicated.  There are 8 registers on
3942          * the controller which we write to to tell it 8 different
3943          * sizes of commands which there may be.  It's a way of
3944          * reducing the DMA done to fetch each command.  Encoded into
3945          * each command's tag are 3 bits which communicate to the controller
3946          * which of the eight sizes that command fits within.  The size of
3947          * each command depends on how many scatter gather entries there are.
3948          * Each SG entry requires 16 bytes.  The eight registers are programmed
3949          * with the number of 16-byte blocks a command of that size requires.
3950          * The smallest command possible requires 5 such 16 byte blocks.
3951          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3952          * blocks.  Note, this only extends to the SG entries contained
3953          * within the command block, and does not extend to chained blocks
3954          * of SG elements.   bft[] contains the eight values we write to
3955          * the registers.  They are not evenly distributed, but have more
3956          * sizes for small commands, and fewer sizes for larger commands.
3957          */
3958         __u32 trans_offset;
3959         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3960                         /*
3961                          *  5 = 1 s/g entry or 4k
3962                          *  6 = 2 s/g entry or 8k
3963                          *  8 = 4 s/g entry or 16k
3964                          * 10 = 6 s/g entry or 24k
3965                          */
3966         unsigned long register_value;
3967         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3968
3969         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3970
3971         /* Controller spec: zero out this buffer. */
3972         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3973         h->reply_pool_head = h->reply_pool;
3974
3975         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3976         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3977                                 h->blockFetchTable);
3978         writel(bft[0], &h->transtable->BlockFetch0);
3979         writel(bft[1], &h->transtable->BlockFetch1);
3980         writel(bft[2], &h->transtable->BlockFetch2);
3981         writel(bft[3], &h->transtable->BlockFetch3);
3982         writel(bft[4], &h->transtable->BlockFetch4);
3983         writel(bft[5], &h->transtable->BlockFetch5);
3984         writel(bft[6], &h->transtable->BlockFetch6);
3985         writel(bft[7], &h->transtable->BlockFetch7);
3986
3987         /* size of controller ring buffer */
3988         writel(h->max_commands, &h->transtable->RepQSize);
3989         writel(1, &h->transtable->RepQCount);
3990         writel(0, &h->transtable->RepQCtrAddrLow32);
3991         writel(0, &h->transtable->RepQCtrAddrHigh32);
3992         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3993         writel(0, &h->transtable->RepQAddr0High32);
3994         writel(CFGTBL_Trans_Performant | use_short_tags,
3995                         &(h->cfgtable->HostWrite.TransportRequest));
3996
3997         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3998         cciss_wait_for_mode_change_ack(h);
3999         register_value = readl(&(h->cfgtable->TransportActive));
4000         if (!(register_value & CFGTBL_Trans_Performant))
4001                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
4002                                         " performant mode\n");
4003 }
4004
4005 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
4006 {
4007         __u32 trans_support;
4008
4009         if (cciss_simple_mode)
4010                 return;
4011
4012         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
4013         /* Attempt to put controller into performant mode if supported */
4014         /* Does board support performant mode? */
4015         trans_support = readl(&(h->cfgtable->TransportSupport));
4016         if (!(trans_support & PERFORMANT_MODE))
4017                 return;
4018
4019         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
4020         /* Performant mode demands commands on a 32 byte boundary
4021          * pci_alloc_consistent aligns on page boundarys already.
4022          * Just need to check if divisible by 32
4023          */
4024         if ((sizeof(CommandList_struct) % 32) != 0) {
4025                 dev_warn(&h->pdev->dev, "%s %d %s\n",
4026                         "cciss info: command size[",
4027                         (int)sizeof(CommandList_struct),
4028                         "] not divisible by 32, no performant mode..\n");
4029                 return;
4030         }
4031
4032         /* Performant mode ring buffer and supporting data structures */
4033         h->reply_pool = (__u64 *)pci_alloc_consistent(
4034                 h->pdev, h->max_commands * sizeof(__u64),
4035                 &(h->reply_pool_dhandle));
4036
4037         /* Need a block fetch table for performant mode */
4038         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
4039                 sizeof(__u32)), GFP_KERNEL);
4040
4041         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
4042                 goto clean_up;
4043
4044         cciss_enter_performant_mode(h,
4045                 trans_support & CFGTBL_Trans_use_short_tags);
4046
4047         /* Change the access methods to the performant access methods */
4048         h->access = SA5_performant_access;
4049         h->transMethod = CFGTBL_Trans_Performant;
4050
4051         return;
4052 clean_up:
4053         kfree(h->blockFetchTable);
4054         if (h->reply_pool)
4055                 pci_free_consistent(h->pdev,
4056                                 h->max_commands * sizeof(__u64),
4057                                 h->reply_pool,
4058                                 h->reply_pool_dhandle);
4059         return;
4060
4061 } /* cciss_put_controller_into_performant_mode */
4062
4063 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
4064  * controllers that are capable. If not, we use IO-APIC mode.
4065  */
4066
4067 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
4068 {
4069 #ifdef CONFIG_PCI_MSI
4070         int err;
4071         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
4072         {0, 2}, {0, 3}
4073         };
4074
4075         /* Some boards advertise MSI but don't really support it */
4076         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
4077             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
4078                 goto default_int_mode;
4079
4080         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
4081                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
4082                 if (!err) {
4083                         h->intr[0] = cciss_msix_entries[0].vector;
4084                         h->intr[1] = cciss_msix_entries[1].vector;
4085                         h->intr[2] = cciss_msix_entries[2].vector;
4086                         h->intr[3] = cciss_msix_entries[3].vector;
4087                         h->msix_vector = 1;
4088                         return;
4089                 }
4090                 if (err > 0) {
4091                         dev_warn(&h->pdev->dev,
4092                                 "only %d MSI-X vectors available\n", err);
4093                         goto default_int_mode;
4094                 } else {
4095                         dev_warn(&h->pdev->dev,
4096                                 "MSI-X init failed %d\n", err);
4097                         goto default_int_mode;
4098                 }
4099         }
4100         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
4101                 if (!pci_enable_msi(h->pdev))
4102                         h->msi_vector = 1;
4103                 else
4104                         dev_warn(&h->pdev->dev, "MSI init failed\n");
4105         }
4106 default_int_mode:
4107 #endif                          /* CONFIG_PCI_MSI */
4108         /* if we get here we're going to use the default interrupt mode */
4109         h->intr[h->intr_mode] = h->pdev->irq;
4110         return;
4111 }
4112
4113 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
4114 {
4115         int i;
4116         u32 subsystem_vendor_id, subsystem_device_id;
4117
4118         subsystem_vendor_id = pdev->subsystem_vendor;
4119         subsystem_device_id = pdev->subsystem_device;
4120         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
4121                         subsystem_vendor_id;
4122
4123         for (i = 0; i < ARRAY_SIZE(products); i++)
4124                 if (*board_id == products[i].board_id)
4125                         return i;
4126         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4127                 *board_id);
4128         return -ENODEV;
4129 }
4130
4131 static inline bool cciss_board_disabled(ctlr_info_t *h)
4132 {
4133         u16 command;
4134
4135         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4136         return ((command & PCI_COMMAND_MEMORY) == 0);
4137 }
4138
4139 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4140         unsigned long *memory_bar)
4141 {
4142         int i;
4143
4144         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4145                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4146                         /* addressing mode bits already removed */
4147                         *memory_bar = pci_resource_start(pdev, i);
4148                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4149                                 *memory_bar);
4150                         return 0;
4151                 }
4152         dev_warn(&pdev->dev, "no memory BAR found\n");
4153         return -ENODEV;
4154 }
4155
4156 static int __devinit cciss_wait_for_board_state(struct pci_dev *pdev,
4157         void __iomem *vaddr, int wait_for_ready)
4158 #define BOARD_READY 1
4159 #define BOARD_NOT_READY 0
4160 {
4161         int i, iterations;
4162         u32 scratchpad;
4163
4164         if (wait_for_ready)
4165                 iterations = CCISS_BOARD_READY_ITERATIONS;
4166         else
4167                 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
4168
4169         for (i = 0; i < iterations; i++) {
4170                 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
4171                 if (wait_for_ready) {
4172                         if (scratchpad == CCISS_FIRMWARE_READY)
4173                                 return 0;
4174                 } else {
4175                         if (scratchpad != CCISS_FIRMWARE_READY)
4176                                 return 0;
4177                 }
4178                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4179         }
4180         dev_warn(&pdev->dev, "board not ready, timed out.\n");
4181         return -ENODEV;
4182 }
4183
4184 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4185         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4186         u64 *cfg_offset)
4187 {
4188         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4189         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4190         *cfg_base_addr &= (u32) 0x0000ffff;
4191         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4192         if (*cfg_base_addr_index == -1) {
4193                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4194                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4195                 return -ENODEV;
4196         }
4197         return 0;
4198 }
4199
4200 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4201 {
4202         u64 cfg_offset;
4203         u32 cfg_base_addr;
4204         u64 cfg_base_addr_index;
4205         u32 trans_offset;
4206         int rc;
4207
4208         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4209                 &cfg_base_addr_index, &cfg_offset);
4210         if (rc)
4211                 return rc;
4212         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4213                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4214         if (!h->cfgtable)
4215                 return -ENOMEM;
4216         rc = write_driver_ver_to_cfgtable(h->cfgtable);
4217         if (rc)
4218                 return rc;
4219         /* Find performant mode table. */
4220         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4221         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4222                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4223                                 sizeof(*h->transtable));
4224         if (!h->transtable)
4225                 return -ENOMEM;
4226         return 0;
4227 }
4228
4229 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4230 {
4231         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4232
4233         /* Limit commands in memory limited kdump scenario. */
4234         if (reset_devices && h->max_commands > 32)
4235                 h->max_commands = 32;
4236
4237         if (h->max_commands < 16) {
4238                 dev_warn(&h->pdev->dev, "Controller reports "
4239                         "max supported commands of %d, an obvious lie. "
4240                         "Using 16.  Ensure that firmware is up to date.\n",
4241                         h->max_commands);
4242                 h->max_commands = 16;
4243         }
4244 }
4245
4246 /* Interrogate the hardware for some limits:
4247  * max commands, max SG elements without chaining, and with chaining,
4248  * SG chain block size, etc.
4249  */
4250 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4251 {
4252         cciss_get_max_perf_mode_cmds(h);
4253         h->nr_cmds = h->max_commands - 4 - cciss_tape_cmds;
4254         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4255         /*
4256          * Limit in-command s/g elements to 32 save dma'able memory.
4257          * Howvever spec says if 0, use 31
4258          */
4259         h->max_cmd_sgentries = 31;
4260         if (h->maxsgentries > 512) {
4261                 h->max_cmd_sgentries = 32;
4262                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4263                 h->maxsgentries--; /* save one for chain pointer */
4264         } else {
4265                 h->maxsgentries = 31; /* default to traditional values */
4266                 h->chainsize = 0;
4267         }
4268 }
4269
4270 static inline bool CISS_signature_present(ctlr_info_t *h)
4271 {
4272         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4273             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4274             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4275             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4276                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4277                 return false;
4278         }
4279         return true;
4280 }
4281
4282 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4283 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4284 {
4285 #ifdef CONFIG_X86
4286         u32 prefetch;
4287
4288         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4289         prefetch |= 0x100;
4290         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4291 #endif
4292 }
4293
4294 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4295  * in a prefetch beyond physical memory.
4296  */
4297 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4298 {
4299         u32 dma_prefetch;
4300         __u32 dma_refetch;
4301
4302         if (h->board_id != 0x3225103C)
4303                 return;
4304         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4305         dma_prefetch |= 0x8000;
4306         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4307         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4308         dma_refetch |= 0x1;
4309         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4310 }
4311
4312 static int __devinit cciss_pci_init(ctlr_info_t *h)
4313 {
4314         int prod_index, err;
4315
4316         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4317         if (prod_index < 0)
4318                 return -ENODEV;
4319         h->product_name = products[prod_index].product_name;
4320         h->access = *(products[prod_index].access);
4321
4322         if (cciss_board_disabled(h)) {
4323                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4324                 return -ENODEV;
4325         }
4326
4327         pci_disable_link_state(h->pdev, PCIE_LINK_STATE_L0S |
4328                                 PCIE_LINK_STATE_L1 | PCIE_LINK_STATE_CLKPM);
4329
4330         err = pci_enable_device(h->pdev);
4331         if (err) {
4332                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4333                 return err;
4334         }
4335
4336         err = pci_request_regions(h->pdev, "cciss");
4337         if (err) {
4338                 dev_warn(&h->pdev->dev,
4339                         "Cannot obtain PCI resources, aborting\n");
4340                 return err;
4341         }
4342
4343         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4344         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4345
4346 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4347  * else we use the IO-APIC interrupt assigned to us by system ROM.
4348  */
4349         cciss_interrupt_mode(h);
4350         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4351         if (err)
4352                 goto err_out_free_res;
4353         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4354         if (!h->vaddr) {
4355                 err = -ENOMEM;
4356                 goto err_out_free_res;
4357         }
4358         err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4359         if (err)
4360                 goto err_out_free_res;
4361         err = cciss_find_cfgtables(h);
4362         if (err)
4363                 goto err_out_free_res;
4364         print_cfg_table(h);
4365         cciss_find_board_params(h);
4366
4367         if (!CISS_signature_present(h)) {
4368                 err = -ENODEV;
4369                 goto err_out_free_res;
4370         }
4371         cciss_enable_scsi_prefetch(h);
4372         cciss_p600_dma_prefetch_quirk(h);
4373         err = cciss_enter_simple_mode(h);
4374         if (err)
4375                 goto err_out_free_res;
4376         cciss_put_controller_into_performant_mode(h);
4377         return 0;
4378
4379 err_out_free_res:
4380         /*
4381          * Deliberately omit pci_disable_device(): it does something nasty to
4382          * Smart Array controllers that pci_enable_device does not undo
4383          */
4384         if (h->transtable)
4385                 iounmap(h->transtable);
4386         if (h->cfgtable)
4387                 iounmap(h->cfgtable);
4388         if (h->vaddr)
4389                 iounmap(h->vaddr);
4390         pci_release_regions(h->pdev);
4391         return err;
4392 }
4393
4394 /* Function to find the first free pointer into our hba[] array
4395  * Returns -1 if no free entries are left.
4396  */
4397 static int alloc_cciss_hba(struct pci_dev *pdev)
4398 {
4399         int i;
4400
4401         for (i = 0; i < MAX_CTLR; i++) {
4402                 if (!hba[i]) {
4403                         ctlr_info_t *h;
4404
4405                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4406                         if (!h)
4407                                 goto Enomem;
4408                         hba[i] = h;
4409                         return i;
4410                 }
4411         }
4412         dev_warn(&pdev->dev, "This driver supports a maximum"
4413                " of %d controllers.\n", MAX_CTLR);
4414         return -1;
4415 Enomem:
4416         dev_warn(&pdev->dev, "out of memory.\n");
4417         return -1;
4418 }
4419
4420 static void free_hba(ctlr_info_t *h)
4421 {
4422         int i;
4423
4424         hba[h->ctlr] = NULL;
4425         for (i = 0; i < h->highest_lun + 1; i++)
4426                 if (h->gendisk[i] != NULL)
4427                         put_disk(h->gendisk[i]);
4428         kfree(h);
4429 }
4430
4431 /* Send a message CDB to the firmware. */
4432 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4433 {
4434         typedef struct {
4435                 CommandListHeader_struct CommandHeader;
4436                 RequestBlock_struct Request;
4437                 ErrDescriptor_struct ErrorDescriptor;
4438         } Command;
4439         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4440         Command *cmd;
4441         dma_addr_t paddr64;
4442         uint32_t paddr32, tag;
4443         void __iomem *vaddr;
4444         int i, err;
4445
4446         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4447         if (vaddr == NULL)
4448                 return -ENOMEM;
4449
4450         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4451            CCISS commands, so they must be allocated from the lower 4GiB of
4452            memory. */
4453         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4454         if (err) {
4455                 iounmap(vaddr);
4456                 return -ENOMEM;
4457         }
4458
4459         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4460         if (cmd == NULL) {
4461                 iounmap(vaddr);
4462                 return -ENOMEM;
4463         }
4464
4465         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4466            although there's no guarantee, we assume that the address is at
4467            least 4-byte aligned (most likely, it's page-aligned). */
4468         paddr32 = paddr64;
4469
4470         cmd->CommandHeader.ReplyQueue = 0;
4471         cmd->CommandHeader.SGList = 0;
4472         cmd->CommandHeader.SGTotal = 0;
4473         cmd->CommandHeader.Tag.lower = paddr32;
4474         cmd->CommandHeader.Tag.upper = 0;
4475         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4476
4477         cmd->Request.CDBLen = 16;
4478         cmd->Request.Type.Type = TYPE_MSG;
4479         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4480         cmd->Request.Type.Direction = XFER_NONE;
4481         cmd->Request.Timeout = 0; /* Don't time out */
4482         cmd->Request.CDB[0] = opcode;
4483         cmd->Request.CDB[1] = type;
4484         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4485
4486         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4487         cmd->ErrorDescriptor.Addr.upper = 0;
4488         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4489
4490         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4491
4492         for (i = 0; i < 10; i++) {
4493                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4494                 if ((tag & ~3) == paddr32)
4495                         break;
4496                 msleep(CCISS_POST_RESET_NOOP_TIMEOUT_MSECS);
4497         }
4498
4499         iounmap(vaddr);
4500
4501         /* we leak the DMA buffer here ... no choice since the controller could
4502            still complete the command. */
4503         if (i == 10) {
4504                 dev_err(&pdev->dev,
4505                         "controller message %02x:%02x timed out\n",
4506                         opcode, type);
4507                 return -ETIMEDOUT;
4508         }
4509
4510         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4511
4512         if (tag & 2) {
4513                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4514                         opcode, type);
4515                 return -EIO;
4516         }
4517
4518         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4519                 opcode, type);
4520         return 0;
4521 }
4522
4523 #define cciss_noop(p) cciss_message(p, 3, 0)
4524
4525 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4526         void * __iomem vaddr, u32 use_doorbell)
4527 {
4528         u16 pmcsr;
4529         int pos;
4530
4531         if (use_doorbell) {
4532                 /* For everything after the P600, the PCI power state method
4533                  * of resetting the controller doesn't work, so we have this
4534                  * other way using the doorbell register.
4535                  */
4536                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4537                 writel(use_doorbell, vaddr + SA5_DOORBELL);
4538         } else { /* Try to do it the PCI power state way */
4539
4540                 /* Quoting from the Open CISS Specification: "The Power
4541                  * Management Control/Status Register (CSR) controls the power
4542                  * state of the device.  The normal operating state is D0,
4543                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4544                  * the controller, place the interface device in D3 then to D0,
4545                  * this causes a secondary PCI reset which will reset the
4546                  * controller." */
4547
4548                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4549                 if (pos == 0) {
4550                         dev_err(&pdev->dev,
4551                                 "cciss_controller_hard_reset: "
4552                                 "PCI PM not supported\n");
4553                         return -ENODEV;
4554                 }
4555                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4556                 /* enter the D3hot power management state */
4557                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4558                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4559                 pmcsr |= PCI_D3hot;
4560                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4561
4562                 msleep(500);
4563
4564                 /* enter the D0 power management state */
4565                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4566                 pmcsr |= PCI_D0;
4567                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4568
4569                 /*
4570                  * The P600 requires a small delay when changing states.
4571                  * Otherwise we may think the board did not reset and we bail.
4572                  * This for kdump only and is particular to the P600.
4573                  */
4574                 msleep(500);
4575         }
4576         return 0;
4577 }
4578
4579 static __devinit void init_driver_version(char *driver_version, int len)
4580 {
4581         memset(driver_version, 0, len);
4582         strncpy(driver_version, "cciss " DRIVER_NAME, len - 1);
4583 }
4584
4585 static __devinit int write_driver_ver_to_cfgtable(
4586         CfgTable_struct __iomem *cfgtable)
4587 {
4588         char *driver_version;
4589         int i, size = sizeof(cfgtable->driver_version);
4590
4591         driver_version = kmalloc(size, GFP_KERNEL);
4592         if (!driver_version)
4593                 return -ENOMEM;
4594
4595         init_driver_version(driver_version, size);
4596         for (i = 0; i < size; i++)
4597                 writeb(driver_version[i], &cfgtable->driver_version[i]);
4598         kfree(driver_version);
4599         return 0;
4600 }
4601
4602 static __devinit void read_driver_ver_from_cfgtable(
4603         CfgTable_struct __iomem *cfgtable, unsigned char *driver_ver)
4604 {
4605         int i;
4606
4607         for (i = 0; i < sizeof(cfgtable->driver_version); i++)
4608                 driver_ver[i] = readb(&cfgtable->driver_version[i]);
4609 }
4610
4611 static __devinit int controller_reset_failed(
4612         CfgTable_struct __iomem *cfgtable)
4613 {
4614
4615         char *driver_ver, *old_driver_ver;
4616         int rc, size = sizeof(cfgtable->driver_version);
4617
4618         old_driver_ver = kmalloc(2 * size, GFP_KERNEL);
4619         if (!old_driver_ver)
4620                 return -ENOMEM;
4621         driver_ver = old_driver_ver + size;
4622
4623         /* After a reset, the 32 bytes of "driver version" in the cfgtable
4624          * should have been changed, otherwise we know the reset failed.
4625          */
4626         init_driver_version(old_driver_ver, size);
4627         read_driver_ver_from_cfgtable(cfgtable, driver_ver);
4628         rc = !memcmp(driver_ver, old_driver_ver, size);
4629         kfree(old_driver_ver);
4630         return rc;
4631 }
4632
4633 /* This does a hard reset of the controller using PCI power management
4634  * states or using the doorbell register. */
4635 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4636 {
4637         u64 cfg_offset;
4638         u32 cfg_base_addr;
4639         u64 cfg_base_addr_index;
4640         void __iomem *vaddr;
4641         unsigned long paddr;
4642         u32 misc_fw_support;
4643         int rc;
4644         CfgTable_struct __iomem *cfgtable;
4645         u32 use_doorbell;
4646         u32 board_id;
4647         u16 command_register;
4648
4649         /* For controllers as old a the p600, this is very nearly
4650          * the same thing as
4651          *
4652          * pci_save_state(pci_dev);
4653          * pci_set_power_state(pci_dev, PCI_D3hot);
4654          * pci_set_power_state(pci_dev, PCI_D0);
4655          * pci_restore_state(pci_dev);
4656          *
4657          * For controllers newer than the P600, the pci power state
4658          * method of resetting doesn't work so we have another way
4659          * using the doorbell register.
4660          */
4661
4662         /* Exclude 640x boards.  These are two pci devices in one slot
4663          * which share a battery backed cache module.  One controls the
4664          * cache, the other accesses the cache through the one that controls
4665          * it.  If we reset the one controlling the cache, the other will
4666          * likely not be happy.  Just forbid resetting this conjoined mess.
4667          */
4668         cciss_lookup_board_id(pdev, &board_id);
4669         if (!ctlr_is_resettable(board_id)) {
4670                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4671                                 "due to shared cache module.");
4672                 return -ENODEV;
4673         }
4674
4675         /* if controller is soft- but not hard resettable... */
4676         if (!ctlr_is_hard_resettable(board_id))
4677                 return -ENOTSUPP; /* try soft reset later. */
4678
4679         /* Save the PCI command register */
4680         pci_read_config_word(pdev, 4, &command_register);
4681         /* Turn the board off.  This is so that later pci_restore_state()
4682          * won't turn the board on before the rest of config space is ready.
4683          */
4684         pci_disable_device(pdev);
4685         pci_save_state(pdev);
4686
4687         /* find the first memory BAR, so we can find the cfg table */
4688         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4689         if (rc)
4690                 return rc;
4691         vaddr = remap_pci_mem(paddr, 0x250);
4692         if (!vaddr)
4693                 return -ENOMEM;
4694
4695         /* find cfgtable in order to check if reset via doorbell is supported */
4696         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4697                                         &cfg_base_addr_index, &cfg_offset);
4698         if (rc)
4699                 goto unmap_vaddr;
4700         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4701                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4702         if (!cfgtable) {
4703                 rc = -ENOMEM;
4704                 goto unmap_vaddr;
4705         }
4706         rc = write_driver_ver_to_cfgtable(cfgtable);
4707         if (rc)
4708                 goto unmap_vaddr;
4709
4710         /* If reset via doorbell register is supported, use that.
4711          * There are two such methods.  Favor the newest method.
4712          */
4713         misc_fw_support = readl(&cfgtable->misc_fw_support);
4714         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET2;
4715         if (use_doorbell) {
4716                 use_doorbell = DOORBELL_CTLR_RESET2;
4717         } else {
4718                 use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4719                 if (use_doorbell) {
4720                         dev_warn(&pdev->dev, "Controller claims that "
4721                                 "'Bit 2 doorbell reset' is "
4722                                 "supported, but not 'bit 5 doorbell reset'.  "
4723                                 "Firmware update is recommended.\n");
4724                         rc = -ENOTSUPP; /* use the soft reset */
4725                         goto unmap_cfgtable;
4726                 }
4727         }
4728
4729         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4730         if (rc)
4731                 goto unmap_cfgtable;
4732         pci_restore_state(pdev);
4733         rc = pci_enable_device(pdev);
4734         if (rc) {
4735                 dev_warn(&pdev->dev, "failed to enable device.\n");
4736                 goto unmap_cfgtable;
4737         }
4738         pci_write_config_word(pdev, 4, command_register);
4739
4740         /* Some devices (notably the HP Smart Array 5i Controller)
4741            need a little pause here */
4742         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4743
4744         /* Wait for board to become not ready, then ready. */
4745         dev_info(&pdev->dev, "Waiting for board to reset.\n");
4746         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4747         if (rc) {
4748                 dev_warn(&pdev->dev, "Failed waiting for board to hard reset."
4749                                 "  Will try soft reset.\n");
4750                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4751                 goto unmap_cfgtable;
4752         }
4753         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4754         if (rc) {
4755                 dev_warn(&pdev->dev,
4756                         "failed waiting for board to become ready "
4757                         "after hard reset\n");
4758                 goto unmap_cfgtable;
4759         }
4760
4761         rc = controller_reset_failed(vaddr);
4762         if (rc < 0)
4763                 goto unmap_cfgtable;
4764         if (rc) {
4765                 dev_warn(&pdev->dev, "Unable to successfully hard reset "
4766                         "controller. Will try soft reset.\n");
4767                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4768         } else {
4769                 dev_info(&pdev->dev, "Board ready after hard reset.\n");
4770         }
4771
4772 unmap_cfgtable:
4773         iounmap(cfgtable);
4774
4775 unmap_vaddr:
4776         iounmap(vaddr);
4777         return rc;
4778 }
4779
4780 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4781 {
4782         int rc, i;
4783
4784         if (!reset_devices)
4785                 return 0;
4786
4787         /* Reset the controller with a PCI power-cycle or via doorbell */
4788         rc = cciss_kdump_hard_reset_controller(pdev);
4789
4790         /* -ENOTSUPP here means we cannot reset the controller
4791          * but it's already (and still) up and running in
4792          * "performant mode".  Or, it might be 640x, which can't reset
4793          * due to concerns about shared bbwc between 6402/6404 pair.
4794          */
4795         if (rc == -ENOTSUPP)
4796                 return rc; /* just try to do the kdump anyhow. */
4797         if (rc)
4798                 return -ENODEV;
4799
4800         /* Now try to get the controller to respond to a no-op */
4801         dev_warn(&pdev->dev, "Waiting for controller to respond to no-op\n");
4802         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4803                 if (cciss_noop(pdev) == 0)
4804                         break;
4805                 else
4806                         dev_warn(&pdev->dev, "no-op failed%s\n",
4807                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4808                                         "; re-trying" : ""));
4809                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4810         }
4811         return 0;
4812 }
4813
4814 static __devinit int cciss_allocate_cmd_pool(ctlr_info_t *h)
4815 {
4816         h->cmd_pool_bits = kmalloc(
4817                 DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG) *
4818                 sizeof(unsigned long), GFP_KERNEL);
4819         h->cmd_pool = pci_alloc_consistent(h->pdev,
4820                 h->nr_cmds * sizeof(CommandList_struct),
4821                 &(h->cmd_pool_dhandle));
4822         h->errinfo_pool = pci_alloc_consistent(h->pdev,
4823                 h->nr_cmds * sizeof(ErrorInfo_struct),
4824                 &(h->errinfo_pool_dhandle));
4825         if ((h->cmd_pool_bits == NULL)
4826                 || (h->cmd_pool == NULL)
4827                 || (h->errinfo_pool == NULL)) {
4828                 dev_err(&h->pdev->dev, "out of memory");
4829                 return -ENOMEM;
4830         }
4831         return 0;
4832 }
4833
4834 static __devinit int cciss_allocate_scatterlists(ctlr_info_t *h)
4835 {
4836         int i;
4837
4838         /* zero it, so that on free we need not know how many were alloc'ed */
4839         h->scatter_list = kzalloc(h->max_commands *
4840                                 sizeof(struct scatterlist *), GFP_KERNEL);
4841         if (!h->scatter_list)
4842                 return -ENOMEM;
4843
4844         for (i = 0; i < h->nr_cmds; i++) {
4845                 h->scatter_list[i] = kmalloc(sizeof(struct scatterlist) *
4846                                                 h->maxsgentries, GFP_KERNEL);
4847                 if (h->scatter_list[i] == NULL) {
4848                         dev_err(&h->pdev->dev, "could not allocate "
4849                                 "s/g lists\n");
4850                         return -ENOMEM;
4851                 }
4852         }
4853         return 0;
4854 }
4855
4856 static void cciss_free_scatterlists(ctlr_info_t *h)
4857 {
4858         int i;
4859
4860         if (h->scatter_list) {
4861                 for (i = 0; i < h->nr_cmds; i++)
4862                         kfree(h->scatter_list[i]);
4863                 kfree(h->scatter_list);
4864         }
4865 }
4866
4867 static void cciss_free_cmd_pool(ctlr_info_t *h)
4868 {
4869         kfree(h->cmd_pool_bits);
4870         if (h->cmd_pool)
4871                 pci_free_consistent(h->pdev,
4872                         h->nr_cmds * sizeof(CommandList_struct),
4873                         h->cmd_pool, h->cmd_pool_dhandle);
4874         if (h->errinfo_pool)
4875                 pci_free_consistent(h->pdev,
4876                         h->nr_cmds * sizeof(ErrorInfo_struct),
4877                         h->errinfo_pool, h->errinfo_pool_dhandle);
4878 }
4879
4880 static int cciss_request_irq(ctlr_info_t *h,
4881         irqreturn_t (*msixhandler)(int, void *),
4882         irqreturn_t (*intxhandler)(int, void *))
4883 {
4884         if (h->msix_vector || h->msi_vector) {
4885                 if (!request_irq(h->intr[h->intr_mode], msixhandler,
4886                                 0, h->devname, h))
4887                         return 0;
4888                 dev_err(&h->pdev->dev, "Unable to get msi irq %d"
4889                         " for %s\n", h->intr[h->intr_mode],
4890                         h->devname);
4891                 return -1;
4892         }
4893
4894         if (!request_irq(h->intr[h->intr_mode], intxhandler,
4895                         IRQF_SHARED, h->devname, h))
4896                 return 0;
4897         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4898                 h->intr[h->intr_mode], h->devname);
4899         return -1;
4900 }
4901
4902 static int __devinit cciss_kdump_soft_reset(ctlr_info_t *h)
4903 {
4904         if (cciss_send_reset(h, CTLR_LUNID, CCISS_RESET_TYPE_CONTROLLER)) {
4905                 dev_warn(&h->pdev->dev, "Resetting array controller failed.\n");
4906                 return -EIO;
4907         }
4908
4909         dev_info(&h->pdev->dev, "Waiting for board to soft reset.\n");
4910         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_NOT_READY)) {
4911                 dev_warn(&h->pdev->dev, "Soft reset had no effect.\n");
4912                 return -1;
4913         }
4914
4915         dev_info(&h->pdev->dev, "Board reset, awaiting READY status.\n");
4916         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY)) {
4917                 dev_warn(&h->pdev->dev, "Board failed to become ready "
4918                         "after soft reset.\n");
4919                 return -1;
4920         }
4921
4922         return 0;
4923 }
4924
4925 static void cciss_undo_allocations_after_kdump_soft_reset(ctlr_info_t *h)
4926 {
4927         int ctlr = h->ctlr;
4928
4929         free_irq(h->intr[h->intr_mode], h);
4930 #ifdef CONFIG_PCI_MSI
4931         if (h->msix_vector)
4932                 pci_disable_msix(h->pdev);
4933         else if (h->msi_vector)
4934                 pci_disable_msi(h->pdev);
4935 #endif /* CONFIG_PCI_MSI */
4936         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4937         cciss_free_scatterlists(h);
4938         cciss_free_cmd_pool(h);
4939         kfree(h->blockFetchTable);
4940         if (h->reply_pool)
4941                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
4942                                 h->reply_pool, h->reply_pool_dhandle);
4943         if (h->transtable)
4944                 iounmap(h->transtable);
4945         if (h->cfgtable)
4946                 iounmap(h->cfgtable);
4947         if (h->vaddr)
4948                 iounmap(h->vaddr);
4949         unregister_blkdev(h->major, h->devname);
4950         cciss_destroy_hba_sysfs_entry(h);
4951         pci_release_regions(h->pdev);
4952         kfree(h);
4953         hba[ctlr] = NULL;
4954 }
4955
4956 /*
4957  *  This is it.  Find all the controllers and register them.  I really hate
4958  *  stealing all these major device numbers.
4959  *  returns the number of block devices registered.
4960  */
4961 static int __devinit cciss_init_one(struct pci_dev *pdev,
4962                                     const struct pci_device_id *ent)
4963 {
4964         int i;
4965         int j = 0;
4966         int rc;
4967         int try_soft_reset = 0;
4968         int dac, return_code;
4969         InquiryData_struct *inq_buff;
4970         ctlr_info_t *h;
4971         unsigned long flags;
4972
4973         rc = cciss_init_reset_devices(pdev);
4974         if (rc) {
4975                 if (rc != -ENOTSUPP)
4976                         return rc;
4977                 /* If the reset fails in a particular way (it has no way to do
4978                  * a proper hard reset, so returns -ENOTSUPP) we can try to do
4979                  * a soft reset once we get the controller configured up to the
4980                  * point that it can accept a command.
4981                  */
4982                 try_soft_reset = 1;
4983                 rc = 0;
4984         }
4985
4986 reinit_after_soft_reset:
4987
4988         i = alloc_cciss_hba(pdev);
4989         if (i < 0)
4990                 return -1;
4991
4992         h = hba[i];
4993         h->pdev = pdev;
4994         h->busy_initializing = 1;
4995         h->intr_mode = cciss_simple_mode ? SIMPLE_MODE_INT : PERF_MODE_INT;
4996         INIT_LIST_HEAD(&h->cmpQ);
4997         INIT_LIST_HEAD(&h->reqQ);
4998         mutex_init(&h->busy_shutting_down);
4999
5000         if (cciss_pci_init(h) != 0)
5001                 goto clean_no_release_regions;
5002
5003         sprintf(h->devname, "cciss%d", i);
5004         h->ctlr = i;
5005
5006         if (cciss_tape_cmds < 2)
5007                 cciss_tape_cmds = 2;
5008         if (cciss_tape_cmds > 16)
5009                 cciss_tape_cmds = 16;
5010
5011         init_completion(&h->scan_wait);
5012
5013         if (cciss_create_hba_sysfs_entry(h))
5014                 goto clean0;
5015
5016         /* configure PCI DMA stuff */
5017         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
5018                 dac = 1;
5019         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
5020                 dac = 0;
5021         else {
5022                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
5023                 goto clean1;
5024         }
5025
5026         /*
5027          * register with the major number, or get a dynamic major number
5028          * by passing 0 as argument.  This is done for greater than
5029          * 8 controller support.
5030          */
5031         if (i < MAX_CTLR_ORIG)
5032                 h->major = COMPAQ_CISS_MAJOR + i;
5033         rc = register_blkdev(h->major, h->devname);
5034         if (rc == -EBUSY || rc == -EINVAL) {
5035                 dev_err(&h->pdev->dev,
5036                        "Unable to get major number %d for %s "
5037                        "on hba %d\n", h->major, h->devname, i);
5038                 goto clean1;
5039         } else {
5040                 if (i >= MAX_CTLR_ORIG)
5041                         h->major = rc;
5042         }
5043
5044         /* make sure the board interrupts are off */
5045         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5046         rc = cciss_request_irq(h, do_cciss_msix_intr, do_cciss_intx);
5047         if (rc)
5048                 goto clean2;
5049
5050         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
5051                h->devname, pdev->device, pci_name(pdev),
5052                h->intr[h->intr_mode], dac ? "" : " not");
5053
5054         if (cciss_allocate_cmd_pool(h))
5055                 goto clean4;
5056
5057         if (cciss_allocate_scatterlists(h))
5058                 goto clean4;
5059
5060         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
5061                 h->chainsize, h->nr_cmds);
5062         if (!h->cmd_sg_list && h->chainsize > 0)
5063                 goto clean4;
5064
5065         spin_lock_init(&h->lock);
5066
5067         /* Initialize the pdev driver private data.
5068            have it point to h.  */
5069         pci_set_drvdata(pdev, h);
5070         /* command and error info recs zeroed out before
5071            they are used */
5072         memset(h->cmd_pool_bits, 0,
5073                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
5074                         * sizeof(unsigned long));
5075
5076         h->num_luns = 0;
5077         h->highest_lun = -1;
5078         for (j = 0; j < CISS_MAX_LUN; j++) {
5079                 h->drv[j] = NULL;
5080                 h->gendisk[j] = NULL;
5081         }
5082
5083         /* At this point, the controller is ready to take commands.
5084          * Now, if reset_devices and the hard reset didn't work, try
5085          * the soft reset and see if that works.
5086          */
5087         if (try_soft_reset) {
5088
5089                 /* This is kind of gross.  We may or may not get a completion
5090                  * from the soft reset command, and if we do, then the value
5091                  * from the fifo may or may not be valid.  So, we wait 10 secs
5092                  * after the reset throwing away any completions we get during
5093                  * that time.  Unregister the interrupt handler and register
5094                  * fake ones to scoop up any residual completions.
5095                  */
5096                 spin_lock_irqsave(&h->lock, flags);
5097                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5098                 spin_unlock_irqrestore(&h->lock, flags);
5099                 free_irq(h->intr[h->intr_mode], h);
5100                 rc = cciss_request_irq(h, cciss_msix_discard_completions,
5101                                         cciss_intx_discard_completions);
5102                 if (rc) {
5103                         dev_warn(&h->pdev->dev, "Failed to request_irq after "
5104                                 "soft reset.\n");
5105                         goto clean4;
5106                 }
5107
5108                 rc = cciss_kdump_soft_reset(h);
5109                 if (rc) {
5110                         dev_warn(&h->pdev->dev, "Soft reset failed.\n");
5111                         goto clean4;
5112                 }
5113
5114                 dev_info(&h->pdev->dev, "Board READY.\n");
5115                 dev_info(&h->pdev->dev,
5116                         "Waiting for stale completions to drain.\n");
5117                 h->access.set_intr_mask(h, CCISS_INTR_ON);
5118                 msleep(10000);
5119                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5120
5121                 rc = controller_reset_failed(h->cfgtable);
5122                 if (rc)
5123                         dev_info(&h->pdev->dev,
5124                                 "Soft reset appears to have failed.\n");
5125
5126                 /* since the controller's reset, we have to go back and re-init
5127                  * everything.  Easiest to just forget what we've done and do it
5128                  * all over again.
5129                  */
5130                 cciss_undo_allocations_after_kdump_soft_reset(h);
5131                 try_soft_reset = 0;
5132                 if (rc)
5133                         /* don't go to clean4, we already unallocated */
5134                         return -ENODEV;
5135
5136                 goto reinit_after_soft_reset;
5137         }
5138
5139         cciss_scsi_setup(h);
5140
5141         /* Turn the interrupts on so we can service requests */
5142         h->access.set_intr_mask(h, CCISS_INTR_ON);
5143
5144         /* Get the firmware version */
5145         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
5146         if (inq_buff == NULL) {
5147                 dev_err(&h->pdev->dev, "out of memory\n");
5148                 goto clean4;
5149         }
5150
5151         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
5152                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
5153         if (return_code == IO_OK) {
5154                 h->firm_ver[0] = inq_buff->data_byte[32];
5155                 h->firm_ver[1] = inq_buff->data_byte[33];
5156                 h->firm_ver[2] = inq_buff->data_byte[34];
5157                 h->firm_ver[3] = inq_buff->data_byte[35];
5158         } else {         /* send command failed */
5159                 dev_warn(&h->pdev->dev, "unable to determine firmware"
5160                         " version of controller\n");
5161         }
5162         kfree(inq_buff);
5163
5164         cciss_procinit(h);
5165
5166         h->cciss_max_sectors = 8192;
5167
5168         rebuild_lun_table(h, 1, 0);
5169         cciss_engage_scsi(h);
5170         h->busy_initializing = 0;
5171         return 1;
5172
5173 clean4:
5174         cciss_free_cmd_pool(h);
5175         cciss_free_scatterlists(h);
5176         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5177         free_irq(h->intr[h->intr_mode], h);
5178 clean2:
5179         unregister_blkdev(h->major, h->devname);
5180 clean1:
5181         cciss_destroy_hba_sysfs_entry(h);
5182 clean0:
5183         pci_release_regions(pdev);
5184 clean_no_release_regions:
5185         h->busy_initializing = 0;
5186
5187         /*
5188          * Deliberately omit pci_disable_device(): it does something nasty to
5189          * Smart Array controllers that pci_enable_device does not undo
5190          */
5191         pci_set_drvdata(pdev, NULL);
5192         free_hba(h);
5193         return -1;
5194 }
5195
5196 static void cciss_shutdown(struct pci_dev *pdev)
5197 {
5198         ctlr_info_t *h;
5199         char *flush_buf;
5200         int return_code;
5201
5202         h = pci_get_drvdata(pdev);
5203         flush_buf = kzalloc(4, GFP_KERNEL);
5204         if (!flush_buf) {
5205                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
5206                 return;
5207         }
5208         /* write all data in the battery backed cache to disk */
5209         memset(flush_buf, 0, 4);
5210         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
5211                 4, 0, CTLR_LUNID, TYPE_CMD);
5212         kfree(flush_buf);
5213         if (return_code != IO_OK)
5214                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
5215         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5216         free_irq(h->intr[h->intr_mode], h);
5217 }
5218
5219 static int __devinit cciss_enter_simple_mode(struct ctlr_info *h)
5220 {
5221         u32 trans_support;
5222
5223         trans_support = readl(&(h->cfgtable->TransportSupport));
5224         if (!(trans_support & SIMPLE_MODE))
5225                 return -ENOTSUPP;
5226
5227         h->max_commands = readl(&(h->cfgtable->CmdsOutMax));
5228         writel(CFGTBL_Trans_Simple, &(h->cfgtable->HostWrite.TransportRequest));
5229         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
5230         cciss_wait_for_mode_change_ack(h);
5231         print_cfg_table(h);
5232         if (!(readl(&(h->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
5233                 dev_warn(&h->pdev->dev, "unable to get board into simple mode\n");
5234                 return -ENODEV;
5235         }
5236         h->transMethod = CFGTBL_Trans_Simple;
5237         return 0;
5238 }
5239
5240
5241 static void __devexit cciss_remove_one(struct pci_dev *pdev)
5242 {
5243         ctlr_info_t *h;
5244         int i, j;
5245
5246         if (pci_get_drvdata(pdev) == NULL) {
5247                 dev_err(&pdev->dev, "Unable to remove device\n");
5248                 return;
5249         }
5250
5251         h = pci_get_drvdata(pdev);
5252         i = h->ctlr;
5253         if (hba[i] == NULL) {
5254                 dev_err(&pdev->dev, "device appears to already be removed\n");
5255                 return;
5256         }
5257
5258         mutex_lock(&h->busy_shutting_down);
5259
5260         remove_from_scan_list(h);
5261         remove_proc_entry(h->devname, proc_cciss);
5262         unregister_blkdev(h->major, h->devname);
5263
5264         /* remove it from the disk list */
5265         for (j = 0; j < CISS_MAX_LUN; j++) {
5266                 struct gendisk *disk = h->gendisk[j];
5267                 if (disk) {
5268                         struct request_queue *q = disk->queue;
5269
5270                         if (disk->flags & GENHD_FL_UP) {
5271                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
5272                                 del_gendisk(disk);
5273                         }
5274                         if (q)
5275                                 blk_cleanup_queue(q);
5276                 }
5277         }
5278
5279 #ifdef CONFIG_CISS_SCSI_TAPE
5280         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
5281 #endif
5282
5283         cciss_shutdown(pdev);
5284
5285 #ifdef CONFIG_PCI_MSI
5286         if (h->msix_vector)
5287                 pci_disable_msix(h->pdev);
5288         else if (h->msi_vector)
5289                 pci_disable_msi(h->pdev);
5290 #endif                          /* CONFIG_PCI_MSI */
5291
5292         iounmap(h->transtable);
5293         iounmap(h->cfgtable);
5294         iounmap(h->vaddr);
5295
5296         cciss_free_cmd_pool(h);
5297         /* Free up sg elements */
5298         for (j = 0; j < h->nr_cmds; j++)
5299                 kfree(h->scatter_list[j]);
5300         kfree(h->scatter_list);
5301         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5302         kfree(h->blockFetchTable);
5303         if (h->reply_pool)
5304                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
5305                                 h->reply_pool, h->reply_pool_dhandle);
5306         /*
5307          * Deliberately omit pci_disable_device(): it does something nasty to
5308          * Smart Array controllers that pci_enable_device does not undo
5309          */
5310         pci_release_regions(pdev);
5311         pci_set_drvdata(pdev, NULL);
5312         cciss_destroy_hba_sysfs_entry(h);
5313         mutex_unlock(&h->busy_shutting_down);
5314         free_hba(h);
5315 }
5316
5317 static struct pci_driver cciss_pci_driver = {
5318         .name = "cciss",
5319         .probe = cciss_init_one,
5320         .remove = __devexit_p(cciss_remove_one),
5321         .id_table = cciss_pci_device_id,        /* id_table */
5322         .shutdown = cciss_shutdown,
5323 };
5324
5325 /*
5326  *  This is it.  Register the PCI driver information for the cards we control
5327  *  the OS will call our registered routines when it finds one of our cards.
5328  */
5329 static int __init cciss_init(void)
5330 {
5331         int err;
5332
5333         /*
5334          * The hardware requires that commands are aligned on a 64-bit
5335          * boundary. Given that we use pci_alloc_consistent() to allocate an
5336          * array of them, the size must be a multiple of 8 bytes.
5337          */
5338         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
5339         printk(KERN_INFO DRIVER_NAME "\n");
5340
5341         err = bus_register(&cciss_bus_type);
5342         if (err)
5343                 return err;
5344
5345         /* Start the scan thread */
5346         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
5347         if (IS_ERR(cciss_scan_thread)) {
5348                 err = PTR_ERR(cciss_scan_thread);
5349                 goto err_bus_unregister;
5350         }
5351
5352         /* Register for our PCI devices */
5353         err = pci_register_driver(&cciss_pci_driver);
5354         if (err)
5355                 goto err_thread_stop;
5356
5357         return err;
5358
5359 err_thread_stop:
5360         kthread_stop(cciss_scan_thread);
5361 err_bus_unregister:
5362         bus_unregister(&cciss_bus_type);
5363
5364         return err;
5365 }
5366
5367 static void __exit cciss_cleanup(void)
5368 {
5369         int i;
5370
5371         pci_unregister_driver(&cciss_pci_driver);
5372         /* double check that all controller entrys have been removed */
5373         for (i = 0; i < MAX_CTLR; i++) {
5374                 if (hba[i] != NULL) {
5375                         dev_warn(&hba[i]->pdev->dev,
5376                                 "had to remove controller\n");
5377                         cciss_remove_one(hba[i]->pdev);
5378                 }
5379         }
5380         kthread_stop(cciss_scan_thread);
5381         if (proc_cciss)
5382                 remove_proc_entry("driver/cciss", NULL);
5383         bus_unregister(&cciss_bus_type);
5384 }
5385
5386 module_init(cciss_init);
5387 module_exit(cciss_cleanup);