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/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/smp_lock.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 3.6.20)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 20)
  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 SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
  66                         " SA6i P600 P800 P400 P400i E200 E200i E500 P700m"
  67                         " Smart Array G2 Series SAS/SATA Controllers");
  68 MODULE_VERSION("3.6.20");
  69 MODULE_LICENSE("GPL");
  70
  71 static int cciss_allow_hpsa;
  72 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  73 MODULE_PARM_DESC(cciss_allow_hpsa,
  74         "Prevent cciss driver from accessing hardware known to be "
  75         " supported by the hpsa driver");
  76
  77 #include "cciss_cmd.h"
  78 #include "cciss.h"
  79 #include <linux/cciss_ioctl.h>
  80
  81 /* define the PCI info for the cards we can control */
  82 static const struct pci_device_id cciss_pci_device_id[] = {
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  90         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  91         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 110         {0,}
 111 };
 112
 113 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 114
 115 /*  board_id = Subsystem Device ID & Vendor ID
 116  *  product = Marketing Name for the board
 117  *  access = Address of the struct of function pointers
 118  */
 119 static struct board_type products[] = {
 120         {0x40700E11, "Smart Array 5300", &SA5_access},
 121         {0x40800E11, "Smart Array 5i", &SA5B_access},
 122         {0x40820E11, "Smart Array 532", &SA5B_access},
 123         {0x40830E11, "Smart Array 5312", &SA5B_access},
 124         {0x409A0E11, "Smart Array 641", &SA5_access},
 125         {0x409B0E11, "Smart Array 642", &SA5_access},
 126         {0x409C0E11, "Smart Array 6400", &SA5_access},
 127         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 128         {0x40910E11, "Smart Array 6i", &SA5_access},
 129         {0x3225103C, "Smart Array P600", &SA5_access},
 130         {0x3235103C, "Smart Array P400i", &SA5_access},
 131         {0x3211103C, "Smart Array E200i", &SA5_access},
 132         {0x3212103C, "Smart Array E200", &SA5_access},
 133         {0x3213103C, "Smart Array E200i", &SA5_access},
 134         {0x3214103C, "Smart Array E200i", &SA5_access},
 135         {0x3215103C, "Smart Array E200i", &SA5_access},
 136         {0x3237103C, "Smart Array E500", &SA5_access},
 137 /* controllers below this line are also supported by the hpsa driver. */
 138 #define HPSA_BOUNDARY 0x3223103C
 139         {0x3223103C, "Smart Array P800", &SA5_access},
 140         {0x3234103C, "Smart Array P400", &SA5_access},
 141         {0x323D103C, "Smart Array P700m", &SA5_access},
 142         {0x3241103C, "Smart Array P212", &SA5_access},
 143         {0x3243103C, "Smart Array P410", &SA5_access},
 144         {0x3245103C, "Smart Array P410i", &SA5_access},
 145         {0x3247103C, "Smart Array P411", &SA5_access},
 146         {0x3249103C, "Smart Array P812", &SA5_access},
 147         {0x324A103C, "Smart Array P712m", &SA5_access},
 148         {0x324B103C, "Smart Array P711m", &SA5_access},
 149 };
 150
 151 /* How long to wait (in milliseconds) for board to go into simple mode */
 152 #define MAX_CONFIG_WAIT 30000
 153 #define MAX_IOCTL_CONFIG_WAIT 1000
 154
 155 /*define how many times we will try a command because of bus resets */
 156 #define MAX_CMD_RETRIES 3
 157
 158 #define MAX_CTLR        32
 159
 160 /* Originally cciss driver only supports 8 major numbers */
 161 #define MAX_CTLR_ORIG   8
 162
 163 static ctlr_info_t *hba[MAX_CTLR];
 164
 165 static struct task_struct *cciss_scan_thread;
 166 static DEFINE_MUTEX(scan_mutex);
 167 static LIST_HEAD(scan_q);
 168
 169 static void do_cciss_request(struct request_queue *q);
 170 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 171 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 172 static int cciss_open(struct block_device *bdev, fmode_t mode);
 173 static int cciss_release(struct gendisk *disk, fmode_t mode);
 174 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 175                        unsigned int cmd, unsigned long arg);
 176 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 177
 178 static int cciss_revalidate(struct gendisk *disk);
 179 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 180 static int deregister_disk(ctlr_info_t *h, int drv_index,
 181                            int clear_all, int via_ioctl);
 182
 183 static void cciss_read_capacity(int ctlr, int logvol,
 184                         sector_t *total_size, unsigned int *block_size);
 185 static void cciss_read_capacity_16(int ctlr, int logvol,
 186                         sector_t *total_size, unsigned int *block_size);
 187 static void cciss_geometry_inquiry(int ctlr, int logvol,
 188                         sector_t total_size,
 189                         unsigned int block_size, InquiryData_struct *inq_buff,
 190                                    drive_info_struct *drv);
 191 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 192                                            __u32);
 193 static void start_io(ctlr_info_t *h);
 194 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 195                         __u8 page_code, unsigned char scsi3addr[],
 196                         int cmd_type);
 197 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 198         int attempt_retry);
 199 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 200
 201 static int add_to_scan_list(struct ctlr_info *h);
 202 static int scan_thread(void *data);
 203 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 204 static void cciss_hba_release(struct device *dev);
 205 static void cciss_device_release(struct device *dev);
 206 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 207 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 208
 209 /* performant mode helper functions */
 210 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 211                                 int *bucket_map);
 212 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 213
 214 #ifdef CONFIG_PROC_FS
 215 static void cciss_procinit(int i);
 216 #else
 217 static void cciss_procinit(int i)
 218 {
 219 }
 220 #endif                          /* CONFIG_PROC_FS */
 221
 222 #ifdef CONFIG_COMPAT
 223 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 224                               unsigned, unsigned long);
 225 #endif
 226
 227 static const struct block_device_operations cciss_fops = {
 228         .owner = THIS_MODULE,
 229         .open = cciss_open,
 230         .release = cciss_release,
 231         .locked_ioctl = cciss_ioctl,
 232         .getgeo = cciss_getgeo,
 233 #ifdef CONFIG_COMPAT
 234         .compat_ioctl = cciss_compat_ioctl,
 235 #endif
 236         .revalidate_disk = cciss_revalidate,
 237 };
 238
 239 /* set_performant_mode: Modify the tag for cciss performant
 240  * set bit 0 for pull model, bits 3-1 for block fetch
 241  * register number
 242  */
 243 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 244 {
 245         if (likely(h->transMethod == CFGTBL_Trans_Performant))
 246                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 247 }
 248
 249 /*
 250  * Enqueuing and dequeuing functions for cmdlists.
 251  */
 252 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 253 {
 254         hlist_add_head(&c->list, list);
 255 }
 256
 257 static inline void removeQ(CommandList_struct *c)
 258 {
 259         /*
 260          * After kexec/dump some commands might still
 261          * be in flight, which the firmware will try
 262          * to complete. Resetting the firmware doesn't work
 263          * with old fw revisions, so we have to mark
 264          * them off as 'stale' to prevent the driver from
 265          * falling over.
 266          */
 267         if (WARN_ON(hlist_unhashed(&c->list))) {
 268                 c->cmd_type = CMD_MSG_STALE;
 269                 return;
 270         }
 271
 272         hlist_del_init(&c->list);
 273 }
 274
 275 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 276         CommandList_struct *c)
 277 {
 278         unsigned long flags;
 279         set_performant_mode(h, c);
 280         spin_lock_irqsave(&h->lock, flags);
 281         addQ(&h->reqQ, c);
 282         h->Qdepth++;
 283         start_io(h);
 284         spin_unlock_irqrestore(&h->lock, flags);
 285 }
 286
 287 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 288         int nr_cmds)
 289 {
 290         int i;
 291
 292         if (!cmd_sg_list)
 293                 return;
 294         for (i = 0; i < nr_cmds; i++) {
 295                 kfree(cmd_sg_list[i]);
 296                 cmd_sg_list[i] = NULL;
 297         }
 298         kfree(cmd_sg_list);
 299 }
 300
 301 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 302         ctlr_info_t *h, int chainsize, int nr_cmds)
 303 {
 304         int j;
 305         SGDescriptor_struct **cmd_sg_list;
 306
 307         if (chainsize <= 0)
 308                 return NULL;
 309
 310         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 311         if (!cmd_sg_list)
 312                 return NULL;
 313
 314         /* Build up chain blocks for each command */
 315         for (j = 0; j < nr_cmds; j++) {
 316                 /* Need a block of chainsized s/g elements. */
 317                 cmd_sg_list[j] = kmalloc((chainsize *
 318                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 319                 if (!cmd_sg_list[j]) {
 320                         dev_err(&h->pdev->dev, "Cannot get memory "
 321                                 "for s/g chains.\n");
 322                         goto clean;
 323                 }
 324         }
 325         return cmd_sg_list;
 326 clean:
 327         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 328         return NULL;
 329 }
 330
 331 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 332 {
 333         SGDescriptor_struct *chain_sg;
 334         u64bit temp64;
 335
 336         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 337                 return;
 338
 339         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 340         temp64.val32.lower = chain_sg->Addr.lower;
 341         temp64.val32.upper = chain_sg->Addr.upper;
 342         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 343 }
 344
 345 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 346         SGDescriptor_struct *chain_block, int len)
 347 {
 348         SGDescriptor_struct *chain_sg;
 349         u64bit temp64;
 350
 351         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 352         chain_sg->Ext = CCISS_SG_CHAIN;
 353         chain_sg->Len = len;
 354         temp64.val = pci_map_single(h->pdev, chain_block, len,
 355                                 PCI_DMA_TODEVICE);
 356         chain_sg->Addr.lower = temp64.val32.lower;
 357         chain_sg->Addr.upper = temp64.val32.upper;
 358 }
 359
 360 #include "cciss_scsi.c"         /* For SCSI tape support */
 361
 362 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 363         "UNKNOWN"
 364 };
 365 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
 366
 367 #ifdef CONFIG_PROC_FS
 368
 369 static inline u32 next_command(ctlr_info_t *h)
 370 {
 371         u32 a;
 372
 373         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
 374                 return h->access.command_completed(h);
 375
 376         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
 377                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
 378                 (h->reply_pool_head)++;
 379                 h->commands_outstanding--;
 380         } else {
 381                 a = FIFO_EMPTY;
 382         }
 383         /* Check for wraparound */
 384         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
 385                 h->reply_pool_head = h->reply_pool;
 386                 h->reply_pool_wraparound ^= 1;
 387         }
 388         return a;
 389 }
 390
 391 /*
 392  * Report information about this controller.
 393  */
 394 #define ENG_GIG 1000000000
 395 #define ENG_GIG_FACTOR (ENG_GIG/512)
 396 #define ENGAGE_SCSI     "engage scsi"
 397
 398 static struct proc_dir_entry *proc_cciss;
 399
 400 static void cciss_seq_show_header(struct seq_file *seq)
 401 {
 402         ctlr_info_t *h = seq->private;
 403
 404         seq_printf(seq, "%s: HP %s Controller\n"
 405                 "Board ID: 0x%08lx\n"
 406                 "Firmware Version: %c%c%c%c\n"
 407                 "IRQ: %d\n"
 408                 "Logical drives: %d\n"
 409                 "Current Q depth: %d\n"
 410                 "Current # commands on controller: %d\n"
 411                 "Max Q depth since init: %d\n"
 412                 "Max # commands on controller since init: %d\n"
 413                 "Max SG entries since init: %d\n",
 414                 h->devname,
 415                 h->product_name,
 416                 (unsigned long)h->board_id,
 417                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 418                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 419                 h->num_luns,
 420                 h->Qdepth, h->commands_outstanding,
 421                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 422
 423 #ifdef CONFIG_CISS_SCSI_TAPE
 424         cciss_seq_tape_report(seq, h->ctlr);
 425 #endif /* CONFIG_CISS_SCSI_TAPE */
 426 }
 427
 428 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 429 {
 430         ctlr_info_t *h = seq->private;
 431         unsigned ctlr = h->ctlr;
 432         unsigned long flags;
 433
 434         /* prevent displaying bogus info during configuration
 435          * or deconfiguration of a logical volume
 436          */
 437         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 438         if (h->busy_configuring) {
 439                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 440                 return ERR_PTR(-EBUSY);
 441         }
 442         h->busy_configuring = 1;
 443         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 444
 445         if (*pos == 0)
 446                 cciss_seq_show_header(seq);
 447
 448         return pos;
 449 }
 450
 451 static int cciss_seq_show(struct seq_file *seq, void *v)
 452 {
 453         sector_t vol_sz, vol_sz_frac;
 454         ctlr_info_t *h = seq->private;
 455         unsigned ctlr = h->ctlr;
 456         loff_t *pos = v;
 457         drive_info_struct *drv = h->drv[*pos];
 458
 459         if (*pos > h->highest_lun)
 460                 return 0;
 461
 462         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 463                 return 0;
 464
 465         if (drv->heads == 0)
 466                 return 0;
 467
 468         vol_sz = drv->nr_blocks;
 469         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 470         vol_sz_frac *= 100;
 471         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 472
 473         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 474                 drv->raid_level = RAID_UNKNOWN;
 475         seq_printf(seq, "cciss/c%dd%d:"
 476                         "\t%4u.%02uGB\tRAID %s\n",
 477                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 478                         raid_label[drv->raid_level]);
 479         return 0;
 480 }
 481
 482 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 483 {
 484         ctlr_info_t *h = seq->private;
 485
 486         if (*pos > h->highest_lun)
 487                 return NULL;
 488         *pos += 1;
 489
 490         return pos;
 491 }
 492
 493 static void cciss_seq_stop(struct seq_file *seq, void *v)
 494 {
 495         ctlr_info_t *h = seq->private;
 496
 497         /* Only reset h->busy_configuring if we succeeded in setting
 498          * it during cciss_seq_start. */
 499         if (v == ERR_PTR(-EBUSY))
 500                 return;
 501
 502         h->busy_configuring = 0;
 503 }
 504
 505 static const struct seq_operations cciss_seq_ops = {
 506         .start = cciss_seq_start,
 507         .show  = cciss_seq_show,
 508         .next  = cciss_seq_next,
 509         .stop  = cciss_seq_stop,
 510 };
 511
 512 static int cciss_seq_open(struct inode *inode, struct file *file)
 513 {
 514         int ret = seq_open(file, &cciss_seq_ops);
 515         struct seq_file *seq = file->private_data;
 516
 517         if (!ret)
 518                 seq->private = PDE(inode)->data;
 519
 520         return ret;
 521 }
 522
 523 static ssize_t
 524 cciss_proc_write(struct file *file, const char __user *buf,
 525                  size_t length, loff_t *ppos)
 526 {
 527         int err;
 528         char *buffer;
 529
 530 #ifndef CONFIG_CISS_SCSI_TAPE
 531         return -EINVAL;
 532 #endif
 533
 534         if (!buf || length > PAGE_SIZE - 1)
 535                 return -EINVAL;
 536
 537         buffer = (char *)__get_free_page(GFP_KERNEL);
 538         if (!buffer)
 539                 return -ENOMEM;
 540
 541         err = -EFAULT;
 542         if (copy_from_user(buffer, buf, length))
 543                 goto out;
 544         buffer[length] = '\0';
 545
 546 #ifdef CONFIG_CISS_SCSI_TAPE
 547         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 548                 struct seq_file *seq = file->private_data;
 549                 ctlr_info_t *h = seq->private;
 550
 551                 err = cciss_engage_scsi(h->ctlr);
 552                 if (err == 0)
 553                         err = length;
 554         } else
 555 #endif /* CONFIG_CISS_SCSI_TAPE */
 556                 err = -EINVAL;
 557         /* might be nice to have "disengage" too, but it's not
 558            safely possible. (only 1 module use count, lock issues.) */
 559
 560 out:
 561         free_page((unsigned long)buffer);
 562         return err;
 563 }
 564
 565 static const struct file_operations cciss_proc_fops = {
 566         .owner   = THIS_MODULE,
 567         .open    = cciss_seq_open,
 568         .read    = seq_read,
 569         .llseek  = seq_lseek,
 570         .release = seq_release,
 571         .write   = cciss_proc_write,
 572 };
 573
 574 static void __devinit cciss_procinit(int i)
 575 {
 576         struct proc_dir_entry *pde;
 577
 578         if (proc_cciss == NULL)
 579                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 580         if (!proc_cciss)
 581                 return;
 582         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 583                                         S_IROTH, proc_cciss,
 584                                         &cciss_proc_fops, hba[i]);
 585 }
 586 #endif                          /* CONFIG_PROC_FS */
 587
 588 #define MAX_PRODUCT_NAME_LEN 19
 589
 590 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 591 #define to_drv(n) container_of(n, drive_info_struct, dev)
 592
 593 static ssize_t host_store_rescan(struct device *dev,
 594                                  struct device_attribute *attr,
 595                                  const char *buf, size_t count)
 596 {
 597         struct ctlr_info *h = to_hba(dev);
 598
 599         add_to_scan_list(h);
 600         wake_up_process(cciss_scan_thread);
 601         wait_for_completion_interruptible(&h->scan_wait);
 602
 603         return count;
 604 }
 605 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 606
 607 static ssize_t dev_show_unique_id(struct device *dev,
 608                                  struct device_attribute *attr,
 609                                  char *buf)
 610 {
 611         drive_info_struct *drv = to_drv(dev);
 612         struct ctlr_info *h = to_hba(drv->dev.parent);
 613         __u8 sn[16];
 614         unsigned long flags;
 615         int ret = 0;
 616
 617         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 618         if (h->busy_configuring)
 619                 ret = -EBUSY;
 620         else
 621                 memcpy(sn, drv->serial_no, sizeof(sn));
 622         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 623
 624         if (ret)
 625                 return ret;
 626         else
 627                 return snprintf(buf, 16 * 2 + 2,
 628                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 629                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 630                                 sn[0], sn[1], sn[2], sn[3],
 631                                 sn[4], sn[5], sn[6], sn[7],
 632                                 sn[8], sn[9], sn[10], sn[11],
 633                                 sn[12], sn[13], sn[14], sn[15]);
 634 }
 635 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 636
 637 static ssize_t dev_show_vendor(struct device *dev,
 638                                struct device_attribute *attr,
 639                                char *buf)
 640 {
 641         drive_info_struct *drv = to_drv(dev);
 642         struct ctlr_info *h = to_hba(drv->dev.parent);
 643         char vendor[VENDOR_LEN + 1];
 644         unsigned long flags;
 645         int ret = 0;
 646
 647         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 648         if (h->busy_configuring)
 649                 ret = -EBUSY;
 650         else
 651                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 652         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 653
 654         if (ret)
 655                 return ret;
 656         else
 657                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 658 }
 659 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 660
 661 static ssize_t dev_show_model(struct device *dev,
 662                               struct device_attribute *attr,
 663                               char *buf)
 664 {
 665         drive_info_struct *drv = to_drv(dev);
 666         struct ctlr_info *h = to_hba(drv->dev.parent);
 667         char model[MODEL_LEN + 1];
 668         unsigned long flags;
 669         int ret = 0;
 670
 671         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 672         if (h->busy_configuring)
 673                 ret = -EBUSY;
 674         else
 675                 memcpy(model, drv->model, MODEL_LEN + 1);
 676         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 677
 678         if (ret)
 679                 return ret;
 680         else
 681                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 682 }
 683 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 684
 685 static ssize_t dev_show_rev(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 rev[REV_LEN + 1];
 692         unsigned long flags;
 693         int ret = 0;
 694
 695         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 696         if (h->busy_configuring)
 697                 ret = -EBUSY;
 698         else
 699                 memcpy(rev, drv->rev, REV_LEN + 1);
 700         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 701
 702         if (ret)
 703                 return ret;
 704         else
 705                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 706 }
 707 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 708
 709 static ssize_t cciss_show_lunid(struct device *dev,
 710                                 struct device_attribute *attr, char *buf)
 711 {
 712         drive_info_struct *drv = to_drv(dev);
 713         struct ctlr_info *h = to_hba(drv->dev.parent);
 714         unsigned long flags;
 715         unsigned char lunid[8];
 716
 717         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 718         if (h->busy_configuring) {
 719                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 720                 return -EBUSY;
 721         }
 722         if (!drv->heads) {
 723                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 724                 return -ENOTTY;
 725         }
 726         memcpy(lunid, drv->LunID, sizeof(lunid));
 727         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 728         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 729                 lunid[0], lunid[1], lunid[2], lunid[3],
 730                 lunid[4], lunid[5], lunid[6], lunid[7]);
 731 }
 732 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 733
 734 static ssize_t cciss_show_raid_level(struct device *dev,
 735                                      struct device_attribute *attr, char *buf)
 736 {
 737         drive_info_struct *drv = to_drv(dev);
 738         struct ctlr_info *h = to_hba(drv->dev.parent);
 739         int raid;
 740         unsigned long flags;
 741
 742         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 743         if (h->busy_configuring) {
 744                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 745                 return -EBUSY;
 746         }
 747         raid = drv->raid_level;
 748         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 749         if (raid < 0 || raid > RAID_UNKNOWN)
 750                 raid = RAID_UNKNOWN;
 751
 752         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 753                         raid_label[raid]);
 754 }
 755 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 756
 757 static ssize_t cciss_show_usage_count(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         int count;
 764
 765         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 766         if (h->busy_configuring) {
 767                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 768                 return -EBUSY;
 769         }
 770         count = drv->usage_count;
 771         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 772         return snprintf(buf, 20, "%d\n", count);
 773 }
 774 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 775
 776 static struct attribute *cciss_host_attrs[] = {
 777         &dev_attr_rescan.attr,
 778         NULL
 779 };
 780
 781 static struct attribute_group cciss_host_attr_group = {
 782         .attrs = cciss_host_attrs,
 783 };
 784
 785 static const struct attribute_group *cciss_host_attr_groups[] = {
 786         &cciss_host_attr_group,
 787         NULL
 788 };
 789
 790 static struct device_type cciss_host_type = {
 791         .name           = "cciss_host",
 792         .groups         = cciss_host_attr_groups,
 793         .release        = cciss_hba_release,
 794 };
 795
 796 static struct attribute *cciss_dev_attrs[] = {
 797         &dev_attr_unique_id.attr,
 798         &dev_attr_model.attr,
 799         &dev_attr_vendor.attr,
 800         &dev_attr_rev.attr,
 801         &dev_attr_lunid.attr,
 802         &dev_attr_raid_level.attr,
 803         &dev_attr_usage_count.attr,
 804         NULL
 805 };
 806
 807 static struct attribute_group cciss_dev_attr_group = {
 808         .attrs = cciss_dev_attrs,
 809 };
 810
 811 static const struct attribute_group *cciss_dev_attr_groups[] = {
 812         &cciss_dev_attr_group,
 813         NULL
 814 };
 815
 816 static struct device_type cciss_dev_type = {
 817         .name           = "cciss_device",
 818         .groups         = cciss_dev_attr_groups,
 819         .release        = cciss_device_release,
 820 };
 821
 822 static struct bus_type cciss_bus_type = {
 823         .name           = "cciss",
 824 };
 825
 826 /*
 827  * cciss_hba_release is called when the reference count
 828  * of h->dev goes to zero.
 829  */
 830 static void cciss_hba_release(struct device *dev)
 831 {
 832         /*
 833          * nothing to do, but need this to avoid a warning
 834          * about not having a release handler from lib/kref.c.
 835          */
 836 }
 837
 838 /*
 839  * Initialize sysfs entry for each controller.  This sets up and registers
 840  * the 'cciss#' directory for each individual controller under
 841  * /sys/bus/pci/devices/<dev>/.
 842  */
 843 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 844 {
 845         device_initialize(&h->dev);
 846         h->dev.type = &cciss_host_type;
 847         h->dev.bus = &cciss_bus_type;
 848         dev_set_name(&h->dev, "%s", h->devname);
 849         h->dev.parent = &h->pdev->dev;
 850
 851         return device_add(&h->dev);
 852 }
 853
 854 /*
 855  * Remove sysfs entries for an hba.
 856  */
 857 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 858 {
 859         device_del(&h->dev);
 860         put_device(&h->dev); /* final put. */
 861 }
 862
 863 /* cciss_device_release is called when the reference count
 864  * of h->drv[x]dev goes to zero.
 865  */
 866 static void cciss_device_release(struct device *dev)
 867 {
 868         drive_info_struct *drv = to_drv(dev);
 869         kfree(drv);
 870 }
 871
 872 /*
 873  * Initialize sysfs for each logical drive.  This sets up and registers
 874  * the 'c#d#' directory for each individual logical drive under
 875  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 876  * /sys/block/cciss!c#d# to this entry.
 877  */
 878 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 879                                        int drv_index)
 880 {
 881         struct device *dev;
 882
 883         if (h->drv[drv_index]->device_initialized)
 884                 return 0;
 885
 886         dev = &h->drv[drv_index]->dev;
 887         device_initialize(dev);
 888         dev->type = &cciss_dev_type;
 889         dev->bus = &cciss_bus_type;
 890         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 891         dev->parent = &h->dev;
 892         h->drv[drv_index]->device_initialized = 1;
 893         return device_add(dev);
 894 }
 895
 896 /*
 897  * Remove sysfs entries for a logical drive.
 898  */
 899 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 900         int ctlr_exiting)
 901 {
 902         struct device *dev = &h->drv[drv_index]->dev;
 903
 904         /* special case for c*d0, we only destroy it on controller exit */
 905         if (drv_index == 0 && !ctlr_exiting)
 906                 return;
 907
 908         device_del(dev);
 909         put_device(dev); /* the "final" put. */
 910         h->drv[drv_index] = NULL;
 911 }
 912
 913 /*
 914  * For operations that cannot sleep, a command block is allocated at init,
 915  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 916  * which ones are free or in use.  For operations that can wait for kmalloc
 917  * to possible sleep, this routine can be called with get_from_pool set to 0.
 918  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 919  */
 920 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 921 {
 922         CommandList_struct *c;
 923         int i;
 924         u64bit temp64;
 925         dma_addr_t cmd_dma_handle, err_dma_handle;
 926
 927         if (!get_from_pool) {
 928                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 929                         sizeof(CommandList_struct), &cmd_dma_handle);
 930                 if (c == NULL)
 931                         return NULL;
 932                 memset(c, 0, sizeof(CommandList_struct));
 933
 934                 c->cmdindex = -1;
 935
 936                 c->err_info = (ErrorInfo_struct *)
 937                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 938                             &err_dma_handle);
 939
 940                 if (c->err_info == NULL) {
 941                         pci_free_consistent(h->pdev,
 942                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 943                         return NULL;
 944                 }
 945                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 946         } else {                /* get it out of the controllers pool */
 947
 948                 do {
 949                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 950                         if (i == h->nr_cmds)
 951                                 return NULL;
 952                 } while (test_and_set_bit
 953                          (i & (BITS_PER_LONG - 1),
 954                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 955 #ifdef CCISS_DEBUG
 956                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 957 #endif
 958                 c = h->cmd_pool + i;
 959                 memset(c, 0, sizeof(CommandList_struct));
 960                 cmd_dma_handle = h->cmd_pool_dhandle
 961                     + i * sizeof(CommandList_struct);
 962                 c->err_info = h->errinfo_pool + i;
 963                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 964                 err_dma_handle = h->errinfo_pool_dhandle
 965                     + i * sizeof(ErrorInfo_struct);
 966                 h->nr_allocs++;
 967
 968                 c->cmdindex = i;
 969         }
 970
 971         INIT_HLIST_NODE(&c->list);
 972         c->busaddr = (__u32) cmd_dma_handle;
 973         temp64.val = (__u64) err_dma_handle;
 974         c->ErrDesc.Addr.lower = temp64.val32.lower;
 975         c->ErrDesc.Addr.upper = temp64.val32.upper;
 976         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 977
 978         c->ctlr = h->ctlr;
 979         return c;
 980 }
 981
 982 /*
 983  * Frees a command block that was previously allocated with cmd_alloc().
 984  */
 985 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 986 {
 987         int i;
 988         u64bit temp64;
 989
 990         if (!got_from_pool) {
 991                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 992                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 993                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 994                                     c->err_info, (dma_addr_t) temp64.val);
 995                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 996                                     c, (dma_addr_t) c->busaddr);
 997         } else {
 998                 i = c - h->cmd_pool;
 999                 clear_bit(i & (BITS_PER_LONG - 1),
1000                           h->cmd_pool_bits + (i / BITS_PER_LONG));
1001                 h->nr_frees++;
1002         }
1003 }
1004
1005 static inline ctlr_info_t *get_host(struct gendisk *disk)
1006 {
1007         return disk->queue->queuedata;
1008 }
1009
1010 static inline drive_info_struct *get_drv(struct gendisk *disk)
1011 {
1012         return disk->private_data;
1013 }
1014
1015 /*
1016  * Open.  Make sure the device is really there.
1017  */
1018 static int cciss_open(struct block_device *bdev, fmode_t mode)
1019 {
1020         ctlr_info_t *host = get_host(bdev->bd_disk);
1021         drive_info_struct *drv = get_drv(bdev->bd_disk);
1022
1023 #ifdef CCISS_DEBUG
1024         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
1025 #endif                          /* CCISS_DEBUG */
1026
1027         if (drv->busy_configuring)
1028                 return -EBUSY;
1029         /*
1030          * Root is allowed to open raw volume zero even if it's not configured
1031          * so array config can still work. Root is also allowed to open any
1032          * volume that has a LUN ID, so it can issue IOCTL to reread the
1033          * disk information.  I don't think I really like this
1034          * but I'm already using way to many device nodes to claim another one
1035          * for "raw controller".
1036          */
1037         if (drv->heads == 0) {
1038                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1039                         /* if not node 0 make sure it is a partition = 0 */
1040                         if (MINOR(bdev->bd_dev) & 0x0f) {
1041                                 return -ENXIO;
1042                                 /* if it is, make sure we have a LUN ID */
1043                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1044                                 sizeof(drv->LunID))) {
1045                                 return -ENXIO;
1046                         }
1047                 }
1048                 if (!capable(CAP_SYS_ADMIN))
1049                         return -EPERM;
1050         }
1051         drv->usage_count++;
1052         host->usage_count++;
1053         return 0;
1054 }
1055
1056 /*
1057  * Close.  Sync first.
1058  */
1059 static int cciss_release(struct gendisk *disk, fmode_t mode)
1060 {
1061         ctlr_info_t *host = get_host(disk);
1062         drive_info_struct *drv = get_drv(disk);
1063
1064 #ifdef CCISS_DEBUG
1065         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
1066 #endif                          /* CCISS_DEBUG */
1067
1068         drv->usage_count--;
1069         host->usage_count--;
1070         return 0;
1071 }
1072
1073 #ifdef CONFIG_COMPAT
1074
1075 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1076                     unsigned cmd, unsigned long arg)
1077 {
1078         int ret;
1079         lock_kernel();
1080         ret = cciss_ioctl(bdev, mode, cmd, arg);
1081         unlock_kernel();
1082         return ret;
1083 }
1084
1085 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1086                                   unsigned cmd, unsigned long arg);
1087 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1088                                       unsigned cmd, unsigned long arg);
1089
1090 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1091                               unsigned cmd, unsigned long arg)
1092 {
1093         switch (cmd) {
1094         case CCISS_GETPCIINFO:
1095         case CCISS_GETINTINFO:
1096         case CCISS_SETINTINFO:
1097         case CCISS_GETNODENAME:
1098         case CCISS_SETNODENAME:
1099         case CCISS_GETHEARTBEAT:
1100         case CCISS_GETBUSTYPES:
1101         case CCISS_GETFIRMVER:
1102         case CCISS_GETDRIVVER:
1103         case CCISS_REVALIDVOLS:
1104         case CCISS_DEREGDISK:
1105         case CCISS_REGNEWDISK:
1106         case CCISS_REGNEWD:
1107         case CCISS_RESCANDISK:
1108         case CCISS_GETLUNINFO:
1109                 return do_ioctl(bdev, mode, cmd, arg);
1110
1111         case CCISS_PASSTHRU32:
1112                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1113         case CCISS_BIG_PASSTHRU32:
1114                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1115
1116         default:
1117                 return -ENOIOCTLCMD;
1118         }
1119 }
1120
1121 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1122                                   unsigned cmd, unsigned long arg)
1123 {
1124         IOCTL32_Command_struct __user *arg32 =
1125             (IOCTL32_Command_struct __user *) arg;
1126         IOCTL_Command_struct arg64;
1127         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1128         int err;
1129         u32 cp;
1130
1131         err = 0;
1132         err |=
1133             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1134                            sizeof(arg64.LUN_info));
1135         err |=
1136             copy_from_user(&arg64.Request, &arg32->Request,
1137                            sizeof(arg64.Request));
1138         err |=
1139             copy_from_user(&arg64.error_info, &arg32->error_info,
1140                            sizeof(arg64.error_info));
1141         err |= get_user(arg64.buf_size, &arg32->buf_size);
1142         err |= get_user(cp, &arg32->buf);
1143         arg64.buf = compat_ptr(cp);
1144         err |= copy_to_user(p, &arg64, sizeof(arg64));
1145
1146         if (err)
1147                 return -EFAULT;
1148
1149         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1150         if (err)
1151                 return err;
1152         err |=
1153             copy_in_user(&arg32->error_info, &p->error_info,
1154                          sizeof(arg32->error_info));
1155         if (err)
1156                 return -EFAULT;
1157         return err;
1158 }
1159
1160 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1161                                       unsigned cmd, unsigned long arg)
1162 {
1163         BIG_IOCTL32_Command_struct __user *arg32 =
1164             (BIG_IOCTL32_Command_struct __user *) arg;
1165         BIG_IOCTL_Command_struct arg64;
1166         BIG_IOCTL_Command_struct __user *p =
1167             compat_alloc_user_space(sizeof(arg64));
1168         int err;
1169         u32 cp;
1170
1171         err = 0;
1172         err |=
1173             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1174                            sizeof(arg64.LUN_info));
1175         err |=
1176             copy_from_user(&arg64.Request, &arg32->Request,
1177                            sizeof(arg64.Request));
1178         err |=
1179             copy_from_user(&arg64.error_info, &arg32->error_info,
1180                            sizeof(arg64.error_info));
1181         err |= get_user(arg64.buf_size, &arg32->buf_size);
1182         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1183         err |= get_user(cp, &arg32->buf);
1184         arg64.buf = compat_ptr(cp);
1185         err |= copy_to_user(p, &arg64, sizeof(arg64));
1186
1187         if (err)
1188                 return -EFAULT;
1189
1190         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1191         if (err)
1192                 return err;
1193         err |=
1194             copy_in_user(&arg32->error_info, &p->error_info,
1195                          sizeof(arg32->error_info));
1196         if (err)
1197                 return -EFAULT;
1198         return err;
1199 }
1200 #endif
1201
1202 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1203 {
1204         drive_info_struct *drv = get_drv(bdev->bd_disk);
1205
1206         if (!drv->cylinders)
1207                 return -ENXIO;
1208
1209         geo->heads = drv->heads;
1210         geo->sectors = drv->sectors;
1211         geo->cylinders = drv->cylinders;
1212         return 0;
1213 }
1214
1215 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1216 {
1217         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1218                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1219                 (void)check_for_unit_attention(host, c);
1220 }
1221 /*
1222  * ioctl
1223  */
1224 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1225                        unsigned int cmd, unsigned long arg)
1226 {
1227         struct gendisk *disk = bdev->bd_disk;
1228         ctlr_info_t *host = get_host(disk);
1229         drive_info_struct *drv = get_drv(disk);
1230         int ctlr = host->ctlr;
1231         void __user *argp = (void __user *)arg;
1232
1233 #ifdef CCISS_DEBUG
1234         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1235 #endif                          /* CCISS_DEBUG */
1236
1237         switch (cmd) {
1238         case CCISS_GETPCIINFO:
1239                 {
1240                         cciss_pci_info_struct pciinfo;
1241
1242                         if (!arg)
1243                                 return -EINVAL;
1244                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1245                         pciinfo.bus = host->pdev->bus->number;
1246                         pciinfo.dev_fn = host->pdev->devfn;
1247                         pciinfo.board_id = host->board_id;
1248                         if (copy_to_user
1249                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1250                                 return -EFAULT;
1251                         return 0;
1252                 }
1253         case CCISS_GETINTINFO:
1254                 {
1255                         cciss_coalint_struct intinfo;
1256                         if (!arg)
1257                                 return -EINVAL;
1258                         intinfo.delay =
1259                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1260                         intinfo.count =
1261                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1262                         if (copy_to_user
1263                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1264                                 return -EFAULT;
1265                         return 0;
1266                 }
1267         case CCISS_SETINTINFO:
1268                 {
1269                         cciss_coalint_struct intinfo;
1270                         unsigned long flags;
1271                         int i;
1272
1273                         if (!arg)
1274                                 return -EINVAL;
1275                         if (!capable(CAP_SYS_ADMIN))
1276                                 return -EPERM;
1277                         if (copy_from_user
1278                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1279                                 return -EFAULT;
1280                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1281                         {
1282 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1283                                 return -EINVAL;
1284                         }
1285                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1286                         /* Update the field, and then ring the doorbell */
1287                         writel(intinfo.delay,
1288                                &(host->cfgtable->HostWrite.CoalIntDelay));
1289                         writel(intinfo.count,
1290                                &(host->cfgtable->HostWrite.CoalIntCount));
1291                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1292
1293                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1294                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1295                                       & CFGTBL_ChangeReq))
1296                                         break;
1297                                 /* delay and try again */
1298                                 udelay(1000);
1299                         }
1300                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1301                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1302                                 return -EAGAIN;
1303                         return 0;
1304                 }
1305         case CCISS_GETNODENAME:
1306                 {
1307                         NodeName_type NodeName;
1308                         int i;
1309
1310                         if (!arg)
1311                                 return -EINVAL;
1312                         for (i = 0; i < 16; i++)
1313                                 NodeName[i] =
1314                                     readb(&host->cfgtable->ServerName[i]);
1315                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1316                                 return -EFAULT;
1317                         return 0;
1318                 }
1319         case CCISS_SETNODENAME:
1320                 {
1321                         NodeName_type NodeName;
1322                         unsigned long flags;
1323                         int i;
1324
1325                         if (!arg)
1326                                 return -EINVAL;
1327                         if (!capable(CAP_SYS_ADMIN))
1328                                 return -EPERM;
1329
1330                         if (copy_from_user
1331                             (NodeName, argp, sizeof(NodeName_type)))
1332                                 return -EFAULT;
1333
1334                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1335
1336                         /* Update the field, and then ring the doorbell */
1337                         for (i = 0; i < 16; i++)
1338                                 writeb(NodeName[i],
1339                                        &host->cfgtable->ServerName[i]);
1340
1341                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1342
1343                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1344                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1345                                       & CFGTBL_ChangeReq))
1346                                         break;
1347                                 /* delay and try again */
1348                                 udelay(1000);
1349                         }
1350                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1351                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1352                                 return -EAGAIN;
1353                         return 0;
1354                 }
1355
1356         case CCISS_GETHEARTBEAT:
1357                 {
1358                         Heartbeat_type heartbeat;
1359
1360                         if (!arg)
1361                                 return -EINVAL;
1362                         heartbeat = readl(&host->cfgtable->HeartBeat);
1363                         if (copy_to_user
1364                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1365                                 return -EFAULT;
1366                         return 0;
1367                 }
1368         case CCISS_GETBUSTYPES:
1369                 {
1370                         BusTypes_type BusTypes;
1371
1372                         if (!arg)
1373                                 return -EINVAL;
1374                         BusTypes = readl(&host->cfgtable->BusTypes);
1375                         if (copy_to_user
1376                             (argp, &BusTypes, sizeof(BusTypes_type)))
1377                                 return -EFAULT;
1378                         return 0;
1379                 }
1380         case CCISS_GETFIRMVER:
1381                 {
1382                         FirmwareVer_type firmware;
1383
1384                         if (!arg)
1385                                 return -EINVAL;
1386                         memcpy(firmware, host->firm_ver, 4);
1387
1388                         if (copy_to_user
1389                             (argp, firmware, sizeof(FirmwareVer_type)))
1390                                 return -EFAULT;
1391                         return 0;
1392                 }
1393         case CCISS_GETDRIVVER:
1394                 {
1395                         DriverVer_type DriverVer = DRIVER_VERSION;
1396
1397                         if (!arg)
1398                                 return -EINVAL;
1399
1400                         if (copy_to_user
1401                             (argp, &DriverVer, sizeof(DriverVer_type)))
1402                                 return -EFAULT;
1403                         return 0;
1404                 }
1405
1406         case CCISS_DEREGDISK:
1407         case CCISS_REGNEWD:
1408         case CCISS_REVALIDVOLS:
1409                 return rebuild_lun_table(host, 0, 1);
1410
1411         case CCISS_GETLUNINFO:{
1412                         LogvolInfo_struct luninfo;
1413
1414                         memcpy(&luninfo.LunID, drv->LunID,
1415                                 sizeof(luninfo.LunID));
1416                         luninfo.num_opens = drv->usage_count;
1417                         luninfo.num_parts = 0;
1418                         if (copy_to_user(argp, &luninfo,
1419                                          sizeof(LogvolInfo_struct)))
1420                                 return -EFAULT;
1421                         return 0;
1422                 }
1423         case CCISS_PASSTHRU:
1424                 {
1425                         IOCTL_Command_struct iocommand;
1426                         CommandList_struct *c;
1427                         char *buff = NULL;
1428                         u64bit temp64;
1429                         DECLARE_COMPLETION_ONSTACK(wait);
1430
1431                         if (!arg)
1432                                 return -EINVAL;
1433
1434                         if (!capable(CAP_SYS_RAWIO))
1435                                 return -EPERM;
1436
1437                         if (copy_from_user
1438                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1439                                 return -EFAULT;
1440                         if ((iocommand.buf_size < 1) &&
1441                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1442                                 return -EINVAL;
1443                         }
1444 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1445                         /* Check kmalloc limits */
1446                         if (iocommand.buf_size > 128000)
1447                                 return -EINVAL;
1448 #endif
1449                         if (iocommand.buf_size > 0) {
1450                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1451                                 if (buff == NULL)
1452                                         return -EFAULT;
1453                         }
1454                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1455                                 /* Copy the data into the buffer we created */
1456                                 if (copy_from_user
1457                                     (buff, iocommand.buf, iocommand.buf_size)) {
1458                                         kfree(buff);
1459                                         return -EFAULT;
1460                                 }
1461                         } else {
1462                                 memset(buff, 0, iocommand.buf_size);
1463                         }
1464                         if ((c = cmd_alloc(host, 0)) == NULL) {
1465                                 kfree(buff);
1466                                 return -ENOMEM;
1467                         }
1468                         /* Fill in the command type */
1469                         c->cmd_type = CMD_IOCTL_PEND;
1470                         /* Fill in Command Header */
1471                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1472                         if (iocommand.buf_size > 0) /* buffer to fill */
1473                         {
1474                                 c->Header.SGList = 1;
1475                                 c->Header.SGTotal = 1;
1476                         } else /* no buffers to fill */
1477                         {
1478                                 c->Header.SGList = 0;
1479                                 c->Header.SGTotal = 0;
1480                         }
1481                         c->Header.LUN = iocommand.LUN_info;
1482                         /* use the kernel address the cmd block for tag */
1483                         c->Header.Tag.lower = c->busaddr;
1484
1485                         /* Fill in Request block */
1486                         c->Request = iocommand.Request;
1487
1488                         /* Fill in the scatter gather information */
1489                         if (iocommand.buf_size > 0) {
1490                                 temp64.val = pci_map_single(host->pdev, buff,
1491                                         iocommand.buf_size,
1492                                         PCI_DMA_BIDIRECTIONAL);
1493                                 c->SG[0].Addr.lower = temp64.val32.lower;
1494                                 c->SG[0].Addr.upper = temp64.val32.upper;
1495                                 c->SG[0].Len = iocommand.buf_size;
1496                                 c->SG[0].Ext = 0;  /* we are not chaining */
1497                         }
1498                         c->waiting = &wait;
1499
1500                         enqueue_cmd_and_start_io(host, c);
1501                         wait_for_completion(&wait);
1502
1503                         /* unlock the buffers from DMA */
1504                         temp64.val32.lower = c->SG[0].Addr.lower;
1505                         temp64.val32.upper = c->SG[0].Addr.upper;
1506                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1507                                          iocommand.buf_size,
1508                                          PCI_DMA_BIDIRECTIONAL);
1509
1510                         check_ioctl_unit_attention(host, c);
1511
1512                         /* Copy the error information out */
1513                         iocommand.error_info = *(c->err_info);
1514                         if (copy_to_user
1515                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1516                                 kfree(buff);
1517                                 cmd_free(host, c, 0);
1518                                 return -EFAULT;
1519                         }
1520
1521                         if (iocommand.Request.Type.Direction == XFER_READ) {
1522                                 /* Copy the data out of the buffer we created */
1523                                 if (copy_to_user
1524                                     (iocommand.buf, buff, iocommand.buf_size)) {
1525                                         kfree(buff);
1526                                         cmd_free(host, c, 0);
1527                                         return -EFAULT;
1528                                 }
1529                         }
1530                         kfree(buff);
1531                         cmd_free(host, c, 0);
1532                         return 0;
1533                 }
1534         case CCISS_BIG_PASSTHRU:{
1535                         BIG_IOCTL_Command_struct *ioc;
1536                         CommandList_struct *c;
1537                         unsigned char **buff = NULL;
1538                         int *buff_size = NULL;
1539                         u64bit temp64;
1540                         BYTE sg_used = 0;
1541                         int status = 0;
1542                         int i;
1543                         DECLARE_COMPLETION_ONSTACK(wait);
1544                         __u32 left;
1545                         __u32 sz;
1546                         BYTE __user *data_ptr;
1547
1548                         if (!arg)
1549                                 return -EINVAL;
1550                         if (!capable(CAP_SYS_RAWIO))
1551                                 return -EPERM;
1552                         ioc = (BIG_IOCTL_Command_struct *)
1553                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1554                         if (!ioc) {
1555                                 status = -ENOMEM;
1556                                 goto cleanup1;
1557                         }
1558                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1559                                 status = -EFAULT;
1560                                 goto cleanup1;
1561                         }
1562                         if ((ioc->buf_size < 1) &&
1563                             (ioc->Request.Type.Direction != XFER_NONE)) {
1564                                 status = -EINVAL;
1565                                 goto cleanup1;
1566                         }
1567                         /* Check kmalloc limits  using all SGs */
1568                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1569                                 status = -EINVAL;
1570                                 goto cleanup1;
1571                         }
1572                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1573                                 status = -EINVAL;
1574                                 goto cleanup1;
1575                         }
1576                         buff =
1577                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1578                         if (!buff) {
1579                                 status = -ENOMEM;
1580                                 goto cleanup1;
1581                         }
1582                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1583                                                    GFP_KERNEL);
1584                         if (!buff_size) {
1585                                 status = -ENOMEM;
1586                                 goto cleanup1;
1587                         }
1588                         left = ioc->buf_size;
1589                         data_ptr = ioc->buf;
1590                         while (left) {
1591                                 sz = (left >
1592                                       ioc->malloc_size) ? ioc->
1593                                     malloc_size : left;
1594                                 buff_size[sg_used] = sz;
1595                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1596                                 if (buff[sg_used] == NULL) {
1597                                         status = -ENOMEM;
1598                                         goto cleanup1;
1599                                 }
1600                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1601                                         if (copy_from_user
1602                                             (buff[sg_used], data_ptr, sz)) {
1603                                                 status = -EFAULT;
1604                                                 goto cleanup1;
1605                                         }
1606                                 } else {
1607                                         memset(buff[sg_used], 0, sz);
1608                                 }
1609                                 left -= sz;
1610                                 data_ptr += sz;
1611                                 sg_used++;
1612                         }
1613                         if ((c = cmd_alloc(host, 0)) == NULL) {
1614                                 status = -ENOMEM;
1615                                 goto cleanup1;
1616                         }
1617                         c->cmd_type = CMD_IOCTL_PEND;
1618                         c->Header.ReplyQueue = 0;
1619
1620                         if (ioc->buf_size > 0) {
1621                                 c->Header.SGList = sg_used;
1622                                 c->Header.SGTotal = sg_used;
1623                         } else {
1624                                 c->Header.SGList = 0;
1625                                 c->Header.SGTotal = 0;
1626                         }
1627                         c->Header.LUN = ioc->LUN_info;
1628                         c->Header.Tag.lower = c->busaddr;
1629
1630                         c->Request = ioc->Request;
1631                         if (ioc->buf_size > 0) {
1632                                 for (i = 0; i < sg_used; i++) {
1633                                         temp64.val =
1634                                             pci_map_single(host->pdev, buff[i],
1635                                                     buff_size[i],
1636                                                     PCI_DMA_BIDIRECTIONAL);
1637                                         c->SG[i].Addr.lower =
1638                                             temp64.val32.lower;
1639                                         c->SG[i].Addr.upper =
1640                                             temp64.val32.upper;
1641                                         c->SG[i].Len = buff_size[i];
1642                                         c->SG[i].Ext = 0;       /* we are not chaining */
1643                                 }
1644                         }
1645                         c->waiting = &wait;
1646                         enqueue_cmd_and_start_io(host, c);
1647                         wait_for_completion(&wait);
1648                         /* unlock the buffers from DMA */
1649                         for (i = 0; i < sg_used; i++) {
1650                                 temp64.val32.lower = c->SG[i].Addr.lower;
1651                                 temp64.val32.upper = c->SG[i].Addr.upper;
1652                                 pci_unmap_single(host->pdev,
1653                                         (dma_addr_t) temp64.val, buff_size[i],
1654                                         PCI_DMA_BIDIRECTIONAL);
1655                         }
1656                         check_ioctl_unit_attention(host, c);
1657                         /* Copy the error information out */
1658                         ioc->error_info = *(c->err_info);
1659                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1660                                 cmd_free(host, c, 0);
1661                                 status = -EFAULT;
1662                                 goto cleanup1;
1663                         }
1664                         if (ioc->Request.Type.Direction == XFER_READ) {
1665                                 /* Copy the data out of the buffer we created */
1666                                 BYTE __user *ptr = ioc->buf;
1667                                 for (i = 0; i < sg_used; i++) {
1668                                         if (copy_to_user
1669                                             (ptr, buff[i], buff_size[i])) {
1670                                                 cmd_free(host, c, 0);
1671                                                 status = -EFAULT;
1672                                                 goto cleanup1;
1673                                         }
1674                                         ptr += buff_size[i];
1675                                 }
1676                         }
1677                         cmd_free(host, c, 0);
1678                         status = 0;
1679                       cleanup1:
1680                         if (buff) {
1681                                 for (i = 0; i < sg_used; i++)
1682                                         kfree(buff[i]);
1683                                 kfree(buff);
1684                         }
1685                         kfree(buff_size);
1686                         kfree(ioc);
1687                         return status;
1688                 }
1689
1690         /* scsi_cmd_ioctl handles these, below, though some are not */
1691         /* very meaningful for cciss.  SG_IO is the main one people want. */
1692
1693         case SG_GET_VERSION_NUM:
1694         case SG_SET_TIMEOUT:
1695         case SG_GET_TIMEOUT:
1696         case SG_GET_RESERVED_SIZE:
1697         case SG_SET_RESERVED_SIZE:
1698         case SG_EMULATED_HOST:
1699         case SG_IO:
1700         case SCSI_IOCTL_SEND_COMMAND:
1701                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1702
1703         /* scsi_cmd_ioctl would normally handle these, below, but */
1704         /* they aren't a good fit for cciss, as CD-ROMs are */
1705         /* not supported, and we don't have any bus/target/lun */
1706         /* which we present to the kernel. */
1707
1708         case CDROM_SEND_PACKET:
1709         case CDROMCLOSETRAY:
1710         case CDROMEJECT:
1711         case SCSI_IOCTL_GET_IDLUN:
1712         case SCSI_IOCTL_GET_BUS_NUMBER:
1713         default:
1714                 return -ENOTTY;
1715         }
1716 }
1717
1718 static void cciss_check_queues(ctlr_info_t *h)
1719 {
1720         int start_queue = h->next_to_run;
1721         int i;
1722
1723         /* check to see if we have maxed out the number of commands that can
1724          * be placed on the queue.  If so then exit.  We do this check here
1725          * in case the interrupt we serviced was from an ioctl and did not
1726          * free any new commands.
1727          */
1728         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1729                 return;
1730
1731         /* We have room on the queue for more commands.  Now we need to queue
1732          * them up.  We will also keep track of the next queue to run so
1733          * that every queue gets a chance to be started first.
1734          */
1735         for (i = 0; i < h->highest_lun + 1; i++) {
1736                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1737                 /* make sure the disk has been added and the drive is real
1738                  * because this can be called from the middle of init_one.
1739                  */
1740                 if (!h->drv[curr_queue])
1741                         continue;
1742                 if (!(h->drv[curr_queue]->queue) ||
1743                         !(h->drv[curr_queue]->heads))
1744                         continue;
1745                 blk_start_queue(h->gendisk[curr_queue]->queue);
1746
1747                 /* check to see if we have maxed out the number of commands
1748                  * that can be placed on the queue.
1749                  */
1750                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1751                         if (curr_queue == start_queue) {
1752                                 h->next_to_run =
1753                                     (start_queue + 1) % (h->highest_lun + 1);
1754                                 break;
1755                         } else {
1756                                 h->next_to_run = curr_queue;
1757                                 break;
1758                         }
1759                 }
1760         }
1761 }
1762
1763 static void cciss_softirq_done(struct request *rq)
1764 {
1765         CommandList_struct *cmd = rq->completion_data;
1766         ctlr_info_t *h = hba[cmd->ctlr];
1767         SGDescriptor_struct *curr_sg = cmd->SG;
1768         u64bit temp64;
1769         unsigned long flags;
1770         int i, ddir;
1771         int sg_index = 0;
1772
1773         if (cmd->Request.Type.Direction == XFER_READ)
1774                 ddir = PCI_DMA_FROMDEVICE;
1775         else
1776                 ddir = PCI_DMA_TODEVICE;
1777
1778         /* command did not need to be retried */
1779         /* unmap the DMA mapping for all the scatter gather elements */
1780         for (i = 0; i < cmd->Header.SGList; i++) {
1781                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1782                         cciss_unmap_sg_chain_block(h, cmd);
1783                         /* Point to the next block */
1784                         curr_sg = h->cmd_sg_list[cmd->cmdindex];
1785                         sg_index = 0;
1786                 }
1787                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1788                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1789                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1790                                 ddir);
1791                 ++sg_index;
1792         }
1793
1794 #ifdef CCISS_DEBUG
1795         printk("Done with %p\n", rq);
1796 #endif                          /* CCISS_DEBUG */
1797
1798         /* set the residual count for pc requests */
1799         if (blk_pc_request(rq))
1800                 rq->resid_len = cmd->err_info->ResidualCnt;
1801
1802         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1803
1804         spin_lock_irqsave(&h->lock, flags);
1805         cmd_free(h, cmd, 1);
1806         cciss_check_queues(h);
1807         spin_unlock_irqrestore(&h->lock, flags);
1808 }
1809
1810 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1811         unsigned char scsi3addr[], uint32_t log_unit)
1812 {
1813         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1814                 sizeof(h->drv[log_unit]->LunID));
1815 }
1816
1817 /* This function gets the SCSI vendor, model, and revision of a logical drive
1818  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1819  * they cannot be read.
1820  */
1821 static void cciss_get_device_descr(int ctlr, int logvol,
1822                                    char *vendor, char *model, char *rev)
1823 {
1824         int rc;
1825         InquiryData_struct *inq_buf;
1826         unsigned char scsi3addr[8];
1827
1828         *vendor = '\0';
1829         *model = '\0';
1830         *rev = '\0';
1831
1832         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1833         if (!inq_buf)
1834                 return;
1835
1836         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1837         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1838                         scsi3addr, TYPE_CMD);
1839         if (rc == IO_OK) {
1840                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1841                 vendor[VENDOR_LEN] = '\0';
1842                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1843                 model[MODEL_LEN] = '\0';
1844                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1845                 rev[REV_LEN] = '\0';
1846         }
1847
1848         kfree(inq_buf);
1849         return;
1850 }
1851
1852 /* This function gets the serial number of a logical drive via
1853  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1854  * number cannot be had, for whatever reason, 16 bytes of 0xff
1855  * are returned instead.
1856  */
1857 static void cciss_get_serial_no(int ctlr, int logvol,
1858                                 unsigned char *serial_no, int buflen)
1859 {
1860 #define PAGE_83_INQ_BYTES 64
1861         int rc;
1862         unsigned char *buf;
1863         unsigned char scsi3addr[8];
1864
1865         if (buflen > 16)
1866                 buflen = 16;
1867         memset(serial_no, 0xff, buflen);
1868         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1869         if (!buf)
1870                 return;
1871         memset(serial_no, 0, buflen);
1872         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1873         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1874                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1875         if (rc == IO_OK)
1876                 memcpy(serial_no, &buf[8], buflen);
1877         kfree(buf);
1878         return;
1879 }
1880
1881 /*
1882  * cciss_add_disk sets up the block device queue for a logical drive
1883  */
1884 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1885                                 int drv_index)
1886 {
1887         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1888         if (!disk->queue)
1889                 goto init_queue_failure;
1890         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1891         disk->major = h->major;
1892         disk->first_minor = drv_index << NWD_SHIFT;
1893         disk->fops = &cciss_fops;
1894         if (cciss_create_ld_sysfs_entry(h, drv_index))
1895                 goto cleanup_queue;
1896         disk->private_data = h->drv[drv_index];
1897         disk->driverfs_dev = &h->drv[drv_index]->dev;
1898
1899         /* Set up queue information */
1900         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1901
1902         /* This is a hardware imposed limit. */
1903         blk_queue_max_segments(disk->queue, h->maxsgentries);
1904
1905         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1906
1907         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1908
1909         disk->queue->queuedata = h;
1910
1911         blk_queue_logical_block_size(disk->queue,
1912                                      h->drv[drv_index]->block_size);
1913
1914         /* Make sure all queue data is written out before */
1915         /* setting h->drv[drv_index]->queue, as setting this */
1916         /* allows the interrupt handler to start the queue */
1917         wmb();
1918         h->drv[drv_index]->queue = disk->queue;
1919         add_disk(disk);
1920         return 0;
1921
1922 cleanup_queue:
1923         blk_cleanup_queue(disk->queue);
1924         disk->queue = NULL;
1925 init_queue_failure:
1926         return -1;
1927 }
1928
1929 /* This function will check the usage_count of the drive to be updated/added.
1930  * If the usage_count is zero and it is a heretofore unknown drive, or,
1931  * the drive's capacity, geometry, or serial number has changed,
1932  * then the drive information will be updated and the disk will be
1933  * re-registered with the kernel.  If these conditions don't hold,
1934  * then it will be left alone for the next reboot.  The exception to this
1935  * is disk 0 which will always be left registered with the kernel since it
1936  * is also the controller node.  Any changes to disk 0 will show up on
1937  * the next reboot.
1938  */
1939 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1940         int via_ioctl)
1941 {
1942         ctlr_info_t *h = hba[ctlr];
1943         struct gendisk *disk;
1944         InquiryData_struct *inq_buff = NULL;
1945         unsigned int block_size;
1946         sector_t total_size;
1947         unsigned long flags = 0;
1948         int ret = 0;
1949         drive_info_struct *drvinfo;
1950
1951         /* Get information about the disk and modify the driver structure */
1952         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1953         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1954         if (inq_buff == NULL || drvinfo == NULL)
1955                 goto mem_msg;
1956
1957         /* testing to see if 16-byte CDBs are already being used */
1958         if (h->cciss_read == CCISS_READ_16) {
1959                 cciss_read_capacity_16(h->ctlr, drv_index,
1960                         &total_size, &block_size);
1961
1962         } else {
1963                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1964                 /* if read_capacity returns all F's this volume is >2TB */
1965                 /* in size so we switch to 16-byte CDB's for all */
1966                 /* read/write ops */
1967                 if (total_size == 0xFFFFFFFFULL) {
1968                         cciss_read_capacity_16(ctlr, drv_index,
1969                         &total_size, &block_size);
1970                         h->cciss_read = CCISS_READ_16;
1971                         h->cciss_write = CCISS_WRITE_16;
1972                 } else {
1973                         h->cciss_read = CCISS_READ_10;
1974                         h->cciss_write = CCISS_WRITE_10;
1975                 }
1976         }
1977
1978         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1979                                inq_buff, drvinfo);
1980         drvinfo->block_size = block_size;
1981         drvinfo->nr_blocks = total_size + 1;
1982
1983         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1984                                 drvinfo->model, drvinfo->rev);
1985         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1986                         sizeof(drvinfo->serial_no));
1987         /* Save the lunid in case we deregister the disk, below. */
1988         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1989                 sizeof(drvinfo->LunID));
1990
1991         /* Is it the same disk we already know, and nothing's changed? */
1992         if (h->drv[drv_index]->raid_level != -1 &&
1993                 ((memcmp(drvinfo->serial_no,
1994                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1995                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1996                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1997                 drvinfo->heads == h->drv[drv_index]->heads &&
1998                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1999                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2000                         /* The disk is unchanged, nothing to update */
2001                         goto freeret;
2002
2003         /* If we get here it's not the same disk, or something's changed,
2004          * so we need to * deregister it, and re-register it, if it's not
2005          * in use.
2006          * If the disk already exists then deregister it before proceeding
2007          * (unless it's the first disk (for the controller node).
2008          */
2009         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2010                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
2011                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2012                 h->drv[drv_index]->busy_configuring = 1;
2013                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2014
2015                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2016                  * which keeps the interrupt handler from starting
2017                  * the queue.
2018                  */
2019                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2020         }
2021
2022         /* If the disk is in use return */
2023         if (ret)
2024                 goto freeret;
2025
2026         /* Save the new information from cciss_geometry_inquiry
2027          * and serial number inquiry.  If the disk was deregistered
2028          * above, then h->drv[drv_index] will be NULL.
2029          */
2030         if (h->drv[drv_index] == NULL) {
2031                 drvinfo->device_initialized = 0;
2032                 h->drv[drv_index] = drvinfo;
2033                 drvinfo = NULL; /* so it won't be freed below. */
2034         } else {
2035                 /* special case for cxd0 */
2036                 h->drv[drv_index]->block_size = drvinfo->block_size;
2037                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2038                 h->drv[drv_index]->heads = drvinfo->heads;
2039                 h->drv[drv_index]->sectors = drvinfo->sectors;
2040                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2041                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2042                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2043                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2044                         VENDOR_LEN + 1);
2045                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2046                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2047         }
2048
2049         ++h->num_luns;
2050         disk = h->gendisk[drv_index];
2051         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2052
2053         /* If it's not disk 0 (drv_index != 0)
2054          * or if it was disk 0, but there was previously
2055          * no actual corresponding configured logical drive
2056          * (raid_leve == -1) then we want to update the
2057          * logical drive's information.
2058          */
2059         if (drv_index || first_time) {
2060                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2061                         cciss_free_gendisk(h, drv_index);
2062                         cciss_free_drive_info(h, drv_index);
2063                         printk(KERN_WARNING "cciss:%d could not update "
2064                                 "disk %d\n", h->ctlr, drv_index);
2065                         --h->num_luns;
2066                 }
2067         }
2068
2069 freeret:
2070         kfree(inq_buff);
2071         kfree(drvinfo);
2072         return;
2073 mem_msg:
2074         printk(KERN_ERR "cciss: out of memory\n");
2075         goto freeret;
2076 }
2077
2078 /* This function will find the first index of the controllers drive array
2079  * that has a null drv pointer and allocate the drive info struct and
2080  * will return that index   This is where new drives will be added.
2081  * If the index to be returned is greater than the highest_lun index for
2082  * the controller then highest_lun is set * to this new index.
2083  * If there are no available indexes or if tha allocation fails, then -1
2084  * is returned.  * "controller_node" is used to know if this is a real
2085  * logical drive, or just the controller node, which determines if this
2086  * counts towards highest_lun.
2087  */
2088 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2089 {
2090         int i;
2091         drive_info_struct *drv;
2092
2093         /* Search for an empty slot for our drive info */
2094         for (i = 0; i < CISS_MAX_LUN; i++) {
2095
2096                 /* if not cxd0 case, and it's occupied, skip it. */
2097                 if (h->drv[i] && i != 0)
2098                         continue;
2099                 /*
2100                  * If it's cxd0 case, and drv is alloc'ed already, and a
2101                  * disk is configured there, skip it.
2102                  */
2103                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2104                         continue;
2105
2106                 /*
2107                  * We've found an empty slot.  Update highest_lun
2108                  * provided this isn't just the fake cxd0 controller node.
2109                  */
2110                 if (i > h->highest_lun && !controller_node)
2111                         h->highest_lun = i;
2112
2113                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2114                 if (i == 0 && h->drv[i] != NULL)
2115                         return i;
2116
2117                 /*
2118                  * Found an empty slot, not already alloc'ed.  Allocate it.
2119                  * Mark it with raid_level == -1, so we know it's new later on.
2120                  */
2121                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2122                 if (!drv)
2123                         return -1;
2124                 drv->raid_level = -1; /* so we know it's new */
2125                 h->drv[i] = drv;
2126                 return i;
2127         }
2128         return -1;
2129 }
2130
2131 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2132 {
2133         kfree(h->drv[drv_index]);
2134         h->drv[drv_index] = NULL;
2135 }
2136
2137 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2138 {
2139         put_disk(h->gendisk[drv_index]);
2140         h->gendisk[drv_index] = NULL;
2141 }
2142
2143 /* cciss_add_gendisk finds a free hba[]->drv structure
2144  * and allocates a gendisk if needed, and sets the lunid
2145  * in the drvinfo structure.   It returns the index into
2146  * the ->drv[] array, or -1 if none are free.
2147  * is_controller_node indicates whether highest_lun should
2148  * count this disk, or if it's only being added to provide
2149  * a means to talk to the controller in case no logical
2150  * drives have yet been configured.
2151  */
2152 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2153         int controller_node)
2154 {
2155         int drv_index;
2156
2157         drv_index = cciss_alloc_drive_info(h, controller_node);
2158         if (drv_index == -1)
2159                 return -1;
2160
2161         /*Check if the gendisk needs to be allocated */
2162         if (!h->gendisk[drv_index]) {
2163                 h->gendisk[drv_index] =
2164                         alloc_disk(1 << NWD_SHIFT);
2165                 if (!h->gendisk[drv_index]) {
2166                         printk(KERN_ERR "cciss%d: could not "
2167                                 "allocate a new disk %d\n",
2168                                 h->ctlr, drv_index);
2169                         goto err_free_drive_info;
2170                 }
2171         }
2172         memcpy(h->drv[drv_index]->LunID, lunid,
2173                 sizeof(h->drv[drv_index]->LunID));
2174         if (cciss_create_ld_sysfs_entry(h, drv_index))
2175                 goto err_free_disk;
2176         /* Don't need to mark this busy because nobody */
2177         /* else knows about this disk yet to contend */
2178         /* for access to it. */
2179         h->drv[drv_index]->busy_configuring = 0;
2180         wmb();
2181         return drv_index;
2182
2183 err_free_disk:
2184         cciss_free_gendisk(h, drv_index);
2185 err_free_drive_info:
2186         cciss_free_drive_info(h, drv_index);
2187         return -1;
2188 }
2189
2190 /* This is for the special case of a controller which
2191  * has no logical drives.  In this case, we still need
2192  * to register a disk so the controller can be accessed
2193  * by the Array Config Utility.
2194  */
2195 static void cciss_add_controller_node(ctlr_info_t *h)
2196 {
2197         struct gendisk *disk;
2198         int drv_index;
2199
2200         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2201                 return;
2202
2203         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2204         if (drv_index == -1)
2205                 goto error;
2206         h->drv[drv_index]->block_size = 512;
2207         h->drv[drv_index]->nr_blocks = 0;
2208         h->drv[drv_index]->heads = 0;
2209         h->drv[drv_index]->sectors = 0;
2210         h->drv[drv_index]->cylinders = 0;
2211         h->drv[drv_index]->raid_level = -1;
2212         memset(h->drv[drv_index]->serial_no, 0, 16);
2213         disk = h->gendisk[drv_index];
2214         if (cciss_add_disk(h, disk, drv_index) == 0)
2215                 return;
2216         cciss_free_gendisk(h, drv_index);
2217         cciss_free_drive_info(h, drv_index);
2218 error:
2219         printk(KERN_WARNING "cciss%d: could not "
2220                 "add disk 0.\n", h->ctlr);
2221         return;
2222 }
2223
2224 /* This function will add and remove logical drives from the Logical
2225  * drive array of the controller and maintain persistency of ordering
2226  * so that mount points are preserved until the next reboot.  This allows
2227  * for the removal of logical drives in the middle of the drive array
2228  * without a re-ordering of those drives.
2229  * INPUT
2230  * h            = The controller to perform the operations on
2231  */
2232 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2233         int via_ioctl)
2234 {
2235         int ctlr = h->ctlr;
2236         int num_luns;
2237         ReportLunData_struct *ld_buff = NULL;
2238         int return_code;
2239         int listlength = 0;
2240         int i;
2241         int drv_found;
2242         int drv_index = 0;
2243         unsigned char lunid[8] = CTLR_LUNID;
2244         unsigned long flags;
2245
2246         if (!capable(CAP_SYS_RAWIO))
2247                 return -EPERM;
2248
2249         /* Set busy_configuring flag for this operation */
2250         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2251         if (h->busy_configuring) {
2252                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2253                 return -EBUSY;
2254         }
2255         h->busy_configuring = 1;
2256         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2257
2258         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2259         if (ld_buff == NULL)
2260                 goto mem_msg;
2261
2262         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2263                                       sizeof(ReportLunData_struct),
2264                                       0, CTLR_LUNID, TYPE_CMD);
2265
2266         if (return_code == IO_OK)
2267                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2268         else {  /* reading number of logical volumes failed */
2269                 printk(KERN_WARNING "cciss: report logical volume"
2270                        " command failed\n");
2271                 listlength = 0;
2272                 goto freeret;
2273         }
2274
2275         num_luns = listlength / 8;      /* 8 bytes per entry */
2276         if (num_luns > CISS_MAX_LUN) {
2277                 num_luns = CISS_MAX_LUN;
2278                 printk(KERN_WARNING "cciss: more luns configured"
2279                        " on controller than can be handled by"
2280                        " this driver.\n");
2281         }
2282
2283         if (num_luns == 0)
2284                 cciss_add_controller_node(h);
2285
2286         /* Compare controller drive array to driver's drive array
2287          * to see if any drives are missing on the controller due
2288          * to action of Array Config Utility (user deletes drive)
2289          * and deregister logical drives which have disappeared.
2290          */
2291         for (i = 0; i <= h->highest_lun; i++) {
2292                 int j;
2293                 drv_found = 0;
2294
2295                 /* skip holes in the array from already deleted drives */
2296                 if (h->drv[i] == NULL)
2297                         continue;
2298
2299                 for (j = 0; j < num_luns; j++) {
2300                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2301                         if (memcmp(h->drv[i]->LunID, lunid,
2302                                 sizeof(lunid)) == 0) {
2303                                 drv_found = 1;
2304                                 break;
2305                         }
2306                 }
2307                 if (!drv_found) {
2308                         /* Deregister it from the OS, it's gone. */
2309                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2310                         h->drv[i]->busy_configuring = 1;
2311                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2312                         return_code = deregister_disk(h, i, 1, via_ioctl);
2313                         if (h->drv[i] != NULL)
2314                                 h->drv[i]->busy_configuring = 0;
2315                 }
2316         }
2317
2318         /* Compare controller drive array to driver's drive array.
2319          * Check for updates in the drive information and any new drives
2320          * on the controller due to ACU adding logical drives, or changing
2321          * a logical drive's size, etc.  Reregister any new/changed drives
2322          */
2323         for (i = 0; i < num_luns; i++) {
2324                 int j;
2325
2326                 drv_found = 0;
2327
2328                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2329                 /* Find if the LUN is already in the drive array
2330                  * of the driver.  If so then update its info
2331                  * if not in use.  If it does not exist then find
2332                  * the first free index and add it.
2333                  */
2334                 for (j = 0; j <= h->highest_lun; j++) {
2335                         if (h->drv[j] != NULL &&
2336                                 memcmp(h->drv[j]->LunID, lunid,
2337                                         sizeof(h->drv[j]->LunID)) == 0) {
2338                                 drv_index = j;
2339                                 drv_found = 1;
2340                                 break;
2341                         }
2342                 }
2343
2344                 /* check if the drive was found already in the array */
2345                 if (!drv_found) {
2346                         drv_index = cciss_add_gendisk(h, lunid, 0);
2347                         if (drv_index == -1)
2348                                 goto freeret;
2349                 }
2350                 cciss_update_drive_info(ctlr, drv_index, first_time,
2351                         via_ioctl);
2352         }               /* end for */
2353
2354 freeret:
2355         kfree(ld_buff);
2356         h->busy_configuring = 0;
2357         /* We return -1 here to tell the ACU that we have registered/updated
2358          * all of the drives that we can and to keep it from calling us
2359          * additional times.
2360          */
2361         return -1;
2362 mem_msg:
2363         printk(KERN_ERR "cciss: out of memory\n");
2364         h->busy_configuring = 0;
2365         goto freeret;
2366 }
2367
2368 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2369 {
2370         /* zero out the disk size info */
2371         drive_info->nr_blocks = 0;
2372         drive_info->block_size = 0;
2373         drive_info->heads = 0;
2374         drive_info->sectors = 0;
2375         drive_info->cylinders = 0;
2376         drive_info->raid_level = -1;
2377         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2378         memset(drive_info->model, 0, sizeof(drive_info->model));
2379         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2380         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2381         /*
2382          * don't clear the LUNID though, we need to remember which
2383          * one this one is.
2384          */
2385 }
2386
2387 /* This function will deregister the disk and it's queue from the
2388  * kernel.  It must be called with the controller lock held and the
2389  * drv structures busy_configuring flag set.  It's parameters are:
2390  *
2391  * disk = This is the disk to be deregistered
2392  * drv  = This is the drive_info_struct associated with the disk to be
2393  *        deregistered.  It contains information about the disk used
2394  *        by the driver.
2395  * clear_all = This flag determines whether or not the disk information
2396  *             is going to be completely cleared out and the highest_lun
2397  *             reset.  Sometimes we want to clear out information about
2398  *             the disk in preparation for re-adding it.  In this case
2399  *             the highest_lun should be left unchanged and the LunID
2400  *             should not be cleared.
2401  * via_ioctl
2402  *    This indicates whether we've reached this path via ioctl.
2403  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2404  *    If this path is reached via ioctl(), then the max_usage_count will
2405  *    be 1, as the process calling ioctl() has got to have the device open.
2406  *    If we get here via sysfs, then the max usage count will be zero.
2407 */
2408 static int deregister_disk(ctlr_info_t *h, int drv_index,
2409                            int clear_all, int via_ioctl)
2410 {
2411         int i;
2412         struct gendisk *disk;
2413         drive_info_struct *drv;
2414         int recalculate_highest_lun;
2415
2416         if (!capable(CAP_SYS_RAWIO))
2417                 return -EPERM;
2418
2419         drv = h->drv[drv_index];
2420         disk = h->gendisk[drv_index];
2421
2422         /* make sure logical volume is NOT is use */
2423         if (clear_all || (h->gendisk[0] == disk)) {
2424                 if (drv->usage_count > via_ioctl)
2425                         return -EBUSY;
2426         } else if (drv->usage_count > 0)
2427                 return -EBUSY;
2428
2429         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2430
2431         /* invalidate the devices and deregister the disk.  If it is disk
2432          * zero do not deregister it but just zero out it's values.  This
2433          * allows us to delete disk zero but keep the controller registered.
2434          */
2435         if (h->gendisk[0] != disk) {
2436                 struct request_queue *q = disk->queue;
2437                 if (disk->flags & GENHD_FL_UP) {
2438                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2439                         del_gendisk(disk);
2440                 }
2441                 if (q)
2442                         blk_cleanup_queue(q);
2443                 /* If clear_all is set then we are deleting the logical
2444                  * drive, not just refreshing its info.  For drives
2445                  * other than disk 0 we will call put_disk.  We do not
2446                  * do this for disk 0 as we need it to be able to
2447                  * configure the controller.
2448                  */
2449                 if (clear_all){
2450                         /* This isn't pretty, but we need to find the
2451                          * disk in our array and NULL our the pointer.
2452                          * This is so that we will call alloc_disk if
2453                          * this index is used again later.
2454                          */
2455                         for (i=0; i < CISS_MAX_LUN; i++){
2456                                 if (h->gendisk[i] == disk) {
2457                                         h->gendisk[i] = NULL;
2458                                         break;
2459                                 }
2460                         }
2461                         put_disk(disk);
2462                 }
2463         } else {
2464                 set_capacity(disk, 0);
2465                 cciss_clear_drive_info(drv);
2466         }
2467
2468         --h->num_luns;
2469
2470         /* if it was the last disk, find the new hightest lun */
2471         if (clear_all && recalculate_highest_lun) {
2472                 int newhighest = -1;
2473                 for (i = 0; i <= h->highest_lun; i++) {
2474                         /* if the disk has size > 0, it is available */
2475                         if (h->drv[i] && h->drv[i]->heads)
2476                                 newhighest = i;
2477                 }
2478                 h->highest_lun = newhighest;
2479         }
2480         return 0;
2481 }
2482
2483 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2484                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2485                 int cmd_type)
2486 {
2487         ctlr_info_t *h = hba[ctlr];
2488         u64bit buff_dma_handle;
2489         int status = IO_OK;
2490
2491         c->cmd_type = CMD_IOCTL_PEND;
2492         c->Header.ReplyQueue = 0;
2493         if (buff != NULL) {
2494                 c->Header.SGList = 1;
2495                 c->Header.SGTotal = 1;
2496         } else {
2497                 c->Header.SGList = 0;
2498                 c->Header.SGTotal = 0;
2499         }
2500         c->Header.Tag.lower = c->busaddr;
2501         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2502
2503         c->Request.Type.Type = cmd_type;
2504         if (cmd_type == TYPE_CMD) {
2505                 switch (cmd) {
2506                 case CISS_INQUIRY:
2507                         /* are we trying to read a vital product page */
2508                         if (page_code != 0) {
2509                                 c->Request.CDB[1] = 0x01;
2510                                 c->Request.CDB[2] = page_code;
2511                         }
2512                         c->Request.CDBLen = 6;
2513                         c->Request.Type.Attribute = ATTR_SIMPLE;
2514                         c->Request.Type.Direction = XFER_READ;
2515                         c->Request.Timeout = 0;
2516                         c->Request.CDB[0] = CISS_INQUIRY;
2517                         c->Request.CDB[4] = size & 0xFF;
2518                         break;
2519                 case CISS_REPORT_LOG:
2520                 case CISS_REPORT_PHYS:
2521                         /* Talking to controller so It's a physical command
2522                            mode = 00 target = 0.  Nothing to write.
2523                          */
2524                         c->Request.CDBLen = 12;
2525                         c->Request.Type.Attribute = ATTR_SIMPLE;
2526                         c->Request.Type.Direction = XFER_READ;
2527                         c->Request.Timeout = 0;
2528                         c->Request.CDB[0] = cmd;
2529                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2530                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2531                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2532                         c->Request.CDB[9] = size & 0xFF;
2533                         break;
2534
2535                 case CCISS_READ_CAPACITY:
2536                         c->Request.CDBLen = 10;
2537                         c->Request.Type.Attribute = ATTR_SIMPLE;
2538                         c->Request.Type.Direction = XFER_READ;
2539                         c->Request.Timeout = 0;
2540                         c->Request.CDB[0] = cmd;
2541                         break;
2542                 case CCISS_READ_CAPACITY_16:
2543                         c->Request.CDBLen = 16;
2544                         c->Request.Type.Attribute = ATTR_SIMPLE;
2545                         c->Request.Type.Direction = XFER_READ;
2546                         c->Request.Timeout = 0;
2547                         c->Request.CDB[0] = cmd;
2548                         c->Request.CDB[1] = 0x10;
2549                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2550                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2551                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2552                         c->Request.CDB[13] = size & 0xFF;
2553                         c->Request.Timeout = 0;
2554                         c->Request.CDB[0] = cmd;
2555                         break;
2556                 case CCISS_CACHE_FLUSH:
2557                         c->Request.CDBLen = 12;
2558                         c->Request.Type.Attribute = ATTR_SIMPLE;
2559                         c->Request.Type.Direction = XFER_WRITE;
2560                         c->Request.Timeout = 0;
2561                         c->Request.CDB[0] = BMIC_WRITE;
2562                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2563                         break;
2564                 case TEST_UNIT_READY:
2565                         c->Request.CDBLen = 6;
2566                         c->Request.Type.Attribute = ATTR_SIMPLE;
2567                         c->Request.Type.Direction = XFER_NONE;
2568                         c->Request.Timeout = 0;
2569                         break;
2570                 default:
2571                         printk(KERN_WARNING
2572                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2573                         return IO_ERROR;
2574                 }
2575         } else if (cmd_type == TYPE_MSG) {
2576                 switch (cmd) {
2577                 case 0: /* ABORT message */
2578                         c->Request.CDBLen = 12;
2579                         c->Request.Type.Attribute = ATTR_SIMPLE;
2580                         c->Request.Type.Direction = XFER_WRITE;
2581                         c->Request.Timeout = 0;
2582                         c->Request.CDB[0] = cmd;        /* abort */
2583                         c->Request.CDB[1] = 0;  /* abort a command */
2584                         /* buff contains the tag of the command to abort */
2585                         memcpy(&c->Request.CDB[4], buff, 8);
2586                         break;
2587                 case 1: /* RESET message */
2588                         c->Request.CDBLen = 16;
2589                         c->Request.Type.Attribute = ATTR_SIMPLE;
2590                         c->Request.Type.Direction = XFER_NONE;
2591                         c->Request.Timeout = 0;
2592                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2593                         c->Request.CDB[0] = cmd;        /* reset */
2594                         c->Request.CDB[1] = 0x03;       /* reset a target */
2595                         break;
2596                 case 3: /* No-Op message */
2597                         c->Request.CDBLen = 1;
2598                         c->Request.Type.Attribute = ATTR_SIMPLE;
2599                         c->Request.Type.Direction = XFER_WRITE;
2600                         c->Request.Timeout = 0;
2601                         c->Request.CDB[0] = cmd;
2602                         break;
2603                 default:
2604                         printk(KERN_WARNING
2605                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2606                         return IO_ERROR;
2607                 }
2608         } else {
2609                 printk(KERN_WARNING
2610                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2611                 return IO_ERROR;
2612         }
2613         /* Fill in the scatter gather information */
2614         if (size > 0) {
2615                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2616                                                              buff, size,
2617                                                              PCI_DMA_BIDIRECTIONAL);
2618                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2619                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2620                 c->SG[0].Len = size;
2621                 c->SG[0].Ext = 0;       /* we are not chaining */
2622         }
2623         return status;
2624 }
2625
2626 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2627 {
2628         switch (c->err_info->ScsiStatus) {
2629         case SAM_STAT_GOOD:
2630                 return IO_OK;
2631         case SAM_STAT_CHECK_CONDITION:
2632                 switch (0xf & c->err_info->SenseInfo[2]) {
2633                 case 0: return IO_OK; /* no sense */
2634                 case 1: return IO_OK; /* recovered error */
2635                 default:
2636                         if (check_for_unit_attention(h, c))
2637                                 return IO_NEEDS_RETRY;
2638                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2639                                 "check condition, sense key = 0x%02x\n",
2640                                 h->ctlr, c->Request.CDB[0],
2641                                 c->err_info->SenseInfo[2]);
2642                 }
2643                 break;
2644         default:
2645                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2646                         "scsi status = 0x%02x\n", h->ctlr,
2647                         c->Request.CDB[0], c->err_info->ScsiStatus);
2648                 break;
2649         }
2650         return IO_ERROR;
2651 }
2652
2653 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2654 {
2655         int return_status = IO_OK;
2656
2657         if (c->err_info->CommandStatus == CMD_SUCCESS)
2658                 return IO_OK;
2659
2660         switch (c->err_info->CommandStatus) {
2661         case CMD_TARGET_STATUS:
2662                 return_status = check_target_status(h, c);
2663                 break;
2664         case CMD_DATA_UNDERRUN:
2665         case CMD_DATA_OVERRUN:
2666                 /* expected for inquiry and report lun commands */
2667                 break;
2668         case CMD_INVALID:
2669                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2670                        "reported invalid\n", c->Request.CDB[0]);
2671                 return_status = IO_ERROR;
2672                 break;
2673         case CMD_PROTOCOL_ERR:
2674                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2675                        "protocol error \n", c->Request.CDB[0]);
2676                 return_status = IO_ERROR;
2677                 break;
2678         case CMD_HARDWARE_ERR:
2679                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2680                        " hardware error\n", c->Request.CDB[0]);
2681                 return_status = IO_ERROR;
2682                 break;
2683         case CMD_CONNECTION_LOST:
2684                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2685                        "connection lost\n", c->Request.CDB[0]);
2686                 return_status = IO_ERROR;
2687                 break;
2688         case CMD_ABORTED:
2689                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2690                        "aborted\n", c->Request.CDB[0]);
2691                 return_status = IO_ERROR;
2692                 break;
2693         case CMD_ABORT_FAILED:
2694                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2695                        "abort failed\n", c->Request.CDB[0]);
2696                 return_status = IO_ERROR;
2697                 break;
2698         case CMD_UNSOLICITED_ABORT:
2699                 printk(KERN_WARNING
2700                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2701                         c->Request.CDB[0]);
2702                 return_status = IO_NEEDS_RETRY;
2703                 break;
2704         default:
2705                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2706                        "unknown status %x\n", c->Request.CDB[0],
2707                        c->err_info->CommandStatus);
2708                 return_status = IO_ERROR;
2709         }
2710         return return_status;
2711 }
2712
2713 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2714         int attempt_retry)
2715 {
2716         DECLARE_COMPLETION_ONSTACK(wait);
2717         u64bit buff_dma_handle;
2718         int return_status = IO_OK;
2719
2720 resend_cmd2:
2721         c->waiting = &wait;
2722         enqueue_cmd_and_start_io(h, c);
2723
2724         wait_for_completion(&wait);
2725
2726         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2727                 goto command_done;
2728
2729         return_status = process_sendcmd_error(h, c);
2730
2731         if (return_status == IO_NEEDS_RETRY &&
2732                 c->retry_count < MAX_CMD_RETRIES) {
2733                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2734                         c->Request.CDB[0]);
2735                 c->retry_count++;
2736                 /* erase the old error information */
2737                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2738                 return_status = IO_OK;
2739                 INIT_COMPLETION(wait);
2740                 goto resend_cmd2;
2741         }
2742
2743 command_done:
2744         /* unlock the buffers from DMA */
2745         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2746         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2747         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2748                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2749         return return_status;
2750 }
2751
2752 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2753                            __u8 page_code, unsigned char scsi3addr[],
2754                         int cmd_type)
2755 {
2756         ctlr_info_t *h = hba[ctlr];
2757         CommandList_struct *c;
2758         int return_status;
2759
2760         c = cmd_alloc(h, 0);
2761         if (!c)
2762                 return -ENOMEM;
2763         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2764                 scsi3addr, cmd_type);
2765         if (return_status == IO_OK)
2766                 return_status = sendcmd_withirq_core(h, c, 1);
2767
2768         cmd_free(h, c, 0);
2769         return return_status;
2770 }
2771
2772 static void cciss_geometry_inquiry(int ctlr, int logvol,
2773                                    sector_t total_size,
2774                                    unsigned int block_size,
2775                                    InquiryData_struct *inq_buff,
2776                                    drive_info_struct *drv)
2777 {
2778         int return_code;
2779         unsigned long t;
2780         unsigned char scsi3addr[8];
2781
2782         memset(inq_buff, 0, sizeof(InquiryData_struct));
2783         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2784         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2785                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2786         if (return_code == IO_OK) {
2787                 if (inq_buff->data_byte[8] == 0xFF) {
2788                         printk(KERN_WARNING
2789                                "cciss: reading geometry failed, volume "
2790                                "does not support reading geometry\n");
2791                         drv->heads = 255;
2792                         drv->sectors = 32;      /* Sectors per track */
2793                         drv->cylinders = total_size + 1;
2794                         drv->raid_level = RAID_UNKNOWN;
2795                 } else {
2796                         drv->heads = inq_buff->data_byte[6];
2797                         drv->sectors = inq_buff->data_byte[7];
2798                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2799                         drv->cylinders += inq_buff->data_byte[5];
2800                         drv->raid_level = inq_buff->data_byte[8];
2801                 }
2802                 drv->block_size = block_size;
2803                 drv->nr_blocks = total_size + 1;
2804                 t = drv->heads * drv->sectors;
2805                 if (t > 1) {
2806                         sector_t real_size = total_size + 1;
2807                         unsigned long rem = sector_div(real_size, t);
2808                         if (rem)
2809                                 real_size++;
2810                         drv->cylinders = real_size;
2811                 }
2812         } else {                /* Get geometry failed */
2813                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2814         }
2815 }
2816
2817 static void
2818 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2819                     unsigned int *block_size)
2820 {
2821         ReadCapdata_struct *buf;
2822         int return_code;
2823         unsigned char scsi3addr[8];
2824
2825         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2826         if (!buf) {
2827                 printk(KERN_WARNING "cciss: out of memory\n");
2828                 return;
2829         }
2830
2831         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2832         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2833                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2834         if (return_code == IO_OK) {
2835                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2836                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2837         } else {                /* read capacity command failed */
2838                 printk(KERN_WARNING "cciss: read capacity failed\n");
2839                 *total_size = 0;
2840                 *block_size = BLOCK_SIZE;
2841         }
2842         kfree(buf);
2843 }
2844
2845 static void cciss_read_capacity_16(int ctlr, int logvol,
2846         sector_t *total_size, unsigned int *block_size)
2847 {
2848         ReadCapdata_struct_16 *buf;
2849         int return_code;
2850         unsigned char scsi3addr[8];
2851
2852         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2853         if (!buf) {
2854                 printk(KERN_WARNING "cciss: out of memory\n");
2855                 return;
2856         }
2857
2858         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2859         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2860                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2861                         0, scsi3addr, TYPE_CMD);
2862         if (return_code == IO_OK) {
2863                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2864                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2865         } else {                /* read capacity command failed */
2866                 printk(KERN_WARNING "cciss: read capacity failed\n");
2867                 *total_size = 0;
2868                 *block_size = BLOCK_SIZE;
2869         }
2870         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2871                (unsigned long long)*total_size+1, *block_size);
2872         kfree(buf);
2873 }
2874
2875 static int cciss_revalidate(struct gendisk *disk)
2876 {
2877         ctlr_info_t *h = get_host(disk);
2878         drive_info_struct *drv = get_drv(disk);
2879         int logvol;
2880         int FOUND = 0;
2881         unsigned int block_size;
2882         sector_t total_size;
2883         InquiryData_struct *inq_buff = NULL;
2884
2885         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2886                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2887                         sizeof(drv->LunID)) == 0) {
2888                         FOUND = 1;
2889                         break;
2890                 }
2891         }
2892
2893         if (!FOUND)
2894                 return 1;
2895
2896         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2897         if (inq_buff == NULL) {
2898                 printk(KERN_WARNING "cciss: out of memory\n");
2899                 return 1;
2900         }
2901         if (h->cciss_read == CCISS_READ_10) {
2902                 cciss_read_capacity(h->ctlr, logvol,
2903                                         &total_size, &block_size);
2904         } else {
2905                 cciss_read_capacity_16(h->ctlr, logvol,
2906                                         &total_size, &block_size);
2907         }
2908         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2909                                inq_buff, drv);
2910
2911         blk_queue_logical_block_size(drv->queue, drv->block_size);
2912         set_capacity(disk, drv->nr_blocks);
2913
2914         kfree(inq_buff);
2915         return 0;
2916 }
2917
2918 /*
2919  * Map (physical) PCI mem into (virtual) kernel space
2920  */
2921 static void __iomem *remap_pci_mem(ulong base, ulong size)
2922 {
2923         ulong page_base = ((ulong) base) & PAGE_MASK;
2924         ulong page_offs = ((ulong) base) - page_base;
2925         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2926
2927         return page_remapped ? (page_remapped + page_offs) : NULL;
2928 }
2929
2930 /*
2931  * Takes jobs of the Q and sends them to the hardware, then puts it on
2932  * the Q to wait for completion.
2933  */
2934 static void start_io(ctlr_info_t *h)
2935 {
2936         CommandList_struct *c;
2937
2938         while (!hlist_empty(&h->reqQ)) {
2939                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2940                 /* can't do anything if fifo is full */
2941                 if ((h->access.fifo_full(h))) {
2942                         printk(KERN_WARNING "cciss: fifo full\n");
2943                         break;
2944                 }
2945
2946                 /* Get the first entry from the Request Q */
2947                 removeQ(c);
2948                 h->Qdepth--;
2949
2950                 /* Tell the controller execute command */
2951                 h->access.submit_command(h, c);
2952
2953                 /* Put job onto the completed Q */
2954                 addQ(&h->cmpQ, c);
2955         }
2956 }
2957
2958 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2959 /* Zeros out the error record and then resends the command back */
2960 /* to the controller */
2961 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2962 {
2963         /* erase the old error information */
2964         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2965
2966         /* add it to software queue and then send it to the controller */
2967         addQ(&h->reqQ, c);
2968         h->Qdepth++;
2969         if (h->Qdepth > h->maxQsinceinit)
2970                 h->maxQsinceinit = h->Qdepth;
2971
2972         start_io(h);
2973 }
2974
2975 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2976         unsigned int msg_byte, unsigned int host_byte,
2977         unsigned int driver_byte)
2978 {
2979         /* inverse of macros in scsi.h */
2980         return (scsi_status_byte & 0xff) |
2981                 ((msg_byte & 0xff) << 8) |
2982                 ((host_byte & 0xff) << 16) |
2983                 ((driver_byte & 0xff) << 24);
2984 }
2985
2986 static inline int evaluate_target_status(ctlr_info_t *h,
2987                         CommandList_struct *cmd, int *retry_cmd)
2988 {
2989         unsigned char sense_key;
2990         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2991         int error_value;
2992
2993         *retry_cmd = 0;
2994         /* If we get in here, it means we got "target status", that is, scsi status */
2995         status_byte = cmd->err_info->ScsiStatus;
2996         driver_byte = DRIVER_OK;
2997         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2998
2999         if (blk_pc_request(cmd->rq))
3000                 host_byte = DID_PASSTHROUGH;
3001         else
3002                 host_byte = DID_OK;
3003
3004         error_value = make_status_bytes(status_byte, msg_byte,
3005                 host_byte, driver_byte);
3006
3007         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3008                 if (!blk_pc_request(cmd->rq))
3009                         printk(KERN_WARNING "cciss: cmd %p "
3010                                "has SCSI Status 0x%x\n",
3011                                cmd, cmd->err_info->ScsiStatus);
3012                 return error_value;
3013         }
3014
3015         /* check the sense key */
3016         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3017         /* no status or recovered error */
3018         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
3019                 error_value = 0;
3020
3021         if (check_for_unit_attention(h, cmd)) {
3022                 *retry_cmd = !blk_pc_request(cmd->rq);
3023                 return 0;
3024         }
3025
3026         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
3027                 if (error_value != 0)
3028                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3029                                " sense key = 0x%x\n", cmd, sense_key);
3030                 return error_value;
3031         }
3032
3033         /* SG_IO or similar, copy sense data back */
3034         if (cmd->rq->sense) {
3035                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3036                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3037                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3038                         cmd->rq->sense_len);
3039         } else
3040                 cmd->rq->sense_len = 0;
3041
3042         return error_value;
3043 }
3044
3045 /* checks the status of the job and calls complete buffers to mark all
3046  * buffers for the completed job. Note that this function does not need
3047  * to hold the hba/queue lock.
3048  */
3049 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3050                                     int timeout)
3051 {
3052         int retry_cmd = 0;
3053         struct request *rq = cmd->rq;
3054
3055         rq->errors = 0;
3056
3057         if (timeout)
3058                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3059
3060         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3061                 goto after_error_processing;
3062
3063         switch (cmd->err_info->CommandStatus) {
3064         case CMD_TARGET_STATUS:
3065                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3066                 break;
3067         case CMD_DATA_UNDERRUN:
3068                 if (blk_fs_request(cmd->rq)) {
3069                         printk(KERN_WARNING "cciss: cmd %p has"
3070                                " completed with data underrun "
3071                                "reported\n", cmd);
3072                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3073                 }
3074                 break;
3075         case CMD_DATA_OVERRUN:
3076                 if (blk_fs_request(cmd->rq))
3077                         printk(KERN_WARNING "cciss: cmd %p has"
3078                                " completed with data overrun "
3079                                "reported\n", cmd);
3080                 break;
3081         case CMD_INVALID:
3082                 printk(KERN_WARNING "cciss: cmd %p is "
3083                        "reported invalid\n", cmd);
3084                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3085                         cmd->err_info->CommandStatus, DRIVER_OK,
3086                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3087                 break;
3088         case CMD_PROTOCOL_ERR:
3089                 printk(KERN_WARNING "cciss: cmd %p has "
3090                        "protocol error \n", cmd);
3091                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3092                         cmd->err_info->CommandStatus, DRIVER_OK,
3093                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3094                 break;
3095         case CMD_HARDWARE_ERR:
3096                 printk(KERN_WARNING "cciss: cmd %p had "
3097                        " hardware error\n", cmd);
3098                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3099                         cmd->err_info->CommandStatus, DRIVER_OK,
3100                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3101                 break;
3102         case CMD_CONNECTION_LOST:
3103                 printk(KERN_WARNING "cciss: cmd %p had "
3104                        "connection lost\n", cmd);
3105                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3106                         cmd->err_info->CommandStatus, DRIVER_OK,
3107                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3108                 break;
3109         case CMD_ABORTED:
3110                 printk(KERN_WARNING "cciss: cmd %p was "
3111                        "aborted\n", cmd);
3112                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3113                         cmd->err_info->CommandStatus, DRIVER_OK,
3114                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3115                 break;
3116         case CMD_ABORT_FAILED:
3117                 printk(KERN_WARNING "cciss: cmd %p reports "
3118                        "abort failed\n", cmd);
3119                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3120                         cmd->err_info->CommandStatus, DRIVER_OK,
3121                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3122                 break;
3123         case CMD_UNSOLICITED_ABORT:
3124                 printk(KERN_WARNING "cciss%d: unsolicited "
3125                        "abort %p\n", h->ctlr, cmd);
3126                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3127                         retry_cmd = 1;
3128                         printk(KERN_WARNING
3129                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3130                         cmd->retry_count++;
3131                 } else
3132                         printk(KERN_WARNING
3133                                "cciss%d: %p retried too "
3134                                "many times\n", h->ctlr, cmd);
3135                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3136                         cmd->err_info->CommandStatus, DRIVER_OK,
3137                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3138                 break;
3139         case CMD_TIMEOUT:
3140                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3141                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3142                         cmd->err_info->CommandStatus, DRIVER_OK,
3143                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3144                 break;
3145         default:
3146                 printk(KERN_WARNING "cciss: cmd %p returned "
3147                        "unknown status %x\n", cmd,
3148                        cmd->err_info->CommandStatus);
3149                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3150                         cmd->err_info->CommandStatus, DRIVER_OK,
3151                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3152         }
3153
3154 after_error_processing:
3155
3156         /* We need to return this command */
3157         if (retry_cmd) {
3158                 resend_cciss_cmd(h, cmd);
3159                 return;
3160         }
3161         cmd->rq->completion_data = cmd;
3162         blk_complete_request(cmd->rq);
3163 }
3164
3165 static inline u32 cciss_tag_contains_index(u32 tag)
3166 {
3167 #define DIRECT_LOOKUP_BIT 0x10
3168         return tag & DIRECT_LOOKUP_BIT;
3169 }
3170
3171 static inline u32 cciss_tag_to_index(u32 tag)
3172 {
3173 #define DIRECT_LOOKUP_SHIFT 5
3174         return tag >> DIRECT_LOOKUP_SHIFT;
3175 }
3176
3177 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3178 {
3179 #define CCISS_ERROR_BITS 0x03
3180         return tag & ~CCISS_ERROR_BITS;
3181 }
3182
3183 static inline void cciss_mark_tag_indexed(u32 *tag)
3184 {
3185         *tag |= DIRECT_LOOKUP_BIT;
3186 }
3187
3188 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3189 {
3190         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3191 }
3192
3193 /*
3194  * Get a request and submit it to the controller.
3195  */
3196 static void do_cciss_request(struct request_queue *q)
3197 {
3198         ctlr_info_t *h = q->queuedata;
3199         CommandList_struct *c;
3200         sector_t start_blk;
3201         int seg;
3202         struct request *creq;
3203         u64bit temp64;
3204         struct scatterlist *tmp_sg;
3205         SGDescriptor_struct *curr_sg;
3206         drive_info_struct *drv;
3207         int i, dir;
3208         int sg_index = 0;
3209         int chained = 0;
3210
3211         /* We call start_io here in case there is a command waiting on the
3212          * queue that has not been sent.
3213          */
3214         if (blk_queue_plugged(q))
3215                 goto startio;
3216
3217       queue:
3218         creq = blk_peek_request(q);
3219         if (!creq)
3220                 goto startio;
3221
3222         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3223
3224         if ((c = cmd_alloc(h, 1)) == NULL)
3225                 goto full;
3226
3227         blk_start_request(creq);
3228
3229         tmp_sg = h->scatter_list[c->cmdindex];
3230         spin_unlock_irq(q->queue_lock);
3231
3232         c->cmd_type = CMD_RWREQ;
3233         c->rq = creq;
3234
3235         /* fill in the request */
3236         drv = creq->rq_disk->private_data;
3237         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3238         /* got command from pool, so use the command block index instead */
3239         /* for direct lookups. */
3240         /* The first 2 bits are reserved for controller error reporting. */
3241         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3242         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3243         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3244         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3245         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3246         c->Request.Type.Attribute = ATTR_SIMPLE;
3247         c->Request.Type.Direction =
3248             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3249         c->Request.Timeout = 0; /* Don't time out */
3250         c->Request.CDB[0] =
3251             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3252         start_blk = blk_rq_pos(creq);
3253 #ifdef CCISS_DEBUG
3254         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3255                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3256 #endif                          /* CCISS_DEBUG */
3257
3258         sg_init_table(tmp_sg, h->maxsgentries);
3259         seg = blk_rq_map_sg(q, creq, tmp_sg);
3260
3261         /* get the DMA records for the setup */
3262         if (c->Request.Type.Direction == XFER_READ)
3263                 dir = PCI_DMA_FROMDEVICE;
3264         else
3265                 dir = PCI_DMA_TODEVICE;
3266
3267         curr_sg = c->SG;
3268         sg_index = 0;
3269         chained = 0;
3270
3271         for (i = 0; i < seg; i++) {
3272                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3273                         !chained && ((seg - i) > 1)) {
3274                         /* Point to next chain block. */
3275                         curr_sg = h->cmd_sg_list[c->cmdindex];
3276                         sg_index = 0;
3277                         chained = 1;
3278                 }
3279                 curr_sg[sg_index].Len = tmp_sg[i].length;
3280                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3281                                                 tmp_sg[i].offset,
3282                                                 tmp_sg[i].length, dir);
3283                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3284                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3285                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3286                 ++sg_index;
3287         }
3288         if (chained)
3289                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3290                         (seg - (h->max_cmd_sgentries - 1)) *
3291                                 sizeof(SGDescriptor_struct));
3292
3293         /* track how many SG entries we are using */
3294         if (seg > h->maxSG)
3295                 h->maxSG = seg;
3296
3297 #ifdef CCISS_DEBUG
3298         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3299                         "chained[%d]\n",
3300                         blk_rq_sectors(creq), seg, chained);
3301 #endif                          /* CCISS_DEBUG */
3302
3303         c->Header.SGTotal = seg + chained;
3304         if (seg <= h->max_cmd_sgentries)
3305                 c->Header.SGList = c->Header.SGTotal;
3306         else
3307                 c->Header.SGList = h->max_cmd_sgentries;
3308         set_performant_mode(h, c);
3309
3310         if (likely(blk_fs_request(creq))) {
3311                 if(h->cciss_read == CCISS_READ_10) {
3312                         c->Request.CDB[1] = 0;
3313                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3314                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3315                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3316                         c->Request.CDB[5] = start_blk & 0xff;
3317                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3318                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3319                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3320                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3321                 } else {
3322                         u32 upper32 = upper_32_bits(start_blk);
3323
3324                         c->Request.CDBLen = 16;
3325                         c->Request.CDB[1]= 0;
3326                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3327                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3328                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3329                         c->Request.CDB[5]= upper32 & 0xff;
3330                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3331                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3332                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3333                         c->Request.CDB[9]= start_blk & 0xff;
3334                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3335                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3336                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3337                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3338                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3339                 }
3340         } else if (blk_pc_request(creq)) {
3341                 c->Request.CDBLen = creq->cmd_len;
3342                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3343         } else {
3344                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3345                 BUG();
3346         }
3347
3348         spin_lock_irq(q->queue_lock);
3349
3350         addQ(&h->reqQ, c);
3351         h->Qdepth++;
3352         if (h->Qdepth > h->maxQsinceinit)
3353                 h->maxQsinceinit = h->Qdepth;
3354
3355         goto queue;
3356 full:
3357         blk_stop_queue(q);
3358 startio:
3359         /* We will already have the driver lock here so not need
3360          * to lock it.
3361          */
3362         start_io(h);
3363 }
3364
3365 static inline unsigned long get_next_completion(ctlr_info_t *h)
3366 {
3367         return h->access.command_completed(h);
3368 }
3369
3370 static inline int interrupt_pending(ctlr_info_t *h)
3371 {
3372         return h->access.intr_pending(h);
3373 }
3374
3375 static inline long interrupt_not_for_us(ctlr_info_t *h)
3376 {
3377         return !(h->msi_vector || h->msix_vector) &&
3378                 ((h->access.intr_pending(h) == 0) ||
3379                 (h->interrupts_enabled == 0));
3380 }
3381
3382 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3383                         u32 raw_tag)
3384 {
3385         if (unlikely(tag_index >= h->nr_cmds)) {
3386                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3387                 return 1;
3388         }
3389         return 0;
3390 }
3391
3392 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3393                                 u32 raw_tag)
3394 {
3395         removeQ(c);
3396         if (likely(c->cmd_type == CMD_RWREQ))
3397                 complete_command(h, c, 0);
3398         else if (c->cmd_type == CMD_IOCTL_PEND)
3399                 complete(c->waiting);
3400 #ifdef CONFIG_CISS_SCSI_TAPE
3401         else if (c->cmd_type == CMD_SCSI)
3402                 complete_scsi_command(c, 0, raw_tag);
3403 #endif
3404 }
3405
3406 /* process completion of an indexed ("direct lookup") command */
3407 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3408 {
3409         u32 tag_index;
3410         CommandList_struct *c;
3411
3412         tag_index = cciss_tag_to_index(raw_tag);
3413         if (bad_tag(h, tag_index, raw_tag))
3414                 return next_command(h);
3415         c = h->cmd_pool + tag_index;
3416         finish_cmd(h, c, raw_tag);
3417         return next_command(h);
3418 }
3419
3420 /* process completion of a non-indexed command */
3421 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3422 {
3423         u32 tag;
3424         CommandList_struct *c = NULL;
3425         struct hlist_node *tmp;
3426         __u32 busaddr_masked, tag_masked;
3427
3428         tag = cciss_tag_discard_error_bits(raw_tag);
3429         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3430                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3431                 tag_masked = cciss_tag_discard_error_bits(tag);
3432                 if (busaddr_masked == tag_masked) {
3433                         finish_cmd(h, c, raw_tag);
3434                         return next_command(h);
3435                 }
3436         }
3437         bad_tag(h, h->nr_cmds + 1, raw_tag);
3438         return next_command(h);
3439 }
3440
3441 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3442 {
3443         ctlr_info_t *h = dev_id;
3444         unsigned long flags;
3445         u32 raw_tag;
3446
3447         if (interrupt_not_for_us(h))
3448                 return IRQ_NONE;
3449         /*
3450          * If there are completed commands in the completion queue,
3451          * we had better do something about it.
3452          */
3453         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3454         while (interrupt_pending(h)) {
3455                 raw_tag = get_next_completion(h);
3456                 while (raw_tag != FIFO_EMPTY) {
3457                         if (cciss_tag_contains_index(raw_tag))
3458                                 raw_tag = process_indexed_cmd(h, raw_tag);
3459                         else
3460                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3461                 }
3462         }
3463
3464         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3465         return IRQ_HANDLED;
3466 }
3467
3468 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3469  * check the interrupt pending register because it is not set.
3470  */
3471 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3472 {
3473         ctlr_info_t *h = dev_id;
3474         unsigned long flags;
3475         u32 raw_tag;
3476
3477         if (interrupt_not_for_us(h))
3478                 return IRQ_NONE;
3479         /*
3480          * If there are completed commands in the completion queue,
3481          * we had better do something about it.
3482          */
3483         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3484         raw_tag = get_next_completion(h);
3485         while (raw_tag != FIFO_EMPTY) {
3486                 if (cciss_tag_contains_index(raw_tag))
3487                         raw_tag = process_indexed_cmd(h, raw_tag);
3488                 else
3489                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3490         }
3491
3492         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3493         return IRQ_HANDLED;
3494 }
3495
3496 /**
3497  * add_to_scan_list() - add controller to rescan queue
3498  * @h:                Pointer to the controller.
3499  *
3500  * Adds the controller to the rescan queue if not already on the queue.
3501  *
3502  * returns 1 if added to the queue, 0 if skipped (could be on the
3503  * queue already, or the controller could be initializing or shutting
3504  * down).
3505  **/
3506 static int add_to_scan_list(struct ctlr_info *h)
3507 {
3508         struct ctlr_info *test_h;
3509         int found = 0;
3510         int ret = 0;
3511
3512         if (h->busy_initializing)
3513                 return 0;
3514
3515         if (!mutex_trylock(&h->busy_shutting_down))
3516                 return 0;
3517
3518         mutex_lock(&scan_mutex);
3519         list_for_each_entry(test_h, &scan_q, scan_list) {
3520                 if (test_h == h) {
3521                         found = 1;
3522                         break;
3523                 }
3524         }
3525         if (!found && !h->busy_scanning) {
3526                 INIT_COMPLETION(h->scan_wait);
3527                 list_add_tail(&h->scan_list, &scan_q);
3528                 ret = 1;
3529         }
3530         mutex_unlock(&scan_mutex);
3531         mutex_unlock(&h->busy_shutting_down);
3532
3533         return ret;
3534 }
3535
3536 /**
3537  * remove_from_scan_list() - remove controller from rescan queue
3538  * @h:                     Pointer to the controller.
3539  *
3540  * Removes the controller from the rescan queue if present. Blocks if
3541  * the controller is currently conducting a rescan.  The controller
3542  * can be in one of three states:
3543  * 1. Doesn't need a scan
3544  * 2. On the scan list, but not scanning yet (we remove it)
3545  * 3. Busy scanning (and not on the list). In this case we want to wait for
3546  *    the scan to complete to make sure the scanning thread for this
3547  *    controller is completely idle.
3548  **/
3549 static void remove_from_scan_list(struct ctlr_info *h)
3550 {
3551         struct ctlr_info *test_h, *tmp_h;
3552
3553         mutex_lock(&scan_mutex);
3554         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3555                 if (test_h == h) { /* state 2. */
3556                         list_del(&h->scan_list);
3557                         complete_all(&h->scan_wait);
3558                         mutex_unlock(&scan_mutex);
3559                         return;
3560                 }
3561         }
3562         if (h->busy_scanning) { /* state 3. */
3563                 mutex_unlock(&scan_mutex);
3564                 wait_for_completion(&h->scan_wait);
3565         } else { /* state 1, nothing to do. */
3566                 mutex_unlock(&scan_mutex);
3567         }
3568 }
3569
3570 /**
3571  * scan_thread() - kernel thread used to rescan controllers
3572  * @data:        Ignored.
3573  *
3574  * A kernel thread used scan for drive topology changes on
3575  * controllers. The thread processes only one controller at a time
3576  * using a queue.  Controllers are added to the queue using
3577  * add_to_scan_list() and removed from the queue either after done
3578  * processing or using remove_from_scan_list().
3579  *
3580  * returns 0.
3581  **/
3582 static int scan_thread(void *data)
3583 {
3584         struct ctlr_info *h;
3585
3586         while (1) {
3587                 set_current_state(TASK_INTERRUPTIBLE);
3588                 schedule();
3589                 if (kthread_should_stop())
3590                         break;
3591
3592                 while (1) {
3593                         mutex_lock(&scan_mutex);
3594                         if (list_empty(&scan_q)) {
3595                                 mutex_unlock(&scan_mutex);
3596                                 break;
3597                         }
3598
3599                         h = list_entry(scan_q.next,
3600                                        struct ctlr_info,
3601                                        scan_list);
3602                         list_del(&h->scan_list);
3603                         h->busy_scanning = 1;
3604                         mutex_unlock(&scan_mutex);
3605
3606                         rebuild_lun_table(h, 0, 0);
3607                         complete_all(&h->scan_wait);
3608                         mutex_lock(&scan_mutex);
3609                         h->busy_scanning = 0;
3610                         mutex_unlock(&scan_mutex);
3611                 }
3612         }
3613
3614         return 0;
3615 }
3616
3617 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3618 {
3619         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3620                 return 0;
3621
3622         switch (c->err_info->SenseInfo[12]) {
3623         case STATE_CHANGED:
3624                 printk(KERN_WARNING "cciss%d: a state change "
3625                         "detected, command retried\n", h->ctlr);
3626                 return 1;
3627         break;
3628         case LUN_FAILED:
3629                 printk(KERN_WARNING "cciss%d: LUN failure "
3630                         "detected, action required\n", h->ctlr);
3631                 return 1;
3632         break;
3633         case REPORT_LUNS_CHANGED:
3634                 printk(KERN_WARNING "cciss%d: report LUN data "
3635                         "changed\n", h->ctlr);
3636         /*
3637          * Here, we could call add_to_scan_list and wake up the scan thread,
3638          * except that it's quite likely that we will get more than one
3639          * REPORT_LUNS_CHANGED condition in quick succession, which means
3640          * that those which occur after the first one will likely happen
3641          * *during* the scan_thread's rescan.  And the rescan code is not
3642          * robust enough to restart in the middle, undoing what it has already
3643          * done, and it's not clear that it's even possible to do this, since
3644          * part of what it does is notify the block layer, which starts
3645          * doing it's own i/o to read partition tables and so on, and the
3646          * driver doesn't have visibility to know what might need undoing.
3647          * In any event, if possible, it is horribly complicated to get right
3648          * so we just don't do it for now.
3649          *
3650          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3651          */
3652                 return 1;
3653         break;
3654         case POWER_OR_RESET:
3655                 printk(KERN_WARNING "cciss%d: a power on "
3656                         "or device reset detected\n", h->ctlr);
3657                 return 1;
3658         break;
3659         case UNIT_ATTENTION_CLEARED:
3660                 printk(KERN_WARNING "cciss%d: unit attention "
3661                     "cleared by another initiator\n", h->ctlr);
3662                 return 1;
3663         break;
3664         default:
3665                 printk(KERN_WARNING "cciss%d: unknown "
3666                         "unit attention detected\n", h->ctlr);
3667                                 return 1;
3668         }
3669 }
3670
3671 /*
3672  *  We cannot read the structure directly, for portability we must use
3673  *   the io functions.
3674  *   This is for debug only.
3675  */
3676 #ifdef CCISS_DEBUG
3677 static void print_cfg_table(CfgTable_struct *tb)
3678 {
3679         int i;
3680         char temp_name[17];
3681
3682         printk("Controller Configuration information\n");
3683         printk("------------------------------------\n");
3684         for (i = 0; i < 4; i++)
3685                 temp_name[i] = readb(&(tb->Signature[i]));
3686         temp_name[4] = '\0';
3687         printk("   Signature = %s\n", temp_name);
3688         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3689         printk("   Transport methods supported = 0x%x\n",
3690                readl(&(tb->TransportSupport)));
3691         printk("   Transport methods active = 0x%x\n",
3692                readl(&(tb->TransportActive)));
3693         printk("   Requested transport Method = 0x%x\n",
3694                readl(&(tb->HostWrite.TransportRequest)));
3695         printk("   Coalesce Interrupt Delay = 0x%x\n",
3696                readl(&(tb->HostWrite.CoalIntDelay)));
3697         printk("   Coalesce Interrupt Count = 0x%x\n",
3698                readl(&(tb->HostWrite.CoalIntCount)));
3699         printk("   Max outstanding commands = 0x%d\n",
3700                readl(&(tb->CmdsOutMax)));
3701         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3702         for (i = 0; i < 16; i++)
3703                 temp_name[i] = readb(&(tb->ServerName[i]));
3704         temp_name[16] = '\0';
3705         printk("   Server Name = %s\n", temp_name);
3706         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3707 }
3708 #endif                          /* CCISS_DEBUG */
3709
3710 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3711 {
3712         int i, offset, mem_type, bar_type;
3713         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3714                 return 0;
3715         offset = 0;
3716         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3717                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3718                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3719                         offset += 4;
3720                 else {
3721                         mem_type = pci_resource_flags(pdev, i) &
3722                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3723                         switch (mem_type) {
3724                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3725                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3726                                 offset += 4;    /* 32 bit */
3727                                 break;
3728                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3729                                 offset += 8;
3730                                 break;
3731                         default:        /* reserved in PCI 2.2 */
3732                                 printk(KERN_WARNING
3733                                        "Base address is invalid\n");
3734                                 return -1;
3735                                 break;
3736                         }
3737                 }
3738                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3739                         return i + 1;
3740         }
3741         return -1;
3742 }
3743
3744 /* Fill in bucket_map[], given nsgs (the max number of
3745  * scatter gather elements supported) and bucket[],
3746  * which is an array of 8 integers.  The bucket[] array
3747  * contains 8 different DMA transfer sizes (in 16
3748  * byte increments) which the controller uses to fetch
3749  * commands.  This function fills in bucket_map[], which
3750  * maps a given number of scatter gather elements to one of
3751  * the 8 DMA transfer sizes.  The point of it is to allow the
3752  * controller to only do as much DMA as needed to fetch the
3753  * command, with the DMA transfer size encoded in the lower
3754  * bits of the command address.
3755  */
3756 static void  calc_bucket_map(int bucket[], int num_buckets,
3757         int nsgs, int *bucket_map)
3758 {
3759         int i, j, b, size;
3760
3761         /* even a command with 0 SGs requires 4 blocks */
3762 #define MINIMUM_TRANSFER_BLOCKS 4
3763 #define NUM_BUCKETS 8
3764         /* Note, bucket_map must have nsgs+1 entries. */
3765         for (i = 0; i <= nsgs; i++) {
3766                 /* Compute size of a command with i SG entries */
3767                 size = i + MINIMUM_TRANSFER_BLOCKS;
3768                 b = num_buckets; /* Assume the biggest bucket */
3769                 /* Find the bucket that is just big enough */
3770                 for (j = 0; j < 8; j++) {
3771                         if (bucket[j] >= size) {
3772                                 b = j;
3773                                 break;
3774                         }
3775                 }
3776                 /* for a command with i SG entries, use bucket b. */
3777                 bucket_map[i] = b;
3778         }
3779 }
3780
3781 static void
3782 cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3783 {
3784         int l = 0;
3785         __u32 trans_support;
3786         __u32 trans_offset;
3787                         /*
3788                          *  5 = 1 s/g entry or 4k
3789                          *  6 = 2 s/g entry or 8k
3790                          *  8 = 4 s/g entry or 16k
3791                          * 10 = 6 s/g entry or 24k
3792                          */
3793         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3794         unsigned long register_value;
3795
3796         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3797
3798         /* Attempt to put controller into performant mode if supported */
3799         /* Does board support performant mode? */
3800         trans_support = readl(&(h->cfgtable->TransportSupport));
3801         if (!(trans_support & PERFORMANT_MODE))
3802                 return;
3803
3804         printk(KERN_WARNING "cciss%d: Placing controller into "
3805                                 "performant mode\n", h->ctlr);
3806         /* Performant mode demands commands on a 32 byte boundary
3807          * pci_alloc_consistent aligns on page boundarys already.
3808          * Just need to check if divisible by 32
3809          */
3810         if ((sizeof(CommandList_struct) % 32) != 0) {
3811                 printk(KERN_WARNING "%s %d %s\n",
3812                         "cciss info: command size[",
3813                         (int)sizeof(CommandList_struct),
3814                         "] not divisible by 32, no performant mode..\n");
3815                 return;
3816         }
3817
3818         /* Performant mode ring buffer and supporting data structures */
3819         h->reply_pool = (__u64 *)pci_alloc_consistent(
3820                 h->pdev, h->max_commands * sizeof(__u64),
3821                 &(h->reply_pool_dhandle));
3822
3823         /* Need a block fetch table for performant mode */
3824         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3825                 sizeof(__u32)), GFP_KERNEL);
3826
3827         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3828                 goto clean_up;
3829
3830         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3831
3832         /* Controller spec: zero out this buffer. */
3833         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3834         h->reply_pool_head = h->reply_pool;
3835
3836         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3837         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3838                                 h->blockFetchTable);
3839         writel(bft[0], &h->transtable->BlockFetch0);
3840         writel(bft[1], &h->transtable->BlockFetch1);
3841         writel(bft[2], &h->transtable->BlockFetch2);
3842         writel(bft[3], &h->transtable->BlockFetch3);
3843         writel(bft[4], &h->transtable->BlockFetch4);
3844         writel(bft[5], &h->transtable->BlockFetch5);
3845         writel(bft[6], &h->transtable->BlockFetch6);
3846         writel(bft[7], &h->transtable->BlockFetch7);
3847
3848         /* size of controller ring buffer */
3849         writel(h->max_commands, &h->transtable->RepQSize);
3850         writel(1, &h->transtable->RepQCount);
3851         writel(0, &h->transtable->RepQCtrAddrLow32);
3852         writel(0, &h->transtable->RepQCtrAddrHigh32);
3853         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3854         writel(0, &h->transtable->RepQAddr0High32);
3855         writel(CFGTBL_Trans_Performant,
3856                         &(h->cfgtable->HostWrite.TransportRequest));
3857
3858         h->transMethod = CFGTBL_Trans_Performant;
3859         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3860         /* under certain very rare conditions, this can take awhile.
3861          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3862          * as we enter this code.) */
3863         for (l = 0; l < MAX_CONFIG_WAIT; l++) {
3864                 register_value = readl(h->vaddr + SA5_DOORBELL);
3865                 if (!(register_value & CFGTBL_ChangeReq))
3866                         break;
3867                 /* delay and try again */
3868                 set_current_state(TASK_INTERRUPTIBLE);
3869                 schedule_timeout(10);
3870         }
3871         register_value = readl(&(h->cfgtable->TransportActive));
3872         if (!(register_value & CFGTBL_Trans_Performant)) {
3873                 printk(KERN_WARNING "cciss: unable to get board into"
3874                                         " performant mode\n");
3875                 return;
3876         }
3877
3878         /* Change the access methods to the performant access methods */
3879         h->access = SA5_performant_access;
3880
3881         return;
3882 clean_up:
3883         kfree(h->blockFetchTable);
3884         if (h->reply_pool)
3885                 pci_free_consistent(h->pdev,
3886                                 h->max_commands * sizeof(__u64),
3887                                 h->reply_pool,
3888                                 h->reply_pool_dhandle);
3889         return;
3890
3891 } /* cciss_put_controller_into_performant_mode */
3892
3893 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3894  * controllers that are capable. If not, we use IO-APIC mode.
3895  */
3896
3897 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3898                                            struct pci_dev *pdev, __u32 board_id)
3899 {
3900 #ifdef CONFIG_PCI_MSI
3901         int err;
3902         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3903         {0, 2}, {0, 3}
3904         };
3905
3906         /* Some boards advertise MSI but don't really support it */
3907         if ((board_id == 0x40700E11) ||
3908             (board_id == 0x40800E11) ||
3909             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3910                 goto default_int_mode;
3911
3912         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3913                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3914                 if (!err) {
3915                         c->intr[0] = cciss_msix_entries[0].vector;
3916                         c->intr[1] = cciss_msix_entries[1].vector;
3917                         c->intr[2] = cciss_msix_entries[2].vector;
3918                         c->intr[3] = cciss_msix_entries[3].vector;
3919                         c->msix_vector = 1;
3920                         return;
3921                 }
3922                 if (err > 0) {
3923                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3924                                "available\n", err);
3925                         goto default_int_mode;
3926                 } else {
3927                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3928                                err);
3929                         goto default_int_mode;
3930                 }
3931         }
3932         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3933                 if (!pci_enable_msi(pdev)) {
3934                         c->msi_vector = 1;
3935                 } else {
3936                         printk(KERN_WARNING "cciss: MSI init failed\n");
3937                 }
3938         }
3939 default_int_mode:
3940 #endif                          /* CONFIG_PCI_MSI */
3941         /* if we get here we're going to use the default interrupt mode */
3942         c->intr[PERF_MODE_INT] = pdev->irq;
3943         return;
3944 }
3945
3946 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3947 {
3948         ushort subsystem_vendor_id, subsystem_device_id, command;
3949         __u32 board_id, scratchpad = 0;
3950         __u64 cfg_offset;
3951         __u32 cfg_base_addr;
3952         __u64 cfg_base_addr_index;
3953         int i, prod_index, err;
3954         __u32 trans_offset;
3955
3956         subsystem_vendor_id = pdev->subsystem_vendor;
3957         subsystem_device_id = pdev->subsystem_device;
3958         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3959                     subsystem_vendor_id);
3960
3961         for (i = 0; i < ARRAY_SIZE(products); i++) {
3962                 /* Stand aside for hpsa driver on request */
3963                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3964                         return -ENODEV;
3965                 if (board_id == products[i].board_id)
3966                         break;
3967         }
3968         prod_index = i;
3969         if (prod_index == ARRAY_SIZE(products)) {
3970                 dev_warn(&pdev->dev,
3971                         "unrecognized board ID: 0x%08lx, ignoring.\n",
3972                         (unsigned long) board_id);
3973                 return -ENODEV;
3974         }
3975
3976         /* check to see if controller has been disabled */
3977         /* BEFORE trying to enable it */
3978         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3979         if (!(command & 0x02)) {
3980                 printk(KERN_WARNING
3981                        "cciss: controller appears to be disabled\n");
3982                 return -ENODEV;
3983         }
3984
3985         err = pci_enable_device(pdev);
3986         if (err) {
3987                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3988                 return err;
3989         }
3990
3991         err = pci_request_regions(pdev, "cciss");
3992         if (err) {
3993                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3994                        "aborting\n");
3995                 return err;
3996         }
3997
3998 #ifdef CCISS_DEBUG
3999         printk("command = %x\n", command);
4000         printk("irq = %x\n", pdev->irq);
4001         printk("board_id = %x\n", board_id);
4002 #endif                          /* CCISS_DEBUG */
4003
4004 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4005  * else we use the IO-APIC interrupt assigned to us by system ROM.
4006  */
4007         cciss_interrupt_mode(c, pdev, board_id);
4008
4009         /* find the memory BAR */
4010         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
4011                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
4012                         break;
4013         }
4014         if (i == DEVICE_COUNT_RESOURCE) {
4015                 printk(KERN_WARNING "cciss: No memory BAR found\n");
4016                 err = -ENODEV;
4017                 goto err_out_free_res;
4018         }
4019
4020         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
4021                                                  * already removed
4022                                                  */
4023
4024 #ifdef CCISS_DEBUG
4025         printk("address 0 = %lx\n", c->paddr);
4026 #endif                          /* CCISS_DEBUG */
4027         c->vaddr = remap_pci_mem(c->paddr, 0x250);
4028
4029         /* Wait for the board to become ready.  (PCI hotplug needs this.)
4030          * We poll for up to 120 secs, once per 100ms. */
4031         for (i = 0; i < 1200; i++) {
4032                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
4033                 if (scratchpad == CCISS_FIRMWARE_READY)
4034                         break;
4035                 set_current_state(TASK_INTERRUPTIBLE);
4036                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
4037         }
4038         if (scratchpad != CCISS_FIRMWARE_READY) {
4039                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
4040                 err = -ENODEV;
4041                 goto err_out_free_res;
4042         }
4043
4044         /* get the address index number */
4045         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
4046         cfg_base_addr &= (__u32) 0x0000ffff;
4047 #ifdef CCISS_DEBUG
4048         printk("cfg base address = %x\n", cfg_base_addr);
4049 #endif                          /* CCISS_DEBUG */
4050         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
4051 #ifdef CCISS_DEBUG
4052         printk("cfg base address index = %llx\n",
4053                 (unsigned long long)cfg_base_addr_index);
4054 #endif                          /* CCISS_DEBUG */
4055         if (cfg_base_addr_index == -1) {
4056                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
4057                 err = -ENODEV;
4058                 goto err_out_free_res;
4059         }
4060
4061         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
4062 #ifdef CCISS_DEBUG
4063         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
4064 #endif                          /* CCISS_DEBUG */
4065         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
4066                                                        cfg_base_addr_index) +
4067                                     cfg_offset, sizeof(CfgTable_struct));
4068         /* Find performant mode table. */
4069         trans_offset = readl(&(c->cfgtable->TransMethodOffset));
4070         c->transtable = remap_pci_mem(pci_resource_start(pdev,
4071                 cfg_base_addr_index) + cfg_offset+trans_offset,
4072                 sizeof(*c->transtable));
4073         c->board_id = board_id;
4074
4075         #ifdef CCISS_DEBUG
4076                 print_cfg_table(c->cfgtable);
4077         #endif                          /* CCISS_DEBUG */
4078
4079         /* Some controllers support Zero Memory Raid (ZMR).
4080          * When configured in ZMR mode the number of supported
4081          * commands drops to 64. So instead of just setting an
4082          * arbitrary value we make the driver a little smarter.
4083          * We read the config table to tell us how many commands
4084          * are supported on the controller then subtract 4 to
4085          * leave a little room for ioctl calls.
4086          */
4087         c->max_commands = readl(&(c->cfgtable->MaxPerformantModeCommands));
4088         c->maxsgentries = readl(&(c->cfgtable->MaxSGElements));
4089
4090         /*
4091          * Limit native command to 32 s/g elements to save dma'able memory.
4092          * Howvever spec says if 0, use 31
4093          */
4094
4095         c->max_cmd_sgentries = 31;
4096         if (c->maxsgentries > 512) {
4097                 c->max_cmd_sgentries = 32;
4098                 c->chainsize = c->maxsgentries - c->max_cmd_sgentries + 1;
4099                 c->maxsgentries -= 1;   /* account for chain pointer */
4100         } else {
4101                 c->maxsgentries = 31;   /* Default to traditional value */
4102                 c->chainsize = 0;       /* traditional */
4103         }
4104
4105         c->product_name = products[prod_index].product_name;
4106         c->access = *(products[prod_index].access);
4107         c->nr_cmds = c->max_commands - 4;
4108         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
4109             (readb(&c->cfgtable->Signature[1]) != 'I') ||
4110             (readb(&c->cfgtable->Signature[2]) != 'S') ||
4111             (readb(&c->cfgtable->Signature[3]) != 'S')) {
4112                 printk("Does not appear to be a valid CISS config table\n");
4113                 err = -ENODEV;
4114                 goto err_out_free_res;
4115         }
4116 #ifdef CONFIG_X86
4117         {
4118                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4119                 __u32 prefetch;
4120                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
4121                 prefetch |= 0x100;
4122                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
4123         }
4124 #endif
4125
4126         /* Disabling DMA prefetch and refetch for the P600.
4127          * An ASIC bug may result in accesses to invalid memory addresses.
4128          * We've disabled prefetch for some time now. Testing with XEN
4129          * kernels revealed a bug in the refetch if dom0 resides on a P600.
4130          */
4131         if(board_id == 0x3225103C) {
4132                         __u32 dma_prefetch;
4133                 __u32 dma_refetch;
4134                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
4135                 dma_prefetch |= 0x8000;
4136                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
4137                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
4138                 dma_refetch |= 0x1;
4139                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
4140         }
4141
4142 #ifdef CCISS_DEBUG
4143         printk(KERN_WARNING "Trying to put board into Performant mode\n");
4144 #endif                          /* CCISS_DEBUG */
4145         return 0;
4146
4147 err_out_free_res:
4148         /*
4149          * Deliberately omit pci_disable_device(): it does something nasty to
4150          * Smart Array controllers that pci_enable_device does not undo
4151          */
4152         pci_release_regions(pdev);
4153         cciss_put_controller_into_performant_mode(c);
4154         return err;
4155 }
4156
4157 /* Function to find the first free pointer into our hba[] array
4158  * Returns -1 if no free entries are left.
4159  */
4160 static int alloc_cciss_hba(void)
4161 {
4162         int i;
4163
4164         for (i = 0; i < MAX_CTLR; i++) {
4165                 if (!hba[i]) {
4166                         ctlr_info_t *p;
4167
4168                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4169                         if (!p)
4170                                 goto Enomem;
4171                         hba[i] = p;
4172                         return i;
4173                 }
4174         }
4175         printk(KERN_WARNING "cciss: This driver supports a maximum"
4176                " of %d controllers.\n", MAX_CTLR);
4177         return -1;
4178 Enomem:
4179         printk(KERN_ERR "cciss: out of memory.\n");
4180         return -1;
4181 }
4182
4183 static void free_hba(int n)
4184 {
4185         ctlr_info_t *h = hba[n];
4186         int i;
4187
4188         hba[n] = NULL;
4189         for (i = 0; i < h->highest_lun + 1; i++)
4190                 if (h->gendisk[i] != NULL)
4191                         put_disk(h->gendisk[i]);
4192         kfree(h);
4193 }
4194
4195 /* Send a message CDB to the firmware. */
4196 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4197 {
4198         typedef struct {
4199                 CommandListHeader_struct CommandHeader;
4200                 RequestBlock_struct Request;
4201                 ErrDescriptor_struct ErrorDescriptor;
4202         } Command;
4203         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4204         Command *cmd;
4205         dma_addr_t paddr64;
4206         uint32_t paddr32, tag;
4207         void __iomem *vaddr;
4208         int i, err;
4209
4210         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4211         if (vaddr == NULL)
4212                 return -ENOMEM;
4213
4214         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4215            CCISS commands, so they must be allocated from the lower 4GiB of
4216            memory. */
4217         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4218         if (err) {
4219                 iounmap(vaddr);
4220                 return -ENOMEM;
4221         }
4222
4223         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4224         if (cmd == NULL) {
4225                 iounmap(vaddr);
4226                 return -ENOMEM;
4227         }
4228
4229         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4230            although there's no guarantee, we assume that the address is at
4231            least 4-byte aligned (most likely, it's page-aligned). */
4232         paddr32 = paddr64;
4233
4234         cmd->CommandHeader.ReplyQueue = 0;
4235         cmd->CommandHeader.SGList = 0;
4236         cmd->CommandHeader.SGTotal = 0;
4237         cmd->CommandHeader.Tag.lower = paddr32;
4238         cmd->CommandHeader.Tag.upper = 0;
4239         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4240
4241         cmd->Request.CDBLen = 16;
4242         cmd->Request.Type.Type = TYPE_MSG;
4243         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4244         cmd->Request.Type.Direction = XFER_NONE;
4245         cmd->Request.Timeout = 0; /* Don't time out */
4246         cmd->Request.CDB[0] = opcode;
4247         cmd->Request.CDB[1] = type;
4248         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4249
4250         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4251         cmd->ErrorDescriptor.Addr.upper = 0;
4252         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4253
4254         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4255
4256         for (i = 0; i < 10; i++) {
4257                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4258                 if ((tag & ~3) == paddr32)
4259                         break;
4260                 schedule_timeout_uninterruptible(HZ);
4261         }
4262
4263         iounmap(vaddr);
4264
4265         /* we leak the DMA buffer here ... no choice since the controller could
4266            still complete the command. */
4267         if (i == 10) {
4268                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4269                         opcode, type);
4270                 return -ETIMEDOUT;
4271         }
4272
4273         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4274
4275         if (tag & 2) {
4276                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4277                         opcode, type);
4278                 return -EIO;
4279         }
4280
4281         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4282                 opcode, type);
4283         return 0;
4284 }
4285
4286 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4287 #define cciss_noop(p) cciss_message(p, 3, 0)
4288
4289 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4290 {
4291 /* the #defines are stolen from drivers/pci/msi.h. */
4292 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4293 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4294
4295         int pos;
4296         u16 control = 0;
4297
4298         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4299         if (pos) {
4300                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4301                 if (control & PCI_MSI_FLAGS_ENABLE) {
4302                         printk(KERN_INFO "cciss: resetting MSI\n");
4303                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4304                 }
4305         }
4306
4307         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4308         if (pos) {
4309                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4310                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4311                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4312                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4313                 }
4314         }
4315
4316         return 0;
4317 }
4318
4319 /* This does a hard reset of the controller using PCI power management
4320  * states. */
4321 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4322 {
4323         u16 pmcsr, saved_config_space[32];
4324         int i, pos;
4325
4326         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4327
4328         /* This is very nearly the same thing as
4329
4330            pci_save_state(pci_dev);
4331            pci_set_power_state(pci_dev, PCI_D3hot);
4332            pci_set_power_state(pci_dev, PCI_D0);
4333            pci_restore_state(pci_dev);
4334
4335            but we can't use these nice canned kernel routines on
4336            kexec, because they also check the MSI/MSI-X state in PCI
4337            configuration space and do the wrong thing when it is
4338            set/cleared.  Also, the pci_save/restore_state functions
4339            violate the ordering requirements for restoring the
4340            configuration space from the CCISS document (see the
4341            comment below).  So we roll our own .... */
4342
4343         for (i = 0; i < 32; i++)
4344                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4345
4346         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4347         if (pos == 0) {
4348                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4349                 return -ENODEV;
4350         }
4351
4352         /* Quoting from the Open CISS Specification: "The Power
4353          * Management Control/Status Register (CSR) controls the power
4354          * state of the device.  The normal operating state is D0,
4355          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4356          * the controller, place the interface device in D3 then to
4357          * D0, this causes a secondary PCI reset which will reset the
4358          * controller." */
4359
4360         /* enter the D3hot power management state */
4361         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4362         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4363         pmcsr |= PCI_D3hot;
4364         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4365
4366         schedule_timeout_uninterruptible(HZ >> 1);
4367
4368         /* enter the D0 power management state */
4369         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4370         pmcsr |= PCI_D0;
4371         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4372
4373         schedule_timeout_uninterruptible(HZ >> 1);
4374
4375         /* Restore the PCI configuration space.  The Open CISS
4376          * Specification says, "Restore the PCI Configuration
4377          * Registers, offsets 00h through 60h. It is important to
4378          * restore the command register, 16-bits at offset 04h,
4379          * last. Do not restore the configuration status register,
4380          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4381         for (i = 0; i < 32; i++) {
4382                 if (i == 2 || i == 3)
4383                         continue;
4384                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4385         }
4386         wmb();
4387         pci_write_config_word(pdev, 4, saved_config_space[2]);
4388
4389         return 0;
4390 }
4391
4392 /*
4393  *  This is it.  Find all the controllers and register them.  I really hate
4394  *  stealing all these major device numbers.
4395  *  returns the number of block devices registered.
4396  */
4397 static int __devinit cciss_init_one(struct pci_dev *pdev,
4398                                     const struct pci_device_id *ent)
4399 {
4400         int i;
4401         int j = 0;
4402         int k = 0;
4403         int rc;
4404         int dac, return_code;
4405         InquiryData_struct *inq_buff;
4406
4407         if (reset_devices) {
4408                 /* Reset the controller with a PCI power-cycle */
4409                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4410                         return -ENODEV;
4411
4412                 /* Now try to get the controller to respond to a no-op. Some
4413                    devices (notably the HP Smart Array 5i Controller) need
4414                    up to 30 seconds to respond. */
4415                 for (i=0; i<30; i++) {
4416                         if (cciss_noop(pdev) == 0)
4417                                 break;
4418
4419                         schedule_timeout_uninterruptible(HZ);
4420                 }
4421                 if (i == 30) {
4422                         printk(KERN_ERR "cciss: controller seems dead\n");
4423                         return -EBUSY;
4424                 }
4425         }
4426
4427         i = alloc_cciss_hba();
4428         if (i < 0)
4429                 return -1;
4430         hba[i]->busy_initializing = 1;
4431         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4432         INIT_HLIST_HEAD(&hba[i]->reqQ);
4433         mutex_init(&hba[i]->busy_shutting_down);
4434
4435         if (cciss_pci_init(hba[i], pdev) != 0)
4436                 goto clean_no_release_regions;
4437
4438         sprintf(hba[i]->devname, "cciss%d", i);
4439         hba[i]->ctlr = i;
4440         hba[i]->pdev = pdev;
4441
4442         init_completion(&hba[i]->scan_wait);
4443
4444         if (cciss_create_hba_sysfs_entry(hba[i]))
4445                 goto clean0;
4446
4447         /* configure PCI DMA stuff */
4448         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4449                 dac = 1;
4450         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4451                 dac = 0;
4452         else {
4453                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4454                 goto clean1;
4455         }
4456
4457         /*
4458          * register with the major number, or get a dynamic major number
4459          * by passing 0 as argument.  This is done for greater than
4460          * 8 controller support.
4461          */
4462         if (i < MAX_CTLR_ORIG)
4463                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4464         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4465         if (rc == -EBUSY || rc == -EINVAL) {
4466                 printk(KERN_ERR
4467                        "cciss:  Unable to get major number %d for %s "
4468                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4469                 goto clean1;
4470         } else {
4471                 if (i >= MAX_CTLR_ORIG)
4472                         hba[i]->major = rc;
4473         }
4474
4475         /* make sure the board interrupts are off */
4476         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4477         if (hba[i]->msi_vector || hba[i]->msix_vector) {
4478                 if (request_irq(hba[i]->intr[SIMPLE_MODE_INT],
4479                                 do_cciss_msix_intr,
4480                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4481                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4482                                hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4483                         goto clean2;
4484                 }
4485         } else {
4486                 if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intx,
4487                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4488                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4489                                hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4490                         goto clean2;
4491                 }
4492         }
4493
4494         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4495                hba[i]->devname, pdev->device, pci_name(pdev),
4496                hba[i]->intr[PERF_MODE_INT], dac ? "" : " not");
4497
4498         hba[i]->cmd_pool_bits =
4499             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4500                         * sizeof(unsigned long), GFP_KERNEL);
4501         hba[i]->cmd_pool = (CommandList_struct *)
4502             pci_alloc_consistent(hba[i]->pdev,
4503                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4504                     &(hba[i]->cmd_pool_dhandle));
4505         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4506             pci_alloc_consistent(hba[i]->pdev,
4507                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4508                     &(hba[i]->errinfo_pool_dhandle));
4509         if ((hba[i]->cmd_pool_bits == NULL)
4510             || (hba[i]->cmd_pool == NULL)
4511             || (hba[i]->errinfo_pool == NULL)) {
4512                 printk(KERN_ERR "cciss: out of memory");
4513                 goto clean4;
4514         }
4515
4516         /* Need space for temp scatter list */
4517         hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4518                                                 sizeof(struct scatterlist *),
4519                                                 GFP_KERNEL);
4520         for (k = 0; k < hba[i]->nr_cmds; k++) {
4521                 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4522                                                         hba[i]->maxsgentries,
4523                                                         GFP_KERNEL);
4524                 if (hba[i]->scatter_list[k] == NULL) {
4525                         printk(KERN_ERR "cciss%d: could not allocate "
4526                                 "s/g lists\n", i);
4527                         goto clean4;
4528                 }
4529         }
4530         hba[i]->cmd_sg_list = cciss_allocate_sg_chain_blocks(hba[i],
4531                 hba[i]->chainsize, hba[i]->nr_cmds);
4532         if (!hba[i]->cmd_sg_list && hba[i]->chainsize > 0)
4533                 goto clean4;
4534
4535         spin_lock_init(&hba[i]->lock);
4536
4537         /* Initialize the pdev driver private data.
4538            have it point to hba[i].  */
4539         pci_set_drvdata(pdev, hba[i]);
4540         /* command and error info recs zeroed out before
4541            they are used */
4542         memset(hba[i]->cmd_pool_bits, 0,
4543                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4544                         * sizeof(unsigned long));
4545
4546         hba[i]->num_luns = 0;
4547         hba[i]->highest_lun = -1;
4548         for (j = 0; j < CISS_MAX_LUN; j++) {
4549                 hba[i]->drv[j] = NULL;
4550                 hba[i]->gendisk[j] = NULL;
4551         }
4552
4553         cciss_scsi_setup(i);
4554
4555         /* Turn the interrupts on so we can service requests */
4556         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4557
4558         /* Get the firmware version */
4559         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4560         if (inq_buff == NULL) {
4561                 printk(KERN_ERR "cciss: out of memory\n");
4562                 goto clean4;
4563         }
4564
4565         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4566                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4567         if (return_code == IO_OK) {
4568                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4569                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4570                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4571                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4572         } else {         /* send command failed */
4573                 printk(KERN_WARNING "cciss: unable to determine firmware"
4574                         " version of controller\n");
4575         }
4576         kfree(inq_buff);
4577
4578         cciss_procinit(i);
4579
4580         hba[i]->cciss_max_sectors = 8192;
4581
4582         rebuild_lun_table(hba[i], 1, 0);
4583         hba[i]->busy_initializing = 0;
4584         return 1;
4585
4586 clean4:
4587         kfree(hba[i]->cmd_pool_bits);
4588         /* Free up sg elements */
4589         for (k = 0; k < hba[i]->nr_cmds; k++)
4590                 kfree(hba[i]->scatter_list[k]);
4591         kfree(hba[i]->scatter_list);
4592         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4593         if (hba[i]->cmd_pool)
4594                 pci_free_consistent(hba[i]->pdev,
4595                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4596                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4597         if (hba[i]->errinfo_pool)
4598                 pci_free_consistent(hba[i]->pdev,
4599                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4600                                     hba[i]->errinfo_pool,
4601                                     hba[i]->errinfo_pool_dhandle);
4602         free_irq(hba[i]->intr[PERF_MODE_INT], hba[i]);
4603 clean2:
4604         unregister_blkdev(hba[i]->major, hba[i]->devname);
4605 clean1:
4606         cciss_destroy_hba_sysfs_entry(hba[i]);
4607 clean0:
4608         pci_release_regions(pdev);
4609 clean_no_release_regions:
4610         hba[i]->busy_initializing = 0;
4611
4612         /*
4613          * Deliberately omit pci_disable_device(): it does something nasty to
4614          * Smart Array controllers that pci_enable_device does not undo
4615          */
4616         pci_set_drvdata(pdev, NULL);
4617         free_hba(i);
4618         return -1;
4619 }
4620
4621 static void cciss_shutdown(struct pci_dev *pdev)
4622 {
4623         ctlr_info_t *h;
4624         char *flush_buf;
4625         int return_code;
4626
4627         h = pci_get_drvdata(pdev);
4628         flush_buf = kzalloc(4, GFP_KERNEL);
4629         if (!flush_buf) {
4630                 printk(KERN_WARNING
4631                         "cciss:%d cache not flushed, out of memory.\n",
4632                         h->ctlr);
4633                 return;
4634         }
4635         /* write all data in the battery backed cache to disk */
4636         memset(flush_buf, 0, 4);
4637         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4638                 4, 0, CTLR_LUNID, TYPE_CMD);
4639         kfree(flush_buf);
4640         if (return_code != IO_OK)
4641                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4642                         h->ctlr);
4643         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4644         free_irq(h->intr[PERF_MODE_INT], h);
4645 }
4646
4647 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4648 {
4649         ctlr_info_t *tmp_ptr;
4650         int i, j;
4651
4652         if (pci_get_drvdata(pdev) == NULL) {
4653                 printk(KERN_ERR "cciss: Unable to remove device \n");
4654                 return;
4655         }
4656
4657         tmp_ptr = pci_get_drvdata(pdev);
4658         i = tmp_ptr->ctlr;
4659         if (hba[i] == NULL) {
4660                 printk(KERN_ERR "cciss: device appears to "
4661                        "already be removed \n");
4662                 return;
4663         }
4664
4665         mutex_lock(&hba[i]->busy_shutting_down);
4666
4667         remove_from_scan_list(hba[i]);
4668         remove_proc_entry(hba[i]->devname, proc_cciss);
4669         unregister_blkdev(hba[i]->major, hba[i]->devname);
4670
4671         /* remove it from the disk list */
4672         for (j = 0; j < CISS_MAX_LUN; j++) {
4673                 struct gendisk *disk = hba[i]->gendisk[j];
4674                 if (disk) {
4675                         struct request_queue *q = disk->queue;
4676
4677                         if (disk->flags & GENHD_FL_UP) {
4678                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4679                                 del_gendisk(disk);
4680                         }
4681                         if (q)
4682                                 blk_cleanup_queue(q);
4683                 }
4684         }
4685
4686 #ifdef CONFIG_CISS_SCSI_TAPE
4687         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4688 #endif
4689
4690         cciss_shutdown(pdev);
4691
4692 #ifdef CONFIG_PCI_MSI
4693         if (hba[i]->msix_vector)
4694                 pci_disable_msix(hba[i]->pdev);
4695         else if (hba[i]->msi_vector)
4696                 pci_disable_msi(hba[i]->pdev);
4697 #endif                          /* CONFIG_PCI_MSI */
4698
4699         iounmap(hba[i]->vaddr);
4700
4701         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4702                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4703         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4704                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4705         kfree(hba[i]->cmd_pool_bits);
4706         /* Free up sg elements */
4707         for (j = 0; j < hba[i]->nr_cmds; j++)
4708                 kfree(hba[i]->scatter_list[j]);
4709         kfree(hba[i]->scatter_list);
4710         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4711         /*
4712          * Deliberately omit pci_disable_device(): it does something nasty to
4713          * Smart Array controllers that pci_enable_device does not undo
4714          */
4715         pci_release_regions(pdev);
4716         pci_set_drvdata(pdev, NULL);
4717         cciss_destroy_hba_sysfs_entry(hba[i]);
4718         mutex_unlock(&hba[i]->busy_shutting_down);
4719         free_hba(i);
4720 }
4721
4722 static struct pci_driver cciss_pci_driver = {
4723         .name = "cciss",
4724         .probe = cciss_init_one,
4725         .remove = __devexit_p(cciss_remove_one),
4726         .id_table = cciss_pci_device_id,        /* id_table */
4727         .shutdown = cciss_shutdown,
4728 };
4729
4730 /*
4731  *  This is it.  Register the PCI driver information for the cards we control
4732  *  the OS will call our registered routines when it finds one of our cards.
4733  */
4734 static int __init cciss_init(void)
4735 {
4736         int err;
4737
4738         /*
4739          * The hardware requires that commands are aligned on a 64-bit
4740          * boundary. Given that we use pci_alloc_consistent() to allocate an
4741          * array of them, the size must be a multiple of 8 bytes.
4742          */
4743         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4744         printk(KERN_INFO DRIVER_NAME "\n");
4745
4746         err = bus_register(&cciss_bus_type);
4747         if (err)
4748                 return err;
4749
4750         /* Start the scan thread */
4751         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4752         if (IS_ERR(cciss_scan_thread)) {
4753                 err = PTR_ERR(cciss_scan_thread);
4754                 goto err_bus_unregister;
4755         }
4756
4757         /* Register for our PCI devices */
4758         err = pci_register_driver(&cciss_pci_driver);
4759         if (err)
4760                 goto err_thread_stop;
4761
4762         return err;
4763
4764 err_thread_stop:
4765         kthread_stop(cciss_scan_thread);
4766 err_bus_unregister:
4767         bus_unregister(&cciss_bus_type);
4768
4769         return err;
4770 }
4771
4772 static void __exit cciss_cleanup(void)
4773 {
4774         int i;
4775
4776         pci_unregister_driver(&cciss_pci_driver);
4777         /* double check that all controller entrys have been removed */
4778         for (i = 0; i < MAX_CTLR; i++) {
4779                 if (hba[i] != NULL) {
4780                         printk(KERN_WARNING "cciss: had to remove"
4781                                " controller %d\n", i);
4782                         cciss_remove_one(hba[i]->pdev);
4783                 }
4784         }
4785         kthread_stop(cciss_scan_thread);
4786         remove_proc_entry("driver/cciss", NULL);
4787         bus_unregister(&cciss_bus_type);
4788 }
4789
4790 module_init(cciss_init);
4791 module_exit(cciss_cleanup);