git.oblomov.eu Git - linux-2.6/blob - 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/delay.h>
  30 #include <linux/major.h>
  31 #include <linux/fs.h>
  32 #include <linux/bio.h>
  33 #include <linux/blkpg.h>
  34 #include <linux/timer.h>
  35 #include <linux/proc_fs.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/init.h>
  38 #include <linux/hdreg.h>
  39 #include <linux/spinlock.h>
  40 #include <linux/compat.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/io.h>
  43
  44 #include <linux/dma-mapping.h>
  45 #include <linux/blkdev.h>
  46 #include <linux/genhd.h>
  47 #include <linux/completion.h>
  48 #include <scsi/scsi.h>
  49 #include <scsi/sg.h>
  50 #include <scsi/scsi_ioctl.h>
  51 #include <linux/cdrom.h>
  52 #include <linux/scatterlist.h>
  53 #include <linux/kthread.h>
  54
  55 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  56 #define DRIVER_NAME "HP CISS Driver (v 3.6.20)"
  57 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 20)
  58
  59 /* Embedded module documentation macros - see modules.h */
  60 MODULE_AUTHOR("Hewlett-Packard Company");
  61 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  62 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
  63                         " SA6i P600 P800 P400 P400i E200 E200i E500 P700m"
  64                         " Smart Array G2 Series SAS/SATA Controllers");
  65 MODULE_VERSION("3.6.20");
  66 MODULE_LICENSE("GPL");
  67
  68 #include "cciss_cmd.h"
  69 #include "cciss.h"
  70 #include <linux/cciss_ioctl.h>
  71
  72 /* define the PCI info for the cards we can control */
  73 static const struct pci_device_id cciss_pci_device_id[] = {
  74         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  75         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  76         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  77         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  78         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  79         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  80         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  83         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  84         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  85         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  86         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  87         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  88         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  89         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
  91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 101         {PCI_VENDOR_ID_HP,     PCI_ANY_ID,      PCI_ANY_ID, PCI_ANY_ID,
 102                 PCI_CLASS_STORAGE_RAID << 8, 0xffff << 8, 0},
 103         {0,}
 104 };
 105
 106 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 107
 108 /*  board_id = Subsystem Device ID & Vendor ID
 109  *  product = Marketing Name for the board
 110  *  access = Address of the struct of function pointers
 111  */
 112 static struct board_type products[] = {
 113         {0x40700E11, "Smart Array 5300", &SA5_access},
 114         {0x40800E11, "Smart Array 5i", &SA5B_access},
 115         {0x40820E11, "Smart Array 532", &SA5B_access},
 116         {0x40830E11, "Smart Array 5312", &SA5B_access},
 117         {0x409A0E11, "Smart Array 641", &SA5_access},
 118         {0x409B0E11, "Smart Array 642", &SA5_access},
 119         {0x409C0E11, "Smart Array 6400", &SA5_access},
 120         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 121         {0x40910E11, "Smart Array 6i", &SA5_access},
 122         {0x3225103C, "Smart Array P600", &SA5_access},
 123         {0x3223103C, "Smart Array P800", &SA5_access},
 124         {0x3234103C, "Smart Array P400", &SA5_access},
 125         {0x3235103C, "Smart Array P400i", &SA5_access},
 126         {0x3211103C, "Smart Array E200i", &SA5_access},
 127         {0x3212103C, "Smart Array E200", &SA5_access},
 128         {0x3213103C, "Smart Array E200i", &SA5_access},
 129         {0x3214103C, "Smart Array E200i", &SA5_access},
 130         {0x3215103C, "Smart Array E200i", &SA5_access},
 131         {0x3237103C, "Smart Array E500", &SA5_access},
 132         {0x323D103C, "Smart Array P700m", &SA5_access},
 133         {0x3241103C, "Smart Array P212", &SA5_access},
 134         {0x3243103C, "Smart Array P410", &SA5_access},
 135         {0x3245103C, "Smart Array P410i", &SA5_access},
 136         {0x3247103C, "Smart Array P411", &SA5_access},
 137         {0x3249103C, "Smart Array P812", &SA5_access},
 138         {0x324A103C, "Smart Array P712m", &SA5_access},
 139         {0x324B103C, "Smart Array P711m", &SA5_access},
 140         {0xFFFF103C, "Unknown Smart Array", &SA5_access},
 141 };
 142
 143 /* How long to wait (in milliseconds) for board to go into simple mode */
 144 #define MAX_CONFIG_WAIT 30000
 145 #define MAX_IOCTL_CONFIG_WAIT 1000
 146
 147 /*define how many times we will try a command because of bus resets */
 148 #define MAX_CMD_RETRIES 3
 149
 150 #define MAX_CTLR        32
 151
 152 /* Originally cciss driver only supports 8 major numbers */
 153 #define MAX_CTLR_ORIG   8
 154
 155 static ctlr_info_t *hba[MAX_CTLR];
 156
 157 static void do_cciss_request(struct request_queue *q);
 158 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
 159 static int cciss_open(struct block_device *bdev, fmode_t mode);
 160 static int cciss_release(struct gendisk *disk, fmode_t mode);
 161 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 162                        unsigned int cmd, unsigned long arg);
 163 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 164
 165 static int cciss_revalidate(struct gendisk *disk);
 166 static int rebuild_lun_table(ctlr_info_t *h, int first_time);
 167 static int deregister_disk(ctlr_info_t *h, int drv_index,
 168                            int clear_all);
 169
 170 static void cciss_read_capacity(int ctlr, int logvol, int withirq,
 171                         sector_t *total_size, unsigned int *block_size);
 172 static void cciss_read_capacity_16(int ctlr, int logvol, int withirq,
 173                         sector_t *total_size, unsigned int *block_size);
 174 static void cciss_geometry_inquiry(int ctlr, int logvol,
 175                         int withirq, sector_t total_size,
 176                         unsigned int block_size, InquiryData_struct *inq_buff,
 177                                    drive_info_struct *drv);
 178 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 179                                            __u32);
 180 static void start_io(ctlr_info_t *h);
 181 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
 182                    __u8 page_code, unsigned char *scsi3addr, int cmd_type);
 183 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 184                         __u8 page_code, unsigned char scsi3addr[],
 185                         int cmd_type);
 186 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 187         int attempt_retry);
 188 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 189
 190 static void fail_all_cmds(unsigned long ctlr);
 191 static int scan_thread(void *data);
 192 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 193
 194 #ifdef CONFIG_PROC_FS
 195 static void cciss_procinit(int i);
 196 #else
 197 static void cciss_procinit(int i)
 198 {
 199 }
 200 #endif                          /* CONFIG_PROC_FS */
 201
 202 #ifdef CONFIG_COMPAT
 203 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 204                               unsigned, unsigned long);
 205 #endif
 206
 207 static struct block_device_operations cciss_fops = {
 208         .owner = THIS_MODULE,
 209         .open = cciss_open,
 210         .release = cciss_release,
 211         .locked_ioctl = cciss_ioctl,
 212         .getgeo = cciss_getgeo,
 213 #ifdef CONFIG_COMPAT
 214         .compat_ioctl = cciss_compat_ioctl,
 215 #endif
 216         .revalidate_disk = cciss_revalidate,
 217 };
 218
 219 /*
 220  * Enqueuing and dequeuing functions for cmdlists.
 221  */
 222 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 223 {
 224         hlist_add_head(&c->list, list);
 225 }
 226
 227 static inline void removeQ(CommandList_struct *c)
 228 {
 229         if (WARN_ON(hlist_unhashed(&c->list)))
 230                 return;
 231
 232         hlist_del_init(&c->list);
 233 }
 234
 235 #include "cciss_scsi.c"         /* For SCSI tape support */
 236
 237 #define RAID_UNKNOWN 6
 238
 239 #ifdef CONFIG_PROC_FS
 240
 241 /*
 242  * Report information about this controller.
 243  */
 244 #define ENG_GIG 1000000000
 245 #define ENG_GIG_FACTOR (ENG_GIG/512)
 246 #define ENGAGE_SCSI     "engage scsi"
 247 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 248         "UNKNOWN"
 249 };
 250
 251 static struct proc_dir_entry *proc_cciss;
 252
 253 static void cciss_seq_show_header(struct seq_file *seq)
 254 {
 255         ctlr_info_t *h = seq->private;
 256
 257         seq_printf(seq, "%s: HP %s Controller\n"
 258                 "Board ID: 0x%08lx\n"
 259                 "Firmware Version: %c%c%c%c\n"
 260                 "IRQ: %d\n"
 261                 "Logical drives: %d\n"
 262                 "Current Q depth: %d\n"
 263                 "Current # commands on controller: %d\n"
 264                 "Max Q depth since init: %d\n"
 265                 "Max # commands on controller since init: %d\n"
 266                 "Max SG entries since init: %d\n",
 267                 h->devname,
 268                 h->product_name,
 269                 (unsigned long)h->board_id,
 270                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 271                 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
 272                 h->num_luns,
 273                 h->Qdepth, h->commands_outstanding,
 274                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 275
 276 #ifdef CONFIG_CISS_SCSI_TAPE
 277         cciss_seq_tape_report(seq, h->ctlr);
 278 #endif /* CONFIG_CISS_SCSI_TAPE */
 279 }
 280
 281 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 282 {
 283         ctlr_info_t *h = seq->private;
 284         unsigned ctlr = h->ctlr;
 285         unsigned long flags;
 286
 287         /* prevent displaying bogus info during configuration
 288          * or deconfiguration of a logical volume
 289          */
 290         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 291         if (h->busy_configuring) {
 292                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 293                 return ERR_PTR(-EBUSY);
 294         }
 295         h->busy_configuring = 1;
 296         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 297
 298         if (*pos == 0)
 299                 cciss_seq_show_header(seq);
 300
 301         return pos;
 302 }
 303
 304 static int cciss_seq_show(struct seq_file *seq, void *v)
 305 {
 306         sector_t vol_sz, vol_sz_frac;
 307         ctlr_info_t *h = seq->private;
 308         unsigned ctlr = h->ctlr;
 309         loff_t *pos = v;
 310         drive_info_struct *drv = &h->drv[*pos];
 311
 312         if (*pos > h->highest_lun)
 313                 return 0;
 314
 315         if (drv->heads == 0)
 316                 return 0;
 317
 318         vol_sz = drv->nr_blocks;
 319         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 320         vol_sz_frac *= 100;
 321         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 322
 323         if (drv->raid_level > 5)
 324                 drv->raid_level = RAID_UNKNOWN;
 325         seq_printf(seq, "cciss/c%dd%d:"
 326                         "\t%4u.%02uGB\tRAID %s\n",
 327                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 328                         raid_label[drv->raid_level]);
 329         return 0;
 330 }
 331
 332 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 333 {
 334         ctlr_info_t *h = seq->private;
 335
 336         if (*pos > h->highest_lun)
 337                 return NULL;
 338         *pos += 1;
 339
 340         return pos;
 341 }
 342
 343 static void cciss_seq_stop(struct seq_file *seq, void *v)
 344 {
 345         ctlr_info_t *h = seq->private;
 346
 347         /* Only reset h->busy_configuring if we succeeded in setting
 348          * it during cciss_seq_start. */
 349         if (v == ERR_PTR(-EBUSY))
 350                 return;
 351
 352         h->busy_configuring = 0;
 353 }
 354
 355 static struct seq_operations cciss_seq_ops = {
 356         .start = cciss_seq_start,
 357         .show  = cciss_seq_show,
 358         .next  = cciss_seq_next,
 359         .stop  = cciss_seq_stop,
 360 };
 361
 362 static int cciss_seq_open(struct inode *inode, struct file *file)
 363 {
 364         int ret = seq_open(file, &cciss_seq_ops);
 365         struct seq_file *seq = file->private_data;
 366
 367         if (!ret)
 368                 seq->private = PDE(inode)->data;
 369
 370         return ret;
 371 }
 372
 373 static ssize_t
 374 cciss_proc_write(struct file *file, const char __user *buf,
 375                  size_t length, loff_t *ppos)
 376 {
 377         int err;
 378         char *buffer;
 379
 380 #ifndef CONFIG_CISS_SCSI_TAPE
 381         return -EINVAL;
 382 #endif
 383
 384         if (!buf || length > PAGE_SIZE - 1)
 385                 return -EINVAL;
 386
 387         buffer = (char *)__get_free_page(GFP_KERNEL);
 388         if (!buffer)
 389                 return -ENOMEM;
 390
 391         err = -EFAULT;
 392         if (copy_from_user(buffer, buf, length))
 393                 goto out;
 394         buffer[length] = '\0';
 395
 396 #ifdef CONFIG_CISS_SCSI_TAPE
 397         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 398                 struct seq_file *seq = file->private_data;
 399                 ctlr_info_t *h = seq->private;
 400                 int rc;
 401
 402                 rc = cciss_engage_scsi(h->ctlr);
 403                 if (rc != 0)
 404                         err = -rc;
 405                 else
 406                         err = length;
 407         } else
 408 #endif /* CONFIG_CISS_SCSI_TAPE */
 409                 err = -EINVAL;
 410         /* might be nice to have "disengage" too, but it's not
 411            safely possible. (only 1 module use count, lock issues.) */
 412
 413 out:
 414         free_page((unsigned long)buffer);
 415         return err;
 416 }
 417
 418 static struct file_operations cciss_proc_fops = {
 419         .owner   = THIS_MODULE,
 420         .open    = cciss_seq_open,
 421         .read    = seq_read,
 422         .llseek  = seq_lseek,
 423         .release = seq_release,
 424         .write   = cciss_proc_write,
 425 };
 426
 427 static void __devinit cciss_procinit(int i)
 428 {
 429         struct proc_dir_entry *pde;
 430
 431         if (proc_cciss == NULL)
 432                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 433         if (!proc_cciss)
 434                 return;
 435         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 436                                         S_IROTH, proc_cciss,
 437                                         &cciss_proc_fops, hba[i]);
 438 }
 439 #endif                          /* CONFIG_PROC_FS */
 440
 441 #define MAX_PRODUCT_NAME_LEN 19
 442
 443 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 444 #define to_drv(n) container_of(n, drive_info_struct, dev)
 445
 446 static struct device_type cciss_host_type = {
 447         .name           = "cciss_host",
 448 };
 449
 450 static ssize_t dev_show_unique_id(struct device *dev,
 451                                  struct device_attribute *attr,
 452                                  char *buf)
 453 {
 454         drive_info_struct *drv = to_drv(dev);
 455         struct ctlr_info *h = to_hba(drv->dev.parent);
 456         __u8 sn[16];
 457         unsigned long flags;
 458         int ret = 0;
 459
 460         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 461         if (h->busy_configuring)
 462                 ret = -EBUSY;
 463         else
 464                 memcpy(sn, drv->serial_no, sizeof(sn));
 465         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 466
 467         if (ret)
 468                 return ret;
 469         else
 470                 return snprintf(buf, 16 * 2 + 2,
 471                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 472                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 473                                 sn[0], sn[1], sn[2], sn[3],
 474                                 sn[4], sn[5], sn[6], sn[7],
 475                                 sn[8], sn[9], sn[10], sn[11],
 476                                 sn[12], sn[13], sn[14], sn[15]);
 477 }
 478 DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 479
 480 static ssize_t dev_show_vendor(struct device *dev,
 481                                struct device_attribute *attr,
 482                                char *buf)
 483 {
 484         drive_info_struct *drv = to_drv(dev);
 485         struct ctlr_info *h = to_hba(drv->dev.parent);
 486         char vendor[VENDOR_LEN + 1];
 487         unsigned long flags;
 488         int ret = 0;
 489
 490         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 491         if (h->busy_configuring)
 492                 ret = -EBUSY;
 493         else
 494                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 495         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 496
 497         if (ret)
 498                 return ret;
 499         else
 500                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 501 }
 502 DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 503
 504 static ssize_t dev_show_model(struct device *dev,
 505                               struct device_attribute *attr,
 506                               char *buf)
 507 {
 508         drive_info_struct *drv = to_drv(dev);
 509         struct ctlr_info *h = to_hba(drv->dev.parent);
 510         char model[MODEL_LEN + 1];
 511         unsigned long flags;
 512         int ret = 0;
 513
 514         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 515         if (h->busy_configuring)
 516                 ret = -EBUSY;
 517         else
 518                 memcpy(model, drv->model, MODEL_LEN + 1);
 519         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 520
 521         if (ret)
 522                 return ret;
 523         else
 524                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 525 }
 526 DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 527
 528 static ssize_t dev_show_rev(struct device *dev,
 529                             struct device_attribute *attr,
 530                             char *buf)
 531 {
 532         drive_info_struct *drv = to_drv(dev);
 533         struct ctlr_info *h = to_hba(drv->dev.parent);
 534         char rev[REV_LEN + 1];
 535         unsigned long flags;
 536         int ret = 0;
 537
 538         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 539         if (h->busy_configuring)
 540                 ret = -EBUSY;
 541         else
 542                 memcpy(rev, drv->rev, REV_LEN + 1);
 543         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 544
 545         if (ret)
 546                 return ret;
 547         else
 548                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 549 }
 550 DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 551
 552 static struct attribute *cciss_dev_attrs[] = {
 553         &dev_attr_unique_id.attr,
 554         &dev_attr_model.attr,
 555         &dev_attr_vendor.attr,
 556         &dev_attr_rev.attr,
 557         NULL
 558 };
 559
 560 static struct attribute_group cciss_dev_attr_group = {
 561         .attrs = cciss_dev_attrs,
 562 };
 563
 564 static struct attribute_group *cciss_dev_attr_groups[] = {
 565         &cciss_dev_attr_group,
 566         NULL
 567 };
 568
 569 static struct device_type cciss_dev_type = {
 570         .name           = "cciss_device",
 571         .groups         = cciss_dev_attr_groups,
 572 };
 573
 574 static struct bus_type cciss_bus_type = {
 575         .name           = "cciss",
 576 };
 577
 578
 579 /*
 580  * Initialize sysfs entry for each controller.  This sets up and registers
 581  * the 'cciss#' directory for each individual controller under
 582  * /sys/bus/pci/devices/<dev>/.
 583  */
 584 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 585 {
 586         device_initialize(&h->dev);
 587         h->dev.type = &cciss_host_type;
 588         h->dev.bus = &cciss_bus_type;
 589         dev_set_name(&h->dev, "%s", h->devname);
 590         h->dev.parent = &h->pdev->dev;
 591
 592         return device_add(&h->dev);
 593 }
 594
 595 /*
 596  * Remove sysfs entries for an hba.
 597  */
 598 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 599 {
 600         device_del(&h->dev);
 601 }
 602
 603 /*
 604  * Initialize sysfs for each logical drive.  This sets up and registers
 605  * the 'c#d#' directory for each individual logical drive under
 606  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 607  * /sys/block/cciss!c#d# to this entry.
 608  */
 609 static int cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 610                                        drive_info_struct *drv,
 611                                        int drv_index)
 612 {
 613         device_initialize(&drv->dev);
 614         drv->dev.type = &cciss_dev_type;
 615         drv->dev.bus = &cciss_bus_type;
 616         dev_set_name(&drv->dev, "c%dd%d", h->ctlr, drv_index);
 617         drv->dev.parent = &h->dev;
 618         return device_add(&drv->dev);
 619 }
 620
 621 /*
 622  * Remove sysfs entries for a logical drive.
 623  */
 624 static void cciss_destroy_ld_sysfs_entry(drive_info_struct *drv)
 625 {
 626         device_del(&drv->dev);
 627 }
 628
 629 /*
 630  * For operations that cannot sleep, a command block is allocated at init,
 631  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 632  * which ones are free or in use.  For operations that can wait for kmalloc
 633  * to possible sleep, this routine can be called with get_from_pool set to 0.
 634  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 635  */
 636 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 637 {
 638         CommandList_struct *c;
 639         int i;
 640         u64bit temp64;
 641         dma_addr_t cmd_dma_handle, err_dma_handle;
 642
 643         if (!get_from_pool) {
 644                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 645                         sizeof(CommandList_struct), &cmd_dma_handle);
 646                 if (c == NULL)
 647                         return NULL;
 648                 memset(c, 0, sizeof(CommandList_struct));
 649
 650                 c->cmdindex = -1;
 651
 652                 c->err_info = (ErrorInfo_struct *)
 653                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 654                             &err_dma_handle);
 655
 656                 if (c->err_info == NULL) {
 657                         pci_free_consistent(h->pdev,
 658                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 659                         return NULL;
 660                 }
 661                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 662         } else {                /* get it out of the controllers pool */
 663
 664                 do {
 665                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 666                         if (i == h->nr_cmds)
 667                                 return NULL;
 668                 } while (test_and_set_bit
 669                          (i & (BITS_PER_LONG - 1),
 670                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 671 #ifdef CCISS_DEBUG
 672                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 673 #endif
 674                 c = h->cmd_pool + i;
 675                 memset(c, 0, sizeof(CommandList_struct));
 676                 cmd_dma_handle = h->cmd_pool_dhandle
 677                     + i * sizeof(CommandList_struct);
 678                 c->err_info = h->errinfo_pool + i;
 679                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 680                 err_dma_handle = h->errinfo_pool_dhandle
 681                     + i * sizeof(ErrorInfo_struct);
 682                 h->nr_allocs++;
 683
 684                 c->cmdindex = i;
 685         }
 686
 687         INIT_HLIST_NODE(&c->list);
 688         c->busaddr = (__u32) cmd_dma_handle;
 689         temp64.val = (__u64) err_dma_handle;
 690         c->ErrDesc.Addr.lower = temp64.val32.lower;
 691         c->ErrDesc.Addr.upper = temp64.val32.upper;
 692         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 693
 694         c->ctlr = h->ctlr;
 695         return c;
 696 }
 697
 698 /*
 699  * Frees a command block that was previously allocated with cmd_alloc().
 700  */
 701 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 702 {
 703         int i;
 704         u64bit temp64;
 705
 706         if (!got_from_pool) {
 707                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 708                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 709                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 710                                     c->err_info, (dma_addr_t) temp64.val);
 711                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 712                                     c, (dma_addr_t) c->busaddr);
 713         } else {
 714                 i = c - h->cmd_pool;
 715                 clear_bit(i & (BITS_PER_LONG - 1),
 716                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 717                 h->nr_frees++;
 718         }
 719 }
 720
 721 static inline ctlr_info_t *get_host(struct gendisk *disk)
 722 {
 723         return disk->queue->queuedata;
 724 }
 725
 726 static inline drive_info_struct *get_drv(struct gendisk *disk)
 727 {
 728         return disk->private_data;
 729 }
 730
 731 /*
 732  * Open.  Make sure the device is really there.
 733  */
 734 static int cciss_open(struct block_device *bdev, fmode_t mode)
 735 {
 736         ctlr_info_t *host = get_host(bdev->bd_disk);
 737         drive_info_struct *drv = get_drv(bdev->bd_disk);
 738
 739 #ifdef CCISS_DEBUG
 740         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
 741 #endif                          /* CCISS_DEBUG */
 742
 743         if (host->busy_initializing || drv->busy_configuring)
 744                 return -EBUSY;
 745         /*
 746          * Root is allowed to open raw volume zero even if it's not configured
 747          * so array config can still work. Root is also allowed to open any
 748          * volume that has a LUN ID, so it can issue IOCTL to reread the
 749          * disk information.  I don't think I really like this
 750          * but I'm already using way to many device nodes to claim another one
 751          * for "raw controller".
 752          */
 753         if (drv->heads == 0) {
 754                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
 755                         /* if not node 0 make sure it is a partition = 0 */
 756                         if (MINOR(bdev->bd_dev) & 0x0f) {
 757                                 return -ENXIO;
 758                                 /* if it is, make sure we have a LUN ID */
 759                         } else if (drv->LunID == 0) {
 760                                 return -ENXIO;
 761                         }
 762                 }
 763                 if (!capable(CAP_SYS_ADMIN))
 764                         return -EPERM;
 765         }
 766         drv->usage_count++;
 767         host->usage_count++;
 768         return 0;
 769 }
 770
 771 /*
 772  * Close.  Sync first.
 773  */
 774 static int cciss_release(struct gendisk *disk, fmode_t mode)
 775 {
 776         ctlr_info_t *host = get_host(disk);
 777         drive_info_struct *drv = get_drv(disk);
 778
 779 #ifdef CCISS_DEBUG
 780         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
 781 #endif                          /* CCISS_DEBUG */
 782
 783         drv->usage_count--;
 784         host->usage_count--;
 785         return 0;
 786 }
 787
 788 #ifdef CONFIG_COMPAT
 789
 790 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 791                     unsigned cmd, unsigned long arg)
 792 {
 793         int ret;
 794         lock_kernel();
 795         ret = cciss_ioctl(bdev, mode, cmd, arg);
 796         unlock_kernel();
 797         return ret;
 798 }
 799
 800 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 801                                   unsigned cmd, unsigned long arg);
 802 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
 803                                       unsigned cmd, unsigned long arg);
 804
 805 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
 806                               unsigned cmd, unsigned long arg)
 807 {
 808         switch (cmd) {
 809         case CCISS_GETPCIINFO:
 810         case CCISS_GETINTINFO:
 811         case CCISS_SETINTINFO:
 812         case CCISS_GETNODENAME:
 813         case CCISS_SETNODENAME:
 814         case CCISS_GETHEARTBEAT:
 815         case CCISS_GETBUSTYPES:
 816         case CCISS_GETFIRMVER:
 817         case CCISS_GETDRIVVER:
 818         case CCISS_REVALIDVOLS:
 819         case CCISS_DEREGDISK:
 820         case CCISS_REGNEWDISK:
 821         case CCISS_REGNEWD:
 822         case CCISS_RESCANDISK:
 823         case CCISS_GETLUNINFO:
 824                 return do_ioctl(bdev, mode, cmd, arg);
 825
 826         case CCISS_PASSTHRU32:
 827                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
 828         case CCISS_BIG_PASSTHRU32:
 829                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
 830
 831         default:
 832                 return -ENOIOCTLCMD;
 833         }
 834 }
 835
 836 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 837                                   unsigned cmd, unsigned long arg)
 838 {
 839         IOCTL32_Command_struct __user *arg32 =
 840             (IOCTL32_Command_struct __user *) arg;
 841         IOCTL_Command_struct arg64;
 842         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
 843         int err;
 844         u32 cp;
 845
 846         err = 0;
 847         err |=
 848             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
 849                            sizeof(arg64.LUN_info));
 850         err |=
 851             copy_from_user(&arg64.Request, &arg32->Request,
 852                            sizeof(arg64.Request));
 853         err |=
 854             copy_from_user(&arg64.error_info, &arg32->error_info,
 855                            sizeof(arg64.error_info));
 856         err |= get_user(arg64.buf_size, &arg32->buf_size);
 857         err |= get_user(cp, &arg32->buf);
 858         arg64.buf = compat_ptr(cp);
 859         err |= copy_to_user(p, &arg64, sizeof(arg64));
 860
 861         if (err)
 862                 return -EFAULT;
 863
 864         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
 865         if (err)
 866                 return err;
 867         err |=
 868             copy_in_user(&arg32->error_info, &p->error_info,
 869                          sizeof(arg32->error_info));
 870         if (err)
 871                 return -EFAULT;
 872         return err;
 873 }
 874
 875 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
 876                                       unsigned cmd, unsigned long arg)
 877 {
 878         BIG_IOCTL32_Command_struct __user *arg32 =
 879             (BIG_IOCTL32_Command_struct __user *) arg;
 880         BIG_IOCTL_Command_struct arg64;
 881         BIG_IOCTL_Command_struct __user *p =
 882             compat_alloc_user_space(sizeof(arg64));
 883         int err;
 884         u32 cp;
 885
 886         err = 0;
 887         err |=
 888             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
 889                            sizeof(arg64.LUN_info));
 890         err |=
 891             copy_from_user(&arg64.Request, &arg32->Request,
 892                            sizeof(arg64.Request));
 893         err |=
 894             copy_from_user(&arg64.error_info, &arg32->error_info,
 895                            sizeof(arg64.error_info));
 896         err |= get_user(arg64.buf_size, &arg32->buf_size);
 897         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
 898         err |= get_user(cp, &arg32->buf);
 899         arg64.buf = compat_ptr(cp);
 900         err |= copy_to_user(p, &arg64, sizeof(arg64));
 901
 902         if (err)
 903                 return -EFAULT;
 904
 905         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
 906         if (err)
 907                 return err;
 908         err |=
 909             copy_in_user(&arg32->error_info, &p->error_info,
 910                          sizeof(arg32->error_info));
 911         if (err)
 912                 return -EFAULT;
 913         return err;
 914 }
 915 #endif
 916
 917 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 918 {
 919         drive_info_struct *drv = get_drv(bdev->bd_disk);
 920
 921         if (!drv->cylinders)
 922                 return -ENXIO;
 923
 924         geo->heads = drv->heads;
 925         geo->sectors = drv->sectors;
 926         geo->cylinders = drv->cylinders;
 927         return 0;
 928 }
 929
 930 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
 931 {
 932         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
 933                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
 934                 (void)check_for_unit_attention(host, c);
 935 }
 936 /*
 937  * ioctl
 938  */
 939 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 940                        unsigned int cmd, unsigned long arg)
 941 {
 942         struct gendisk *disk = bdev->bd_disk;
 943         ctlr_info_t *host = get_host(disk);
 944         drive_info_struct *drv = get_drv(disk);
 945         int ctlr = host->ctlr;
 946         void __user *argp = (void __user *)arg;
 947
 948 #ifdef CCISS_DEBUG
 949         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
 950 #endif                          /* CCISS_DEBUG */
 951
 952         switch (cmd) {
 953         case CCISS_GETPCIINFO:
 954                 {
 955                         cciss_pci_info_struct pciinfo;
 956
 957                         if (!arg)
 958                                 return -EINVAL;
 959                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
 960                         pciinfo.bus = host->pdev->bus->number;
 961                         pciinfo.dev_fn = host->pdev->devfn;
 962                         pciinfo.board_id = host->board_id;
 963                         if (copy_to_user
 964                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
 965                                 return -EFAULT;
 966                         return 0;
 967                 }
 968         case CCISS_GETINTINFO:
 969                 {
 970                         cciss_coalint_struct intinfo;
 971                         if (!arg)
 972                                 return -EINVAL;
 973                         intinfo.delay =
 974                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
 975                         intinfo.count =
 976                             readl(&host->cfgtable->HostWrite.CoalIntCount);
 977                         if (copy_to_user
 978                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
 979                                 return -EFAULT;
 980                         return 0;
 981                 }
 982         case CCISS_SETINTINFO:
 983                 {
 984                         cciss_coalint_struct intinfo;
 985                         unsigned long flags;
 986                         int i;
 987
 988                         if (!arg)
 989                                 return -EINVAL;
 990                         if (!capable(CAP_SYS_ADMIN))
 991                                 return -EPERM;
 992                         if (copy_from_user
 993                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
 994                                 return -EFAULT;
 995                         if ((intinfo.delay == 0) && (intinfo.count == 0))
 996                         {
 997 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
 998                                 return -EINVAL;
 999                         }
1000                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1001                         /* Update the field, and then ring the doorbell */
1002                         writel(intinfo.delay,
1003                                &(host->cfgtable->HostWrite.CoalIntDelay));
1004                         writel(intinfo.count,
1005                                &(host->cfgtable->HostWrite.CoalIntCount));
1006                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1007
1008                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1009                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1010                                       & CFGTBL_ChangeReq))
1011                                         break;
1012                                 /* delay and try again */
1013                                 udelay(1000);
1014                         }
1015                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1016                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1017                                 return -EAGAIN;
1018                         return 0;
1019                 }
1020         case CCISS_GETNODENAME:
1021                 {
1022                         NodeName_type NodeName;
1023                         int i;
1024
1025                         if (!arg)
1026                                 return -EINVAL;
1027                         for (i = 0; i < 16; i++)
1028                                 NodeName[i] =
1029                                     readb(&host->cfgtable->ServerName[i]);
1030                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1031                                 return -EFAULT;
1032                         return 0;
1033                 }
1034         case CCISS_SETNODENAME:
1035                 {
1036                         NodeName_type NodeName;
1037                         unsigned long flags;
1038                         int i;
1039
1040                         if (!arg)
1041                                 return -EINVAL;
1042                         if (!capable(CAP_SYS_ADMIN))
1043                                 return -EPERM;
1044
1045                         if (copy_from_user
1046                             (NodeName, argp, sizeof(NodeName_type)))
1047                                 return -EFAULT;
1048
1049                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1050
1051                         /* Update the field, and then ring the doorbell */
1052                         for (i = 0; i < 16; i++)
1053                                 writeb(NodeName[i],
1054                                        &host->cfgtable->ServerName[i]);
1055
1056                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1057
1058                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1059                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1060                                       & CFGTBL_ChangeReq))
1061                                         break;
1062                                 /* delay and try again */
1063                                 udelay(1000);
1064                         }
1065                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1066                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1067                                 return -EAGAIN;
1068                         return 0;
1069                 }
1070
1071         case CCISS_GETHEARTBEAT:
1072                 {
1073                         Heartbeat_type heartbeat;
1074
1075                         if (!arg)
1076                                 return -EINVAL;
1077                         heartbeat = readl(&host->cfgtable->HeartBeat);
1078                         if (copy_to_user
1079                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1080                                 return -EFAULT;
1081                         return 0;
1082                 }
1083         case CCISS_GETBUSTYPES:
1084                 {
1085                         BusTypes_type BusTypes;
1086
1087                         if (!arg)
1088                                 return -EINVAL;
1089                         BusTypes = readl(&host->cfgtable->BusTypes);
1090                         if (copy_to_user
1091                             (argp, &BusTypes, sizeof(BusTypes_type)))
1092                                 return -EFAULT;
1093                         return 0;
1094                 }
1095         case CCISS_GETFIRMVER:
1096                 {
1097                         FirmwareVer_type firmware;
1098
1099                         if (!arg)
1100                                 return -EINVAL;
1101                         memcpy(firmware, host->firm_ver, 4);
1102
1103                         if (copy_to_user
1104                             (argp, firmware, sizeof(FirmwareVer_type)))
1105                                 return -EFAULT;
1106                         return 0;
1107                 }
1108         case CCISS_GETDRIVVER:
1109                 {
1110                         DriverVer_type DriverVer = DRIVER_VERSION;
1111
1112                         if (!arg)
1113                                 return -EINVAL;
1114
1115                         if (copy_to_user
1116                             (argp, &DriverVer, sizeof(DriverVer_type)))
1117                                 return -EFAULT;
1118                         return 0;
1119                 }
1120
1121         case CCISS_DEREGDISK:
1122         case CCISS_REGNEWD:
1123         case CCISS_REVALIDVOLS:
1124                 return rebuild_lun_table(host, 0);
1125
1126         case CCISS_GETLUNINFO:{
1127                         LogvolInfo_struct luninfo;
1128
1129                         luninfo.LunID = drv->LunID;
1130                         luninfo.num_opens = drv->usage_count;
1131                         luninfo.num_parts = 0;
1132                         if (copy_to_user(argp, &luninfo,
1133                                          sizeof(LogvolInfo_struct)))
1134                                 return -EFAULT;
1135                         return 0;
1136                 }
1137         case CCISS_PASSTHRU:
1138                 {
1139                         IOCTL_Command_struct iocommand;
1140                         CommandList_struct *c;
1141                         char *buff = NULL;
1142                         u64bit temp64;
1143                         unsigned long flags;
1144                         DECLARE_COMPLETION_ONSTACK(wait);
1145
1146                         if (!arg)
1147                                 return -EINVAL;
1148
1149                         if (!capable(CAP_SYS_RAWIO))
1150                                 return -EPERM;
1151
1152                         if (copy_from_user
1153                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1154                                 return -EFAULT;
1155                         if ((iocommand.buf_size < 1) &&
1156                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1157                                 return -EINVAL;
1158                         }
1159 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1160                         /* Check kmalloc limits */
1161                         if (iocommand.buf_size > 128000)
1162                                 return -EINVAL;
1163 #endif
1164                         if (iocommand.buf_size > 0) {
1165                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1166                                 if (buff == NULL)
1167                                         return -EFAULT;
1168                         }
1169                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1170                                 /* Copy the data into the buffer we created */
1171                                 if (copy_from_user
1172                                     (buff, iocommand.buf, iocommand.buf_size)) {
1173                                         kfree(buff);
1174                                         return -EFAULT;
1175                                 }
1176                         } else {
1177                                 memset(buff, 0, iocommand.buf_size);
1178                         }
1179                         if ((c = cmd_alloc(host, 0)) == NULL) {
1180                                 kfree(buff);
1181                                 return -ENOMEM;
1182                         }
1183                         // Fill in the command type
1184                         c->cmd_type = CMD_IOCTL_PEND;
1185                         // Fill in Command Header
1186                         c->Header.ReplyQueue = 0;       // unused in simple mode
1187                         if (iocommand.buf_size > 0)     // buffer to fill
1188                         {
1189                                 c->Header.SGList = 1;
1190                                 c->Header.SGTotal = 1;
1191                         } else  // no buffers to fill
1192                         {
1193                                 c->Header.SGList = 0;
1194                                 c->Header.SGTotal = 0;
1195                         }
1196                         c->Header.LUN = iocommand.LUN_info;
1197                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
1198
1199                         // Fill in Request block
1200                         c->Request = iocommand.Request;
1201
1202                         // Fill in the scatter gather information
1203                         if (iocommand.buf_size > 0) {
1204                                 temp64.val = pci_map_single(host->pdev, buff,
1205                                         iocommand.buf_size,
1206                                         PCI_DMA_BIDIRECTIONAL);
1207                                 c->SG[0].Addr.lower = temp64.val32.lower;
1208                                 c->SG[0].Addr.upper = temp64.val32.upper;
1209                                 c->SG[0].Len = iocommand.buf_size;
1210                                 c->SG[0].Ext = 0;       // we are not chaining
1211                         }
1212                         c->waiting = &wait;
1213
1214                         /* Put the request on the tail of the request queue */
1215                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1216                         addQ(&host->reqQ, c);
1217                         host->Qdepth++;
1218                         start_io(host);
1219                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1220
1221                         wait_for_completion(&wait);
1222
1223                         /* unlock the buffers from DMA */
1224                         temp64.val32.lower = c->SG[0].Addr.lower;
1225                         temp64.val32.upper = c->SG[0].Addr.upper;
1226                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1227                                          iocommand.buf_size,
1228                                          PCI_DMA_BIDIRECTIONAL);
1229
1230                         check_ioctl_unit_attention(host, c);
1231
1232                         /* Copy the error information out */
1233                         iocommand.error_info = *(c->err_info);
1234                         if (copy_to_user
1235                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1236                                 kfree(buff);
1237                                 cmd_free(host, c, 0);
1238                                 return -EFAULT;
1239                         }
1240
1241                         if (iocommand.Request.Type.Direction == XFER_READ) {
1242                                 /* Copy the data out of the buffer we created */
1243                                 if (copy_to_user
1244                                     (iocommand.buf, buff, iocommand.buf_size)) {
1245                                         kfree(buff);
1246                                         cmd_free(host, c, 0);
1247                                         return -EFAULT;
1248                                 }
1249                         }
1250                         kfree(buff);
1251                         cmd_free(host, c, 0);
1252                         return 0;
1253                 }
1254         case CCISS_BIG_PASSTHRU:{
1255                         BIG_IOCTL_Command_struct *ioc;
1256                         CommandList_struct *c;
1257                         unsigned char **buff = NULL;
1258                         int *buff_size = NULL;
1259                         u64bit temp64;
1260                         unsigned long flags;
1261                         BYTE sg_used = 0;
1262                         int status = 0;
1263                         int i;
1264                         DECLARE_COMPLETION_ONSTACK(wait);
1265                         __u32 left;
1266                         __u32 sz;
1267                         BYTE __user *data_ptr;
1268
1269                         if (!arg)
1270                                 return -EINVAL;
1271                         if (!capable(CAP_SYS_RAWIO))
1272                                 return -EPERM;
1273                         ioc = (BIG_IOCTL_Command_struct *)
1274                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1275                         if (!ioc) {
1276                                 status = -ENOMEM;
1277                                 goto cleanup1;
1278                         }
1279                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1280                                 status = -EFAULT;
1281                                 goto cleanup1;
1282                         }
1283                         if ((ioc->buf_size < 1) &&
1284                             (ioc->Request.Type.Direction != XFER_NONE)) {
1285                                 status = -EINVAL;
1286                                 goto cleanup1;
1287                         }
1288                         /* Check kmalloc limits  using all SGs */
1289                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1290                                 status = -EINVAL;
1291                                 goto cleanup1;
1292                         }
1293                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1294                                 status = -EINVAL;
1295                                 goto cleanup1;
1296                         }
1297                         buff =
1298                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1299                         if (!buff) {
1300                                 status = -ENOMEM;
1301                                 goto cleanup1;
1302                         }
1303                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1304                                                    GFP_KERNEL);
1305                         if (!buff_size) {
1306                                 status = -ENOMEM;
1307                                 goto cleanup1;
1308                         }
1309                         left = ioc->buf_size;
1310                         data_ptr = ioc->buf;
1311                         while (left) {
1312                                 sz = (left >
1313                                       ioc->malloc_size) ? ioc->
1314                                     malloc_size : left;
1315                                 buff_size[sg_used] = sz;
1316                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1317                                 if (buff[sg_used] == NULL) {
1318                                         status = -ENOMEM;
1319                                         goto cleanup1;
1320                                 }
1321                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1322                                         if (copy_from_user
1323                                             (buff[sg_used], data_ptr, sz)) {
1324                                                 status = -EFAULT;
1325                                                 goto cleanup1;
1326                                         }
1327                                 } else {
1328                                         memset(buff[sg_used], 0, sz);
1329                                 }
1330                                 left -= sz;
1331                                 data_ptr += sz;
1332                                 sg_used++;
1333                         }
1334                         if ((c = cmd_alloc(host, 0)) == NULL) {
1335                                 status = -ENOMEM;
1336                                 goto cleanup1;
1337                         }
1338                         c->cmd_type = CMD_IOCTL_PEND;
1339                         c->Header.ReplyQueue = 0;
1340
1341                         if (ioc->buf_size > 0) {
1342                                 c->Header.SGList = sg_used;
1343                                 c->Header.SGTotal = sg_used;
1344                         } else {
1345                                 c->Header.SGList = 0;
1346                                 c->Header.SGTotal = 0;
1347                         }
1348                         c->Header.LUN = ioc->LUN_info;
1349                         c->Header.Tag.lower = c->busaddr;
1350
1351                         c->Request = ioc->Request;
1352                         if (ioc->buf_size > 0) {
1353                                 int i;
1354                                 for (i = 0; i < sg_used; i++) {
1355                                         temp64.val =
1356                                             pci_map_single(host->pdev, buff[i],
1357                                                     buff_size[i],
1358                                                     PCI_DMA_BIDIRECTIONAL);
1359                                         c->SG[i].Addr.lower =
1360                                             temp64.val32.lower;
1361                                         c->SG[i].Addr.upper =
1362                                             temp64.val32.upper;
1363                                         c->SG[i].Len = buff_size[i];
1364                                         c->SG[i].Ext = 0;       /* we are not chaining */
1365                                 }
1366                         }
1367                         c->waiting = &wait;
1368                         /* Put the request on the tail of the request queue */
1369                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1370                         addQ(&host->reqQ, c);
1371                         host->Qdepth++;
1372                         start_io(host);
1373                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1374                         wait_for_completion(&wait);
1375                         /* unlock the buffers from DMA */
1376                         for (i = 0; i < sg_used; i++) {
1377                                 temp64.val32.lower = c->SG[i].Addr.lower;
1378                                 temp64.val32.upper = c->SG[i].Addr.upper;
1379                                 pci_unmap_single(host->pdev,
1380                                         (dma_addr_t) temp64.val, buff_size[i],
1381                                         PCI_DMA_BIDIRECTIONAL);
1382                         }
1383                         check_ioctl_unit_attention(host, c);
1384                         /* Copy the error information out */
1385                         ioc->error_info = *(c->err_info);
1386                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1387                                 cmd_free(host, c, 0);
1388                                 status = -EFAULT;
1389                                 goto cleanup1;
1390                         }
1391                         if (ioc->Request.Type.Direction == XFER_READ) {
1392                                 /* Copy the data out of the buffer we created */
1393                                 BYTE __user *ptr = ioc->buf;
1394                                 for (i = 0; i < sg_used; i++) {
1395                                         if (copy_to_user
1396                                             (ptr, buff[i], buff_size[i])) {
1397                                                 cmd_free(host, c, 0);
1398                                                 status = -EFAULT;
1399                                                 goto cleanup1;
1400                                         }
1401                                         ptr += buff_size[i];
1402                                 }
1403                         }
1404                         cmd_free(host, c, 0);
1405                         status = 0;
1406                       cleanup1:
1407                         if (buff) {
1408                                 for (i = 0; i < sg_used; i++)
1409                                         kfree(buff[i]);
1410                                 kfree(buff);
1411                         }
1412                         kfree(buff_size);
1413                         kfree(ioc);
1414                         return status;
1415                 }
1416
1417         /* scsi_cmd_ioctl handles these, below, though some are not */
1418         /* very meaningful for cciss.  SG_IO is the main one people want. */
1419
1420         case SG_GET_VERSION_NUM:
1421         case SG_SET_TIMEOUT:
1422         case SG_GET_TIMEOUT:
1423         case SG_GET_RESERVED_SIZE:
1424         case SG_SET_RESERVED_SIZE:
1425         case SG_EMULATED_HOST:
1426         case SG_IO:
1427         case SCSI_IOCTL_SEND_COMMAND:
1428                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1429
1430         /* scsi_cmd_ioctl would normally handle these, below, but */
1431         /* they aren't a good fit for cciss, as CD-ROMs are */
1432         /* not supported, and we don't have any bus/target/lun */
1433         /* which we present to the kernel. */
1434
1435         case CDROM_SEND_PACKET:
1436         case CDROMCLOSETRAY:
1437         case CDROMEJECT:
1438         case SCSI_IOCTL_GET_IDLUN:
1439         case SCSI_IOCTL_GET_BUS_NUMBER:
1440         default:
1441                 return -ENOTTY;
1442         }
1443 }
1444
1445 static void cciss_check_queues(ctlr_info_t *h)
1446 {
1447         int start_queue = h->next_to_run;
1448         int i;
1449
1450         /* check to see if we have maxed out the number of commands that can
1451          * be placed on the queue.  If so then exit.  We do this check here
1452          * in case the interrupt we serviced was from an ioctl and did not
1453          * free any new commands.
1454          */
1455         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1456                 return;
1457
1458         /* We have room on the queue for more commands.  Now we need to queue
1459          * them up.  We will also keep track of the next queue to run so
1460          * that every queue gets a chance to be started first.
1461          */
1462         for (i = 0; i < h->highest_lun + 1; i++) {
1463                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1464                 /* make sure the disk has been added and the drive is real
1465                  * because this can be called from the middle of init_one.
1466                  */
1467                 if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
1468                         continue;
1469                 blk_start_queue(h->gendisk[curr_queue]->queue);
1470
1471                 /* check to see if we have maxed out the number of commands
1472                  * that can be placed on the queue.
1473                  */
1474                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1475                         if (curr_queue == start_queue) {
1476                                 h->next_to_run =
1477                                     (start_queue + 1) % (h->highest_lun + 1);
1478                                 break;
1479                         } else {
1480                                 h->next_to_run = curr_queue;
1481                                 break;
1482                         }
1483                 }
1484         }
1485 }
1486
1487 static void cciss_softirq_done(struct request *rq)
1488 {
1489         CommandList_struct *cmd = rq->completion_data;
1490         ctlr_info_t *h = hba[cmd->ctlr];
1491         unsigned long flags;
1492         u64bit temp64;
1493         int i, ddir;
1494
1495         if (cmd->Request.Type.Direction == XFER_READ)
1496                 ddir = PCI_DMA_FROMDEVICE;
1497         else
1498                 ddir = PCI_DMA_TODEVICE;
1499
1500         /* command did not need to be retried */
1501         /* unmap the DMA mapping for all the scatter gather elements */
1502         for (i = 0; i < cmd->Header.SGList; i++) {
1503                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1504                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1505                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1506         }
1507
1508 #ifdef CCISS_DEBUG
1509         printk("Done with %p\n", rq);
1510 #endif                          /* CCISS_DEBUG */
1511
1512         /* set the residual count for pc requests */
1513         if (blk_pc_request(rq))
1514                 rq->resid_len = cmd->err_info->ResidualCnt;
1515
1516         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1517
1518         spin_lock_irqsave(&h->lock, flags);
1519         cmd_free(h, cmd, 1);
1520         cciss_check_queues(h);
1521         spin_unlock_irqrestore(&h->lock, flags);
1522 }
1523
1524 static void log_unit_to_scsi3addr(ctlr_info_t *h, unsigned char scsi3addr[],
1525         uint32_t log_unit)
1526 {
1527         log_unit = h->drv[log_unit].LunID & 0x03fff;
1528         memset(&scsi3addr[4], 0, 4);
1529         memcpy(&scsi3addr[0], &log_unit, 4);
1530         scsi3addr[3] |= 0x40;
1531 }
1532
1533 /* This function gets the SCSI vendor, model, and revision of a logical drive
1534  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1535  * they cannot be read.
1536  */
1537 static void cciss_get_device_descr(int ctlr, int logvol, int withirq,
1538                                    char *vendor, char *model, char *rev)
1539 {
1540         int rc;
1541         InquiryData_struct *inq_buf;
1542         unsigned char scsi3addr[8];
1543
1544         *vendor = '\0';
1545         *model = '\0';
1546         *rev = '\0';
1547
1548         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1549         if (!inq_buf)
1550                 return;
1551
1552         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1553         if (withirq)
1554                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf,
1555                              sizeof(InquiryData_struct), 0,
1556                                 scsi3addr, TYPE_CMD);
1557         else
1558                 rc = sendcmd(CISS_INQUIRY, ctlr, inq_buf,
1559                              sizeof(InquiryData_struct), 0,
1560                                 scsi3addr, TYPE_CMD);
1561         if (rc == IO_OK) {
1562                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1563                 vendor[VENDOR_LEN] = '\0';
1564                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1565                 model[MODEL_LEN] = '\0';
1566                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1567                 rev[REV_LEN] = '\0';
1568         }
1569
1570         kfree(inq_buf);
1571         return;
1572 }
1573
1574 /* This function gets the serial number of a logical drive via
1575  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1576  * number cannot be had, for whatever reason, 16 bytes of 0xff
1577  * are returned instead.
1578  */
1579 static void cciss_get_serial_no(int ctlr, int logvol, int withirq,
1580                                 unsigned char *serial_no, int buflen)
1581 {
1582 #define PAGE_83_INQ_BYTES 64
1583         int rc;
1584         unsigned char *buf;
1585         unsigned char scsi3addr[8];
1586
1587         if (buflen > 16)
1588                 buflen = 16;
1589         memset(serial_no, 0xff, buflen);
1590         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1591         if (!buf)
1592                 return;
1593         memset(serial_no, 0, buflen);
1594         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1595         if (withirq)
1596                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1597                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1598         else
1599                 rc = sendcmd(CISS_INQUIRY, ctlr, buf,
1600                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1601         if (rc == IO_OK)
1602                 memcpy(serial_no, &buf[8], buflen);
1603         kfree(buf);
1604         return;
1605 }
1606
1607 static void cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1608                                 int drv_index)
1609 {
1610         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1611         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1612         disk->major = h->major;
1613         disk->first_minor = drv_index << NWD_SHIFT;
1614         disk->fops = &cciss_fops;
1615         disk->private_data = &h->drv[drv_index];
1616         disk->driverfs_dev = &h->drv[drv_index].dev;
1617
1618         /* Set up queue information */
1619         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1620
1621         /* This is a hardware imposed limit. */
1622         blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1623
1624         /* This is a limit in the driver and could be eliminated. */
1625         blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1626
1627         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1628
1629         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1630
1631         disk->queue->queuedata = h;
1632
1633         blk_queue_logical_block_size(disk->queue,
1634                                      h->drv[drv_index].block_size);
1635
1636         /* Make sure all queue data is written out before */
1637         /* setting h->drv[drv_index].queue, as setting this */
1638         /* allows the interrupt handler to start the queue */
1639         wmb();
1640         h->drv[drv_index].queue = disk->queue;
1641         add_disk(disk);
1642 }
1643
1644 /* This function will check the usage_count of the drive to be updated/added.
1645  * If the usage_count is zero and it is a heretofore unknown drive, or,
1646  * the drive's capacity, geometry, or serial number has changed,
1647  * then the drive information will be updated and the disk will be
1648  * re-registered with the kernel.  If these conditions don't hold,
1649  * then it will be left alone for the next reboot.  The exception to this
1650  * is disk 0 which will always be left registered with the kernel since it
1651  * is also the controller node.  Any changes to disk 0 will show up on
1652  * the next reboot.
1653  */
1654 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time)
1655 {
1656         ctlr_info_t *h = hba[ctlr];
1657         struct gendisk *disk;
1658         InquiryData_struct *inq_buff = NULL;
1659         unsigned int block_size;
1660         sector_t total_size;
1661         unsigned long flags = 0;
1662         int ret = 0;
1663         drive_info_struct *drvinfo;
1664         int was_only_controller_node;
1665
1666         /* Get information about the disk and modify the driver structure */
1667         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1668         drvinfo = kmalloc(sizeof(*drvinfo), GFP_KERNEL);
1669         if (inq_buff == NULL || drvinfo == NULL)
1670                 goto mem_msg;
1671
1672         /* See if we're trying to update the "controller node"
1673          * this will happen the when the first logical drive gets
1674          * created by ACU.
1675          */
1676         was_only_controller_node = (drv_index == 0 &&
1677                                 h->drv[0].raid_level == -1);
1678
1679         /* testing to see if 16-byte CDBs are already being used */
1680         if (h->cciss_read == CCISS_READ_16) {
1681                 cciss_read_capacity_16(h->ctlr, drv_index, 1,
1682                         &total_size, &block_size);
1683
1684         } else {
1685                 cciss_read_capacity(ctlr, drv_index, 1,
1686                                     &total_size, &block_size);
1687
1688                 /* if read_capacity returns all F's this volume is >2TB */
1689                 /* in size so we switch to 16-byte CDB's for all */
1690                 /* read/write ops */
1691                 if (total_size == 0xFFFFFFFFULL) {
1692                         cciss_read_capacity_16(ctlr, drv_index, 1,
1693                         &total_size, &block_size);
1694                         h->cciss_read = CCISS_READ_16;
1695                         h->cciss_write = CCISS_WRITE_16;
1696                 } else {
1697                         h->cciss_read = CCISS_READ_10;
1698                         h->cciss_write = CCISS_WRITE_10;
1699                 }
1700         }
1701
1702         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1703                                inq_buff, drvinfo);
1704         drvinfo->block_size = block_size;
1705         drvinfo->nr_blocks = total_size + 1;
1706
1707         cciss_get_device_descr(ctlr, drv_index, 1, drvinfo->vendor,
1708                                 drvinfo->model, drvinfo->rev);
1709         cciss_get_serial_no(ctlr, drv_index, 1, drvinfo->serial_no,
1710                         sizeof(drvinfo->serial_no));
1711
1712         /* Is it the same disk we already know, and nothing's changed? */
1713         if (h->drv[drv_index].raid_level != -1 &&
1714                 ((memcmp(drvinfo->serial_no,
1715                                 h->drv[drv_index].serial_no, 16) == 0) &&
1716                 drvinfo->block_size == h->drv[drv_index].block_size &&
1717                 drvinfo->nr_blocks == h->drv[drv_index].nr_blocks &&
1718                 drvinfo->heads == h->drv[drv_index].heads &&
1719                 drvinfo->sectors == h->drv[drv_index].sectors &&
1720                 drvinfo->cylinders == h->drv[drv_index].cylinders))
1721                         /* The disk is unchanged, nothing to update */
1722                         goto freeret;
1723
1724         /* If we get here it's not the same disk, or something's changed,
1725          * so we need to * deregister it, and re-register it, if it's not
1726          * in use.
1727          * If the disk already exists then deregister it before proceeding
1728          * (unless it's the first disk (for the controller node).
1729          */
1730         if (h->drv[drv_index].raid_level != -1 && drv_index != 0) {
1731                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1732                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1733                 h->drv[drv_index].busy_configuring = 1;
1734                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1735
1736                 /* deregister_disk sets h->drv[drv_index].queue = NULL
1737                  * which keeps the interrupt handler from starting
1738                  * the queue.
1739                  */
1740                 ret = deregister_disk(h, drv_index, 0);
1741                 h->drv[drv_index].busy_configuring = 0;
1742         }
1743
1744         /* If the disk is in use return */
1745         if (ret)
1746                 goto freeret;
1747
1748         /* Save the new information from cciss_geometry_inquiry
1749          * and serial number inquiry.
1750          */
1751         h->drv[drv_index].block_size = drvinfo->block_size;
1752         h->drv[drv_index].nr_blocks = drvinfo->nr_blocks;
1753         h->drv[drv_index].heads = drvinfo->heads;
1754         h->drv[drv_index].sectors = drvinfo->sectors;
1755         h->drv[drv_index].cylinders = drvinfo->cylinders;
1756         h->drv[drv_index].raid_level = drvinfo->raid_level;
1757         memcpy(h->drv[drv_index].serial_no, drvinfo->serial_no, 16);
1758         memcpy(h->drv[drv_index].vendor, drvinfo->vendor, VENDOR_LEN + 1);
1759         memcpy(h->drv[drv_index].model, drvinfo->model, MODEL_LEN + 1);
1760         memcpy(h->drv[drv_index].rev, drvinfo->rev, REV_LEN + 1);
1761
1762         ++h->num_luns;
1763         disk = h->gendisk[drv_index];
1764         set_capacity(disk, h->drv[drv_index].nr_blocks);
1765
1766         /* If it's not disk 0 (drv_index != 0)
1767          * or if it was disk 0, but there was previously
1768          * no actual corresponding configured logical drive
1769          * (raid_leve == -1) then we want to update the
1770          * logical drive's information.
1771          */
1772         if (drv_index || first_time)
1773                 cciss_add_disk(h, disk, drv_index);
1774
1775 freeret:
1776         kfree(inq_buff);
1777         kfree(drvinfo);
1778         return;
1779 mem_msg:
1780         printk(KERN_ERR "cciss: out of memory\n");
1781         goto freeret;
1782 }
1783
1784 /* This function will find the first index of the controllers drive array
1785  * that has a -1 for the raid_level and will return that index.  This is
1786  * where new drives will be added.  If the index to be returned is greater
1787  * than the highest_lun index for the controller then highest_lun is set
1788  * to this new index.  If there are no available indexes then -1 is returned.
1789  * "controller_node" is used to know if this is a real logical drive, or just
1790  * the controller node, which determines if this counts towards highest_lun.
1791  */
1792 static int cciss_find_free_drive_index(int ctlr, int controller_node)
1793 {
1794         int i;
1795
1796         for (i = 0; i < CISS_MAX_LUN; i++) {
1797                 if (hba[ctlr]->drv[i].raid_level == -1) {
1798                         if (i > hba[ctlr]->highest_lun)
1799                                 if (!controller_node)
1800                                         hba[ctlr]->highest_lun = i;
1801                         return i;
1802                 }
1803         }
1804         return -1;
1805 }
1806
1807 /* cciss_add_gendisk finds a free hba[]->drv structure
1808  * and allocates a gendisk if needed, and sets the lunid
1809  * in the drvinfo structure.   It returns the index into
1810  * the ->drv[] array, or -1 if none are free.
1811  * is_controller_node indicates whether highest_lun should
1812  * count this disk, or if it's only being added to provide
1813  * a means to talk to the controller in case no logical
1814  * drives have yet been configured.
1815  */
1816 static int cciss_add_gendisk(ctlr_info_t *h, __u32 lunid, int controller_node)
1817 {
1818         int drv_index;
1819
1820         drv_index = cciss_find_free_drive_index(h->ctlr, controller_node);
1821         if (drv_index == -1)
1822                 return -1;
1823         /*Check if the gendisk needs to be allocated */
1824         if (!h->gendisk[drv_index]) {
1825                 h->gendisk[drv_index] =
1826                         alloc_disk(1 << NWD_SHIFT);
1827                 if (!h->gendisk[drv_index]) {
1828                         printk(KERN_ERR "cciss%d: could not "
1829                                 "allocate a new disk %d\n",
1830                                 h->ctlr, drv_index);
1831                         return -1;
1832                 }
1833         }
1834         h->drv[drv_index].LunID = lunid;
1835         if (cciss_create_ld_sysfs_entry(h, &h->drv[drv_index], drv_index))
1836                 goto err_free_disk;
1837
1838         /* Don't need to mark this busy because nobody */
1839         /* else knows about this disk yet to contend */
1840         /* for access to it. */
1841         h->drv[drv_index].busy_configuring = 0;
1842         wmb();
1843         return drv_index;
1844
1845 err_free_disk:
1846         put_disk(h->gendisk[drv_index]);
1847         h->gendisk[drv_index] = NULL;
1848         return -1;
1849 }
1850
1851 /* This is for the special case of a controller which
1852  * has no logical drives.  In this case, we still need
1853  * to register a disk so the controller can be accessed
1854  * by the Array Config Utility.
1855  */
1856 static void cciss_add_controller_node(ctlr_info_t *h)
1857 {
1858         struct gendisk *disk;
1859         int drv_index;
1860
1861         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
1862                 return;
1863
1864         drv_index = cciss_add_gendisk(h, 0, 1);
1865         if (drv_index == -1) {
1866                 printk(KERN_WARNING "cciss%d: could not "
1867                         "add disk 0.\n", h->ctlr);
1868                 return;
1869         }
1870         h->drv[drv_index].block_size = 512;
1871         h->drv[drv_index].nr_blocks = 0;
1872         h->drv[drv_index].heads = 0;
1873         h->drv[drv_index].sectors = 0;
1874         h->drv[drv_index].cylinders = 0;
1875         h->drv[drv_index].raid_level = -1;
1876         memset(h->drv[drv_index].serial_no, 0, 16);
1877         disk = h->gendisk[drv_index];
1878         cciss_add_disk(h, disk, drv_index);
1879 }
1880
1881 /* This function will add and remove logical drives from the Logical
1882  * drive array of the controller and maintain persistency of ordering
1883  * so that mount points are preserved until the next reboot.  This allows
1884  * for the removal of logical drives in the middle of the drive array
1885  * without a re-ordering of those drives.
1886  * INPUT
1887  * h            = The controller to perform the operations on
1888  */
1889 static int rebuild_lun_table(ctlr_info_t *h, int first_time)
1890 {
1891         int ctlr = h->ctlr;
1892         int num_luns;
1893         ReportLunData_struct *ld_buff = NULL;
1894         int return_code;
1895         int listlength = 0;
1896         int i;
1897         int drv_found;
1898         int drv_index = 0;
1899         __u32 lunid = 0;
1900         unsigned long flags;
1901
1902         if (!capable(CAP_SYS_RAWIO))
1903                 return -EPERM;
1904
1905         /* Set busy_configuring flag for this operation */
1906         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1907         if (h->busy_configuring) {
1908                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1909                 return -EBUSY;
1910         }
1911         h->busy_configuring = 1;
1912         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1913
1914         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1915         if (ld_buff == NULL)
1916                 goto mem_msg;
1917
1918         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
1919                                       sizeof(ReportLunData_struct),
1920                                       0, CTLR_LUNID, TYPE_CMD);
1921
1922         if (return_code == IO_OK)
1923                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
1924         else {  /* reading number of logical volumes failed */
1925                 printk(KERN_WARNING "cciss: report logical volume"
1926                        " command failed\n");
1927                 listlength = 0;
1928                 goto freeret;
1929         }
1930
1931         num_luns = listlength / 8;      /* 8 bytes per entry */
1932         if (num_luns > CISS_MAX_LUN) {
1933                 num_luns = CISS_MAX_LUN;
1934                 printk(KERN_WARNING "cciss: more luns configured"
1935                        " on controller than can be handled by"
1936                        " this driver.\n");
1937         }
1938
1939         if (num_luns == 0)
1940                 cciss_add_controller_node(h);
1941
1942         /* Compare controller drive array to driver's drive array
1943          * to see if any drives are missing on the controller due
1944          * to action of Array Config Utility (user deletes drive)
1945          * and deregister logical drives which have disappeared.
1946          */
1947         for (i = 0; i <= h->highest_lun; i++) {
1948                 int j;
1949                 drv_found = 0;
1950
1951                 /* skip holes in the array from already deleted drives */
1952                 if (h->drv[i].raid_level == -1)
1953                         continue;
1954
1955                 for (j = 0; j < num_luns; j++) {
1956                         memcpy(&lunid, &ld_buff->LUN[j][0], 4);
1957                         lunid = le32_to_cpu(lunid);
1958                         if (h->drv[i].LunID == lunid) {
1959                                 drv_found = 1;
1960                                 break;
1961                         }
1962                 }
1963                 if (!drv_found) {
1964                         /* Deregister it from the OS, it's gone. */
1965                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1966                         h->drv[i].busy_configuring = 1;
1967                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1968                         return_code = deregister_disk(h, i, 1);
1969                         cciss_destroy_ld_sysfs_entry(&h->drv[i]);
1970                         h->drv[i].busy_configuring = 0;
1971                 }
1972         }
1973
1974         /* Compare controller drive array to driver's drive array.
1975          * Check for updates in the drive information and any new drives
1976          * on the controller due to ACU adding logical drives, or changing
1977          * a logical drive's size, etc.  Reregister any new/changed drives
1978          */
1979         for (i = 0; i < num_luns; i++) {
1980                 int j;
1981
1982                 drv_found = 0;
1983
1984                 memcpy(&lunid, &ld_buff->LUN[i][0], 4);
1985                 lunid = le32_to_cpu(lunid);
1986
1987                 /* Find if the LUN is already in the drive array
1988                  * of the driver.  If so then update its info
1989                  * if not in use.  If it does not exist then find
1990                  * the first free index and add it.
1991                  */
1992                 for (j = 0; j <= h->highest_lun; j++) {
1993                         if (h->drv[j].raid_level != -1 &&
1994                                 h->drv[j].LunID == lunid) {
1995                                 drv_index = j;
1996                                 drv_found = 1;
1997                                 break;
1998                         }
1999                 }
2000
2001                 /* check if the drive was found already in the array */
2002                 if (!drv_found) {
2003                         drv_index = cciss_add_gendisk(h, lunid, 0);
2004                         if (drv_index == -1)
2005                                 goto freeret;
2006                 }
2007                 cciss_update_drive_info(ctlr, drv_index, first_time);
2008         }               /* end for */
2009
2010 freeret:
2011         kfree(ld_buff);
2012         h->busy_configuring = 0;
2013         /* We return -1 here to tell the ACU that we have registered/updated
2014          * all of the drives that we can and to keep it from calling us
2015          * additional times.
2016          */
2017         return -1;
2018 mem_msg:
2019         printk(KERN_ERR "cciss: out of memory\n");
2020         h->busy_configuring = 0;
2021         goto freeret;
2022 }
2023
2024 /* This function will deregister the disk and it's queue from the
2025  * kernel.  It must be called with the controller lock held and the
2026  * drv structures busy_configuring flag set.  It's parameters are:
2027  *
2028  * disk = This is the disk to be deregistered
2029  * drv  = This is the drive_info_struct associated with the disk to be
2030  *        deregistered.  It contains information about the disk used
2031  *        by the driver.
2032  * clear_all = This flag determines whether or not the disk information
2033  *             is going to be completely cleared out and the highest_lun
2034  *             reset.  Sometimes we want to clear out information about
2035  *             the disk in preparation for re-adding it.  In this case
2036  *             the highest_lun should be left unchanged and the LunID
2037  *             should not be cleared.
2038 */
2039 static int deregister_disk(ctlr_info_t *h, int drv_index,
2040                            int clear_all)
2041 {
2042         int i;
2043         struct gendisk *disk;
2044         drive_info_struct *drv;
2045
2046         if (!capable(CAP_SYS_RAWIO))
2047                 return -EPERM;
2048
2049         drv = &h->drv[drv_index];
2050         disk = h->gendisk[drv_index];
2051
2052         /* make sure logical volume is NOT is use */
2053         if (clear_all || (h->gendisk[0] == disk)) {
2054                 if (drv->usage_count > 1)
2055                         return -EBUSY;
2056         } else if (drv->usage_count > 0)
2057                 return -EBUSY;
2058
2059         /* invalidate the devices and deregister the disk.  If it is disk
2060          * zero do not deregister it but just zero out it's values.  This
2061          * allows us to delete disk zero but keep the controller registered.
2062          */
2063         if (h->gendisk[0] != disk) {
2064                 struct request_queue *q = disk->queue;
2065                 if (disk->flags & GENHD_FL_UP)
2066                         del_gendisk(disk);
2067                 if (q) {
2068                         blk_cleanup_queue(q);
2069                         /* Set drv->queue to NULL so that we do not try
2070                          * to call blk_start_queue on this queue in the
2071                          * interrupt handler
2072                          */
2073                         drv->queue = NULL;
2074                 }
2075                 /* If clear_all is set then we are deleting the logical
2076                  * drive, not just refreshing its info.  For drives
2077                  * other than disk 0 we will call put_disk.  We do not
2078                  * do this for disk 0 as we need it to be able to
2079                  * configure the controller.
2080                  */
2081                 if (clear_all){
2082                         /* This isn't pretty, but we need to find the
2083                          * disk in our array and NULL our the pointer.
2084                          * This is so that we will call alloc_disk if
2085                          * this index is used again later.
2086                          */
2087                         for (i=0; i < CISS_MAX_LUN; i++){
2088                                 if (h->gendisk[i] == disk) {
2089                                         h->gendisk[i] = NULL;
2090                                         break;
2091                                 }
2092                         }
2093                         put_disk(disk);
2094                 }
2095         } else {
2096                 set_capacity(disk, 0);
2097         }
2098
2099         --h->num_luns;
2100         /* zero out the disk size info */
2101         drv->nr_blocks = 0;
2102         drv->block_size = 0;
2103         drv->heads = 0;
2104         drv->sectors = 0;
2105         drv->cylinders = 0;
2106         drv->raid_level = -1;   /* This can be used as a flag variable to
2107                                  * indicate that this element of the drive
2108                                  * array is free.
2109                                  */
2110
2111         if (clear_all) {
2112                 /* check to see if it was the last disk */
2113                 if (drv == h->drv + h->highest_lun) {
2114                         /* if so, find the new hightest lun */
2115                         int i, newhighest = -1;
2116                         for (i = 0; i <= h->highest_lun; i++) {
2117                                 /* if the disk has size > 0, it is available */
2118                                 if (h->drv[i].heads)
2119                                         newhighest = i;
2120                         }
2121                         h->highest_lun = newhighest;
2122                 }
2123
2124                 drv->LunID = 0;
2125         }
2126         return 0;
2127 }
2128
2129 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2130                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2131                 int cmd_type)
2132 {
2133         ctlr_info_t *h = hba[ctlr];
2134         u64bit buff_dma_handle;
2135         int status = IO_OK;
2136
2137         c->cmd_type = CMD_IOCTL_PEND;
2138         c->Header.ReplyQueue = 0;
2139         if (buff != NULL) {
2140                 c->Header.SGList = 1;
2141                 c->Header.SGTotal = 1;
2142         } else {
2143                 c->Header.SGList = 0;
2144                 c->Header.SGTotal = 0;
2145         }
2146         c->Header.Tag.lower = c->busaddr;
2147         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2148
2149         c->Request.Type.Type = cmd_type;
2150         if (cmd_type == TYPE_CMD) {
2151                 switch (cmd) {
2152                 case CISS_INQUIRY:
2153                         /* are we trying to read a vital product page */
2154                         if (page_code != 0) {
2155                                 c->Request.CDB[1] = 0x01;
2156                                 c->Request.CDB[2] = page_code;
2157                         }
2158                         c->Request.CDBLen = 6;
2159                         c->Request.Type.Attribute = ATTR_SIMPLE;
2160                         c->Request.Type.Direction = XFER_READ;
2161                         c->Request.Timeout = 0;
2162                         c->Request.CDB[0] = CISS_INQUIRY;
2163                         c->Request.CDB[4] = size & 0xFF;
2164                         break;
2165                 case CISS_REPORT_LOG:
2166                 case CISS_REPORT_PHYS:
2167                         /* Talking to controller so It's a physical command
2168                            mode = 00 target = 0.  Nothing to write.
2169                          */
2170                         c->Request.CDBLen = 12;
2171                         c->Request.Type.Attribute = ATTR_SIMPLE;
2172                         c->Request.Type.Direction = XFER_READ;
2173                         c->Request.Timeout = 0;
2174                         c->Request.CDB[0] = cmd;
2175                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
2176                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2177                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2178                         c->Request.CDB[9] = size & 0xFF;
2179                         break;
2180
2181                 case CCISS_READ_CAPACITY:
2182                         c->Request.CDBLen = 10;
2183                         c->Request.Type.Attribute = ATTR_SIMPLE;
2184                         c->Request.Type.Direction = XFER_READ;
2185                         c->Request.Timeout = 0;
2186                         c->Request.CDB[0] = cmd;
2187                         break;
2188                 case CCISS_READ_CAPACITY_16:
2189                         c->Request.CDBLen = 16;
2190                         c->Request.Type.Attribute = ATTR_SIMPLE;
2191                         c->Request.Type.Direction = XFER_READ;
2192                         c->Request.Timeout = 0;
2193                         c->Request.CDB[0] = cmd;
2194                         c->Request.CDB[1] = 0x10;
2195                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2196                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2197                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2198                         c->Request.CDB[13] = size & 0xFF;
2199                         c->Request.Timeout = 0;
2200                         c->Request.CDB[0] = cmd;
2201                         break;
2202                 case CCISS_CACHE_FLUSH:
2203                         c->Request.CDBLen = 12;
2204                         c->Request.Type.Attribute = ATTR_SIMPLE;
2205                         c->Request.Type.Direction = XFER_WRITE;
2206                         c->Request.Timeout = 0;
2207                         c->Request.CDB[0] = BMIC_WRITE;
2208                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2209                         break;
2210                 case TEST_UNIT_READY:
2211                         c->Request.CDBLen = 6;
2212                         c->Request.Type.Attribute = ATTR_SIMPLE;
2213                         c->Request.Type.Direction = XFER_NONE;
2214                         c->Request.Timeout = 0;
2215                         break;
2216                 default:
2217                         printk(KERN_WARNING
2218                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2219                         return IO_ERROR;
2220                 }
2221         } else if (cmd_type == TYPE_MSG) {
2222                 switch (cmd) {
2223                 case 0: /* ABORT message */
2224                         c->Request.CDBLen = 12;
2225                         c->Request.Type.Attribute = ATTR_SIMPLE;
2226                         c->Request.Type.Direction = XFER_WRITE;
2227                         c->Request.Timeout = 0;
2228                         c->Request.CDB[0] = cmd;        /* abort */
2229                         c->Request.CDB[1] = 0;  /* abort a command */
2230                         /* buff contains the tag of the command to abort */
2231                         memcpy(&c->Request.CDB[4], buff, 8);
2232                         break;
2233                 case 1: /* RESET message */
2234                         c->Request.CDBLen = 16;
2235                         c->Request.Type.Attribute = ATTR_SIMPLE;
2236                         c->Request.Type.Direction = XFER_NONE;
2237                         c->Request.Timeout = 0;
2238                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2239                         c->Request.CDB[0] = cmd;        /* reset */
2240                         c->Request.CDB[1] = 0x03;       /* reset a target */
2241                         break;
2242                 case 3: /* No-Op message */
2243                         c->Request.CDBLen = 1;
2244                         c->Request.Type.Attribute = ATTR_SIMPLE;
2245                         c->Request.Type.Direction = XFER_WRITE;
2246                         c->Request.Timeout = 0;
2247                         c->Request.CDB[0] = cmd;
2248                         break;
2249                 default:
2250                         printk(KERN_WARNING
2251                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2252                         return IO_ERROR;
2253                 }
2254         } else {
2255                 printk(KERN_WARNING
2256                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2257                 return IO_ERROR;
2258         }
2259         /* Fill in the scatter gather information */
2260         if (size > 0) {
2261                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2262                                                              buff, size,
2263                                                              PCI_DMA_BIDIRECTIONAL);
2264                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2265                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2266                 c->SG[0].Len = size;
2267                 c->SG[0].Ext = 0;       /* we are not chaining */
2268         }
2269         return status;
2270 }
2271
2272 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2273 {
2274         switch (c->err_info->ScsiStatus) {
2275         case SAM_STAT_GOOD:
2276                 return IO_OK;
2277         case SAM_STAT_CHECK_CONDITION:
2278                 switch (0xf & c->err_info->SenseInfo[2]) {
2279                 case 0: return IO_OK; /* no sense */
2280                 case 1: return IO_OK; /* recovered error */
2281                 default:
2282                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2283                                 "check condition, sense key = 0x%02x\n",
2284                                 h->ctlr, c->Request.CDB[0],
2285                                 c->err_info->SenseInfo[2]);
2286                 }
2287                 break;
2288         default:
2289                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2290                         "scsi status = 0x%02x\n", h->ctlr,
2291                         c->Request.CDB[0], c->err_info->ScsiStatus);
2292                 break;
2293         }
2294         return IO_ERROR;
2295 }
2296
2297 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2298 {
2299         int return_status = IO_OK;
2300
2301         if (c->err_info->CommandStatus == CMD_SUCCESS)
2302                 return IO_OK;
2303
2304         switch (c->err_info->CommandStatus) {
2305         case CMD_TARGET_STATUS:
2306                 return_status = check_target_status(h, c);
2307                 break;
2308         case CMD_DATA_UNDERRUN:
2309         case CMD_DATA_OVERRUN:
2310                 /* expected for inquiry and report lun commands */
2311                 break;
2312         case CMD_INVALID:
2313                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2314                        "reported invalid\n", c->Request.CDB[0]);
2315                 return_status = IO_ERROR;
2316                 break;
2317         case CMD_PROTOCOL_ERR:
2318                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2319                        "protocol error \n", c->Request.CDB[0]);
2320                 return_status = IO_ERROR;
2321                 break;
2322         case CMD_HARDWARE_ERR:
2323                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2324                        " hardware error\n", c->Request.CDB[0]);
2325                 return_status = IO_ERROR;
2326                 break;
2327         case CMD_CONNECTION_LOST:
2328                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2329                        "connection lost\n", c->Request.CDB[0]);
2330                 return_status = IO_ERROR;
2331                 break;
2332         case CMD_ABORTED:
2333                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2334                        "aborted\n", c->Request.CDB[0]);
2335                 return_status = IO_ERROR;
2336                 break;
2337         case CMD_ABORT_FAILED:
2338                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2339                        "abort failed\n", c->Request.CDB[0]);
2340                 return_status = IO_ERROR;
2341                 break;
2342         case CMD_UNSOLICITED_ABORT:
2343                 printk(KERN_WARNING
2344                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2345                         c->Request.CDB[0]);
2346                 return_status = IO_NEEDS_RETRY;
2347                 break;
2348         default:
2349                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2350                        "unknown status %x\n", c->Request.CDB[0],
2351                        c->err_info->CommandStatus);
2352                 return_status = IO_ERROR;
2353         }
2354         return return_status;
2355 }
2356
2357 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2358         int attempt_retry)
2359 {
2360         DECLARE_COMPLETION_ONSTACK(wait);
2361         u64bit buff_dma_handle;
2362         unsigned long flags;
2363         int return_status = IO_OK;
2364
2365 resend_cmd2:
2366         c->waiting = &wait;
2367         /* Put the request on the tail of the queue and send it */
2368         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2369         addQ(&h->reqQ, c);
2370         h->Qdepth++;
2371         start_io(h);
2372         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2373
2374         wait_for_completion(&wait);
2375
2376         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2377                 goto command_done;
2378
2379         return_status = process_sendcmd_error(h, c);
2380
2381         if (return_status == IO_NEEDS_RETRY &&
2382                 c->retry_count < MAX_CMD_RETRIES) {
2383                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2384                         c->Request.CDB[0]);
2385                 c->retry_count++;
2386                 /* erase the old error information */
2387                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2388                 return_status = IO_OK;
2389                 INIT_COMPLETION(wait);
2390                 goto resend_cmd2;
2391         }
2392
2393 command_done:
2394         /* unlock the buffers from DMA */
2395         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2396         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2397         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2398                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2399         return return_status;
2400 }
2401
2402 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2403                            __u8 page_code, unsigned char scsi3addr[],
2404                         int cmd_type)
2405 {
2406         ctlr_info_t *h = hba[ctlr];
2407         CommandList_struct *c;
2408         int return_status;
2409
2410         c = cmd_alloc(h, 0);
2411         if (!c)
2412                 return -ENOMEM;
2413         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2414                 scsi3addr, cmd_type);
2415         if (return_status == IO_OK)
2416                 return_status = sendcmd_withirq_core(h, c, 1);
2417
2418         cmd_free(h, c, 0);
2419         return return_status;
2420 }
2421
2422 static void cciss_geometry_inquiry(int ctlr, int logvol,
2423                                    int withirq, sector_t total_size,
2424                                    unsigned int block_size,
2425                                    InquiryData_struct *inq_buff,
2426                                    drive_info_struct *drv)
2427 {
2428         int return_code;
2429         unsigned long t;
2430         unsigned char scsi3addr[8];
2431
2432         memset(inq_buff, 0, sizeof(InquiryData_struct));
2433         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2434         if (withirq)
2435                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
2436                                               inq_buff, sizeof(*inq_buff),
2437                                               0xC1, scsi3addr, TYPE_CMD);
2438         else
2439                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
2440                                       sizeof(*inq_buff), 0xC1, scsi3addr,
2441                                       TYPE_CMD);
2442         if (return_code == IO_OK) {
2443                 if (inq_buff->data_byte[8] == 0xFF) {
2444                         printk(KERN_WARNING
2445                                "cciss: reading geometry failed, volume "
2446                                "does not support reading geometry\n");
2447                         drv->heads = 255;
2448                         drv->sectors = 32;      // Sectors per track
2449                         drv->cylinders = total_size + 1;
2450                         drv->raid_level = RAID_UNKNOWN;
2451                 } else {
2452                         drv->heads = inq_buff->data_byte[6];
2453                         drv->sectors = inq_buff->data_byte[7];
2454                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2455                         drv->cylinders += inq_buff->data_byte[5];
2456                         drv->raid_level = inq_buff->data_byte[8];
2457                 }
2458                 drv->block_size = block_size;
2459                 drv->nr_blocks = total_size + 1;
2460                 t = drv->heads * drv->sectors;
2461                 if (t > 1) {
2462                         sector_t real_size = total_size + 1;
2463                         unsigned long rem = sector_div(real_size, t);
2464                         if (rem)
2465                                 real_size++;
2466                         drv->cylinders = real_size;
2467                 }
2468         } else {                /* Get geometry failed */
2469                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2470         }
2471         printk(KERN_INFO "      heads=%d, sectors=%d, cylinders=%d\n\n",
2472                drv->heads, drv->sectors, drv->cylinders);
2473 }
2474
2475 static void
2476 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
2477                     unsigned int *block_size)
2478 {
2479         ReadCapdata_struct *buf;
2480         int return_code;
2481         unsigned char scsi3addr[8];
2482
2483         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2484         if (!buf) {
2485                 printk(KERN_WARNING "cciss: out of memory\n");
2486                 return;
2487         }
2488
2489         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2490         if (withirq)
2491                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
2492                                 ctlr, buf, sizeof(ReadCapdata_struct),
2493                                         0, scsi3addr, TYPE_CMD);
2494         else
2495                 return_code = sendcmd(CCISS_READ_CAPACITY,
2496                                 ctlr, buf, sizeof(ReadCapdata_struct),
2497                                         0, scsi3addr, TYPE_CMD);
2498         if (return_code == IO_OK) {
2499                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2500                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2501         } else {                /* read capacity command failed */
2502                 printk(KERN_WARNING "cciss: read capacity failed\n");
2503                 *total_size = 0;
2504                 *block_size = BLOCK_SIZE;
2505         }
2506         if (*total_size != 0)
2507                 printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2508                 (unsigned long long)*total_size+1, *block_size);
2509         kfree(buf);
2510 }
2511
2512 static void
2513 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size,                                 unsigned int *block_size)
2514 {
2515         ReadCapdata_struct_16 *buf;
2516         int return_code;
2517         unsigned char scsi3addr[8];
2518
2519         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2520         if (!buf) {
2521                 printk(KERN_WARNING "cciss: out of memory\n");
2522                 return;
2523         }
2524
2525         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2526         if (withirq) {
2527                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2528                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2529                                 0, scsi3addr, TYPE_CMD);
2530         }
2531         else {
2532                 return_code = sendcmd(CCISS_READ_CAPACITY_16,
2533                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2534                                 0, scsi3addr, TYPE_CMD);
2535         }
2536         if (return_code == IO_OK) {
2537                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2538                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2539         } else {                /* read capacity command failed */
2540                 printk(KERN_WARNING "cciss: read capacity failed\n");
2541                 *total_size = 0;
2542                 *block_size = BLOCK_SIZE;
2543         }
2544         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2545                (unsigned long long)*total_size+1, *block_size);
2546         kfree(buf);
2547 }
2548
2549 static int cciss_revalidate(struct gendisk *disk)
2550 {
2551         ctlr_info_t *h = get_host(disk);
2552         drive_info_struct *drv = get_drv(disk);
2553         int logvol;
2554         int FOUND = 0;
2555         unsigned int block_size;
2556         sector_t total_size;
2557         InquiryData_struct *inq_buff = NULL;
2558
2559         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2560                 if (h->drv[logvol].LunID == drv->LunID) {
2561                         FOUND = 1;
2562                         break;
2563                 }
2564         }
2565
2566         if (!FOUND)
2567                 return 1;
2568
2569         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2570         if (inq_buff == NULL) {
2571                 printk(KERN_WARNING "cciss: out of memory\n");
2572                 return 1;
2573         }
2574         if (h->cciss_read == CCISS_READ_10) {
2575                 cciss_read_capacity(h->ctlr, logvol, 1,
2576                                         &total_size, &block_size);
2577         } else {
2578                 cciss_read_capacity_16(h->ctlr, logvol, 1,
2579                                         &total_size, &block_size);
2580         }
2581         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2582                                inq_buff, drv);
2583
2584         blk_queue_logical_block_size(drv->queue, drv->block_size);
2585         set_capacity(disk, drv->nr_blocks);
2586
2587         kfree(inq_buff);
2588         return 0;
2589 }
2590
2591 /*
2592  *   Wait polling for a command to complete.
2593  *   The memory mapped FIFO is polled for the completion.
2594  *   Used only at init time, interrupts from the HBA are disabled.
2595  */
2596 static unsigned long pollcomplete(int ctlr)
2597 {
2598         unsigned long done;
2599         int i;
2600
2601         /* Wait (up to 20 seconds) for a command to complete */
2602
2603         for (i = 20 * HZ; i > 0; i--) {
2604                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2605                 if (done == FIFO_EMPTY)
2606                         schedule_timeout_uninterruptible(1);
2607                 else
2608                         return done;
2609         }
2610         /* Invalid address to tell caller we ran out of time */
2611         return 1;
2612 }
2613
2614 /* Send command c to controller h and poll for it to complete.
2615  * Turns interrupts off on the board.  Used at driver init time
2616  * and during SCSI error recovery.
2617  */
2618 static int sendcmd_core(ctlr_info_t *h, CommandList_struct *c)
2619 {
2620         int i;
2621         unsigned long complete;
2622         int status = IO_ERROR;
2623         u64bit buff_dma_handle;
2624
2625 resend_cmd1:
2626
2627         /* Disable interrupt on the board. */
2628         h->access.set_intr_mask(h, CCISS_INTR_OFF);
2629
2630         /* Make sure there is room in the command FIFO */
2631         /* Actually it should be completely empty at this time */
2632         /* unless we are in here doing error handling for the scsi */
2633         /* tape side of the driver. */
2634         for (i = 200000; i > 0; i--) {
2635                 /* if fifo isn't full go */
2636                 if (!(h->access.fifo_full(h)))
2637                         break;
2638                 udelay(10);
2639                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2640                        " waiting!\n", h->ctlr);
2641         }
2642         h->access.submit_command(h, c); /* Send the cmd */
2643         do {
2644                 complete = pollcomplete(h->ctlr);
2645
2646 #ifdef CCISS_DEBUG
2647                 printk(KERN_DEBUG "cciss: command completed\n");
2648 #endif                          /* CCISS_DEBUG */
2649
2650                 if (complete == 1) {
2651                         printk(KERN_WARNING
2652                                "cciss cciss%d: SendCmd Timeout out, "
2653                                "No command list address returned!\n", h->ctlr);
2654                         status = IO_ERROR;
2655                         break;
2656                 }
2657
2658                 /* Make sure it's the command we're expecting. */
2659                 if ((complete & ~CISS_ERROR_BIT) != c->busaddr) {
2660                         printk(KERN_WARNING "cciss%d: Unexpected command "
2661                                 "completion.\n", h->ctlr);
2662                         continue;
2663                 }
2664
2665                 /* It is our command.  If no error, we're done. */
2666                 if (!(complete & CISS_ERROR_BIT)) {
2667                         status = IO_OK;
2668                         break;
2669                 }
2670
2671                 /* There is an error... */
2672
2673                 /* if data overrun or underun on Report command ignore it */
2674                 if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2675                      (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2676                      (c->Request.CDB[0] == CISS_INQUIRY)) &&
2677                         ((c->err_info->CommandStatus == CMD_DATA_OVERRUN) ||
2678                          (c->err_info->CommandStatus == CMD_DATA_UNDERRUN))) {
2679                         complete = c->busaddr;
2680                         status = IO_OK;
2681                         break;
2682                 }
2683
2684                 if (c->err_info->CommandStatus == CMD_UNSOLICITED_ABORT) {
2685                         printk(KERN_WARNING "cciss%d: unsolicited abort %p\n",
2686                                 h->ctlr, c);
2687                         if (c->retry_count < MAX_CMD_RETRIES) {
2688                                 printk(KERN_WARNING "cciss%d: retrying %p\n",
2689                                    h->ctlr, c);
2690                                 c->retry_count++;
2691                                 /* erase the old error information */
2692                                 memset(c->err_info, 0, sizeof(c->err_info));
2693                                 goto resend_cmd1;
2694                         }
2695                         printk(KERN_WARNING "cciss%d: retried %p too many "
2696                                 "times\n", h->ctlr, c);
2697                         status = IO_ERROR;
2698                         break;
2699                 }
2700
2701                 if (c->err_info->CommandStatus == CMD_UNABORTABLE) {
2702                         printk(KERN_WARNING "cciss%d: command could not be "
2703                                 "aborted.\n", h->ctlr);
2704                         status = IO_ERROR;
2705                         break;
2706                 }
2707
2708                 if (c->err_info->CommandStatus == CMD_TARGET_STATUS) {
2709                         status = check_target_status(h, c);
2710                         break;
2711                 }
2712
2713                 printk(KERN_WARNING "cciss%d: sendcmd error\n", h->ctlr);
2714                 printk(KERN_WARNING "cmd = 0x%02x, CommandStatus = 0x%02x\n",
2715                         c->Request.CDB[0], c->err_info->CommandStatus);
2716                 status = IO_ERROR;
2717                 break;
2718
2719         } while (1);
2720
2721         /* unlock the data buffer from DMA */
2722         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2723         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2724         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2725                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2726         return status;
2727 }
2728
2729 /*
2730  * Send a command to the controller, and wait for it to complete.
2731  * Used at init time, and during SCSI error recovery.
2732  */
2733 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
2734         __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2735 {
2736         CommandList_struct *c;
2737         int status;
2738
2739         c = cmd_alloc(hba[ctlr], 1);
2740         if (!c) {
2741                 printk(KERN_WARNING "cciss: unable to get memory");
2742                 return IO_ERROR;
2743         }
2744         status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2745                 scsi3addr, cmd_type);
2746         if (status == IO_OK)
2747                 status = sendcmd_core(hba[ctlr], c);
2748         cmd_free(hba[ctlr], c, 1);
2749         return status;
2750 }
2751
2752 /*
2753  * Map (physical) PCI mem into (virtual) kernel space
2754  */
2755 static void __iomem *remap_pci_mem(ulong base, ulong size)
2756 {
2757         ulong page_base = ((ulong) base) & PAGE_MASK;
2758         ulong page_offs = ((ulong) base) - page_base;
2759         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2760
2761         return page_remapped ? (page_remapped + page_offs) : NULL;
2762 }
2763
2764 /*
2765  * Takes jobs of the Q and sends them to the hardware, then puts it on
2766  * the Q to wait for completion.
2767  */
2768 static void start_io(ctlr_info_t *h)
2769 {
2770         CommandList_struct *c;
2771
2772         while (!hlist_empty(&h->reqQ)) {
2773                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2774                 /* can't do anything if fifo is full */
2775                 if ((h->access.fifo_full(h))) {
2776                         printk(KERN_WARNING "cciss: fifo full\n");
2777                         break;
2778                 }
2779
2780                 /* Get the first entry from the Request Q */
2781                 removeQ(c);
2782                 h->Qdepth--;
2783
2784                 /* Tell the controller execute command */
2785                 h->access.submit_command(h, c);
2786
2787                 /* Put job onto the completed Q */
2788                 addQ(&h->cmpQ, c);
2789         }
2790 }
2791
2792 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2793 /* Zeros out the error record and then resends the command back */
2794 /* to the controller */
2795 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2796 {
2797         /* erase the old error information */
2798         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2799
2800         /* add it to software queue and then send it to the controller */
2801         addQ(&h->reqQ, c);
2802         h->Qdepth++;
2803         if (h->Qdepth > h->maxQsinceinit)
2804                 h->maxQsinceinit = h->Qdepth;
2805
2806         start_io(h);
2807 }
2808
2809 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2810         unsigned int msg_byte, unsigned int host_byte,
2811         unsigned int driver_byte)
2812 {
2813         /* inverse of macros in scsi.h */
2814         return (scsi_status_byte & 0xff) |
2815                 ((msg_byte & 0xff) << 8) |
2816                 ((host_byte & 0xff) << 16) |
2817                 ((driver_byte & 0xff) << 24);
2818 }
2819
2820 static inline int evaluate_target_status(ctlr_info_t *h,
2821                         CommandList_struct *cmd, int *retry_cmd)
2822 {
2823         unsigned char sense_key;
2824         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2825         int error_value;
2826
2827         *retry_cmd = 0;
2828         /* If we get in here, it means we got "target status", that is, scsi status */
2829         status_byte = cmd->err_info->ScsiStatus;
2830         driver_byte = DRIVER_OK;
2831         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2832
2833         if (blk_pc_request(cmd->rq))
2834                 host_byte = DID_PASSTHROUGH;
2835         else
2836                 host_byte = DID_OK;
2837
2838         error_value = make_status_bytes(status_byte, msg_byte,
2839                 host_byte, driver_byte);
2840
2841         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2842                 if (!blk_pc_request(cmd->rq))
2843                         printk(KERN_WARNING "cciss: cmd %p "
2844                                "has SCSI Status 0x%x\n",
2845                                cmd, cmd->err_info->ScsiStatus);
2846                 return error_value;
2847         }
2848
2849         /* check the sense key */
2850         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2851         /* no status or recovered error */
2852         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2853                 error_value = 0;
2854
2855         if (check_for_unit_attention(h, cmd)) {
2856                 *retry_cmd = !blk_pc_request(cmd->rq);
2857                 return 0;
2858         }
2859
2860         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2861                 if (error_value != 0)
2862                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2863                                " sense key = 0x%x\n", cmd, sense_key);
2864                 return error_value;
2865         }
2866
2867         /* SG_IO or similar, copy sense data back */
2868         if (cmd->rq->sense) {
2869                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2870                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2871                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2872                         cmd->rq->sense_len);
2873         } else
2874                 cmd->rq->sense_len = 0;
2875
2876         return error_value;
2877 }
2878
2879 /* checks the status of the job and calls complete buffers to mark all
2880  * buffers for the completed job. Note that this function does not need
2881  * to hold the hba/queue lock.
2882  */
2883 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2884                                     int timeout)
2885 {
2886         int retry_cmd = 0;
2887         struct request *rq = cmd->rq;
2888
2889         rq->errors = 0;
2890
2891         if (timeout)
2892                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
2893
2894         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
2895                 goto after_error_processing;
2896
2897         switch (cmd->err_info->CommandStatus) {
2898         case CMD_TARGET_STATUS:
2899                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
2900                 break;
2901         case CMD_DATA_UNDERRUN:
2902                 if (blk_fs_request(cmd->rq)) {
2903                         printk(KERN_WARNING "cciss: cmd %p has"
2904                                " completed with data underrun "
2905                                "reported\n", cmd);
2906                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
2907                 }
2908                 break;
2909         case CMD_DATA_OVERRUN:
2910                 if (blk_fs_request(cmd->rq))
2911                         printk(KERN_WARNING "cciss: cmd %p has"
2912                                " completed with data overrun "
2913                                "reported\n", cmd);
2914                 break;
2915         case CMD_INVALID:
2916                 printk(KERN_WARNING "cciss: cmd %p is "
2917                        "reported invalid\n", cmd);
2918                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2919                         cmd->err_info->CommandStatus, DRIVER_OK,
2920                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2921                 break;
2922         case CMD_PROTOCOL_ERR:
2923                 printk(KERN_WARNING "cciss: cmd %p has "
2924                        "protocol error \n", cmd);
2925                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2926                         cmd->err_info->CommandStatus, DRIVER_OK,
2927                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2928                 break;
2929         case CMD_HARDWARE_ERR:
2930                 printk(KERN_WARNING "cciss: cmd %p had "
2931                        " hardware error\n", cmd);
2932                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2933                         cmd->err_info->CommandStatus, DRIVER_OK,
2934                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2935                 break;
2936         case CMD_CONNECTION_LOST:
2937                 printk(KERN_WARNING "cciss: cmd %p had "
2938                        "connection lost\n", cmd);
2939                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2940                         cmd->err_info->CommandStatus, DRIVER_OK,
2941                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2942                 break;
2943         case CMD_ABORTED:
2944                 printk(KERN_WARNING "cciss: cmd %p was "
2945                        "aborted\n", cmd);
2946                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2947                         cmd->err_info->CommandStatus, DRIVER_OK,
2948                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
2949                 break;
2950         case CMD_ABORT_FAILED:
2951                 printk(KERN_WARNING "cciss: cmd %p reports "
2952                        "abort failed\n", cmd);
2953                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2954                         cmd->err_info->CommandStatus, DRIVER_OK,
2955                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2956                 break;
2957         case CMD_UNSOLICITED_ABORT:
2958                 printk(KERN_WARNING "cciss%d: unsolicited "
2959                        "abort %p\n", h->ctlr, cmd);
2960                 if (cmd->retry_count < MAX_CMD_RETRIES) {
2961                         retry_cmd = 1;
2962                         printk(KERN_WARNING
2963                                "cciss%d: retrying %p\n", h->ctlr, cmd);
2964                         cmd->retry_count++;
2965                 } else
2966                         printk(KERN_WARNING
2967                                "cciss%d: %p retried too "
2968                                "many times\n", h->ctlr, cmd);
2969                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2970                         cmd->err_info->CommandStatus, DRIVER_OK,
2971                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
2972                 break;
2973         case CMD_TIMEOUT:
2974                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
2975                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2976                         cmd->err_info->CommandStatus, DRIVER_OK,
2977                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2978                 break;
2979         default:
2980                 printk(KERN_WARNING "cciss: cmd %p returned "
2981                        "unknown status %x\n", cmd,
2982                        cmd->err_info->CommandStatus);
2983                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2984                         cmd->err_info->CommandStatus, DRIVER_OK,
2985                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2986         }
2987
2988 after_error_processing:
2989
2990         /* We need to return this command */
2991         if (retry_cmd) {
2992                 resend_cciss_cmd(h, cmd);
2993                 return;
2994         }
2995         cmd->rq->completion_data = cmd;
2996         blk_complete_request(cmd->rq);
2997 }
2998
2999 /*
3000  * Get a request and submit it to the controller.
3001  */
3002 static void do_cciss_request(struct request_queue *q)
3003 {
3004         ctlr_info_t *h = q->queuedata;
3005         CommandList_struct *c;
3006         sector_t start_blk;
3007         int seg;
3008         struct request *creq;
3009         u64bit temp64;
3010         struct scatterlist tmp_sg[MAXSGENTRIES];
3011         drive_info_struct *drv;
3012         int i, dir;
3013
3014         /* We call start_io here in case there is a command waiting on the
3015          * queue that has not been sent.
3016          */
3017         if (blk_queue_plugged(q))
3018                 goto startio;
3019
3020       queue:
3021         creq = blk_peek_request(q);
3022         if (!creq)
3023                 goto startio;
3024
3025         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
3026
3027         if ((c = cmd_alloc(h, 1)) == NULL)
3028                 goto full;
3029
3030         blk_start_request(creq);
3031
3032         spin_unlock_irq(q->queue_lock);
3033
3034         c->cmd_type = CMD_RWREQ;
3035         c->rq = creq;
3036
3037         /* fill in the request */
3038         drv = creq->rq_disk->private_data;
3039         c->Header.ReplyQueue = 0;       // unused in simple mode
3040         /* got command from pool, so use the command block index instead */
3041         /* for direct lookups. */
3042         /* The first 2 bits are reserved for controller error reporting. */
3043         c->Header.Tag.lower = (c->cmdindex << 3);
3044         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
3045         c->Header.LUN.LogDev.VolId = drv->LunID;
3046         c->Header.LUN.LogDev.Mode = 1;
3047         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
3048         c->Request.Type.Type = TYPE_CMD;        // It is a command.
3049         c->Request.Type.Attribute = ATTR_SIMPLE;
3050         c->Request.Type.Direction =
3051             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3052         c->Request.Timeout = 0; // Don't time out
3053         c->Request.CDB[0] =
3054             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3055         start_blk = blk_rq_pos(creq);
3056 #ifdef CCISS_DEBUG
3057         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3058                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3059 #endif                          /* CCISS_DEBUG */
3060
3061         sg_init_table(tmp_sg, MAXSGENTRIES);
3062         seg = blk_rq_map_sg(q, creq, tmp_sg);
3063
3064         /* get the DMA records for the setup */
3065         if (c->Request.Type.Direction == XFER_READ)
3066                 dir = PCI_DMA_FROMDEVICE;
3067         else
3068                 dir = PCI_DMA_TODEVICE;
3069
3070         for (i = 0; i < seg; i++) {
3071                 c->SG[i].Len = tmp_sg[i].length;
3072                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3073                                                   tmp_sg[i].offset,
3074                                                   tmp_sg[i].length, dir);
3075                 c->SG[i].Addr.lower = temp64.val32.lower;
3076                 c->SG[i].Addr.upper = temp64.val32.upper;
3077                 c->SG[i].Ext = 0;       // we are not chaining
3078         }
3079         /* track how many SG entries we are using */
3080         if (seg > h->maxSG)
3081                 h->maxSG = seg;
3082
3083 #ifdef CCISS_DEBUG
3084         printk(KERN_DEBUG "cciss: Submitting %u sectors in %d segments\n",
3085                blk_rq_sectors(creq), seg);
3086 #endif                          /* CCISS_DEBUG */
3087
3088         c->Header.SGList = c->Header.SGTotal = seg;
3089         if (likely(blk_fs_request(creq))) {
3090                 if(h->cciss_read == CCISS_READ_10) {
3091                         c->Request.CDB[1] = 0;
3092                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
3093                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3094                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3095                         c->Request.CDB[5] = start_blk & 0xff;
3096                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
3097                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3098                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3099                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3100                 } else {
3101                         u32 upper32 = upper_32_bits(start_blk);
3102
3103                         c->Request.CDBLen = 16;
3104                         c->Request.CDB[1]= 0;
3105                         c->Request.CDB[2]= (upper32 >> 24) & 0xff;      //MSB
3106                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3107                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3108                         c->Request.CDB[5]= upper32 & 0xff;
3109                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3110                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3111                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3112                         c->Request.CDB[9]= start_blk & 0xff;
3113                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3114                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3115                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3116                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3117                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3118                 }
3119         } else if (blk_pc_request(creq)) {
3120                 c->Request.CDBLen = creq->cmd_len;
3121                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3122         } else {
3123                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3124                 BUG();
3125         }
3126
3127         spin_lock_irq(q->queue_lock);
3128
3129         addQ(&h->reqQ, c);
3130         h->Qdepth++;
3131         if (h->Qdepth > h->maxQsinceinit)
3132                 h->maxQsinceinit = h->Qdepth;
3133
3134         goto queue;
3135 full:
3136         blk_stop_queue(q);
3137 startio:
3138         /* We will already have the driver lock here so not need
3139          * to lock it.
3140          */
3141         start_io(h);
3142 }
3143
3144 static inline unsigned long get_next_completion(ctlr_info_t *h)
3145 {
3146         return h->access.command_completed(h);
3147 }
3148
3149 static inline int interrupt_pending(ctlr_info_t *h)
3150 {
3151         return h->access.intr_pending(h);
3152 }
3153
3154 static inline long interrupt_not_for_us(ctlr_info_t *h)
3155 {
3156         return (((h->access.intr_pending(h) == 0) ||
3157                  (h->interrupts_enabled == 0)));
3158 }
3159
3160 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3161 {
3162         ctlr_info_t *h = dev_id;
3163         CommandList_struct *c;
3164         unsigned long flags;
3165         __u32 a, a1, a2;
3166
3167         if (interrupt_not_for_us(h))
3168                 return IRQ_NONE;
3169         /*
3170          * If there are completed commands in the completion queue,
3171          * we had better do something about it.
3172          */
3173         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3174         while (interrupt_pending(h)) {
3175                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3176                         a1 = a;
3177                         if ((a & 0x04)) {
3178                                 a2 = (a >> 3);
3179                                 if (a2 >= h->nr_cmds) {
3180                                         printk(KERN_WARNING
3181                                                "cciss: controller cciss%d failed, stopping.\n",
3182                                                h->ctlr);
3183                                         fail_all_cmds(h->ctlr);
3184                                         return IRQ_HANDLED;
3185                                 }
3186
3187                                 c = h->cmd_pool + a2;
3188                                 a = c->busaddr;
3189
3190                         } else {
3191                                 struct hlist_node *tmp;
3192
3193                                 a &= ~3;
3194                                 c = NULL;
3195                                 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3196                                         if (c->busaddr == a)
3197                                                 break;
3198                                 }
3199                         }
3200                         /*
3201                          * If we've found the command, take it off the
3202                          * completion Q and free it
3203                          */
3204                         if (c && c->busaddr == a) {
3205                                 removeQ(c);
3206                                 if (c->cmd_type == CMD_RWREQ) {
3207                                         complete_command(h, c, 0);
3208                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3209                                         complete(c->waiting);
3210                                 }
3211 #                               ifdef CONFIG_CISS_SCSI_TAPE
3212                                 else if (c->cmd_type == CMD_SCSI)
3213                                         complete_scsi_command(c, 0, a1);
3214 #                               endif
3215                                 continue;
3216                         }
3217                 }
3218         }
3219
3220         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3221         return IRQ_HANDLED;
3222 }
3223
3224 static int scan_thread(void *data)
3225 {
3226         ctlr_info_t *h = data;
3227         int rc;
3228         DECLARE_COMPLETION_ONSTACK(wait);
3229         h->rescan_wait = &wait;
3230
3231         for (;;) {
3232                 rc = wait_for_completion_interruptible(&wait);
3233                 if (kthread_should_stop())
3234                         break;
3235                 if (!rc)
3236                         rebuild_lun_table(h, 0);
3237         }
3238         return 0;
3239 }
3240
3241 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3242 {
3243         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3244                 return 0;
3245
3246         switch (c->err_info->SenseInfo[12]) {
3247         case STATE_CHANGED:
3248                 printk(KERN_WARNING "cciss%d: a state change "
3249                         "detected, command retried\n", h->ctlr);
3250                 return 1;
3251         break;
3252         case LUN_FAILED:
3253                 printk(KERN_WARNING "cciss%d: LUN failure "
3254                         "detected, action required\n", h->ctlr);
3255                 return 1;
3256         break;
3257         case REPORT_LUNS_CHANGED:
3258                 printk(KERN_WARNING "cciss%d: report LUN data "
3259                         "changed\n", h->ctlr);
3260                 if (h->rescan_wait)
3261                         complete(h->rescan_wait);
3262                 return 1;
3263         break;
3264         case POWER_OR_RESET:
3265                 printk(KERN_WARNING "cciss%d: a power on "
3266                         "or device reset detected\n", h->ctlr);
3267                 return 1;
3268         break;
3269         case UNIT_ATTENTION_CLEARED:
3270                 printk(KERN_WARNING "cciss%d: unit attention "
3271                     "cleared by another initiator\n", h->ctlr);
3272                 return 1;
3273         break;
3274         default:
3275                 printk(KERN_WARNING "cciss%d: unknown "
3276                         "unit attention detected\n", h->ctlr);
3277                                 return 1;
3278         }
3279 }
3280
3281 /*
3282  *  We cannot read the structure directly, for portability we must use
3283  *   the io functions.
3284  *   This is for debug only.
3285  */
3286 #ifdef CCISS_DEBUG
3287 static void print_cfg_table(CfgTable_struct *tb)
3288 {
3289         int i;
3290         char temp_name[17];
3291
3292         printk("Controller Configuration information\n");
3293         printk("------------------------------------\n");
3294         for (i = 0; i < 4; i++)
3295                 temp_name[i] = readb(&(tb->Signature[i]));
3296         temp_name[4] = '\0';
3297         printk("   Signature = %s\n", temp_name);
3298         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3299         printk("   Transport methods supported = 0x%x\n",
3300                readl(&(tb->TransportSupport)));
3301         printk("   Transport methods active = 0x%x\n",
3302                readl(&(tb->TransportActive)));
3303         printk("   Requested transport Method = 0x%x\n",
3304                readl(&(tb->HostWrite.TransportRequest)));
3305         printk("   Coalesce Interrupt Delay = 0x%x\n",
3306                readl(&(tb->HostWrite.CoalIntDelay)));
3307         printk("   Coalesce Interrupt Count = 0x%x\n",
3308                readl(&(tb->HostWrite.CoalIntCount)));
3309         printk("   Max outstanding commands = 0x%d\n",
3310                readl(&(tb->CmdsOutMax)));
3311         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3312         for (i = 0; i < 16; i++)
3313                 temp_name[i] = readb(&(tb->ServerName[i]));
3314         temp_name[16] = '\0';
3315         printk("   Server Name = %s\n", temp_name);
3316         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3317 }
3318 #endif                          /* CCISS_DEBUG */
3319
3320 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3321 {
3322         int i, offset, mem_type, bar_type;
3323         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3324                 return 0;
3325         offset = 0;
3326         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3327                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3328                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3329                         offset += 4;
3330                 else {
3331                         mem_type = pci_resource_flags(pdev, i) &
3332                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3333                         switch (mem_type) {
3334                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3335                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3336                                 offset += 4;    /* 32 bit */
3337                                 break;
3338                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3339                                 offset += 8;
3340                                 break;
3341                         default:        /* reserved in PCI 2.2 */
3342                                 printk(KERN_WARNING
3343                                        "Base address is invalid\n");
3344                                 return -1;
3345                                 break;
3346                         }
3347                 }
3348                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3349                         return i + 1;
3350         }
3351         return -1;
3352 }
3353
3354 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3355  * controllers that are capable. If not, we use IO-APIC mode.
3356  */
3357
3358 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3359                                            struct pci_dev *pdev, __u32 board_id)
3360 {
3361 #ifdef CONFIG_PCI_MSI
3362         int err;
3363         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3364         {0, 2}, {0, 3}
3365         };
3366
3367         /* Some boards advertise MSI but don't really support it */
3368         if ((board_id == 0x40700E11) ||
3369             (board_id == 0x40800E11) ||
3370             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3371                 goto default_int_mode;
3372
3373         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3374                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3375                 if (!err) {
3376                         c->intr[0] = cciss_msix_entries[0].vector;
3377                         c->intr[1] = cciss_msix_entries[1].vector;
3378                         c->intr[2] = cciss_msix_entries[2].vector;
3379                         c->intr[3] = cciss_msix_entries[3].vector;
3380                         c->msix_vector = 1;
3381                         return;
3382                 }
3383                 if (err > 0) {
3384                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3385                                "available\n", err);
3386                         goto default_int_mode;
3387                 } else {
3388                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3389                                err);
3390                         goto default_int_mode;
3391                 }
3392         }
3393         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3394                 if (!pci_enable_msi(pdev)) {
3395                         c->msi_vector = 1;
3396                 } else {
3397                         printk(KERN_WARNING "cciss: MSI init failed\n");
3398                 }
3399         }
3400 default_int_mode:
3401 #endif                          /* CONFIG_PCI_MSI */
3402         /* if we get here we're going to use the default interrupt mode */
3403         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3404         return;
3405 }
3406
3407 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3408 {
3409         ushort subsystem_vendor_id, subsystem_device_id, command;
3410         __u32 board_id, scratchpad = 0;
3411         __u64 cfg_offset;
3412         __u32 cfg_base_addr;
3413         __u64 cfg_base_addr_index;
3414         int i, err;
3415
3416         /* check to see if controller has been disabled */
3417         /* BEFORE trying to enable it */
3418         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3419         if (!(command & 0x02)) {
3420                 printk(KERN_WARNING
3421                        "cciss: controller appears to be disabled\n");
3422                 return -ENODEV;
3423         }
3424
3425         err = pci_enable_device(pdev);
3426         if (err) {
3427                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3428                 return err;
3429         }
3430
3431         err = pci_request_regions(pdev, "cciss");
3432         if (err) {
3433                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3434                        "aborting\n");
3435                 return err;
3436         }
3437
3438         subsystem_vendor_id = pdev->subsystem_vendor;
3439         subsystem_device_id = pdev->subsystem_device;
3440         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3441                     subsystem_vendor_id);
3442
3443 #ifdef CCISS_DEBUG
3444         printk("command = %x\n", command);
3445         printk("irq = %x\n", pdev->irq);
3446         printk("board_id = %x\n", board_id);
3447 #endif                          /* CCISS_DEBUG */
3448
3449 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3450  * else we use the IO-APIC interrupt assigned to us by system ROM.
3451  */
3452         cciss_interrupt_mode(c, pdev, board_id);
3453
3454         /* find the memory BAR */
3455         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3456                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3457                         break;
3458         }
3459         if (i == DEVICE_COUNT_RESOURCE) {
3460                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3461                 err = -ENODEV;
3462                 goto err_out_free_res;
3463         }
3464
3465         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3466                                                  * already removed
3467                                                  */
3468
3469 #ifdef CCISS_DEBUG
3470         printk("address 0 = %lx\n", c->paddr);
3471 #endif                          /* CCISS_DEBUG */
3472         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3473
3474         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3475          * We poll for up to 120 secs, once per 100ms. */
3476         for (i = 0; i < 1200; i++) {
3477                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3478                 if (scratchpad == CCISS_FIRMWARE_READY)
3479                         break;
3480                 set_current_state(TASK_INTERRUPTIBLE);
3481                 schedule_timeout(HZ / 10);      /* wait 100ms */
3482         }
3483         if (scratchpad != CCISS_FIRMWARE_READY) {
3484                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3485                 err = -ENODEV;
3486                 goto err_out_free_res;
3487         }
3488
3489         /* get the address index number */
3490         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3491         cfg_base_addr &= (__u32) 0x0000ffff;
3492 #ifdef CCISS_DEBUG
3493         printk("cfg base address = %x\n", cfg_base_addr);
3494 #endif                          /* CCISS_DEBUG */
3495         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3496 #ifdef CCISS_DEBUG
3497         printk("cfg base address index = %llx\n",
3498                 (unsigned long long)cfg_base_addr_index);
3499 #endif                          /* CCISS_DEBUG */
3500         if (cfg_base_addr_index == -1) {
3501                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3502                 err = -ENODEV;
3503                 goto err_out_free_res;
3504         }
3505
3506         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3507 #ifdef CCISS_DEBUG
3508         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3509 #endif                          /* CCISS_DEBUG */
3510         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3511                                                        cfg_base_addr_index) +
3512                                     cfg_offset, sizeof(CfgTable_struct));
3513         c->board_id = board_id;
3514
3515 #ifdef CCISS_DEBUG
3516         print_cfg_table(c->cfgtable);
3517 #endif                          /* CCISS_DEBUG */
3518
3519         /* Some controllers support Zero Memory Raid (ZMR).
3520          * When configured in ZMR mode the number of supported
3521          * commands drops to 64. So instead of just setting an
3522          * arbitrary value we make the driver a little smarter.
3523          * We read the config table to tell us how many commands
3524          * are supported on the controller then subtract 4 to
3525          * leave a little room for ioctl calls.
3526          */
3527         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3528         for (i = 0; i < ARRAY_SIZE(products); i++) {
3529                 if (board_id == products[i].board_id) {
3530                         c->product_name = products[i].product_name;
3531                         c->access = *(products[i].access);
3532                         c->nr_cmds = c->max_commands - 4;
3533                         break;
3534                 }
3535         }
3536         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3537             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3538             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3539             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3540                 printk("Does not appear to be a valid CISS config table\n");
3541                 err = -ENODEV;
3542                 goto err_out_free_res;
3543         }
3544         /* We didn't find the controller in our list. We know the
3545          * signature is valid. If it's an HP device let's try to
3546          * bind to the device and fire it up. Otherwise we bail.
3547          */
3548         if (i == ARRAY_SIZE(products)) {
3549                 if (subsystem_vendor_id == PCI_VENDOR_ID_HP) {
3550                         c->product_name = products[i-1].product_name;
3551                         c->access = *(products[i-1].access);
3552                         c->nr_cmds = c->max_commands - 4;
3553                         printk(KERN_WARNING "cciss: This is an unknown "
3554                                 "Smart Array controller.\n"
3555                                 "cciss: Please update to the latest driver "
3556                                 "available from www.hp.com.\n");
3557                 } else {
3558                         printk(KERN_WARNING "cciss: Sorry, I don't know how"
3559                                 " to access the Smart Array controller %08lx\n"
3560                                         , (unsigned long)board_id);
3561                         err = -ENODEV;
3562                         goto err_out_free_res;
3563                 }
3564         }
3565 #ifdef CONFIG_X86
3566         {
3567                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3568                 __u32 prefetch;
3569                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3570                 prefetch |= 0x100;
3571                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3572         }
3573 #endif
3574
3575         /* Disabling DMA prefetch and refetch for the P600.
3576          * An ASIC bug may result in accesses to invalid memory addresses.
3577          * We've disabled prefetch for some time now. Testing with XEN
3578          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3579          */
3580         if(board_id == 0x3225103C) {
3581                 __u32 dma_prefetch;
3582                 __u32 dma_refetch;
3583                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3584                 dma_prefetch |= 0x8000;
3585                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3586                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3587                 dma_refetch |= 0x1;
3588                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3589         }
3590
3591 #ifdef CCISS_DEBUG
3592         printk("Trying to put board into Simple mode\n");
3593 #endif                          /* CCISS_DEBUG */
3594         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3595         /* Update the field, and then ring the doorbell */
3596         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3597         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3598
3599         /* under certain very rare conditions, this can take awhile.
3600          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3601          * as we enter this code.) */
3602         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3603                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3604                         break;
3605                 /* delay and try again */
3606                 set_current_state(TASK_INTERRUPTIBLE);
3607                 schedule_timeout(10);
3608         }
3609
3610 #ifdef CCISS_DEBUG
3611         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3612                readl(c->vaddr + SA5_DOORBELL));
3613 #endif                          /* CCISS_DEBUG */
3614 #ifdef CCISS_DEBUG
3615         print_cfg_table(c->cfgtable);
3616 #endif                          /* CCISS_DEBUG */
3617
3618         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3619                 printk(KERN_WARNING "cciss: unable to get board into"
3620                        " simple mode\n");
3621                 err = -ENODEV;
3622                 goto err_out_free_res;
3623         }
3624         return 0;
3625
3626 err_out_free_res:
3627         /*
3628          * Deliberately omit pci_disable_device(): it does something nasty to
3629          * Smart Array controllers that pci_enable_device does not undo
3630          */
3631         pci_release_regions(pdev);
3632         return err;
3633 }
3634
3635 /* Function to find the first free pointer into our hba[] array
3636  * Returns -1 if no free entries are left.
3637  */
3638 static int alloc_cciss_hba(void)
3639 {
3640         int i;
3641
3642         for (i = 0; i < MAX_CTLR; i++) {
3643                 if (!hba[i]) {
3644                         ctlr_info_t *p;
3645
3646                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3647                         if (!p)
3648                                 goto Enomem;
3649                         hba[i] = p;
3650                         return i;
3651                 }
3652         }
3653         printk(KERN_WARNING "cciss: This driver supports a maximum"
3654                " of %d controllers.\n", MAX_CTLR);
3655         return -1;
3656 Enomem:
3657         printk(KERN_ERR "cciss: out of memory.\n");
3658         return -1;
3659 }
3660
3661 static void free_hba(int i)
3662 {
3663         ctlr_info_t *p = hba[i];
3664         int n;
3665
3666         hba[i] = NULL;
3667         for (n = 0; n < CISS_MAX_LUN; n++)
3668                 put_disk(p->gendisk[n]);
3669         kfree(p);
3670 }
3671
3672 /* Send a message CDB to the firmware. */
3673 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3674 {
3675         typedef struct {
3676                 CommandListHeader_struct CommandHeader;
3677                 RequestBlock_struct Request;
3678                 ErrDescriptor_struct ErrorDescriptor;
3679         } Command;
3680         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3681         Command *cmd;
3682         dma_addr_t paddr64;
3683         uint32_t paddr32, tag;
3684         void __iomem *vaddr;
3685         int i, err;
3686
3687         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3688         if (vaddr == NULL)
3689                 return -ENOMEM;
3690
3691         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3692            CCISS commands, so they must be allocated from the lower 4GiB of
3693            memory. */
3694         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3695         if (err) {
3696                 iounmap(vaddr);
3697                 return -ENOMEM;
3698         }
3699
3700         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3701         if (cmd == NULL) {
3702                 iounmap(vaddr);
3703                 return -ENOMEM;
3704         }
3705
3706         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
3707            although there's no guarantee, we assume that the address is at
3708            least 4-byte aligned (most likely, it's page-aligned). */
3709         paddr32 = paddr64;
3710
3711         cmd->CommandHeader.ReplyQueue = 0;
3712         cmd->CommandHeader.SGList = 0;
3713         cmd->CommandHeader.SGTotal = 0;
3714         cmd->CommandHeader.Tag.lower = paddr32;
3715         cmd->CommandHeader.Tag.upper = 0;
3716         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
3717
3718         cmd->Request.CDBLen = 16;
3719         cmd->Request.Type.Type = TYPE_MSG;
3720         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
3721         cmd->Request.Type.Direction = XFER_NONE;
3722         cmd->Request.Timeout = 0; /* Don't time out */
3723         cmd->Request.CDB[0] = opcode;
3724         cmd->Request.CDB[1] = type;
3725         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
3726
3727         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
3728         cmd->ErrorDescriptor.Addr.upper = 0;
3729         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
3730
3731         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
3732
3733         for (i = 0; i < 10; i++) {
3734                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
3735                 if ((tag & ~3) == paddr32)
3736                         break;
3737                 schedule_timeout_uninterruptible(HZ);
3738         }
3739
3740         iounmap(vaddr);
3741
3742         /* we leak the DMA buffer here ... no choice since the controller could
3743            still complete the command. */
3744         if (i == 10) {
3745                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
3746                         opcode, type);
3747                 return -ETIMEDOUT;
3748         }
3749
3750         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
3751
3752         if (tag & 2) {
3753                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
3754                         opcode, type);
3755                 return -EIO;
3756         }
3757
3758         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
3759                 opcode, type);
3760         return 0;
3761 }
3762
3763 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
3764 #define cciss_noop(p) cciss_message(p, 3, 0)
3765
3766 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
3767 {
3768 /* the #defines are stolen from drivers/pci/msi.h. */
3769 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
3770 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
3771
3772         int pos;
3773         u16 control = 0;
3774
3775         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
3776         if (pos) {
3777                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
3778                 if (control & PCI_MSI_FLAGS_ENABLE) {
3779                         printk(KERN_INFO "cciss: resetting MSI\n");
3780                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
3781                 }
3782         }
3783
3784         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
3785         if (pos) {
3786                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
3787                 if (control & PCI_MSIX_FLAGS_ENABLE) {
3788                         printk(KERN_INFO "cciss: resetting MSI-X\n");
3789                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
3790                 }
3791         }
3792
3793         return 0;
3794 }
3795
3796 /* This does a hard reset of the controller using PCI power management
3797  * states. */
3798 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
3799 {
3800         u16 pmcsr, saved_config_space[32];
3801         int i, pos;
3802
3803         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
3804
3805         /* This is very nearly the same thing as
3806
3807            pci_save_state(pci_dev);
3808            pci_set_power_state(pci_dev, PCI_D3hot);
3809            pci_set_power_state(pci_dev, PCI_D0);
3810            pci_restore_state(pci_dev);
3811
3812            but we can't use these nice canned kernel routines on
3813            kexec, because they also check the MSI/MSI-X state in PCI
3814            configuration space and do the wrong thing when it is
3815            set/cleared.  Also, the pci_save/restore_state functions
3816            violate the ordering requirements for restoring the
3817            configuration space from the CCISS document (see the
3818            comment below).  So we roll our own .... */
3819
3820         for (i = 0; i < 32; i++)
3821                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
3822
3823         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
3824         if (pos == 0) {
3825                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
3826                 return -ENODEV;
3827         }
3828
3829         /* Quoting from the Open CISS Specification: "The Power
3830          * Management Control/Status Register (CSR) controls the power
3831          * state of the device.  The normal operating state is D0,
3832          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
3833          * the controller, place the interface device in D3 then to
3834          * D0, this causes a secondary PCI reset which will reset the
3835          * controller." */
3836
3837         /* enter the D3hot power management state */
3838         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
3839         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3840         pmcsr |= PCI_D3hot;
3841         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
3842
3843         schedule_timeout_uninterruptible(HZ >> 1);
3844
3845         /* enter the D0 power management state */
3846         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3847         pmcsr |= PCI_D0;
3848         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
3849
3850         schedule_timeout_uninterruptible(HZ >> 1);
3851
3852         /* Restore the PCI configuration space.  The Open CISS
3853          * Specification says, "Restore the PCI Configuration
3854          * Registers, offsets 00h through 60h. It is important to
3855          * restore the command register, 16-bits at offset 04h,
3856          * last. Do not restore the configuration status register,
3857          * 16-bits at offset 06h."  Note that the offset is 2*i. */
3858         for (i = 0; i < 32; i++) {
3859                 if (i == 2 || i == 3)
3860                         continue;
3861                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
3862         }
3863         wmb();
3864         pci_write_config_word(pdev, 4, saved_config_space[2]);
3865
3866         return 0;
3867 }
3868
3869 /*
3870  *  This is it.  Find all the controllers and register them.  I really hate
3871  *  stealing all these major device numbers.
3872  *  returns the number of block devices registered.
3873  */
3874 static int __devinit cciss_init_one(struct pci_dev *pdev,
3875                                     const struct pci_device_id *ent)
3876 {
3877         int i;
3878         int j = 0;
3879         int rc;
3880         int dac, return_code;
3881         InquiryData_struct *inq_buff = NULL;
3882
3883         if (reset_devices) {
3884                 /* Reset the controller with a PCI power-cycle */
3885                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
3886                         return -ENODEV;
3887
3888                 /* Now try to get the controller to respond to a no-op. Some
3889                    devices (notably the HP Smart Array 5i Controller) need
3890                    up to 30 seconds to respond. */
3891                 for (i=0; i<30; i++) {
3892                         if (cciss_noop(pdev) == 0)
3893                                 break;
3894
3895                         schedule_timeout_uninterruptible(HZ);
3896                 }
3897                 if (i == 30) {
3898                         printk(KERN_ERR "cciss: controller seems dead\n");
3899                         return -EBUSY;
3900                 }
3901         }
3902
3903         i = alloc_cciss_hba();
3904         if (i < 0)
3905                 return -1;
3906
3907         hba[i]->busy_initializing = 1;
3908         INIT_HLIST_HEAD(&hba[i]->cmpQ);
3909         INIT_HLIST_HEAD(&hba[i]->reqQ);
3910
3911         if (cciss_pci_init(hba[i], pdev) != 0)
3912                 goto clean0;
3913
3914         sprintf(hba[i]->devname, "cciss%d", i);
3915         hba[i]->ctlr = i;
3916         hba[i]->pdev = pdev;
3917
3918         if (cciss_create_hba_sysfs_entry(hba[i]))
3919                 goto clean0;
3920
3921         /* configure PCI DMA stuff */
3922         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
3923                 dac = 1;
3924         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
3925                 dac = 0;
3926         else {
3927                 printk(KERN_ERR "cciss: no suitable DMA available\n");
3928                 goto clean1;
3929         }
3930
3931         /*
3932          * register with the major number, or get a dynamic major number
3933          * by passing 0 as argument.  This is done for greater than
3934          * 8 controller support.
3935          */
3936         if (i < MAX_CTLR_ORIG)
3937                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
3938         rc = register_blkdev(hba[i]->major, hba[i]->devname);
3939         if (rc == -EBUSY || rc == -EINVAL) {
3940                 printk(KERN_ERR
3941                        "cciss:  Unable to get major number %d for %s "
3942                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
3943                 goto clean1;
3944         } else {
3945                 if (i >= MAX_CTLR_ORIG)
3946                         hba[i]->major = rc;
3947         }
3948
3949         /* make sure the board interrupts are off */
3950         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
3951         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
3952                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
3953                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
3954                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
3955                 goto clean2;
3956         }
3957
3958         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
3959                hba[i]->devname, pdev->device, pci_name(pdev),
3960                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
3961
3962         hba[i]->cmd_pool_bits =
3963             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
3964                         * sizeof(unsigned long), GFP_KERNEL);
3965         hba[i]->cmd_pool = (CommandList_struct *)
3966             pci_alloc_consistent(hba[i]->pdev,
3967                     hba[i]->nr_cmds * sizeof(CommandList_struct),
3968                     &(hba[i]->cmd_pool_dhandle));
3969         hba[i]->errinfo_pool = (ErrorInfo_struct *)
3970             pci_alloc_consistent(hba[i]->pdev,
3971                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3972                     &(hba[i]->errinfo_pool_dhandle));
3973         if ((hba[i]->cmd_pool_bits == NULL)
3974             || (hba[i]->cmd_pool == NULL)
3975             || (hba[i]->errinfo_pool == NULL)) {
3976                 printk(KERN_ERR "cciss: out of memory");
3977                 goto clean4;
3978         }
3979         spin_lock_init(&hba[i]->lock);
3980
3981         /* Initialize the pdev driver private data.
3982            have it point to hba[i].  */
3983         pci_set_drvdata(pdev, hba[i]);
3984         /* command and error info recs zeroed out before
3985            they are used */
3986         memset(hba[i]->cmd_pool_bits, 0,
3987                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
3988                         * sizeof(unsigned long));
3989
3990         hba[i]->num_luns = 0;
3991         hba[i]->highest_lun = -1;
3992         for (j = 0; j < CISS_MAX_LUN; j++) {
3993                 hba[i]->drv[j].raid_level = -1;
3994                 hba[i]->drv[j].queue = NULL;
3995                 hba[i]->gendisk[j] = NULL;
3996         }
3997
3998         cciss_scsi_setup(i);
3999
4000         /* Turn the interrupts on so we can service requests */
4001         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4002
4003         /* Get the firmware version */
4004         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4005         if (inq_buff == NULL) {
4006                 printk(KERN_ERR "cciss: out of memory\n");
4007                 goto clean4;
4008         }
4009
4010         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4011                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4012         if (return_code == IO_OK) {
4013                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4014                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4015                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4016                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4017         } else {         /* send command failed */
4018                 printk(KERN_WARNING "cciss: unable to determine firmware"
4019                         " version of controller\n");
4020         }
4021
4022         cciss_procinit(i);
4023
4024         hba[i]->cciss_max_sectors = 2048;
4025
4026         hba[i]->busy_initializing = 0;
4027
4028         rebuild_lun_table(hba[i], 1);
4029         hba[i]->cciss_scan_thread = kthread_run(scan_thread, hba[i],
4030                                 "cciss_scan%02d", i);
4031         if (IS_ERR(hba[i]->cciss_scan_thread))
4032                 return PTR_ERR(hba[i]->cciss_scan_thread);
4033
4034         return 1;
4035
4036 clean4:
4037         kfree(inq_buff);
4038         kfree(hba[i]->cmd_pool_bits);
4039         if (hba[i]->cmd_pool)
4040                 pci_free_consistent(hba[i]->pdev,
4041                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4042                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4043         if (hba[i]->errinfo_pool)
4044                 pci_free_consistent(hba[i]->pdev,
4045                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4046                                     hba[i]->errinfo_pool,
4047                                     hba[i]->errinfo_pool_dhandle);
4048         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4049 clean2:
4050         unregister_blkdev(hba[i]->major, hba[i]->devname);
4051 clean1:
4052         cciss_destroy_hba_sysfs_entry(hba[i]);
4053 clean0:
4054         hba[i]->busy_initializing = 0;
4055         /* cleanup any queues that may have been initialized */
4056         for (j=0; j <= hba[i]->highest_lun; j++){
4057                 drive_info_struct *drv = &(hba[i]->drv[j]);
4058                 if (drv->queue)
4059                         blk_cleanup_queue(drv->queue);
4060         }
4061         /*
4062          * Deliberately omit pci_disable_device(): it does something nasty to
4063          * Smart Array controllers that pci_enable_device does not undo
4064          */
4065         pci_release_regions(pdev);
4066         pci_set_drvdata(pdev, NULL);
4067         free_hba(i);
4068         return -1;
4069 }
4070
4071 static void cciss_shutdown(struct pci_dev *pdev)
4072 {
4073         ctlr_info_t *tmp_ptr;
4074         int i;
4075         char flush_buf[4];
4076         int return_code;
4077
4078         tmp_ptr = pci_get_drvdata(pdev);
4079         if (tmp_ptr == NULL)
4080                 return;
4081         i = tmp_ptr->ctlr;
4082         if (hba[i] == NULL)
4083                 return;
4084
4085         /* Turn board interrupts off  and send the flush cache command */
4086         /* sendcmd will turn off interrupt, and send the flush...
4087          * To write all data in the battery backed cache to disks */
4088         memset(flush_buf, 0, 4);
4089         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0,
4090                 CTLR_LUNID, TYPE_CMD);
4091         if (return_code == IO_OK) {
4092                 printk(KERN_INFO "Completed flushing cache on controller %d\n", i);
4093         } else {
4094                 printk(KERN_WARNING "Error flushing cache on controller %d\n", i);
4095         }
4096         free_irq(hba[i]->intr[2], hba[i]);
4097 }
4098
4099 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4100 {
4101         ctlr_info_t *tmp_ptr;
4102         int i, j;
4103
4104         if (pci_get_drvdata(pdev) == NULL) {
4105                 printk(KERN_ERR "cciss: Unable to remove device \n");
4106                 return;
4107         }
4108
4109         tmp_ptr = pci_get_drvdata(pdev);
4110         i = tmp_ptr->ctlr;
4111         if (hba[i] == NULL) {
4112                 printk(KERN_ERR "cciss: device appears to "
4113                        "already be removed \n");
4114                 return;
4115         }
4116
4117         kthread_stop(hba[i]->cciss_scan_thread);
4118
4119         remove_proc_entry(hba[i]->devname, proc_cciss);
4120         unregister_blkdev(hba[i]->major, hba[i]->devname);
4121
4122         /* remove it from the disk list */
4123         for (j = 0; j < CISS_MAX_LUN; j++) {
4124                 struct gendisk *disk = hba[i]->gendisk[j];
4125                 if (disk) {
4126                         struct request_queue *q = disk->queue;
4127
4128                         if (disk->flags & GENHD_FL_UP)
4129                                 del_gendisk(disk);
4130                         if (q)
4131                                 blk_cleanup_queue(q);
4132                 }
4133         }
4134
4135 #ifdef CONFIG_CISS_SCSI_TAPE
4136         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4137 #endif
4138
4139         cciss_shutdown(pdev);
4140
4141 #ifdef CONFIG_PCI_MSI
4142         if (hba[i]->msix_vector)
4143                 pci_disable_msix(hba[i]->pdev);
4144         else if (hba[i]->msi_vector)
4145                 pci_disable_msi(hba[i]->pdev);
4146 #endif                          /* CONFIG_PCI_MSI */
4147
4148         iounmap(hba[i]->vaddr);
4149
4150         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4151                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4152         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4153                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4154         kfree(hba[i]->cmd_pool_bits);
4155         /*
4156          * Deliberately omit pci_disable_device(): it does something nasty to
4157          * Smart Array controllers that pci_enable_device does not undo
4158          */
4159         pci_release_regions(pdev);
4160         pci_set_drvdata(pdev, NULL);
4161         cciss_destroy_hba_sysfs_entry(hba[i]);
4162         free_hba(i);
4163 }
4164
4165 static struct pci_driver cciss_pci_driver = {
4166         .name = "cciss",
4167         .probe = cciss_init_one,
4168         .remove = __devexit_p(cciss_remove_one),
4169         .id_table = cciss_pci_device_id,        /* id_table */
4170         .shutdown = cciss_shutdown,
4171 };
4172
4173 /*
4174  *  This is it.  Register the PCI driver information for the cards we control
4175  *  the OS will call our registered routines when it finds one of our cards.
4176  */
4177 static int __init cciss_init(void)
4178 {
4179         int err;
4180
4181         /*
4182          * The hardware requires that commands are aligned on a 64-bit
4183          * boundary. Given that we use pci_alloc_consistent() to allocate an
4184          * array of them, the size must be a multiple of 8 bytes.
4185          */
4186         BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4187
4188         printk(KERN_INFO DRIVER_NAME "\n");
4189
4190         err = bus_register(&cciss_bus_type);
4191         if (err)
4192                 return err;
4193
4194         /* Register for our PCI devices */
4195         err = pci_register_driver(&cciss_pci_driver);
4196         if (err)
4197                 goto err_bus_register;
4198
4199         return 0;
4200
4201 err_bus_register:
4202         bus_unregister(&cciss_bus_type);
4203         return err;
4204 }
4205
4206 static void __exit cciss_cleanup(void)
4207 {
4208         int i;
4209
4210         pci_unregister_driver(&cciss_pci_driver);
4211         /* double check that all controller entrys have been removed */
4212         for (i = 0; i < MAX_CTLR; i++) {
4213                 if (hba[i] != NULL) {
4214                         printk(KERN_WARNING "cciss: had to remove"
4215                                " controller %d\n", i);
4216                         cciss_remove_one(hba[i]->pdev);
4217                 }
4218         }
4219         remove_proc_entry("driver/cciss", NULL);
4220         bus_unregister(&cciss_bus_type);
4221 }
4222
4223 static void fail_all_cmds(unsigned long ctlr)
4224 {
4225         /* If we get here, the board is apparently dead. */
4226         ctlr_info_t *h = hba[ctlr];
4227         CommandList_struct *c;
4228         unsigned long flags;
4229
4230         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4231         h->alive = 0;           /* the controller apparently died... */
4232
4233         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4234
4235         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
4236
4237         /* move everything off the request queue onto the completed queue */
4238         while (!hlist_empty(&h->reqQ)) {
4239                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4240                 removeQ(c);
4241                 h->Qdepth--;
4242                 addQ(&h->cmpQ, c);
4243         }
4244
4245         /* Now, fail everything on the completed queue with a HW error */
4246         while (!hlist_empty(&h->cmpQ)) {
4247                 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4248                 removeQ(c);
4249                 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4250                 if (c->cmd_type == CMD_RWREQ) {
4251                         complete_command(h, c, 0);
4252                 } else if (c->cmd_type == CMD_IOCTL_PEND)
4253                         complete(c->waiting);
4254 #ifdef CONFIG_CISS_SCSI_TAPE
4255                 else if (c->cmd_type == CMD_SCSI)
4256                         complete_scsi_command(c, 0, 0);
4257 #endif
4258         }
4259         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4260         return;
4261 }
4262
4263 module_init(cciss_init);
4264 module_exit(cciss_cleanup);