drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/compat.h>
  43 #include <linux/mutex.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/io.h>
  46
  47 #include <linux/dma-mapping.h>
  48 #include <linux/blkdev.h>
  49 #include <linux/genhd.h>
  50 #include <linux/completion.h>
  51 #include <scsi/scsi.h>
  52 #include <scsi/sg.h>
  53 #include <scsi/scsi_ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/scatterlist.h>
  56 #include <linux/kthread.h>
  57
  58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  59 #define DRIVER_NAME "HP CISS Driver (v 3.6.20)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 20)
  61
  62 /* Embedded module documentation macros - see modules.h */
  63 MODULE_AUTHOR("Hewlett-Packard Company");
  64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  65 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
  66                         " SA6i P600 P800 P400 P400i E200 E200i E500 P700m"
  67                         " Smart Array G2 Series SAS/SATA Controllers");
  68 MODULE_VERSION("3.6.20");
  69 MODULE_LICENSE("GPL");
  70
  71 static int cciss_allow_hpsa;
  72 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  73 MODULE_PARM_DESC(cciss_allow_hpsa,
  74         "Prevent cciss driver from accessing hardware known to be "
  75         " supported by the hpsa driver");
  76
  77 #include "cciss_cmd.h"
  78 #include "cciss.h"
  79 #include <linux/cciss_ioctl.h>
  80
  81 /* define the PCI info for the cards we can control */
  82 static const struct pci_device_id cciss_pci_device_id[] = {
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  90         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  91         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 110         {0,}
 111 };
 112
 113 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 114
 115 /*  board_id = Subsystem Device ID & Vendor ID
 116  *  product = Marketing Name for the board
 117  *  access = Address of the struct of function pointers
 118  */
 119 static struct board_type products[] = {
 120         {0x40700E11, "Smart Array 5300", &SA5_access},
 121         {0x40800E11, "Smart Array 5i", &SA5B_access},
 122         {0x40820E11, "Smart Array 532", &SA5B_access},
 123         {0x40830E11, "Smart Array 5312", &SA5B_access},
 124         {0x409A0E11, "Smart Array 641", &SA5_access},
 125         {0x409B0E11, "Smart Array 642", &SA5_access},
 126         {0x409C0E11, "Smart Array 6400", &SA5_access},
 127         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 128         {0x40910E11, "Smart Array 6i", &SA5_access},
 129         {0x3225103C, "Smart Array P600", &SA5_access},
 130         {0x3235103C, "Smart Array P400i", &SA5_access},
 131         {0x3211103C, "Smart Array E200i", &SA5_access},
 132         {0x3212103C, "Smart Array E200", &SA5_access},
 133         {0x3213103C, "Smart Array E200i", &SA5_access},
 134         {0x3214103C, "Smart Array E200i", &SA5_access},
 135         {0x3215103C, "Smart Array E200i", &SA5_access},
 136         {0x3237103C, "Smart Array E500", &SA5_access},
 137 /* controllers below this line are also supported by the hpsa driver. */
 138 #define HPSA_BOUNDARY 0x3223103C
 139         {0x3223103C, "Smart Array P800", &SA5_access},
 140         {0x3234103C, "Smart Array P400", &SA5_access},
 141         {0x323D103C, "Smart Array P700m", &SA5_access},
 142         {0x3241103C, "Smart Array P212", &SA5_access},
 143         {0x3243103C, "Smart Array P410", &SA5_access},
 144         {0x3245103C, "Smart Array P410i", &SA5_access},
 145         {0x3247103C, "Smart Array P411", &SA5_access},
 146         {0x3249103C, "Smart Array P812", &SA5_access},
 147         {0x324A103C, "Smart Array P712m", &SA5_access},
 148         {0x324B103C, "Smart Array P711m", &SA5_access},
 149 };
 150
 151 /* How long to wait (in milliseconds) for board to go into simple mode */
 152 #define MAX_CONFIG_WAIT 30000
 153 #define MAX_IOCTL_CONFIG_WAIT 1000
 154
 155 /*define how many times we will try a command because of bus resets */
 156 #define MAX_CMD_RETRIES 3
 157
 158 #define MAX_CTLR        32
 159
 160 /* Originally cciss driver only supports 8 major numbers */
 161 #define MAX_CTLR_ORIG   8
 162
 163 static ctlr_info_t *hba[MAX_CTLR];
 164
 165 static struct task_struct *cciss_scan_thread;
 166 static DEFINE_MUTEX(scan_mutex);
 167 static LIST_HEAD(scan_q);
 168
 169 static void do_cciss_request(struct request_queue *q);
 170 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
 171 static int cciss_open(struct block_device *bdev, fmode_t mode);
 172 static int cciss_release(struct gendisk *disk, fmode_t mode);
 173 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 174                        unsigned int cmd, unsigned long arg);
 175 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 176
 177 static int cciss_revalidate(struct gendisk *disk);
 178 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 179 static int deregister_disk(ctlr_info_t *h, int drv_index,
 180                            int clear_all, int via_ioctl);
 181
 182 static void cciss_read_capacity(int ctlr, int logvol,
 183                         sector_t *total_size, unsigned int *block_size);
 184 static void cciss_read_capacity_16(int ctlr, int logvol,
 185                         sector_t *total_size, unsigned int *block_size);
 186 static void cciss_geometry_inquiry(int ctlr, int logvol,
 187                         sector_t total_size,
 188                         unsigned int block_size, InquiryData_struct *inq_buff,
 189                                    drive_info_struct *drv);
 190 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 191                                            __u32);
 192 static void start_io(ctlr_info_t *h);
 193 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 194                         __u8 page_code, unsigned char scsi3addr[],
 195                         int cmd_type);
 196 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 197         int attempt_retry);
 198 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 199
 200 static void fail_all_cmds(unsigned long ctlr);
 201 static int add_to_scan_list(struct ctlr_info *h);
 202 static int scan_thread(void *data);
 203 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 204 static void cciss_hba_release(struct device *dev);
 205 static void cciss_device_release(struct device *dev);
 206 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 207 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 208
 209 #ifdef CONFIG_PROC_FS
 210 static void cciss_procinit(int i);
 211 #else
 212 static void cciss_procinit(int i)
 213 {
 214 }
 215 #endif                          /* CONFIG_PROC_FS */
 216
 217 #ifdef CONFIG_COMPAT
 218 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 219                               unsigned, unsigned long);
 220 #endif
 221
 222 static const struct block_device_operations cciss_fops = {
 223         .owner = THIS_MODULE,
 224         .open = cciss_open,
 225         .release = cciss_release,
 226         .locked_ioctl = cciss_ioctl,
 227         .getgeo = cciss_getgeo,
 228 #ifdef CONFIG_COMPAT
 229         .compat_ioctl = cciss_compat_ioctl,
 230 #endif
 231         .revalidate_disk = cciss_revalidate,
 232 };
 233
 234 /*
 235  * Enqueuing and dequeuing functions for cmdlists.
 236  */
 237 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 238 {
 239         hlist_add_head(&c->list, list);
 240 }
 241
 242 static inline void removeQ(CommandList_struct *c)
 243 {
 244         /*
 245          * After kexec/dump some commands might still
 246          * be in flight, which the firmware will try
 247          * to complete. Resetting the firmware doesn't work
 248          * with old fw revisions, so we have to mark
 249          * them off as 'stale' to prevent the driver from
 250          * falling over.
 251          */
 252         if (WARN_ON(hlist_unhashed(&c->list))) {
 253                 c->cmd_type = CMD_MSG_STALE;
 254                 return;
 255         }
 256
 257         hlist_del_init(&c->list);
 258 }
 259
 260 #include "cciss_scsi.c"         /* For SCSI tape support */
 261
 262 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 263         "UNKNOWN"
 264 };
 265 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
 266
 267 #ifdef CONFIG_PROC_FS
 268
 269 /*
 270  * Report information about this controller.
 271  */
 272 #define ENG_GIG 1000000000
 273 #define ENG_GIG_FACTOR (ENG_GIG/512)
 274 #define ENGAGE_SCSI     "engage scsi"
 275
 276 static struct proc_dir_entry *proc_cciss;
 277
 278 static void cciss_seq_show_header(struct seq_file *seq)
 279 {
 280         ctlr_info_t *h = seq->private;
 281
 282         seq_printf(seq, "%s: HP %s Controller\n"
 283                 "Board ID: 0x%08lx\n"
 284                 "Firmware Version: %c%c%c%c\n"
 285                 "IRQ: %d\n"
 286                 "Logical drives: %d\n"
 287                 "Current Q depth: %d\n"
 288                 "Current # commands on controller: %d\n"
 289                 "Max Q depth since init: %d\n"
 290                 "Max # commands on controller since init: %d\n"
 291                 "Max SG entries since init: %d\n",
 292                 h->devname,
 293                 h->product_name,
 294                 (unsigned long)h->board_id,
 295                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 296                 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
 297                 h->num_luns,
 298                 h->Qdepth, h->commands_outstanding,
 299                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 300
 301 #ifdef CONFIG_CISS_SCSI_TAPE
 302         cciss_seq_tape_report(seq, h->ctlr);
 303 #endif /* CONFIG_CISS_SCSI_TAPE */
 304 }
 305
 306 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 307 {
 308         ctlr_info_t *h = seq->private;
 309         unsigned ctlr = h->ctlr;
 310         unsigned long flags;
 311
 312         /* prevent displaying bogus info during configuration
 313          * or deconfiguration of a logical volume
 314          */
 315         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 316         if (h->busy_configuring) {
 317                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 318                 return ERR_PTR(-EBUSY);
 319         }
 320         h->busy_configuring = 1;
 321         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 322
 323         if (*pos == 0)
 324                 cciss_seq_show_header(seq);
 325
 326         return pos;
 327 }
 328
 329 static int cciss_seq_show(struct seq_file *seq, void *v)
 330 {
 331         sector_t vol_sz, vol_sz_frac;
 332         ctlr_info_t *h = seq->private;
 333         unsigned ctlr = h->ctlr;
 334         loff_t *pos = v;
 335         drive_info_struct *drv = h->drv[*pos];
 336
 337         if (*pos > h->highest_lun)
 338                 return 0;
 339
 340         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 341                 return 0;
 342
 343         if (drv->heads == 0)
 344                 return 0;
 345
 346         vol_sz = drv->nr_blocks;
 347         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 348         vol_sz_frac *= 100;
 349         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 350
 351         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 352                 drv->raid_level = RAID_UNKNOWN;
 353         seq_printf(seq, "cciss/c%dd%d:"
 354                         "\t%4u.%02uGB\tRAID %s\n",
 355                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 356                         raid_label[drv->raid_level]);
 357         return 0;
 358 }
 359
 360 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 361 {
 362         ctlr_info_t *h = seq->private;
 363
 364         if (*pos > h->highest_lun)
 365                 return NULL;
 366         *pos += 1;
 367
 368         return pos;
 369 }
 370
 371 static void cciss_seq_stop(struct seq_file *seq, void *v)
 372 {
 373         ctlr_info_t *h = seq->private;
 374
 375         /* Only reset h->busy_configuring if we succeeded in setting
 376          * it during cciss_seq_start. */
 377         if (v == ERR_PTR(-EBUSY))
 378                 return;
 379
 380         h->busy_configuring = 0;
 381 }
 382
 383 static const struct seq_operations cciss_seq_ops = {
 384         .start = cciss_seq_start,
 385         .show  = cciss_seq_show,
 386         .next  = cciss_seq_next,
 387         .stop  = cciss_seq_stop,
 388 };
 389
 390 static int cciss_seq_open(struct inode *inode, struct file *file)
 391 {
 392         int ret = seq_open(file, &cciss_seq_ops);
 393         struct seq_file *seq = file->private_data;
 394
 395         if (!ret)
 396                 seq->private = PDE(inode)->data;
 397
 398         return ret;
 399 }
 400
 401 static ssize_t
 402 cciss_proc_write(struct file *file, const char __user *buf,
 403                  size_t length, loff_t *ppos)
 404 {
 405         int err;
 406         char *buffer;
 407
 408 #ifndef CONFIG_CISS_SCSI_TAPE
 409         return -EINVAL;
 410 #endif
 411
 412         if (!buf || length > PAGE_SIZE - 1)
 413                 return -EINVAL;
 414
 415         buffer = (char *)__get_free_page(GFP_KERNEL);
 416         if (!buffer)
 417                 return -ENOMEM;
 418
 419         err = -EFAULT;
 420         if (copy_from_user(buffer, buf, length))
 421                 goto out;
 422         buffer[length] = '\0';
 423
 424 #ifdef CONFIG_CISS_SCSI_TAPE
 425         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 426                 struct seq_file *seq = file->private_data;
 427                 ctlr_info_t *h = seq->private;
 428
 429                 err = cciss_engage_scsi(h->ctlr);
 430                 if (err == 0)
 431                         err = length;
 432         } else
 433 #endif /* CONFIG_CISS_SCSI_TAPE */
 434                 err = -EINVAL;
 435         /* might be nice to have "disengage" too, but it's not
 436            safely possible. (only 1 module use count, lock issues.) */
 437
 438 out:
 439         free_page((unsigned long)buffer);
 440         return err;
 441 }
 442
 443 static const struct file_operations cciss_proc_fops = {
 444         .owner   = THIS_MODULE,
 445         .open    = cciss_seq_open,
 446         .read    = seq_read,
 447         .llseek  = seq_lseek,
 448         .release = seq_release,
 449         .write   = cciss_proc_write,
 450 };
 451
 452 static void __devinit cciss_procinit(int i)
 453 {
 454         struct proc_dir_entry *pde;
 455
 456         if (proc_cciss == NULL)
 457                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 458         if (!proc_cciss)
 459                 return;
 460         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 461                                         S_IROTH, proc_cciss,
 462                                         &cciss_proc_fops, hba[i]);
 463 }
 464 #endif                          /* CONFIG_PROC_FS */
 465
 466 #define MAX_PRODUCT_NAME_LEN 19
 467
 468 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 469 #define to_drv(n) container_of(n, drive_info_struct, dev)
 470
 471 static ssize_t host_store_rescan(struct device *dev,
 472                                  struct device_attribute *attr,
 473                                  const char *buf, size_t count)
 474 {
 475         struct ctlr_info *h = to_hba(dev);
 476
 477         add_to_scan_list(h);
 478         wake_up_process(cciss_scan_thread);
 479         wait_for_completion_interruptible(&h->scan_wait);
 480
 481         return count;
 482 }
 483 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 484
 485 static ssize_t dev_show_unique_id(struct device *dev,
 486                                  struct device_attribute *attr,
 487                                  char *buf)
 488 {
 489         drive_info_struct *drv = to_drv(dev);
 490         struct ctlr_info *h = to_hba(drv->dev.parent);
 491         __u8 sn[16];
 492         unsigned long flags;
 493         int ret = 0;
 494
 495         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 496         if (h->busy_configuring)
 497                 ret = -EBUSY;
 498         else
 499                 memcpy(sn, drv->serial_no, sizeof(sn));
 500         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 501
 502         if (ret)
 503                 return ret;
 504         else
 505                 return snprintf(buf, 16 * 2 + 2,
 506                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 507                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 508                                 sn[0], sn[1], sn[2], sn[3],
 509                                 sn[4], sn[5], sn[6], sn[7],
 510                                 sn[8], sn[9], sn[10], sn[11],
 511                                 sn[12], sn[13], sn[14], sn[15]);
 512 }
 513 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 514
 515 static ssize_t dev_show_vendor(struct device *dev,
 516                                struct device_attribute *attr,
 517                                char *buf)
 518 {
 519         drive_info_struct *drv = to_drv(dev);
 520         struct ctlr_info *h = to_hba(drv->dev.parent);
 521         char vendor[VENDOR_LEN + 1];
 522         unsigned long flags;
 523         int ret = 0;
 524
 525         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 526         if (h->busy_configuring)
 527                 ret = -EBUSY;
 528         else
 529                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 530         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 531
 532         if (ret)
 533                 return ret;
 534         else
 535                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 536 }
 537 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 538
 539 static ssize_t dev_show_model(struct device *dev,
 540                               struct device_attribute *attr,
 541                               char *buf)
 542 {
 543         drive_info_struct *drv = to_drv(dev);
 544         struct ctlr_info *h = to_hba(drv->dev.parent);
 545         char model[MODEL_LEN + 1];
 546         unsigned long flags;
 547         int ret = 0;
 548
 549         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 550         if (h->busy_configuring)
 551                 ret = -EBUSY;
 552         else
 553                 memcpy(model, drv->model, MODEL_LEN + 1);
 554         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 555
 556         if (ret)
 557                 return ret;
 558         else
 559                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 560 }
 561 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 562
 563 static ssize_t dev_show_rev(struct device *dev,
 564                             struct device_attribute *attr,
 565                             char *buf)
 566 {
 567         drive_info_struct *drv = to_drv(dev);
 568         struct ctlr_info *h = to_hba(drv->dev.parent);
 569         char rev[REV_LEN + 1];
 570         unsigned long flags;
 571         int ret = 0;
 572
 573         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 574         if (h->busy_configuring)
 575                 ret = -EBUSY;
 576         else
 577                 memcpy(rev, drv->rev, REV_LEN + 1);
 578         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 579
 580         if (ret)
 581                 return ret;
 582         else
 583                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 584 }
 585 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 586
 587 static ssize_t cciss_show_lunid(struct device *dev,
 588                                 struct device_attribute *attr, char *buf)
 589 {
 590         drive_info_struct *drv = to_drv(dev);
 591         struct ctlr_info *h = to_hba(drv->dev.parent);
 592         unsigned long flags;
 593         unsigned char lunid[8];
 594
 595         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 596         if (h->busy_configuring) {
 597                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 598                 return -EBUSY;
 599         }
 600         if (!drv->heads) {
 601                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 602                 return -ENOTTY;
 603         }
 604         memcpy(lunid, drv->LunID, sizeof(lunid));
 605         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 606         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 607                 lunid[0], lunid[1], lunid[2], lunid[3],
 608                 lunid[4], lunid[5], lunid[6], lunid[7]);
 609 }
 610 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 611
 612 static ssize_t cciss_show_raid_level(struct device *dev,
 613                                      struct device_attribute *attr, char *buf)
 614 {
 615         drive_info_struct *drv = to_drv(dev);
 616         struct ctlr_info *h = to_hba(drv->dev.parent);
 617         int raid;
 618         unsigned long flags;
 619
 620         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 621         if (h->busy_configuring) {
 622                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 623                 return -EBUSY;
 624         }
 625         raid = drv->raid_level;
 626         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 627         if (raid < 0 || raid > RAID_UNKNOWN)
 628                 raid = RAID_UNKNOWN;
 629
 630         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 631                         raid_label[raid]);
 632 }
 633 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 634
 635 static ssize_t cciss_show_usage_count(struct device *dev,
 636                                       struct device_attribute *attr, char *buf)
 637 {
 638         drive_info_struct *drv = to_drv(dev);
 639         struct ctlr_info *h = to_hba(drv->dev.parent);
 640         unsigned long flags;
 641         int count;
 642
 643         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 644         if (h->busy_configuring) {
 645                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 646                 return -EBUSY;
 647         }
 648         count = drv->usage_count;
 649         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 650         return snprintf(buf, 20, "%d\n", count);
 651 }
 652 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 653
 654 static struct attribute *cciss_host_attrs[] = {
 655         &dev_attr_rescan.attr,
 656         NULL
 657 };
 658
 659 static struct attribute_group cciss_host_attr_group = {
 660         .attrs = cciss_host_attrs,
 661 };
 662
 663 static const struct attribute_group *cciss_host_attr_groups[] = {
 664         &cciss_host_attr_group,
 665         NULL
 666 };
 667
 668 static struct device_type cciss_host_type = {
 669         .name           = "cciss_host",
 670         .groups         = cciss_host_attr_groups,
 671         .release        = cciss_hba_release,
 672 };
 673
 674 static struct attribute *cciss_dev_attrs[] = {
 675         &dev_attr_unique_id.attr,
 676         &dev_attr_model.attr,
 677         &dev_attr_vendor.attr,
 678         &dev_attr_rev.attr,
 679         &dev_attr_lunid.attr,
 680         &dev_attr_raid_level.attr,
 681         &dev_attr_usage_count.attr,
 682         NULL
 683 };
 684
 685 static struct attribute_group cciss_dev_attr_group = {
 686         .attrs = cciss_dev_attrs,
 687 };
 688
 689 static const struct attribute_group *cciss_dev_attr_groups[] = {
 690         &cciss_dev_attr_group,
 691         NULL
 692 };
 693
 694 static struct device_type cciss_dev_type = {
 695         .name           = "cciss_device",
 696         .groups         = cciss_dev_attr_groups,
 697         .release        = cciss_device_release,
 698 };
 699
 700 static struct bus_type cciss_bus_type = {
 701         .name           = "cciss",
 702 };
 703
 704 /*
 705  * cciss_hba_release is called when the reference count
 706  * of h->dev goes to zero.
 707  */
 708 static void cciss_hba_release(struct device *dev)
 709 {
 710         /*
 711          * nothing to do, but need this to avoid a warning
 712          * about not having a release handler from lib/kref.c.
 713          */
 714 }
 715
 716 /*
 717  * Initialize sysfs entry for each controller.  This sets up and registers
 718  * the 'cciss#' directory for each individual controller under
 719  * /sys/bus/pci/devices/<dev>/.
 720  */
 721 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 722 {
 723         device_initialize(&h->dev);
 724         h->dev.type = &cciss_host_type;
 725         h->dev.bus = &cciss_bus_type;
 726         dev_set_name(&h->dev, "%s", h->devname);
 727         h->dev.parent = &h->pdev->dev;
 728
 729         return device_add(&h->dev);
 730 }
 731
 732 /*
 733  * Remove sysfs entries for an hba.
 734  */
 735 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 736 {
 737         device_del(&h->dev);
 738         put_device(&h->dev); /* final put. */
 739 }
 740
 741 /* cciss_device_release is called when the reference count
 742  * of h->drv[x]dev goes to zero.
 743  */
 744 static void cciss_device_release(struct device *dev)
 745 {
 746         drive_info_struct *drv = to_drv(dev);
 747         kfree(drv);
 748 }
 749
 750 /*
 751  * Initialize sysfs for each logical drive.  This sets up and registers
 752  * the 'c#d#' directory for each individual logical drive under
 753  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 754  * /sys/block/cciss!c#d# to this entry.
 755  */
 756 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 757                                        int drv_index)
 758 {
 759         struct device *dev;
 760
 761         if (h->drv[drv_index]->device_initialized)
 762                 return 0;
 763
 764         dev = &h->drv[drv_index]->dev;
 765         device_initialize(dev);
 766         dev->type = &cciss_dev_type;
 767         dev->bus = &cciss_bus_type;
 768         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 769         dev->parent = &h->dev;
 770         h->drv[drv_index]->device_initialized = 1;
 771         return device_add(dev);
 772 }
 773
 774 /*
 775  * Remove sysfs entries for a logical drive.
 776  */
 777 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 778         int ctlr_exiting)
 779 {
 780         struct device *dev = &h->drv[drv_index]->dev;
 781
 782         /* special case for c*d0, we only destroy it on controller exit */
 783         if (drv_index == 0 && !ctlr_exiting)
 784                 return;
 785
 786         device_del(dev);
 787         put_device(dev); /* the "final" put. */
 788         h->drv[drv_index] = NULL;
 789 }
 790
 791 /*
 792  * For operations that cannot sleep, a command block is allocated at init,
 793  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 794  * which ones are free or in use.  For operations that can wait for kmalloc
 795  * to possible sleep, this routine can be called with get_from_pool set to 0.
 796  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 797  */
 798 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 799 {
 800         CommandList_struct *c;
 801         int i;
 802         u64bit temp64;
 803         dma_addr_t cmd_dma_handle, err_dma_handle;
 804
 805         if (!get_from_pool) {
 806                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 807                         sizeof(CommandList_struct), &cmd_dma_handle);
 808                 if (c == NULL)
 809                         return NULL;
 810                 memset(c, 0, sizeof(CommandList_struct));
 811
 812                 c->cmdindex = -1;
 813
 814                 c->err_info = (ErrorInfo_struct *)
 815                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 816                             &err_dma_handle);
 817
 818                 if (c->err_info == NULL) {
 819                         pci_free_consistent(h->pdev,
 820                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 821                         return NULL;
 822                 }
 823                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 824         } else {                /* get it out of the controllers pool */
 825
 826                 do {
 827                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 828                         if (i == h->nr_cmds)
 829                                 return NULL;
 830                 } while (test_and_set_bit
 831                          (i & (BITS_PER_LONG - 1),
 832                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 833 #ifdef CCISS_DEBUG
 834                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 835 #endif
 836                 c = h->cmd_pool + i;
 837                 memset(c, 0, sizeof(CommandList_struct));
 838                 cmd_dma_handle = h->cmd_pool_dhandle
 839                     + i * sizeof(CommandList_struct);
 840                 c->err_info = h->errinfo_pool + i;
 841                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 842                 err_dma_handle = h->errinfo_pool_dhandle
 843                     + i * sizeof(ErrorInfo_struct);
 844                 h->nr_allocs++;
 845
 846                 c->cmdindex = i;
 847         }
 848
 849         INIT_HLIST_NODE(&c->list);
 850         c->busaddr = (__u32) cmd_dma_handle;
 851         temp64.val = (__u64) err_dma_handle;
 852         c->ErrDesc.Addr.lower = temp64.val32.lower;
 853         c->ErrDesc.Addr.upper = temp64.val32.upper;
 854         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 855
 856         c->ctlr = h->ctlr;
 857         return c;
 858 }
 859
 860 /*
 861  * Frees a command block that was previously allocated with cmd_alloc().
 862  */
 863 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 864 {
 865         int i;
 866         u64bit temp64;
 867
 868         if (!got_from_pool) {
 869                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 870                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 871                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 872                                     c->err_info, (dma_addr_t) temp64.val);
 873                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 874                                     c, (dma_addr_t) c->busaddr);
 875         } else {
 876                 i = c - h->cmd_pool;
 877                 clear_bit(i & (BITS_PER_LONG - 1),
 878                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 879                 h->nr_frees++;
 880         }
 881 }
 882
 883 static inline ctlr_info_t *get_host(struct gendisk *disk)
 884 {
 885         return disk->queue->queuedata;
 886 }
 887
 888 static inline drive_info_struct *get_drv(struct gendisk *disk)
 889 {
 890         return disk->private_data;
 891 }
 892
 893 /*
 894  * Open.  Make sure the device is really there.
 895  */
 896 static int cciss_open(struct block_device *bdev, fmode_t mode)
 897 {
 898         ctlr_info_t *host = get_host(bdev->bd_disk);
 899         drive_info_struct *drv = get_drv(bdev->bd_disk);
 900
 901 #ifdef CCISS_DEBUG
 902         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
 903 #endif                          /* CCISS_DEBUG */
 904
 905         if (drv->busy_configuring)
 906                 return -EBUSY;
 907         /*
 908          * Root is allowed to open raw volume zero even if it's not configured
 909          * so array config can still work. Root is also allowed to open any
 910          * volume that has a LUN ID, so it can issue IOCTL to reread the
 911          * disk information.  I don't think I really like this
 912          * but I'm already using way to many device nodes to claim another one
 913          * for "raw controller".
 914          */
 915         if (drv->heads == 0) {
 916                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
 917                         /* if not node 0 make sure it is a partition = 0 */
 918                         if (MINOR(bdev->bd_dev) & 0x0f) {
 919                                 return -ENXIO;
 920                                 /* if it is, make sure we have a LUN ID */
 921                         } else if (memcmp(drv->LunID, CTLR_LUNID,
 922                                 sizeof(drv->LunID))) {
 923                                 return -ENXIO;
 924                         }
 925                 }
 926                 if (!capable(CAP_SYS_ADMIN))
 927                         return -EPERM;
 928         }
 929         drv->usage_count++;
 930         host->usage_count++;
 931         return 0;
 932 }
 933
 934 /*
 935  * Close.  Sync first.
 936  */
 937 static int cciss_release(struct gendisk *disk, fmode_t mode)
 938 {
 939         ctlr_info_t *host = get_host(disk);
 940         drive_info_struct *drv = get_drv(disk);
 941
 942 #ifdef CCISS_DEBUG
 943         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
 944 #endif                          /* CCISS_DEBUG */
 945
 946         drv->usage_count--;
 947         host->usage_count--;
 948         return 0;
 949 }
 950
 951 #ifdef CONFIG_COMPAT
 952
 953 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 954                     unsigned cmd, unsigned long arg)
 955 {
 956         int ret;
 957         lock_kernel();
 958         ret = cciss_ioctl(bdev, mode, cmd, arg);
 959         unlock_kernel();
 960         return ret;
 961 }
 962
 963 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 964                                   unsigned cmd, unsigned long arg);
 965 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
 966                                       unsigned cmd, unsigned long arg);
 967
 968 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
 969                               unsigned cmd, unsigned long arg)
 970 {
 971         switch (cmd) {
 972         case CCISS_GETPCIINFO:
 973         case CCISS_GETINTINFO:
 974         case CCISS_SETINTINFO:
 975         case CCISS_GETNODENAME:
 976         case CCISS_SETNODENAME:
 977         case CCISS_GETHEARTBEAT:
 978         case CCISS_GETBUSTYPES:
 979         case CCISS_GETFIRMVER:
 980         case CCISS_GETDRIVVER:
 981         case CCISS_REVALIDVOLS:
 982         case CCISS_DEREGDISK:
 983         case CCISS_REGNEWDISK:
 984         case CCISS_REGNEWD:
 985         case CCISS_RESCANDISK:
 986         case CCISS_GETLUNINFO:
 987                 return do_ioctl(bdev, mode, cmd, arg);
 988
 989         case CCISS_PASSTHRU32:
 990                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
 991         case CCISS_BIG_PASSTHRU32:
 992                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
 993
 994         default:
 995                 return -ENOIOCTLCMD;
 996         }
 997 }
 998
 999 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1000                                   unsigned cmd, unsigned long arg)
1001 {
1002         IOCTL32_Command_struct __user *arg32 =
1003             (IOCTL32_Command_struct __user *) arg;
1004         IOCTL_Command_struct arg64;
1005         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1006         int err;
1007         u32 cp;
1008
1009         err = 0;
1010         err |=
1011             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1012                            sizeof(arg64.LUN_info));
1013         err |=
1014             copy_from_user(&arg64.Request, &arg32->Request,
1015                            sizeof(arg64.Request));
1016         err |=
1017             copy_from_user(&arg64.error_info, &arg32->error_info,
1018                            sizeof(arg64.error_info));
1019         err |= get_user(arg64.buf_size, &arg32->buf_size);
1020         err |= get_user(cp, &arg32->buf);
1021         arg64.buf = compat_ptr(cp);
1022         err |= copy_to_user(p, &arg64, sizeof(arg64));
1023
1024         if (err)
1025                 return -EFAULT;
1026
1027         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1028         if (err)
1029                 return err;
1030         err |=
1031             copy_in_user(&arg32->error_info, &p->error_info,
1032                          sizeof(arg32->error_info));
1033         if (err)
1034                 return -EFAULT;
1035         return err;
1036 }
1037
1038 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1039                                       unsigned cmd, unsigned long arg)
1040 {
1041         BIG_IOCTL32_Command_struct __user *arg32 =
1042             (BIG_IOCTL32_Command_struct __user *) arg;
1043         BIG_IOCTL_Command_struct arg64;
1044         BIG_IOCTL_Command_struct __user *p =
1045             compat_alloc_user_space(sizeof(arg64));
1046         int err;
1047         u32 cp;
1048
1049         err = 0;
1050         err |=
1051             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1052                            sizeof(arg64.LUN_info));
1053         err |=
1054             copy_from_user(&arg64.Request, &arg32->Request,
1055                            sizeof(arg64.Request));
1056         err |=
1057             copy_from_user(&arg64.error_info, &arg32->error_info,
1058                            sizeof(arg64.error_info));
1059         err |= get_user(arg64.buf_size, &arg32->buf_size);
1060         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1061         err |= get_user(cp, &arg32->buf);
1062         arg64.buf = compat_ptr(cp);
1063         err |= copy_to_user(p, &arg64, sizeof(arg64));
1064
1065         if (err)
1066                 return -EFAULT;
1067
1068         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1069         if (err)
1070                 return err;
1071         err |=
1072             copy_in_user(&arg32->error_info, &p->error_info,
1073                          sizeof(arg32->error_info));
1074         if (err)
1075                 return -EFAULT;
1076         return err;
1077 }
1078 #endif
1079
1080 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1081 {
1082         drive_info_struct *drv = get_drv(bdev->bd_disk);
1083
1084         if (!drv->cylinders)
1085                 return -ENXIO;
1086
1087         geo->heads = drv->heads;
1088         geo->sectors = drv->sectors;
1089         geo->cylinders = drv->cylinders;
1090         return 0;
1091 }
1092
1093 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1094 {
1095         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1096                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1097                 (void)check_for_unit_attention(host, c);
1098 }
1099 /*
1100  * ioctl
1101  */
1102 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1103                        unsigned int cmd, unsigned long arg)
1104 {
1105         struct gendisk *disk = bdev->bd_disk;
1106         ctlr_info_t *host = get_host(disk);
1107         drive_info_struct *drv = get_drv(disk);
1108         int ctlr = host->ctlr;
1109         void __user *argp = (void __user *)arg;
1110
1111 #ifdef CCISS_DEBUG
1112         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1113 #endif                          /* CCISS_DEBUG */
1114
1115         switch (cmd) {
1116         case CCISS_GETPCIINFO:
1117                 {
1118                         cciss_pci_info_struct pciinfo;
1119
1120                         if (!arg)
1121                                 return -EINVAL;
1122                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1123                         pciinfo.bus = host->pdev->bus->number;
1124                         pciinfo.dev_fn = host->pdev->devfn;
1125                         pciinfo.board_id = host->board_id;
1126                         if (copy_to_user
1127                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1128                                 return -EFAULT;
1129                         return 0;
1130                 }
1131         case CCISS_GETINTINFO:
1132                 {
1133                         cciss_coalint_struct intinfo;
1134                         if (!arg)
1135                                 return -EINVAL;
1136                         intinfo.delay =
1137                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1138                         intinfo.count =
1139                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1140                         if (copy_to_user
1141                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1142                                 return -EFAULT;
1143                         return 0;
1144                 }
1145         case CCISS_SETINTINFO:
1146                 {
1147                         cciss_coalint_struct intinfo;
1148                         unsigned long flags;
1149                         int i;
1150
1151                         if (!arg)
1152                                 return -EINVAL;
1153                         if (!capable(CAP_SYS_ADMIN))
1154                                 return -EPERM;
1155                         if (copy_from_user
1156                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1157                                 return -EFAULT;
1158                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1159                         {
1160 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1161                                 return -EINVAL;
1162                         }
1163                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1164                         /* Update the field, and then ring the doorbell */
1165                         writel(intinfo.delay,
1166                                &(host->cfgtable->HostWrite.CoalIntDelay));
1167                         writel(intinfo.count,
1168                                &(host->cfgtable->HostWrite.CoalIntCount));
1169                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1170
1171                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1172                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1173                                       & CFGTBL_ChangeReq))
1174                                         break;
1175                                 /* delay and try again */
1176                                 udelay(1000);
1177                         }
1178                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1179                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1180                                 return -EAGAIN;
1181                         return 0;
1182                 }
1183         case CCISS_GETNODENAME:
1184                 {
1185                         NodeName_type NodeName;
1186                         int i;
1187
1188                         if (!arg)
1189                                 return -EINVAL;
1190                         for (i = 0; i < 16; i++)
1191                                 NodeName[i] =
1192                                     readb(&host->cfgtable->ServerName[i]);
1193                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1194                                 return -EFAULT;
1195                         return 0;
1196                 }
1197         case CCISS_SETNODENAME:
1198                 {
1199                         NodeName_type NodeName;
1200                         unsigned long flags;
1201                         int i;
1202
1203                         if (!arg)
1204                                 return -EINVAL;
1205                         if (!capable(CAP_SYS_ADMIN))
1206                                 return -EPERM;
1207
1208                         if (copy_from_user
1209                             (NodeName, argp, sizeof(NodeName_type)))
1210                                 return -EFAULT;
1211
1212                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1213
1214                         /* Update the field, and then ring the doorbell */
1215                         for (i = 0; i < 16; i++)
1216                                 writeb(NodeName[i],
1217                                        &host->cfgtable->ServerName[i]);
1218
1219                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1220
1221                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1222                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1223                                       & CFGTBL_ChangeReq))
1224                                         break;
1225                                 /* delay and try again */
1226                                 udelay(1000);
1227                         }
1228                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1229                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1230                                 return -EAGAIN;
1231                         return 0;
1232                 }
1233
1234         case CCISS_GETHEARTBEAT:
1235                 {
1236                         Heartbeat_type heartbeat;
1237
1238                         if (!arg)
1239                                 return -EINVAL;
1240                         heartbeat = readl(&host->cfgtable->HeartBeat);
1241                         if (copy_to_user
1242                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1243                                 return -EFAULT;
1244                         return 0;
1245                 }
1246         case CCISS_GETBUSTYPES:
1247                 {
1248                         BusTypes_type BusTypes;
1249
1250                         if (!arg)
1251                                 return -EINVAL;
1252                         BusTypes = readl(&host->cfgtable->BusTypes);
1253                         if (copy_to_user
1254                             (argp, &BusTypes, sizeof(BusTypes_type)))
1255                                 return -EFAULT;
1256                         return 0;
1257                 }
1258         case CCISS_GETFIRMVER:
1259                 {
1260                         FirmwareVer_type firmware;
1261
1262                         if (!arg)
1263                                 return -EINVAL;
1264                         memcpy(firmware, host->firm_ver, 4);
1265
1266                         if (copy_to_user
1267                             (argp, firmware, sizeof(FirmwareVer_type)))
1268                                 return -EFAULT;
1269                         return 0;
1270                 }
1271         case CCISS_GETDRIVVER:
1272                 {
1273                         DriverVer_type DriverVer = DRIVER_VERSION;
1274
1275                         if (!arg)
1276                                 return -EINVAL;
1277
1278                         if (copy_to_user
1279                             (argp, &DriverVer, sizeof(DriverVer_type)))
1280                                 return -EFAULT;
1281                         return 0;
1282                 }
1283
1284         case CCISS_DEREGDISK:
1285         case CCISS_REGNEWD:
1286         case CCISS_REVALIDVOLS:
1287                 return rebuild_lun_table(host, 0, 1);
1288
1289         case CCISS_GETLUNINFO:{
1290                         LogvolInfo_struct luninfo;
1291
1292                         memcpy(&luninfo.LunID, drv->LunID,
1293                                 sizeof(luninfo.LunID));
1294                         luninfo.num_opens = drv->usage_count;
1295                         luninfo.num_parts = 0;
1296                         if (copy_to_user(argp, &luninfo,
1297                                          sizeof(LogvolInfo_struct)))
1298                                 return -EFAULT;
1299                         return 0;
1300                 }
1301         case CCISS_PASSTHRU:
1302                 {
1303                         IOCTL_Command_struct iocommand;
1304                         CommandList_struct *c;
1305                         char *buff = NULL;
1306                         u64bit temp64;
1307                         unsigned long flags;
1308                         DECLARE_COMPLETION_ONSTACK(wait);
1309
1310                         if (!arg)
1311                                 return -EINVAL;
1312
1313                         if (!capable(CAP_SYS_RAWIO))
1314                                 return -EPERM;
1315
1316                         if (copy_from_user
1317                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1318                                 return -EFAULT;
1319                         if ((iocommand.buf_size < 1) &&
1320                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1321                                 return -EINVAL;
1322                         }
1323 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1324                         /* Check kmalloc limits */
1325                         if (iocommand.buf_size > 128000)
1326                                 return -EINVAL;
1327 #endif
1328                         if (iocommand.buf_size > 0) {
1329                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1330                                 if (buff == NULL)
1331                                         return -EFAULT;
1332                         }
1333                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1334                                 /* Copy the data into the buffer we created */
1335                                 if (copy_from_user
1336                                     (buff, iocommand.buf, iocommand.buf_size)) {
1337                                         kfree(buff);
1338                                         return -EFAULT;
1339                                 }
1340                         } else {
1341                                 memset(buff, 0, iocommand.buf_size);
1342                         }
1343                         if ((c = cmd_alloc(host, 0)) == NULL) {
1344                                 kfree(buff);
1345                                 return -ENOMEM;
1346                         }
1347                         // Fill in the command type
1348                         c->cmd_type = CMD_IOCTL_PEND;
1349                         // Fill in Command Header
1350                         c->Header.ReplyQueue = 0;       // unused in simple mode
1351                         if (iocommand.buf_size > 0)     // buffer to fill
1352                         {
1353                                 c->Header.SGList = 1;
1354                                 c->Header.SGTotal = 1;
1355                         } else  // no buffers to fill
1356                         {
1357                                 c->Header.SGList = 0;
1358                                 c->Header.SGTotal = 0;
1359                         }
1360                         c->Header.LUN = iocommand.LUN_info;
1361                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
1362
1363                         // Fill in Request block
1364                         c->Request = iocommand.Request;
1365
1366                         // Fill in the scatter gather information
1367                         if (iocommand.buf_size > 0) {
1368                                 temp64.val = pci_map_single(host->pdev, buff,
1369                                         iocommand.buf_size,
1370                                         PCI_DMA_BIDIRECTIONAL);
1371                                 c->SG[0].Addr.lower = temp64.val32.lower;
1372                                 c->SG[0].Addr.upper = temp64.val32.upper;
1373                                 c->SG[0].Len = iocommand.buf_size;
1374                                 c->SG[0].Ext = 0;       // we are not chaining
1375                         }
1376                         c->waiting = &wait;
1377
1378                         /* Put the request on the tail of the request queue */
1379                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1380                         addQ(&host->reqQ, c);
1381                         host->Qdepth++;
1382                         start_io(host);
1383                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1384
1385                         wait_for_completion(&wait);
1386
1387                         /* unlock the buffers from DMA */
1388                         temp64.val32.lower = c->SG[0].Addr.lower;
1389                         temp64.val32.upper = c->SG[0].Addr.upper;
1390                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1391                                          iocommand.buf_size,
1392                                          PCI_DMA_BIDIRECTIONAL);
1393
1394                         check_ioctl_unit_attention(host, c);
1395
1396                         /* Copy the error information out */
1397                         iocommand.error_info = *(c->err_info);
1398                         if (copy_to_user
1399                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1400                                 kfree(buff);
1401                                 cmd_free(host, c, 0);
1402                                 return -EFAULT;
1403                         }
1404
1405                         if (iocommand.Request.Type.Direction == XFER_READ) {
1406                                 /* Copy the data out of the buffer we created */
1407                                 if (copy_to_user
1408                                     (iocommand.buf, buff, iocommand.buf_size)) {
1409                                         kfree(buff);
1410                                         cmd_free(host, c, 0);
1411                                         return -EFAULT;
1412                                 }
1413                         }
1414                         kfree(buff);
1415                         cmd_free(host, c, 0);
1416                         return 0;
1417                 }
1418         case CCISS_BIG_PASSTHRU:{
1419                         BIG_IOCTL_Command_struct *ioc;
1420                         CommandList_struct *c;
1421                         unsigned char **buff = NULL;
1422                         int *buff_size = NULL;
1423                         u64bit temp64;
1424                         unsigned long flags;
1425                         BYTE sg_used = 0;
1426                         int status = 0;
1427                         int i;
1428                         DECLARE_COMPLETION_ONSTACK(wait);
1429                         __u32 left;
1430                         __u32 sz;
1431                         BYTE __user *data_ptr;
1432
1433                         if (!arg)
1434                                 return -EINVAL;
1435                         if (!capable(CAP_SYS_RAWIO))
1436                                 return -EPERM;
1437                         ioc = (BIG_IOCTL_Command_struct *)
1438                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1439                         if (!ioc) {
1440                                 status = -ENOMEM;
1441                                 goto cleanup1;
1442                         }
1443                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1444                                 status = -EFAULT;
1445                                 goto cleanup1;
1446                         }
1447                         if ((ioc->buf_size < 1) &&
1448                             (ioc->Request.Type.Direction != XFER_NONE)) {
1449                                 status = -EINVAL;
1450                                 goto cleanup1;
1451                         }
1452                         /* Check kmalloc limits  using all SGs */
1453                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1454                                 status = -EINVAL;
1455                                 goto cleanup1;
1456                         }
1457                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1458                                 status = -EINVAL;
1459                                 goto cleanup1;
1460                         }
1461                         buff =
1462                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1463                         if (!buff) {
1464                                 status = -ENOMEM;
1465                                 goto cleanup1;
1466                         }
1467                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1468                                                    GFP_KERNEL);
1469                         if (!buff_size) {
1470                                 status = -ENOMEM;
1471                                 goto cleanup1;
1472                         }
1473                         left = ioc->buf_size;
1474                         data_ptr = ioc->buf;
1475                         while (left) {
1476                                 sz = (left >
1477                                       ioc->malloc_size) ? ioc->
1478                                     malloc_size : left;
1479                                 buff_size[sg_used] = sz;
1480                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1481                                 if (buff[sg_used] == NULL) {
1482                                         status = -ENOMEM;
1483                                         goto cleanup1;
1484                                 }
1485                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1486                                         if (copy_from_user
1487                                             (buff[sg_used], data_ptr, sz)) {
1488                                                 status = -EFAULT;
1489                                                 goto cleanup1;
1490                                         }
1491                                 } else {
1492                                         memset(buff[sg_used], 0, sz);
1493                                 }
1494                                 left -= sz;
1495                                 data_ptr += sz;
1496                                 sg_used++;
1497                         }
1498                         if ((c = cmd_alloc(host, 0)) == NULL) {
1499                                 status = -ENOMEM;
1500                                 goto cleanup1;
1501                         }
1502                         c->cmd_type = CMD_IOCTL_PEND;
1503                         c->Header.ReplyQueue = 0;
1504
1505                         if (ioc->buf_size > 0) {
1506                                 c->Header.SGList = sg_used;
1507                                 c->Header.SGTotal = sg_used;
1508                         } else {
1509                                 c->Header.SGList = 0;
1510                                 c->Header.SGTotal = 0;
1511                         }
1512                         c->Header.LUN = ioc->LUN_info;
1513                         c->Header.Tag.lower = c->busaddr;
1514
1515                         c->Request = ioc->Request;
1516                         if (ioc->buf_size > 0) {
1517                                 int i;
1518                                 for (i = 0; i < sg_used; i++) {
1519                                         temp64.val =
1520                                             pci_map_single(host->pdev, buff[i],
1521                                                     buff_size[i],
1522                                                     PCI_DMA_BIDIRECTIONAL);
1523                                         c->SG[i].Addr.lower =
1524                                             temp64.val32.lower;
1525                                         c->SG[i].Addr.upper =
1526                                             temp64.val32.upper;
1527                                         c->SG[i].Len = buff_size[i];
1528                                         c->SG[i].Ext = 0;       /* we are not chaining */
1529                                 }
1530                         }
1531                         c->waiting = &wait;
1532                         /* Put the request on the tail of the request queue */
1533                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1534                         addQ(&host->reqQ, c);
1535                         host->Qdepth++;
1536                         start_io(host);
1537                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1538                         wait_for_completion(&wait);
1539                         /* unlock the buffers from DMA */
1540                         for (i = 0; i < sg_used; i++) {
1541                                 temp64.val32.lower = c->SG[i].Addr.lower;
1542                                 temp64.val32.upper = c->SG[i].Addr.upper;
1543                                 pci_unmap_single(host->pdev,
1544                                         (dma_addr_t) temp64.val, buff_size[i],
1545                                         PCI_DMA_BIDIRECTIONAL);
1546                         }
1547                         check_ioctl_unit_attention(host, c);
1548                         /* Copy the error information out */
1549                         ioc->error_info = *(c->err_info);
1550                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1551                                 cmd_free(host, c, 0);
1552                                 status = -EFAULT;
1553                                 goto cleanup1;
1554                         }
1555                         if (ioc->Request.Type.Direction == XFER_READ) {
1556                                 /* Copy the data out of the buffer we created */
1557                                 BYTE __user *ptr = ioc->buf;
1558                                 for (i = 0; i < sg_used; i++) {
1559                                         if (copy_to_user
1560                                             (ptr, buff[i], buff_size[i])) {
1561                                                 cmd_free(host, c, 0);
1562                                                 status = -EFAULT;
1563                                                 goto cleanup1;
1564                                         }
1565                                         ptr += buff_size[i];
1566                                 }
1567                         }
1568                         cmd_free(host, c, 0);
1569                         status = 0;
1570                       cleanup1:
1571                         if (buff) {
1572                                 for (i = 0; i < sg_used; i++)
1573                                         kfree(buff[i]);
1574                                 kfree(buff);
1575                         }
1576                         kfree(buff_size);
1577                         kfree(ioc);
1578                         return status;
1579                 }
1580
1581         /* scsi_cmd_ioctl handles these, below, though some are not */
1582         /* very meaningful for cciss.  SG_IO is the main one people want. */
1583
1584         case SG_GET_VERSION_NUM:
1585         case SG_SET_TIMEOUT:
1586         case SG_GET_TIMEOUT:
1587         case SG_GET_RESERVED_SIZE:
1588         case SG_SET_RESERVED_SIZE:
1589         case SG_EMULATED_HOST:
1590         case SG_IO:
1591         case SCSI_IOCTL_SEND_COMMAND:
1592                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1593
1594         /* scsi_cmd_ioctl would normally handle these, below, but */
1595         /* they aren't a good fit for cciss, as CD-ROMs are */
1596         /* not supported, and we don't have any bus/target/lun */
1597         /* which we present to the kernel. */
1598
1599         case CDROM_SEND_PACKET:
1600         case CDROMCLOSETRAY:
1601         case CDROMEJECT:
1602         case SCSI_IOCTL_GET_IDLUN:
1603         case SCSI_IOCTL_GET_BUS_NUMBER:
1604         default:
1605                 return -ENOTTY;
1606         }
1607 }
1608
1609 static void cciss_check_queues(ctlr_info_t *h)
1610 {
1611         int start_queue = h->next_to_run;
1612         int i;
1613
1614         /* check to see if we have maxed out the number of commands that can
1615          * be placed on the queue.  If so then exit.  We do this check here
1616          * in case the interrupt we serviced was from an ioctl and did not
1617          * free any new commands.
1618          */
1619         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1620                 return;
1621
1622         /* We have room on the queue for more commands.  Now we need to queue
1623          * them up.  We will also keep track of the next queue to run so
1624          * that every queue gets a chance to be started first.
1625          */
1626         for (i = 0; i < h->highest_lun + 1; i++) {
1627                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1628                 /* make sure the disk has been added and the drive is real
1629                  * because this can be called from the middle of init_one.
1630                  */
1631                 if (!h->drv[curr_queue])
1632                         continue;
1633                 if (!(h->drv[curr_queue]->queue) ||
1634                         !(h->drv[curr_queue]->heads))
1635                         continue;
1636                 blk_start_queue(h->gendisk[curr_queue]->queue);
1637
1638                 /* check to see if we have maxed out the number of commands
1639                  * that can be placed on the queue.
1640                  */
1641                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1642                         if (curr_queue == start_queue) {
1643                                 h->next_to_run =
1644                                     (start_queue + 1) % (h->highest_lun + 1);
1645                                 break;
1646                         } else {
1647                                 h->next_to_run = curr_queue;
1648                                 break;
1649                         }
1650                 }
1651         }
1652 }
1653
1654 static void cciss_softirq_done(struct request *rq)
1655 {
1656         CommandList_struct *cmd = rq->completion_data;
1657         ctlr_info_t *h = hba[cmd->ctlr];
1658         SGDescriptor_struct *curr_sg = cmd->SG;
1659         unsigned long flags;
1660         u64bit temp64;
1661         int i, ddir;
1662         int sg_index = 0;
1663
1664         if (cmd->Request.Type.Direction == XFER_READ)
1665                 ddir = PCI_DMA_FROMDEVICE;
1666         else
1667                 ddir = PCI_DMA_TODEVICE;
1668
1669         /* command did not need to be retried */
1670         /* unmap the DMA mapping for all the scatter gather elements */
1671         for (i = 0; i < cmd->Header.SGList; i++) {
1672                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1673                         temp64.val32.lower = cmd->SG[i].Addr.lower;
1674                         temp64.val32.upper = cmd->SG[i].Addr.upper;
1675                         pci_dma_sync_single_for_cpu(h->pdev, temp64.val,
1676                                                 cmd->SG[i].Len, ddir);
1677                         pci_unmap_single(h->pdev, temp64.val,
1678                                                 cmd->SG[i].Len, ddir);
1679                         /* Point to the next block */
1680                         curr_sg = h->cmd_sg_list[cmd->cmdindex]->sgchain;
1681                         sg_index = 0;
1682                 }
1683                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1684                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1685                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1686                                 ddir);
1687                 ++sg_index;
1688         }
1689
1690 #ifdef CCISS_DEBUG
1691         printk("Done with %p\n", rq);
1692 #endif                          /* CCISS_DEBUG */
1693
1694         /* set the residual count for pc requests */
1695         if (blk_pc_request(rq))
1696                 rq->resid_len = cmd->err_info->ResidualCnt;
1697
1698         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1699
1700         spin_lock_irqsave(&h->lock, flags);
1701         cmd_free(h, cmd, 1);
1702         cciss_check_queues(h);
1703         spin_unlock_irqrestore(&h->lock, flags);
1704 }
1705
1706 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1707         unsigned char scsi3addr[], uint32_t log_unit)
1708 {
1709         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1710                 sizeof(h->drv[log_unit]->LunID));
1711 }
1712
1713 /* This function gets the SCSI vendor, model, and revision of a logical drive
1714  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1715  * they cannot be read.
1716  */
1717 static void cciss_get_device_descr(int ctlr, int logvol,
1718                                    char *vendor, char *model, char *rev)
1719 {
1720         int rc;
1721         InquiryData_struct *inq_buf;
1722         unsigned char scsi3addr[8];
1723
1724         *vendor = '\0';
1725         *model = '\0';
1726         *rev = '\0';
1727
1728         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1729         if (!inq_buf)
1730                 return;
1731
1732         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1733         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1734                         scsi3addr, TYPE_CMD);
1735         if (rc == IO_OK) {
1736                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1737                 vendor[VENDOR_LEN] = '\0';
1738                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1739                 model[MODEL_LEN] = '\0';
1740                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1741                 rev[REV_LEN] = '\0';
1742         }
1743
1744         kfree(inq_buf);
1745         return;
1746 }
1747
1748 /* This function gets the serial number of a logical drive via
1749  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1750  * number cannot be had, for whatever reason, 16 bytes of 0xff
1751  * are returned instead.
1752  */
1753 static void cciss_get_serial_no(int ctlr, int logvol,
1754                                 unsigned char *serial_no, int buflen)
1755 {
1756 #define PAGE_83_INQ_BYTES 64
1757         int rc;
1758         unsigned char *buf;
1759         unsigned char scsi3addr[8];
1760
1761         if (buflen > 16)
1762                 buflen = 16;
1763         memset(serial_no, 0xff, buflen);
1764         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1765         if (!buf)
1766                 return;
1767         memset(serial_no, 0, buflen);
1768         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1769         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1770                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1771         if (rc == IO_OK)
1772                 memcpy(serial_no, &buf[8], buflen);
1773         kfree(buf);
1774         return;
1775 }
1776
1777 /*
1778  * cciss_add_disk sets up the block device queue for a logical drive
1779  */
1780 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1781                                 int drv_index)
1782 {
1783         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1784         if (!disk->queue)
1785                 goto init_queue_failure;
1786         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1787         disk->major = h->major;
1788         disk->first_minor = drv_index << NWD_SHIFT;
1789         disk->fops = &cciss_fops;
1790         if (cciss_create_ld_sysfs_entry(h, drv_index))
1791                 goto cleanup_queue;
1792         disk->private_data = h->drv[drv_index];
1793         disk->driverfs_dev = &h->drv[drv_index]->dev;
1794
1795         /* Set up queue information */
1796         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1797
1798         /* This is a hardware imposed limit. */
1799         blk_queue_max_hw_segments(disk->queue, h->maxsgentries);
1800
1801         /* This is a limit in the driver and could be eliminated. */
1802         blk_queue_max_phys_segments(disk->queue, h->maxsgentries);
1803
1804         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1805
1806         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1807
1808         disk->queue->queuedata = h;
1809
1810         blk_queue_logical_block_size(disk->queue,
1811                                      h->drv[drv_index]->block_size);
1812
1813         /* Make sure all queue data is written out before */
1814         /* setting h->drv[drv_index]->queue, as setting this */
1815         /* allows the interrupt handler to start the queue */
1816         wmb();
1817         h->drv[drv_index]->queue = disk->queue;
1818         add_disk(disk);
1819         return 0;
1820
1821 cleanup_queue:
1822         blk_cleanup_queue(disk->queue);
1823         disk->queue = NULL;
1824 init_queue_failure:
1825         return -1;
1826 }
1827
1828 /* This function will check the usage_count of the drive to be updated/added.
1829  * If the usage_count is zero and it is a heretofore unknown drive, or,
1830  * the drive's capacity, geometry, or serial number has changed,
1831  * then the drive information will be updated and the disk will be
1832  * re-registered with the kernel.  If these conditions don't hold,
1833  * then it will be left alone for the next reboot.  The exception to this
1834  * is disk 0 which will always be left registered with the kernel since it
1835  * is also the controller node.  Any changes to disk 0 will show up on
1836  * the next reboot.
1837  */
1838 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1839         int via_ioctl)
1840 {
1841         ctlr_info_t *h = hba[ctlr];
1842         struct gendisk *disk;
1843         InquiryData_struct *inq_buff = NULL;
1844         unsigned int block_size;
1845         sector_t total_size;
1846         unsigned long flags = 0;
1847         int ret = 0;
1848         drive_info_struct *drvinfo;
1849
1850         /* Get information about the disk and modify the driver structure */
1851         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1852         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1853         if (inq_buff == NULL || drvinfo == NULL)
1854                 goto mem_msg;
1855
1856         /* testing to see if 16-byte CDBs are already being used */
1857         if (h->cciss_read == CCISS_READ_16) {
1858                 cciss_read_capacity_16(h->ctlr, drv_index,
1859                         &total_size, &block_size);
1860
1861         } else {
1862                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1863                 /* if read_capacity returns all F's this volume is >2TB */
1864                 /* in size so we switch to 16-byte CDB's for all */
1865                 /* read/write ops */
1866                 if (total_size == 0xFFFFFFFFULL) {
1867                         cciss_read_capacity_16(ctlr, drv_index,
1868                         &total_size, &block_size);
1869                         h->cciss_read = CCISS_READ_16;
1870                         h->cciss_write = CCISS_WRITE_16;
1871                 } else {
1872                         h->cciss_read = CCISS_READ_10;
1873                         h->cciss_write = CCISS_WRITE_10;
1874                 }
1875         }
1876
1877         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1878                                inq_buff, drvinfo);
1879         drvinfo->block_size = block_size;
1880         drvinfo->nr_blocks = total_size + 1;
1881
1882         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1883                                 drvinfo->model, drvinfo->rev);
1884         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1885                         sizeof(drvinfo->serial_no));
1886         /* Save the lunid in case we deregister the disk, below. */
1887         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1888                 sizeof(drvinfo->LunID));
1889
1890         /* Is it the same disk we already know, and nothing's changed? */
1891         if (h->drv[drv_index]->raid_level != -1 &&
1892                 ((memcmp(drvinfo->serial_no,
1893                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1894                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1895                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1896                 drvinfo->heads == h->drv[drv_index]->heads &&
1897                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1898                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1899                         /* The disk is unchanged, nothing to update */
1900                         goto freeret;
1901
1902         /* If we get here it's not the same disk, or something's changed,
1903          * so we need to * deregister it, and re-register it, if it's not
1904          * in use.
1905          * If the disk already exists then deregister it before proceeding
1906          * (unless it's the first disk (for the controller node).
1907          */
1908         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1909                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1910                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1911                 h->drv[drv_index]->busy_configuring = 1;
1912                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1913
1914                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1915                  * which keeps the interrupt handler from starting
1916                  * the queue.
1917                  */
1918                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1919         }
1920
1921         /* If the disk is in use return */
1922         if (ret)
1923                 goto freeret;
1924
1925         /* Save the new information from cciss_geometry_inquiry
1926          * and serial number inquiry.  If the disk was deregistered
1927          * above, then h->drv[drv_index] will be NULL.
1928          */
1929         if (h->drv[drv_index] == NULL) {
1930                 drvinfo->device_initialized = 0;
1931                 h->drv[drv_index] = drvinfo;
1932                 drvinfo = NULL; /* so it won't be freed below. */
1933         } else {
1934                 /* special case for cxd0 */
1935                 h->drv[drv_index]->block_size = drvinfo->block_size;
1936                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
1937                 h->drv[drv_index]->heads = drvinfo->heads;
1938                 h->drv[drv_index]->sectors = drvinfo->sectors;
1939                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
1940                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
1941                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
1942                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
1943                         VENDOR_LEN + 1);
1944                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
1945                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
1946         }
1947
1948         ++h->num_luns;
1949         disk = h->gendisk[drv_index];
1950         set_capacity(disk, h->drv[drv_index]->nr_blocks);
1951
1952         /* If it's not disk 0 (drv_index != 0)
1953          * or if it was disk 0, but there was previously
1954          * no actual corresponding configured logical drive
1955          * (raid_leve == -1) then we want to update the
1956          * logical drive's information.
1957          */
1958         if (drv_index || first_time) {
1959                 if (cciss_add_disk(h, disk, drv_index) != 0) {
1960                         cciss_free_gendisk(h, drv_index);
1961                         cciss_free_drive_info(h, drv_index);
1962                         printk(KERN_WARNING "cciss:%d could not update "
1963                                 "disk %d\n", h->ctlr, drv_index);
1964                         --h->num_luns;
1965                 }
1966         }
1967
1968 freeret:
1969         kfree(inq_buff);
1970         kfree(drvinfo);
1971         return;
1972 mem_msg:
1973         printk(KERN_ERR "cciss: out of memory\n");
1974         goto freeret;
1975 }
1976
1977 /* This function will find the first index of the controllers drive array
1978  * that has a null drv pointer and allocate the drive info struct and
1979  * will return that index   This is where new drives will be added.
1980  * If the index to be returned is greater than the highest_lun index for
1981  * the controller then highest_lun is set * to this new index.
1982  * If there are no available indexes or if tha allocation fails, then -1
1983  * is returned.  * "controller_node" is used to know if this is a real
1984  * logical drive, or just the controller node, which determines if this
1985  * counts towards highest_lun.
1986  */
1987 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
1988 {
1989         int i;
1990         drive_info_struct *drv;
1991
1992         /* Search for an empty slot for our drive info */
1993         for (i = 0; i < CISS_MAX_LUN; i++) {
1994
1995                 /* if not cxd0 case, and it's occupied, skip it. */
1996                 if (h->drv[i] && i != 0)
1997                         continue;
1998                 /*
1999                  * If it's cxd0 case, and drv is alloc'ed already, and a
2000                  * disk is configured there, skip it.
2001                  */
2002                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2003                         continue;
2004
2005                 /*
2006                  * We've found an empty slot.  Update highest_lun
2007                  * provided this isn't just the fake cxd0 controller node.
2008                  */
2009                 if (i > h->highest_lun && !controller_node)
2010                         h->highest_lun = i;
2011
2012                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2013                 if (i == 0 && h->drv[i] != NULL)
2014                         return i;
2015
2016                 /*
2017                  * Found an empty slot, not already alloc'ed.  Allocate it.
2018                  * Mark it with raid_level == -1, so we know it's new later on.
2019                  */
2020                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2021                 if (!drv)
2022                         return -1;
2023                 drv->raid_level = -1; /* so we know it's new */
2024                 h->drv[i] = drv;
2025                 return i;
2026         }
2027         return -1;
2028 }
2029
2030 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2031 {
2032         kfree(h->drv[drv_index]);
2033         h->drv[drv_index] = NULL;
2034 }
2035
2036 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2037 {
2038         put_disk(h->gendisk[drv_index]);
2039         h->gendisk[drv_index] = NULL;
2040 }
2041
2042 /* cciss_add_gendisk finds a free hba[]->drv structure
2043  * and allocates a gendisk if needed, and sets the lunid
2044  * in the drvinfo structure.   It returns the index into
2045  * the ->drv[] array, or -1 if none are free.
2046  * is_controller_node indicates whether highest_lun should
2047  * count this disk, or if it's only being added to provide
2048  * a means to talk to the controller in case no logical
2049  * drives have yet been configured.
2050  */
2051 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2052         int controller_node)
2053 {
2054         int drv_index;
2055
2056         drv_index = cciss_alloc_drive_info(h, controller_node);
2057         if (drv_index == -1)
2058                 return -1;
2059
2060         /*Check if the gendisk needs to be allocated */
2061         if (!h->gendisk[drv_index]) {
2062                 h->gendisk[drv_index] =
2063                         alloc_disk(1 << NWD_SHIFT);
2064                 if (!h->gendisk[drv_index]) {
2065                         printk(KERN_ERR "cciss%d: could not "
2066                                 "allocate a new disk %d\n",
2067                                 h->ctlr, drv_index);
2068                         goto err_free_drive_info;
2069                 }
2070         }
2071         memcpy(h->drv[drv_index]->LunID, lunid,
2072                 sizeof(h->drv[drv_index]->LunID));
2073         if (cciss_create_ld_sysfs_entry(h, drv_index))
2074                 goto err_free_disk;
2075         /* Don't need to mark this busy because nobody */
2076         /* else knows about this disk yet to contend */
2077         /* for access to it. */
2078         h->drv[drv_index]->busy_configuring = 0;
2079         wmb();
2080         return drv_index;
2081
2082 err_free_disk:
2083         cciss_free_gendisk(h, drv_index);
2084 err_free_drive_info:
2085         cciss_free_drive_info(h, drv_index);
2086         return -1;
2087 }
2088
2089 /* This is for the special case of a controller which
2090  * has no logical drives.  In this case, we still need
2091  * to register a disk so the controller can be accessed
2092  * by the Array Config Utility.
2093  */
2094 static void cciss_add_controller_node(ctlr_info_t *h)
2095 {
2096         struct gendisk *disk;
2097         int drv_index;
2098
2099         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2100                 return;
2101
2102         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2103         if (drv_index == -1)
2104                 goto error;
2105         h->drv[drv_index]->block_size = 512;
2106         h->drv[drv_index]->nr_blocks = 0;
2107         h->drv[drv_index]->heads = 0;
2108         h->drv[drv_index]->sectors = 0;
2109         h->drv[drv_index]->cylinders = 0;
2110         h->drv[drv_index]->raid_level = -1;
2111         memset(h->drv[drv_index]->serial_no, 0, 16);
2112         disk = h->gendisk[drv_index];
2113         if (cciss_add_disk(h, disk, drv_index) == 0)
2114                 return;
2115         cciss_free_gendisk(h, drv_index);
2116         cciss_free_drive_info(h, drv_index);
2117 error:
2118         printk(KERN_WARNING "cciss%d: could not "
2119                 "add disk 0.\n", h->ctlr);
2120         return;
2121 }
2122
2123 /* This function will add and remove logical drives from the Logical
2124  * drive array of the controller and maintain persistency of ordering
2125  * so that mount points are preserved until the next reboot.  This allows
2126  * for the removal of logical drives in the middle of the drive array
2127  * without a re-ordering of those drives.
2128  * INPUT
2129  * h            = The controller to perform the operations on
2130  */
2131 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2132         int via_ioctl)
2133 {
2134         int ctlr = h->ctlr;
2135         int num_luns;
2136         ReportLunData_struct *ld_buff = NULL;
2137         int return_code;
2138         int listlength = 0;
2139         int i;
2140         int drv_found;
2141         int drv_index = 0;
2142         unsigned char lunid[8] = CTLR_LUNID;
2143         unsigned long flags;
2144
2145         if (!capable(CAP_SYS_RAWIO))
2146                 return -EPERM;
2147
2148         /* Set busy_configuring flag for this operation */
2149         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2150         if (h->busy_configuring) {
2151                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2152                 return -EBUSY;
2153         }
2154         h->busy_configuring = 1;
2155         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2156
2157         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2158         if (ld_buff == NULL)
2159                 goto mem_msg;
2160
2161         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2162                                       sizeof(ReportLunData_struct),
2163                                       0, CTLR_LUNID, TYPE_CMD);
2164
2165         if (return_code == IO_OK)
2166                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2167         else {  /* reading number of logical volumes failed */
2168                 printk(KERN_WARNING "cciss: report logical volume"
2169                        " command failed\n");
2170                 listlength = 0;
2171                 goto freeret;
2172         }
2173
2174         num_luns = listlength / 8;      /* 8 bytes per entry */
2175         if (num_luns > CISS_MAX_LUN) {
2176                 num_luns = CISS_MAX_LUN;
2177                 printk(KERN_WARNING "cciss: more luns configured"
2178                        " on controller than can be handled by"
2179                        " this driver.\n");
2180         }
2181
2182         if (num_luns == 0)
2183                 cciss_add_controller_node(h);
2184
2185         /* Compare controller drive array to driver's drive array
2186          * to see if any drives are missing on the controller due
2187          * to action of Array Config Utility (user deletes drive)
2188          * and deregister logical drives which have disappeared.
2189          */
2190         for (i = 0; i <= h->highest_lun; i++) {
2191                 int j;
2192                 drv_found = 0;
2193
2194                 /* skip holes in the array from already deleted drives */
2195                 if (h->drv[i] == NULL)
2196                         continue;
2197
2198                 for (j = 0; j < num_luns; j++) {
2199                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2200                         if (memcmp(h->drv[i]->LunID, lunid,
2201                                 sizeof(lunid)) == 0) {
2202                                 drv_found = 1;
2203                                 break;
2204                         }
2205                 }
2206                 if (!drv_found) {
2207                         /* Deregister it from the OS, it's gone. */
2208                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2209                         h->drv[i]->busy_configuring = 1;
2210                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2211                         return_code = deregister_disk(h, i, 1, via_ioctl);
2212                         if (h->drv[i] != NULL)
2213                                 h->drv[i]->busy_configuring = 0;
2214                 }
2215         }
2216
2217         /* Compare controller drive array to driver's drive array.
2218          * Check for updates in the drive information and any new drives
2219          * on the controller due to ACU adding logical drives, or changing
2220          * a logical drive's size, etc.  Reregister any new/changed drives
2221          */
2222         for (i = 0; i < num_luns; i++) {
2223                 int j;
2224
2225                 drv_found = 0;
2226
2227                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2228                 /* Find if the LUN is already in the drive array
2229                  * of the driver.  If so then update its info
2230                  * if not in use.  If it does not exist then find
2231                  * the first free index and add it.
2232                  */
2233                 for (j = 0; j <= h->highest_lun; j++) {
2234                         if (h->drv[j] != NULL &&
2235                                 memcmp(h->drv[j]->LunID, lunid,
2236                                         sizeof(h->drv[j]->LunID)) == 0) {
2237                                 drv_index = j;
2238                                 drv_found = 1;
2239                                 break;
2240                         }
2241                 }
2242
2243                 /* check if the drive was found already in the array */
2244                 if (!drv_found) {
2245                         drv_index = cciss_add_gendisk(h, lunid, 0);
2246                         if (drv_index == -1)
2247                                 goto freeret;
2248                 }
2249                 cciss_update_drive_info(ctlr, drv_index, first_time,
2250                         via_ioctl);
2251         }               /* end for */
2252
2253 freeret:
2254         kfree(ld_buff);
2255         h->busy_configuring = 0;
2256         /* We return -1 here to tell the ACU that we have registered/updated
2257          * all of the drives that we can and to keep it from calling us
2258          * additional times.
2259          */
2260         return -1;
2261 mem_msg:
2262         printk(KERN_ERR "cciss: out of memory\n");
2263         h->busy_configuring = 0;
2264         goto freeret;
2265 }
2266
2267 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2268 {
2269         /* zero out the disk size info */
2270         drive_info->nr_blocks = 0;
2271         drive_info->block_size = 0;
2272         drive_info->heads = 0;
2273         drive_info->sectors = 0;
2274         drive_info->cylinders = 0;
2275         drive_info->raid_level = -1;
2276         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2277         memset(drive_info->model, 0, sizeof(drive_info->model));
2278         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2279         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2280         /*
2281          * don't clear the LUNID though, we need to remember which
2282          * one this one is.
2283          */
2284 }
2285
2286 /* This function will deregister the disk and it's queue from the
2287  * kernel.  It must be called with the controller lock held and the
2288  * drv structures busy_configuring flag set.  It's parameters are:
2289  *
2290  * disk = This is the disk to be deregistered
2291  * drv  = This is the drive_info_struct associated with the disk to be
2292  *        deregistered.  It contains information about the disk used
2293  *        by the driver.
2294  * clear_all = This flag determines whether or not the disk information
2295  *             is going to be completely cleared out and the highest_lun
2296  *             reset.  Sometimes we want to clear out information about
2297  *             the disk in preparation for re-adding it.  In this case
2298  *             the highest_lun should be left unchanged and the LunID
2299  *             should not be cleared.
2300  * via_ioctl
2301  *    This indicates whether we've reached this path via ioctl.
2302  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2303  *    If this path is reached via ioctl(), then the max_usage_count will
2304  *    be 1, as the process calling ioctl() has got to have the device open.
2305  *    If we get here via sysfs, then the max usage count will be zero.
2306 */
2307 static int deregister_disk(ctlr_info_t *h, int drv_index,
2308                            int clear_all, int via_ioctl)
2309 {
2310         int i;
2311         struct gendisk *disk;
2312         drive_info_struct *drv;
2313         int recalculate_highest_lun;
2314
2315         if (!capable(CAP_SYS_RAWIO))
2316                 return -EPERM;
2317
2318         drv = h->drv[drv_index];
2319         disk = h->gendisk[drv_index];
2320
2321         /* make sure logical volume is NOT is use */
2322         if (clear_all || (h->gendisk[0] == disk)) {
2323                 if (drv->usage_count > via_ioctl)
2324                         return -EBUSY;
2325         } else if (drv->usage_count > 0)
2326                 return -EBUSY;
2327
2328         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2329
2330         /* invalidate the devices and deregister the disk.  If it is disk
2331          * zero do not deregister it but just zero out it's values.  This
2332          * allows us to delete disk zero but keep the controller registered.
2333          */
2334         if (h->gendisk[0] != disk) {
2335                 struct request_queue *q = disk->queue;
2336                 if (disk->flags & GENHD_FL_UP) {
2337                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2338                         del_gendisk(disk);
2339                 }
2340                 if (q)
2341                         blk_cleanup_queue(q);
2342                 /* If clear_all is set then we are deleting the logical
2343                  * drive, not just refreshing its info.  For drives
2344                  * other than disk 0 we will call put_disk.  We do not
2345                  * do this for disk 0 as we need it to be able to
2346                  * configure the controller.
2347                  */
2348                 if (clear_all){
2349                         /* This isn't pretty, but we need to find the
2350                          * disk in our array and NULL our the pointer.
2351                          * This is so that we will call alloc_disk if
2352                          * this index is used again later.
2353                          */
2354                         for (i=0; i < CISS_MAX_LUN; i++){
2355                                 if (h->gendisk[i] == disk) {
2356                                         h->gendisk[i] = NULL;
2357                                         break;
2358                                 }
2359                         }
2360                         put_disk(disk);
2361                 }
2362         } else {
2363                 set_capacity(disk, 0);
2364                 cciss_clear_drive_info(drv);
2365         }
2366
2367         --h->num_luns;
2368
2369         /* if it was the last disk, find the new hightest lun */
2370         if (clear_all && recalculate_highest_lun) {
2371                 int i, newhighest = -1;
2372                 for (i = 0; i <= h->highest_lun; i++) {
2373                         /* if the disk has size > 0, it is available */
2374                         if (h->drv[i] && h->drv[i]->heads)
2375                                 newhighest = i;
2376                 }
2377                 h->highest_lun = newhighest;
2378         }
2379         return 0;
2380 }
2381
2382 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2383                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2384                 int cmd_type)
2385 {
2386         ctlr_info_t *h = hba[ctlr];
2387         u64bit buff_dma_handle;
2388         int status = IO_OK;
2389
2390         c->cmd_type = CMD_IOCTL_PEND;
2391         c->Header.ReplyQueue = 0;
2392         if (buff != NULL) {
2393                 c->Header.SGList = 1;
2394                 c->Header.SGTotal = 1;
2395         } else {
2396                 c->Header.SGList = 0;
2397                 c->Header.SGTotal = 0;
2398         }
2399         c->Header.Tag.lower = c->busaddr;
2400         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2401
2402         c->Request.Type.Type = cmd_type;
2403         if (cmd_type == TYPE_CMD) {
2404                 switch (cmd) {
2405                 case CISS_INQUIRY:
2406                         /* are we trying to read a vital product page */
2407                         if (page_code != 0) {
2408                                 c->Request.CDB[1] = 0x01;
2409                                 c->Request.CDB[2] = page_code;
2410                         }
2411                         c->Request.CDBLen = 6;
2412                         c->Request.Type.Attribute = ATTR_SIMPLE;
2413                         c->Request.Type.Direction = XFER_READ;
2414                         c->Request.Timeout = 0;
2415                         c->Request.CDB[0] = CISS_INQUIRY;
2416                         c->Request.CDB[4] = size & 0xFF;
2417                         break;
2418                 case CISS_REPORT_LOG:
2419                 case CISS_REPORT_PHYS:
2420                         /* Talking to controller so It's a physical command
2421                            mode = 00 target = 0.  Nothing to write.
2422                          */
2423                         c->Request.CDBLen = 12;
2424                         c->Request.Type.Attribute = ATTR_SIMPLE;
2425                         c->Request.Type.Direction = XFER_READ;
2426                         c->Request.Timeout = 0;
2427                         c->Request.CDB[0] = cmd;
2428                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
2429                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2430                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2431                         c->Request.CDB[9] = size & 0xFF;
2432                         break;
2433
2434                 case CCISS_READ_CAPACITY:
2435                         c->Request.CDBLen = 10;
2436                         c->Request.Type.Attribute = ATTR_SIMPLE;
2437                         c->Request.Type.Direction = XFER_READ;
2438                         c->Request.Timeout = 0;
2439                         c->Request.CDB[0] = cmd;
2440                         break;
2441                 case CCISS_READ_CAPACITY_16:
2442                         c->Request.CDBLen = 16;
2443                         c->Request.Type.Attribute = ATTR_SIMPLE;
2444                         c->Request.Type.Direction = XFER_READ;
2445                         c->Request.Timeout = 0;
2446                         c->Request.CDB[0] = cmd;
2447                         c->Request.CDB[1] = 0x10;
2448                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2449                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2450                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2451                         c->Request.CDB[13] = size & 0xFF;
2452                         c->Request.Timeout = 0;
2453                         c->Request.CDB[0] = cmd;
2454                         break;
2455                 case CCISS_CACHE_FLUSH:
2456                         c->Request.CDBLen = 12;
2457                         c->Request.Type.Attribute = ATTR_SIMPLE;
2458                         c->Request.Type.Direction = XFER_WRITE;
2459                         c->Request.Timeout = 0;
2460                         c->Request.CDB[0] = BMIC_WRITE;
2461                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2462                         break;
2463                 case TEST_UNIT_READY:
2464                         c->Request.CDBLen = 6;
2465                         c->Request.Type.Attribute = ATTR_SIMPLE;
2466                         c->Request.Type.Direction = XFER_NONE;
2467                         c->Request.Timeout = 0;
2468                         break;
2469                 default:
2470                         printk(KERN_WARNING
2471                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2472                         return IO_ERROR;
2473                 }
2474         } else if (cmd_type == TYPE_MSG) {
2475                 switch (cmd) {
2476                 case 0: /* ABORT message */
2477                         c->Request.CDBLen = 12;
2478                         c->Request.Type.Attribute = ATTR_SIMPLE;
2479                         c->Request.Type.Direction = XFER_WRITE;
2480                         c->Request.Timeout = 0;
2481                         c->Request.CDB[0] = cmd;        /* abort */
2482                         c->Request.CDB[1] = 0;  /* abort a command */
2483                         /* buff contains the tag of the command to abort */
2484                         memcpy(&c->Request.CDB[4], buff, 8);
2485                         break;
2486                 case 1: /* RESET message */
2487                         c->Request.CDBLen = 16;
2488                         c->Request.Type.Attribute = ATTR_SIMPLE;
2489                         c->Request.Type.Direction = XFER_NONE;
2490                         c->Request.Timeout = 0;
2491                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2492                         c->Request.CDB[0] = cmd;        /* reset */
2493                         c->Request.CDB[1] = 0x03;       /* reset a target */
2494                         break;
2495                 case 3: /* No-Op message */
2496                         c->Request.CDBLen = 1;
2497                         c->Request.Type.Attribute = ATTR_SIMPLE;
2498                         c->Request.Type.Direction = XFER_WRITE;
2499                         c->Request.Timeout = 0;
2500                         c->Request.CDB[0] = cmd;
2501                         break;
2502                 default:
2503                         printk(KERN_WARNING
2504                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2505                         return IO_ERROR;
2506                 }
2507         } else {
2508                 printk(KERN_WARNING
2509                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2510                 return IO_ERROR;
2511         }
2512         /* Fill in the scatter gather information */
2513         if (size > 0) {
2514                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2515                                                              buff, size,
2516                                                              PCI_DMA_BIDIRECTIONAL);
2517                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2518                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2519                 c->SG[0].Len = size;
2520                 c->SG[0].Ext = 0;       /* we are not chaining */
2521         }
2522         return status;
2523 }
2524
2525 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2526 {
2527         switch (c->err_info->ScsiStatus) {
2528         case SAM_STAT_GOOD:
2529                 return IO_OK;
2530         case SAM_STAT_CHECK_CONDITION:
2531                 switch (0xf & c->err_info->SenseInfo[2]) {
2532                 case 0: return IO_OK; /* no sense */
2533                 case 1: return IO_OK; /* recovered error */
2534                 default:
2535                         if (check_for_unit_attention(h, c))
2536                                 return IO_NEEDS_RETRY;
2537                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2538                                 "check condition, sense key = 0x%02x\n",
2539                                 h->ctlr, c->Request.CDB[0],
2540                                 c->err_info->SenseInfo[2]);
2541                 }
2542                 break;
2543         default:
2544                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2545                         "scsi status = 0x%02x\n", h->ctlr,
2546                         c->Request.CDB[0], c->err_info->ScsiStatus);
2547                 break;
2548         }
2549         return IO_ERROR;
2550 }
2551
2552 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2553 {
2554         int return_status = IO_OK;
2555
2556         if (c->err_info->CommandStatus == CMD_SUCCESS)
2557                 return IO_OK;
2558
2559         switch (c->err_info->CommandStatus) {
2560         case CMD_TARGET_STATUS:
2561                 return_status = check_target_status(h, c);
2562                 break;
2563         case CMD_DATA_UNDERRUN:
2564         case CMD_DATA_OVERRUN:
2565                 /* expected for inquiry and report lun commands */
2566                 break;
2567         case CMD_INVALID:
2568                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2569                        "reported invalid\n", c->Request.CDB[0]);
2570                 return_status = IO_ERROR;
2571                 break;
2572         case CMD_PROTOCOL_ERR:
2573                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2574                        "protocol error \n", c->Request.CDB[0]);
2575                 return_status = IO_ERROR;
2576                 break;
2577         case CMD_HARDWARE_ERR:
2578                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2579                        " hardware error\n", c->Request.CDB[0]);
2580                 return_status = IO_ERROR;
2581                 break;
2582         case CMD_CONNECTION_LOST:
2583                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2584                        "connection lost\n", c->Request.CDB[0]);
2585                 return_status = IO_ERROR;
2586                 break;
2587         case CMD_ABORTED:
2588                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2589                        "aborted\n", c->Request.CDB[0]);
2590                 return_status = IO_ERROR;
2591                 break;
2592         case CMD_ABORT_FAILED:
2593                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2594                        "abort failed\n", c->Request.CDB[0]);
2595                 return_status = IO_ERROR;
2596                 break;
2597         case CMD_UNSOLICITED_ABORT:
2598                 printk(KERN_WARNING
2599                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2600                         c->Request.CDB[0]);
2601                 return_status = IO_NEEDS_RETRY;
2602                 break;
2603         default:
2604                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2605                        "unknown status %x\n", c->Request.CDB[0],
2606                        c->err_info->CommandStatus);
2607                 return_status = IO_ERROR;
2608         }
2609         return return_status;
2610 }
2611
2612 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2613         int attempt_retry)
2614 {
2615         DECLARE_COMPLETION_ONSTACK(wait);
2616         u64bit buff_dma_handle;
2617         unsigned long flags;
2618         int return_status = IO_OK;
2619
2620 resend_cmd2:
2621         c->waiting = &wait;
2622         /* Put the request on the tail of the queue and send it */
2623         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2624         addQ(&h->reqQ, c);
2625         h->Qdepth++;
2626         start_io(h);
2627         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2628
2629         wait_for_completion(&wait);
2630
2631         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2632                 goto command_done;
2633
2634         return_status = process_sendcmd_error(h, c);
2635
2636         if (return_status == IO_NEEDS_RETRY &&
2637                 c->retry_count < MAX_CMD_RETRIES) {
2638                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2639                         c->Request.CDB[0]);
2640                 c->retry_count++;
2641                 /* erase the old error information */
2642                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2643                 return_status = IO_OK;
2644                 INIT_COMPLETION(wait);
2645                 goto resend_cmd2;
2646         }
2647
2648 command_done:
2649         /* unlock the buffers from DMA */
2650         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2651         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2652         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2653                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2654         return return_status;
2655 }
2656
2657 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2658                            __u8 page_code, unsigned char scsi3addr[],
2659                         int cmd_type)
2660 {
2661         ctlr_info_t *h = hba[ctlr];
2662         CommandList_struct *c;
2663         int return_status;
2664
2665         c = cmd_alloc(h, 0);
2666         if (!c)
2667                 return -ENOMEM;
2668         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2669                 scsi3addr, cmd_type);
2670         if (return_status == IO_OK)
2671                 return_status = sendcmd_withirq_core(h, c, 1);
2672
2673         cmd_free(h, c, 0);
2674         return return_status;
2675 }
2676
2677 static void cciss_geometry_inquiry(int ctlr, int logvol,
2678                                    sector_t total_size,
2679                                    unsigned int block_size,
2680                                    InquiryData_struct *inq_buff,
2681                                    drive_info_struct *drv)
2682 {
2683         int return_code;
2684         unsigned long t;
2685         unsigned char scsi3addr[8];
2686
2687         memset(inq_buff, 0, sizeof(InquiryData_struct));
2688         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2689         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2690                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2691         if (return_code == IO_OK) {
2692                 if (inq_buff->data_byte[8] == 0xFF) {
2693                         printk(KERN_WARNING
2694                                "cciss: reading geometry failed, volume "
2695                                "does not support reading geometry\n");
2696                         drv->heads = 255;
2697                         drv->sectors = 32;      // Sectors per track
2698                         drv->cylinders = total_size + 1;
2699                         drv->raid_level = RAID_UNKNOWN;
2700                 } else {
2701                         drv->heads = inq_buff->data_byte[6];
2702                         drv->sectors = inq_buff->data_byte[7];
2703                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2704                         drv->cylinders += inq_buff->data_byte[5];
2705                         drv->raid_level = inq_buff->data_byte[8];
2706                 }
2707                 drv->block_size = block_size;
2708                 drv->nr_blocks = total_size + 1;
2709                 t = drv->heads * drv->sectors;
2710                 if (t > 1) {
2711                         sector_t real_size = total_size + 1;
2712                         unsigned long rem = sector_div(real_size, t);
2713                         if (rem)
2714                                 real_size++;
2715                         drv->cylinders = real_size;
2716                 }
2717         } else {                /* Get geometry failed */
2718                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2719         }
2720 }
2721
2722 static void
2723 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2724                     unsigned int *block_size)
2725 {
2726         ReadCapdata_struct *buf;
2727         int return_code;
2728         unsigned char scsi3addr[8];
2729
2730         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2731         if (!buf) {
2732                 printk(KERN_WARNING "cciss: out of memory\n");
2733                 return;
2734         }
2735
2736         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2737         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2738                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2739         if (return_code == IO_OK) {
2740                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2741                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2742         } else {                /* read capacity command failed */
2743                 printk(KERN_WARNING "cciss: read capacity failed\n");
2744                 *total_size = 0;
2745                 *block_size = BLOCK_SIZE;
2746         }
2747         kfree(buf);
2748 }
2749
2750 static void cciss_read_capacity_16(int ctlr, int logvol,
2751         sector_t *total_size, unsigned int *block_size)
2752 {
2753         ReadCapdata_struct_16 *buf;
2754         int return_code;
2755         unsigned char scsi3addr[8];
2756
2757         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2758         if (!buf) {
2759                 printk(KERN_WARNING "cciss: out of memory\n");
2760                 return;
2761         }
2762
2763         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2764         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2765                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2766                         0, scsi3addr, TYPE_CMD);
2767         if (return_code == IO_OK) {
2768                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2769                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2770         } else {                /* read capacity command failed */
2771                 printk(KERN_WARNING "cciss: read capacity failed\n");
2772                 *total_size = 0;
2773                 *block_size = BLOCK_SIZE;
2774         }
2775         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2776                (unsigned long long)*total_size+1, *block_size);
2777         kfree(buf);
2778 }
2779
2780 static int cciss_revalidate(struct gendisk *disk)
2781 {
2782         ctlr_info_t *h = get_host(disk);
2783         drive_info_struct *drv = get_drv(disk);
2784         int logvol;
2785         int FOUND = 0;
2786         unsigned int block_size;
2787         sector_t total_size;
2788         InquiryData_struct *inq_buff = NULL;
2789
2790         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2791                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2792                         sizeof(drv->LunID)) == 0) {
2793                         FOUND = 1;
2794                         break;
2795                 }
2796         }
2797
2798         if (!FOUND)
2799                 return 1;
2800
2801         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2802         if (inq_buff == NULL) {
2803                 printk(KERN_WARNING "cciss: out of memory\n");
2804                 return 1;
2805         }
2806         if (h->cciss_read == CCISS_READ_10) {
2807                 cciss_read_capacity(h->ctlr, logvol,
2808                                         &total_size, &block_size);
2809         } else {
2810                 cciss_read_capacity_16(h->ctlr, logvol,
2811                                         &total_size, &block_size);
2812         }
2813         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2814                                inq_buff, drv);
2815
2816         blk_queue_logical_block_size(drv->queue, drv->block_size);
2817         set_capacity(disk, drv->nr_blocks);
2818
2819         kfree(inq_buff);
2820         return 0;
2821 }
2822
2823 /*
2824  * Map (physical) PCI mem into (virtual) kernel space
2825  */
2826 static void __iomem *remap_pci_mem(ulong base, ulong size)
2827 {
2828         ulong page_base = ((ulong) base) & PAGE_MASK;
2829         ulong page_offs = ((ulong) base) - page_base;
2830         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2831
2832         return page_remapped ? (page_remapped + page_offs) : NULL;
2833 }
2834
2835 /*
2836  * Takes jobs of the Q and sends them to the hardware, then puts it on
2837  * the Q to wait for completion.
2838  */
2839 static void start_io(ctlr_info_t *h)
2840 {
2841         CommandList_struct *c;
2842
2843         while (!hlist_empty(&h->reqQ)) {
2844                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2845                 /* can't do anything if fifo is full */
2846                 if ((h->access.fifo_full(h))) {
2847                         printk(KERN_WARNING "cciss: fifo full\n");
2848                         break;
2849                 }
2850
2851                 /* Get the first entry from the Request Q */
2852                 removeQ(c);
2853                 h->Qdepth--;
2854
2855                 /* Tell the controller execute command */
2856                 h->access.submit_command(h, c);
2857
2858                 /* Put job onto the completed Q */
2859                 addQ(&h->cmpQ, c);
2860         }
2861 }
2862
2863 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2864 /* Zeros out the error record and then resends the command back */
2865 /* to the controller */
2866 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2867 {
2868         /* erase the old error information */
2869         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2870
2871         /* add it to software queue and then send it to the controller */
2872         addQ(&h->reqQ, c);
2873         h->Qdepth++;
2874         if (h->Qdepth > h->maxQsinceinit)
2875                 h->maxQsinceinit = h->Qdepth;
2876
2877         start_io(h);
2878 }
2879
2880 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2881         unsigned int msg_byte, unsigned int host_byte,
2882         unsigned int driver_byte)
2883 {
2884         /* inverse of macros in scsi.h */
2885         return (scsi_status_byte & 0xff) |
2886                 ((msg_byte & 0xff) << 8) |
2887                 ((host_byte & 0xff) << 16) |
2888                 ((driver_byte & 0xff) << 24);
2889 }
2890
2891 static inline int evaluate_target_status(ctlr_info_t *h,
2892                         CommandList_struct *cmd, int *retry_cmd)
2893 {
2894         unsigned char sense_key;
2895         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2896         int error_value;
2897
2898         *retry_cmd = 0;
2899         /* If we get in here, it means we got "target status", that is, scsi status */
2900         status_byte = cmd->err_info->ScsiStatus;
2901         driver_byte = DRIVER_OK;
2902         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2903
2904         if (blk_pc_request(cmd->rq))
2905                 host_byte = DID_PASSTHROUGH;
2906         else
2907                 host_byte = DID_OK;
2908
2909         error_value = make_status_bytes(status_byte, msg_byte,
2910                 host_byte, driver_byte);
2911
2912         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2913                 if (!blk_pc_request(cmd->rq))
2914                         printk(KERN_WARNING "cciss: cmd %p "
2915                                "has SCSI Status 0x%x\n",
2916                                cmd, cmd->err_info->ScsiStatus);
2917                 return error_value;
2918         }
2919
2920         /* check the sense key */
2921         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2922         /* no status or recovered error */
2923         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2924                 error_value = 0;
2925
2926         if (check_for_unit_attention(h, cmd)) {
2927                 *retry_cmd = !blk_pc_request(cmd->rq);
2928                 return 0;
2929         }
2930
2931         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2932                 if (error_value != 0)
2933                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2934                                " sense key = 0x%x\n", cmd, sense_key);
2935                 return error_value;
2936         }
2937
2938         /* SG_IO or similar, copy sense data back */
2939         if (cmd->rq->sense) {
2940                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2941                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2942                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2943                         cmd->rq->sense_len);
2944         } else
2945                 cmd->rq->sense_len = 0;
2946
2947         return error_value;
2948 }
2949
2950 /* checks the status of the job and calls complete buffers to mark all
2951  * buffers for the completed job. Note that this function does not need
2952  * to hold the hba/queue lock.
2953  */
2954 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2955                                     int timeout)
2956 {
2957         int retry_cmd = 0;
2958         struct request *rq = cmd->rq;
2959
2960         rq->errors = 0;
2961
2962         if (timeout)
2963                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
2964
2965         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
2966                 goto after_error_processing;
2967
2968         switch (cmd->err_info->CommandStatus) {
2969         case CMD_TARGET_STATUS:
2970                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
2971                 break;
2972         case CMD_DATA_UNDERRUN:
2973                 if (blk_fs_request(cmd->rq)) {
2974                         printk(KERN_WARNING "cciss: cmd %p has"
2975                                " completed with data underrun "
2976                                "reported\n", cmd);
2977                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
2978                 }
2979                 break;
2980         case CMD_DATA_OVERRUN:
2981                 if (blk_fs_request(cmd->rq))
2982                         printk(KERN_WARNING "cciss: cmd %p has"
2983                                " completed with data overrun "
2984                                "reported\n", cmd);
2985                 break;
2986         case CMD_INVALID:
2987                 printk(KERN_WARNING "cciss: cmd %p is "
2988                        "reported invalid\n", cmd);
2989                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2990                         cmd->err_info->CommandStatus, DRIVER_OK,
2991                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2992                 break;
2993         case CMD_PROTOCOL_ERR:
2994                 printk(KERN_WARNING "cciss: cmd %p has "
2995                        "protocol error \n", cmd);
2996                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2997                         cmd->err_info->CommandStatus, DRIVER_OK,
2998                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2999                 break;
3000         case CMD_HARDWARE_ERR:
3001                 printk(KERN_WARNING "cciss: cmd %p had "
3002                        " hardware error\n", cmd);
3003                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3004                         cmd->err_info->CommandStatus, DRIVER_OK,
3005                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3006                 break;
3007         case CMD_CONNECTION_LOST:
3008                 printk(KERN_WARNING "cciss: cmd %p had "
3009                        "connection lost\n", cmd);
3010                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3011                         cmd->err_info->CommandStatus, DRIVER_OK,
3012                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3013                 break;
3014         case CMD_ABORTED:
3015                 printk(KERN_WARNING "cciss: cmd %p was "
3016                        "aborted\n", cmd);
3017                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3018                         cmd->err_info->CommandStatus, DRIVER_OK,
3019                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3020                 break;
3021         case CMD_ABORT_FAILED:
3022                 printk(KERN_WARNING "cciss: cmd %p reports "
3023                        "abort failed\n", cmd);
3024                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3025                         cmd->err_info->CommandStatus, DRIVER_OK,
3026                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3027                 break;
3028         case CMD_UNSOLICITED_ABORT:
3029                 printk(KERN_WARNING "cciss%d: unsolicited "
3030                        "abort %p\n", h->ctlr, cmd);
3031                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3032                         retry_cmd = 1;
3033                         printk(KERN_WARNING
3034                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3035                         cmd->retry_count++;
3036                 } else
3037                         printk(KERN_WARNING
3038                                "cciss%d: %p retried too "
3039                                "many times\n", h->ctlr, cmd);
3040                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3041                         cmd->err_info->CommandStatus, DRIVER_OK,
3042                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3043                 break;
3044         case CMD_TIMEOUT:
3045                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3046                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3047                         cmd->err_info->CommandStatus, DRIVER_OK,
3048                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3049                 break;
3050         default:
3051                 printk(KERN_WARNING "cciss: cmd %p returned "
3052                        "unknown status %x\n", cmd,
3053                        cmd->err_info->CommandStatus);
3054                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3055                         cmd->err_info->CommandStatus, DRIVER_OK,
3056                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3057         }
3058
3059 after_error_processing:
3060
3061         /* We need to return this command */
3062         if (retry_cmd) {
3063                 resend_cciss_cmd(h, cmd);
3064                 return;
3065         }
3066         cmd->rq->completion_data = cmd;
3067         blk_complete_request(cmd->rq);
3068 }
3069
3070 /*
3071  * Get a request and submit it to the controller.
3072  */
3073 static void do_cciss_request(struct request_queue *q)
3074 {
3075         ctlr_info_t *h = q->queuedata;
3076         CommandList_struct *c;
3077         sector_t start_blk;
3078         int seg;
3079         struct request *creq;
3080         u64bit temp64;
3081         struct scatterlist *tmp_sg;
3082         SGDescriptor_struct *curr_sg;
3083         drive_info_struct *drv;
3084         int i, dir;
3085         int nseg = 0;
3086         int sg_index = 0;
3087         int chained = 0;
3088
3089         /* We call start_io here in case there is a command waiting on the
3090          * queue that has not been sent.
3091          */
3092         if (blk_queue_plugged(q))
3093                 goto startio;
3094
3095       queue:
3096         creq = blk_peek_request(q);
3097         if (!creq)
3098                 goto startio;
3099
3100         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3101
3102         if ((c = cmd_alloc(h, 1)) == NULL)
3103                 goto full;
3104
3105         blk_start_request(creq);
3106
3107         tmp_sg = h->scatter_list[c->cmdindex];
3108         spin_unlock_irq(q->queue_lock);
3109
3110         c->cmd_type = CMD_RWREQ;
3111         c->rq = creq;
3112
3113         /* fill in the request */
3114         drv = creq->rq_disk->private_data;
3115         c->Header.ReplyQueue = 0;       // unused in simple mode
3116         /* got command from pool, so use the command block index instead */
3117         /* for direct lookups. */
3118         /* The first 2 bits are reserved for controller error reporting. */
3119         c->Header.Tag.lower = (c->cmdindex << 3);
3120         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
3121         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3122         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
3123         c->Request.Type.Type = TYPE_CMD;        // It is a command.
3124         c->Request.Type.Attribute = ATTR_SIMPLE;
3125         c->Request.Type.Direction =
3126             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3127         c->Request.Timeout = 0; // Don't time out
3128         c->Request.CDB[0] =
3129             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3130         start_blk = blk_rq_pos(creq);
3131 #ifdef CCISS_DEBUG
3132         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3133                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3134 #endif                          /* CCISS_DEBUG */
3135
3136         sg_init_table(tmp_sg, h->maxsgentries);
3137         seg = blk_rq_map_sg(q, creq, tmp_sg);
3138
3139         /* get the DMA records for the setup */
3140         if (c->Request.Type.Direction == XFER_READ)
3141                 dir = PCI_DMA_FROMDEVICE;
3142         else
3143                 dir = PCI_DMA_TODEVICE;
3144
3145         curr_sg = c->SG;
3146         sg_index = 0;
3147         chained = 0;
3148
3149         for (i = 0; i < seg; i++) {
3150                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3151                         !chained && ((seg - i) > 1)) {
3152                         nseg = seg - i;
3153                         curr_sg[sg_index].Len = (nseg) *
3154                                         sizeof(SGDescriptor_struct);
3155                         curr_sg[sg_index].Ext = CCISS_SG_CHAIN;
3156
3157                         /* Point to next chain block. */
3158                         curr_sg = h->cmd_sg_list[c->cmdindex]->sgchain;
3159                         sg_index = 0;
3160                         chained = 1;
3161                 }
3162                 curr_sg[sg_index].Len = tmp_sg[i].length;
3163                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3164                                                 tmp_sg[i].offset,
3165                                                 tmp_sg[i].length, dir);
3166                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3167                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3168                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3169
3170                 ++sg_index;
3171         }
3172
3173         if (chained) {
3174                 int len;
3175                 curr_sg = c->SG;
3176                 sg_index = h->max_cmd_sgentries - 1;
3177                 len = curr_sg[sg_index].Len;
3178                 /* Setup pointer to next chain block.
3179                  * Fill out last element in current chain
3180                  * block with address of next chain block.
3181                  */
3182                 temp64.val = pci_map_single(h->pdev,
3183                                         h->cmd_sg_list[c->cmdindex]->sgchain,
3184                                         len, dir);
3185
3186                 h->cmd_sg_list[c->cmdindex]->sg_chain_dma = temp64.val;
3187                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3188                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3189
3190                 pci_dma_sync_single_for_device(h->pdev,
3191                                 h->cmd_sg_list[c->cmdindex]->sg_chain_dma,
3192                                 len, dir);
3193         }
3194
3195         /* track how many SG entries we are using */
3196         if (seg > h->maxSG)
3197                 h->maxSG = seg;
3198
3199 #ifdef CCISS_DEBUG
3200         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3201                         "chained[%d]\n",
3202                         blk_rq_sectors(creq), seg, chained);
3203 #endif                          /* CCISS_DEBUG */
3204
3205         c->Header.SGList = c->Header.SGTotal = seg + chained;
3206         if (seg > h->max_cmd_sgentries)
3207                 c->Header.SGList = h->max_cmd_sgentries;
3208
3209         if (likely(blk_fs_request(creq))) {
3210                 if(h->cciss_read == CCISS_READ_10) {
3211                         c->Request.CDB[1] = 0;
3212                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
3213                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3214                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3215                         c->Request.CDB[5] = start_blk & 0xff;
3216                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
3217                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3218                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3219                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3220                 } else {
3221                         u32 upper32 = upper_32_bits(start_blk);
3222
3223                         c->Request.CDBLen = 16;
3224                         c->Request.CDB[1]= 0;
3225                         c->Request.CDB[2]= (upper32 >> 24) & 0xff;      //MSB
3226                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3227                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3228                         c->Request.CDB[5]= upper32 & 0xff;
3229                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3230                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3231                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3232                         c->Request.CDB[9]= start_blk & 0xff;
3233                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3234                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3235                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3236                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3237                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3238                 }
3239         } else if (blk_pc_request(creq)) {
3240                 c->Request.CDBLen = creq->cmd_len;
3241                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3242         } else {
3243                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3244                 BUG();
3245         }
3246
3247         spin_lock_irq(q->queue_lock);
3248
3249         addQ(&h->reqQ, c);
3250         h->Qdepth++;
3251         if (h->Qdepth > h->maxQsinceinit)
3252                 h->maxQsinceinit = h->Qdepth;
3253
3254         goto queue;
3255 full:
3256         blk_stop_queue(q);
3257 startio:
3258         /* We will already have the driver lock here so not need
3259          * to lock it.
3260          */
3261         start_io(h);
3262 }
3263
3264 static inline unsigned long get_next_completion(ctlr_info_t *h)
3265 {
3266         return h->access.command_completed(h);
3267 }
3268
3269 static inline int interrupt_pending(ctlr_info_t *h)
3270 {
3271         return h->access.intr_pending(h);
3272 }
3273
3274 static inline long interrupt_not_for_us(ctlr_info_t *h)
3275 {
3276         return (((h->access.intr_pending(h) == 0) ||
3277                  (h->interrupts_enabled == 0)));
3278 }
3279
3280 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3281 {
3282         ctlr_info_t *h = dev_id;
3283         CommandList_struct *c;
3284         unsigned long flags;
3285         __u32 a, a1, a2;
3286
3287         if (interrupt_not_for_us(h))
3288                 return IRQ_NONE;
3289         /*
3290          * If there are completed commands in the completion queue,
3291          * we had better do something about it.
3292          */
3293         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3294         while (interrupt_pending(h)) {
3295                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3296                         a1 = a;
3297                         if ((a & 0x04)) {
3298                                 a2 = (a >> 3);
3299                                 if (a2 >= h->nr_cmds) {
3300                                         printk(KERN_WARNING
3301                                                "cciss: controller cciss%d failed, stopping.\n",
3302                                                h->ctlr);
3303                                         fail_all_cmds(h->ctlr);
3304                                         return IRQ_HANDLED;
3305                                 }
3306
3307                                 c = h->cmd_pool + a2;
3308                                 a = c->busaddr;
3309
3310                         } else {
3311                                 struct hlist_node *tmp;
3312
3313                                 a &= ~3;
3314                                 c = NULL;
3315                                 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3316                                         if (c->busaddr == a)
3317                                                 break;
3318                                 }
3319                         }
3320                         /*
3321                          * If we've found the command, take it off the
3322                          * completion Q and free it
3323                          */
3324                         if (c && c->busaddr == a) {
3325                                 removeQ(c);
3326                                 if (c->cmd_type == CMD_RWREQ) {
3327                                         complete_command(h, c, 0);
3328                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3329                                         complete(c->waiting);
3330                                 }
3331 #                               ifdef CONFIG_CISS_SCSI_TAPE
3332                                 else if (c->cmd_type == CMD_SCSI)
3333                                         complete_scsi_command(c, 0, a1);
3334 #                               endif
3335                                 continue;
3336                         }
3337                 }
3338         }
3339
3340         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3341         return IRQ_HANDLED;
3342 }
3343
3344 /**
3345  * add_to_scan_list() - add controller to rescan queue
3346  * @h:                Pointer to the controller.
3347  *
3348  * Adds the controller to the rescan queue if not already on the queue.
3349  *
3350  * returns 1 if added to the queue, 0 if skipped (could be on the
3351  * queue already, or the controller could be initializing or shutting
3352  * down).
3353  **/
3354 static int add_to_scan_list(struct ctlr_info *h)
3355 {
3356         struct ctlr_info *test_h;
3357         int found = 0;
3358         int ret = 0;
3359
3360         if (h->busy_initializing)
3361                 return 0;
3362
3363         if (!mutex_trylock(&h->busy_shutting_down))
3364                 return 0;
3365
3366         mutex_lock(&scan_mutex);
3367         list_for_each_entry(test_h, &scan_q, scan_list) {
3368                 if (test_h == h) {
3369                         found = 1;
3370                         break;
3371                 }
3372         }
3373         if (!found && !h->busy_scanning) {
3374                 INIT_COMPLETION(h->scan_wait);
3375                 list_add_tail(&h->scan_list, &scan_q);
3376                 ret = 1;
3377         }
3378         mutex_unlock(&scan_mutex);
3379         mutex_unlock(&h->busy_shutting_down);
3380
3381         return ret;
3382 }
3383
3384 /**
3385  * remove_from_scan_list() - remove controller from rescan queue
3386  * @h:                     Pointer to the controller.
3387  *
3388  * Removes the controller from the rescan queue if present. Blocks if
3389  * the controller is currently conducting a rescan.  The controller
3390  * can be in one of three states:
3391  * 1. Doesn't need a scan
3392  * 2. On the scan list, but not scanning yet (we remove it)
3393  * 3. Busy scanning (and not on the list). In this case we want to wait for
3394  *    the scan to complete to make sure the scanning thread for this
3395  *    controller is completely idle.
3396  **/
3397 static void remove_from_scan_list(struct ctlr_info *h)
3398 {
3399         struct ctlr_info *test_h, *tmp_h;
3400
3401         mutex_lock(&scan_mutex);
3402         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3403                 if (test_h == h) { /* state 2. */
3404                         list_del(&h->scan_list);
3405                         complete_all(&h->scan_wait);
3406                         mutex_unlock(&scan_mutex);
3407                         return;
3408                 }
3409         }
3410         if (h->busy_scanning) { /* state 3. */
3411                 mutex_unlock(&scan_mutex);
3412                 wait_for_completion(&h->scan_wait);
3413         } else { /* state 1, nothing to do. */
3414                 mutex_unlock(&scan_mutex);
3415         }
3416 }
3417
3418 /**
3419  * scan_thread() - kernel thread used to rescan controllers
3420  * @data:        Ignored.
3421  *
3422  * A kernel thread used scan for drive topology changes on
3423  * controllers. The thread processes only one controller at a time
3424  * using a queue.  Controllers are added to the queue using
3425  * add_to_scan_list() and removed from the queue either after done
3426  * processing or using remove_from_scan_list().
3427  *
3428  * returns 0.
3429  **/
3430 static int scan_thread(void *data)
3431 {
3432         struct ctlr_info *h;
3433
3434         while (1) {
3435                 set_current_state(TASK_INTERRUPTIBLE);
3436                 schedule();
3437                 if (kthread_should_stop())
3438                         break;
3439
3440                 while (1) {
3441                         mutex_lock(&scan_mutex);
3442                         if (list_empty(&scan_q)) {
3443                                 mutex_unlock(&scan_mutex);
3444                                 break;
3445                         }
3446
3447                         h = list_entry(scan_q.next,
3448                                        struct ctlr_info,
3449                                        scan_list);
3450                         list_del(&h->scan_list);
3451                         h->busy_scanning = 1;
3452                         mutex_unlock(&scan_mutex);
3453
3454                         rebuild_lun_table(h, 0, 0);
3455                         complete_all(&h->scan_wait);
3456                         mutex_lock(&scan_mutex);
3457                         h->busy_scanning = 0;
3458                         mutex_unlock(&scan_mutex);
3459                 }
3460         }
3461
3462         return 0;
3463 }
3464
3465 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3466 {
3467         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3468                 return 0;
3469
3470         switch (c->err_info->SenseInfo[12]) {
3471         case STATE_CHANGED:
3472                 printk(KERN_WARNING "cciss%d: a state change "
3473                         "detected, command retried\n", h->ctlr);
3474                 return 1;
3475         break;
3476         case LUN_FAILED:
3477                 printk(KERN_WARNING "cciss%d: LUN failure "
3478                         "detected, action required\n", h->ctlr);
3479                 return 1;
3480         break;
3481         case REPORT_LUNS_CHANGED:
3482                 printk(KERN_WARNING "cciss%d: report LUN data "
3483                         "changed\n", h->ctlr);
3484         /*
3485          * Here, we could call add_to_scan_list and wake up the scan thread,
3486          * except that it's quite likely that we will get more than one
3487          * REPORT_LUNS_CHANGED condition in quick succession, which means
3488          * that those which occur after the first one will likely happen
3489          * *during* the scan_thread's rescan.  And the rescan code is not
3490          * robust enough to restart in the middle, undoing what it has already
3491          * done, and it's not clear that it's even possible to do this, since
3492          * part of what it does is notify the block layer, which starts
3493          * doing it's own i/o to read partition tables and so on, and the
3494          * driver doesn't have visibility to know what might need undoing.
3495          * In any event, if possible, it is horribly complicated to get right
3496          * so we just don't do it for now.
3497          *
3498          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3499          */
3500                 return 1;
3501         break;
3502         case POWER_OR_RESET:
3503                 printk(KERN_WARNING "cciss%d: a power on "
3504                         "or device reset detected\n", h->ctlr);
3505                 return 1;
3506         break;
3507         case UNIT_ATTENTION_CLEARED:
3508                 printk(KERN_WARNING "cciss%d: unit attention "
3509                     "cleared by another initiator\n", h->ctlr);
3510                 return 1;
3511         break;
3512         default:
3513                 printk(KERN_WARNING "cciss%d: unknown "
3514                         "unit attention detected\n", h->ctlr);
3515                                 return 1;
3516         }
3517 }
3518
3519 /*
3520  *  We cannot read the structure directly, for portability we must use
3521  *   the io functions.
3522  *   This is for debug only.
3523  */
3524 #ifdef CCISS_DEBUG
3525 static void print_cfg_table(CfgTable_struct *tb)
3526 {
3527         int i;
3528         char temp_name[17];
3529
3530         printk("Controller Configuration information\n");
3531         printk("------------------------------------\n");
3532         for (i = 0; i < 4; i++)
3533                 temp_name[i] = readb(&(tb->Signature[i]));
3534         temp_name[4] = '\0';
3535         printk("   Signature = %s\n", temp_name);
3536         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3537         printk("   Transport methods supported = 0x%x\n",
3538                readl(&(tb->TransportSupport)));
3539         printk("   Transport methods active = 0x%x\n",
3540                readl(&(tb->TransportActive)));
3541         printk("   Requested transport Method = 0x%x\n",
3542                readl(&(tb->HostWrite.TransportRequest)));
3543         printk("   Coalesce Interrupt Delay = 0x%x\n",
3544                readl(&(tb->HostWrite.CoalIntDelay)));
3545         printk("   Coalesce Interrupt Count = 0x%x\n",
3546                readl(&(tb->HostWrite.CoalIntCount)));
3547         printk("   Max outstanding commands = 0x%d\n",
3548                readl(&(tb->CmdsOutMax)));
3549         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3550         for (i = 0; i < 16; i++)
3551                 temp_name[i] = readb(&(tb->ServerName[i]));
3552         temp_name[16] = '\0';
3553         printk("   Server Name = %s\n", temp_name);
3554         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3555 }
3556 #endif                          /* CCISS_DEBUG */
3557
3558 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3559 {
3560         int i, offset, mem_type, bar_type;
3561         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3562                 return 0;
3563         offset = 0;
3564         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3565                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3566                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3567                         offset += 4;
3568                 else {
3569                         mem_type = pci_resource_flags(pdev, i) &
3570                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3571                         switch (mem_type) {
3572                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3573                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3574                                 offset += 4;    /* 32 bit */
3575                                 break;
3576                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3577                                 offset += 8;
3578                                 break;
3579                         default:        /* reserved in PCI 2.2 */
3580                                 printk(KERN_WARNING
3581                                        "Base address is invalid\n");
3582                                 return -1;
3583                                 break;
3584                         }
3585                 }
3586                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3587                         return i + 1;
3588         }
3589         return -1;
3590 }
3591
3592 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3593  * controllers that are capable. If not, we use IO-APIC mode.
3594  */
3595
3596 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3597                                            struct pci_dev *pdev, __u32 board_id)
3598 {
3599 #ifdef CONFIG_PCI_MSI
3600         int err;
3601         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3602         {0, 2}, {0, 3}
3603         };
3604
3605         /* Some boards advertise MSI but don't really support it */
3606         if ((board_id == 0x40700E11) ||
3607             (board_id == 0x40800E11) ||
3608             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3609                 goto default_int_mode;
3610
3611         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3612                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3613                 if (!err) {
3614                         c->intr[0] = cciss_msix_entries[0].vector;
3615                         c->intr[1] = cciss_msix_entries[1].vector;
3616                         c->intr[2] = cciss_msix_entries[2].vector;
3617                         c->intr[3] = cciss_msix_entries[3].vector;
3618                         c->msix_vector = 1;
3619                         return;
3620                 }
3621                 if (err > 0) {
3622                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3623                                "available\n", err);
3624                         goto default_int_mode;
3625                 } else {
3626                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3627                                err);
3628                         goto default_int_mode;
3629                 }
3630         }
3631         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3632                 if (!pci_enable_msi(pdev)) {
3633                         c->msi_vector = 1;
3634                 } else {
3635                         printk(KERN_WARNING "cciss: MSI init failed\n");
3636                 }
3637         }
3638 default_int_mode:
3639 #endif                          /* CONFIG_PCI_MSI */
3640         /* if we get here we're going to use the default interrupt mode */
3641         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3642         return;
3643 }
3644
3645 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3646 {
3647         ushort subsystem_vendor_id, subsystem_device_id, command;
3648         __u32 board_id, scratchpad = 0;
3649         __u64 cfg_offset;
3650         __u32 cfg_base_addr;
3651         __u64 cfg_base_addr_index;
3652         int i, prod_index, err;
3653
3654         subsystem_vendor_id = pdev->subsystem_vendor;
3655         subsystem_device_id = pdev->subsystem_device;
3656         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3657                     subsystem_vendor_id);
3658
3659         for (i = 0; i < ARRAY_SIZE(products); i++) {
3660                 /* Stand aside for hpsa driver on request */
3661                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3662                         return -ENODEV;
3663                 if (board_id == products[i].board_id)
3664                         break;
3665         }
3666         prod_index = i;
3667         if (prod_index == ARRAY_SIZE(products)) {
3668                 dev_warn(&pdev->dev,
3669                         "unrecognized board ID: 0x%08lx, ignoring.\n",
3670                         (unsigned long) board_id);
3671                 return -ENODEV;
3672         }
3673
3674         /* check to see if controller has been disabled */
3675         /* BEFORE trying to enable it */
3676         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3677         if (!(command & 0x02)) {
3678                 printk(KERN_WARNING
3679                        "cciss: controller appears to be disabled\n");
3680                 return -ENODEV;
3681         }
3682
3683         err = pci_enable_device(pdev);
3684         if (err) {
3685                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3686                 return err;
3687         }
3688
3689         err = pci_request_regions(pdev, "cciss");
3690         if (err) {
3691                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3692                        "aborting\n");
3693                 return err;
3694         }
3695
3696 #ifdef CCISS_DEBUG
3697         printk("command = %x\n", command);
3698         printk("irq = %x\n", pdev->irq);
3699         printk("board_id = %x\n", board_id);
3700 #endif                          /* CCISS_DEBUG */
3701
3702 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3703  * else we use the IO-APIC interrupt assigned to us by system ROM.
3704  */
3705         cciss_interrupt_mode(c, pdev, board_id);
3706
3707         /* find the memory BAR */
3708         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3709                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3710                         break;
3711         }
3712         if (i == DEVICE_COUNT_RESOURCE) {
3713                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3714                 err = -ENODEV;
3715                 goto err_out_free_res;
3716         }
3717
3718         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3719                                                  * already removed
3720                                                  */
3721
3722 #ifdef CCISS_DEBUG
3723         printk("address 0 = %lx\n", c->paddr);
3724 #endif                          /* CCISS_DEBUG */
3725         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3726
3727         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3728          * We poll for up to 120 secs, once per 100ms. */
3729         for (i = 0; i < 1200; i++) {
3730                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3731                 if (scratchpad == CCISS_FIRMWARE_READY)
3732                         break;
3733                 set_current_state(TASK_INTERRUPTIBLE);
3734                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
3735         }
3736         if (scratchpad != CCISS_FIRMWARE_READY) {
3737                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3738                 err = -ENODEV;
3739                 goto err_out_free_res;
3740         }
3741
3742         /* get the address index number */
3743         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3744         cfg_base_addr &= (__u32) 0x0000ffff;
3745 #ifdef CCISS_DEBUG
3746         printk("cfg base address = %x\n", cfg_base_addr);
3747 #endif                          /* CCISS_DEBUG */
3748         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3749 #ifdef CCISS_DEBUG
3750         printk("cfg base address index = %llx\n",
3751                 (unsigned long long)cfg_base_addr_index);
3752 #endif                          /* CCISS_DEBUG */
3753         if (cfg_base_addr_index == -1) {
3754                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3755                 err = -ENODEV;
3756                 goto err_out_free_res;
3757         }
3758
3759         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3760 #ifdef CCISS_DEBUG
3761         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3762 #endif                          /* CCISS_DEBUG */
3763         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3764                                                        cfg_base_addr_index) +
3765                                     cfg_offset, sizeof(CfgTable_struct));
3766         c->board_id = board_id;
3767
3768 #ifdef CCISS_DEBUG
3769         print_cfg_table(c->cfgtable);
3770 #endif                          /* CCISS_DEBUG */
3771
3772         /* Some controllers support Zero Memory Raid (ZMR).
3773          * When configured in ZMR mode the number of supported
3774          * commands drops to 64. So instead of just setting an
3775          * arbitrary value we make the driver a little smarter.
3776          * We read the config table to tell us how many commands
3777          * are supported on the controller then subtract 4 to
3778          * leave a little room for ioctl calls.
3779          */
3780         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3781         c->maxsgentries = readl(&(c->cfgtable->MaxSGElements));
3782
3783         /*
3784          * Limit native command to 32 s/g elements to save dma'able memory.
3785          * Howvever spec says if 0, use 31
3786          */
3787
3788         c->max_cmd_sgentries = 31;
3789         if (c->maxsgentries > 512) {
3790                 c->max_cmd_sgentries = 32;
3791                 c->chainsize = c->maxsgentries - c->max_cmd_sgentries + 1;
3792                 c->maxsgentries -= 1;   /* account for chain pointer */
3793         } else {
3794                 c->maxsgentries = 31;   /* Default to traditional value */
3795                 c->chainsize = 0;       /* traditional */
3796         }
3797
3798         c->product_name = products[prod_index].product_name;
3799         c->access = *(products[prod_index].access);
3800         c->nr_cmds = c->max_commands - 4;
3801         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3802             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3803             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3804             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3805                 printk("Does not appear to be a valid CISS config table\n");
3806                 err = -ENODEV;
3807                 goto err_out_free_res;
3808         }
3809 #ifdef CONFIG_X86
3810         {
3811                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3812                 __u32 prefetch;
3813                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3814                 prefetch |= 0x100;
3815                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3816         }
3817 #endif
3818
3819         /* Disabling DMA prefetch and refetch for the P600.
3820          * An ASIC bug may result in accesses to invalid memory addresses.
3821          * We've disabled prefetch for some time now. Testing with XEN
3822          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3823          */
3824         if(board_id == 0x3225103C) {
3825                 __u32 dma_prefetch;
3826                 __u32 dma_refetch;
3827                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3828                 dma_prefetch |= 0x8000;
3829                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3830                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3831                 dma_refetch |= 0x1;
3832                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3833         }
3834
3835 #ifdef CCISS_DEBUG
3836         printk("Trying to put board into Simple mode\n");
3837 #endif                          /* CCISS_DEBUG */
3838         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3839         /* Update the field, and then ring the doorbell */
3840         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3841         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3842
3843         /* under certain very rare conditions, this can take awhile.
3844          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3845          * as we enter this code.) */
3846         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3847                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3848                         break;
3849                 /* delay and try again */
3850                 set_current_state(TASK_INTERRUPTIBLE);
3851                 schedule_timeout(msecs_to_jiffies(1));
3852         }
3853
3854 #ifdef CCISS_DEBUG
3855         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3856                readl(c->vaddr + SA5_DOORBELL));
3857 #endif                          /* CCISS_DEBUG */
3858 #ifdef CCISS_DEBUG
3859         print_cfg_table(c->cfgtable);
3860 #endif                          /* CCISS_DEBUG */
3861
3862         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3863                 printk(KERN_WARNING "cciss: unable to get board into"
3864                        " simple mode\n");
3865                 err = -ENODEV;
3866                 goto err_out_free_res;
3867         }
3868         return 0;
3869
3870 err_out_free_res:
3871         /*
3872          * Deliberately omit pci_disable_device(): it does something nasty to
3873          * Smart Array controllers that pci_enable_device does not undo
3874          */
3875         pci_release_regions(pdev);
3876         return err;
3877 }
3878
3879 /* Function to find the first free pointer into our hba[] array
3880  * Returns -1 if no free entries are left.
3881  */
3882 static int alloc_cciss_hba(void)
3883 {
3884         int i;
3885
3886         for (i = 0; i < MAX_CTLR; i++) {
3887                 if (!hba[i]) {
3888                         ctlr_info_t *p;
3889
3890                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3891                         if (!p)
3892                                 goto Enomem;
3893                         hba[i] = p;
3894                         return i;
3895                 }
3896         }
3897         printk(KERN_WARNING "cciss: This driver supports a maximum"
3898                " of %d controllers.\n", MAX_CTLR);
3899         return -1;
3900 Enomem:
3901         printk(KERN_ERR "cciss: out of memory.\n");
3902         return -1;
3903 }
3904
3905 static void free_hba(int n)
3906 {
3907         ctlr_info_t *h = hba[n];
3908         int i;
3909
3910         hba[n] = NULL;
3911         for (i = 0; i < h->highest_lun + 1; i++)
3912                 if (h->gendisk[i] != NULL)
3913                         put_disk(h->gendisk[i]);
3914         kfree(h);
3915 }
3916
3917 /* Send a message CDB to the firmware. */
3918 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3919 {
3920         typedef struct {
3921                 CommandListHeader_struct CommandHeader;
3922                 RequestBlock_struct Request;
3923                 ErrDescriptor_struct ErrorDescriptor;
3924         } Command;
3925         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3926         Command *cmd;
3927         dma_addr_t paddr64;
3928         uint32_t paddr32, tag;
3929         void __iomem *vaddr;
3930         int i, err;
3931
3932         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3933         if (vaddr == NULL)
3934                 return -ENOMEM;
3935
3936         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3937            CCISS commands, so they must be allocated from the lower 4GiB of
3938            memory. */
3939         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3940         if (err) {
3941                 iounmap(vaddr);
3942                 return -ENOMEM;
3943         }
3944
3945         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3946         if (cmd == NULL) {
3947                 iounmap(vaddr);
3948                 return -ENOMEM;
3949         }
3950
3951         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
3952            although there's no guarantee, we assume that the address is at
3953            least 4-byte aligned (most likely, it's page-aligned). */
3954         paddr32 = paddr64;
3955
3956         cmd->CommandHeader.ReplyQueue = 0;
3957         cmd->CommandHeader.SGList = 0;
3958         cmd->CommandHeader.SGTotal = 0;
3959         cmd->CommandHeader.Tag.lower = paddr32;
3960         cmd->CommandHeader.Tag.upper = 0;
3961         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
3962
3963         cmd->Request.CDBLen = 16;
3964         cmd->Request.Type.Type = TYPE_MSG;
3965         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
3966         cmd->Request.Type.Direction = XFER_NONE;
3967         cmd->Request.Timeout = 0; /* Don't time out */
3968         cmd->Request.CDB[0] = opcode;
3969         cmd->Request.CDB[1] = type;
3970         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
3971
3972         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
3973         cmd->ErrorDescriptor.Addr.upper = 0;
3974         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
3975
3976         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
3977
3978         for (i = 0; i < 10; i++) {
3979                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
3980                 if ((tag & ~3) == paddr32)
3981                         break;
3982                 schedule_timeout_uninterruptible(HZ);
3983         }
3984
3985         iounmap(vaddr);
3986
3987         /* we leak the DMA buffer here ... no choice since the controller could
3988            still complete the command. */
3989         if (i == 10) {
3990                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
3991                         opcode, type);
3992                 return -ETIMEDOUT;
3993         }
3994
3995         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
3996
3997         if (tag & 2) {
3998                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
3999                         opcode, type);
4000                 return -EIO;
4001         }
4002
4003         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4004                 opcode, type);
4005         return 0;
4006 }
4007
4008 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4009 #define cciss_noop(p) cciss_message(p, 3, 0)
4010
4011 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4012 {
4013 /* the #defines are stolen from drivers/pci/msi.h. */
4014 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4015 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4016
4017         int pos;
4018         u16 control = 0;
4019
4020         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4021         if (pos) {
4022                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4023                 if (control & PCI_MSI_FLAGS_ENABLE) {
4024                         printk(KERN_INFO "cciss: resetting MSI\n");
4025                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4026                 }
4027         }
4028
4029         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4030         if (pos) {
4031                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4032                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4033                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4034                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4035                 }
4036         }
4037
4038         return 0;
4039 }
4040
4041 /* This does a hard reset of the controller using PCI power management
4042  * states. */
4043 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4044 {
4045         u16 pmcsr, saved_config_space[32];
4046         int i, pos;
4047
4048         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4049
4050         /* This is very nearly the same thing as
4051
4052            pci_save_state(pci_dev);
4053            pci_set_power_state(pci_dev, PCI_D3hot);
4054            pci_set_power_state(pci_dev, PCI_D0);
4055            pci_restore_state(pci_dev);
4056
4057            but we can't use these nice canned kernel routines on
4058            kexec, because they also check the MSI/MSI-X state in PCI
4059            configuration space and do the wrong thing when it is
4060            set/cleared.  Also, the pci_save/restore_state functions
4061            violate the ordering requirements for restoring the
4062            configuration space from the CCISS document (see the
4063            comment below).  So we roll our own .... */
4064
4065         for (i = 0; i < 32; i++)
4066                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4067
4068         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4069         if (pos == 0) {
4070                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4071                 return -ENODEV;
4072         }
4073
4074         /* Quoting from the Open CISS Specification: "The Power
4075          * Management Control/Status Register (CSR) controls the power
4076          * state of the device.  The normal operating state is D0,
4077          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4078          * the controller, place the interface device in D3 then to
4079          * D0, this causes a secondary PCI reset which will reset the
4080          * controller." */
4081
4082         /* enter the D3hot power management state */
4083         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4084         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4085         pmcsr |= PCI_D3hot;
4086         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4087
4088         schedule_timeout_uninterruptible(HZ >> 1);
4089
4090         /* enter the D0 power management state */
4091         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4092         pmcsr |= PCI_D0;
4093         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4094
4095         schedule_timeout_uninterruptible(HZ >> 1);
4096
4097         /* Restore the PCI configuration space.  The Open CISS
4098          * Specification says, "Restore the PCI Configuration
4099          * Registers, offsets 00h through 60h. It is important to
4100          * restore the command register, 16-bits at offset 04h,
4101          * last. Do not restore the configuration status register,
4102          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4103         for (i = 0; i < 32; i++) {
4104                 if (i == 2 || i == 3)
4105                         continue;
4106                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4107         }
4108         wmb();
4109         pci_write_config_word(pdev, 4, saved_config_space[2]);
4110
4111         return 0;
4112 }
4113
4114 /*
4115  *  This is it.  Find all the controllers and register them.  I really hate
4116  *  stealing all these major device numbers.
4117  *  returns the number of block devices registered.
4118  */
4119 static int __devinit cciss_init_one(struct pci_dev *pdev,
4120                                     const struct pci_device_id *ent)
4121 {
4122         int i;
4123         int j = 0;
4124         int k = 0;
4125         int rc;
4126         int dac, return_code;
4127         InquiryData_struct *inq_buff;
4128
4129         if (reset_devices) {
4130                 /* Reset the controller with a PCI power-cycle */
4131                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4132                         return -ENODEV;
4133
4134                 /* Now try to get the controller to respond to a no-op. Some
4135                    devices (notably the HP Smart Array 5i Controller) need
4136                    up to 30 seconds to respond. */
4137                 for (i=0; i<30; i++) {
4138                         if (cciss_noop(pdev) == 0)
4139                                 break;
4140
4141                         schedule_timeout_uninterruptible(HZ);
4142                 }
4143                 if (i == 30) {
4144                         printk(KERN_ERR "cciss: controller seems dead\n");
4145                         return -EBUSY;
4146                 }
4147         }
4148
4149         i = alloc_cciss_hba();
4150         if (i < 0)
4151                 return -1;
4152
4153         hba[i]->busy_initializing = 1;
4154         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4155         INIT_HLIST_HEAD(&hba[i]->reqQ);
4156         mutex_init(&hba[i]->busy_shutting_down);
4157
4158         if (cciss_pci_init(hba[i], pdev) != 0)
4159                 goto clean_no_release_regions;
4160
4161         sprintf(hba[i]->devname, "cciss%d", i);
4162         hba[i]->ctlr = i;
4163         hba[i]->pdev = pdev;
4164
4165         init_completion(&hba[i]->scan_wait);
4166
4167         if (cciss_create_hba_sysfs_entry(hba[i]))
4168                 goto clean0;
4169
4170         /* configure PCI DMA stuff */
4171         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4172                 dac = 1;
4173         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4174                 dac = 0;
4175         else {
4176                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4177                 goto clean1;
4178         }
4179
4180         /*
4181          * register with the major number, or get a dynamic major number
4182          * by passing 0 as argument.  This is done for greater than
4183          * 8 controller support.
4184          */
4185         if (i < MAX_CTLR_ORIG)
4186                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4187         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4188         if (rc == -EBUSY || rc == -EINVAL) {
4189                 printk(KERN_ERR
4190                        "cciss:  Unable to get major number %d for %s "
4191                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4192                 goto clean1;
4193         } else {
4194                 if (i >= MAX_CTLR_ORIG)
4195                         hba[i]->major = rc;
4196         }
4197
4198         /* make sure the board interrupts are off */
4199         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4200         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
4201                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
4202                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4203                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4204                 goto clean2;
4205         }
4206
4207         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4208                hba[i]->devname, pdev->device, pci_name(pdev),
4209                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4210
4211         hba[i]->cmd_pool_bits =
4212             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4213                         * sizeof(unsigned long), GFP_KERNEL);
4214         hba[i]->cmd_pool = (CommandList_struct *)
4215             pci_alloc_consistent(hba[i]->pdev,
4216                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4217                     &(hba[i]->cmd_pool_dhandle));
4218         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4219             pci_alloc_consistent(hba[i]->pdev,
4220                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4221                     &(hba[i]->errinfo_pool_dhandle));
4222         if ((hba[i]->cmd_pool_bits == NULL)
4223             || (hba[i]->cmd_pool == NULL)
4224             || (hba[i]->errinfo_pool == NULL)) {
4225                 printk(KERN_ERR "cciss: out of memory");
4226                 goto clean4;
4227         }
4228
4229         /* Need space for temp scatter list */
4230         hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4231                                                 sizeof(struct scatterlist *),
4232                                                 GFP_KERNEL);
4233         for (k = 0; k < hba[i]->nr_cmds; k++) {
4234                 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4235                                                         hba[i]->maxsgentries,
4236                                                         GFP_KERNEL);
4237                 if (hba[i]->scatter_list[k] == NULL) {
4238                         printk(KERN_ERR "cciss%d: could not allocate "
4239                                 "s/g lists\n", i);
4240                         goto clean4;
4241                 }
4242         }
4243         hba[i]->cmd_sg_list = kmalloc(sizeof(struct Cmd_sg_list *) *
4244                                                 hba[i]->nr_cmds,
4245                                                 GFP_KERNEL);
4246         if (!hba[i]->cmd_sg_list) {
4247                 printk(KERN_ERR "cciss%d: Cannot get memory for "
4248                         "s/g chaining.\n", i);
4249                 goto clean4;
4250         }
4251         /* Build up chain blocks for each command */
4252         if (hba[i]->chainsize > 0) {
4253                 for (j = 0; j < hba[i]->nr_cmds; j++) {
4254                         hba[i]->cmd_sg_list[j] =
4255                                         kmalloc(sizeof(struct Cmd_sg_list),
4256                                                         GFP_KERNEL);
4257                         if (!hba[i]->cmd_sg_list[j]) {
4258                                 printk(KERN_ERR "cciss%d: Cannot get memory "
4259                                         "for chain block.\n", i);
4260                                 goto clean4;
4261                         }
4262                         /* Need a block of chainsized s/g elements. */
4263                         hba[i]->cmd_sg_list[j]->sgchain =
4264                                         kmalloc((hba[i]->chainsize *
4265                                                 sizeof(SGDescriptor_struct)),
4266                                                 GFP_KERNEL);
4267                         if (!hba[i]->cmd_sg_list[j]->sgchain) {
4268                                 printk(KERN_ERR "cciss%d: Cannot get memory "
4269                                         "for s/g chains\n", i);
4270                                 goto clean4;
4271                         }
4272                 }
4273         }
4274
4275         spin_lock_init(&hba[i]->lock);
4276
4277         /* Initialize the pdev driver private data.
4278            have it point to hba[i].  */
4279         pci_set_drvdata(pdev, hba[i]);
4280         /* command and error info recs zeroed out before
4281            they are used */
4282         memset(hba[i]->cmd_pool_bits, 0,
4283                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4284                         * sizeof(unsigned long));
4285
4286         hba[i]->num_luns = 0;
4287         hba[i]->highest_lun = -1;
4288         for (j = 0; j < CISS_MAX_LUN; j++) {
4289                 hba[i]->drv[j] = NULL;
4290                 hba[i]->gendisk[j] = NULL;
4291         }
4292
4293         cciss_scsi_setup(i);
4294
4295         /* Turn the interrupts on so we can service requests */
4296         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4297
4298         /* Get the firmware version */
4299         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4300         if (inq_buff == NULL) {
4301                 printk(KERN_ERR "cciss: out of memory\n");
4302                 goto clean4;
4303         }
4304
4305         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4306                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4307         if (return_code == IO_OK) {
4308                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4309                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4310                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4311                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4312         } else {         /* send command failed */
4313                 printk(KERN_WARNING "cciss: unable to determine firmware"
4314                         " version of controller\n");
4315         }
4316         kfree(inq_buff);
4317
4318         cciss_procinit(i);
4319
4320         hba[i]->cciss_max_sectors = 8192;
4321
4322         rebuild_lun_table(hba[i], 1, 0);
4323         hba[i]->busy_initializing = 0;
4324         return 1;
4325
4326 clean4:
4327         kfree(hba[i]->cmd_pool_bits);
4328         /* Free up sg elements */
4329         for (k = 0; k < hba[i]->nr_cmds; k++)
4330                 kfree(hba[i]->scatter_list[k]);
4331         kfree(hba[i]->scatter_list);
4332         /* Only free up extra s/g lists if controller supports them */
4333         if (hba[i]->chainsize > 0) {
4334                 for (j = 0; j < hba[i]->nr_cmds; j++) {
4335                         if (hba[i]->cmd_sg_list[j]) {
4336                                 kfree(hba[i]->cmd_sg_list[j]->sgchain);
4337                                 kfree(hba[i]->cmd_sg_list[j]);
4338                         }
4339                 }
4340                 kfree(hba[i]->cmd_sg_list);
4341         }
4342         if (hba[i]->cmd_pool)
4343                 pci_free_consistent(hba[i]->pdev,
4344                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4345                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4346         if (hba[i]->errinfo_pool)
4347                 pci_free_consistent(hba[i]->pdev,
4348                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4349                                     hba[i]->errinfo_pool,
4350                                     hba[i]->errinfo_pool_dhandle);
4351         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4352 clean2:
4353         unregister_blkdev(hba[i]->major, hba[i]->devname);
4354 clean1:
4355         cciss_destroy_hba_sysfs_entry(hba[i]);
4356 clean0:
4357         pci_release_regions(pdev);
4358 clean_no_release_regions:
4359         hba[i]->busy_initializing = 0;
4360
4361         /*
4362          * Deliberately omit pci_disable_device(): it does something nasty to
4363          * Smart Array controllers that pci_enable_device does not undo
4364          */
4365         pci_set_drvdata(pdev, NULL);
4366         free_hba(i);
4367         return -1;
4368 }
4369
4370 static void cciss_shutdown(struct pci_dev *pdev)
4371 {
4372         ctlr_info_t *h;
4373         char *flush_buf;
4374         int return_code;
4375
4376         h = pci_get_drvdata(pdev);
4377         flush_buf = kzalloc(4, GFP_KERNEL);
4378         if (!flush_buf) {
4379                 printk(KERN_WARNING
4380                         "cciss:%d cache not flushed, out of memory.\n",
4381                         h->ctlr);
4382                 return;
4383         }
4384         /* write all data in the battery backed cache to disk */
4385         memset(flush_buf, 0, 4);
4386         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4387                 4, 0, CTLR_LUNID, TYPE_CMD);
4388         kfree(flush_buf);
4389         if (return_code != IO_OK)
4390                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4391                         h->ctlr);
4392         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4393         free_irq(h->intr[2], h);
4394 }
4395
4396 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4397 {
4398         ctlr_info_t *tmp_ptr;
4399         int i, j;
4400
4401         if (pci_get_drvdata(pdev) == NULL) {
4402                 printk(KERN_ERR "cciss: Unable to remove device \n");
4403                 return;
4404         }
4405
4406         tmp_ptr = pci_get_drvdata(pdev);
4407         i = tmp_ptr->ctlr;
4408         if (hba[i] == NULL) {
4409                 printk(KERN_ERR "cciss: device appears to "
4410                        "already be removed \n");
4411                 return;
4412         }
4413
4414         mutex_lock(&hba[i]->busy_shutting_down);
4415
4416         remove_from_scan_list(hba[i]);
4417         remove_proc_entry(hba[i]->devname, proc_cciss);
4418         unregister_blkdev(hba[i]->major, hba[i]->devname);
4419
4420         /* remove it from the disk list */
4421         for (j = 0; j < CISS_MAX_LUN; j++) {
4422                 struct gendisk *disk = hba[i]->gendisk[j];
4423                 if (disk) {
4424                         struct request_queue *q = disk->queue;
4425
4426                         if (disk->flags & GENHD_FL_UP) {
4427                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4428                                 del_gendisk(disk);
4429                         }
4430                         if (q)
4431                                 blk_cleanup_queue(q);
4432                 }
4433         }
4434
4435 #ifdef CONFIG_CISS_SCSI_TAPE
4436         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4437 #endif
4438
4439         cciss_shutdown(pdev);
4440
4441 #ifdef CONFIG_PCI_MSI
4442         if (hba[i]->msix_vector)
4443                 pci_disable_msix(hba[i]->pdev);
4444         else if (hba[i]->msi_vector)
4445                 pci_disable_msi(hba[i]->pdev);
4446 #endif                          /* CONFIG_PCI_MSI */
4447
4448         iounmap(hba[i]->vaddr);
4449
4450         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4451                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4452         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4453                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4454         kfree(hba[i]->cmd_pool_bits);
4455         /* Free up sg elements */
4456         for (j = 0; j < hba[i]->nr_cmds; j++)
4457                 kfree(hba[i]->scatter_list[j]);
4458         kfree(hba[i]->scatter_list);
4459         /* Only free up extra s/g lists if controller supports them */
4460         if (hba[i]->chainsize > 0) {
4461                 for (j = 0; j < hba[i]->nr_cmds; j++) {
4462                         if (hba[i]->cmd_sg_list[j]) {
4463                                 kfree(hba[i]->cmd_sg_list[j]->sgchain);
4464                                 kfree(hba[i]->cmd_sg_list[j]);
4465                         }
4466                 }
4467                 kfree(hba[i]->cmd_sg_list);
4468         }
4469         /*
4470          * Deliberately omit pci_disable_device(): it does something nasty to
4471          * Smart Array controllers that pci_enable_device does not undo
4472          */
4473         pci_release_regions(pdev);
4474         pci_set_drvdata(pdev, NULL);
4475         cciss_destroy_hba_sysfs_entry(hba[i]);
4476         mutex_unlock(&hba[i]->busy_shutting_down);
4477         free_hba(i);
4478 }
4479
4480 static struct pci_driver cciss_pci_driver = {
4481         .name = "cciss",
4482         .probe = cciss_init_one,
4483         .remove = __devexit_p(cciss_remove_one),
4484         .id_table = cciss_pci_device_id,        /* id_table */
4485         .shutdown = cciss_shutdown,
4486 };
4487
4488 /*
4489  *  This is it.  Register the PCI driver information for the cards we control
4490  *  the OS will call our registered routines when it finds one of our cards.
4491  */
4492 static int __init cciss_init(void)
4493 {
4494         int err;
4495
4496         /*
4497          * The hardware requires that commands are aligned on a 64-bit
4498          * boundary. Given that we use pci_alloc_consistent() to allocate an
4499          * array of them, the size must be a multiple of 8 bytes.
4500          */
4501         BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4502
4503         printk(KERN_INFO DRIVER_NAME "\n");
4504
4505         err = bus_register(&cciss_bus_type);
4506         if (err)
4507                 return err;
4508
4509         /* Start the scan thread */
4510         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4511         if (IS_ERR(cciss_scan_thread)) {
4512                 err = PTR_ERR(cciss_scan_thread);
4513                 goto err_bus_unregister;
4514         }
4515
4516         /* Register for our PCI devices */
4517         err = pci_register_driver(&cciss_pci_driver);
4518         if (err)
4519                 goto err_thread_stop;
4520
4521         return err;
4522
4523 err_thread_stop:
4524         kthread_stop(cciss_scan_thread);
4525 err_bus_unregister:
4526         bus_unregister(&cciss_bus_type);
4527
4528         return err;
4529 }
4530
4531 static void __exit cciss_cleanup(void)
4532 {
4533         int i;
4534
4535         pci_unregister_driver(&cciss_pci_driver);
4536         /* double check that all controller entrys have been removed */
4537         for (i = 0; i < MAX_CTLR; i++) {
4538                 if (hba[i] != NULL) {
4539                         printk(KERN_WARNING "cciss: had to remove"
4540                                " controller %d\n", i);
4541                         cciss_remove_one(hba[i]->pdev);
4542                 }
4543         }
4544         kthread_stop(cciss_scan_thread);
4545         remove_proc_entry("driver/cciss", NULL);
4546         bus_unregister(&cciss_bus_type);
4547 }
4548
4549 static void fail_all_cmds(unsigned long ctlr)
4550 {
4551         /* If we get here, the board is apparently dead. */
4552         ctlr_info_t *h = hba[ctlr];
4553         CommandList_struct *c;
4554         unsigned long flags;
4555
4556         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4557         h->alive = 0;           /* the controller apparently died... */
4558
4559         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4560
4561         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
4562
4563         /* move everything off the request queue onto the completed queue */
4564         while (!hlist_empty(&h->reqQ)) {
4565                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4566                 removeQ(c);
4567                 h->Qdepth--;
4568                 addQ(&h->cmpQ, c);
4569         }
4570
4571         /* Now, fail everything on the completed queue with a HW error */
4572         while (!hlist_empty(&h->cmpQ)) {
4573                 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4574                 removeQ(c);
4575                 if (c->cmd_type != CMD_MSG_STALE)
4576                         c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4577                 if (c->cmd_type == CMD_RWREQ) {
4578                         complete_command(h, c, 0);
4579                 } else if (c->cmd_type == CMD_IOCTL_PEND)
4580                         complete(c->waiting);
4581 #ifdef CONFIG_CISS_SCSI_TAPE
4582                 else if (c->cmd_type == CMD_SCSI)
4583                         complete_scsi_command(c, 0, 0);
4584 #endif
4585         }
4586         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4587         return;
4588 }
4589
4590 module_init(cciss_init);
4591 module_exit(cciss_cleanup);