SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/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                         /* use the kernel address the cmd block for tag */
1362                         c->Header.Tag.lower = c->busaddr;
1363
1364                         /* Fill in Request block */
1365                         c->Request = iocommand.Request;
1366
1367                         /* Fill in the scatter gather information */
1368                         if (iocommand.buf_size > 0) {
1369                                 temp64.val = pci_map_single(host->pdev, buff,
1370                                         iocommand.buf_size,
1371                                         PCI_DMA_BIDIRECTIONAL);
1372                                 c->SG[0].Addr.lower = temp64.val32.lower;
1373                                 c->SG[0].Addr.upper = temp64.val32.upper;
1374                                 c->SG[0].Len = iocommand.buf_size;
1375                                 c->SG[0].Ext = 0;  /* we are not chaining */
1376                         }
1377                         c->waiting = &wait;
1378
1379                         /* Put the request on the tail of the request queue */
1380                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1381                         addQ(&host->reqQ, c);
1382                         host->Qdepth++;
1383                         start_io(host);
1384                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1385
1386                         wait_for_completion(&wait);
1387
1388                         /* unlock the buffers from DMA */
1389                         temp64.val32.lower = c->SG[0].Addr.lower;
1390                         temp64.val32.upper = c->SG[0].Addr.upper;
1391                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1392                                          iocommand.buf_size,
1393                                          PCI_DMA_BIDIRECTIONAL);
1394
1395                         check_ioctl_unit_attention(host, c);
1396
1397                         /* Copy the error information out */
1398                         iocommand.error_info = *(c->err_info);
1399                         if (copy_to_user
1400                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1401                                 kfree(buff);
1402                                 cmd_free(host, c, 0);
1403                                 return -EFAULT;
1404                         }
1405
1406                         if (iocommand.Request.Type.Direction == XFER_READ) {
1407                                 /* Copy the data out of the buffer we created */
1408                                 if (copy_to_user
1409                                     (iocommand.buf, buff, iocommand.buf_size)) {
1410                                         kfree(buff);
1411                                         cmd_free(host, c, 0);
1412                                         return -EFAULT;
1413                                 }
1414                         }
1415                         kfree(buff);
1416                         cmd_free(host, c, 0);
1417                         return 0;
1418                 }
1419         case CCISS_BIG_PASSTHRU:{
1420                         BIG_IOCTL_Command_struct *ioc;
1421                         CommandList_struct *c;
1422                         unsigned char **buff = NULL;
1423                         int *buff_size = NULL;
1424                         u64bit temp64;
1425                         unsigned long flags;
1426                         BYTE sg_used = 0;
1427                         int status = 0;
1428                         int i;
1429                         DECLARE_COMPLETION_ONSTACK(wait);
1430                         __u32 left;
1431                         __u32 sz;
1432                         BYTE __user *data_ptr;
1433
1434                         if (!arg)
1435                                 return -EINVAL;
1436                         if (!capable(CAP_SYS_RAWIO))
1437                                 return -EPERM;
1438                         ioc = (BIG_IOCTL_Command_struct *)
1439                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1440                         if (!ioc) {
1441                                 status = -ENOMEM;
1442                                 goto cleanup1;
1443                         }
1444                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1445                                 status = -EFAULT;
1446                                 goto cleanup1;
1447                         }
1448                         if ((ioc->buf_size < 1) &&
1449                             (ioc->Request.Type.Direction != XFER_NONE)) {
1450                                 status = -EINVAL;
1451                                 goto cleanup1;
1452                         }
1453                         /* Check kmalloc limits  using all SGs */
1454                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1455                                 status = -EINVAL;
1456                                 goto cleanup1;
1457                         }
1458                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1459                                 status = -EINVAL;
1460                                 goto cleanup1;
1461                         }
1462                         buff =
1463                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1464                         if (!buff) {
1465                                 status = -ENOMEM;
1466                                 goto cleanup1;
1467                         }
1468                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1469                                                    GFP_KERNEL);
1470                         if (!buff_size) {
1471                                 status = -ENOMEM;
1472                                 goto cleanup1;
1473                         }
1474                         left = ioc->buf_size;
1475                         data_ptr = ioc->buf;
1476                         while (left) {
1477                                 sz = (left >
1478                                       ioc->malloc_size) ? ioc->
1479                                     malloc_size : left;
1480                                 buff_size[sg_used] = sz;
1481                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1482                                 if (buff[sg_used] == NULL) {
1483                                         status = -ENOMEM;
1484                                         goto cleanup1;
1485                                 }
1486                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1487                                         if (copy_from_user
1488                                             (buff[sg_used], data_ptr, sz)) {
1489                                                 status = -EFAULT;
1490                                                 goto cleanup1;
1491                                         }
1492                                 } else {
1493                                         memset(buff[sg_used], 0, sz);
1494                                 }
1495                                 left -= sz;
1496                                 data_ptr += sz;
1497                                 sg_used++;
1498                         }
1499                         if ((c = cmd_alloc(host, 0)) == NULL) {
1500                                 status = -ENOMEM;
1501                                 goto cleanup1;
1502                         }
1503                         c->cmd_type = CMD_IOCTL_PEND;
1504                         c->Header.ReplyQueue = 0;
1505
1506                         if (ioc->buf_size > 0) {
1507                                 c->Header.SGList = sg_used;
1508                                 c->Header.SGTotal = sg_used;
1509                         } else {
1510                                 c->Header.SGList = 0;
1511                                 c->Header.SGTotal = 0;
1512                         }
1513                         c->Header.LUN = ioc->LUN_info;
1514                         c->Header.Tag.lower = c->busaddr;
1515
1516                         c->Request = ioc->Request;
1517                         if (ioc->buf_size > 0) {
1518                                 int i;
1519                                 for (i = 0; i < sg_used; i++) {
1520                                         temp64.val =
1521                                             pci_map_single(host->pdev, buff[i],
1522                                                     buff_size[i],
1523                                                     PCI_DMA_BIDIRECTIONAL);
1524                                         c->SG[i].Addr.lower =
1525                                             temp64.val32.lower;
1526                                         c->SG[i].Addr.upper =
1527                                             temp64.val32.upper;
1528                                         c->SG[i].Len = buff_size[i];
1529                                         c->SG[i].Ext = 0;       /* we are not chaining */
1530                                 }
1531                         }
1532                         c->waiting = &wait;
1533                         /* Put the request on the tail of the request queue */
1534                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1535                         addQ(&host->reqQ, c);
1536                         host->Qdepth++;
1537                         start_io(host);
1538                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1539                         wait_for_completion(&wait);
1540                         /* unlock the buffers from DMA */
1541                         for (i = 0; i < sg_used; i++) {
1542                                 temp64.val32.lower = c->SG[i].Addr.lower;
1543                                 temp64.val32.upper = c->SG[i].Addr.upper;
1544                                 pci_unmap_single(host->pdev,
1545                                         (dma_addr_t) temp64.val, buff_size[i],
1546                                         PCI_DMA_BIDIRECTIONAL);
1547                         }
1548                         check_ioctl_unit_attention(host, c);
1549                         /* Copy the error information out */
1550                         ioc->error_info = *(c->err_info);
1551                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1552                                 cmd_free(host, c, 0);
1553                                 status = -EFAULT;
1554                                 goto cleanup1;
1555                         }
1556                         if (ioc->Request.Type.Direction == XFER_READ) {
1557                                 /* Copy the data out of the buffer we created */
1558                                 BYTE __user *ptr = ioc->buf;
1559                                 for (i = 0; i < sg_used; i++) {
1560                                         if (copy_to_user
1561                                             (ptr, buff[i], buff_size[i])) {
1562                                                 cmd_free(host, c, 0);
1563                                                 status = -EFAULT;
1564                                                 goto cleanup1;
1565                                         }
1566                                         ptr += buff_size[i];
1567                                 }
1568                         }
1569                         cmd_free(host, c, 0);
1570                         status = 0;
1571                       cleanup1:
1572                         if (buff) {
1573                                 for (i = 0; i < sg_used; i++)
1574                                         kfree(buff[i]);
1575                                 kfree(buff);
1576                         }
1577                         kfree(buff_size);
1578                         kfree(ioc);
1579                         return status;
1580                 }
1581
1582         /* scsi_cmd_ioctl handles these, below, though some are not */
1583         /* very meaningful for cciss.  SG_IO is the main one people want. */
1584
1585         case SG_GET_VERSION_NUM:
1586         case SG_SET_TIMEOUT:
1587         case SG_GET_TIMEOUT:
1588         case SG_GET_RESERVED_SIZE:
1589         case SG_SET_RESERVED_SIZE:
1590         case SG_EMULATED_HOST:
1591         case SG_IO:
1592         case SCSI_IOCTL_SEND_COMMAND:
1593                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1594
1595         /* scsi_cmd_ioctl would normally handle these, below, but */
1596         /* they aren't a good fit for cciss, as CD-ROMs are */
1597         /* not supported, and we don't have any bus/target/lun */
1598         /* which we present to the kernel. */
1599
1600         case CDROM_SEND_PACKET:
1601         case CDROMCLOSETRAY:
1602         case CDROMEJECT:
1603         case SCSI_IOCTL_GET_IDLUN:
1604         case SCSI_IOCTL_GET_BUS_NUMBER:
1605         default:
1606                 return -ENOTTY;
1607         }
1608 }
1609
1610 static void cciss_check_queues(ctlr_info_t *h)
1611 {
1612         int start_queue = h->next_to_run;
1613         int i;
1614
1615         /* check to see if we have maxed out the number of commands that can
1616          * be placed on the queue.  If so then exit.  We do this check here
1617          * in case the interrupt we serviced was from an ioctl and did not
1618          * free any new commands.
1619          */
1620         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1621                 return;
1622
1623         /* We have room on the queue for more commands.  Now we need to queue
1624          * them up.  We will also keep track of the next queue to run so
1625          * that every queue gets a chance to be started first.
1626          */
1627         for (i = 0; i < h->highest_lun + 1; i++) {
1628                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1629                 /* make sure the disk has been added and the drive is real
1630                  * because this can be called from the middle of init_one.
1631                  */
1632                 if (!h->drv[curr_queue])
1633                         continue;
1634                 if (!(h->drv[curr_queue]->queue) ||
1635                         !(h->drv[curr_queue]->heads))
1636                         continue;
1637                 blk_start_queue(h->gendisk[curr_queue]->queue);
1638
1639                 /* check to see if we have maxed out the number of commands
1640                  * that can be placed on the queue.
1641                  */
1642                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1643                         if (curr_queue == start_queue) {
1644                                 h->next_to_run =
1645                                     (start_queue + 1) % (h->highest_lun + 1);
1646                                 break;
1647                         } else {
1648                                 h->next_to_run = curr_queue;
1649                                 break;
1650                         }
1651                 }
1652         }
1653 }
1654
1655 static void cciss_softirq_done(struct request *rq)
1656 {
1657         CommandList_struct *cmd = rq->completion_data;
1658         ctlr_info_t *h = hba[cmd->ctlr];
1659         SGDescriptor_struct *curr_sg = cmd->SG;
1660         unsigned long flags;
1661         u64bit temp64;
1662         int i, ddir;
1663         int sg_index = 0;
1664
1665         if (cmd->Request.Type.Direction == XFER_READ)
1666                 ddir = PCI_DMA_FROMDEVICE;
1667         else
1668                 ddir = PCI_DMA_TODEVICE;
1669
1670         /* command did not need to be retried */
1671         /* unmap the DMA mapping for all the scatter gather elements */
1672         for (i = 0; i < cmd->Header.SGList; i++) {
1673                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1674                         temp64.val32.lower = cmd->SG[i].Addr.lower;
1675                         temp64.val32.upper = cmd->SG[i].Addr.upper;
1676                         pci_dma_sync_single_for_cpu(h->pdev, temp64.val,
1677                                                 cmd->SG[i].Len, ddir);
1678                         pci_unmap_single(h->pdev, temp64.val,
1679                                                 cmd->SG[i].Len, ddir);
1680                         /* Point to the next block */
1681                         curr_sg = h->cmd_sg_list[cmd->cmdindex]->sgchain;
1682                         sg_index = 0;
1683                 }
1684                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1685                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1686                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1687                                 ddir);
1688                 ++sg_index;
1689         }
1690
1691 #ifdef CCISS_DEBUG
1692         printk("Done with %p\n", rq);
1693 #endif                          /* CCISS_DEBUG */
1694
1695         /* set the residual count for pc requests */
1696         if (blk_pc_request(rq))
1697                 rq->resid_len = cmd->err_info->ResidualCnt;
1698
1699         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1700
1701         spin_lock_irqsave(&h->lock, flags);
1702         cmd_free(h, cmd, 1);
1703         cciss_check_queues(h);
1704         spin_unlock_irqrestore(&h->lock, flags);
1705 }
1706
1707 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1708         unsigned char scsi3addr[], uint32_t log_unit)
1709 {
1710         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1711                 sizeof(h->drv[log_unit]->LunID));
1712 }
1713
1714 /* This function gets the SCSI vendor, model, and revision of a logical drive
1715  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1716  * they cannot be read.
1717  */
1718 static void cciss_get_device_descr(int ctlr, int logvol,
1719                                    char *vendor, char *model, char *rev)
1720 {
1721         int rc;
1722         InquiryData_struct *inq_buf;
1723         unsigned char scsi3addr[8];
1724
1725         *vendor = '\0';
1726         *model = '\0';
1727         *rev = '\0';
1728
1729         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1730         if (!inq_buf)
1731                 return;
1732
1733         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1734         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1735                         scsi3addr, TYPE_CMD);
1736         if (rc == IO_OK) {
1737                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1738                 vendor[VENDOR_LEN] = '\0';
1739                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1740                 model[MODEL_LEN] = '\0';
1741                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1742                 rev[REV_LEN] = '\0';
1743         }
1744
1745         kfree(inq_buf);
1746         return;
1747 }
1748
1749 /* This function gets the serial number of a logical drive via
1750  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1751  * number cannot be had, for whatever reason, 16 bytes of 0xff
1752  * are returned instead.
1753  */
1754 static void cciss_get_serial_no(int ctlr, int logvol,
1755                                 unsigned char *serial_no, int buflen)
1756 {
1757 #define PAGE_83_INQ_BYTES 64
1758         int rc;
1759         unsigned char *buf;
1760         unsigned char scsi3addr[8];
1761
1762         if (buflen > 16)
1763                 buflen = 16;
1764         memset(serial_no, 0xff, buflen);
1765         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1766         if (!buf)
1767                 return;
1768         memset(serial_no, 0, buflen);
1769         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1770         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1771                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1772         if (rc == IO_OK)
1773                 memcpy(serial_no, &buf[8], buflen);
1774         kfree(buf);
1775         return;
1776 }
1777
1778 /*
1779  * cciss_add_disk sets up the block device queue for a logical drive
1780  */
1781 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1782                                 int drv_index)
1783 {
1784         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1785         if (!disk->queue)
1786                 goto init_queue_failure;
1787         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1788         disk->major = h->major;
1789         disk->first_minor = drv_index << NWD_SHIFT;
1790         disk->fops = &cciss_fops;
1791         if (cciss_create_ld_sysfs_entry(h, drv_index))
1792                 goto cleanup_queue;
1793         disk->private_data = h->drv[drv_index];
1794         disk->driverfs_dev = &h->drv[drv_index]->dev;
1795
1796         /* Set up queue information */
1797         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1798
1799         /* This is a hardware imposed limit. */
1800         blk_queue_max_hw_segments(disk->queue, h->maxsgentries);
1801
1802         /* This is a limit in the driver and could be eliminated. */
1803         blk_queue_max_phys_segments(disk->queue, h->maxsgentries);
1804
1805         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1806
1807         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1808
1809         disk->queue->queuedata = h;
1810
1811         blk_queue_logical_block_size(disk->queue,
1812                                      h->drv[drv_index]->block_size);
1813
1814         /* Make sure all queue data is written out before */
1815         /* setting h->drv[drv_index]->queue, as setting this */
1816         /* allows the interrupt handler to start the queue */
1817         wmb();
1818         h->drv[drv_index]->queue = disk->queue;
1819         add_disk(disk);
1820         return 0;
1821
1822 cleanup_queue:
1823         blk_cleanup_queue(disk->queue);
1824         disk->queue = NULL;
1825 init_queue_failure:
1826         return -1;
1827 }
1828
1829 /* This function will check the usage_count of the drive to be updated/added.
1830  * If the usage_count is zero and it is a heretofore unknown drive, or,
1831  * the drive's capacity, geometry, or serial number has changed,
1832  * then the drive information will be updated and the disk will be
1833  * re-registered with the kernel.  If these conditions don't hold,
1834  * then it will be left alone for the next reboot.  The exception to this
1835  * is disk 0 which will always be left registered with the kernel since it
1836  * is also the controller node.  Any changes to disk 0 will show up on
1837  * the next reboot.
1838  */
1839 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1840         int via_ioctl)
1841 {
1842         ctlr_info_t *h = hba[ctlr];
1843         struct gendisk *disk;
1844         InquiryData_struct *inq_buff = NULL;
1845         unsigned int block_size;
1846         sector_t total_size;
1847         unsigned long flags = 0;
1848         int ret = 0;
1849         drive_info_struct *drvinfo;
1850
1851         /* Get information about the disk and modify the driver structure */
1852         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1853         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1854         if (inq_buff == NULL || drvinfo == NULL)
1855                 goto mem_msg;
1856
1857         /* testing to see if 16-byte CDBs are already being used */
1858         if (h->cciss_read == CCISS_READ_16) {
1859                 cciss_read_capacity_16(h->ctlr, drv_index,
1860                         &total_size, &block_size);
1861
1862         } else {
1863                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1864                 /* if read_capacity returns all F's this volume is >2TB */
1865                 /* in size so we switch to 16-byte CDB's for all */
1866                 /* read/write ops */
1867                 if (total_size == 0xFFFFFFFFULL) {
1868                         cciss_read_capacity_16(ctlr, drv_index,
1869                         &total_size, &block_size);
1870                         h->cciss_read = CCISS_READ_16;
1871                         h->cciss_write = CCISS_WRITE_16;
1872                 } else {
1873                         h->cciss_read = CCISS_READ_10;
1874                         h->cciss_write = CCISS_WRITE_10;
1875                 }
1876         }
1877
1878         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1879                                inq_buff, drvinfo);
1880         drvinfo->block_size = block_size;
1881         drvinfo->nr_blocks = total_size + 1;
1882
1883         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1884                                 drvinfo->model, drvinfo->rev);
1885         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1886                         sizeof(drvinfo->serial_no));
1887         /* Save the lunid in case we deregister the disk, below. */
1888         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1889                 sizeof(drvinfo->LunID));
1890
1891         /* Is it the same disk we already know, and nothing's changed? */
1892         if (h->drv[drv_index]->raid_level != -1 &&
1893                 ((memcmp(drvinfo->serial_no,
1894                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1895                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1896                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1897                 drvinfo->heads == h->drv[drv_index]->heads &&
1898                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1899                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1900                         /* The disk is unchanged, nothing to update */
1901                         goto freeret;
1902
1903         /* If we get here it's not the same disk, or something's changed,
1904          * so we need to * deregister it, and re-register it, if it's not
1905          * in use.
1906          * If the disk already exists then deregister it before proceeding
1907          * (unless it's the first disk (for the controller node).
1908          */
1909         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1910                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1911                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1912                 h->drv[drv_index]->busy_configuring = 1;
1913                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1914
1915                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1916                  * which keeps the interrupt handler from starting
1917                  * the queue.
1918                  */
1919                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1920         }
1921
1922         /* If the disk is in use return */
1923         if (ret)
1924                 goto freeret;
1925
1926         /* Save the new information from cciss_geometry_inquiry
1927          * and serial number inquiry.  If the disk was deregistered
1928          * above, then h->drv[drv_index] will be NULL.
1929          */
1930         if (h->drv[drv_index] == NULL) {
1931                 drvinfo->device_initialized = 0;
1932                 h->drv[drv_index] = drvinfo;
1933                 drvinfo = NULL; /* so it won't be freed below. */
1934         } else {
1935                 /* special case for cxd0 */
1936                 h->drv[drv_index]->block_size = drvinfo->block_size;
1937                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
1938                 h->drv[drv_index]->heads = drvinfo->heads;
1939                 h->drv[drv_index]->sectors = drvinfo->sectors;
1940                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
1941                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
1942                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
1943                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
1944                         VENDOR_LEN + 1);
1945                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
1946                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
1947         }
1948
1949         ++h->num_luns;
1950         disk = h->gendisk[drv_index];
1951         set_capacity(disk, h->drv[drv_index]->nr_blocks);
1952
1953         /* If it's not disk 0 (drv_index != 0)
1954          * or if it was disk 0, but there was previously
1955          * no actual corresponding configured logical drive
1956          * (raid_leve == -1) then we want to update the
1957          * logical drive's information.
1958          */
1959         if (drv_index || first_time) {
1960                 if (cciss_add_disk(h, disk, drv_index) != 0) {
1961                         cciss_free_gendisk(h, drv_index);
1962                         cciss_free_drive_info(h, drv_index);
1963                         printk(KERN_WARNING "cciss:%d could not update "
1964                                 "disk %d\n", h->ctlr, drv_index);
1965                         --h->num_luns;
1966                 }
1967         }
1968
1969 freeret:
1970         kfree(inq_buff);
1971         kfree(drvinfo);
1972         return;
1973 mem_msg:
1974         printk(KERN_ERR "cciss: out of memory\n");
1975         goto freeret;
1976 }
1977
1978 /* This function will find the first index of the controllers drive array
1979  * that has a null drv pointer and allocate the drive info struct and
1980  * will return that index   This is where new drives will be added.
1981  * If the index to be returned is greater than the highest_lun index for
1982  * the controller then highest_lun is set * to this new index.
1983  * If there are no available indexes or if tha allocation fails, then -1
1984  * is returned.  * "controller_node" is used to know if this is a real
1985  * logical drive, or just the controller node, which determines if this
1986  * counts towards highest_lun.
1987  */
1988 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
1989 {
1990         int i;
1991         drive_info_struct *drv;
1992
1993         /* Search for an empty slot for our drive info */
1994         for (i = 0; i < CISS_MAX_LUN; i++) {
1995
1996                 /* if not cxd0 case, and it's occupied, skip it. */
1997                 if (h->drv[i] && i != 0)
1998                         continue;
1999                 /*
2000                  * If it's cxd0 case, and drv is alloc'ed already, and a
2001                  * disk is configured there, skip it.
2002                  */
2003                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2004                         continue;
2005
2006                 /*
2007                  * We've found an empty slot.  Update highest_lun
2008                  * provided this isn't just the fake cxd0 controller node.
2009                  */
2010                 if (i > h->highest_lun && !controller_node)
2011                         h->highest_lun = i;
2012
2013                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2014                 if (i == 0 && h->drv[i] != NULL)
2015                         return i;
2016
2017                 /*
2018                  * Found an empty slot, not already alloc'ed.  Allocate it.
2019                  * Mark it with raid_level == -1, so we know it's new later on.
2020                  */
2021                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2022                 if (!drv)
2023                         return -1;
2024                 drv->raid_level = -1; /* so we know it's new */
2025                 h->drv[i] = drv;
2026                 return i;
2027         }
2028         return -1;
2029 }
2030
2031 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2032 {
2033         kfree(h->drv[drv_index]);
2034         h->drv[drv_index] = NULL;
2035 }
2036
2037 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2038 {
2039         put_disk(h->gendisk[drv_index]);
2040         h->gendisk[drv_index] = NULL;
2041 }
2042
2043 /* cciss_add_gendisk finds a free hba[]->drv structure
2044  * and allocates a gendisk if needed, and sets the lunid
2045  * in the drvinfo structure.   It returns the index into
2046  * the ->drv[] array, or -1 if none are free.
2047  * is_controller_node indicates whether highest_lun should
2048  * count this disk, or if it's only being added to provide
2049  * a means to talk to the controller in case no logical
2050  * drives have yet been configured.
2051  */
2052 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2053         int controller_node)
2054 {
2055         int drv_index;
2056
2057         drv_index = cciss_alloc_drive_info(h, controller_node);
2058         if (drv_index == -1)
2059                 return -1;
2060
2061         /*Check if the gendisk needs to be allocated */
2062         if (!h->gendisk[drv_index]) {
2063                 h->gendisk[drv_index] =
2064                         alloc_disk(1 << NWD_SHIFT);
2065                 if (!h->gendisk[drv_index]) {
2066                         printk(KERN_ERR "cciss%d: could not "
2067                                 "allocate a new disk %d\n",
2068                                 h->ctlr, drv_index);
2069                         goto err_free_drive_info;
2070                 }
2071         }
2072         memcpy(h->drv[drv_index]->LunID, lunid,
2073                 sizeof(h->drv[drv_index]->LunID));
2074         if (cciss_create_ld_sysfs_entry(h, drv_index))
2075                 goto err_free_disk;
2076         /* Don't need to mark this busy because nobody */
2077         /* else knows about this disk yet to contend */
2078         /* for access to it. */
2079         h->drv[drv_index]->busy_configuring = 0;
2080         wmb();
2081         return drv_index;
2082
2083 err_free_disk:
2084         cciss_free_gendisk(h, drv_index);
2085 err_free_drive_info:
2086         cciss_free_drive_info(h, drv_index);
2087         return -1;
2088 }
2089
2090 /* This is for the special case of a controller which
2091  * has no logical drives.  In this case, we still need
2092  * to register a disk so the controller can be accessed
2093  * by the Array Config Utility.
2094  */
2095 static void cciss_add_controller_node(ctlr_info_t *h)
2096 {
2097         struct gendisk *disk;
2098         int drv_index;
2099
2100         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2101                 return;
2102
2103         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2104         if (drv_index == -1)
2105                 goto error;
2106         h->drv[drv_index]->block_size = 512;
2107         h->drv[drv_index]->nr_blocks = 0;
2108         h->drv[drv_index]->heads = 0;
2109         h->drv[drv_index]->sectors = 0;
2110         h->drv[drv_index]->cylinders = 0;
2111         h->drv[drv_index]->raid_level = -1;
2112         memset(h->drv[drv_index]->serial_no, 0, 16);
2113         disk = h->gendisk[drv_index];
2114         if (cciss_add_disk(h, disk, drv_index) == 0)
2115                 return;
2116         cciss_free_gendisk(h, drv_index);
2117         cciss_free_drive_info(h, drv_index);
2118 error:
2119         printk(KERN_WARNING "cciss%d: could not "
2120                 "add disk 0.\n", h->ctlr);
2121         return;
2122 }
2123
2124 /* This function will add and remove logical drives from the Logical
2125  * drive array of the controller and maintain persistency of ordering
2126  * so that mount points are preserved until the next reboot.  This allows
2127  * for the removal of logical drives in the middle of the drive array
2128  * without a re-ordering of those drives.
2129  * INPUT
2130  * h            = The controller to perform the operations on
2131  */
2132 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2133         int via_ioctl)
2134 {
2135         int ctlr = h->ctlr;
2136         int num_luns;
2137         ReportLunData_struct *ld_buff = NULL;
2138         int return_code;
2139         int listlength = 0;
2140         int i;
2141         int drv_found;
2142         int drv_index = 0;
2143         unsigned char lunid[8] = CTLR_LUNID;
2144         unsigned long flags;
2145
2146         if (!capable(CAP_SYS_RAWIO))
2147                 return -EPERM;
2148
2149         /* Set busy_configuring flag for this operation */
2150         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2151         if (h->busy_configuring) {
2152                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2153                 return -EBUSY;
2154         }
2155         h->busy_configuring = 1;
2156         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2157
2158         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2159         if (ld_buff == NULL)
2160                 goto mem_msg;
2161
2162         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2163                                       sizeof(ReportLunData_struct),
2164                                       0, CTLR_LUNID, TYPE_CMD);
2165
2166         if (return_code == IO_OK)
2167                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2168         else {  /* reading number of logical volumes failed */
2169                 printk(KERN_WARNING "cciss: report logical volume"
2170                        " command failed\n");
2171                 listlength = 0;
2172                 goto freeret;
2173         }
2174
2175         num_luns = listlength / 8;      /* 8 bytes per entry */
2176         if (num_luns > CISS_MAX_LUN) {
2177                 num_luns = CISS_MAX_LUN;
2178                 printk(KERN_WARNING "cciss: more luns configured"
2179                        " on controller than can be handled by"
2180                        " this driver.\n");
2181         }
2182
2183         if (num_luns == 0)
2184                 cciss_add_controller_node(h);
2185
2186         /* Compare controller drive array to driver's drive array
2187          * to see if any drives are missing on the controller due
2188          * to action of Array Config Utility (user deletes drive)
2189          * and deregister logical drives which have disappeared.
2190          */
2191         for (i = 0; i <= h->highest_lun; i++) {
2192                 int j;
2193                 drv_found = 0;
2194
2195                 /* skip holes in the array from already deleted drives */
2196                 if (h->drv[i] == NULL)
2197                         continue;
2198
2199                 for (j = 0; j < num_luns; j++) {
2200                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2201                         if (memcmp(h->drv[i]->LunID, lunid,
2202                                 sizeof(lunid)) == 0) {
2203                                 drv_found = 1;
2204                                 break;
2205                         }
2206                 }
2207                 if (!drv_found) {
2208                         /* Deregister it from the OS, it's gone. */
2209                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2210                         h->drv[i]->busy_configuring = 1;
2211                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2212                         return_code = deregister_disk(h, i, 1, via_ioctl);
2213                         if (h->drv[i] != NULL)
2214                                 h->drv[i]->busy_configuring = 0;
2215                 }
2216         }
2217
2218         /* Compare controller drive array to driver's drive array.
2219          * Check for updates in the drive information and any new drives
2220          * on the controller due to ACU adding logical drives, or changing
2221          * a logical drive's size, etc.  Reregister any new/changed drives
2222          */
2223         for (i = 0; i < num_luns; i++) {
2224                 int j;
2225
2226                 drv_found = 0;
2227
2228                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2229                 /* Find if the LUN is already in the drive array
2230                  * of the driver.  If so then update its info
2231                  * if not in use.  If it does not exist then find
2232                  * the first free index and add it.
2233                  */
2234                 for (j = 0; j <= h->highest_lun; j++) {
2235                         if (h->drv[j] != NULL &&
2236                                 memcmp(h->drv[j]->LunID, lunid,
2237                                         sizeof(h->drv[j]->LunID)) == 0) {
2238                                 drv_index = j;
2239                                 drv_found = 1;
2240                                 break;
2241                         }
2242                 }
2243
2244                 /* check if the drive was found already in the array */
2245                 if (!drv_found) {
2246                         drv_index = cciss_add_gendisk(h, lunid, 0);
2247                         if (drv_index == -1)
2248                                 goto freeret;
2249                 }
2250                 cciss_update_drive_info(ctlr, drv_index, first_time,
2251                         via_ioctl);
2252         }               /* end for */
2253
2254 freeret:
2255         kfree(ld_buff);
2256         h->busy_configuring = 0;
2257         /* We return -1 here to tell the ACU that we have registered/updated
2258          * all of the drives that we can and to keep it from calling us
2259          * additional times.
2260          */
2261         return -1;
2262 mem_msg:
2263         printk(KERN_ERR "cciss: out of memory\n");
2264         h->busy_configuring = 0;
2265         goto freeret;
2266 }
2267
2268 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2269 {
2270         /* zero out the disk size info */
2271         drive_info->nr_blocks = 0;
2272         drive_info->block_size = 0;
2273         drive_info->heads = 0;
2274         drive_info->sectors = 0;
2275         drive_info->cylinders = 0;
2276         drive_info->raid_level = -1;
2277         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2278         memset(drive_info->model, 0, sizeof(drive_info->model));
2279         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2280         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2281         /*
2282          * don't clear the LUNID though, we need to remember which
2283          * one this one is.
2284          */
2285 }
2286
2287 /* This function will deregister the disk and it's queue from the
2288  * kernel.  It must be called with the controller lock held and the
2289  * drv structures busy_configuring flag set.  It's parameters are:
2290  *
2291  * disk = This is the disk to be deregistered
2292  * drv  = This is the drive_info_struct associated with the disk to be
2293  *        deregistered.  It contains information about the disk used
2294  *        by the driver.
2295  * clear_all = This flag determines whether or not the disk information
2296  *             is going to be completely cleared out and the highest_lun
2297  *             reset.  Sometimes we want to clear out information about
2298  *             the disk in preparation for re-adding it.  In this case
2299  *             the highest_lun should be left unchanged and the LunID
2300  *             should not be cleared.
2301  * via_ioctl
2302  *    This indicates whether we've reached this path via ioctl.
2303  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2304  *    If this path is reached via ioctl(), then the max_usage_count will
2305  *    be 1, as the process calling ioctl() has got to have the device open.
2306  *    If we get here via sysfs, then the max usage count will be zero.
2307 */
2308 static int deregister_disk(ctlr_info_t *h, int drv_index,
2309                            int clear_all, int via_ioctl)
2310 {
2311         int i;
2312         struct gendisk *disk;
2313         drive_info_struct *drv;
2314         int recalculate_highest_lun;
2315
2316         if (!capable(CAP_SYS_RAWIO))
2317                 return -EPERM;
2318
2319         drv = h->drv[drv_index];
2320         disk = h->gendisk[drv_index];
2321
2322         /* make sure logical volume is NOT is use */
2323         if (clear_all || (h->gendisk[0] == disk)) {
2324                 if (drv->usage_count > via_ioctl)
2325                         return -EBUSY;
2326         } else if (drv->usage_count > 0)
2327                 return -EBUSY;
2328
2329         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2330
2331         /* invalidate the devices and deregister the disk.  If it is disk
2332          * zero do not deregister it but just zero out it's values.  This
2333          * allows us to delete disk zero but keep the controller registered.
2334          */
2335         if (h->gendisk[0] != disk) {
2336                 struct request_queue *q = disk->queue;
2337                 if (disk->flags & GENHD_FL_UP) {
2338                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2339                         del_gendisk(disk);
2340                 }
2341                 if (q)
2342                         blk_cleanup_queue(q);
2343                 /* If clear_all is set then we are deleting the logical
2344                  * drive, not just refreshing its info.  For drives
2345                  * other than disk 0 we will call put_disk.  We do not
2346                  * do this for disk 0 as we need it to be able to
2347                  * configure the controller.
2348                  */
2349                 if (clear_all){
2350                         /* This isn't pretty, but we need to find the
2351                          * disk in our array and NULL our the pointer.
2352                          * This is so that we will call alloc_disk if
2353                          * this index is used again later.
2354                          */
2355                         for (i=0; i < CISS_MAX_LUN; i++){
2356                                 if (h->gendisk[i] == disk) {
2357                                         h->gendisk[i] = NULL;
2358                                         break;
2359                                 }
2360                         }
2361                         put_disk(disk);
2362                 }
2363         } else {
2364                 set_capacity(disk, 0);
2365                 cciss_clear_drive_info(drv);
2366         }
2367
2368         --h->num_luns;
2369
2370         /* if it was the last disk, find the new hightest lun */
2371         if (clear_all && recalculate_highest_lun) {
2372                 int i, newhighest = -1;
2373                 for (i = 0; i <= h->highest_lun; i++) {
2374                         /* if the disk has size > 0, it is available */
2375                         if (h->drv[i] && h->drv[i]->heads)
2376                                 newhighest = i;
2377                 }
2378                 h->highest_lun = newhighest;
2379         }
2380         return 0;
2381 }
2382
2383 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2384                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2385                 int cmd_type)
2386 {
2387         ctlr_info_t *h = hba[ctlr];
2388         u64bit buff_dma_handle;
2389         int status = IO_OK;
2390
2391         c->cmd_type = CMD_IOCTL_PEND;
2392         c->Header.ReplyQueue = 0;
2393         if (buff != NULL) {
2394                 c->Header.SGList = 1;
2395                 c->Header.SGTotal = 1;
2396         } else {
2397                 c->Header.SGList = 0;
2398                 c->Header.SGTotal = 0;
2399         }
2400         c->Header.Tag.lower = c->busaddr;
2401         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2402
2403         c->Request.Type.Type = cmd_type;
2404         if (cmd_type == TYPE_CMD) {
2405                 switch (cmd) {
2406                 case CISS_INQUIRY:
2407                         /* are we trying to read a vital product page */
2408                         if (page_code != 0) {
2409                                 c->Request.CDB[1] = 0x01;
2410                                 c->Request.CDB[2] = page_code;
2411                         }
2412                         c->Request.CDBLen = 6;
2413                         c->Request.Type.Attribute = ATTR_SIMPLE;
2414                         c->Request.Type.Direction = XFER_READ;
2415                         c->Request.Timeout = 0;
2416                         c->Request.CDB[0] = CISS_INQUIRY;
2417                         c->Request.CDB[4] = size & 0xFF;
2418                         break;
2419                 case CISS_REPORT_LOG:
2420                 case CISS_REPORT_PHYS:
2421                         /* Talking to controller so It's a physical command
2422                            mode = 00 target = 0.  Nothing to write.
2423                          */
2424                         c->Request.CDBLen = 12;
2425                         c->Request.Type.Attribute = ATTR_SIMPLE;
2426                         c->Request.Type.Direction = XFER_READ;
2427                         c->Request.Timeout = 0;
2428                         c->Request.CDB[0] = cmd;
2429                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2430                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2431                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2432                         c->Request.CDB[9] = size & 0xFF;
2433                         break;
2434
2435                 case CCISS_READ_CAPACITY:
2436                         c->Request.CDBLen = 10;
2437                         c->Request.Type.Attribute = ATTR_SIMPLE;
2438                         c->Request.Type.Direction = XFER_READ;
2439                         c->Request.Timeout = 0;
2440                         c->Request.CDB[0] = cmd;
2441                         break;
2442                 case CCISS_READ_CAPACITY_16:
2443                         c->Request.CDBLen = 16;
2444                         c->Request.Type.Attribute = ATTR_SIMPLE;
2445                         c->Request.Type.Direction = XFER_READ;
2446                         c->Request.Timeout = 0;
2447                         c->Request.CDB[0] = cmd;
2448                         c->Request.CDB[1] = 0x10;
2449                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2450                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2451                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2452                         c->Request.CDB[13] = size & 0xFF;
2453                         c->Request.Timeout = 0;
2454                         c->Request.CDB[0] = cmd;
2455                         break;
2456                 case CCISS_CACHE_FLUSH:
2457                         c->Request.CDBLen = 12;
2458                         c->Request.Type.Attribute = ATTR_SIMPLE;
2459                         c->Request.Type.Direction = XFER_WRITE;
2460                         c->Request.Timeout = 0;
2461                         c->Request.CDB[0] = BMIC_WRITE;
2462                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2463                         break;
2464                 case TEST_UNIT_READY:
2465                         c->Request.CDBLen = 6;
2466                         c->Request.Type.Attribute = ATTR_SIMPLE;
2467                         c->Request.Type.Direction = XFER_NONE;
2468                         c->Request.Timeout = 0;
2469                         break;
2470                 default:
2471                         printk(KERN_WARNING
2472                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2473                         return IO_ERROR;
2474                 }
2475         } else if (cmd_type == TYPE_MSG) {
2476                 switch (cmd) {
2477                 case 0: /* ABORT message */
2478                         c->Request.CDBLen = 12;
2479                         c->Request.Type.Attribute = ATTR_SIMPLE;
2480                         c->Request.Type.Direction = XFER_WRITE;
2481                         c->Request.Timeout = 0;
2482                         c->Request.CDB[0] = cmd;        /* abort */
2483                         c->Request.CDB[1] = 0;  /* abort a command */
2484                         /* buff contains the tag of the command to abort */
2485                         memcpy(&c->Request.CDB[4], buff, 8);
2486                         break;
2487                 case 1: /* RESET message */
2488                         c->Request.CDBLen = 16;
2489                         c->Request.Type.Attribute = ATTR_SIMPLE;
2490                         c->Request.Type.Direction = XFER_NONE;
2491                         c->Request.Timeout = 0;
2492                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2493                         c->Request.CDB[0] = cmd;        /* reset */
2494                         c->Request.CDB[1] = 0x03;       /* reset a target */
2495                         break;
2496                 case 3: /* No-Op message */
2497                         c->Request.CDBLen = 1;
2498                         c->Request.Type.Attribute = ATTR_SIMPLE;
2499                         c->Request.Type.Direction = XFER_WRITE;
2500                         c->Request.Timeout = 0;
2501                         c->Request.CDB[0] = cmd;
2502                         break;
2503                 default:
2504                         printk(KERN_WARNING
2505                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2506                         return IO_ERROR;
2507                 }
2508         } else {
2509                 printk(KERN_WARNING
2510                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2511                 return IO_ERROR;
2512         }
2513         /* Fill in the scatter gather information */
2514         if (size > 0) {
2515                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2516                                                              buff, size,
2517                                                              PCI_DMA_BIDIRECTIONAL);
2518                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2519                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2520                 c->SG[0].Len = size;
2521                 c->SG[0].Ext = 0;       /* we are not chaining */
2522         }
2523         return status;
2524 }
2525
2526 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2527 {
2528         switch (c->err_info->ScsiStatus) {
2529         case SAM_STAT_GOOD:
2530                 return IO_OK;
2531         case SAM_STAT_CHECK_CONDITION:
2532                 switch (0xf & c->err_info->SenseInfo[2]) {
2533                 case 0: return IO_OK; /* no sense */
2534                 case 1: return IO_OK; /* recovered error */
2535                 default:
2536                         if (check_for_unit_attention(h, c))
2537                                 return IO_NEEDS_RETRY;
2538                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2539                                 "check condition, sense key = 0x%02x\n",
2540                                 h->ctlr, c->Request.CDB[0],
2541                                 c->err_info->SenseInfo[2]);
2542                 }
2543                 break;
2544         default:
2545                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2546                         "scsi status = 0x%02x\n", h->ctlr,
2547                         c->Request.CDB[0], c->err_info->ScsiStatus);
2548                 break;
2549         }
2550         return IO_ERROR;
2551 }
2552
2553 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2554 {
2555         int return_status = IO_OK;
2556
2557         if (c->err_info->CommandStatus == CMD_SUCCESS)
2558                 return IO_OK;
2559
2560         switch (c->err_info->CommandStatus) {
2561         case CMD_TARGET_STATUS:
2562                 return_status = check_target_status(h, c);
2563                 break;
2564         case CMD_DATA_UNDERRUN:
2565         case CMD_DATA_OVERRUN:
2566                 /* expected for inquiry and report lun commands */
2567                 break;
2568         case CMD_INVALID:
2569                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2570                        "reported invalid\n", c->Request.CDB[0]);
2571                 return_status = IO_ERROR;
2572                 break;
2573         case CMD_PROTOCOL_ERR:
2574                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2575                        "protocol error \n", c->Request.CDB[0]);
2576                 return_status = IO_ERROR;
2577                 break;
2578         case CMD_HARDWARE_ERR:
2579                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2580                        " hardware error\n", c->Request.CDB[0]);
2581                 return_status = IO_ERROR;
2582                 break;
2583         case CMD_CONNECTION_LOST:
2584                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2585                        "connection lost\n", c->Request.CDB[0]);
2586                 return_status = IO_ERROR;
2587                 break;
2588         case CMD_ABORTED:
2589                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2590                        "aborted\n", c->Request.CDB[0]);
2591                 return_status = IO_ERROR;
2592                 break;
2593         case CMD_ABORT_FAILED:
2594                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2595                        "abort failed\n", c->Request.CDB[0]);
2596                 return_status = IO_ERROR;
2597                 break;
2598         case CMD_UNSOLICITED_ABORT:
2599                 printk(KERN_WARNING
2600                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2601                         c->Request.CDB[0]);
2602                 return_status = IO_NEEDS_RETRY;
2603                 break;
2604         default:
2605                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2606                        "unknown status %x\n", c->Request.CDB[0],
2607                        c->err_info->CommandStatus);
2608                 return_status = IO_ERROR;
2609         }
2610         return return_status;
2611 }
2612
2613 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2614         int attempt_retry)
2615 {
2616         DECLARE_COMPLETION_ONSTACK(wait);
2617         u64bit buff_dma_handle;
2618         unsigned long flags;
2619         int return_status = IO_OK;
2620
2621 resend_cmd2:
2622         c->waiting = &wait;
2623         /* Put the request on the tail of the queue and send it */
2624         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2625         addQ(&h->reqQ, c);
2626         h->Qdepth++;
2627         start_io(h);
2628         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2629
2630         wait_for_completion(&wait);
2631
2632         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2633                 goto command_done;
2634
2635         return_status = process_sendcmd_error(h, c);
2636
2637         if (return_status == IO_NEEDS_RETRY &&
2638                 c->retry_count < MAX_CMD_RETRIES) {
2639                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2640                         c->Request.CDB[0]);
2641                 c->retry_count++;
2642                 /* erase the old error information */
2643                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2644                 return_status = IO_OK;
2645                 INIT_COMPLETION(wait);
2646                 goto resend_cmd2;
2647         }
2648
2649 command_done:
2650         /* unlock the buffers from DMA */
2651         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2652         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2653         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2654                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2655         return return_status;
2656 }
2657
2658 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2659                            __u8 page_code, unsigned char scsi3addr[],
2660                         int cmd_type)
2661 {
2662         ctlr_info_t *h = hba[ctlr];
2663         CommandList_struct *c;
2664         int return_status;
2665
2666         c = cmd_alloc(h, 0);
2667         if (!c)
2668                 return -ENOMEM;
2669         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2670                 scsi3addr, cmd_type);
2671         if (return_status == IO_OK)
2672                 return_status = sendcmd_withirq_core(h, c, 1);
2673
2674         cmd_free(h, c, 0);
2675         return return_status;
2676 }
2677
2678 static void cciss_geometry_inquiry(int ctlr, int logvol,
2679                                    sector_t total_size,
2680                                    unsigned int block_size,
2681                                    InquiryData_struct *inq_buff,
2682                                    drive_info_struct *drv)
2683 {
2684         int return_code;
2685         unsigned long t;
2686         unsigned char scsi3addr[8];
2687
2688         memset(inq_buff, 0, sizeof(InquiryData_struct));
2689         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2690         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2691                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2692         if (return_code == IO_OK) {
2693                 if (inq_buff->data_byte[8] == 0xFF) {
2694                         printk(KERN_WARNING
2695                                "cciss: reading geometry failed, volume "
2696                                "does not support reading geometry\n");
2697                         drv->heads = 255;
2698                         drv->sectors = 32;      /* Sectors per track */
2699                         drv->cylinders = total_size + 1;
2700                         drv->raid_level = RAID_UNKNOWN;
2701                 } else {
2702                         drv->heads = inq_buff->data_byte[6];
2703                         drv->sectors = inq_buff->data_byte[7];
2704                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2705                         drv->cylinders += inq_buff->data_byte[5];
2706                         drv->raid_level = inq_buff->data_byte[8];
2707                 }
2708                 drv->block_size = block_size;
2709                 drv->nr_blocks = total_size + 1;
2710                 t = drv->heads * drv->sectors;
2711                 if (t > 1) {
2712                         sector_t real_size = total_size + 1;
2713                         unsigned long rem = sector_div(real_size, t);
2714                         if (rem)
2715                                 real_size++;
2716                         drv->cylinders = real_size;
2717                 }
2718         } else {                /* Get geometry failed */
2719                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2720         }
2721 }
2722
2723 static void
2724 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2725                     unsigned int *block_size)
2726 {
2727         ReadCapdata_struct *buf;
2728         int return_code;
2729         unsigned char scsi3addr[8];
2730
2731         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2732         if (!buf) {
2733                 printk(KERN_WARNING "cciss: out of memory\n");
2734                 return;
2735         }
2736
2737         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2738         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2739                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2740         if (return_code == IO_OK) {
2741                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2742                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2743         } else {                /* read capacity command failed */
2744                 printk(KERN_WARNING "cciss: read capacity failed\n");
2745                 *total_size = 0;
2746                 *block_size = BLOCK_SIZE;
2747         }
2748         kfree(buf);
2749 }
2750
2751 static void cciss_read_capacity_16(int ctlr, int logvol,
2752         sector_t *total_size, unsigned int *block_size)
2753 {
2754         ReadCapdata_struct_16 *buf;
2755         int return_code;
2756         unsigned char scsi3addr[8];
2757
2758         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2759         if (!buf) {
2760                 printk(KERN_WARNING "cciss: out of memory\n");
2761                 return;
2762         }
2763
2764         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2765         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2766                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2767                         0, scsi3addr, TYPE_CMD);
2768         if (return_code == IO_OK) {
2769                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2770                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2771         } else {                /* read capacity command failed */
2772                 printk(KERN_WARNING "cciss: read capacity failed\n");
2773                 *total_size = 0;
2774                 *block_size = BLOCK_SIZE;
2775         }
2776         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2777                (unsigned long long)*total_size+1, *block_size);
2778         kfree(buf);
2779 }
2780
2781 static int cciss_revalidate(struct gendisk *disk)
2782 {
2783         ctlr_info_t *h = get_host(disk);
2784         drive_info_struct *drv = get_drv(disk);
2785         int logvol;
2786         int FOUND = 0;
2787         unsigned int block_size;
2788         sector_t total_size;
2789         InquiryData_struct *inq_buff = NULL;
2790
2791         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2792                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2793                         sizeof(drv->LunID)) == 0) {
2794                         FOUND = 1;
2795                         break;
2796                 }
2797         }
2798
2799         if (!FOUND)
2800                 return 1;
2801
2802         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2803         if (inq_buff == NULL) {
2804                 printk(KERN_WARNING "cciss: out of memory\n");
2805                 return 1;
2806         }
2807         if (h->cciss_read == CCISS_READ_10) {
2808                 cciss_read_capacity(h->ctlr, logvol,
2809                                         &total_size, &block_size);
2810         } else {
2811                 cciss_read_capacity_16(h->ctlr, logvol,
2812                                         &total_size, &block_size);
2813         }
2814         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2815                                inq_buff, drv);
2816
2817         blk_queue_logical_block_size(drv->queue, drv->block_size);
2818         set_capacity(disk, drv->nr_blocks);
2819
2820         kfree(inq_buff);
2821         return 0;
2822 }
2823
2824 /*
2825  * Map (physical) PCI mem into (virtual) kernel space
2826  */
2827 static void __iomem *remap_pci_mem(ulong base, ulong size)
2828 {
2829         ulong page_base = ((ulong) base) & PAGE_MASK;
2830         ulong page_offs = ((ulong) base) - page_base;
2831         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2832
2833         return page_remapped ? (page_remapped + page_offs) : NULL;
2834 }
2835
2836 /*
2837  * Takes jobs of the Q and sends them to the hardware, then puts it on
2838  * the Q to wait for completion.
2839  */
2840 static void start_io(ctlr_info_t *h)
2841 {
2842         CommandList_struct *c;
2843
2844         while (!hlist_empty(&h->reqQ)) {
2845                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2846                 /* can't do anything if fifo is full */
2847                 if ((h->access.fifo_full(h))) {
2848                         printk(KERN_WARNING "cciss: fifo full\n");
2849                         break;
2850                 }
2851
2852                 /* Get the first entry from the Request Q */
2853                 removeQ(c);
2854                 h->Qdepth--;
2855
2856                 /* Tell the controller execute command */
2857                 h->access.submit_command(h, c);
2858
2859                 /* Put job onto the completed Q */
2860                 addQ(&h->cmpQ, c);
2861         }
2862 }
2863
2864 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2865 /* Zeros out the error record and then resends the command back */
2866 /* to the controller */
2867 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2868 {
2869         /* erase the old error information */
2870         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2871
2872         /* add it to software queue and then send it to the controller */
2873         addQ(&h->reqQ, c);
2874         h->Qdepth++;
2875         if (h->Qdepth > h->maxQsinceinit)
2876                 h->maxQsinceinit = h->Qdepth;
2877
2878         start_io(h);
2879 }
2880
2881 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2882         unsigned int msg_byte, unsigned int host_byte,
2883         unsigned int driver_byte)
2884 {
2885         /* inverse of macros in scsi.h */
2886         return (scsi_status_byte & 0xff) |
2887                 ((msg_byte & 0xff) << 8) |
2888                 ((host_byte & 0xff) << 16) |
2889                 ((driver_byte & 0xff) << 24);
2890 }
2891
2892 static inline int evaluate_target_status(ctlr_info_t *h,
2893                         CommandList_struct *cmd, int *retry_cmd)
2894 {
2895         unsigned char sense_key;
2896         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2897         int error_value;
2898
2899         *retry_cmd = 0;
2900         /* If we get in here, it means we got "target status", that is, scsi status */
2901         status_byte = cmd->err_info->ScsiStatus;
2902         driver_byte = DRIVER_OK;
2903         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2904
2905         if (blk_pc_request(cmd->rq))
2906                 host_byte = DID_PASSTHROUGH;
2907         else
2908                 host_byte = DID_OK;
2909
2910         error_value = make_status_bytes(status_byte, msg_byte,
2911                 host_byte, driver_byte);
2912
2913         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2914                 if (!blk_pc_request(cmd->rq))
2915                         printk(KERN_WARNING "cciss: cmd %p "
2916                                "has SCSI Status 0x%x\n",
2917                                cmd, cmd->err_info->ScsiStatus);
2918                 return error_value;
2919         }
2920
2921         /* check the sense key */
2922         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2923         /* no status or recovered error */
2924         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2925                 error_value = 0;
2926
2927         if (check_for_unit_attention(h, cmd)) {
2928                 *retry_cmd = !blk_pc_request(cmd->rq);
2929                 return 0;
2930         }
2931
2932         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2933                 if (error_value != 0)
2934                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2935                                " sense key = 0x%x\n", cmd, sense_key);
2936                 return error_value;
2937         }
2938
2939         /* SG_IO or similar, copy sense data back */
2940         if (cmd->rq->sense) {
2941                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2942                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2943                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2944                         cmd->rq->sense_len);
2945         } else
2946                 cmd->rq->sense_len = 0;
2947
2948         return error_value;
2949 }
2950
2951 /* checks the status of the job and calls complete buffers to mark all
2952  * buffers for the completed job. Note that this function does not need
2953  * to hold the hba/queue lock.
2954  */
2955 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2956                                     int timeout)
2957 {
2958         int retry_cmd = 0;
2959         struct request *rq = cmd->rq;
2960
2961         rq->errors = 0;
2962
2963         if (timeout)
2964                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
2965
2966         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
2967                 goto after_error_processing;
2968
2969         switch (cmd->err_info->CommandStatus) {
2970         case CMD_TARGET_STATUS:
2971                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
2972                 break;
2973         case CMD_DATA_UNDERRUN:
2974                 if (blk_fs_request(cmd->rq)) {
2975                         printk(KERN_WARNING "cciss: cmd %p has"
2976                                " completed with data underrun "
2977                                "reported\n", cmd);
2978                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
2979                 }
2980                 break;
2981         case CMD_DATA_OVERRUN:
2982                 if (blk_fs_request(cmd->rq))
2983                         printk(KERN_WARNING "cciss: cmd %p has"
2984                                " completed with data overrun "
2985                                "reported\n", cmd);
2986                 break;
2987         case CMD_INVALID:
2988                 printk(KERN_WARNING "cciss: cmd %p is "
2989                        "reported invalid\n", cmd);
2990                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2991                         cmd->err_info->CommandStatus, DRIVER_OK,
2992                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2993                 break;
2994         case CMD_PROTOCOL_ERR:
2995                 printk(KERN_WARNING "cciss: cmd %p has "
2996                        "protocol error \n", cmd);
2997                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2998                         cmd->err_info->CommandStatus, DRIVER_OK,
2999                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3000                 break;
3001         case CMD_HARDWARE_ERR:
3002                 printk(KERN_WARNING "cciss: cmd %p had "
3003                        " hardware error\n", cmd);
3004                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3005                         cmd->err_info->CommandStatus, DRIVER_OK,
3006                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3007                 break;
3008         case CMD_CONNECTION_LOST:
3009                 printk(KERN_WARNING "cciss: cmd %p had "
3010                        "connection lost\n", cmd);
3011                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3012                         cmd->err_info->CommandStatus, DRIVER_OK,
3013                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3014                 break;
3015         case CMD_ABORTED:
3016                 printk(KERN_WARNING "cciss: cmd %p was "
3017                        "aborted\n", cmd);
3018                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3019                         cmd->err_info->CommandStatus, DRIVER_OK,
3020                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3021                 break;
3022         case CMD_ABORT_FAILED:
3023                 printk(KERN_WARNING "cciss: cmd %p reports "
3024                        "abort failed\n", cmd);
3025                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3026                         cmd->err_info->CommandStatus, DRIVER_OK,
3027                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3028                 break;
3029         case CMD_UNSOLICITED_ABORT:
3030                 printk(KERN_WARNING "cciss%d: unsolicited "
3031                        "abort %p\n", h->ctlr, cmd);
3032                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3033                         retry_cmd = 1;
3034                         printk(KERN_WARNING
3035                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3036                         cmd->retry_count++;
3037                 } else
3038                         printk(KERN_WARNING
3039                                "cciss%d: %p retried too "
3040                                "many times\n", h->ctlr, cmd);
3041                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3042                         cmd->err_info->CommandStatus, DRIVER_OK,
3043                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3044                 break;
3045         case CMD_TIMEOUT:
3046                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3047                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3048                         cmd->err_info->CommandStatus, DRIVER_OK,
3049                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3050                 break;
3051         default:
3052                 printk(KERN_WARNING "cciss: cmd %p returned "
3053                        "unknown status %x\n", cmd,
3054                        cmd->err_info->CommandStatus);
3055                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3056                         cmd->err_info->CommandStatus, DRIVER_OK,
3057                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3058         }
3059
3060 after_error_processing:
3061
3062         /* We need to return this command */
3063         if (retry_cmd) {
3064                 resend_cciss_cmd(h, cmd);
3065                 return;
3066         }
3067         cmd->rq->completion_data = cmd;
3068         blk_complete_request(cmd->rq);
3069 }
3070
3071 /*
3072  * Get a request and submit it to the controller.
3073  */
3074 static void do_cciss_request(struct request_queue *q)
3075 {
3076         ctlr_info_t *h = q->queuedata;
3077         CommandList_struct *c;
3078         sector_t start_blk;
3079         int seg;
3080         struct request *creq;
3081         u64bit temp64;
3082         struct scatterlist *tmp_sg;
3083         SGDescriptor_struct *curr_sg;
3084         drive_info_struct *drv;
3085         int i, dir;
3086         int nseg = 0;
3087         int sg_index = 0;
3088         int chained = 0;
3089
3090         /* We call start_io here in case there is a command waiting on the
3091          * queue that has not been sent.
3092          */
3093         if (blk_queue_plugged(q))
3094                 goto startio;
3095
3096       queue:
3097         creq = blk_peek_request(q);
3098         if (!creq)
3099                 goto startio;
3100
3101         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3102
3103         if ((c = cmd_alloc(h, 1)) == NULL)
3104                 goto full;
3105
3106         blk_start_request(creq);
3107
3108         tmp_sg = h->scatter_list[c->cmdindex];
3109         spin_unlock_irq(q->queue_lock);
3110
3111         c->cmd_type = CMD_RWREQ;
3112         c->rq = creq;
3113
3114         /* fill in the request */
3115         drv = creq->rq_disk->private_data;
3116         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3117         /* got command from pool, so use the command block index instead */
3118         /* for direct lookups. */
3119         /* The first 2 bits are reserved for controller error reporting. */
3120         c->Header.Tag.lower = (c->cmdindex << 3);
3121         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
3122         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3123         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3124         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3125         c->Request.Type.Attribute = ATTR_SIMPLE;
3126         c->Request.Type.Direction =
3127             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3128         c->Request.Timeout = 0; /* Don't time out */
3129         c->Request.CDB[0] =
3130             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3131         start_blk = blk_rq_pos(creq);
3132 #ifdef CCISS_DEBUG
3133         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3134                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3135 #endif                          /* CCISS_DEBUG */
3136
3137         sg_init_table(tmp_sg, h->maxsgentries);
3138         seg = blk_rq_map_sg(q, creq, tmp_sg);
3139
3140         /* get the DMA records for the setup */
3141         if (c->Request.Type.Direction == XFER_READ)
3142                 dir = PCI_DMA_FROMDEVICE;
3143         else
3144                 dir = PCI_DMA_TODEVICE;
3145
3146         curr_sg = c->SG;
3147         sg_index = 0;
3148         chained = 0;
3149
3150         for (i = 0; i < seg; i++) {
3151                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3152                         !chained && ((seg - i) > 1)) {
3153                         nseg = seg - i;
3154                         curr_sg[sg_index].Len = (nseg) *
3155                                         sizeof(SGDescriptor_struct);
3156                         curr_sg[sg_index].Ext = CCISS_SG_CHAIN;
3157
3158                         /* Point to next chain block. */
3159                         curr_sg = h->cmd_sg_list[c->cmdindex]->sgchain;
3160                         sg_index = 0;
3161                         chained = 1;
3162                 }
3163                 curr_sg[sg_index].Len = tmp_sg[i].length;
3164                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3165                                                 tmp_sg[i].offset,
3166                                                 tmp_sg[i].length, dir);
3167                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3168                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3169                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3170
3171                 ++sg_index;
3172         }
3173
3174         if (chained) {
3175                 int len;
3176                 curr_sg = c->SG;
3177                 sg_index = h->max_cmd_sgentries - 1;
3178                 len = curr_sg[sg_index].Len;
3179                 /* Setup pointer to next chain block.
3180                  * Fill out last element in current chain
3181                  * block with address of next chain block.
3182                  */
3183                 temp64.val = pci_map_single(h->pdev,
3184                                         h->cmd_sg_list[c->cmdindex]->sgchain,
3185                                         len, dir);
3186
3187                 h->cmd_sg_list[c->cmdindex]->sg_chain_dma = temp64.val;
3188                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3189                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3190
3191                 pci_dma_sync_single_for_device(h->pdev,
3192                                 h->cmd_sg_list[c->cmdindex]->sg_chain_dma,
3193                                 len, dir);
3194         }
3195
3196         /* track how many SG entries we are using */
3197         if (seg > h->maxSG)
3198                 h->maxSG = seg;
3199
3200 #ifdef CCISS_DEBUG
3201         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3202                         "chained[%d]\n",
3203                         blk_rq_sectors(creq), seg, chained);
3204 #endif                          /* CCISS_DEBUG */
3205
3206         c->Header.SGList = c->Header.SGTotal = seg + chained;
3207         if (seg > h->max_cmd_sgentries)
3208                 c->Header.SGList = h->max_cmd_sgentries;
3209
3210         if (likely(blk_fs_request(creq))) {
3211                 if(h->cciss_read == CCISS_READ_10) {
3212                         c->Request.CDB[1] = 0;
3213                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3214                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3215                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3216                         c->Request.CDB[5] = start_blk & 0xff;
3217                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3218                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3219                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3220                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3221                 } else {
3222                         u32 upper32 = upper_32_bits(start_blk);
3223
3224                         c->Request.CDBLen = 16;
3225                         c->Request.CDB[1]= 0;
3226                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3227                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3228                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3229                         c->Request.CDB[5]= upper32 & 0xff;
3230                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3231                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3232                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3233                         c->Request.CDB[9]= start_blk & 0xff;
3234                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3235                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3236                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3237                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3238                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3239                 }
3240         } else if (blk_pc_request(creq)) {
3241                 c->Request.CDBLen = creq->cmd_len;
3242                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3243         } else {
3244                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3245                 BUG();
3246         }
3247
3248         spin_lock_irq(q->queue_lock);
3249
3250         addQ(&h->reqQ, c);
3251         h->Qdepth++;
3252         if (h->Qdepth > h->maxQsinceinit)
3253                 h->maxQsinceinit = h->Qdepth;
3254
3255         goto queue;
3256 full:
3257         blk_stop_queue(q);
3258 startio:
3259         /* We will already have the driver lock here so not need
3260          * to lock it.
3261          */
3262         start_io(h);
3263 }
3264
3265 static inline unsigned long get_next_completion(ctlr_info_t *h)
3266 {
3267         return h->access.command_completed(h);
3268 }
3269
3270 static inline int interrupt_pending(ctlr_info_t *h)
3271 {
3272         return h->access.intr_pending(h);
3273 }
3274
3275 static inline long interrupt_not_for_us(ctlr_info_t *h)
3276 {
3277         return (((h->access.intr_pending(h) == 0) ||
3278                  (h->interrupts_enabled == 0)));
3279 }
3280
3281 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3282 {
3283         ctlr_info_t *h = dev_id;
3284         CommandList_struct *c;
3285         unsigned long flags;
3286         __u32 a, a1, a2;
3287
3288         if (interrupt_not_for_us(h))
3289                 return IRQ_NONE;
3290         /*
3291          * If there are completed commands in the completion queue,
3292          * we had better do something about it.
3293          */
3294         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3295         while (interrupt_pending(h)) {
3296                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3297                         a1 = a;
3298                         if ((a & 0x04)) {
3299                                 a2 = (a >> 3);
3300                                 if (a2 >= h->nr_cmds) {
3301                                         printk(KERN_WARNING
3302                                                "cciss: controller cciss%d failed, stopping.\n",
3303                                                h->ctlr);
3304                                         fail_all_cmds(h->ctlr);
3305                                         return IRQ_HANDLED;
3306                                 }
3307
3308                                 c = h->cmd_pool + a2;
3309                                 a = c->busaddr;
3310
3311                         } else {
3312                                 struct hlist_node *tmp;
3313
3314                                 a &= ~3;
3315                                 c = NULL;
3316                                 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3317                                         if (c->busaddr == a)
3318                                                 break;
3319                                 }
3320                         }
3321                         /*
3322                          * If we've found the command, take it off the
3323                          * completion Q and free it
3324                          */
3325                         if (c && c->busaddr == a) {
3326                                 removeQ(c);
3327                                 if (c->cmd_type == CMD_RWREQ) {
3328                                         complete_command(h, c, 0);
3329                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3330                                         complete(c->waiting);
3331                                 }
3332 #                               ifdef CONFIG_CISS_SCSI_TAPE
3333                                 else if (c->cmd_type == CMD_SCSI)
3334                                         complete_scsi_command(c, 0, a1);
3335 #                               endif
3336                                 continue;
3337                         }
3338                 }
3339         }
3340
3341         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3342         return IRQ_HANDLED;
3343 }
3344
3345 /**
3346  * add_to_scan_list() - add controller to rescan queue
3347  * @h:                Pointer to the controller.
3348  *
3349  * Adds the controller to the rescan queue if not already on the queue.
3350  *
3351  * returns 1 if added to the queue, 0 if skipped (could be on the
3352  * queue already, or the controller could be initializing or shutting
3353  * down).
3354  **/
3355 static int add_to_scan_list(struct ctlr_info *h)
3356 {
3357         struct ctlr_info *test_h;
3358         int found = 0;
3359         int ret = 0;
3360
3361         if (h->busy_initializing)
3362                 return 0;
3363
3364         if (!mutex_trylock(&h->busy_shutting_down))
3365                 return 0;
3366
3367         mutex_lock(&scan_mutex);
3368         list_for_each_entry(test_h, &scan_q, scan_list) {
3369                 if (test_h == h) {
3370                         found = 1;
3371                         break;
3372                 }
3373         }
3374         if (!found && !h->busy_scanning) {
3375                 INIT_COMPLETION(h->scan_wait);
3376                 list_add_tail(&h->scan_list, &scan_q);
3377                 ret = 1;
3378         }
3379         mutex_unlock(&scan_mutex);
3380         mutex_unlock(&h->busy_shutting_down);
3381
3382         return ret;
3383 }
3384
3385 /**
3386  * remove_from_scan_list() - remove controller from rescan queue
3387  * @h:                     Pointer to the controller.
3388  *
3389  * Removes the controller from the rescan queue if present. Blocks if
3390  * the controller is currently conducting a rescan.  The controller
3391  * can be in one of three states:
3392  * 1. Doesn't need a scan
3393  * 2. On the scan list, but not scanning yet (we remove it)
3394  * 3. Busy scanning (and not on the list). In this case we want to wait for
3395  *    the scan to complete to make sure the scanning thread for this
3396  *    controller is completely idle.
3397  **/
3398 static void remove_from_scan_list(struct ctlr_info *h)
3399 {
3400         struct ctlr_info *test_h, *tmp_h;
3401
3402         mutex_lock(&scan_mutex);
3403         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3404                 if (test_h == h) { /* state 2. */
3405                         list_del(&h->scan_list);
3406                         complete_all(&h->scan_wait);
3407                         mutex_unlock(&scan_mutex);
3408                         return;
3409                 }
3410         }
3411         if (h->busy_scanning) { /* state 3. */
3412                 mutex_unlock(&scan_mutex);
3413                 wait_for_completion(&h->scan_wait);
3414         } else { /* state 1, nothing to do. */
3415                 mutex_unlock(&scan_mutex);
3416         }
3417 }
3418
3419 /**
3420  * scan_thread() - kernel thread used to rescan controllers
3421  * @data:        Ignored.
3422  *
3423  * A kernel thread used scan for drive topology changes on
3424  * controllers. The thread processes only one controller at a time
3425  * using a queue.  Controllers are added to the queue using
3426  * add_to_scan_list() and removed from the queue either after done
3427  * processing or using remove_from_scan_list().
3428  *
3429  * returns 0.
3430  **/
3431 static int scan_thread(void *data)
3432 {
3433         struct ctlr_info *h;
3434
3435         while (1) {
3436                 set_current_state(TASK_INTERRUPTIBLE);
3437                 schedule();
3438                 if (kthread_should_stop())
3439                         break;
3440
3441                 while (1) {
3442                         mutex_lock(&scan_mutex);
3443                         if (list_empty(&scan_q)) {
3444                                 mutex_unlock(&scan_mutex);
3445                                 break;
3446                         }
3447
3448                         h = list_entry(scan_q.next,
3449                                        struct ctlr_info,
3450                                        scan_list);
3451                         list_del(&h->scan_list);
3452                         h->busy_scanning = 1;
3453                         mutex_unlock(&scan_mutex);
3454
3455                         rebuild_lun_table(h, 0, 0);
3456                         complete_all(&h->scan_wait);
3457                         mutex_lock(&scan_mutex);
3458                         h->busy_scanning = 0;
3459                         mutex_unlock(&scan_mutex);
3460                 }
3461         }
3462
3463         return 0;
3464 }
3465
3466 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3467 {
3468         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3469                 return 0;
3470
3471         switch (c->err_info->SenseInfo[12]) {
3472         case STATE_CHANGED:
3473                 printk(KERN_WARNING "cciss%d: a state change "
3474                         "detected, command retried\n", h->ctlr);
3475                 return 1;
3476         break;
3477         case LUN_FAILED:
3478                 printk(KERN_WARNING "cciss%d: LUN failure "
3479                         "detected, action required\n", h->ctlr);
3480                 return 1;
3481         break;
3482         case REPORT_LUNS_CHANGED:
3483                 printk(KERN_WARNING "cciss%d: report LUN data "
3484                         "changed\n", h->ctlr);
3485         /*
3486          * Here, we could call add_to_scan_list and wake up the scan thread,
3487          * except that it's quite likely that we will get more than one
3488          * REPORT_LUNS_CHANGED condition in quick succession, which means
3489          * that those which occur after the first one will likely happen
3490          * *during* the scan_thread's rescan.  And the rescan code is not
3491          * robust enough to restart in the middle, undoing what it has already
3492          * done, and it's not clear that it's even possible to do this, since
3493          * part of what it does is notify the block layer, which starts
3494          * doing it's own i/o to read partition tables and so on, and the
3495          * driver doesn't have visibility to know what might need undoing.
3496          * In any event, if possible, it is horribly complicated to get right
3497          * so we just don't do it for now.
3498          *
3499          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3500          */
3501                 return 1;
3502         break;
3503         case POWER_OR_RESET:
3504                 printk(KERN_WARNING "cciss%d: a power on "
3505                         "or device reset detected\n", h->ctlr);
3506                 return 1;
3507         break;
3508         case UNIT_ATTENTION_CLEARED:
3509                 printk(KERN_WARNING "cciss%d: unit attention "
3510                     "cleared by another initiator\n", h->ctlr);
3511                 return 1;
3512         break;
3513         default:
3514                 printk(KERN_WARNING "cciss%d: unknown "
3515                         "unit attention detected\n", h->ctlr);
3516                                 return 1;
3517         }
3518 }
3519
3520 /*
3521  *  We cannot read the structure directly, for portability we must use
3522  *   the io functions.
3523  *   This is for debug only.
3524  */
3525 #ifdef CCISS_DEBUG
3526 static void print_cfg_table(CfgTable_struct *tb)
3527 {
3528         int i;
3529         char temp_name[17];
3530
3531         printk("Controller Configuration information\n");
3532         printk("------------------------------------\n");
3533         for (i = 0; i < 4; i++)
3534                 temp_name[i] = readb(&(tb->Signature[i]));
3535         temp_name[4] = '\0';
3536         printk("   Signature = %s\n", temp_name);
3537         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3538         printk("   Transport methods supported = 0x%x\n",
3539                readl(&(tb->TransportSupport)));
3540         printk("   Transport methods active = 0x%x\n",
3541                readl(&(tb->TransportActive)));
3542         printk("   Requested transport Method = 0x%x\n",
3543                readl(&(tb->HostWrite.TransportRequest)));
3544         printk("   Coalesce Interrupt Delay = 0x%x\n",
3545                readl(&(tb->HostWrite.CoalIntDelay)));
3546         printk("   Coalesce Interrupt Count = 0x%x\n",
3547                readl(&(tb->HostWrite.CoalIntCount)));
3548         printk("   Max outstanding commands = 0x%d\n",
3549                readl(&(tb->CmdsOutMax)));
3550         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3551         for (i = 0; i < 16; i++)
3552                 temp_name[i] = readb(&(tb->ServerName[i]));
3553         temp_name[16] = '\0';
3554         printk("   Server Name = %s\n", temp_name);
3555         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3556 }
3557 #endif                          /* CCISS_DEBUG */
3558
3559 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3560 {
3561         int i, offset, mem_type, bar_type;
3562         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3563                 return 0;
3564         offset = 0;
3565         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3566                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3567                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3568                         offset += 4;
3569                 else {
3570                         mem_type = pci_resource_flags(pdev, i) &
3571                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3572                         switch (mem_type) {
3573                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3574                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3575                                 offset += 4;    /* 32 bit */
3576                                 break;
3577                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3578                                 offset += 8;
3579                                 break;
3580                         default:        /* reserved in PCI 2.2 */
3581                                 printk(KERN_WARNING
3582                                        "Base address is invalid\n");
3583                                 return -1;
3584                                 break;
3585                         }
3586                 }
3587                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3588                         return i + 1;
3589         }
3590         return -1;
3591 }
3592
3593 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3594  * controllers that are capable. If not, we use IO-APIC mode.
3595  */
3596
3597 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3598                                            struct pci_dev *pdev, __u32 board_id)
3599 {
3600 #ifdef CONFIG_PCI_MSI
3601         int err;
3602         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3603         {0, 2}, {0, 3}
3604         };
3605
3606         /* Some boards advertise MSI but don't really support it */
3607         if ((board_id == 0x40700E11) ||
3608             (board_id == 0x40800E11) ||
3609             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3610                 goto default_int_mode;
3611
3612         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3613                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3614                 if (!err) {
3615                         c->intr[0] = cciss_msix_entries[0].vector;
3616                         c->intr[1] = cciss_msix_entries[1].vector;
3617                         c->intr[2] = cciss_msix_entries[2].vector;
3618                         c->intr[3] = cciss_msix_entries[3].vector;
3619                         c->msix_vector = 1;
3620                         return;
3621                 }
3622                 if (err > 0) {
3623                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3624                                "available\n", err);
3625                         goto default_int_mode;
3626                 } else {
3627                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3628                                err);
3629                         goto default_int_mode;
3630                 }
3631         }
3632         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3633                 if (!pci_enable_msi(pdev)) {
3634                         c->msi_vector = 1;
3635                 } else {
3636                         printk(KERN_WARNING "cciss: MSI init failed\n");
3637                 }
3638         }
3639 default_int_mode:
3640 #endif                          /* CONFIG_PCI_MSI */
3641         /* if we get here we're going to use the default interrupt mode */
3642         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3643         return;
3644 }
3645
3646 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3647 {
3648         ushort subsystem_vendor_id, subsystem_device_id, command;
3649         __u32 board_id, scratchpad = 0;
3650         __u64 cfg_offset;
3651         __u32 cfg_base_addr;
3652         __u64 cfg_base_addr_index;
3653         int i, prod_index, err;
3654
3655         subsystem_vendor_id = pdev->subsystem_vendor;
3656         subsystem_device_id = pdev->subsystem_device;
3657         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3658                     subsystem_vendor_id);
3659
3660         for (i = 0; i < ARRAY_SIZE(products); i++) {
3661                 /* Stand aside for hpsa driver on request */
3662                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3663                         return -ENODEV;
3664                 if (board_id == products[i].board_id)
3665                         break;
3666         }
3667         prod_index = i;
3668         if (prod_index == ARRAY_SIZE(products)) {
3669                 dev_warn(&pdev->dev,
3670                         "unrecognized board ID: 0x%08lx, ignoring.\n",
3671                         (unsigned long) board_id);
3672                 return -ENODEV;
3673         }
3674
3675         /* check to see if controller has been disabled */
3676         /* BEFORE trying to enable it */
3677         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3678         if (!(command & 0x02)) {
3679                 printk(KERN_WARNING
3680                        "cciss: controller appears to be disabled\n");
3681                 return -ENODEV;
3682         }
3683
3684         err = pci_enable_device(pdev);
3685         if (err) {
3686                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3687                 return err;
3688         }
3689
3690         err = pci_request_regions(pdev, "cciss");
3691         if (err) {
3692                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3693                        "aborting\n");
3694                 return err;
3695         }
3696
3697 #ifdef CCISS_DEBUG
3698         printk("command = %x\n", command);
3699         printk("irq = %x\n", pdev->irq);
3700         printk("board_id = %x\n", board_id);
3701 #endif                          /* CCISS_DEBUG */
3702
3703 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3704  * else we use the IO-APIC interrupt assigned to us by system ROM.
3705  */
3706         cciss_interrupt_mode(c, pdev, board_id);
3707
3708         /* find the memory BAR */
3709         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3710                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3711                         break;
3712         }
3713         if (i == DEVICE_COUNT_RESOURCE) {
3714                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3715                 err = -ENODEV;
3716                 goto err_out_free_res;
3717         }
3718
3719         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3720                                                  * already removed
3721                                                  */
3722
3723 #ifdef CCISS_DEBUG
3724         printk("address 0 = %lx\n", c->paddr);
3725 #endif                          /* CCISS_DEBUG */
3726         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3727
3728         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3729          * We poll for up to 120 secs, once per 100ms. */
3730         for (i = 0; i < 1200; i++) {
3731                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3732                 if (scratchpad == CCISS_FIRMWARE_READY)
3733                         break;
3734                 set_current_state(TASK_INTERRUPTIBLE);
3735                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
3736         }
3737         if (scratchpad != CCISS_FIRMWARE_READY) {
3738                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3739                 err = -ENODEV;
3740                 goto err_out_free_res;
3741         }
3742
3743         /* get the address index number */
3744         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3745         cfg_base_addr &= (__u32) 0x0000ffff;
3746 #ifdef CCISS_DEBUG
3747         printk("cfg base address = %x\n", cfg_base_addr);
3748 #endif                          /* CCISS_DEBUG */
3749         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3750 #ifdef CCISS_DEBUG
3751         printk("cfg base address index = %llx\n",
3752                 (unsigned long long)cfg_base_addr_index);
3753 #endif                          /* CCISS_DEBUG */
3754         if (cfg_base_addr_index == -1) {
3755                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3756                 err = -ENODEV;
3757                 goto err_out_free_res;
3758         }
3759
3760         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3761 #ifdef CCISS_DEBUG
3762         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3763 #endif                          /* CCISS_DEBUG */
3764         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3765                                                        cfg_base_addr_index) +
3766                                     cfg_offset, sizeof(CfgTable_struct));
3767         c->board_id = board_id;
3768
3769 #ifdef CCISS_DEBUG
3770         print_cfg_table(c->cfgtable);
3771 #endif                          /* CCISS_DEBUG */
3772
3773         /* Some controllers support Zero Memory Raid (ZMR).
3774          * When configured in ZMR mode the number of supported
3775          * commands drops to 64. So instead of just setting an
3776          * arbitrary value we make the driver a little smarter.
3777          * We read the config table to tell us how many commands
3778          * are supported on the controller then subtract 4 to
3779          * leave a little room for ioctl calls.
3780          */
3781         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3782         c->maxsgentries = readl(&(c->cfgtable->MaxSGElements));
3783
3784         /*
3785          * Limit native command to 32 s/g elements to save dma'able memory.
3786          * Howvever spec says if 0, use 31
3787          */
3788
3789         c->max_cmd_sgentries = 31;
3790         if (c->maxsgentries > 512) {
3791                 c->max_cmd_sgentries = 32;
3792                 c->chainsize = c->maxsgentries - c->max_cmd_sgentries + 1;
3793                 c->maxsgentries -= 1;   /* account for chain pointer */
3794         } else {
3795                 c->maxsgentries = 31;   /* Default to traditional value */
3796                 c->chainsize = 0;       /* traditional */
3797         }
3798
3799         c->product_name = products[prod_index].product_name;
3800         c->access = *(products[prod_index].access);
3801         c->nr_cmds = c->max_commands - 4;
3802         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3803             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3804             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3805             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3806                 printk("Does not appear to be a valid CISS config table\n");
3807                 err = -ENODEV;
3808                 goto err_out_free_res;
3809         }
3810 #ifdef CONFIG_X86
3811         {
3812                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3813                 __u32 prefetch;
3814                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3815                 prefetch |= 0x100;
3816                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3817         }
3818 #endif
3819
3820         /* Disabling DMA prefetch and refetch for the P600.
3821          * An ASIC bug may result in accesses to invalid memory addresses.
3822          * We've disabled prefetch for some time now. Testing with XEN
3823          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3824          */
3825         if(board_id == 0x3225103C) {
3826                 __u32 dma_prefetch;
3827                 __u32 dma_refetch;
3828                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3829                 dma_prefetch |= 0x8000;
3830                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3831                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3832                 dma_refetch |= 0x1;
3833                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3834         }
3835
3836 #ifdef CCISS_DEBUG
3837         printk("Trying to put board into Simple mode\n");
3838 #endif                          /* CCISS_DEBUG */
3839         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3840         /* Update the field, and then ring the doorbell */
3841         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3842         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3843
3844         /* under certain very rare conditions, this can take awhile.
3845          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3846          * as we enter this code.) */
3847         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3848                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3849                         break;
3850                 /* delay and try again */
3851                 set_current_state(TASK_INTERRUPTIBLE);
3852                 schedule_timeout(msecs_to_jiffies(1));
3853         }
3854
3855 #ifdef CCISS_DEBUG
3856         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3857                readl(c->vaddr + SA5_DOORBELL));
3858 #endif                          /* CCISS_DEBUG */
3859 #ifdef CCISS_DEBUG
3860         print_cfg_table(c->cfgtable);
3861 #endif                          /* CCISS_DEBUG */
3862
3863         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3864                 printk(KERN_WARNING "cciss: unable to get board into"
3865                        " simple mode\n");
3866                 err = -ENODEV;
3867                 goto err_out_free_res;
3868         }
3869         return 0;
3870
3871 err_out_free_res:
3872         /*
3873          * Deliberately omit pci_disable_device(): it does something nasty to
3874          * Smart Array controllers that pci_enable_device does not undo
3875          */
3876         pci_release_regions(pdev);
3877         return err;
3878 }
3879
3880 /* Function to find the first free pointer into our hba[] array
3881  * Returns -1 if no free entries are left.
3882  */
3883 static int alloc_cciss_hba(void)
3884 {
3885         int i;
3886
3887         for (i = 0; i < MAX_CTLR; i++) {
3888                 if (!hba[i]) {
3889                         ctlr_info_t *p;
3890
3891                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3892                         if (!p)
3893                                 goto Enomem;
3894                         hba[i] = p;
3895                         return i;
3896                 }
3897         }
3898         printk(KERN_WARNING "cciss: This driver supports a maximum"
3899                " of %d controllers.\n", MAX_CTLR);
3900         return -1;
3901 Enomem:
3902         printk(KERN_ERR "cciss: out of memory.\n");
3903         return -1;
3904 }
3905
3906 static void free_hba(int n)
3907 {
3908         ctlr_info_t *h = hba[n];
3909         int i;
3910
3911         hba[n] = NULL;
3912         for (i = 0; i < h->highest_lun + 1; i++)
3913                 if (h->gendisk[i] != NULL)
3914                         put_disk(h->gendisk[i]);
3915         kfree(h);
3916 }
3917
3918 /* Send a message CDB to the firmware. */
3919 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3920 {
3921         typedef struct {
3922                 CommandListHeader_struct CommandHeader;
3923                 RequestBlock_struct Request;
3924                 ErrDescriptor_struct ErrorDescriptor;
3925         } Command;
3926         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3927         Command *cmd;
3928         dma_addr_t paddr64;
3929         uint32_t paddr32, tag;
3930         void __iomem *vaddr;
3931         int i, err;
3932
3933         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3934         if (vaddr == NULL)
3935                 return -ENOMEM;
3936
3937         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3938            CCISS commands, so they must be allocated from the lower 4GiB of
3939            memory. */
3940         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3941         if (err) {
3942                 iounmap(vaddr);
3943                 return -ENOMEM;
3944         }
3945
3946         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3947         if (cmd == NULL) {
3948                 iounmap(vaddr);
3949                 return -ENOMEM;
3950         }
3951
3952         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
3953            although there's no guarantee, we assume that the address is at
3954            least 4-byte aligned (most likely, it's page-aligned). */
3955         paddr32 = paddr64;
3956
3957         cmd->CommandHeader.ReplyQueue = 0;
3958         cmd->CommandHeader.SGList = 0;
3959         cmd->CommandHeader.SGTotal = 0;
3960         cmd->CommandHeader.Tag.lower = paddr32;
3961         cmd->CommandHeader.Tag.upper = 0;
3962         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
3963
3964         cmd->Request.CDBLen = 16;
3965         cmd->Request.Type.Type = TYPE_MSG;
3966         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
3967         cmd->Request.Type.Direction = XFER_NONE;
3968         cmd->Request.Timeout = 0; /* Don't time out */
3969         cmd->Request.CDB[0] = opcode;
3970         cmd->Request.CDB[1] = type;
3971         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
3972
3973         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
3974         cmd->ErrorDescriptor.Addr.upper = 0;
3975         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
3976
3977         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
3978
3979         for (i = 0; i < 10; i++) {
3980                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
3981                 if ((tag & ~3) == paddr32)
3982                         break;
3983                 schedule_timeout_uninterruptible(HZ);
3984         }
3985
3986         iounmap(vaddr);
3987
3988         /* we leak the DMA buffer here ... no choice since the controller could
3989            still complete the command. */
3990         if (i == 10) {
3991                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
3992                         opcode, type);
3993                 return -ETIMEDOUT;
3994         }
3995
3996         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
3997
3998         if (tag & 2) {
3999                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4000                         opcode, type);
4001                 return -EIO;
4002         }
4003
4004         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4005                 opcode, type);
4006         return 0;
4007 }
4008
4009 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4010 #define cciss_noop(p) cciss_message(p, 3, 0)
4011
4012 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4013 {
4014 /* the #defines are stolen from drivers/pci/msi.h. */
4015 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4016 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4017
4018         int pos;
4019         u16 control = 0;
4020
4021         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4022         if (pos) {
4023                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4024                 if (control & PCI_MSI_FLAGS_ENABLE) {
4025                         printk(KERN_INFO "cciss: resetting MSI\n");
4026                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4027                 }
4028         }
4029
4030         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4031         if (pos) {
4032                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4033                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4034                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4035                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4036                 }
4037         }
4038
4039         return 0;
4040 }
4041
4042 /* This does a hard reset of the controller using PCI power management
4043  * states. */
4044 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4045 {
4046         u16 pmcsr, saved_config_space[32];
4047         int i, pos;
4048
4049         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4050
4051         /* This is very nearly the same thing as
4052
4053            pci_save_state(pci_dev);
4054            pci_set_power_state(pci_dev, PCI_D3hot);
4055            pci_set_power_state(pci_dev, PCI_D0);
4056            pci_restore_state(pci_dev);
4057
4058            but we can't use these nice canned kernel routines on
4059            kexec, because they also check the MSI/MSI-X state in PCI
4060            configuration space and do the wrong thing when it is
4061            set/cleared.  Also, the pci_save/restore_state functions
4062            violate the ordering requirements for restoring the
4063            configuration space from the CCISS document (see the
4064            comment below).  So we roll our own .... */
4065
4066         for (i = 0; i < 32; i++)
4067                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4068
4069         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4070         if (pos == 0) {
4071                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4072                 return -ENODEV;
4073         }
4074
4075         /* Quoting from the Open CISS Specification: "The Power
4076          * Management Control/Status Register (CSR) controls the power
4077          * state of the device.  The normal operating state is D0,
4078          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4079          * the controller, place the interface device in D3 then to
4080          * D0, this causes a secondary PCI reset which will reset the
4081          * controller." */
4082
4083         /* enter the D3hot power management state */
4084         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4085         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4086         pmcsr |= PCI_D3hot;
4087         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4088
4089         schedule_timeout_uninterruptible(HZ >> 1);
4090
4091         /* enter the D0 power management state */
4092         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4093         pmcsr |= PCI_D0;
4094         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4095
4096         schedule_timeout_uninterruptible(HZ >> 1);
4097
4098         /* Restore the PCI configuration space.  The Open CISS
4099          * Specification says, "Restore the PCI Configuration
4100          * Registers, offsets 00h through 60h. It is important to
4101          * restore the command register, 16-bits at offset 04h,
4102          * last. Do not restore the configuration status register,
4103          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4104         for (i = 0; i < 32; i++) {
4105                 if (i == 2 || i == 3)
4106                         continue;
4107                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4108         }
4109         wmb();
4110         pci_write_config_word(pdev, 4, saved_config_space[2]);
4111
4112         return 0;
4113 }
4114
4115 /*
4116  *  This is it.  Find all the controllers and register them.  I really hate
4117  *  stealing all these major device numbers.
4118  *  returns the number of block devices registered.
4119  */
4120 static int __devinit cciss_init_one(struct pci_dev *pdev,
4121                                     const struct pci_device_id *ent)
4122 {
4123         int i;
4124         int j = 0;
4125         int k = 0;
4126         int rc;
4127         int dac, return_code;
4128         InquiryData_struct *inq_buff;
4129
4130         if (reset_devices) {
4131                 /* Reset the controller with a PCI power-cycle */
4132                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4133                         return -ENODEV;
4134
4135                 /* Now try to get the controller to respond to a no-op. Some
4136                    devices (notably the HP Smart Array 5i Controller) need
4137                    up to 30 seconds to respond. */
4138                 for (i=0; i<30; i++) {
4139                         if (cciss_noop(pdev) == 0)
4140                                 break;
4141
4142                         schedule_timeout_uninterruptible(HZ);
4143                 }
4144                 if (i == 30) {
4145                         printk(KERN_ERR "cciss: controller seems dead\n");
4146                         return -EBUSY;
4147                 }
4148         }
4149
4150         i = alloc_cciss_hba();
4151         if (i < 0)
4152                 return -1;
4153
4154         hba[i]->busy_initializing = 1;
4155         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4156         INIT_HLIST_HEAD(&hba[i]->reqQ);
4157         mutex_init(&hba[i]->busy_shutting_down);
4158
4159         if (cciss_pci_init(hba[i], pdev) != 0)
4160                 goto clean_no_release_regions;
4161
4162         sprintf(hba[i]->devname, "cciss%d", i);
4163         hba[i]->ctlr = i;
4164         hba[i]->pdev = pdev;
4165
4166         init_completion(&hba[i]->scan_wait);
4167
4168         if (cciss_create_hba_sysfs_entry(hba[i]))
4169                 goto clean0;
4170
4171         /* configure PCI DMA stuff */
4172         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4173                 dac = 1;
4174         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4175                 dac = 0;
4176         else {
4177                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4178                 goto clean1;
4179         }
4180
4181         /*
4182          * register with the major number, or get a dynamic major number
4183          * by passing 0 as argument.  This is done for greater than
4184          * 8 controller support.
4185          */
4186         if (i < MAX_CTLR_ORIG)
4187                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4188         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4189         if (rc == -EBUSY || rc == -EINVAL) {
4190                 printk(KERN_ERR
4191                        "cciss:  Unable to get major number %d for %s "
4192                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4193                 goto clean1;
4194         } else {
4195                 if (i >= MAX_CTLR_ORIG)
4196                         hba[i]->major = rc;
4197         }
4198
4199         /* make sure the board interrupts are off */
4200         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4201         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
4202                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
4203                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4204                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4205                 goto clean2;
4206         }
4207
4208         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4209                hba[i]->devname, pdev->device, pci_name(pdev),
4210                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4211
4212         hba[i]->cmd_pool_bits =
4213             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4214                         * sizeof(unsigned long), GFP_KERNEL);
4215         hba[i]->cmd_pool = (CommandList_struct *)
4216             pci_alloc_consistent(hba[i]->pdev,
4217                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4218                     &(hba[i]->cmd_pool_dhandle));
4219         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4220             pci_alloc_consistent(hba[i]->pdev,
4221                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4222                     &(hba[i]->errinfo_pool_dhandle));
4223         if ((hba[i]->cmd_pool_bits == NULL)
4224             || (hba[i]->cmd_pool == NULL)
4225             || (hba[i]->errinfo_pool == NULL)) {
4226                 printk(KERN_ERR "cciss: out of memory");
4227                 goto clean4;
4228         }
4229
4230         /* Need space for temp scatter list */
4231         hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4232                                                 sizeof(struct scatterlist *),
4233                                                 GFP_KERNEL);
4234         for (k = 0; k < hba[i]->nr_cmds; k++) {
4235                 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4236                                                         hba[i]->maxsgentries,
4237                                                         GFP_KERNEL);
4238                 if (hba[i]->scatter_list[k] == NULL) {
4239                         printk(KERN_ERR "cciss%d: could not allocate "
4240                                 "s/g lists\n", i);
4241                         goto clean4;
4242                 }
4243         }
4244         hba[i]->cmd_sg_list = kmalloc(sizeof(struct Cmd_sg_list *) *
4245                                                 hba[i]->nr_cmds,
4246                                                 GFP_KERNEL);
4247         if (!hba[i]->cmd_sg_list) {
4248                 printk(KERN_ERR "cciss%d: Cannot get memory for "
4249                         "s/g chaining.\n", i);
4250                 goto clean4;
4251         }
4252         /* Build up chain blocks for each command */
4253         if (hba[i]->chainsize > 0) {
4254                 for (j = 0; j < hba[i]->nr_cmds; j++) {
4255                         hba[i]->cmd_sg_list[j] =
4256                                         kmalloc(sizeof(struct Cmd_sg_list),
4257                                                         GFP_KERNEL);
4258                         if (!hba[i]->cmd_sg_list[j]) {
4259                                 printk(KERN_ERR "cciss%d: Cannot get memory "
4260                                         "for chain block.\n", i);
4261                                 goto clean4;
4262                         }
4263                         /* Need a block of chainsized s/g elements. */
4264                         hba[i]->cmd_sg_list[j]->sgchain =
4265                                         kmalloc((hba[i]->chainsize *
4266                                                 sizeof(SGDescriptor_struct)),
4267                                                 GFP_KERNEL);
4268                         if (!hba[i]->cmd_sg_list[j]->sgchain) {
4269                                 printk(KERN_ERR "cciss%d: Cannot get memory "
4270                                         "for s/g chains\n", i);
4271                                 goto clean4;
4272                         }
4273                 }
4274         }
4275
4276         spin_lock_init(&hba[i]->lock);
4277
4278         /* Initialize the pdev driver private data.
4279            have it point to hba[i].  */
4280         pci_set_drvdata(pdev, hba[i]);
4281         /* command and error info recs zeroed out before
4282            they are used */
4283         memset(hba[i]->cmd_pool_bits, 0,
4284                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4285                         * sizeof(unsigned long));
4286
4287         hba[i]->num_luns = 0;
4288         hba[i]->highest_lun = -1;
4289         for (j = 0; j < CISS_MAX_LUN; j++) {
4290                 hba[i]->drv[j] = NULL;
4291                 hba[i]->gendisk[j] = NULL;
4292         }
4293
4294         cciss_scsi_setup(i);
4295
4296         /* Turn the interrupts on so we can service requests */
4297         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4298
4299         /* Get the firmware version */
4300         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4301         if (inq_buff == NULL) {
4302                 printk(KERN_ERR "cciss: out of memory\n");
4303                 goto clean4;
4304         }
4305
4306         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4307                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4308         if (return_code == IO_OK) {
4309                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4310                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4311                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4312                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4313         } else {         /* send command failed */
4314                 printk(KERN_WARNING "cciss: unable to determine firmware"
4315                         " version of controller\n");
4316         }
4317         kfree(inq_buff);
4318
4319         cciss_procinit(i);
4320
4321         hba[i]->cciss_max_sectors = 8192;
4322
4323         rebuild_lun_table(hba[i], 1, 0);
4324         hba[i]->busy_initializing = 0;
4325         return 1;
4326
4327 clean4:
4328         kfree(hba[i]->cmd_pool_bits);
4329         /* Free up sg elements */
4330         for (k = 0; k < hba[i]->nr_cmds; k++)
4331                 kfree(hba[i]->scatter_list[k]);
4332         kfree(hba[i]->scatter_list);
4333         /* Only free up extra s/g lists if controller supports them */
4334         if (hba[i]->chainsize > 0) {
4335                 for (j = 0; j < hba[i]->nr_cmds; j++) {
4336                         if (hba[i]->cmd_sg_list[j]) {
4337                                 kfree(hba[i]->cmd_sg_list[j]->sgchain);
4338                                 kfree(hba[i]->cmd_sg_list[j]);
4339                         }
4340                 }
4341                 kfree(hba[i]->cmd_sg_list);
4342         }
4343         if (hba[i]->cmd_pool)
4344                 pci_free_consistent(hba[i]->pdev,
4345                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4346                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4347         if (hba[i]->errinfo_pool)
4348                 pci_free_consistent(hba[i]->pdev,
4349                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4350                                     hba[i]->errinfo_pool,
4351                                     hba[i]->errinfo_pool_dhandle);
4352         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4353 clean2:
4354         unregister_blkdev(hba[i]->major, hba[i]->devname);
4355 clean1:
4356         cciss_destroy_hba_sysfs_entry(hba[i]);
4357 clean0:
4358         pci_release_regions(pdev);
4359 clean_no_release_regions:
4360         hba[i]->busy_initializing = 0;
4361
4362         /*
4363          * Deliberately omit pci_disable_device(): it does something nasty to
4364          * Smart Array controllers that pci_enable_device does not undo
4365          */
4366         pci_set_drvdata(pdev, NULL);
4367         free_hba(i);
4368         return -1;
4369 }
4370
4371 static void cciss_shutdown(struct pci_dev *pdev)
4372 {
4373         ctlr_info_t *h;
4374         char *flush_buf;
4375         int return_code;
4376
4377         h = pci_get_drvdata(pdev);
4378         flush_buf = kzalloc(4, GFP_KERNEL);
4379         if (!flush_buf) {
4380                 printk(KERN_WARNING
4381                         "cciss:%d cache not flushed, out of memory.\n",
4382                         h->ctlr);
4383                 return;
4384         }
4385         /* write all data in the battery backed cache to disk */
4386         memset(flush_buf, 0, 4);
4387         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4388                 4, 0, CTLR_LUNID, TYPE_CMD);
4389         kfree(flush_buf);
4390         if (return_code != IO_OK)
4391                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4392                         h->ctlr);
4393         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4394         free_irq(h->intr[2], h);
4395 }
4396
4397 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4398 {
4399         ctlr_info_t *tmp_ptr;
4400         int i, j;
4401
4402         if (pci_get_drvdata(pdev) == NULL) {
4403                 printk(KERN_ERR "cciss: Unable to remove device \n");
4404                 return;
4405         }
4406
4407         tmp_ptr = pci_get_drvdata(pdev);
4408         i = tmp_ptr->ctlr;
4409         if (hba[i] == NULL) {
4410                 printk(KERN_ERR "cciss: device appears to "
4411                        "already be removed \n");
4412                 return;
4413         }
4414
4415         mutex_lock(&hba[i]->busy_shutting_down);
4416
4417         remove_from_scan_list(hba[i]);
4418         remove_proc_entry(hba[i]->devname, proc_cciss);
4419         unregister_blkdev(hba[i]->major, hba[i]->devname);
4420
4421         /* remove it from the disk list */
4422         for (j = 0; j < CISS_MAX_LUN; j++) {
4423                 struct gendisk *disk = hba[i]->gendisk[j];
4424                 if (disk) {
4425                         struct request_queue *q = disk->queue;
4426
4427                         if (disk->flags & GENHD_FL_UP) {
4428                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4429                                 del_gendisk(disk);
4430                         }
4431                         if (q)
4432                                 blk_cleanup_queue(q);
4433                 }
4434         }
4435
4436 #ifdef CONFIG_CISS_SCSI_TAPE
4437         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4438 #endif
4439
4440         cciss_shutdown(pdev);
4441
4442 #ifdef CONFIG_PCI_MSI
4443         if (hba[i]->msix_vector)
4444                 pci_disable_msix(hba[i]->pdev);
4445         else if (hba[i]->msi_vector)
4446                 pci_disable_msi(hba[i]->pdev);
4447 #endif                          /* CONFIG_PCI_MSI */
4448
4449         iounmap(hba[i]->vaddr);
4450
4451         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4452                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4453         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4454                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4455         kfree(hba[i]->cmd_pool_bits);
4456         /* Free up sg elements */
4457         for (j = 0; j < hba[i]->nr_cmds; j++)
4458                 kfree(hba[i]->scatter_list[j]);
4459         kfree(hba[i]->scatter_list);
4460         /* Only free up extra s/g lists if controller supports them */
4461         if (hba[i]->chainsize > 0) {
4462                 for (j = 0; j < hba[i]->nr_cmds; j++) {
4463                         if (hba[i]->cmd_sg_list[j]) {
4464                                 kfree(hba[i]->cmd_sg_list[j]->sgchain);
4465                                 kfree(hba[i]->cmd_sg_list[j]);
4466                         }
4467                 }
4468                 kfree(hba[i]->cmd_sg_list);
4469         }
4470         /*
4471          * Deliberately omit pci_disable_device(): it does something nasty to
4472          * Smart Array controllers that pci_enable_device does not undo
4473          */
4474         pci_release_regions(pdev);
4475         pci_set_drvdata(pdev, NULL);
4476         cciss_destroy_hba_sysfs_entry(hba[i]);
4477         mutex_unlock(&hba[i]->busy_shutting_down);
4478         free_hba(i);
4479 }
4480
4481 static struct pci_driver cciss_pci_driver = {
4482         .name = "cciss",
4483         .probe = cciss_init_one,
4484         .remove = __devexit_p(cciss_remove_one),
4485         .id_table = cciss_pci_device_id,        /* id_table */
4486         .shutdown = cciss_shutdown,
4487 };
4488
4489 /*
4490  *  This is it.  Register the PCI driver information for the cards we control
4491  *  the OS will call our registered routines when it finds one of our cards.
4492  */
4493 static int __init cciss_init(void)
4494 {
4495         int err;
4496
4497         /*
4498          * The hardware requires that commands are aligned on a 64-bit
4499          * boundary. Given that we use pci_alloc_consistent() to allocate an
4500          * array of them, the size must be a multiple of 8 bytes.
4501          */
4502         BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4503
4504         printk(KERN_INFO DRIVER_NAME "\n");
4505
4506         err = bus_register(&cciss_bus_type);
4507         if (err)
4508                 return err;
4509
4510         /* Start the scan thread */
4511         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4512         if (IS_ERR(cciss_scan_thread)) {
4513                 err = PTR_ERR(cciss_scan_thread);
4514                 goto err_bus_unregister;
4515         }
4516
4517         /* Register for our PCI devices */
4518         err = pci_register_driver(&cciss_pci_driver);
4519         if (err)
4520                 goto err_thread_stop;
4521
4522         return err;
4523
4524 err_thread_stop:
4525         kthread_stop(cciss_scan_thread);
4526 err_bus_unregister:
4527         bus_unregister(&cciss_bus_type);
4528
4529         return err;
4530 }
4531
4532 static void __exit cciss_cleanup(void)
4533 {
4534         int i;
4535
4536         pci_unregister_driver(&cciss_pci_driver);
4537         /* double check that all controller entrys have been removed */
4538         for (i = 0; i < MAX_CTLR; i++) {
4539                 if (hba[i] != NULL) {
4540                         printk(KERN_WARNING "cciss: had to remove"
4541                                " controller %d\n", i);
4542                         cciss_remove_one(hba[i]->pdev);
4543                 }
4544         }
4545         kthread_stop(cciss_scan_thread);
4546         remove_proc_entry("driver/cciss", NULL);
4547         bus_unregister(&cciss_bus_type);
4548 }
4549
4550 static void fail_all_cmds(unsigned long ctlr)
4551 {
4552         /* If we get here, the board is apparently dead. */
4553         ctlr_info_t *h = hba[ctlr];
4554         CommandList_struct *c;
4555         unsigned long flags;
4556
4557         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4558         h->alive = 0;           /* the controller apparently died... */
4559
4560         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4561
4562         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
4563
4564         /* move everything off the request queue onto the completed queue */
4565         while (!hlist_empty(&h->reqQ)) {
4566                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4567                 removeQ(c);
4568                 h->Qdepth--;
4569                 addQ(&h->cmpQ, c);
4570         }
4571
4572         /* Now, fail everything on the completed queue with a HW error */
4573         while (!hlist_empty(&h->cmpQ)) {
4574                 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4575                 removeQ(c);
4576                 if (c->cmd_type != CMD_MSG_STALE)
4577                         c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4578                 if (c->cmd_type == CMD_RWREQ) {
4579                         complete_command(h, c, 0);
4580                 } else if (c->cmd_type == CMD_IOCTL_PEND)
4581                         complete(c->waiting);
4582 #ifdef CONFIG_CISS_SCSI_TAPE
4583                 else if (c->cmd_type == CMD_SCSI)
4584                         complete_scsi_command(c, 0, 0);
4585 #endif
4586         }
4587         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4588         return;
4589 }
4590
4591 module_init(cciss_init);
4592 module_exit(cciss_cleanup);