git.oblomov.eu Git - linux-2.6/blob - drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/delay.h>
  30 #include <linux/major.h>
  31 #include <linux/fs.h>
  32 #include <linux/bio.h>
  33 #include <linux/blkpg.h>
  34 #include <linux/timer.h>
  35 #include <linux/proc_fs.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/init.h>
  38 #include <linux/hdreg.h>
  39 #include <linux/spinlock.h>
  40 #include <linux/compat.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/io.h>
  43
  44 #include <linux/dma-mapping.h>
  45 #include <linux/blkdev.h>
  46 #include <linux/genhd.h>
  47 #include <linux/completion.h>
  48 #include <scsi/scsi.h>
  49 #include <scsi/sg.h>
  50 #include <scsi/scsi_ioctl.h>
  51 #include <linux/cdrom.h>
  52 #include <linux/scatterlist.h>
  53 #include <linux/kthread.h>
  54
  55 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  56 #define DRIVER_NAME "HP CISS Driver (v 3.6.20)"
  57 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 20)
  58
  59 /* Embedded module documentation macros - see modules.h */
  60 MODULE_AUTHOR("Hewlett-Packard Company");
  61 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  62 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
  63                         " SA6i P600 P800 P400 P400i E200 E200i E500 P700m"
  64                         " Smart Array G2 Series SAS/SATA Controllers");
  65 MODULE_VERSION("3.6.20");
  66 MODULE_LICENSE("GPL");
  67
  68 #include "cciss_cmd.h"
  69 #include "cciss.h"
  70 #include <linux/cciss_ioctl.h>
  71
  72 /* define the PCI info for the cards we can control */
  73 static const struct pci_device_id cciss_pci_device_id[] = {
  74         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  75         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  76         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  77         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  78         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  79         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  80         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  83         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  84         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  85         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  86         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  87         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  88         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  89         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
  91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 101         {PCI_VENDOR_ID_HP,     PCI_ANY_ID,      PCI_ANY_ID, PCI_ANY_ID,
 102                 PCI_CLASS_STORAGE_RAID << 8, 0xffff << 8, 0},
 103         {0,}
 104 };
 105
 106 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 107
 108 /*  board_id = Subsystem Device ID & Vendor ID
 109  *  product = Marketing Name for the board
 110  *  access = Address of the struct of function pointers
 111  */
 112 static struct board_type products[] = {
 113         {0x40700E11, "Smart Array 5300", &SA5_access},
 114         {0x40800E11, "Smart Array 5i", &SA5B_access},
 115         {0x40820E11, "Smart Array 532", &SA5B_access},
 116         {0x40830E11, "Smart Array 5312", &SA5B_access},
 117         {0x409A0E11, "Smart Array 641", &SA5_access},
 118         {0x409B0E11, "Smart Array 642", &SA5_access},
 119         {0x409C0E11, "Smart Array 6400", &SA5_access},
 120         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 121         {0x40910E11, "Smart Array 6i", &SA5_access},
 122         {0x3225103C, "Smart Array P600", &SA5_access},
 123         {0x3223103C, "Smart Array P800", &SA5_access},
 124         {0x3234103C, "Smart Array P400", &SA5_access},
 125         {0x3235103C, "Smart Array P400i", &SA5_access},
 126         {0x3211103C, "Smart Array E200i", &SA5_access},
 127         {0x3212103C, "Smart Array E200", &SA5_access},
 128         {0x3213103C, "Smart Array E200i", &SA5_access},
 129         {0x3214103C, "Smart Array E200i", &SA5_access},
 130         {0x3215103C, "Smart Array E200i", &SA5_access},
 131         {0x3237103C, "Smart Array E500", &SA5_access},
 132         {0x323D103C, "Smart Array P700m", &SA5_access},
 133         {0x3241103C, "Smart Array P212", &SA5_access},
 134         {0x3243103C, "Smart Array P410", &SA5_access},
 135         {0x3245103C, "Smart Array P410i", &SA5_access},
 136         {0x3247103C, "Smart Array P411", &SA5_access},
 137         {0x3249103C, "Smart Array P812", &SA5_access},
 138         {0x324A103C, "Smart Array P712m", &SA5_access},
 139         {0x324B103C, "Smart Array P711m", &SA5_access},
 140         {0xFFFF103C, "Unknown Smart Array", &SA5_access},
 141 };
 142
 143 /* How long to wait (in milliseconds) for board to go into simple mode */
 144 #define MAX_CONFIG_WAIT 30000
 145 #define MAX_IOCTL_CONFIG_WAIT 1000
 146
 147 /*define how many times we will try a command because of bus resets */
 148 #define MAX_CMD_RETRIES 3
 149
 150 #define MAX_CTLR        32
 151
 152 /* Originally cciss driver only supports 8 major numbers */
 153 #define MAX_CTLR_ORIG   8
 154
 155 static ctlr_info_t *hba[MAX_CTLR];
 156
 157 static void do_cciss_request(struct request_queue *q);
 158 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
 159 static int cciss_open(struct block_device *bdev, fmode_t mode);
 160 static int cciss_release(struct gendisk *disk, fmode_t mode);
 161 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 162                        unsigned int cmd, unsigned long arg);
 163 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 164
 165 static int cciss_revalidate(struct gendisk *disk);
 166 static int rebuild_lun_table(ctlr_info_t *h, int first_time);
 167 static int deregister_disk(ctlr_info_t *h, int drv_index,
 168                            int clear_all);
 169
 170 static void cciss_read_capacity(int ctlr, int logvol, int withirq,
 171                         sector_t *total_size, unsigned int *block_size);
 172 static void cciss_read_capacity_16(int ctlr, int logvol, int withirq,
 173                         sector_t *total_size, unsigned int *block_size);
 174 static void cciss_geometry_inquiry(int ctlr, int logvol,
 175                         int withirq, sector_t total_size,
 176                         unsigned int block_size, InquiryData_struct *inq_buff,
 177                                    drive_info_struct *drv);
 178 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 179                                            __u32);
 180 static void start_io(ctlr_info_t *h);
 181 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
 182                    __u8 page_code, unsigned char *scsi3addr, int cmd_type);
 183 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 184                         __u8 page_code, unsigned char scsi3addr[],
 185                         int cmd_type);
 186 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 187         int attempt_retry);
 188 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 189
 190 static void fail_all_cmds(unsigned long ctlr);
 191 static int scan_thread(void *data);
 192 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 193
 194 #ifdef CONFIG_PROC_FS
 195 static void cciss_procinit(int i);
 196 #else
 197 static void cciss_procinit(int i)
 198 {
 199 }
 200 #endif                          /* CONFIG_PROC_FS */
 201
 202 #ifdef CONFIG_COMPAT
 203 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 204                               unsigned, unsigned long);
 205 #endif
 206
 207 static struct block_device_operations cciss_fops = {
 208         .owner = THIS_MODULE,
 209         .open = cciss_open,
 210         .release = cciss_release,
 211         .locked_ioctl = cciss_ioctl,
 212         .getgeo = cciss_getgeo,
 213 #ifdef CONFIG_COMPAT
 214         .compat_ioctl = cciss_compat_ioctl,
 215 #endif
 216         .revalidate_disk = cciss_revalidate,
 217 };
 218
 219 /*
 220  * Enqueuing and dequeuing functions for cmdlists.
 221  */
 222 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 223 {
 224         hlist_add_head(&c->list, list);
 225 }
 226
 227 static inline void removeQ(CommandList_struct *c)
 228 {
 229         /*
 230          * After kexec/dump some commands might still
 231          * be in flight, which the firmware will try
 232          * to complete. Resetting the firmware doesn't work
 233          * with old fw revisions, so we have to mark
 234          * them off as 'stale' to prevent the driver from
 235          * falling over.
 236          */
 237         if (WARN_ON(hlist_unhashed(&c->list))) {
 238                 c->cmd_type = CMD_MSG_STALE;
 239                 return;
 240         }
 241
 242         hlist_del_init(&c->list);
 243 }
 244
 245 #include "cciss_scsi.c"         /* For SCSI tape support */
 246
 247 #define RAID_UNKNOWN 6
 248
 249 #ifdef CONFIG_PROC_FS
 250
 251 /*
 252  * Report information about this controller.
 253  */
 254 #define ENG_GIG 1000000000
 255 #define ENG_GIG_FACTOR (ENG_GIG/512)
 256 #define ENGAGE_SCSI     "engage scsi"
 257 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 258         "UNKNOWN"
 259 };
 260
 261 static struct proc_dir_entry *proc_cciss;
 262
 263 static void cciss_seq_show_header(struct seq_file *seq)
 264 {
 265         ctlr_info_t *h = seq->private;
 266
 267         seq_printf(seq, "%s: HP %s Controller\n"
 268                 "Board ID: 0x%08lx\n"
 269                 "Firmware Version: %c%c%c%c\n"
 270                 "IRQ: %d\n"
 271                 "Logical drives: %d\n"
 272                 "Current Q depth: %d\n"
 273                 "Current # commands on controller: %d\n"
 274                 "Max Q depth since init: %d\n"
 275                 "Max # commands on controller since init: %d\n"
 276                 "Max SG entries since init: %d\n",
 277                 h->devname,
 278                 h->product_name,
 279                 (unsigned long)h->board_id,
 280                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 281                 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
 282                 h->num_luns,
 283                 h->Qdepth, h->commands_outstanding,
 284                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 285
 286 #ifdef CONFIG_CISS_SCSI_TAPE
 287         cciss_seq_tape_report(seq, h->ctlr);
 288 #endif /* CONFIG_CISS_SCSI_TAPE */
 289 }
 290
 291 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 292 {
 293         ctlr_info_t *h = seq->private;
 294         unsigned ctlr = h->ctlr;
 295         unsigned long flags;
 296
 297         /* prevent displaying bogus info during configuration
 298          * or deconfiguration of a logical volume
 299          */
 300         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 301         if (h->busy_configuring) {
 302                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 303                 return ERR_PTR(-EBUSY);
 304         }
 305         h->busy_configuring = 1;
 306         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 307
 308         if (*pos == 0)
 309                 cciss_seq_show_header(seq);
 310
 311         return pos;
 312 }
 313
 314 static int cciss_seq_show(struct seq_file *seq, void *v)
 315 {
 316         sector_t vol_sz, vol_sz_frac;
 317         ctlr_info_t *h = seq->private;
 318         unsigned ctlr = h->ctlr;
 319         loff_t *pos = v;
 320         drive_info_struct *drv = &h->drv[*pos];
 321
 322         if (*pos > h->highest_lun)
 323                 return 0;
 324
 325         if (drv->heads == 0)
 326                 return 0;
 327
 328         vol_sz = drv->nr_blocks;
 329         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 330         vol_sz_frac *= 100;
 331         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 332
 333         if (drv->raid_level > 5)
 334                 drv->raid_level = RAID_UNKNOWN;
 335         seq_printf(seq, "cciss/c%dd%d:"
 336                         "\t%4u.%02uGB\tRAID %s\n",
 337                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 338                         raid_label[drv->raid_level]);
 339         return 0;
 340 }
 341
 342 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 343 {
 344         ctlr_info_t *h = seq->private;
 345
 346         if (*pos > h->highest_lun)
 347                 return NULL;
 348         *pos += 1;
 349
 350         return pos;
 351 }
 352
 353 static void cciss_seq_stop(struct seq_file *seq, void *v)
 354 {
 355         ctlr_info_t *h = seq->private;
 356
 357         /* Only reset h->busy_configuring if we succeeded in setting
 358          * it during cciss_seq_start. */
 359         if (v == ERR_PTR(-EBUSY))
 360                 return;
 361
 362         h->busy_configuring = 0;
 363 }
 364
 365 static struct seq_operations cciss_seq_ops = {
 366         .start = cciss_seq_start,
 367         .show  = cciss_seq_show,
 368         .next  = cciss_seq_next,
 369         .stop  = cciss_seq_stop,
 370 };
 371
 372 static int cciss_seq_open(struct inode *inode, struct file *file)
 373 {
 374         int ret = seq_open(file, &cciss_seq_ops);
 375         struct seq_file *seq = file->private_data;
 376
 377         if (!ret)
 378                 seq->private = PDE(inode)->data;
 379
 380         return ret;
 381 }
 382
 383 static ssize_t
 384 cciss_proc_write(struct file *file, const char __user *buf,
 385                  size_t length, loff_t *ppos)
 386 {
 387         int err;
 388         char *buffer;
 389
 390 #ifndef CONFIG_CISS_SCSI_TAPE
 391         return -EINVAL;
 392 #endif
 393
 394         if (!buf || length > PAGE_SIZE - 1)
 395                 return -EINVAL;
 396
 397         buffer = (char *)__get_free_page(GFP_KERNEL);
 398         if (!buffer)
 399                 return -ENOMEM;
 400
 401         err = -EFAULT;
 402         if (copy_from_user(buffer, buf, length))
 403                 goto out;
 404         buffer[length] = '\0';
 405
 406 #ifdef CONFIG_CISS_SCSI_TAPE
 407         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 408                 struct seq_file *seq = file->private_data;
 409                 ctlr_info_t *h = seq->private;
 410                 int rc;
 411
 412                 rc = cciss_engage_scsi(h->ctlr);
 413                 if (rc != 0)
 414                         err = -rc;
 415                 else
 416                         err = length;
 417         } else
 418 #endif /* CONFIG_CISS_SCSI_TAPE */
 419                 err = -EINVAL;
 420         /* might be nice to have "disengage" too, but it's not
 421            safely possible. (only 1 module use count, lock issues.) */
 422
 423 out:
 424         free_page((unsigned long)buffer);
 425         return err;
 426 }
 427
 428 static struct file_operations cciss_proc_fops = {
 429         .owner   = THIS_MODULE,
 430         .open    = cciss_seq_open,
 431         .read    = seq_read,
 432         .llseek  = seq_lseek,
 433         .release = seq_release,
 434         .write   = cciss_proc_write,
 435 };
 436
 437 static void __devinit cciss_procinit(int i)
 438 {
 439         struct proc_dir_entry *pde;
 440
 441         if (proc_cciss == NULL)
 442                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 443         if (!proc_cciss)
 444                 return;
 445         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 446                                         S_IROTH, proc_cciss,
 447                                         &cciss_proc_fops, hba[i]);
 448 }
 449 #endif                          /* CONFIG_PROC_FS */
 450
 451 #define MAX_PRODUCT_NAME_LEN 19
 452
 453 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 454 #define to_drv(n) container_of(n, drive_info_struct, dev)
 455
 456 static struct device_type cciss_host_type = {
 457         .name           = "cciss_host",
 458 };
 459
 460 static ssize_t dev_show_unique_id(struct device *dev,
 461                                  struct device_attribute *attr,
 462                                  char *buf)
 463 {
 464         drive_info_struct *drv = to_drv(dev);
 465         struct ctlr_info *h = to_hba(drv->dev.parent);
 466         __u8 sn[16];
 467         unsigned long flags;
 468         int ret = 0;
 469
 470         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 471         if (h->busy_configuring)
 472                 ret = -EBUSY;
 473         else
 474                 memcpy(sn, drv->serial_no, sizeof(sn));
 475         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 476
 477         if (ret)
 478                 return ret;
 479         else
 480                 return snprintf(buf, 16 * 2 + 2,
 481                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 482                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 483                                 sn[0], sn[1], sn[2], sn[3],
 484                                 sn[4], sn[5], sn[6], sn[7],
 485                                 sn[8], sn[9], sn[10], sn[11],
 486                                 sn[12], sn[13], sn[14], sn[15]);
 487 }
 488 DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 489
 490 static ssize_t dev_show_vendor(struct device *dev,
 491                                struct device_attribute *attr,
 492                                char *buf)
 493 {
 494         drive_info_struct *drv = to_drv(dev);
 495         struct ctlr_info *h = to_hba(drv->dev.parent);
 496         char vendor[VENDOR_LEN + 1];
 497         unsigned long flags;
 498         int ret = 0;
 499
 500         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 501         if (h->busy_configuring)
 502                 ret = -EBUSY;
 503         else
 504                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 505         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 506
 507         if (ret)
 508                 return ret;
 509         else
 510                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 511 }
 512 DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 513
 514 static ssize_t dev_show_model(struct device *dev,
 515                               struct device_attribute *attr,
 516                               char *buf)
 517 {
 518         drive_info_struct *drv = to_drv(dev);
 519         struct ctlr_info *h = to_hba(drv->dev.parent);
 520         char model[MODEL_LEN + 1];
 521         unsigned long flags;
 522         int ret = 0;
 523
 524         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 525         if (h->busy_configuring)
 526                 ret = -EBUSY;
 527         else
 528                 memcpy(model, drv->model, MODEL_LEN + 1);
 529         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 530
 531         if (ret)
 532                 return ret;
 533         else
 534                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 535 }
 536 DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 537
 538 static ssize_t dev_show_rev(struct device *dev,
 539                             struct device_attribute *attr,
 540                             char *buf)
 541 {
 542         drive_info_struct *drv = to_drv(dev);
 543         struct ctlr_info *h = to_hba(drv->dev.parent);
 544         char rev[REV_LEN + 1];
 545         unsigned long flags;
 546         int ret = 0;
 547
 548         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 549         if (h->busy_configuring)
 550                 ret = -EBUSY;
 551         else
 552                 memcpy(rev, drv->rev, REV_LEN + 1);
 553         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 554
 555         if (ret)
 556                 return ret;
 557         else
 558                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 559 }
 560 DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 561
 562 static struct attribute *cciss_dev_attrs[] = {
 563         &dev_attr_unique_id.attr,
 564         &dev_attr_model.attr,
 565         &dev_attr_vendor.attr,
 566         &dev_attr_rev.attr,
 567         NULL
 568 };
 569
 570 static struct attribute_group cciss_dev_attr_group = {
 571         .attrs = cciss_dev_attrs,
 572 };
 573
 574 static struct attribute_group *cciss_dev_attr_groups[] = {
 575         &cciss_dev_attr_group,
 576         NULL
 577 };
 578
 579 static struct device_type cciss_dev_type = {
 580         .name           = "cciss_device",
 581         .groups         = cciss_dev_attr_groups,
 582 };
 583
 584 static struct bus_type cciss_bus_type = {
 585         .name           = "cciss",
 586 };
 587
 588
 589 /*
 590  * Initialize sysfs entry for each controller.  This sets up and registers
 591  * the 'cciss#' directory for each individual controller under
 592  * /sys/bus/pci/devices/<dev>/.
 593  */
 594 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 595 {
 596         device_initialize(&h->dev);
 597         h->dev.type = &cciss_host_type;
 598         h->dev.bus = &cciss_bus_type;
 599         dev_set_name(&h->dev, "%s", h->devname);
 600         h->dev.parent = &h->pdev->dev;
 601
 602         return device_add(&h->dev);
 603 }
 604
 605 /*
 606  * Remove sysfs entries for an hba.
 607  */
 608 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 609 {
 610         device_del(&h->dev);
 611 }
 612
 613 /*
 614  * Initialize sysfs for each logical drive.  This sets up and registers
 615  * the 'c#d#' directory for each individual logical drive under
 616  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 617  * /sys/block/cciss!c#d# to this entry.
 618  */
 619 static int cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 620                                        drive_info_struct *drv,
 621                                        int drv_index)
 622 {
 623         device_initialize(&drv->dev);
 624         drv->dev.type = &cciss_dev_type;
 625         drv->dev.bus = &cciss_bus_type;
 626         dev_set_name(&drv->dev, "c%dd%d", h->ctlr, drv_index);
 627         drv->dev.parent = &h->dev;
 628         return device_add(&drv->dev);
 629 }
 630
 631 /*
 632  * Remove sysfs entries for a logical drive.
 633  */
 634 static void cciss_destroy_ld_sysfs_entry(drive_info_struct *drv)
 635 {
 636         device_del(&drv->dev);
 637 }
 638
 639 /*
 640  * For operations that cannot sleep, a command block is allocated at init,
 641  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 642  * which ones are free or in use.  For operations that can wait for kmalloc
 643  * to possible sleep, this routine can be called with get_from_pool set to 0.
 644  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 645  */
 646 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 647 {
 648         CommandList_struct *c;
 649         int i;
 650         u64bit temp64;
 651         dma_addr_t cmd_dma_handle, err_dma_handle;
 652
 653         if (!get_from_pool) {
 654                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 655                         sizeof(CommandList_struct), &cmd_dma_handle);
 656                 if (c == NULL)
 657                         return NULL;
 658                 memset(c, 0, sizeof(CommandList_struct));
 659
 660                 c->cmdindex = -1;
 661
 662                 c->err_info = (ErrorInfo_struct *)
 663                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 664                             &err_dma_handle);
 665
 666                 if (c->err_info == NULL) {
 667                         pci_free_consistent(h->pdev,
 668                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 669                         return NULL;
 670                 }
 671                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 672         } else {                /* get it out of the controllers pool */
 673
 674                 do {
 675                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 676                         if (i == h->nr_cmds)
 677                                 return NULL;
 678                 } while (test_and_set_bit
 679                          (i & (BITS_PER_LONG - 1),
 680                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 681 #ifdef CCISS_DEBUG
 682                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 683 #endif
 684                 c = h->cmd_pool + i;
 685                 memset(c, 0, sizeof(CommandList_struct));
 686                 cmd_dma_handle = h->cmd_pool_dhandle
 687                     + i * sizeof(CommandList_struct);
 688                 c->err_info = h->errinfo_pool + i;
 689                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 690                 err_dma_handle = h->errinfo_pool_dhandle
 691                     + i * sizeof(ErrorInfo_struct);
 692                 h->nr_allocs++;
 693
 694                 c->cmdindex = i;
 695         }
 696
 697         INIT_HLIST_NODE(&c->list);
 698         c->busaddr = (__u32) cmd_dma_handle;
 699         temp64.val = (__u64) err_dma_handle;
 700         c->ErrDesc.Addr.lower = temp64.val32.lower;
 701         c->ErrDesc.Addr.upper = temp64.val32.upper;
 702         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 703
 704         c->ctlr = h->ctlr;
 705         return c;
 706 }
 707
 708 /*
 709  * Frees a command block that was previously allocated with cmd_alloc().
 710  */
 711 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 712 {
 713         int i;
 714         u64bit temp64;
 715
 716         if (!got_from_pool) {
 717                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 718                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 719                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 720                                     c->err_info, (dma_addr_t) temp64.val);
 721                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 722                                     c, (dma_addr_t) c->busaddr);
 723         } else {
 724                 i = c - h->cmd_pool;
 725                 clear_bit(i & (BITS_PER_LONG - 1),
 726                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 727                 h->nr_frees++;
 728         }
 729 }
 730
 731 static inline ctlr_info_t *get_host(struct gendisk *disk)
 732 {
 733         return disk->queue->queuedata;
 734 }
 735
 736 static inline drive_info_struct *get_drv(struct gendisk *disk)
 737 {
 738         return disk->private_data;
 739 }
 740
 741 /*
 742  * Open.  Make sure the device is really there.
 743  */
 744 static int cciss_open(struct block_device *bdev, fmode_t mode)
 745 {
 746         ctlr_info_t *host = get_host(bdev->bd_disk);
 747         drive_info_struct *drv = get_drv(bdev->bd_disk);
 748
 749 #ifdef CCISS_DEBUG
 750         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
 751 #endif                          /* CCISS_DEBUG */
 752
 753         if (host->busy_initializing || drv->busy_configuring)
 754                 return -EBUSY;
 755         /*
 756          * Root is allowed to open raw volume zero even if it's not configured
 757          * so array config can still work. Root is also allowed to open any
 758          * volume that has a LUN ID, so it can issue IOCTL to reread the
 759          * disk information.  I don't think I really like this
 760          * but I'm already using way to many device nodes to claim another one
 761          * for "raw controller".
 762          */
 763         if (drv->heads == 0) {
 764                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
 765                         /* if not node 0 make sure it is a partition = 0 */
 766                         if (MINOR(bdev->bd_dev) & 0x0f) {
 767                                 return -ENXIO;
 768                                 /* if it is, make sure we have a LUN ID */
 769                         } else if (drv->LunID == 0) {
 770                                 return -ENXIO;
 771                         }
 772                 }
 773                 if (!capable(CAP_SYS_ADMIN))
 774                         return -EPERM;
 775         }
 776         drv->usage_count++;
 777         host->usage_count++;
 778         return 0;
 779 }
 780
 781 /*
 782  * Close.  Sync first.
 783  */
 784 static int cciss_release(struct gendisk *disk, fmode_t mode)
 785 {
 786         ctlr_info_t *host = get_host(disk);
 787         drive_info_struct *drv = get_drv(disk);
 788
 789 #ifdef CCISS_DEBUG
 790         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
 791 #endif                          /* CCISS_DEBUG */
 792
 793         drv->usage_count--;
 794         host->usage_count--;
 795         return 0;
 796 }
 797
 798 #ifdef CONFIG_COMPAT
 799
 800 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 801                     unsigned cmd, unsigned long arg)
 802 {
 803         int ret;
 804         lock_kernel();
 805         ret = cciss_ioctl(bdev, mode, cmd, arg);
 806         unlock_kernel();
 807         return ret;
 808 }
 809
 810 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 811                                   unsigned cmd, unsigned long arg);
 812 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
 813                                       unsigned cmd, unsigned long arg);
 814
 815 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
 816                               unsigned cmd, unsigned long arg)
 817 {
 818         switch (cmd) {
 819         case CCISS_GETPCIINFO:
 820         case CCISS_GETINTINFO:
 821         case CCISS_SETINTINFO:
 822         case CCISS_GETNODENAME:
 823         case CCISS_SETNODENAME:
 824         case CCISS_GETHEARTBEAT:
 825         case CCISS_GETBUSTYPES:
 826         case CCISS_GETFIRMVER:
 827         case CCISS_GETDRIVVER:
 828         case CCISS_REVALIDVOLS:
 829         case CCISS_DEREGDISK:
 830         case CCISS_REGNEWDISK:
 831         case CCISS_REGNEWD:
 832         case CCISS_RESCANDISK:
 833         case CCISS_GETLUNINFO:
 834                 return do_ioctl(bdev, mode, cmd, arg);
 835
 836         case CCISS_PASSTHRU32:
 837                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
 838         case CCISS_BIG_PASSTHRU32:
 839                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
 840
 841         default:
 842                 return -ENOIOCTLCMD;
 843         }
 844 }
 845
 846 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 847                                   unsigned cmd, unsigned long arg)
 848 {
 849         IOCTL32_Command_struct __user *arg32 =
 850             (IOCTL32_Command_struct __user *) arg;
 851         IOCTL_Command_struct arg64;
 852         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
 853         int err;
 854         u32 cp;
 855
 856         err = 0;
 857         err |=
 858             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
 859                            sizeof(arg64.LUN_info));
 860         err |=
 861             copy_from_user(&arg64.Request, &arg32->Request,
 862                            sizeof(arg64.Request));
 863         err |=
 864             copy_from_user(&arg64.error_info, &arg32->error_info,
 865                            sizeof(arg64.error_info));
 866         err |= get_user(arg64.buf_size, &arg32->buf_size);
 867         err |= get_user(cp, &arg32->buf);
 868         arg64.buf = compat_ptr(cp);
 869         err |= copy_to_user(p, &arg64, sizeof(arg64));
 870
 871         if (err)
 872                 return -EFAULT;
 873
 874         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
 875         if (err)
 876                 return err;
 877         err |=
 878             copy_in_user(&arg32->error_info, &p->error_info,
 879                          sizeof(arg32->error_info));
 880         if (err)
 881                 return -EFAULT;
 882         return err;
 883 }
 884
 885 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
 886                                       unsigned cmd, unsigned long arg)
 887 {
 888         BIG_IOCTL32_Command_struct __user *arg32 =
 889             (BIG_IOCTL32_Command_struct __user *) arg;
 890         BIG_IOCTL_Command_struct arg64;
 891         BIG_IOCTL_Command_struct __user *p =
 892             compat_alloc_user_space(sizeof(arg64));
 893         int err;
 894         u32 cp;
 895
 896         err = 0;
 897         err |=
 898             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
 899                            sizeof(arg64.LUN_info));
 900         err |=
 901             copy_from_user(&arg64.Request, &arg32->Request,
 902                            sizeof(arg64.Request));
 903         err |=
 904             copy_from_user(&arg64.error_info, &arg32->error_info,
 905                            sizeof(arg64.error_info));
 906         err |= get_user(arg64.buf_size, &arg32->buf_size);
 907         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
 908         err |= get_user(cp, &arg32->buf);
 909         arg64.buf = compat_ptr(cp);
 910         err |= copy_to_user(p, &arg64, sizeof(arg64));
 911
 912         if (err)
 913                 return -EFAULT;
 914
 915         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
 916         if (err)
 917                 return err;
 918         err |=
 919             copy_in_user(&arg32->error_info, &p->error_info,
 920                          sizeof(arg32->error_info));
 921         if (err)
 922                 return -EFAULT;
 923         return err;
 924 }
 925 #endif
 926
 927 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 928 {
 929         drive_info_struct *drv = get_drv(bdev->bd_disk);
 930
 931         if (!drv->cylinders)
 932                 return -ENXIO;
 933
 934         geo->heads = drv->heads;
 935         geo->sectors = drv->sectors;
 936         geo->cylinders = drv->cylinders;
 937         return 0;
 938 }
 939
 940 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
 941 {
 942         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
 943                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
 944                 (void)check_for_unit_attention(host, c);
 945 }
 946 /*
 947  * ioctl
 948  */
 949 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 950                        unsigned int cmd, unsigned long arg)
 951 {
 952         struct gendisk *disk = bdev->bd_disk;
 953         ctlr_info_t *host = get_host(disk);
 954         drive_info_struct *drv = get_drv(disk);
 955         int ctlr = host->ctlr;
 956         void __user *argp = (void __user *)arg;
 957
 958 #ifdef CCISS_DEBUG
 959         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
 960 #endif                          /* CCISS_DEBUG */
 961
 962         switch (cmd) {
 963         case CCISS_GETPCIINFO:
 964                 {
 965                         cciss_pci_info_struct pciinfo;
 966
 967                         if (!arg)
 968                                 return -EINVAL;
 969                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
 970                         pciinfo.bus = host->pdev->bus->number;
 971                         pciinfo.dev_fn = host->pdev->devfn;
 972                         pciinfo.board_id = host->board_id;
 973                         if (copy_to_user
 974                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
 975                                 return -EFAULT;
 976                         return 0;
 977                 }
 978         case CCISS_GETINTINFO:
 979                 {
 980                         cciss_coalint_struct intinfo;
 981                         if (!arg)
 982                                 return -EINVAL;
 983                         intinfo.delay =
 984                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
 985                         intinfo.count =
 986                             readl(&host->cfgtable->HostWrite.CoalIntCount);
 987                         if (copy_to_user
 988                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
 989                                 return -EFAULT;
 990                         return 0;
 991                 }
 992         case CCISS_SETINTINFO:
 993                 {
 994                         cciss_coalint_struct intinfo;
 995                         unsigned long flags;
 996                         int i;
 997
 998                         if (!arg)
 999                                 return -EINVAL;
1000                         if (!capable(CAP_SYS_ADMIN))
1001                                 return -EPERM;
1002                         if (copy_from_user
1003                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1004                                 return -EFAULT;
1005                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1006                         {
1007 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1008                                 return -EINVAL;
1009                         }
1010                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1011                         /* Update the field, and then ring the doorbell */
1012                         writel(intinfo.delay,
1013                                &(host->cfgtable->HostWrite.CoalIntDelay));
1014                         writel(intinfo.count,
1015                                &(host->cfgtable->HostWrite.CoalIntCount));
1016                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1017
1018                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1019                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1020                                       & CFGTBL_ChangeReq))
1021                                         break;
1022                                 /* delay and try again */
1023                                 udelay(1000);
1024                         }
1025                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1026                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1027                                 return -EAGAIN;
1028                         return 0;
1029                 }
1030         case CCISS_GETNODENAME:
1031                 {
1032                         NodeName_type NodeName;
1033                         int i;
1034
1035                         if (!arg)
1036                                 return -EINVAL;
1037                         for (i = 0; i < 16; i++)
1038                                 NodeName[i] =
1039                                     readb(&host->cfgtable->ServerName[i]);
1040                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1041                                 return -EFAULT;
1042                         return 0;
1043                 }
1044         case CCISS_SETNODENAME:
1045                 {
1046                         NodeName_type NodeName;
1047                         unsigned long flags;
1048                         int i;
1049
1050                         if (!arg)
1051                                 return -EINVAL;
1052                         if (!capable(CAP_SYS_ADMIN))
1053                                 return -EPERM;
1054
1055                         if (copy_from_user
1056                             (NodeName, argp, sizeof(NodeName_type)))
1057                                 return -EFAULT;
1058
1059                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1060
1061                         /* Update the field, and then ring the doorbell */
1062                         for (i = 0; i < 16; i++)
1063                                 writeb(NodeName[i],
1064                                        &host->cfgtable->ServerName[i]);
1065
1066                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1067
1068                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1069                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1070                                       & CFGTBL_ChangeReq))
1071                                         break;
1072                                 /* delay and try again */
1073                                 udelay(1000);
1074                         }
1075                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1076                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1077                                 return -EAGAIN;
1078                         return 0;
1079                 }
1080
1081         case CCISS_GETHEARTBEAT:
1082                 {
1083                         Heartbeat_type heartbeat;
1084
1085                         if (!arg)
1086                                 return -EINVAL;
1087                         heartbeat = readl(&host->cfgtable->HeartBeat);
1088                         if (copy_to_user
1089                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1090                                 return -EFAULT;
1091                         return 0;
1092                 }
1093         case CCISS_GETBUSTYPES:
1094                 {
1095                         BusTypes_type BusTypes;
1096
1097                         if (!arg)
1098                                 return -EINVAL;
1099                         BusTypes = readl(&host->cfgtable->BusTypes);
1100                         if (copy_to_user
1101                             (argp, &BusTypes, sizeof(BusTypes_type)))
1102                                 return -EFAULT;
1103                         return 0;
1104                 }
1105         case CCISS_GETFIRMVER:
1106                 {
1107                         FirmwareVer_type firmware;
1108
1109                         if (!arg)
1110                                 return -EINVAL;
1111                         memcpy(firmware, host->firm_ver, 4);
1112
1113                         if (copy_to_user
1114                             (argp, firmware, sizeof(FirmwareVer_type)))
1115                                 return -EFAULT;
1116                         return 0;
1117                 }
1118         case CCISS_GETDRIVVER:
1119                 {
1120                         DriverVer_type DriverVer = DRIVER_VERSION;
1121
1122                         if (!arg)
1123                                 return -EINVAL;
1124
1125                         if (copy_to_user
1126                             (argp, &DriverVer, sizeof(DriverVer_type)))
1127                                 return -EFAULT;
1128                         return 0;
1129                 }
1130
1131         case CCISS_DEREGDISK:
1132         case CCISS_REGNEWD:
1133         case CCISS_REVALIDVOLS:
1134                 return rebuild_lun_table(host, 0);
1135
1136         case CCISS_GETLUNINFO:{
1137                         LogvolInfo_struct luninfo;
1138
1139                         luninfo.LunID = drv->LunID;
1140                         luninfo.num_opens = drv->usage_count;
1141                         luninfo.num_parts = 0;
1142                         if (copy_to_user(argp, &luninfo,
1143                                          sizeof(LogvolInfo_struct)))
1144                                 return -EFAULT;
1145                         return 0;
1146                 }
1147         case CCISS_PASSTHRU:
1148                 {
1149                         IOCTL_Command_struct iocommand;
1150                         CommandList_struct *c;
1151                         char *buff = NULL;
1152                         u64bit temp64;
1153                         unsigned long flags;
1154                         DECLARE_COMPLETION_ONSTACK(wait);
1155
1156                         if (!arg)
1157                                 return -EINVAL;
1158
1159                         if (!capable(CAP_SYS_RAWIO))
1160                                 return -EPERM;
1161
1162                         if (copy_from_user
1163                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1164                                 return -EFAULT;
1165                         if ((iocommand.buf_size < 1) &&
1166                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1167                                 return -EINVAL;
1168                         }
1169 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1170                         /* Check kmalloc limits */
1171                         if (iocommand.buf_size > 128000)
1172                                 return -EINVAL;
1173 #endif
1174                         if (iocommand.buf_size > 0) {
1175                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1176                                 if (buff == NULL)
1177                                         return -EFAULT;
1178                         }
1179                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1180                                 /* Copy the data into the buffer we created */
1181                                 if (copy_from_user
1182                                     (buff, iocommand.buf, iocommand.buf_size)) {
1183                                         kfree(buff);
1184                                         return -EFAULT;
1185                                 }
1186                         } else {
1187                                 memset(buff, 0, iocommand.buf_size);
1188                         }
1189                         if ((c = cmd_alloc(host, 0)) == NULL) {
1190                                 kfree(buff);
1191                                 return -ENOMEM;
1192                         }
1193                         // Fill in the command type
1194                         c->cmd_type = CMD_IOCTL_PEND;
1195                         // Fill in Command Header
1196                         c->Header.ReplyQueue = 0;       // unused in simple mode
1197                         if (iocommand.buf_size > 0)     // buffer to fill
1198                         {
1199                                 c->Header.SGList = 1;
1200                                 c->Header.SGTotal = 1;
1201                         } else  // no buffers to fill
1202                         {
1203                                 c->Header.SGList = 0;
1204                                 c->Header.SGTotal = 0;
1205                         }
1206                         c->Header.LUN = iocommand.LUN_info;
1207                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
1208
1209                         // Fill in Request block
1210                         c->Request = iocommand.Request;
1211
1212                         // Fill in the scatter gather information
1213                         if (iocommand.buf_size > 0) {
1214                                 temp64.val = pci_map_single(host->pdev, buff,
1215                                         iocommand.buf_size,
1216                                         PCI_DMA_BIDIRECTIONAL);
1217                                 c->SG[0].Addr.lower = temp64.val32.lower;
1218                                 c->SG[0].Addr.upper = temp64.val32.upper;
1219                                 c->SG[0].Len = iocommand.buf_size;
1220                                 c->SG[0].Ext = 0;       // we are not chaining
1221                         }
1222                         c->waiting = &wait;
1223
1224                         /* Put the request on the tail of the request queue */
1225                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1226                         addQ(&host->reqQ, c);
1227                         host->Qdepth++;
1228                         start_io(host);
1229                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1230
1231                         wait_for_completion(&wait);
1232
1233                         /* unlock the buffers from DMA */
1234                         temp64.val32.lower = c->SG[0].Addr.lower;
1235                         temp64.val32.upper = c->SG[0].Addr.upper;
1236                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1237                                          iocommand.buf_size,
1238                                          PCI_DMA_BIDIRECTIONAL);
1239
1240                         check_ioctl_unit_attention(host, c);
1241
1242                         /* Copy the error information out */
1243                         iocommand.error_info = *(c->err_info);
1244                         if (copy_to_user
1245                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1246                                 kfree(buff);
1247                                 cmd_free(host, c, 0);
1248                                 return -EFAULT;
1249                         }
1250
1251                         if (iocommand.Request.Type.Direction == XFER_READ) {
1252                                 /* Copy the data out of the buffer we created */
1253                                 if (copy_to_user
1254                                     (iocommand.buf, buff, iocommand.buf_size)) {
1255                                         kfree(buff);
1256                                         cmd_free(host, c, 0);
1257                                         return -EFAULT;
1258                                 }
1259                         }
1260                         kfree(buff);
1261                         cmd_free(host, c, 0);
1262                         return 0;
1263                 }
1264         case CCISS_BIG_PASSTHRU:{
1265                         BIG_IOCTL_Command_struct *ioc;
1266                         CommandList_struct *c;
1267                         unsigned char **buff = NULL;
1268                         int *buff_size = NULL;
1269                         u64bit temp64;
1270                         unsigned long flags;
1271                         BYTE sg_used = 0;
1272                         int status = 0;
1273                         int i;
1274                         DECLARE_COMPLETION_ONSTACK(wait);
1275                         __u32 left;
1276                         __u32 sz;
1277                         BYTE __user *data_ptr;
1278
1279                         if (!arg)
1280                                 return -EINVAL;
1281                         if (!capable(CAP_SYS_RAWIO))
1282                                 return -EPERM;
1283                         ioc = (BIG_IOCTL_Command_struct *)
1284                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1285                         if (!ioc) {
1286                                 status = -ENOMEM;
1287                                 goto cleanup1;
1288                         }
1289                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1290                                 status = -EFAULT;
1291                                 goto cleanup1;
1292                         }
1293                         if ((ioc->buf_size < 1) &&
1294                             (ioc->Request.Type.Direction != XFER_NONE)) {
1295                                 status = -EINVAL;
1296                                 goto cleanup1;
1297                         }
1298                         /* Check kmalloc limits  using all SGs */
1299                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1300                                 status = -EINVAL;
1301                                 goto cleanup1;
1302                         }
1303                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1304                                 status = -EINVAL;
1305                                 goto cleanup1;
1306                         }
1307                         buff =
1308                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1309                         if (!buff) {
1310                                 status = -ENOMEM;
1311                                 goto cleanup1;
1312                         }
1313                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1314                                                    GFP_KERNEL);
1315                         if (!buff_size) {
1316                                 status = -ENOMEM;
1317                                 goto cleanup1;
1318                         }
1319                         left = ioc->buf_size;
1320                         data_ptr = ioc->buf;
1321                         while (left) {
1322                                 sz = (left >
1323                                       ioc->malloc_size) ? ioc->
1324                                     malloc_size : left;
1325                                 buff_size[sg_used] = sz;
1326                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1327                                 if (buff[sg_used] == NULL) {
1328                                         status = -ENOMEM;
1329                                         goto cleanup1;
1330                                 }
1331                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1332                                         if (copy_from_user
1333                                             (buff[sg_used], data_ptr, sz)) {
1334                                                 status = -EFAULT;
1335                                                 goto cleanup1;
1336                                         }
1337                                 } else {
1338                                         memset(buff[sg_used], 0, sz);
1339                                 }
1340                                 left -= sz;
1341                                 data_ptr += sz;
1342                                 sg_used++;
1343                         }
1344                         if ((c = cmd_alloc(host, 0)) == NULL) {
1345                                 status = -ENOMEM;
1346                                 goto cleanup1;
1347                         }
1348                         c->cmd_type = CMD_IOCTL_PEND;
1349                         c->Header.ReplyQueue = 0;
1350
1351                         if (ioc->buf_size > 0) {
1352                                 c->Header.SGList = sg_used;
1353                                 c->Header.SGTotal = sg_used;
1354                         } else {
1355                                 c->Header.SGList = 0;
1356                                 c->Header.SGTotal = 0;
1357                         }
1358                         c->Header.LUN = ioc->LUN_info;
1359                         c->Header.Tag.lower = c->busaddr;
1360
1361                         c->Request = ioc->Request;
1362                         if (ioc->buf_size > 0) {
1363                                 int i;
1364                                 for (i = 0; i < sg_used; i++) {
1365                                         temp64.val =
1366                                             pci_map_single(host->pdev, buff[i],
1367                                                     buff_size[i],
1368                                                     PCI_DMA_BIDIRECTIONAL);
1369                                         c->SG[i].Addr.lower =
1370                                             temp64.val32.lower;
1371                                         c->SG[i].Addr.upper =
1372                                             temp64.val32.upper;
1373                                         c->SG[i].Len = buff_size[i];
1374                                         c->SG[i].Ext = 0;       /* we are not chaining */
1375                                 }
1376                         }
1377                         c->waiting = &wait;
1378                         /* Put the request on the tail of the request queue */
1379                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1380                         addQ(&host->reqQ, c);
1381                         host->Qdepth++;
1382                         start_io(host);
1383                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1384                         wait_for_completion(&wait);
1385                         /* unlock the buffers from DMA */
1386                         for (i = 0; i < sg_used; i++) {
1387                                 temp64.val32.lower = c->SG[i].Addr.lower;
1388                                 temp64.val32.upper = c->SG[i].Addr.upper;
1389                                 pci_unmap_single(host->pdev,
1390                                         (dma_addr_t) temp64.val, buff_size[i],
1391                                         PCI_DMA_BIDIRECTIONAL);
1392                         }
1393                         check_ioctl_unit_attention(host, c);
1394                         /* Copy the error information out */
1395                         ioc->error_info = *(c->err_info);
1396                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1397                                 cmd_free(host, c, 0);
1398                                 status = -EFAULT;
1399                                 goto cleanup1;
1400                         }
1401                         if (ioc->Request.Type.Direction == XFER_READ) {
1402                                 /* Copy the data out of the buffer we created */
1403                                 BYTE __user *ptr = ioc->buf;
1404                                 for (i = 0; i < sg_used; i++) {
1405                                         if (copy_to_user
1406                                             (ptr, buff[i], buff_size[i])) {
1407                                                 cmd_free(host, c, 0);
1408                                                 status = -EFAULT;
1409                                                 goto cleanup1;
1410                                         }
1411                                         ptr += buff_size[i];
1412                                 }
1413                         }
1414                         cmd_free(host, c, 0);
1415                         status = 0;
1416                       cleanup1:
1417                         if (buff) {
1418                                 for (i = 0; i < sg_used; i++)
1419                                         kfree(buff[i]);
1420                                 kfree(buff);
1421                         }
1422                         kfree(buff_size);
1423                         kfree(ioc);
1424                         return status;
1425                 }
1426
1427         /* scsi_cmd_ioctl handles these, below, though some are not */
1428         /* very meaningful for cciss.  SG_IO is the main one people want. */
1429
1430         case SG_GET_VERSION_NUM:
1431         case SG_SET_TIMEOUT:
1432         case SG_GET_TIMEOUT:
1433         case SG_GET_RESERVED_SIZE:
1434         case SG_SET_RESERVED_SIZE:
1435         case SG_EMULATED_HOST:
1436         case SG_IO:
1437         case SCSI_IOCTL_SEND_COMMAND:
1438                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1439
1440         /* scsi_cmd_ioctl would normally handle these, below, but */
1441         /* they aren't a good fit for cciss, as CD-ROMs are */
1442         /* not supported, and we don't have any bus/target/lun */
1443         /* which we present to the kernel. */
1444
1445         case CDROM_SEND_PACKET:
1446         case CDROMCLOSETRAY:
1447         case CDROMEJECT:
1448         case SCSI_IOCTL_GET_IDLUN:
1449         case SCSI_IOCTL_GET_BUS_NUMBER:
1450         default:
1451                 return -ENOTTY;
1452         }
1453 }
1454
1455 static void cciss_check_queues(ctlr_info_t *h)
1456 {
1457         int start_queue = h->next_to_run;
1458         int i;
1459
1460         /* check to see if we have maxed out the number of commands that can
1461          * be placed on the queue.  If so then exit.  We do this check here
1462          * in case the interrupt we serviced was from an ioctl and did not
1463          * free any new commands.
1464          */
1465         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1466                 return;
1467
1468         /* We have room on the queue for more commands.  Now we need to queue
1469          * them up.  We will also keep track of the next queue to run so
1470          * that every queue gets a chance to be started first.
1471          */
1472         for (i = 0; i < h->highest_lun + 1; i++) {
1473                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1474                 /* make sure the disk has been added and the drive is real
1475                  * because this can be called from the middle of init_one.
1476                  */
1477                 if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
1478                         continue;
1479                 blk_start_queue(h->gendisk[curr_queue]->queue);
1480
1481                 /* check to see if we have maxed out the number of commands
1482                  * that can be placed on the queue.
1483                  */
1484                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1485                         if (curr_queue == start_queue) {
1486                                 h->next_to_run =
1487                                     (start_queue + 1) % (h->highest_lun + 1);
1488                                 break;
1489                         } else {
1490                                 h->next_to_run = curr_queue;
1491                                 break;
1492                         }
1493                 }
1494         }
1495 }
1496
1497 static void cciss_softirq_done(struct request *rq)
1498 {
1499         CommandList_struct *cmd = rq->completion_data;
1500         ctlr_info_t *h = hba[cmd->ctlr];
1501         unsigned long flags;
1502         u64bit temp64;
1503         int i, ddir;
1504
1505         if (cmd->Request.Type.Direction == XFER_READ)
1506                 ddir = PCI_DMA_FROMDEVICE;
1507         else
1508                 ddir = PCI_DMA_TODEVICE;
1509
1510         /* command did not need to be retried */
1511         /* unmap the DMA mapping for all the scatter gather elements */
1512         for (i = 0; i < cmd->Header.SGList; i++) {
1513                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1514                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1515                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1516         }
1517
1518 #ifdef CCISS_DEBUG
1519         printk("Done with %p\n", rq);
1520 #endif                          /* CCISS_DEBUG */
1521
1522         /* set the residual count for pc requests */
1523         if (blk_pc_request(rq))
1524                 rq->resid_len = cmd->err_info->ResidualCnt;
1525
1526         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1527
1528         spin_lock_irqsave(&h->lock, flags);
1529         cmd_free(h, cmd, 1);
1530         cciss_check_queues(h);
1531         spin_unlock_irqrestore(&h->lock, flags);
1532 }
1533
1534 static void log_unit_to_scsi3addr(ctlr_info_t *h, unsigned char scsi3addr[],
1535         uint32_t log_unit)
1536 {
1537         log_unit = h->drv[log_unit].LunID & 0x03fff;
1538         memset(&scsi3addr[4], 0, 4);
1539         memcpy(&scsi3addr[0], &log_unit, 4);
1540         scsi3addr[3] |= 0x40;
1541 }
1542
1543 /* This function gets the SCSI vendor, model, and revision of a logical drive
1544  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1545  * they cannot be read.
1546  */
1547 static void cciss_get_device_descr(int ctlr, int logvol, int withirq,
1548                                    char *vendor, char *model, char *rev)
1549 {
1550         int rc;
1551         InquiryData_struct *inq_buf;
1552         unsigned char scsi3addr[8];
1553
1554         *vendor = '\0';
1555         *model = '\0';
1556         *rev = '\0';
1557
1558         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1559         if (!inq_buf)
1560                 return;
1561
1562         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1563         if (withirq)
1564                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf,
1565                              sizeof(InquiryData_struct), 0,
1566                                 scsi3addr, TYPE_CMD);
1567         else
1568                 rc = sendcmd(CISS_INQUIRY, ctlr, inq_buf,
1569                              sizeof(InquiryData_struct), 0,
1570                                 scsi3addr, TYPE_CMD);
1571         if (rc == IO_OK) {
1572                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1573                 vendor[VENDOR_LEN] = '\0';
1574                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1575                 model[MODEL_LEN] = '\0';
1576                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1577                 rev[REV_LEN] = '\0';
1578         }
1579
1580         kfree(inq_buf);
1581         return;
1582 }
1583
1584 /* This function gets the serial number of a logical drive via
1585  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1586  * number cannot be had, for whatever reason, 16 bytes of 0xff
1587  * are returned instead.
1588  */
1589 static void cciss_get_serial_no(int ctlr, int logvol, int withirq,
1590                                 unsigned char *serial_no, int buflen)
1591 {
1592 #define PAGE_83_INQ_BYTES 64
1593         int rc;
1594         unsigned char *buf;
1595         unsigned char scsi3addr[8];
1596
1597         if (buflen > 16)
1598                 buflen = 16;
1599         memset(serial_no, 0xff, buflen);
1600         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1601         if (!buf)
1602                 return;
1603         memset(serial_no, 0, buflen);
1604         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1605         if (withirq)
1606                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1607                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1608         else
1609                 rc = sendcmd(CISS_INQUIRY, ctlr, buf,
1610                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1611         if (rc == IO_OK)
1612                 memcpy(serial_no, &buf[8], buflen);
1613         kfree(buf);
1614         return;
1615 }
1616
1617 static void cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1618                                 int drv_index)
1619 {
1620         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1621         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1622         disk->major = h->major;
1623         disk->first_minor = drv_index << NWD_SHIFT;
1624         disk->fops = &cciss_fops;
1625         disk->private_data = &h->drv[drv_index];
1626         disk->driverfs_dev = &h->drv[drv_index].dev;
1627
1628         /* Set up queue information */
1629         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1630
1631         /* This is a hardware imposed limit. */
1632         blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1633
1634         /* This is a limit in the driver and could be eliminated. */
1635         blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1636
1637         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1638
1639         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1640
1641         disk->queue->queuedata = h;
1642
1643         blk_queue_logical_block_size(disk->queue,
1644                                      h->drv[drv_index].block_size);
1645
1646         /* Make sure all queue data is written out before */
1647         /* setting h->drv[drv_index].queue, as setting this */
1648         /* allows the interrupt handler to start the queue */
1649         wmb();
1650         h->drv[drv_index].queue = disk->queue;
1651         add_disk(disk);
1652 }
1653
1654 /* This function will check the usage_count of the drive to be updated/added.
1655  * If the usage_count is zero and it is a heretofore unknown drive, or,
1656  * the drive's capacity, geometry, or serial number has changed,
1657  * then the drive information will be updated and the disk will be
1658  * re-registered with the kernel.  If these conditions don't hold,
1659  * then it will be left alone for the next reboot.  The exception to this
1660  * is disk 0 which will always be left registered with the kernel since it
1661  * is also the controller node.  Any changes to disk 0 will show up on
1662  * the next reboot.
1663  */
1664 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time)
1665 {
1666         ctlr_info_t *h = hba[ctlr];
1667         struct gendisk *disk;
1668         InquiryData_struct *inq_buff = NULL;
1669         unsigned int block_size;
1670         sector_t total_size;
1671         unsigned long flags = 0;
1672         int ret = 0;
1673         drive_info_struct *drvinfo;
1674         int was_only_controller_node;
1675
1676         /* Get information about the disk and modify the driver structure */
1677         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1678         drvinfo = kmalloc(sizeof(*drvinfo), GFP_KERNEL);
1679         if (inq_buff == NULL || drvinfo == NULL)
1680                 goto mem_msg;
1681
1682         /* See if we're trying to update the "controller node"
1683          * this will happen the when the first logical drive gets
1684          * created by ACU.
1685          */
1686         was_only_controller_node = (drv_index == 0 &&
1687                                 h->drv[0].raid_level == -1);
1688
1689         /* testing to see if 16-byte CDBs are already being used */
1690         if (h->cciss_read == CCISS_READ_16) {
1691                 cciss_read_capacity_16(h->ctlr, drv_index, 1,
1692                         &total_size, &block_size);
1693
1694         } else {
1695                 cciss_read_capacity(ctlr, drv_index, 1,
1696                                     &total_size, &block_size);
1697
1698                 /* if read_capacity returns all F's this volume is >2TB */
1699                 /* in size so we switch to 16-byte CDB's for all */
1700                 /* read/write ops */
1701                 if (total_size == 0xFFFFFFFFULL) {
1702                         cciss_read_capacity_16(ctlr, drv_index, 1,
1703                         &total_size, &block_size);
1704                         h->cciss_read = CCISS_READ_16;
1705                         h->cciss_write = CCISS_WRITE_16;
1706                 } else {
1707                         h->cciss_read = CCISS_READ_10;
1708                         h->cciss_write = CCISS_WRITE_10;
1709                 }
1710         }
1711
1712         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1713                                inq_buff, drvinfo);
1714         drvinfo->block_size = block_size;
1715         drvinfo->nr_blocks = total_size + 1;
1716
1717         cciss_get_device_descr(ctlr, drv_index, 1, drvinfo->vendor,
1718                                 drvinfo->model, drvinfo->rev);
1719         cciss_get_serial_no(ctlr, drv_index, 1, drvinfo->serial_no,
1720                         sizeof(drvinfo->serial_no));
1721
1722         /* Is it the same disk we already know, and nothing's changed? */
1723         if (h->drv[drv_index].raid_level != -1 &&
1724                 ((memcmp(drvinfo->serial_no,
1725                                 h->drv[drv_index].serial_no, 16) == 0) &&
1726                 drvinfo->block_size == h->drv[drv_index].block_size &&
1727                 drvinfo->nr_blocks == h->drv[drv_index].nr_blocks &&
1728                 drvinfo->heads == h->drv[drv_index].heads &&
1729                 drvinfo->sectors == h->drv[drv_index].sectors &&
1730                 drvinfo->cylinders == h->drv[drv_index].cylinders))
1731                         /* The disk is unchanged, nothing to update */
1732                         goto freeret;
1733
1734         /* If we get here it's not the same disk, or something's changed,
1735          * so we need to * deregister it, and re-register it, if it's not
1736          * in use.
1737          * If the disk already exists then deregister it before proceeding
1738          * (unless it's the first disk (for the controller node).
1739          */
1740         if (h->drv[drv_index].raid_level != -1 && drv_index != 0) {
1741                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1742                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1743                 h->drv[drv_index].busy_configuring = 1;
1744                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1745
1746                 /* deregister_disk sets h->drv[drv_index].queue = NULL
1747                  * which keeps the interrupt handler from starting
1748                  * the queue.
1749                  */
1750                 ret = deregister_disk(h, drv_index, 0);
1751                 h->drv[drv_index].busy_configuring = 0;
1752         }
1753
1754         /* If the disk is in use return */
1755         if (ret)
1756                 goto freeret;
1757
1758         /* Save the new information from cciss_geometry_inquiry
1759          * and serial number inquiry.
1760          */
1761         h->drv[drv_index].block_size = drvinfo->block_size;
1762         h->drv[drv_index].nr_blocks = drvinfo->nr_blocks;
1763         h->drv[drv_index].heads = drvinfo->heads;
1764         h->drv[drv_index].sectors = drvinfo->sectors;
1765         h->drv[drv_index].cylinders = drvinfo->cylinders;
1766         h->drv[drv_index].raid_level = drvinfo->raid_level;
1767         memcpy(h->drv[drv_index].serial_no, drvinfo->serial_no, 16);
1768         memcpy(h->drv[drv_index].vendor, drvinfo->vendor, VENDOR_LEN + 1);
1769         memcpy(h->drv[drv_index].model, drvinfo->model, MODEL_LEN + 1);
1770         memcpy(h->drv[drv_index].rev, drvinfo->rev, REV_LEN + 1);
1771
1772         ++h->num_luns;
1773         disk = h->gendisk[drv_index];
1774         set_capacity(disk, h->drv[drv_index].nr_blocks);
1775
1776         /* If it's not disk 0 (drv_index != 0)
1777          * or if it was disk 0, but there was previously
1778          * no actual corresponding configured logical drive
1779          * (raid_leve == -1) then we want to update the
1780          * logical drive's information.
1781          */
1782         if (drv_index || first_time)
1783                 cciss_add_disk(h, disk, drv_index);
1784
1785 freeret:
1786         kfree(inq_buff);
1787         kfree(drvinfo);
1788         return;
1789 mem_msg:
1790         printk(KERN_ERR "cciss: out of memory\n");
1791         goto freeret;
1792 }
1793
1794 /* This function will find the first index of the controllers drive array
1795  * that has a -1 for the raid_level and will return that index.  This is
1796  * where new drives will be added.  If the index to be returned is greater
1797  * than the highest_lun index for the controller then highest_lun is set
1798  * to this new index.  If there are no available indexes then -1 is returned.
1799  * "controller_node" is used to know if this is a real logical drive, or just
1800  * the controller node, which determines if this counts towards highest_lun.
1801  */
1802 static int cciss_find_free_drive_index(int ctlr, int controller_node)
1803 {
1804         int i;
1805
1806         for (i = 0; i < CISS_MAX_LUN; i++) {
1807                 if (hba[ctlr]->drv[i].raid_level == -1) {
1808                         if (i > hba[ctlr]->highest_lun)
1809                                 if (!controller_node)
1810                                         hba[ctlr]->highest_lun = i;
1811                         return i;
1812                 }
1813         }
1814         return -1;
1815 }
1816
1817 /* cciss_add_gendisk finds a free hba[]->drv structure
1818  * and allocates a gendisk if needed, and sets the lunid
1819  * in the drvinfo structure.   It returns the index into
1820  * the ->drv[] array, or -1 if none are free.
1821  * is_controller_node indicates whether highest_lun should
1822  * count this disk, or if it's only being added to provide
1823  * a means to talk to the controller in case no logical
1824  * drives have yet been configured.
1825  */
1826 static int cciss_add_gendisk(ctlr_info_t *h, __u32 lunid, int controller_node)
1827 {
1828         int drv_index;
1829
1830         drv_index = cciss_find_free_drive_index(h->ctlr, controller_node);
1831         if (drv_index == -1)
1832                 return -1;
1833         /*Check if the gendisk needs to be allocated */
1834         if (!h->gendisk[drv_index]) {
1835                 h->gendisk[drv_index] =
1836                         alloc_disk(1 << NWD_SHIFT);
1837                 if (!h->gendisk[drv_index]) {
1838                         printk(KERN_ERR "cciss%d: could not "
1839                                 "allocate a new disk %d\n",
1840                                 h->ctlr, drv_index);
1841                         return -1;
1842                 }
1843         }
1844         h->drv[drv_index].LunID = lunid;
1845         if (cciss_create_ld_sysfs_entry(h, &h->drv[drv_index], drv_index))
1846                 goto err_free_disk;
1847
1848         /* Don't need to mark this busy because nobody */
1849         /* else knows about this disk yet to contend */
1850         /* for access to it. */
1851         h->drv[drv_index].busy_configuring = 0;
1852         wmb();
1853         return drv_index;
1854
1855 err_free_disk:
1856         put_disk(h->gendisk[drv_index]);
1857         h->gendisk[drv_index] = NULL;
1858         return -1;
1859 }
1860
1861 /* This is for the special case of a controller which
1862  * has no logical drives.  In this case, we still need
1863  * to register a disk so the controller can be accessed
1864  * by the Array Config Utility.
1865  */
1866 static void cciss_add_controller_node(ctlr_info_t *h)
1867 {
1868         struct gendisk *disk;
1869         int drv_index;
1870
1871         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
1872                 return;
1873
1874         drv_index = cciss_add_gendisk(h, 0, 1);
1875         if (drv_index == -1) {
1876                 printk(KERN_WARNING "cciss%d: could not "
1877                         "add disk 0.\n", h->ctlr);
1878                 return;
1879         }
1880         h->drv[drv_index].block_size = 512;
1881         h->drv[drv_index].nr_blocks = 0;
1882         h->drv[drv_index].heads = 0;
1883         h->drv[drv_index].sectors = 0;
1884         h->drv[drv_index].cylinders = 0;
1885         h->drv[drv_index].raid_level = -1;
1886         memset(h->drv[drv_index].serial_no, 0, 16);
1887         disk = h->gendisk[drv_index];
1888         cciss_add_disk(h, disk, drv_index);
1889 }
1890
1891 /* This function will add and remove logical drives from the Logical
1892  * drive array of the controller and maintain persistency of ordering
1893  * so that mount points are preserved until the next reboot.  This allows
1894  * for the removal of logical drives in the middle of the drive array
1895  * without a re-ordering of those drives.
1896  * INPUT
1897  * h            = The controller to perform the operations on
1898  */
1899 static int rebuild_lun_table(ctlr_info_t *h, int first_time)
1900 {
1901         int ctlr = h->ctlr;
1902         int num_luns;
1903         ReportLunData_struct *ld_buff = NULL;
1904         int return_code;
1905         int listlength = 0;
1906         int i;
1907         int drv_found;
1908         int drv_index = 0;
1909         __u32 lunid = 0;
1910         unsigned long flags;
1911
1912         if (!capable(CAP_SYS_RAWIO))
1913                 return -EPERM;
1914
1915         /* Set busy_configuring flag for this operation */
1916         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1917         if (h->busy_configuring) {
1918                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1919                 return -EBUSY;
1920         }
1921         h->busy_configuring = 1;
1922         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1923
1924         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1925         if (ld_buff == NULL)
1926                 goto mem_msg;
1927
1928         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
1929                                       sizeof(ReportLunData_struct),
1930                                       0, CTLR_LUNID, TYPE_CMD);
1931
1932         if (return_code == IO_OK)
1933                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
1934         else {  /* reading number of logical volumes failed */
1935                 printk(KERN_WARNING "cciss: report logical volume"
1936                        " command failed\n");
1937                 listlength = 0;
1938                 goto freeret;
1939         }
1940
1941         num_luns = listlength / 8;      /* 8 bytes per entry */
1942         if (num_luns > CISS_MAX_LUN) {
1943                 num_luns = CISS_MAX_LUN;
1944                 printk(KERN_WARNING "cciss: more luns configured"
1945                        " on controller than can be handled by"
1946                        " this driver.\n");
1947         }
1948
1949         if (num_luns == 0)
1950                 cciss_add_controller_node(h);
1951
1952         /* Compare controller drive array to driver's drive array
1953          * to see if any drives are missing on the controller due
1954          * to action of Array Config Utility (user deletes drive)
1955          * and deregister logical drives which have disappeared.
1956          */
1957         for (i = 0; i <= h->highest_lun; i++) {
1958                 int j;
1959                 drv_found = 0;
1960
1961                 /* skip holes in the array from already deleted drives */
1962                 if (h->drv[i].raid_level == -1)
1963                         continue;
1964
1965                 for (j = 0; j < num_luns; j++) {
1966                         memcpy(&lunid, &ld_buff->LUN[j][0], 4);
1967                         lunid = le32_to_cpu(lunid);
1968                         if (h->drv[i].LunID == lunid) {
1969                                 drv_found = 1;
1970                                 break;
1971                         }
1972                 }
1973                 if (!drv_found) {
1974                         /* Deregister it from the OS, it's gone. */
1975                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1976                         h->drv[i].busy_configuring = 1;
1977                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1978                         return_code = deregister_disk(h, i, 1);
1979                         cciss_destroy_ld_sysfs_entry(&h->drv[i]);
1980                         h->drv[i].busy_configuring = 0;
1981                 }
1982         }
1983
1984         /* Compare controller drive array to driver's drive array.
1985          * Check for updates in the drive information and any new drives
1986          * on the controller due to ACU adding logical drives, or changing
1987          * a logical drive's size, etc.  Reregister any new/changed drives
1988          */
1989         for (i = 0; i < num_luns; i++) {
1990                 int j;
1991
1992                 drv_found = 0;
1993
1994                 memcpy(&lunid, &ld_buff->LUN[i][0], 4);
1995                 lunid = le32_to_cpu(lunid);
1996
1997                 /* Find if the LUN is already in the drive array
1998                  * of the driver.  If so then update its info
1999                  * if not in use.  If it does not exist then find
2000                  * the first free index and add it.
2001                  */
2002                 for (j = 0; j <= h->highest_lun; j++) {
2003                         if (h->drv[j].raid_level != -1 &&
2004                                 h->drv[j].LunID == lunid) {
2005                                 drv_index = j;
2006                                 drv_found = 1;
2007                                 break;
2008                         }
2009                 }
2010
2011                 /* check if the drive was found already in the array */
2012                 if (!drv_found) {
2013                         drv_index = cciss_add_gendisk(h, lunid, 0);
2014                         if (drv_index == -1)
2015                                 goto freeret;
2016                 }
2017                 cciss_update_drive_info(ctlr, drv_index, first_time);
2018         }               /* end for */
2019
2020 freeret:
2021         kfree(ld_buff);
2022         h->busy_configuring = 0;
2023         /* We return -1 here to tell the ACU that we have registered/updated
2024          * all of the drives that we can and to keep it from calling us
2025          * additional times.
2026          */
2027         return -1;
2028 mem_msg:
2029         printk(KERN_ERR "cciss: out of memory\n");
2030         h->busy_configuring = 0;
2031         goto freeret;
2032 }
2033
2034 /* This function will deregister the disk and it's queue from the
2035  * kernel.  It must be called with the controller lock held and the
2036  * drv structures busy_configuring flag set.  It's parameters are:
2037  *
2038  * disk = This is the disk to be deregistered
2039  * drv  = This is the drive_info_struct associated with the disk to be
2040  *        deregistered.  It contains information about the disk used
2041  *        by the driver.
2042  * clear_all = This flag determines whether or not the disk information
2043  *             is going to be completely cleared out and the highest_lun
2044  *             reset.  Sometimes we want to clear out information about
2045  *             the disk in preparation for re-adding it.  In this case
2046  *             the highest_lun should be left unchanged and the LunID
2047  *             should not be cleared.
2048 */
2049 static int deregister_disk(ctlr_info_t *h, int drv_index,
2050                            int clear_all)
2051 {
2052         int i;
2053         struct gendisk *disk;
2054         drive_info_struct *drv;
2055
2056         if (!capable(CAP_SYS_RAWIO))
2057                 return -EPERM;
2058
2059         drv = &h->drv[drv_index];
2060         disk = h->gendisk[drv_index];
2061
2062         /* make sure logical volume is NOT is use */
2063         if (clear_all || (h->gendisk[0] == disk)) {
2064                 if (drv->usage_count > 1)
2065                         return -EBUSY;
2066         } else if (drv->usage_count > 0)
2067                 return -EBUSY;
2068
2069         /* invalidate the devices and deregister the disk.  If it is disk
2070          * zero do not deregister it but just zero out it's values.  This
2071          * allows us to delete disk zero but keep the controller registered.
2072          */
2073         if (h->gendisk[0] != disk) {
2074                 struct request_queue *q = disk->queue;
2075                 if (disk->flags & GENHD_FL_UP)
2076                         del_gendisk(disk);
2077                 if (q) {
2078                         blk_cleanup_queue(q);
2079                         /* Set drv->queue to NULL so that we do not try
2080                          * to call blk_start_queue on this queue in the
2081                          * interrupt handler
2082                          */
2083                         drv->queue = NULL;
2084                 }
2085                 /* If clear_all is set then we are deleting the logical
2086                  * drive, not just refreshing its info.  For drives
2087                  * other than disk 0 we will call put_disk.  We do not
2088                  * do this for disk 0 as we need it to be able to
2089                  * configure the controller.
2090                  */
2091                 if (clear_all){
2092                         /* This isn't pretty, but we need to find the
2093                          * disk in our array and NULL our the pointer.
2094                          * This is so that we will call alloc_disk if
2095                          * this index is used again later.
2096                          */
2097                         for (i=0; i < CISS_MAX_LUN; i++){
2098                                 if (h->gendisk[i] == disk) {
2099                                         h->gendisk[i] = NULL;
2100                                         break;
2101                                 }
2102                         }
2103                         put_disk(disk);
2104                 }
2105         } else {
2106                 set_capacity(disk, 0);
2107         }
2108
2109         --h->num_luns;
2110         /* zero out the disk size info */
2111         drv->nr_blocks = 0;
2112         drv->block_size = 0;
2113         drv->heads = 0;
2114         drv->sectors = 0;
2115         drv->cylinders = 0;
2116         drv->raid_level = -1;   /* This can be used as a flag variable to
2117                                  * indicate that this element of the drive
2118                                  * array is free.
2119                                  */
2120
2121         if (clear_all) {
2122                 /* check to see if it was the last disk */
2123                 if (drv == h->drv + h->highest_lun) {
2124                         /* if so, find the new hightest lun */
2125                         int i, newhighest = -1;
2126                         for (i = 0; i <= h->highest_lun; i++) {
2127                                 /* if the disk has size > 0, it is available */
2128                                 if (h->drv[i].heads)
2129                                         newhighest = i;
2130                         }
2131                         h->highest_lun = newhighest;
2132                 }
2133
2134                 drv->LunID = 0;
2135         }
2136         return 0;
2137 }
2138
2139 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2140                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2141                 int cmd_type)
2142 {
2143         ctlr_info_t *h = hba[ctlr];
2144         u64bit buff_dma_handle;
2145         int status = IO_OK;
2146
2147         c->cmd_type = CMD_IOCTL_PEND;
2148         c->Header.ReplyQueue = 0;
2149         if (buff != NULL) {
2150                 c->Header.SGList = 1;
2151                 c->Header.SGTotal = 1;
2152         } else {
2153                 c->Header.SGList = 0;
2154                 c->Header.SGTotal = 0;
2155         }
2156         c->Header.Tag.lower = c->busaddr;
2157         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2158
2159         c->Request.Type.Type = cmd_type;
2160         if (cmd_type == TYPE_CMD) {
2161                 switch (cmd) {
2162                 case CISS_INQUIRY:
2163                         /* are we trying to read a vital product page */
2164                         if (page_code != 0) {
2165                                 c->Request.CDB[1] = 0x01;
2166                                 c->Request.CDB[2] = page_code;
2167                         }
2168                         c->Request.CDBLen = 6;
2169                         c->Request.Type.Attribute = ATTR_SIMPLE;
2170                         c->Request.Type.Direction = XFER_READ;
2171                         c->Request.Timeout = 0;
2172                         c->Request.CDB[0] = CISS_INQUIRY;
2173                         c->Request.CDB[4] = size & 0xFF;
2174                         break;
2175                 case CISS_REPORT_LOG:
2176                 case CISS_REPORT_PHYS:
2177                         /* Talking to controller so It's a physical command
2178                            mode = 00 target = 0.  Nothing to write.
2179                          */
2180                         c->Request.CDBLen = 12;
2181                         c->Request.Type.Attribute = ATTR_SIMPLE;
2182                         c->Request.Type.Direction = XFER_READ;
2183                         c->Request.Timeout = 0;
2184                         c->Request.CDB[0] = cmd;
2185                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
2186                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2187                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2188                         c->Request.CDB[9] = size & 0xFF;
2189                         break;
2190
2191                 case CCISS_READ_CAPACITY:
2192                         c->Request.CDBLen = 10;
2193                         c->Request.Type.Attribute = ATTR_SIMPLE;
2194                         c->Request.Type.Direction = XFER_READ;
2195                         c->Request.Timeout = 0;
2196                         c->Request.CDB[0] = cmd;
2197                         break;
2198                 case CCISS_READ_CAPACITY_16:
2199                         c->Request.CDBLen = 16;
2200                         c->Request.Type.Attribute = ATTR_SIMPLE;
2201                         c->Request.Type.Direction = XFER_READ;
2202                         c->Request.Timeout = 0;
2203                         c->Request.CDB[0] = cmd;
2204                         c->Request.CDB[1] = 0x10;
2205                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2206                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2207                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2208                         c->Request.CDB[13] = size & 0xFF;
2209                         c->Request.Timeout = 0;
2210                         c->Request.CDB[0] = cmd;
2211                         break;
2212                 case CCISS_CACHE_FLUSH:
2213                         c->Request.CDBLen = 12;
2214                         c->Request.Type.Attribute = ATTR_SIMPLE;
2215                         c->Request.Type.Direction = XFER_WRITE;
2216                         c->Request.Timeout = 0;
2217                         c->Request.CDB[0] = BMIC_WRITE;
2218                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2219                         break;
2220                 case TEST_UNIT_READY:
2221                         c->Request.CDBLen = 6;
2222                         c->Request.Type.Attribute = ATTR_SIMPLE;
2223                         c->Request.Type.Direction = XFER_NONE;
2224                         c->Request.Timeout = 0;
2225                         break;
2226                 default:
2227                         printk(KERN_WARNING
2228                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2229                         return IO_ERROR;
2230                 }
2231         } else if (cmd_type == TYPE_MSG) {
2232                 switch (cmd) {
2233                 case 0: /* ABORT message */
2234                         c->Request.CDBLen = 12;
2235                         c->Request.Type.Attribute = ATTR_SIMPLE;
2236                         c->Request.Type.Direction = XFER_WRITE;
2237                         c->Request.Timeout = 0;
2238                         c->Request.CDB[0] = cmd;        /* abort */
2239                         c->Request.CDB[1] = 0;  /* abort a command */
2240                         /* buff contains the tag of the command to abort */
2241                         memcpy(&c->Request.CDB[4], buff, 8);
2242                         break;
2243                 case 1: /* RESET message */
2244                         c->Request.CDBLen = 16;
2245                         c->Request.Type.Attribute = ATTR_SIMPLE;
2246                         c->Request.Type.Direction = XFER_NONE;
2247                         c->Request.Timeout = 0;
2248                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2249                         c->Request.CDB[0] = cmd;        /* reset */
2250                         c->Request.CDB[1] = 0x03;       /* reset a target */
2251                         break;
2252                 case 3: /* No-Op message */
2253                         c->Request.CDBLen = 1;
2254                         c->Request.Type.Attribute = ATTR_SIMPLE;
2255                         c->Request.Type.Direction = XFER_WRITE;
2256                         c->Request.Timeout = 0;
2257                         c->Request.CDB[0] = cmd;
2258                         break;
2259                 default:
2260                         printk(KERN_WARNING
2261                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2262                         return IO_ERROR;
2263                 }
2264         } else {
2265                 printk(KERN_WARNING
2266                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2267                 return IO_ERROR;
2268         }
2269         /* Fill in the scatter gather information */
2270         if (size > 0) {
2271                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2272                                                              buff, size,
2273                                                              PCI_DMA_BIDIRECTIONAL);
2274                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2275                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2276                 c->SG[0].Len = size;
2277                 c->SG[0].Ext = 0;       /* we are not chaining */
2278         }
2279         return status;
2280 }
2281
2282 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2283 {
2284         switch (c->err_info->ScsiStatus) {
2285         case SAM_STAT_GOOD:
2286                 return IO_OK;
2287         case SAM_STAT_CHECK_CONDITION:
2288                 switch (0xf & c->err_info->SenseInfo[2]) {
2289                 case 0: return IO_OK; /* no sense */
2290                 case 1: return IO_OK; /* recovered error */
2291                 default:
2292                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2293                                 "check condition, sense key = 0x%02x\n",
2294                                 h->ctlr, c->Request.CDB[0],
2295                                 c->err_info->SenseInfo[2]);
2296                 }
2297                 break;
2298         default:
2299                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2300                         "scsi status = 0x%02x\n", h->ctlr,
2301                         c->Request.CDB[0], c->err_info->ScsiStatus);
2302                 break;
2303         }
2304         return IO_ERROR;
2305 }
2306
2307 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2308 {
2309         int return_status = IO_OK;
2310
2311         if (c->err_info->CommandStatus == CMD_SUCCESS)
2312                 return IO_OK;
2313
2314         switch (c->err_info->CommandStatus) {
2315         case CMD_TARGET_STATUS:
2316                 return_status = check_target_status(h, c);
2317                 break;
2318         case CMD_DATA_UNDERRUN:
2319         case CMD_DATA_OVERRUN:
2320                 /* expected for inquiry and report lun commands */
2321                 break;
2322         case CMD_INVALID:
2323                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2324                        "reported invalid\n", c->Request.CDB[0]);
2325                 return_status = IO_ERROR;
2326                 break;
2327         case CMD_PROTOCOL_ERR:
2328                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2329                        "protocol error \n", c->Request.CDB[0]);
2330                 return_status = IO_ERROR;
2331                 break;
2332         case CMD_HARDWARE_ERR:
2333                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2334                        " hardware error\n", c->Request.CDB[0]);
2335                 return_status = IO_ERROR;
2336                 break;
2337         case CMD_CONNECTION_LOST:
2338                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2339                        "connection lost\n", c->Request.CDB[0]);
2340                 return_status = IO_ERROR;
2341                 break;
2342         case CMD_ABORTED:
2343                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2344                        "aborted\n", c->Request.CDB[0]);
2345                 return_status = IO_ERROR;
2346                 break;
2347         case CMD_ABORT_FAILED:
2348                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2349                        "abort failed\n", c->Request.CDB[0]);
2350                 return_status = IO_ERROR;
2351                 break;
2352         case CMD_UNSOLICITED_ABORT:
2353                 printk(KERN_WARNING
2354                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2355                         c->Request.CDB[0]);
2356                 return_status = IO_NEEDS_RETRY;
2357                 break;
2358         default:
2359                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2360                        "unknown status %x\n", c->Request.CDB[0],
2361                        c->err_info->CommandStatus);
2362                 return_status = IO_ERROR;
2363         }
2364         return return_status;
2365 }
2366
2367 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2368         int attempt_retry)
2369 {
2370         DECLARE_COMPLETION_ONSTACK(wait);
2371         u64bit buff_dma_handle;
2372         unsigned long flags;
2373         int return_status = IO_OK;
2374
2375 resend_cmd2:
2376         c->waiting = &wait;
2377         /* Put the request on the tail of the queue and send it */
2378         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2379         addQ(&h->reqQ, c);
2380         h->Qdepth++;
2381         start_io(h);
2382         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2383
2384         wait_for_completion(&wait);
2385
2386         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2387                 goto command_done;
2388
2389         return_status = process_sendcmd_error(h, c);
2390
2391         if (return_status == IO_NEEDS_RETRY &&
2392                 c->retry_count < MAX_CMD_RETRIES) {
2393                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2394                         c->Request.CDB[0]);
2395                 c->retry_count++;
2396                 /* erase the old error information */
2397                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2398                 return_status = IO_OK;
2399                 INIT_COMPLETION(wait);
2400                 goto resend_cmd2;
2401         }
2402
2403 command_done:
2404         /* unlock the buffers from DMA */
2405         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2406         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2407         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2408                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2409         return return_status;
2410 }
2411
2412 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2413                            __u8 page_code, unsigned char scsi3addr[],
2414                         int cmd_type)
2415 {
2416         ctlr_info_t *h = hba[ctlr];
2417         CommandList_struct *c;
2418         int return_status;
2419
2420         c = cmd_alloc(h, 0);
2421         if (!c)
2422                 return -ENOMEM;
2423         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2424                 scsi3addr, cmd_type);
2425         if (return_status == IO_OK)
2426                 return_status = sendcmd_withirq_core(h, c, 1);
2427
2428         cmd_free(h, c, 0);
2429         return return_status;
2430 }
2431
2432 static void cciss_geometry_inquiry(int ctlr, int logvol,
2433                                    int withirq, sector_t total_size,
2434                                    unsigned int block_size,
2435                                    InquiryData_struct *inq_buff,
2436                                    drive_info_struct *drv)
2437 {
2438         int return_code;
2439         unsigned long t;
2440         unsigned char scsi3addr[8];
2441
2442         memset(inq_buff, 0, sizeof(InquiryData_struct));
2443         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2444         if (withirq)
2445                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
2446                                               inq_buff, sizeof(*inq_buff),
2447                                               0xC1, scsi3addr, TYPE_CMD);
2448         else
2449                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
2450                                       sizeof(*inq_buff), 0xC1, scsi3addr,
2451                                       TYPE_CMD);
2452         if (return_code == IO_OK) {
2453                 if (inq_buff->data_byte[8] == 0xFF) {
2454                         printk(KERN_WARNING
2455                                "cciss: reading geometry failed, volume "
2456                                "does not support reading geometry\n");
2457                         drv->heads = 255;
2458                         drv->sectors = 32;      // Sectors per track
2459                         drv->cylinders = total_size + 1;
2460                         drv->raid_level = RAID_UNKNOWN;
2461                 } else {
2462                         drv->heads = inq_buff->data_byte[6];
2463                         drv->sectors = inq_buff->data_byte[7];
2464                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2465                         drv->cylinders += inq_buff->data_byte[5];
2466                         drv->raid_level = inq_buff->data_byte[8];
2467                 }
2468                 drv->block_size = block_size;
2469                 drv->nr_blocks = total_size + 1;
2470                 t = drv->heads * drv->sectors;
2471                 if (t > 1) {
2472                         sector_t real_size = total_size + 1;
2473                         unsigned long rem = sector_div(real_size, t);
2474                         if (rem)
2475                                 real_size++;
2476                         drv->cylinders = real_size;
2477                 }
2478         } else {                /* Get geometry failed */
2479                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2480         }
2481         printk(KERN_INFO "      heads=%d, sectors=%d, cylinders=%d\n\n",
2482                drv->heads, drv->sectors, drv->cylinders);
2483 }
2484
2485 static void
2486 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
2487                     unsigned int *block_size)
2488 {
2489         ReadCapdata_struct *buf;
2490         int return_code;
2491         unsigned char scsi3addr[8];
2492
2493         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2494         if (!buf) {
2495                 printk(KERN_WARNING "cciss: out of memory\n");
2496                 return;
2497         }
2498
2499         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2500         if (withirq)
2501                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
2502                                 ctlr, buf, sizeof(ReadCapdata_struct),
2503                                         0, scsi3addr, TYPE_CMD);
2504         else
2505                 return_code = sendcmd(CCISS_READ_CAPACITY,
2506                                 ctlr, buf, sizeof(ReadCapdata_struct),
2507                                         0, scsi3addr, TYPE_CMD);
2508         if (return_code == IO_OK) {
2509                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2510                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2511         } else {                /* read capacity command failed */
2512                 printk(KERN_WARNING "cciss: read capacity failed\n");
2513                 *total_size = 0;
2514                 *block_size = BLOCK_SIZE;
2515         }
2516         if (*total_size != 0)
2517                 printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2518                 (unsigned long long)*total_size+1, *block_size);
2519         kfree(buf);
2520 }
2521
2522 static void
2523 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size,                                 unsigned int *block_size)
2524 {
2525         ReadCapdata_struct_16 *buf;
2526         int return_code;
2527         unsigned char scsi3addr[8];
2528
2529         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2530         if (!buf) {
2531                 printk(KERN_WARNING "cciss: out of memory\n");
2532                 return;
2533         }
2534
2535         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2536         if (withirq) {
2537                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2538                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2539                                 0, scsi3addr, TYPE_CMD);
2540         }
2541         else {
2542                 return_code = sendcmd(CCISS_READ_CAPACITY_16,
2543                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2544                                 0, scsi3addr, TYPE_CMD);
2545         }
2546         if (return_code == IO_OK) {
2547                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2548                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2549         } else {                /* read capacity command failed */
2550                 printk(KERN_WARNING "cciss: read capacity failed\n");
2551                 *total_size = 0;
2552                 *block_size = BLOCK_SIZE;
2553         }
2554         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2555                (unsigned long long)*total_size+1, *block_size);
2556         kfree(buf);
2557 }
2558
2559 static int cciss_revalidate(struct gendisk *disk)
2560 {
2561         ctlr_info_t *h = get_host(disk);
2562         drive_info_struct *drv = get_drv(disk);
2563         int logvol;
2564         int FOUND = 0;
2565         unsigned int block_size;
2566         sector_t total_size;
2567         InquiryData_struct *inq_buff = NULL;
2568
2569         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2570                 if (h->drv[logvol].LunID == drv->LunID) {
2571                         FOUND = 1;
2572                         break;
2573                 }
2574         }
2575
2576         if (!FOUND)
2577                 return 1;
2578
2579         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2580         if (inq_buff == NULL) {
2581                 printk(KERN_WARNING "cciss: out of memory\n");
2582                 return 1;
2583         }
2584         if (h->cciss_read == CCISS_READ_10) {
2585                 cciss_read_capacity(h->ctlr, logvol, 1,
2586                                         &total_size, &block_size);
2587         } else {
2588                 cciss_read_capacity_16(h->ctlr, logvol, 1,
2589                                         &total_size, &block_size);
2590         }
2591         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2592                                inq_buff, drv);
2593
2594         blk_queue_logical_block_size(drv->queue, drv->block_size);
2595         set_capacity(disk, drv->nr_blocks);
2596
2597         kfree(inq_buff);
2598         return 0;
2599 }
2600
2601 /*
2602  *   Wait polling for a command to complete.
2603  *   The memory mapped FIFO is polled for the completion.
2604  *   Used only at init time, interrupts from the HBA are disabled.
2605  */
2606 static unsigned long pollcomplete(int ctlr)
2607 {
2608         unsigned long done;
2609         int i;
2610
2611         /* Wait (up to 20 seconds) for a command to complete */
2612
2613         for (i = 20 * HZ; i > 0; i--) {
2614                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2615                 if (done == FIFO_EMPTY)
2616                         schedule_timeout_uninterruptible(1);
2617                 else
2618                         return done;
2619         }
2620         /* Invalid address to tell caller we ran out of time */
2621         return 1;
2622 }
2623
2624 /* Send command c to controller h and poll for it to complete.
2625  * Turns interrupts off on the board.  Used at driver init time
2626  * and during SCSI error recovery.
2627  */
2628 static int sendcmd_core(ctlr_info_t *h, CommandList_struct *c)
2629 {
2630         int i;
2631         unsigned long complete;
2632         int status = IO_ERROR;
2633         u64bit buff_dma_handle;
2634
2635 resend_cmd1:
2636
2637         /* Disable interrupt on the board. */
2638         h->access.set_intr_mask(h, CCISS_INTR_OFF);
2639
2640         /* Make sure there is room in the command FIFO */
2641         /* Actually it should be completely empty at this time */
2642         /* unless we are in here doing error handling for the scsi */
2643         /* tape side of the driver. */
2644         for (i = 200000; i > 0; i--) {
2645                 /* if fifo isn't full go */
2646                 if (!(h->access.fifo_full(h)))
2647                         break;
2648                 udelay(10);
2649                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2650                        " waiting!\n", h->ctlr);
2651         }
2652         h->access.submit_command(h, c); /* Send the cmd */
2653         do {
2654                 complete = pollcomplete(h->ctlr);
2655
2656 #ifdef CCISS_DEBUG
2657                 printk(KERN_DEBUG "cciss: command completed\n");
2658 #endif                          /* CCISS_DEBUG */
2659
2660                 if (complete == 1) {
2661                         printk(KERN_WARNING
2662                                "cciss cciss%d: SendCmd Timeout out, "
2663                                "No command list address returned!\n", h->ctlr);
2664                         status = IO_ERROR;
2665                         break;
2666                 }
2667
2668                 /* Make sure it's the command we're expecting. */
2669                 if ((complete & ~CISS_ERROR_BIT) != c->busaddr) {
2670                         printk(KERN_WARNING "cciss%d: Unexpected command "
2671                                 "completion.\n", h->ctlr);
2672                         continue;
2673                 }
2674
2675                 /* It is our command.  If no error, we're done. */
2676                 if (!(complete & CISS_ERROR_BIT)) {
2677                         status = IO_OK;
2678                         break;
2679                 }
2680
2681                 /* There is an error... */
2682
2683                 /* if data overrun or underun on Report command ignore it */
2684                 if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2685                      (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2686                      (c->Request.CDB[0] == CISS_INQUIRY)) &&
2687                         ((c->err_info->CommandStatus == CMD_DATA_OVERRUN) ||
2688                          (c->err_info->CommandStatus == CMD_DATA_UNDERRUN))) {
2689                         complete = c->busaddr;
2690                         status = IO_OK;
2691                         break;
2692                 }
2693
2694                 if (c->err_info->CommandStatus == CMD_UNSOLICITED_ABORT) {
2695                         printk(KERN_WARNING "cciss%d: unsolicited abort %p\n",
2696                                 h->ctlr, c);
2697                         if (c->retry_count < MAX_CMD_RETRIES) {
2698                                 printk(KERN_WARNING "cciss%d: retrying %p\n",
2699                                    h->ctlr, c);
2700                                 c->retry_count++;
2701                                 /* erase the old error information */
2702                                 memset(c->err_info, 0, sizeof(c->err_info));
2703                                 goto resend_cmd1;
2704                         }
2705                         printk(KERN_WARNING "cciss%d: retried %p too many "
2706                                 "times\n", h->ctlr, c);
2707                         status = IO_ERROR;
2708                         break;
2709                 }
2710
2711                 if (c->err_info->CommandStatus == CMD_UNABORTABLE) {
2712                         printk(KERN_WARNING "cciss%d: command could not be "
2713                                 "aborted.\n", h->ctlr);
2714                         status = IO_ERROR;
2715                         break;
2716                 }
2717
2718                 if (c->err_info->CommandStatus == CMD_TARGET_STATUS) {
2719                         status = check_target_status(h, c);
2720                         break;
2721                 }
2722
2723                 printk(KERN_WARNING "cciss%d: sendcmd error\n", h->ctlr);
2724                 printk(KERN_WARNING "cmd = 0x%02x, CommandStatus = 0x%02x\n",
2725                         c->Request.CDB[0], c->err_info->CommandStatus);
2726                 status = IO_ERROR;
2727                 break;
2728
2729         } while (1);
2730
2731         /* unlock the data buffer from DMA */
2732         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2733         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2734         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2735                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2736         return status;
2737 }
2738
2739 /*
2740  * Send a command to the controller, and wait for it to complete.
2741  * Used at init time, and during SCSI error recovery.
2742  */
2743 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
2744         __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2745 {
2746         CommandList_struct *c;
2747         int status;
2748
2749         c = cmd_alloc(hba[ctlr], 1);
2750         if (!c) {
2751                 printk(KERN_WARNING "cciss: unable to get memory");
2752                 return IO_ERROR;
2753         }
2754         status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2755                 scsi3addr, cmd_type);
2756         if (status == IO_OK)
2757                 status = sendcmd_core(hba[ctlr], c);
2758         cmd_free(hba[ctlr], c, 1);
2759         return status;
2760 }
2761
2762 /*
2763  * Map (physical) PCI mem into (virtual) kernel space
2764  */
2765 static void __iomem *remap_pci_mem(ulong base, ulong size)
2766 {
2767         ulong page_base = ((ulong) base) & PAGE_MASK;
2768         ulong page_offs = ((ulong) base) - page_base;
2769         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2770
2771         return page_remapped ? (page_remapped + page_offs) : NULL;
2772 }
2773
2774 /*
2775  * Takes jobs of the Q and sends them to the hardware, then puts it on
2776  * the Q to wait for completion.
2777  */
2778 static void start_io(ctlr_info_t *h)
2779 {
2780         CommandList_struct *c;
2781
2782         while (!hlist_empty(&h->reqQ)) {
2783                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2784                 /* can't do anything if fifo is full */
2785                 if ((h->access.fifo_full(h))) {
2786                         printk(KERN_WARNING "cciss: fifo full\n");
2787                         break;
2788                 }
2789
2790                 /* Get the first entry from the Request Q */
2791                 removeQ(c);
2792                 h->Qdepth--;
2793
2794                 /* Tell the controller execute command */
2795                 h->access.submit_command(h, c);
2796
2797                 /* Put job onto the completed Q */
2798                 addQ(&h->cmpQ, c);
2799         }
2800 }
2801
2802 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2803 /* Zeros out the error record and then resends the command back */
2804 /* to the controller */
2805 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2806 {
2807         /* erase the old error information */
2808         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2809
2810         /* add it to software queue and then send it to the controller */
2811         addQ(&h->reqQ, c);
2812         h->Qdepth++;
2813         if (h->Qdepth > h->maxQsinceinit)
2814                 h->maxQsinceinit = h->Qdepth;
2815
2816         start_io(h);
2817 }
2818
2819 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2820         unsigned int msg_byte, unsigned int host_byte,
2821         unsigned int driver_byte)
2822 {
2823         /* inverse of macros in scsi.h */
2824         return (scsi_status_byte & 0xff) |
2825                 ((msg_byte & 0xff) << 8) |
2826                 ((host_byte & 0xff) << 16) |
2827                 ((driver_byte & 0xff) << 24);
2828 }
2829
2830 static inline int evaluate_target_status(ctlr_info_t *h,
2831                         CommandList_struct *cmd, int *retry_cmd)
2832 {
2833         unsigned char sense_key;
2834         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2835         int error_value;
2836
2837         *retry_cmd = 0;
2838         /* If we get in here, it means we got "target status", that is, scsi status */
2839         status_byte = cmd->err_info->ScsiStatus;
2840         driver_byte = DRIVER_OK;
2841         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2842
2843         if (blk_pc_request(cmd->rq))
2844                 host_byte = DID_PASSTHROUGH;
2845         else
2846                 host_byte = DID_OK;
2847
2848         error_value = make_status_bytes(status_byte, msg_byte,
2849                 host_byte, driver_byte);
2850
2851         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2852                 if (!blk_pc_request(cmd->rq))
2853                         printk(KERN_WARNING "cciss: cmd %p "
2854                                "has SCSI Status 0x%x\n",
2855                                cmd, cmd->err_info->ScsiStatus);
2856                 return error_value;
2857         }
2858
2859         /* check the sense key */
2860         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2861         /* no status or recovered error */
2862         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2863                 error_value = 0;
2864
2865         if (check_for_unit_attention(h, cmd)) {
2866                 *retry_cmd = !blk_pc_request(cmd->rq);
2867                 return 0;
2868         }
2869
2870         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2871                 if (error_value != 0)
2872                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2873                                " sense key = 0x%x\n", cmd, sense_key);
2874                 return error_value;
2875         }
2876
2877         /* SG_IO or similar, copy sense data back */
2878         if (cmd->rq->sense) {
2879                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2880                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2881                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2882                         cmd->rq->sense_len);
2883         } else
2884                 cmd->rq->sense_len = 0;
2885
2886         return error_value;
2887 }
2888
2889 /* checks the status of the job and calls complete buffers to mark all
2890  * buffers for the completed job. Note that this function does not need
2891  * to hold the hba/queue lock.
2892  */
2893 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2894                                     int timeout)
2895 {
2896         int retry_cmd = 0;
2897         struct request *rq = cmd->rq;
2898
2899         rq->errors = 0;
2900
2901         if (timeout)
2902                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
2903
2904         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
2905                 goto after_error_processing;
2906
2907         switch (cmd->err_info->CommandStatus) {
2908         case CMD_TARGET_STATUS:
2909                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
2910                 break;
2911         case CMD_DATA_UNDERRUN:
2912                 if (blk_fs_request(cmd->rq)) {
2913                         printk(KERN_WARNING "cciss: cmd %p has"
2914                                " completed with data underrun "
2915                                "reported\n", cmd);
2916                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
2917                 }
2918                 break;
2919         case CMD_DATA_OVERRUN:
2920                 if (blk_fs_request(cmd->rq))
2921                         printk(KERN_WARNING "cciss: cmd %p has"
2922                                " completed with data overrun "
2923                                "reported\n", cmd);
2924                 break;
2925         case CMD_INVALID:
2926                 printk(KERN_WARNING "cciss: cmd %p is "
2927                        "reported invalid\n", cmd);
2928                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2929                         cmd->err_info->CommandStatus, DRIVER_OK,
2930                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2931                 break;
2932         case CMD_PROTOCOL_ERR:
2933                 printk(KERN_WARNING "cciss: cmd %p has "
2934                        "protocol error \n", cmd);
2935                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2936                         cmd->err_info->CommandStatus, DRIVER_OK,
2937                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2938                 break;
2939         case CMD_HARDWARE_ERR:
2940                 printk(KERN_WARNING "cciss: cmd %p had "
2941                        " hardware error\n", cmd);
2942                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2943                         cmd->err_info->CommandStatus, DRIVER_OK,
2944                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2945                 break;
2946         case CMD_CONNECTION_LOST:
2947                 printk(KERN_WARNING "cciss: cmd %p had "
2948                        "connection lost\n", cmd);
2949                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2950                         cmd->err_info->CommandStatus, DRIVER_OK,
2951                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2952                 break;
2953         case CMD_ABORTED:
2954                 printk(KERN_WARNING "cciss: cmd %p was "
2955                        "aborted\n", cmd);
2956                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2957                         cmd->err_info->CommandStatus, DRIVER_OK,
2958                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
2959                 break;
2960         case CMD_ABORT_FAILED:
2961                 printk(KERN_WARNING "cciss: cmd %p reports "
2962                        "abort failed\n", cmd);
2963                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2964                         cmd->err_info->CommandStatus, DRIVER_OK,
2965                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2966                 break;
2967         case CMD_UNSOLICITED_ABORT:
2968                 printk(KERN_WARNING "cciss%d: unsolicited "
2969                        "abort %p\n", h->ctlr, cmd);
2970                 if (cmd->retry_count < MAX_CMD_RETRIES) {
2971                         retry_cmd = 1;
2972                         printk(KERN_WARNING
2973                                "cciss%d: retrying %p\n", h->ctlr, cmd);
2974                         cmd->retry_count++;
2975                 } else
2976                         printk(KERN_WARNING
2977                                "cciss%d: %p retried too "
2978                                "many times\n", h->ctlr, cmd);
2979                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2980                         cmd->err_info->CommandStatus, DRIVER_OK,
2981                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
2982                 break;
2983         case CMD_TIMEOUT:
2984                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
2985                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2986                         cmd->err_info->CommandStatus, DRIVER_OK,
2987                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2988                 break;
2989         default:
2990                 printk(KERN_WARNING "cciss: cmd %p returned "
2991                        "unknown status %x\n", cmd,
2992                        cmd->err_info->CommandStatus);
2993                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2994                         cmd->err_info->CommandStatus, DRIVER_OK,
2995                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2996         }
2997
2998 after_error_processing:
2999
3000         /* We need to return this command */
3001         if (retry_cmd) {
3002                 resend_cciss_cmd(h, cmd);
3003                 return;
3004         }
3005         cmd->rq->completion_data = cmd;
3006         blk_complete_request(cmd->rq);
3007 }
3008
3009 /*
3010  * Get a request and submit it to the controller.
3011  */
3012 static void do_cciss_request(struct request_queue *q)
3013 {
3014         ctlr_info_t *h = q->queuedata;
3015         CommandList_struct *c;
3016         sector_t start_blk;
3017         int seg;
3018         struct request *creq;
3019         u64bit temp64;
3020         struct scatterlist tmp_sg[MAXSGENTRIES];
3021         drive_info_struct *drv;
3022         int i, dir;
3023
3024         /* We call start_io here in case there is a command waiting on the
3025          * queue that has not been sent.
3026          */
3027         if (blk_queue_plugged(q))
3028                 goto startio;
3029
3030       queue:
3031         creq = blk_peek_request(q);
3032         if (!creq)
3033                 goto startio;
3034
3035         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
3036
3037         if ((c = cmd_alloc(h, 1)) == NULL)
3038                 goto full;
3039
3040         blk_start_request(creq);
3041
3042         spin_unlock_irq(q->queue_lock);
3043
3044         c->cmd_type = CMD_RWREQ;
3045         c->rq = creq;
3046
3047         /* fill in the request */
3048         drv = creq->rq_disk->private_data;
3049         c->Header.ReplyQueue = 0;       // unused in simple mode
3050         /* got command from pool, so use the command block index instead */
3051         /* for direct lookups. */
3052         /* The first 2 bits are reserved for controller error reporting. */
3053         c->Header.Tag.lower = (c->cmdindex << 3);
3054         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
3055         c->Header.LUN.LogDev.VolId = drv->LunID;
3056         c->Header.LUN.LogDev.Mode = 1;
3057         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
3058         c->Request.Type.Type = TYPE_CMD;        // It is a command.
3059         c->Request.Type.Attribute = ATTR_SIMPLE;
3060         c->Request.Type.Direction =
3061             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3062         c->Request.Timeout = 0; // Don't time out
3063         c->Request.CDB[0] =
3064             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3065         start_blk = blk_rq_pos(creq);
3066 #ifdef CCISS_DEBUG
3067         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3068                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3069 #endif                          /* CCISS_DEBUG */
3070
3071         sg_init_table(tmp_sg, MAXSGENTRIES);
3072         seg = blk_rq_map_sg(q, creq, tmp_sg);
3073
3074         /* get the DMA records for the setup */
3075         if (c->Request.Type.Direction == XFER_READ)
3076                 dir = PCI_DMA_FROMDEVICE;
3077         else
3078                 dir = PCI_DMA_TODEVICE;
3079
3080         for (i = 0; i < seg; i++) {
3081                 c->SG[i].Len = tmp_sg[i].length;
3082                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3083                                                   tmp_sg[i].offset,
3084                                                   tmp_sg[i].length, dir);
3085                 c->SG[i].Addr.lower = temp64.val32.lower;
3086                 c->SG[i].Addr.upper = temp64.val32.upper;
3087                 c->SG[i].Ext = 0;       // we are not chaining
3088         }
3089         /* track how many SG entries we are using */
3090         if (seg > h->maxSG)
3091                 h->maxSG = seg;
3092
3093 #ifdef CCISS_DEBUG
3094         printk(KERN_DEBUG "cciss: Submitting %u sectors in %d segments\n",
3095                blk_rq_sectors(creq), seg);
3096 #endif                          /* CCISS_DEBUG */
3097
3098         c->Header.SGList = c->Header.SGTotal = seg;
3099         if (likely(blk_fs_request(creq))) {
3100                 if(h->cciss_read == CCISS_READ_10) {
3101                         c->Request.CDB[1] = 0;
3102                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
3103                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3104                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3105                         c->Request.CDB[5] = start_blk & 0xff;
3106                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
3107                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3108                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3109                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3110                 } else {
3111                         u32 upper32 = upper_32_bits(start_blk);
3112
3113                         c->Request.CDBLen = 16;
3114                         c->Request.CDB[1]= 0;
3115                         c->Request.CDB[2]= (upper32 >> 24) & 0xff;      //MSB
3116                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3117                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3118                         c->Request.CDB[5]= upper32 & 0xff;
3119                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3120                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3121                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3122                         c->Request.CDB[9]= start_blk & 0xff;
3123                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3124                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3125                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3126                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3127                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3128                 }
3129         } else if (blk_pc_request(creq)) {
3130                 c->Request.CDBLen = creq->cmd_len;
3131                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3132         } else {
3133                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3134                 BUG();
3135         }
3136
3137         spin_lock_irq(q->queue_lock);
3138
3139         addQ(&h->reqQ, c);
3140         h->Qdepth++;
3141         if (h->Qdepth > h->maxQsinceinit)
3142                 h->maxQsinceinit = h->Qdepth;
3143
3144         goto queue;
3145 full:
3146         blk_stop_queue(q);
3147 startio:
3148         /* We will already have the driver lock here so not need
3149          * to lock it.
3150          */
3151         start_io(h);
3152 }
3153
3154 static inline unsigned long get_next_completion(ctlr_info_t *h)
3155 {
3156         return h->access.command_completed(h);
3157 }
3158
3159 static inline int interrupt_pending(ctlr_info_t *h)
3160 {
3161         return h->access.intr_pending(h);
3162 }
3163
3164 static inline long interrupt_not_for_us(ctlr_info_t *h)
3165 {
3166         return (((h->access.intr_pending(h) == 0) ||
3167                  (h->interrupts_enabled == 0)));
3168 }
3169
3170 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3171 {
3172         ctlr_info_t *h = dev_id;
3173         CommandList_struct *c;
3174         unsigned long flags;
3175         __u32 a, a1, a2;
3176
3177         if (interrupt_not_for_us(h))
3178                 return IRQ_NONE;
3179         /*
3180          * If there are completed commands in the completion queue,
3181          * we had better do something about it.
3182          */
3183         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3184         while (interrupt_pending(h)) {
3185                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3186                         a1 = a;
3187                         if ((a & 0x04)) {
3188                                 a2 = (a >> 3);
3189                                 if (a2 >= h->nr_cmds) {
3190                                         printk(KERN_WARNING
3191                                                "cciss: controller cciss%d failed, stopping.\n",
3192                                                h->ctlr);
3193                                         fail_all_cmds(h->ctlr);
3194                                         return IRQ_HANDLED;
3195                                 }
3196
3197                                 c = h->cmd_pool + a2;
3198                                 a = c->busaddr;
3199
3200                         } else {
3201                                 struct hlist_node *tmp;
3202
3203                                 a &= ~3;
3204                                 c = NULL;
3205                                 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3206                                         if (c->busaddr == a)
3207                                                 break;
3208                                 }
3209                         }
3210                         /*
3211                          * If we've found the command, take it off the
3212                          * completion Q and free it
3213                          */
3214                         if (c && c->busaddr == a) {
3215                                 removeQ(c);
3216                                 if (c->cmd_type == CMD_RWREQ) {
3217                                         complete_command(h, c, 0);
3218                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3219                                         complete(c->waiting);
3220                                 }
3221 #                               ifdef CONFIG_CISS_SCSI_TAPE
3222                                 else if (c->cmd_type == CMD_SCSI)
3223                                         complete_scsi_command(c, 0, a1);
3224 #                               endif
3225                                 continue;
3226                         }
3227                 }
3228         }
3229
3230         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3231         return IRQ_HANDLED;
3232 }
3233
3234 static int scan_thread(void *data)
3235 {
3236         ctlr_info_t *h = data;
3237         int rc;
3238         DECLARE_COMPLETION_ONSTACK(wait);
3239         h->rescan_wait = &wait;
3240
3241         for (;;) {
3242                 rc = wait_for_completion_interruptible(&wait);
3243                 if (kthread_should_stop())
3244                         break;
3245                 if (!rc)
3246                         rebuild_lun_table(h, 0);
3247         }
3248         return 0;
3249 }
3250
3251 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3252 {
3253         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3254                 return 0;
3255
3256         switch (c->err_info->SenseInfo[12]) {
3257         case STATE_CHANGED:
3258                 printk(KERN_WARNING "cciss%d: a state change "
3259                         "detected, command retried\n", h->ctlr);
3260                 return 1;
3261         break;
3262         case LUN_FAILED:
3263                 printk(KERN_WARNING "cciss%d: LUN failure "
3264                         "detected, action required\n", h->ctlr);
3265                 return 1;
3266         break;
3267         case REPORT_LUNS_CHANGED:
3268                 printk(KERN_WARNING "cciss%d: report LUN data "
3269                         "changed\n", h->ctlr);
3270                 if (h->rescan_wait)
3271                         complete(h->rescan_wait);
3272                 return 1;
3273         break;
3274         case POWER_OR_RESET:
3275                 printk(KERN_WARNING "cciss%d: a power on "
3276                         "or device reset detected\n", h->ctlr);
3277                 return 1;
3278         break;
3279         case UNIT_ATTENTION_CLEARED:
3280                 printk(KERN_WARNING "cciss%d: unit attention "
3281                     "cleared by another initiator\n", h->ctlr);
3282                 return 1;
3283         break;
3284         default:
3285                 printk(KERN_WARNING "cciss%d: unknown "
3286                         "unit attention detected\n", h->ctlr);
3287                                 return 1;
3288         }
3289 }
3290
3291 /*
3292  *  We cannot read the structure directly, for portability we must use
3293  *   the io functions.
3294  *   This is for debug only.
3295  */
3296 #ifdef CCISS_DEBUG
3297 static void print_cfg_table(CfgTable_struct *tb)
3298 {
3299         int i;
3300         char temp_name[17];
3301
3302         printk("Controller Configuration information\n");
3303         printk("------------------------------------\n");
3304         for (i = 0; i < 4; i++)
3305                 temp_name[i] = readb(&(tb->Signature[i]));
3306         temp_name[4] = '\0';
3307         printk("   Signature = %s\n", temp_name);
3308         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3309         printk("   Transport methods supported = 0x%x\n",
3310                readl(&(tb->TransportSupport)));
3311         printk("   Transport methods active = 0x%x\n",
3312                readl(&(tb->TransportActive)));
3313         printk("   Requested transport Method = 0x%x\n",
3314                readl(&(tb->HostWrite.TransportRequest)));
3315         printk("   Coalesce Interrupt Delay = 0x%x\n",
3316                readl(&(tb->HostWrite.CoalIntDelay)));
3317         printk("   Coalesce Interrupt Count = 0x%x\n",
3318                readl(&(tb->HostWrite.CoalIntCount)));
3319         printk("   Max outstanding commands = 0x%d\n",
3320                readl(&(tb->CmdsOutMax)));
3321         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3322         for (i = 0; i < 16; i++)
3323                 temp_name[i] = readb(&(tb->ServerName[i]));
3324         temp_name[16] = '\0';
3325         printk("   Server Name = %s\n", temp_name);
3326         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3327 }
3328 #endif                          /* CCISS_DEBUG */
3329
3330 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3331 {
3332         int i, offset, mem_type, bar_type;
3333         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3334                 return 0;
3335         offset = 0;
3336         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3337                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3338                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3339                         offset += 4;
3340                 else {
3341                         mem_type = pci_resource_flags(pdev, i) &
3342                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3343                         switch (mem_type) {
3344                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3345                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3346                                 offset += 4;    /* 32 bit */
3347                                 break;
3348                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3349                                 offset += 8;
3350                                 break;
3351                         default:        /* reserved in PCI 2.2 */
3352                                 printk(KERN_WARNING
3353                                        "Base address is invalid\n");
3354                                 return -1;
3355                                 break;
3356                         }
3357                 }
3358                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3359                         return i + 1;
3360         }
3361         return -1;
3362 }
3363
3364 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3365  * controllers that are capable. If not, we use IO-APIC mode.
3366  */
3367
3368 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3369                                            struct pci_dev *pdev, __u32 board_id)
3370 {
3371 #ifdef CONFIG_PCI_MSI
3372         int err;
3373         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3374         {0, 2}, {0, 3}
3375         };
3376
3377         /* Some boards advertise MSI but don't really support it */
3378         if ((board_id == 0x40700E11) ||
3379             (board_id == 0x40800E11) ||
3380             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3381                 goto default_int_mode;
3382
3383         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3384                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3385                 if (!err) {
3386                         c->intr[0] = cciss_msix_entries[0].vector;
3387                         c->intr[1] = cciss_msix_entries[1].vector;
3388                         c->intr[2] = cciss_msix_entries[2].vector;
3389                         c->intr[3] = cciss_msix_entries[3].vector;
3390                         c->msix_vector = 1;
3391                         return;
3392                 }
3393                 if (err > 0) {
3394                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3395                                "available\n", err);
3396                         goto default_int_mode;
3397                 } else {
3398                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3399                                err);
3400                         goto default_int_mode;
3401                 }
3402         }
3403         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3404                 if (!pci_enable_msi(pdev)) {
3405                         c->msi_vector = 1;
3406                 } else {
3407                         printk(KERN_WARNING "cciss: MSI init failed\n");
3408                 }
3409         }
3410 default_int_mode:
3411 #endif                          /* CONFIG_PCI_MSI */
3412         /* if we get here we're going to use the default interrupt mode */
3413         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3414         return;
3415 }
3416
3417 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3418 {
3419         ushort subsystem_vendor_id, subsystem_device_id, command;
3420         __u32 board_id, scratchpad = 0;
3421         __u64 cfg_offset;
3422         __u32 cfg_base_addr;
3423         __u64 cfg_base_addr_index;
3424         int i, err;
3425
3426         /* check to see if controller has been disabled */
3427         /* BEFORE trying to enable it */
3428         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3429         if (!(command & 0x02)) {
3430                 printk(KERN_WARNING
3431                        "cciss: controller appears to be disabled\n");
3432                 return -ENODEV;
3433         }
3434
3435         err = pci_enable_device(pdev);
3436         if (err) {
3437                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3438                 return err;
3439         }
3440
3441         err = pci_request_regions(pdev, "cciss");
3442         if (err) {
3443                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3444                        "aborting\n");
3445                 return err;
3446         }
3447
3448         subsystem_vendor_id = pdev->subsystem_vendor;
3449         subsystem_device_id = pdev->subsystem_device;
3450         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3451                     subsystem_vendor_id);
3452
3453 #ifdef CCISS_DEBUG
3454         printk("command = %x\n", command);
3455         printk("irq = %x\n", pdev->irq);
3456         printk("board_id = %x\n", board_id);
3457 #endif                          /* CCISS_DEBUG */
3458
3459 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3460  * else we use the IO-APIC interrupt assigned to us by system ROM.
3461  */
3462         cciss_interrupt_mode(c, pdev, board_id);
3463
3464         /* find the memory BAR */
3465         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3466                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3467                         break;
3468         }
3469         if (i == DEVICE_COUNT_RESOURCE) {
3470                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3471                 err = -ENODEV;
3472                 goto err_out_free_res;
3473         }
3474
3475         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3476                                                  * already removed
3477                                                  */
3478
3479 #ifdef CCISS_DEBUG
3480         printk("address 0 = %lx\n", c->paddr);
3481 #endif                          /* CCISS_DEBUG */
3482         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3483
3484         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3485          * We poll for up to 120 secs, once per 100ms. */
3486         for (i = 0; i < 1200; i++) {
3487                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3488                 if (scratchpad == CCISS_FIRMWARE_READY)
3489                         break;
3490                 set_current_state(TASK_INTERRUPTIBLE);
3491                 schedule_timeout(HZ / 10);      /* wait 100ms */
3492         }
3493         if (scratchpad != CCISS_FIRMWARE_READY) {
3494                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3495                 err = -ENODEV;
3496                 goto err_out_free_res;
3497         }
3498
3499         /* get the address index number */
3500         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3501         cfg_base_addr &= (__u32) 0x0000ffff;
3502 #ifdef CCISS_DEBUG
3503         printk("cfg base address = %x\n", cfg_base_addr);
3504 #endif                          /* CCISS_DEBUG */
3505         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3506 #ifdef CCISS_DEBUG
3507         printk("cfg base address index = %llx\n",
3508                 (unsigned long long)cfg_base_addr_index);
3509 #endif                          /* CCISS_DEBUG */
3510         if (cfg_base_addr_index == -1) {
3511                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3512                 err = -ENODEV;
3513                 goto err_out_free_res;
3514         }
3515
3516         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3517 #ifdef CCISS_DEBUG
3518         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3519 #endif                          /* CCISS_DEBUG */
3520         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3521                                                        cfg_base_addr_index) +
3522                                     cfg_offset, sizeof(CfgTable_struct));
3523         c->board_id = board_id;
3524
3525 #ifdef CCISS_DEBUG
3526         print_cfg_table(c->cfgtable);
3527 #endif                          /* CCISS_DEBUG */
3528
3529         /* Some controllers support Zero Memory Raid (ZMR).
3530          * When configured in ZMR mode the number of supported
3531          * commands drops to 64. So instead of just setting an
3532          * arbitrary value we make the driver a little smarter.
3533          * We read the config table to tell us how many commands
3534          * are supported on the controller then subtract 4 to
3535          * leave a little room for ioctl calls.
3536          */
3537         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3538         for (i = 0; i < ARRAY_SIZE(products); i++) {
3539                 if (board_id == products[i].board_id) {
3540                         c->product_name = products[i].product_name;
3541                         c->access = *(products[i].access);
3542                         c->nr_cmds = c->max_commands - 4;
3543                         break;
3544                 }
3545         }
3546         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3547             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3548             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3549             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3550                 printk("Does not appear to be a valid CISS config table\n");
3551                 err = -ENODEV;
3552                 goto err_out_free_res;
3553         }
3554         /* We didn't find the controller in our list. We know the
3555          * signature is valid. If it's an HP device let's try to
3556          * bind to the device and fire it up. Otherwise we bail.
3557          */
3558         if (i == ARRAY_SIZE(products)) {
3559                 if (subsystem_vendor_id == PCI_VENDOR_ID_HP) {
3560                         c->product_name = products[i-1].product_name;
3561                         c->access = *(products[i-1].access);
3562                         c->nr_cmds = c->max_commands - 4;
3563                         printk(KERN_WARNING "cciss: This is an unknown "
3564                                 "Smart Array controller.\n"
3565                                 "cciss: Please update to the latest driver "
3566                                 "available from www.hp.com.\n");
3567                 } else {
3568                         printk(KERN_WARNING "cciss: Sorry, I don't know how"
3569                                 " to access the Smart Array controller %08lx\n"
3570                                         , (unsigned long)board_id);
3571                         err = -ENODEV;
3572                         goto err_out_free_res;
3573                 }
3574         }
3575 #ifdef CONFIG_X86
3576         {
3577                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3578                 __u32 prefetch;
3579                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3580                 prefetch |= 0x100;
3581                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3582         }
3583 #endif
3584
3585         /* Disabling DMA prefetch and refetch for the P600.
3586          * An ASIC bug may result in accesses to invalid memory addresses.
3587          * We've disabled prefetch for some time now. Testing with XEN
3588          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3589          */
3590         if(board_id == 0x3225103C) {
3591                 __u32 dma_prefetch;
3592                 __u32 dma_refetch;
3593                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3594                 dma_prefetch |= 0x8000;
3595                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3596                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3597                 dma_refetch |= 0x1;
3598                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3599         }
3600
3601 #ifdef CCISS_DEBUG
3602         printk("Trying to put board into Simple mode\n");
3603 #endif                          /* CCISS_DEBUG */
3604         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3605         /* Update the field, and then ring the doorbell */
3606         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3607         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3608
3609         /* under certain very rare conditions, this can take awhile.
3610          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3611          * as we enter this code.) */
3612         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3613                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3614                         break;
3615                 /* delay and try again */
3616                 set_current_state(TASK_INTERRUPTIBLE);
3617                 schedule_timeout(10);
3618         }
3619
3620 #ifdef CCISS_DEBUG
3621         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3622                readl(c->vaddr + SA5_DOORBELL));
3623 #endif                          /* CCISS_DEBUG */
3624 #ifdef CCISS_DEBUG
3625         print_cfg_table(c->cfgtable);
3626 #endif                          /* CCISS_DEBUG */
3627
3628         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3629                 printk(KERN_WARNING "cciss: unable to get board into"
3630                        " simple mode\n");
3631                 err = -ENODEV;
3632                 goto err_out_free_res;
3633         }
3634         return 0;
3635
3636 err_out_free_res:
3637         /*
3638          * Deliberately omit pci_disable_device(): it does something nasty to
3639          * Smart Array controllers that pci_enable_device does not undo
3640          */
3641         pci_release_regions(pdev);
3642         return err;
3643 }
3644
3645 /* Function to find the first free pointer into our hba[] array
3646  * Returns -1 if no free entries are left.
3647  */
3648 static int alloc_cciss_hba(void)
3649 {
3650         int i;
3651
3652         for (i = 0; i < MAX_CTLR; i++) {
3653                 if (!hba[i]) {
3654                         ctlr_info_t *p;
3655
3656                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3657                         if (!p)
3658                                 goto Enomem;
3659                         hba[i] = p;
3660                         return i;
3661                 }
3662         }
3663         printk(KERN_WARNING "cciss: This driver supports a maximum"
3664                " of %d controllers.\n", MAX_CTLR);
3665         return -1;
3666 Enomem:
3667         printk(KERN_ERR "cciss: out of memory.\n");
3668         return -1;
3669 }
3670
3671 static void free_hba(int i)
3672 {
3673         ctlr_info_t *p = hba[i];
3674         int n;
3675
3676         hba[i] = NULL;
3677         for (n = 0; n < CISS_MAX_LUN; n++)
3678                 put_disk(p->gendisk[n]);
3679         kfree(p);
3680 }
3681
3682 /* Send a message CDB to the firmware. */
3683 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3684 {
3685         typedef struct {
3686                 CommandListHeader_struct CommandHeader;
3687                 RequestBlock_struct Request;
3688                 ErrDescriptor_struct ErrorDescriptor;
3689         } Command;
3690         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3691         Command *cmd;
3692         dma_addr_t paddr64;
3693         uint32_t paddr32, tag;
3694         void __iomem *vaddr;
3695         int i, err;
3696
3697         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3698         if (vaddr == NULL)
3699                 return -ENOMEM;
3700
3701         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3702            CCISS commands, so they must be allocated from the lower 4GiB of
3703            memory. */
3704         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3705         if (err) {
3706                 iounmap(vaddr);
3707                 return -ENOMEM;
3708         }
3709
3710         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3711         if (cmd == NULL) {
3712                 iounmap(vaddr);
3713                 return -ENOMEM;
3714         }
3715
3716         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
3717            although there's no guarantee, we assume that the address is at
3718            least 4-byte aligned (most likely, it's page-aligned). */
3719         paddr32 = paddr64;
3720
3721         cmd->CommandHeader.ReplyQueue = 0;
3722         cmd->CommandHeader.SGList = 0;
3723         cmd->CommandHeader.SGTotal = 0;
3724         cmd->CommandHeader.Tag.lower = paddr32;
3725         cmd->CommandHeader.Tag.upper = 0;
3726         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
3727
3728         cmd->Request.CDBLen = 16;
3729         cmd->Request.Type.Type = TYPE_MSG;
3730         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
3731         cmd->Request.Type.Direction = XFER_NONE;
3732         cmd->Request.Timeout = 0; /* Don't time out */
3733         cmd->Request.CDB[0] = opcode;
3734         cmd->Request.CDB[1] = type;
3735         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
3736
3737         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
3738         cmd->ErrorDescriptor.Addr.upper = 0;
3739         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
3740
3741         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
3742
3743         for (i = 0; i < 10; i++) {
3744                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
3745                 if ((tag & ~3) == paddr32)
3746                         break;
3747                 schedule_timeout_uninterruptible(HZ);
3748         }
3749
3750         iounmap(vaddr);
3751
3752         /* we leak the DMA buffer here ... no choice since the controller could
3753            still complete the command. */
3754         if (i == 10) {
3755                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
3756                         opcode, type);
3757                 return -ETIMEDOUT;
3758         }
3759
3760         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
3761
3762         if (tag & 2) {
3763                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
3764                         opcode, type);
3765                 return -EIO;
3766         }
3767
3768         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
3769                 opcode, type);
3770         return 0;
3771 }
3772
3773 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
3774 #define cciss_noop(p) cciss_message(p, 3, 0)
3775
3776 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
3777 {
3778 /* the #defines are stolen from drivers/pci/msi.h. */
3779 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
3780 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
3781
3782         int pos;
3783         u16 control = 0;
3784
3785         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
3786         if (pos) {
3787                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
3788                 if (control & PCI_MSI_FLAGS_ENABLE) {
3789                         printk(KERN_INFO "cciss: resetting MSI\n");
3790                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
3791                 }
3792         }
3793
3794         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
3795         if (pos) {
3796                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
3797                 if (control & PCI_MSIX_FLAGS_ENABLE) {
3798                         printk(KERN_INFO "cciss: resetting MSI-X\n");
3799                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
3800                 }
3801         }
3802
3803         return 0;
3804 }
3805
3806 /* This does a hard reset of the controller using PCI power management
3807  * states. */
3808 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
3809 {
3810         u16 pmcsr, saved_config_space[32];
3811         int i, pos;
3812
3813         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
3814
3815         /* This is very nearly the same thing as
3816
3817            pci_save_state(pci_dev);
3818            pci_set_power_state(pci_dev, PCI_D3hot);
3819            pci_set_power_state(pci_dev, PCI_D0);
3820            pci_restore_state(pci_dev);
3821
3822            but we can't use these nice canned kernel routines on
3823            kexec, because they also check the MSI/MSI-X state in PCI
3824            configuration space and do the wrong thing when it is
3825            set/cleared.  Also, the pci_save/restore_state functions
3826            violate the ordering requirements for restoring the
3827            configuration space from the CCISS document (see the
3828            comment below).  So we roll our own .... */
3829
3830         for (i = 0; i < 32; i++)
3831                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
3832
3833         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
3834         if (pos == 0) {
3835                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
3836                 return -ENODEV;
3837         }
3838
3839         /* Quoting from the Open CISS Specification: "The Power
3840          * Management Control/Status Register (CSR) controls the power
3841          * state of the device.  The normal operating state is D0,
3842          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
3843          * the controller, place the interface device in D3 then to
3844          * D0, this causes a secondary PCI reset which will reset the
3845          * controller." */
3846
3847         /* enter the D3hot power management state */
3848         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
3849         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3850         pmcsr |= PCI_D3hot;
3851         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
3852
3853         schedule_timeout_uninterruptible(HZ >> 1);
3854
3855         /* enter the D0 power management state */
3856         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3857         pmcsr |= PCI_D0;
3858         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
3859
3860         schedule_timeout_uninterruptible(HZ >> 1);
3861
3862         /* Restore the PCI configuration space.  The Open CISS
3863          * Specification says, "Restore the PCI Configuration
3864          * Registers, offsets 00h through 60h. It is important to
3865          * restore the command register, 16-bits at offset 04h,
3866          * last. Do not restore the configuration status register,
3867          * 16-bits at offset 06h."  Note that the offset is 2*i. */
3868         for (i = 0; i < 32; i++) {
3869                 if (i == 2 || i == 3)
3870                         continue;
3871                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
3872         }
3873         wmb();
3874         pci_write_config_word(pdev, 4, saved_config_space[2]);
3875
3876         return 0;
3877 }
3878
3879 /*
3880  *  This is it.  Find all the controllers and register them.  I really hate
3881  *  stealing all these major device numbers.
3882  *  returns the number of block devices registered.
3883  */
3884 static int __devinit cciss_init_one(struct pci_dev *pdev,
3885                                     const struct pci_device_id *ent)
3886 {
3887         int i;
3888         int j = 0;
3889         int rc;
3890         int dac, return_code;
3891         InquiryData_struct *inq_buff = NULL;
3892
3893         if (reset_devices) {
3894                 /* Reset the controller with a PCI power-cycle */
3895                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
3896                         return -ENODEV;
3897
3898                 /* Now try to get the controller to respond to a no-op. Some
3899                    devices (notably the HP Smart Array 5i Controller) need
3900                    up to 30 seconds to respond. */
3901                 for (i=0; i<30; i++) {
3902                         if (cciss_noop(pdev) == 0)
3903                                 break;
3904
3905                         schedule_timeout_uninterruptible(HZ);
3906                 }
3907                 if (i == 30) {
3908                         printk(KERN_ERR "cciss: controller seems dead\n");
3909                         return -EBUSY;
3910                 }
3911         }
3912
3913         i = alloc_cciss_hba();
3914         if (i < 0)
3915                 return -1;
3916
3917         hba[i]->busy_initializing = 1;
3918         INIT_HLIST_HEAD(&hba[i]->cmpQ);
3919         INIT_HLIST_HEAD(&hba[i]->reqQ);
3920
3921         if (cciss_pci_init(hba[i], pdev) != 0)
3922                 goto clean0;
3923
3924         sprintf(hba[i]->devname, "cciss%d", i);
3925         hba[i]->ctlr = i;
3926         hba[i]->pdev = pdev;
3927
3928         if (cciss_create_hba_sysfs_entry(hba[i]))
3929                 goto clean0;
3930
3931         /* configure PCI DMA stuff */
3932         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
3933                 dac = 1;
3934         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
3935                 dac = 0;
3936         else {
3937                 printk(KERN_ERR "cciss: no suitable DMA available\n");
3938                 goto clean1;
3939         }
3940
3941         /*
3942          * register with the major number, or get a dynamic major number
3943          * by passing 0 as argument.  This is done for greater than
3944          * 8 controller support.
3945          */
3946         if (i < MAX_CTLR_ORIG)
3947                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
3948         rc = register_blkdev(hba[i]->major, hba[i]->devname);
3949         if (rc == -EBUSY || rc == -EINVAL) {
3950                 printk(KERN_ERR
3951                        "cciss:  Unable to get major number %d for %s "
3952                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
3953                 goto clean1;
3954         } else {
3955                 if (i >= MAX_CTLR_ORIG)
3956                         hba[i]->major = rc;
3957         }
3958
3959         /* make sure the board interrupts are off */
3960         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
3961         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
3962                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
3963                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
3964                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
3965                 goto clean2;
3966         }
3967
3968         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
3969                hba[i]->devname, pdev->device, pci_name(pdev),
3970                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
3971
3972         hba[i]->cmd_pool_bits =
3973             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
3974                         * sizeof(unsigned long), GFP_KERNEL);
3975         hba[i]->cmd_pool = (CommandList_struct *)
3976             pci_alloc_consistent(hba[i]->pdev,
3977                     hba[i]->nr_cmds * sizeof(CommandList_struct),
3978                     &(hba[i]->cmd_pool_dhandle));
3979         hba[i]->errinfo_pool = (ErrorInfo_struct *)
3980             pci_alloc_consistent(hba[i]->pdev,
3981                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3982                     &(hba[i]->errinfo_pool_dhandle));
3983         if ((hba[i]->cmd_pool_bits == NULL)
3984             || (hba[i]->cmd_pool == NULL)
3985             || (hba[i]->errinfo_pool == NULL)) {
3986                 printk(KERN_ERR "cciss: out of memory");
3987                 goto clean4;
3988         }
3989         spin_lock_init(&hba[i]->lock);
3990
3991         /* Initialize the pdev driver private data.
3992            have it point to hba[i].  */
3993         pci_set_drvdata(pdev, hba[i]);
3994         /* command and error info recs zeroed out before
3995            they are used */
3996         memset(hba[i]->cmd_pool_bits, 0,
3997                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
3998                         * sizeof(unsigned long));
3999
4000         hba[i]->num_luns = 0;
4001         hba[i]->highest_lun = -1;
4002         for (j = 0; j < CISS_MAX_LUN; j++) {
4003                 hba[i]->drv[j].raid_level = -1;
4004                 hba[i]->drv[j].queue = NULL;
4005                 hba[i]->gendisk[j] = NULL;
4006         }
4007
4008         cciss_scsi_setup(i);
4009
4010         /* Turn the interrupts on so we can service requests */
4011         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4012
4013         /* Get the firmware version */
4014         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4015         if (inq_buff == NULL) {
4016                 printk(KERN_ERR "cciss: out of memory\n");
4017                 goto clean4;
4018         }
4019
4020         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4021                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4022         if (return_code == IO_OK) {
4023                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4024                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4025                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4026                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4027         } else {         /* send command failed */
4028                 printk(KERN_WARNING "cciss: unable to determine firmware"
4029                         " version of controller\n");
4030         }
4031
4032         cciss_procinit(i);
4033
4034         hba[i]->cciss_max_sectors = 2048;
4035
4036         hba[i]->busy_initializing = 0;
4037
4038         rebuild_lun_table(hba[i], 1);
4039         hba[i]->cciss_scan_thread = kthread_run(scan_thread, hba[i],
4040                                 "cciss_scan%02d", i);
4041         if (IS_ERR(hba[i]->cciss_scan_thread))
4042                 return PTR_ERR(hba[i]->cciss_scan_thread);
4043
4044         return 1;
4045
4046 clean4:
4047         kfree(inq_buff);
4048         kfree(hba[i]->cmd_pool_bits);
4049         if (hba[i]->cmd_pool)
4050                 pci_free_consistent(hba[i]->pdev,
4051                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4052                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4053         if (hba[i]->errinfo_pool)
4054                 pci_free_consistent(hba[i]->pdev,
4055                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4056                                     hba[i]->errinfo_pool,
4057                                     hba[i]->errinfo_pool_dhandle);
4058         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4059 clean2:
4060         unregister_blkdev(hba[i]->major, hba[i]->devname);
4061 clean1:
4062         cciss_destroy_hba_sysfs_entry(hba[i]);
4063 clean0:
4064         hba[i]->busy_initializing = 0;
4065         /* cleanup any queues that may have been initialized */
4066         for (j=0; j <= hba[i]->highest_lun; j++){
4067                 drive_info_struct *drv = &(hba[i]->drv[j]);
4068                 if (drv->queue)
4069                         blk_cleanup_queue(drv->queue);
4070         }
4071         /*
4072          * Deliberately omit pci_disable_device(): it does something nasty to
4073          * Smart Array controllers that pci_enable_device does not undo
4074          */
4075         pci_release_regions(pdev);
4076         pci_set_drvdata(pdev, NULL);
4077         free_hba(i);
4078         return -1;
4079 }
4080
4081 static void cciss_shutdown(struct pci_dev *pdev)
4082 {
4083         ctlr_info_t *tmp_ptr;
4084         int i;
4085         char flush_buf[4];
4086         int return_code;
4087
4088         tmp_ptr = pci_get_drvdata(pdev);
4089         if (tmp_ptr == NULL)
4090                 return;
4091         i = tmp_ptr->ctlr;
4092         if (hba[i] == NULL)
4093                 return;
4094
4095         /* Turn board interrupts off  and send the flush cache command */
4096         /* sendcmd will turn off interrupt, and send the flush...
4097          * To write all data in the battery backed cache to disks */
4098         memset(flush_buf, 0, 4);
4099         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0,
4100                 CTLR_LUNID, TYPE_CMD);
4101         if (return_code == IO_OK) {
4102                 printk(KERN_INFO "Completed flushing cache on controller %d\n", i);
4103         } else {
4104                 printk(KERN_WARNING "Error flushing cache on controller %d\n", i);
4105         }
4106         free_irq(hba[i]->intr[2], hba[i]);
4107 }
4108
4109 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4110 {
4111         ctlr_info_t *tmp_ptr;
4112         int i, j;
4113
4114         if (pci_get_drvdata(pdev) == NULL) {
4115                 printk(KERN_ERR "cciss: Unable to remove device \n");
4116                 return;
4117         }
4118
4119         tmp_ptr = pci_get_drvdata(pdev);
4120         i = tmp_ptr->ctlr;
4121         if (hba[i] == NULL) {
4122                 printk(KERN_ERR "cciss: device appears to "
4123                        "already be removed \n");
4124                 return;
4125         }
4126
4127         kthread_stop(hba[i]->cciss_scan_thread);
4128
4129         remove_proc_entry(hba[i]->devname, proc_cciss);
4130         unregister_blkdev(hba[i]->major, hba[i]->devname);
4131
4132         /* remove it from the disk list */
4133         for (j = 0; j < CISS_MAX_LUN; j++) {
4134                 struct gendisk *disk = hba[i]->gendisk[j];
4135                 if (disk) {
4136                         struct request_queue *q = disk->queue;
4137
4138                         if (disk->flags & GENHD_FL_UP)
4139                                 del_gendisk(disk);
4140                         if (q)
4141                                 blk_cleanup_queue(q);
4142                 }
4143         }
4144
4145 #ifdef CONFIG_CISS_SCSI_TAPE
4146         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4147 #endif
4148
4149         cciss_shutdown(pdev);
4150
4151 #ifdef CONFIG_PCI_MSI
4152         if (hba[i]->msix_vector)
4153                 pci_disable_msix(hba[i]->pdev);
4154         else if (hba[i]->msi_vector)
4155                 pci_disable_msi(hba[i]->pdev);
4156 #endif                          /* CONFIG_PCI_MSI */
4157
4158         iounmap(hba[i]->vaddr);
4159
4160         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4161                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4162         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4163                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4164         kfree(hba[i]->cmd_pool_bits);
4165         /*
4166          * Deliberately omit pci_disable_device(): it does something nasty to
4167          * Smart Array controllers that pci_enable_device does not undo
4168          */
4169         pci_release_regions(pdev);
4170         pci_set_drvdata(pdev, NULL);
4171         cciss_destroy_hba_sysfs_entry(hba[i]);
4172         free_hba(i);
4173 }
4174
4175 static struct pci_driver cciss_pci_driver = {
4176         .name = "cciss",
4177         .probe = cciss_init_one,
4178         .remove = __devexit_p(cciss_remove_one),
4179         .id_table = cciss_pci_device_id,        /* id_table */
4180         .shutdown = cciss_shutdown,
4181 };
4182
4183 /*
4184  *  This is it.  Register the PCI driver information for the cards we control
4185  *  the OS will call our registered routines when it finds one of our cards.
4186  */
4187 static int __init cciss_init(void)
4188 {
4189         int err;
4190
4191         /*
4192          * The hardware requires that commands are aligned on a 64-bit
4193          * boundary. Given that we use pci_alloc_consistent() to allocate an
4194          * array of them, the size must be a multiple of 8 bytes.
4195          */
4196         BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4197
4198         printk(KERN_INFO DRIVER_NAME "\n");
4199
4200         err = bus_register(&cciss_bus_type);
4201         if (err)
4202                 return err;
4203
4204         /* Register for our PCI devices */
4205         err = pci_register_driver(&cciss_pci_driver);
4206         if (err)
4207                 goto err_bus_register;
4208
4209         return 0;
4210
4211 err_bus_register:
4212         bus_unregister(&cciss_bus_type);
4213         return err;
4214 }
4215
4216 static void __exit cciss_cleanup(void)
4217 {
4218         int i;
4219
4220         pci_unregister_driver(&cciss_pci_driver);
4221         /* double check that all controller entrys have been removed */
4222         for (i = 0; i < MAX_CTLR; i++) {
4223                 if (hba[i] != NULL) {
4224                         printk(KERN_WARNING "cciss: had to remove"
4225                                " controller %d\n", i);
4226                         cciss_remove_one(hba[i]->pdev);
4227                 }
4228         }
4229         remove_proc_entry("driver/cciss", NULL);
4230         bus_unregister(&cciss_bus_type);
4231 }
4232
4233 static void fail_all_cmds(unsigned long ctlr)
4234 {
4235         /* If we get here, the board is apparently dead. */
4236         ctlr_info_t *h = hba[ctlr];
4237         CommandList_struct *c;
4238         unsigned long flags;
4239
4240         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4241         h->alive = 0;           /* the controller apparently died... */
4242
4243         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4244
4245         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
4246
4247         /* move everything off the request queue onto the completed queue */
4248         while (!hlist_empty(&h->reqQ)) {
4249                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4250                 removeQ(c);
4251                 h->Qdepth--;
4252                 addQ(&h->cmpQ, c);
4253         }
4254
4255         /* Now, fail everything on the completed queue with a HW error */
4256         while (!hlist_empty(&h->cmpQ)) {
4257                 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4258                 removeQ(c);
4259                 if (c->cmd_type != CMD_MSG_STALE)
4260                         c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4261                 if (c->cmd_type == CMD_RWREQ) {
4262                         complete_command(h, c, 0);
4263                 } else if (c->cmd_type == CMD_IOCTL_PEND)
4264                         complete(c->waiting);
4265 #ifdef CONFIG_CISS_SCSI_TAPE
4266                 else if (c->cmd_type == CMD_SCSI)
4267                         complete_scsi_command(c, 0, 0);
4268 #endif
4269         }
4270         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4271         return;
4272 }
4273
4274 module_init(cciss_init);
4275 module_exit(cciss_cleanup);