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 <linux/blktrace_api.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/io.h>
  44
  45 #include <linux/dma-mapping.h>
  46 #include <linux/blkdev.h>
  47 #include <linux/genhd.h>
  48 #include <linux/completion.h>
  49 #include <scsi/scsi.h>
  50 #include <scsi/sg.h>
  51 #include <scsi/scsi_ioctl.h>
  52 #include <linux/cdrom.h>
  53 #include <linux/scatterlist.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_ANY_ID,      PCI_ANY_ID, PCI_ANY_ID,
 100                 PCI_CLASS_STORAGE_RAID << 8, 0xffff << 8, 0},
 101         {0,}
 102 };
 103
 104 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 105
 106 /*  board_id = Subsystem Device ID & Vendor ID
 107  *  product = Marketing Name for the board
 108  *  access = Address of the struct of function pointers
 109  */
 110 static struct board_type products[] = {
 111         {0x40700E11, "Smart Array 5300", &SA5_access},
 112         {0x40800E11, "Smart Array 5i", &SA5B_access},
 113         {0x40820E11, "Smart Array 532", &SA5B_access},
 114         {0x40830E11, "Smart Array 5312", &SA5B_access},
 115         {0x409A0E11, "Smart Array 641", &SA5_access},
 116         {0x409B0E11, "Smart Array 642", &SA5_access},
 117         {0x409C0E11, "Smart Array 6400", &SA5_access},
 118         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 119         {0x40910E11, "Smart Array 6i", &SA5_access},
 120         {0x3225103C, "Smart Array P600", &SA5_access},
 121         {0x3223103C, "Smart Array P800", &SA5_access},
 122         {0x3234103C, "Smart Array P400", &SA5_access},
 123         {0x3235103C, "Smart Array P400i", &SA5_access},
 124         {0x3211103C, "Smart Array E200i", &SA5_access},
 125         {0x3212103C, "Smart Array E200", &SA5_access},
 126         {0x3213103C, "Smart Array E200i", &SA5_access},
 127         {0x3214103C, "Smart Array E200i", &SA5_access},
 128         {0x3215103C, "Smart Array E200i", &SA5_access},
 129         {0x3237103C, "Smart Array E500", &SA5_access},
 130         {0x323D103C, "Smart Array P700m", &SA5_access},
 131         {0x3241103C, "Smart Array P212", &SA5_access},
 132         {0x3243103C, "Smart Array P410", &SA5_access},
 133         {0x3245103C, "Smart Array P410i", &SA5_access},
 134         {0x3247103C, "Smart Array P411", &SA5_access},
 135         {0x3249103C, "Smart Array P812", &SA5_access},
 136         {0xFFFF103C, "Unknown Smart Array", &SA5_access},
 137 };
 138
 139 /* How long to wait (in milliseconds) for board to go into simple mode */
 140 #define MAX_CONFIG_WAIT 30000
 141 #define MAX_IOCTL_CONFIG_WAIT 1000
 142
 143 /*define how many times we will try a command because of bus resets */
 144 #define MAX_CMD_RETRIES 3
 145
 146 #define MAX_CTLR        32
 147
 148 /* Originally cciss driver only supports 8 major numbers */
 149 #define MAX_CTLR_ORIG   8
 150
 151 static ctlr_info_t *hba[MAX_CTLR];
 152
 153 static void do_cciss_request(struct request_queue *q);
 154 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
 155 static int cciss_open(struct inode *inode, struct file *filep);
 156 static int cciss_release(struct inode *inode, struct file *filep);
 157 static int cciss_ioctl(struct inode *inode, struct file *filep,
 158                        unsigned int cmd, unsigned long arg);
 159 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 160
 161 static int cciss_revalidate(struct gendisk *disk);
 162 static int rebuild_lun_table(ctlr_info_t *h, int first_time);
 163 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
 164                            int clear_all);
 165
 166 static void cciss_read_capacity(int ctlr, int logvol, int withirq,
 167                         sector_t *total_size, unsigned int *block_size);
 168 static void cciss_read_capacity_16(int ctlr, int logvol, int withirq,
 169                         sector_t *total_size, unsigned int *block_size);
 170 static void cciss_geometry_inquiry(int ctlr, int logvol,
 171                         int withirq, sector_t total_size,
 172                         unsigned int block_size, InquiryData_struct *inq_buff,
 173                                    drive_info_struct *drv);
 174 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 175                                            __u32);
 176 static void start_io(ctlr_info_t *h);
 177 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
 178                    unsigned int use_unit_num, unsigned int log_unit,
 179                    __u8 page_code, unsigned char *scsi3addr, int cmd_type);
 180 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 181                            unsigned int use_unit_num, unsigned int log_unit,
 182                            __u8 page_code, int cmd_type);
 183
 184 static void fail_all_cmds(unsigned long ctlr);
 185
 186 #ifdef CONFIG_PROC_FS
 187 static void cciss_procinit(int i);
 188 #else
 189 static void cciss_procinit(int i)
 190 {
 191 }
 192 #endif                          /* CONFIG_PROC_FS */
 193
 194 #ifdef CONFIG_COMPAT
 195 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg);
 196 #endif
 197
 198 static struct block_device_operations cciss_fops = {
 199         .owner = THIS_MODULE,
 200         .open = cciss_open,
 201         .release = cciss_release,
 202         .ioctl = cciss_ioctl,
 203         .getgeo = cciss_getgeo,
 204 #ifdef CONFIG_COMPAT
 205         .compat_ioctl = cciss_compat_ioctl,
 206 #endif
 207         .revalidate_disk = cciss_revalidate,
 208 };
 209
 210 /*
 211  * Enqueuing and dequeuing functions for cmdlists.
 212  */
 213 static inline void addQ(CommandList_struct **Qptr, CommandList_struct *c)
 214 {
 215         if (*Qptr == NULL) {
 216                 *Qptr = c;
 217                 c->next = c->prev = c;
 218         } else {
 219                 c->prev = (*Qptr)->prev;
 220                 c->next = (*Qptr);
 221                 (*Qptr)->prev->next = c;
 222                 (*Qptr)->prev = c;
 223         }
 224 }
 225
 226 static inline CommandList_struct *removeQ(CommandList_struct **Qptr,
 227                                           CommandList_struct *c)
 228 {
 229         if (c && c->next != c) {
 230                 if (*Qptr == c)
 231                         *Qptr = c->next;
 232                 c->prev->next = c->next;
 233                 c->next->prev = c->prev;
 234         } else {
 235                 *Qptr = NULL;
 236         }
 237         return c;
 238 }
 239
 240 #include "cciss_scsi.c"         /* For SCSI tape support */
 241
 242 #define RAID_UNKNOWN 6
 243
 244 #ifdef CONFIG_PROC_FS
 245
 246 /*
 247  * Report information about this controller.
 248  */
 249 #define ENG_GIG 1000000000
 250 #define ENG_GIG_FACTOR (ENG_GIG/512)
 251 #define ENGAGE_SCSI     "engage scsi"
 252 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 253         "UNKNOWN"
 254 };
 255
 256 static struct proc_dir_entry *proc_cciss;
 257
 258 static void cciss_seq_show_header(struct seq_file *seq)
 259 {
 260         ctlr_info_t *h = seq->private;
 261
 262         seq_printf(seq, "%s: HP %s Controller\n"
 263                 "Board ID: 0x%08lx\n"
 264                 "Firmware Version: %c%c%c%c\n"
 265                 "IRQ: %d\n"
 266                 "Logical drives: %d\n"
 267                 "Current Q depth: %d\n"
 268                 "Current # commands on controller: %d\n"
 269                 "Max Q depth since init: %d\n"
 270                 "Max # commands on controller since init: %d\n"
 271                 "Max SG entries since init: %d\n",
 272                 h->devname,
 273                 h->product_name,
 274                 (unsigned long)h->board_id,
 275                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 276                 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
 277                 h->num_luns,
 278                 h->Qdepth, h->commands_outstanding,
 279                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 280
 281 #ifdef CONFIG_CISS_SCSI_TAPE
 282         cciss_seq_tape_report(seq, h->ctlr);
 283 #endif /* CONFIG_CISS_SCSI_TAPE */
 284 }
 285
 286 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 287 {
 288         ctlr_info_t *h = seq->private;
 289         unsigned ctlr = h->ctlr;
 290         unsigned long flags;
 291
 292         /* prevent displaying bogus info during configuration
 293          * or deconfiguration of a logical volume
 294          */
 295         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 296         if (h->busy_configuring) {
 297                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 298                 return ERR_PTR(-EBUSY);
 299         }
 300         h->busy_configuring = 1;
 301         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 302
 303         if (*pos == 0)
 304                 cciss_seq_show_header(seq);
 305
 306         return pos;
 307 }
 308
 309 static int cciss_seq_show(struct seq_file *seq, void *v)
 310 {
 311         sector_t vol_sz, vol_sz_frac;
 312         ctlr_info_t *h = seq->private;
 313         unsigned ctlr = h->ctlr;
 314         loff_t *pos = v;
 315         drive_info_struct *drv = &h->drv[*pos];
 316
 317         if (*pos > h->highest_lun)
 318                 return 0;
 319
 320         if (drv->heads == 0)
 321                 return 0;
 322
 323         vol_sz = drv->nr_blocks;
 324         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 325         vol_sz_frac *= 100;
 326         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 327
 328         if (drv->raid_level > 5)
 329                 drv->raid_level = RAID_UNKNOWN;
 330         seq_printf(seq, "cciss/c%dd%d:"
 331                         "\t%4u.%02uGB\tRAID %s\n",
 332                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 333                         raid_label[drv->raid_level]);
 334         return 0;
 335 }
 336
 337 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 338 {
 339         ctlr_info_t *h = seq->private;
 340
 341         if (*pos > h->highest_lun)
 342                 return NULL;
 343         *pos += 1;
 344
 345         return pos;
 346 }
 347
 348 static void cciss_seq_stop(struct seq_file *seq, void *v)
 349 {
 350         ctlr_info_t *h = seq->private;
 351
 352         /* Only reset h->busy_configuring if we succeeded in setting
 353          * it during cciss_seq_start. */
 354         if (v == ERR_PTR(-EBUSY))
 355                 return;
 356
 357         h->busy_configuring = 0;
 358 }
 359
 360 static struct seq_operations cciss_seq_ops = {
 361         .start = cciss_seq_start,
 362         .show  = cciss_seq_show,
 363         .next  = cciss_seq_next,
 364         .stop  = cciss_seq_stop,
 365 };
 366
 367 static int cciss_seq_open(struct inode *inode, struct file *file)
 368 {
 369         int ret = seq_open(file, &cciss_seq_ops);
 370         struct seq_file *seq = file->private_data;
 371
 372         if (!ret)
 373                 seq->private = PDE(inode)->data;
 374
 375         return ret;
 376 }
 377
 378 static ssize_t
 379 cciss_proc_write(struct file *file, const char __user *buf,
 380                  size_t length, loff_t *ppos)
 381 {
 382         int err;
 383         char *buffer;
 384
 385 #ifndef CONFIG_CISS_SCSI_TAPE
 386         return -EINVAL;
 387 #endif
 388
 389         if (!buf || length > PAGE_SIZE - 1)
 390                 return -EINVAL;
 391
 392         buffer = (char *)__get_free_page(GFP_KERNEL);
 393         if (!buffer)
 394                 return -ENOMEM;
 395
 396         err = -EFAULT;
 397         if (copy_from_user(buffer, buf, length))
 398                 goto out;
 399         buffer[length] = '\0';
 400
 401 #ifdef CONFIG_CISS_SCSI_TAPE
 402         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 403                 struct seq_file *seq = file->private_data;
 404                 ctlr_info_t *h = seq->private;
 405                 int rc;
 406
 407                 rc = cciss_engage_scsi(h->ctlr);
 408                 if (rc != 0)
 409                         err = -rc;
 410                 else
 411                         err = length;
 412         } else
 413 #endif /* CONFIG_CISS_SCSI_TAPE */
 414                 err = -EINVAL;
 415         /* might be nice to have "disengage" too, but it's not
 416            safely possible. (only 1 module use count, lock issues.) */
 417
 418 out:
 419         free_page((unsigned long)buffer);
 420         return err;
 421 }
 422
 423 static struct file_operations cciss_proc_fops = {
 424         .owner   = THIS_MODULE,
 425         .open    = cciss_seq_open,
 426         .read    = seq_read,
 427         .llseek  = seq_lseek,
 428         .release = seq_release,
 429         .write   = cciss_proc_write,
 430 };
 431
 432 static void __devinit cciss_procinit(int i)
 433 {
 434         struct proc_dir_entry *pde;
 435
 436         if (proc_cciss == NULL)
 437                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 438         if (!proc_cciss)
 439                 return;
 440         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 441                                         S_IROTH, proc_cciss,
 442                                         &cciss_proc_fops, hba[i]);
 443 }
 444 #endif                          /* CONFIG_PROC_FS */
 445
 446 /*
 447  * For operations that cannot sleep, a command block is allocated at init,
 448  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 449  * which ones are free or in use.  For operations that can wait for kmalloc
 450  * to possible sleep, this routine can be called with get_from_pool set to 0.
 451  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 452  */
 453 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 454 {
 455         CommandList_struct *c;
 456         int i;
 457         u64bit temp64;
 458         dma_addr_t cmd_dma_handle, err_dma_handle;
 459
 460         if (!get_from_pool) {
 461                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 462                         sizeof(CommandList_struct), &cmd_dma_handle);
 463                 if (c == NULL)
 464                         return NULL;
 465                 memset(c, 0, sizeof(CommandList_struct));
 466
 467                 c->cmdindex = -1;
 468
 469                 c->err_info = (ErrorInfo_struct *)
 470                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 471                             &err_dma_handle);
 472
 473                 if (c->err_info == NULL) {
 474                         pci_free_consistent(h->pdev,
 475                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 476                         return NULL;
 477                 }
 478                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 479         } else {                /* get it out of the controllers pool */
 480
 481                 do {
 482                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 483                         if (i == h->nr_cmds)
 484                                 return NULL;
 485                 } while (test_and_set_bit
 486                          (i & (BITS_PER_LONG - 1),
 487                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 488 #ifdef CCISS_DEBUG
 489                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 490 #endif
 491                 c = h->cmd_pool + i;
 492                 memset(c, 0, sizeof(CommandList_struct));
 493                 cmd_dma_handle = h->cmd_pool_dhandle
 494                     + i * sizeof(CommandList_struct);
 495                 c->err_info = h->errinfo_pool + i;
 496                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 497                 err_dma_handle = h->errinfo_pool_dhandle
 498                     + i * sizeof(ErrorInfo_struct);
 499                 h->nr_allocs++;
 500
 501                 c->cmdindex = i;
 502         }
 503
 504         c->busaddr = (__u32) cmd_dma_handle;
 505         temp64.val = (__u64) err_dma_handle;
 506         c->ErrDesc.Addr.lower = temp64.val32.lower;
 507         c->ErrDesc.Addr.upper = temp64.val32.upper;
 508         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 509
 510         c->ctlr = h->ctlr;
 511         return c;
 512 }
 513
 514 /*
 515  * Frees a command block that was previously allocated with cmd_alloc().
 516  */
 517 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 518 {
 519         int i;
 520         u64bit temp64;
 521
 522         if (!got_from_pool) {
 523                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 524                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 525                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 526                                     c->err_info, (dma_addr_t) temp64.val);
 527                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 528                                     c, (dma_addr_t) c->busaddr);
 529         } else {
 530                 i = c - h->cmd_pool;
 531                 clear_bit(i & (BITS_PER_LONG - 1),
 532                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 533                 h->nr_frees++;
 534         }
 535 }
 536
 537 static inline ctlr_info_t *get_host(struct gendisk *disk)
 538 {
 539         return disk->queue->queuedata;
 540 }
 541
 542 static inline drive_info_struct *get_drv(struct gendisk *disk)
 543 {
 544         return disk->private_data;
 545 }
 546
 547 /*
 548  * Open.  Make sure the device is really there.
 549  */
 550 static int cciss_open(struct inode *inode, struct file *filep)
 551 {
 552         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
 553         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
 554
 555 #ifdef CCISS_DEBUG
 556         printk(KERN_DEBUG "cciss_open %s\n", inode->i_bdev->bd_disk->disk_name);
 557 #endif                          /* CCISS_DEBUG */
 558
 559         if (host->busy_initializing || drv->busy_configuring)
 560                 return -EBUSY;
 561         /*
 562          * Root is allowed to open raw volume zero even if it's not configured
 563          * so array config can still work. Root is also allowed to open any
 564          * volume that has a LUN ID, so it can issue IOCTL to reread the
 565          * disk information.  I don't think I really like this
 566          * but I'm already using way to many device nodes to claim another one
 567          * for "raw controller".
 568          */
 569         if (drv->heads == 0) {
 570                 if (iminor(inode) != 0) {       /* not node 0? */
 571                         /* if not node 0 make sure it is a partition = 0 */
 572                         if (iminor(inode) & 0x0f) {
 573                                 return -ENXIO;
 574                                 /* if it is, make sure we have a LUN ID */
 575                         } else if (drv->LunID == 0) {
 576                                 return -ENXIO;
 577                         }
 578                 }
 579                 if (!capable(CAP_SYS_ADMIN))
 580                         return -EPERM;
 581         }
 582         drv->usage_count++;
 583         host->usage_count++;
 584         return 0;
 585 }
 586
 587 /*
 588  * Close.  Sync first.
 589  */
 590 static int cciss_release(struct inode *inode, struct file *filep)
 591 {
 592         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
 593         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
 594
 595 #ifdef CCISS_DEBUG
 596         printk(KERN_DEBUG "cciss_release %s\n",
 597                inode->i_bdev->bd_disk->disk_name);
 598 #endif                          /* CCISS_DEBUG */
 599
 600         drv->usage_count--;
 601         host->usage_count--;
 602         return 0;
 603 }
 604
 605 #ifdef CONFIG_COMPAT
 606
 607 static int do_ioctl(struct file *f, unsigned cmd, unsigned long arg)
 608 {
 609         int ret;
 610         lock_kernel();
 611         ret = cciss_ioctl(f->f_path.dentry->d_inode, f, cmd, arg);
 612         unlock_kernel();
 613         return ret;
 614 }
 615
 616 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd,
 617                                   unsigned long arg);
 618 static int cciss_ioctl32_big_passthru(struct file *f, unsigned cmd,
 619                                       unsigned long arg);
 620
 621 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg)
 622 {
 623         switch (cmd) {
 624         case CCISS_GETPCIINFO:
 625         case CCISS_GETINTINFO:
 626         case CCISS_SETINTINFO:
 627         case CCISS_GETNODENAME:
 628         case CCISS_SETNODENAME:
 629         case CCISS_GETHEARTBEAT:
 630         case CCISS_GETBUSTYPES:
 631         case CCISS_GETFIRMVER:
 632         case CCISS_GETDRIVVER:
 633         case CCISS_REVALIDVOLS:
 634         case CCISS_DEREGDISK:
 635         case CCISS_REGNEWDISK:
 636         case CCISS_REGNEWD:
 637         case CCISS_RESCANDISK:
 638         case CCISS_GETLUNINFO:
 639                 return do_ioctl(f, cmd, arg);
 640
 641         case CCISS_PASSTHRU32:
 642                 return cciss_ioctl32_passthru(f, cmd, arg);
 643         case CCISS_BIG_PASSTHRU32:
 644                 return cciss_ioctl32_big_passthru(f, cmd, arg);
 645
 646         default:
 647                 return -ENOIOCTLCMD;
 648         }
 649 }
 650
 651 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd,
 652                                   unsigned long arg)
 653 {
 654         IOCTL32_Command_struct __user *arg32 =
 655             (IOCTL32_Command_struct __user *) arg;
 656         IOCTL_Command_struct arg64;
 657         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
 658         int err;
 659         u32 cp;
 660
 661         err = 0;
 662         err |=
 663             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
 664                            sizeof(arg64.LUN_info));
 665         err |=
 666             copy_from_user(&arg64.Request, &arg32->Request,
 667                            sizeof(arg64.Request));
 668         err |=
 669             copy_from_user(&arg64.error_info, &arg32->error_info,
 670                            sizeof(arg64.error_info));
 671         err |= get_user(arg64.buf_size, &arg32->buf_size);
 672         err |= get_user(cp, &arg32->buf);
 673         arg64.buf = compat_ptr(cp);
 674         err |= copy_to_user(p, &arg64, sizeof(arg64));
 675
 676         if (err)
 677                 return -EFAULT;
 678
 679         err = do_ioctl(f, CCISS_PASSTHRU, (unsigned long)p);
 680         if (err)
 681                 return err;
 682         err |=
 683             copy_in_user(&arg32->error_info, &p->error_info,
 684                          sizeof(arg32->error_info));
 685         if (err)
 686                 return -EFAULT;
 687         return err;
 688 }
 689
 690 static int cciss_ioctl32_big_passthru(struct file *file, unsigned cmd,
 691                                       unsigned long arg)
 692 {
 693         BIG_IOCTL32_Command_struct __user *arg32 =
 694             (BIG_IOCTL32_Command_struct __user *) arg;
 695         BIG_IOCTL_Command_struct arg64;
 696         BIG_IOCTL_Command_struct __user *p =
 697             compat_alloc_user_space(sizeof(arg64));
 698         int err;
 699         u32 cp;
 700
 701         err = 0;
 702         err |=
 703             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
 704                            sizeof(arg64.LUN_info));
 705         err |=
 706             copy_from_user(&arg64.Request, &arg32->Request,
 707                            sizeof(arg64.Request));
 708         err |=
 709             copy_from_user(&arg64.error_info, &arg32->error_info,
 710                            sizeof(arg64.error_info));
 711         err |= get_user(arg64.buf_size, &arg32->buf_size);
 712         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
 713         err |= get_user(cp, &arg32->buf);
 714         arg64.buf = compat_ptr(cp);
 715         err |= copy_to_user(p, &arg64, sizeof(arg64));
 716
 717         if (err)
 718                 return -EFAULT;
 719
 720         err = do_ioctl(file, CCISS_BIG_PASSTHRU, (unsigned long)p);
 721         if (err)
 722                 return err;
 723         err |=
 724             copy_in_user(&arg32->error_info, &p->error_info,
 725                          sizeof(arg32->error_info));
 726         if (err)
 727                 return -EFAULT;
 728         return err;
 729 }
 730 #endif
 731
 732 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 733 {
 734         drive_info_struct *drv = get_drv(bdev->bd_disk);
 735
 736         if (!drv->cylinders)
 737                 return -ENXIO;
 738
 739         geo->heads = drv->heads;
 740         geo->sectors = drv->sectors;
 741         geo->cylinders = drv->cylinders;
 742         return 0;
 743 }
 744
 745 /*
 746  * ioctl
 747  */
 748 static int cciss_ioctl(struct inode *inode, struct file *filep,
 749                        unsigned int cmd, unsigned long arg)
 750 {
 751         struct block_device *bdev = inode->i_bdev;
 752         struct gendisk *disk = bdev->bd_disk;
 753         ctlr_info_t *host = get_host(disk);
 754         drive_info_struct *drv = get_drv(disk);
 755         int ctlr = host->ctlr;
 756         void __user *argp = (void __user *)arg;
 757
 758 #ifdef CCISS_DEBUG
 759         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
 760 #endif                          /* CCISS_DEBUG */
 761
 762         switch (cmd) {
 763         case CCISS_GETPCIINFO:
 764                 {
 765                         cciss_pci_info_struct pciinfo;
 766
 767                         if (!arg)
 768                                 return -EINVAL;
 769                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
 770                         pciinfo.bus = host->pdev->bus->number;
 771                         pciinfo.dev_fn = host->pdev->devfn;
 772                         pciinfo.board_id = host->board_id;
 773                         if (copy_to_user
 774                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
 775                                 return -EFAULT;
 776                         return 0;
 777                 }
 778         case CCISS_GETINTINFO:
 779                 {
 780                         cciss_coalint_struct intinfo;
 781                         if (!arg)
 782                                 return -EINVAL;
 783                         intinfo.delay =
 784                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
 785                         intinfo.count =
 786                             readl(&host->cfgtable->HostWrite.CoalIntCount);
 787                         if (copy_to_user
 788                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
 789                                 return -EFAULT;
 790                         return 0;
 791                 }
 792         case CCISS_SETINTINFO:
 793                 {
 794                         cciss_coalint_struct intinfo;
 795                         unsigned long flags;
 796                         int i;
 797
 798                         if (!arg)
 799                                 return -EINVAL;
 800                         if (!capable(CAP_SYS_ADMIN))
 801                                 return -EPERM;
 802                         if (copy_from_user
 803                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
 804                                 return -EFAULT;
 805                         if ((intinfo.delay == 0) && (intinfo.count == 0))
 806                         {
 807 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
 808                                 return -EINVAL;
 809                         }
 810                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 811                         /* Update the field, and then ring the doorbell */
 812                         writel(intinfo.delay,
 813                                &(host->cfgtable->HostWrite.CoalIntDelay));
 814                         writel(intinfo.count,
 815                                &(host->cfgtable->HostWrite.CoalIntCount));
 816                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
 817
 818                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
 819                                 if (!(readl(host->vaddr + SA5_DOORBELL)
 820                                       & CFGTBL_ChangeReq))
 821                                         break;
 822                                 /* delay and try again */
 823                                 udelay(1000);
 824                         }
 825                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 826                         if (i >= MAX_IOCTL_CONFIG_WAIT)
 827                                 return -EAGAIN;
 828                         return 0;
 829                 }
 830         case CCISS_GETNODENAME:
 831                 {
 832                         NodeName_type NodeName;
 833                         int i;
 834
 835                         if (!arg)
 836                                 return -EINVAL;
 837                         for (i = 0; i < 16; i++)
 838                                 NodeName[i] =
 839                                     readb(&host->cfgtable->ServerName[i]);
 840                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
 841                                 return -EFAULT;
 842                         return 0;
 843                 }
 844         case CCISS_SETNODENAME:
 845                 {
 846                         NodeName_type NodeName;
 847                         unsigned long flags;
 848                         int i;
 849
 850                         if (!arg)
 851                                 return -EINVAL;
 852                         if (!capable(CAP_SYS_ADMIN))
 853                                 return -EPERM;
 854
 855                         if (copy_from_user
 856                             (NodeName, argp, sizeof(NodeName_type)))
 857                                 return -EFAULT;
 858
 859                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 860
 861                         /* Update the field, and then ring the doorbell */
 862                         for (i = 0; i < 16; i++)
 863                                 writeb(NodeName[i],
 864                                        &host->cfgtable->ServerName[i]);
 865
 866                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
 867
 868                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
 869                                 if (!(readl(host->vaddr + SA5_DOORBELL)
 870                                       & CFGTBL_ChangeReq))
 871                                         break;
 872                                 /* delay and try again */
 873                                 udelay(1000);
 874                         }
 875                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 876                         if (i >= MAX_IOCTL_CONFIG_WAIT)
 877                                 return -EAGAIN;
 878                         return 0;
 879                 }
 880
 881         case CCISS_GETHEARTBEAT:
 882                 {
 883                         Heartbeat_type heartbeat;
 884
 885                         if (!arg)
 886                                 return -EINVAL;
 887                         heartbeat = readl(&host->cfgtable->HeartBeat);
 888                         if (copy_to_user
 889                             (argp, &heartbeat, sizeof(Heartbeat_type)))
 890                                 return -EFAULT;
 891                         return 0;
 892                 }
 893         case CCISS_GETBUSTYPES:
 894                 {
 895                         BusTypes_type BusTypes;
 896
 897                         if (!arg)
 898                                 return -EINVAL;
 899                         BusTypes = readl(&host->cfgtable->BusTypes);
 900                         if (copy_to_user
 901                             (argp, &BusTypes, sizeof(BusTypes_type)))
 902                                 return -EFAULT;
 903                         return 0;
 904                 }
 905         case CCISS_GETFIRMVER:
 906                 {
 907                         FirmwareVer_type firmware;
 908
 909                         if (!arg)
 910                                 return -EINVAL;
 911                         memcpy(firmware, host->firm_ver, 4);
 912
 913                         if (copy_to_user
 914                             (argp, firmware, sizeof(FirmwareVer_type)))
 915                                 return -EFAULT;
 916                         return 0;
 917                 }
 918         case CCISS_GETDRIVVER:
 919                 {
 920                         DriverVer_type DriverVer = DRIVER_VERSION;
 921
 922                         if (!arg)
 923                                 return -EINVAL;
 924
 925                         if (copy_to_user
 926                             (argp, &DriverVer, sizeof(DriverVer_type)))
 927                                 return -EFAULT;
 928                         return 0;
 929                 }
 930
 931         case CCISS_DEREGDISK:
 932         case CCISS_REGNEWD:
 933         case CCISS_REVALIDVOLS:
 934                 return rebuild_lun_table(host, 0);
 935
 936         case CCISS_GETLUNINFO:{
 937                         LogvolInfo_struct luninfo;
 938
 939                         luninfo.LunID = drv->LunID;
 940                         luninfo.num_opens = drv->usage_count;
 941                         luninfo.num_parts = 0;
 942                         if (copy_to_user(argp, &luninfo,
 943                                          sizeof(LogvolInfo_struct)))
 944                                 return -EFAULT;
 945                         return 0;
 946                 }
 947         case CCISS_PASSTHRU:
 948                 {
 949                         IOCTL_Command_struct iocommand;
 950                         CommandList_struct *c;
 951                         char *buff = NULL;
 952                         u64bit temp64;
 953                         unsigned long flags;
 954                         DECLARE_COMPLETION_ONSTACK(wait);
 955
 956                         if (!arg)
 957                                 return -EINVAL;
 958
 959                         if (!capable(CAP_SYS_RAWIO))
 960                                 return -EPERM;
 961
 962                         if (copy_from_user
 963                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
 964                                 return -EFAULT;
 965                         if ((iocommand.buf_size < 1) &&
 966                             (iocommand.Request.Type.Direction != XFER_NONE)) {
 967                                 return -EINVAL;
 968                         }
 969 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
 970                         /* Check kmalloc limits */
 971                         if (iocommand.buf_size > 128000)
 972                                 return -EINVAL;
 973 #endif
 974                         if (iocommand.buf_size > 0) {
 975                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
 976                                 if (buff == NULL)
 977                                         return -EFAULT;
 978                         }
 979                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
 980                                 /* Copy the data into the buffer we created */
 981                                 if (copy_from_user
 982                                     (buff, iocommand.buf, iocommand.buf_size)) {
 983                                         kfree(buff);
 984                                         return -EFAULT;
 985                                 }
 986                         } else {
 987                                 memset(buff, 0, iocommand.buf_size);
 988                         }
 989                         if ((c = cmd_alloc(host, 0)) == NULL) {
 990                                 kfree(buff);
 991                                 return -ENOMEM;
 992                         }
 993                         // Fill in the command type
 994                         c->cmd_type = CMD_IOCTL_PEND;
 995                         // Fill in Command Header
 996                         c->Header.ReplyQueue = 0;       // unused in simple mode
 997                         if (iocommand.buf_size > 0)     // buffer to fill
 998                         {
 999                                 c->Header.SGList = 1;
1000                                 c->Header.SGTotal = 1;
1001                         } else  // no buffers to fill
1002                         {
1003                                 c->Header.SGList = 0;
1004                                 c->Header.SGTotal = 0;
1005                         }
1006                         c->Header.LUN = iocommand.LUN_info;
1007                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
1008
1009                         // Fill in Request block
1010                         c->Request = iocommand.Request;
1011
1012                         // Fill in the scatter gather information
1013                         if (iocommand.buf_size > 0) {
1014                                 temp64.val = pci_map_single(host->pdev, buff,
1015                                         iocommand.buf_size,
1016                                         PCI_DMA_BIDIRECTIONAL);
1017                                 c->SG[0].Addr.lower = temp64.val32.lower;
1018                                 c->SG[0].Addr.upper = temp64.val32.upper;
1019                                 c->SG[0].Len = iocommand.buf_size;
1020                                 c->SG[0].Ext = 0;       // we are not chaining
1021                         }
1022                         c->waiting = &wait;
1023
1024                         /* Put the request on the tail of the request queue */
1025                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1026                         addQ(&host->reqQ, c);
1027                         host->Qdepth++;
1028                         start_io(host);
1029                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1030
1031                         wait_for_completion(&wait);
1032
1033                         /* unlock the buffers from DMA */
1034                         temp64.val32.lower = c->SG[0].Addr.lower;
1035                         temp64.val32.upper = c->SG[0].Addr.upper;
1036                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1037                                          iocommand.buf_size,
1038                                          PCI_DMA_BIDIRECTIONAL);
1039
1040                         /* Copy the error information out */
1041                         iocommand.error_info = *(c->err_info);
1042                         if (copy_to_user
1043                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1044                                 kfree(buff);
1045                                 cmd_free(host, c, 0);
1046                                 return -EFAULT;
1047                         }
1048
1049                         if (iocommand.Request.Type.Direction == XFER_READ) {
1050                                 /* Copy the data out of the buffer we created */
1051                                 if (copy_to_user
1052                                     (iocommand.buf, buff, iocommand.buf_size)) {
1053                                         kfree(buff);
1054                                         cmd_free(host, c, 0);
1055                                         return -EFAULT;
1056                                 }
1057                         }
1058                         kfree(buff);
1059                         cmd_free(host, c, 0);
1060                         return 0;
1061                 }
1062         case CCISS_BIG_PASSTHRU:{
1063                         BIG_IOCTL_Command_struct *ioc;
1064                         CommandList_struct *c;
1065                         unsigned char **buff = NULL;
1066                         int *buff_size = NULL;
1067                         u64bit temp64;
1068                         unsigned long flags;
1069                         BYTE sg_used = 0;
1070                         int status = 0;
1071                         int i;
1072                         DECLARE_COMPLETION_ONSTACK(wait);
1073                         __u32 left;
1074                         __u32 sz;
1075                         BYTE __user *data_ptr;
1076
1077                         if (!arg)
1078                                 return -EINVAL;
1079                         if (!capable(CAP_SYS_RAWIO))
1080                                 return -EPERM;
1081                         ioc = (BIG_IOCTL_Command_struct *)
1082                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1083                         if (!ioc) {
1084                                 status = -ENOMEM;
1085                                 goto cleanup1;
1086                         }
1087                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1088                                 status = -EFAULT;
1089                                 goto cleanup1;
1090                         }
1091                         if ((ioc->buf_size < 1) &&
1092                             (ioc->Request.Type.Direction != XFER_NONE)) {
1093                                 status = -EINVAL;
1094                                 goto cleanup1;
1095                         }
1096                         /* Check kmalloc limits  using all SGs */
1097                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1098                                 status = -EINVAL;
1099                                 goto cleanup1;
1100                         }
1101                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1102                                 status = -EINVAL;
1103                                 goto cleanup1;
1104                         }
1105                         buff =
1106                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1107                         if (!buff) {
1108                                 status = -ENOMEM;
1109                                 goto cleanup1;
1110                         }
1111                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1112                                                    GFP_KERNEL);
1113                         if (!buff_size) {
1114                                 status = -ENOMEM;
1115                                 goto cleanup1;
1116                         }
1117                         left = ioc->buf_size;
1118                         data_ptr = ioc->buf;
1119                         while (left) {
1120                                 sz = (left >
1121                                       ioc->malloc_size) ? ioc->
1122                                     malloc_size : left;
1123                                 buff_size[sg_used] = sz;
1124                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1125                                 if (buff[sg_used] == NULL) {
1126                                         status = -ENOMEM;
1127                                         goto cleanup1;
1128                                 }
1129                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1130                                         if (copy_from_user
1131                                             (buff[sg_used], data_ptr, sz)) {
1132                                                 status = -EFAULT;
1133                                                 goto cleanup1;
1134                                         }
1135                                 } else {
1136                                         memset(buff[sg_used], 0, sz);
1137                                 }
1138                                 left -= sz;
1139                                 data_ptr += sz;
1140                                 sg_used++;
1141                         }
1142                         if ((c = cmd_alloc(host, 0)) == NULL) {
1143                                 status = -ENOMEM;
1144                                 goto cleanup1;
1145                         }
1146                         c->cmd_type = CMD_IOCTL_PEND;
1147                         c->Header.ReplyQueue = 0;
1148
1149                         if (ioc->buf_size > 0) {
1150                                 c->Header.SGList = sg_used;
1151                                 c->Header.SGTotal = sg_used;
1152                         } else {
1153                                 c->Header.SGList = 0;
1154                                 c->Header.SGTotal = 0;
1155                         }
1156                         c->Header.LUN = ioc->LUN_info;
1157                         c->Header.Tag.lower = c->busaddr;
1158
1159                         c->Request = ioc->Request;
1160                         if (ioc->buf_size > 0) {
1161                                 int i;
1162                                 for (i = 0; i < sg_used; i++) {
1163                                         temp64.val =
1164                                             pci_map_single(host->pdev, buff[i],
1165                                                     buff_size[i],
1166                                                     PCI_DMA_BIDIRECTIONAL);
1167                                         c->SG[i].Addr.lower =
1168                                             temp64.val32.lower;
1169                                         c->SG[i].Addr.upper =
1170                                             temp64.val32.upper;
1171                                         c->SG[i].Len = buff_size[i];
1172                                         c->SG[i].Ext = 0;       /* we are not chaining */
1173                                 }
1174                         }
1175                         c->waiting = &wait;
1176                         /* Put the request on the tail of the request queue */
1177                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1178                         addQ(&host->reqQ, c);
1179                         host->Qdepth++;
1180                         start_io(host);
1181                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1182                         wait_for_completion(&wait);
1183                         /* unlock the buffers from DMA */
1184                         for (i = 0; i < sg_used; i++) {
1185                                 temp64.val32.lower = c->SG[i].Addr.lower;
1186                                 temp64.val32.upper = c->SG[i].Addr.upper;
1187                                 pci_unmap_single(host->pdev,
1188                                         (dma_addr_t) temp64.val, buff_size[i],
1189                                         PCI_DMA_BIDIRECTIONAL);
1190                         }
1191                         /* Copy the error information out */
1192                         ioc->error_info = *(c->err_info);
1193                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1194                                 cmd_free(host, c, 0);
1195                                 status = -EFAULT;
1196                                 goto cleanup1;
1197                         }
1198                         if (ioc->Request.Type.Direction == XFER_READ) {
1199                                 /* Copy the data out of the buffer we created */
1200                                 BYTE __user *ptr = ioc->buf;
1201                                 for (i = 0; i < sg_used; i++) {
1202                                         if (copy_to_user
1203                                             (ptr, buff[i], buff_size[i])) {
1204                                                 cmd_free(host, c, 0);
1205                                                 status = -EFAULT;
1206                                                 goto cleanup1;
1207                                         }
1208                                         ptr += buff_size[i];
1209                                 }
1210                         }
1211                         cmd_free(host, c, 0);
1212                         status = 0;
1213                       cleanup1:
1214                         if (buff) {
1215                                 for (i = 0; i < sg_used; i++)
1216                                         kfree(buff[i]);
1217                                 kfree(buff);
1218                         }
1219                         kfree(buff_size);
1220                         kfree(ioc);
1221                         return status;
1222                 }
1223
1224         /* scsi_cmd_ioctl handles these, below, though some are not */
1225         /* very meaningful for cciss.  SG_IO is the main one people want. */
1226
1227         case SG_GET_VERSION_NUM:
1228         case SG_SET_TIMEOUT:
1229         case SG_GET_TIMEOUT:
1230         case SG_GET_RESERVED_SIZE:
1231         case SG_SET_RESERVED_SIZE:
1232         case SG_EMULATED_HOST:
1233         case SG_IO:
1234         case SCSI_IOCTL_SEND_COMMAND:
1235                 return scsi_cmd_ioctl(filep, disk->queue, disk, cmd, argp);
1236
1237         /* scsi_cmd_ioctl would normally handle these, below, but */
1238         /* they aren't a good fit for cciss, as CD-ROMs are */
1239         /* not supported, and we don't have any bus/target/lun */
1240         /* which we present to the kernel. */
1241
1242         case CDROM_SEND_PACKET:
1243         case CDROMCLOSETRAY:
1244         case CDROMEJECT:
1245         case SCSI_IOCTL_GET_IDLUN:
1246         case SCSI_IOCTL_GET_BUS_NUMBER:
1247         default:
1248                 return -ENOTTY;
1249         }
1250 }
1251
1252 static void cciss_check_queues(ctlr_info_t *h)
1253 {
1254         int start_queue = h->next_to_run;
1255         int i;
1256
1257         /* check to see if we have maxed out the number of commands that can
1258          * be placed on the queue.  If so then exit.  We do this check here
1259          * in case the interrupt we serviced was from an ioctl and did not
1260          * free any new commands.
1261          */
1262         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1263                 return;
1264
1265         /* We have room on the queue for more commands.  Now we need to queue
1266          * them up.  We will also keep track of the next queue to run so
1267          * that every queue gets a chance to be started first.
1268          */
1269         for (i = 0; i < h->highest_lun + 1; i++) {
1270                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1271                 /* make sure the disk has been added and the drive is real
1272                  * because this can be called from the middle of init_one.
1273                  */
1274                 if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
1275                         continue;
1276                 blk_start_queue(h->gendisk[curr_queue]->queue);
1277
1278                 /* check to see if we have maxed out the number of commands
1279                  * that can be placed on the queue.
1280                  */
1281                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1282                         if (curr_queue == start_queue) {
1283                                 h->next_to_run =
1284                                     (start_queue + 1) % (h->highest_lun + 1);
1285                                 break;
1286                         } else {
1287                                 h->next_to_run = curr_queue;
1288                                 break;
1289                         }
1290                 }
1291         }
1292 }
1293
1294 static void cciss_softirq_done(struct request *rq)
1295 {
1296         CommandList_struct *cmd = rq->completion_data;
1297         ctlr_info_t *h = hba[cmd->ctlr];
1298         unsigned long flags;
1299         u64bit temp64;
1300         int i, ddir;
1301
1302         if (cmd->Request.Type.Direction == XFER_READ)
1303                 ddir = PCI_DMA_FROMDEVICE;
1304         else
1305                 ddir = PCI_DMA_TODEVICE;
1306
1307         /* command did not need to be retried */
1308         /* unmap the DMA mapping for all the scatter gather elements */
1309         for (i = 0; i < cmd->Header.SGList; i++) {
1310                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1311                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1312                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1313         }
1314
1315 #ifdef CCISS_DEBUG
1316         printk("Done with %p\n", rq);
1317 #endif                          /* CCISS_DEBUG */
1318
1319         if (blk_end_request(rq, (rq->errors == 0) ? 0 : -EIO, blk_rq_bytes(rq)))
1320                 BUG();
1321
1322         spin_lock_irqsave(&h->lock, flags);
1323         cmd_free(h, cmd, 1);
1324         cciss_check_queues(h);
1325         spin_unlock_irqrestore(&h->lock, flags);
1326 }
1327
1328 /* This function gets the serial number of a logical drive via
1329  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1330  * number cannot be had, for whatever reason, 16 bytes of 0xff
1331  * are returned instead.
1332  */
1333 static void cciss_get_serial_no(int ctlr, int logvol, int withirq,
1334                                 unsigned char *serial_no, int buflen)
1335 {
1336 #define PAGE_83_INQ_BYTES 64
1337         int rc;
1338         unsigned char *buf;
1339
1340         if (buflen > 16)
1341                 buflen = 16;
1342         memset(serial_no, 0xff, buflen);
1343         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1344         if (!buf)
1345                 return;
1346         memset(serial_no, 0, buflen);
1347         if (withirq)
1348                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1349                         PAGE_83_INQ_BYTES, 1, logvol, 0x83, TYPE_CMD);
1350         else
1351                 rc = sendcmd(CISS_INQUIRY, ctlr, buf,
1352                         PAGE_83_INQ_BYTES, 1, logvol, 0x83, NULL, TYPE_CMD);
1353         if (rc == IO_OK)
1354                 memcpy(serial_no, &buf[8], buflen);
1355         kfree(buf);
1356         return;
1357 }
1358
1359 static void cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1360                                 int drv_index)
1361 {
1362         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1363         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1364         disk->major = h->major;
1365         disk->first_minor = drv_index << NWD_SHIFT;
1366         disk->fops = &cciss_fops;
1367         disk->private_data = &h->drv[drv_index];
1368
1369         /* Set up queue information */
1370         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1371
1372         /* This is a hardware imposed limit. */
1373         blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1374
1375         /* This is a limit in the driver and could be eliminated. */
1376         blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1377
1378         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1379
1380         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1381
1382         disk->queue->queuedata = h;
1383
1384         blk_queue_hardsect_size(disk->queue,
1385                                 h->drv[drv_index].block_size);
1386
1387         /* Make sure all queue data is written out before */
1388         /* setting h->drv[drv_index].queue, as setting this */
1389         /* allows the interrupt handler to start the queue */
1390         wmb();
1391         h->drv[drv_index].queue = disk->queue;
1392         add_disk(disk);
1393 }
1394
1395 /* This function will check the usage_count of the drive to be updated/added.
1396  * If the usage_count is zero and it is a heretofore unknown drive, or,
1397  * the drive's capacity, geometry, or serial number has changed,
1398  * then the drive information will be updated and the disk will be
1399  * re-registered with the kernel.  If these conditions don't hold,
1400  * then it will be left alone for the next reboot.  The exception to this
1401  * is disk 0 which will always be left registered with the kernel since it
1402  * is also the controller node.  Any changes to disk 0 will show up on
1403  * the next reboot.
1404  */
1405 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time)
1406 {
1407         ctlr_info_t *h = hba[ctlr];
1408         struct gendisk *disk;
1409         InquiryData_struct *inq_buff = NULL;
1410         unsigned int block_size;
1411         sector_t total_size;
1412         unsigned long flags = 0;
1413         int ret = 0;
1414         drive_info_struct *drvinfo;
1415         int was_only_controller_node;
1416
1417         /* Get information about the disk and modify the driver structure */
1418         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1419         drvinfo = kmalloc(sizeof(*drvinfo), GFP_KERNEL);
1420         if (inq_buff == NULL || drvinfo == NULL)
1421                 goto mem_msg;
1422
1423         /* See if we're trying to update the "controller node"
1424          * this will happen the when the first logical drive gets
1425          * created by ACU.
1426          */
1427         was_only_controller_node = (drv_index == 0 &&
1428                                 h->drv[0].raid_level == -1);
1429
1430         /* testing to see if 16-byte CDBs are already being used */
1431         if (h->cciss_read == CCISS_READ_16) {
1432                 cciss_read_capacity_16(h->ctlr, drv_index, 1,
1433                         &total_size, &block_size);
1434
1435         } else {
1436                 cciss_read_capacity(ctlr, drv_index, 1,
1437                                     &total_size, &block_size);
1438
1439                 /* if read_capacity returns all F's this volume is >2TB */
1440                 /* in size so we switch to 16-byte CDB's for all */
1441                 /* read/write ops */
1442                 if (total_size == 0xFFFFFFFFULL) {
1443                         cciss_read_capacity_16(ctlr, drv_index, 1,
1444                         &total_size, &block_size);
1445                         h->cciss_read = CCISS_READ_16;
1446                         h->cciss_write = CCISS_WRITE_16;
1447                 } else {
1448                         h->cciss_read = CCISS_READ_10;
1449                         h->cciss_write = CCISS_WRITE_10;
1450                 }
1451         }
1452
1453         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1454                                inq_buff, drvinfo);
1455         drvinfo->block_size = block_size;
1456         drvinfo->nr_blocks = total_size + 1;
1457
1458         cciss_get_serial_no(ctlr, drv_index, 1, drvinfo->serial_no,
1459                         sizeof(drvinfo->serial_no));
1460
1461         /* Is it the same disk we already know, and nothing's changed? */
1462         if (h->drv[drv_index].raid_level != -1 &&
1463                 ((memcmp(drvinfo->serial_no,
1464                                 h->drv[drv_index].serial_no, 16) == 0) &&
1465                 drvinfo->block_size == h->drv[drv_index].block_size &&
1466                 drvinfo->nr_blocks == h->drv[drv_index].nr_blocks &&
1467                 drvinfo->heads == h->drv[drv_index].heads &&
1468                 drvinfo->sectors == h->drv[drv_index].sectors &&
1469                 drvinfo->cylinders == h->drv[drv_index].cylinders))
1470                         /* The disk is unchanged, nothing to update */
1471                         goto freeret;
1472
1473         /* If we get here it's not the same disk, or something's changed,
1474          * so we need to * deregister it, and re-register it, if it's not
1475          * in use.
1476          * If the disk already exists then deregister it before proceeding
1477          * (unless it's the first disk (for the controller node).
1478          */
1479         if (h->drv[drv_index].raid_level != -1 && drv_index != 0) {
1480                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1481                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1482                 h->drv[drv_index].busy_configuring = 1;
1483                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1484
1485                 /* deregister_disk sets h->drv[drv_index].queue = NULL
1486                  * which keeps the interrupt handler from starting
1487                  * the queue.
1488                  */
1489                 ret = deregister_disk(h->gendisk[drv_index],
1490                                       &h->drv[drv_index], 0);
1491                 h->drv[drv_index].busy_configuring = 0;
1492         }
1493
1494         /* If the disk is in use return */
1495         if (ret)
1496                 goto freeret;
1497
1498         /* Save the new information from cciss_geometry_inquiry
1499          * and serial number inquiry.
1500          */
1501         h->drv[drv_index].block_size = drvinfo->block_size;
1502         h->drv[drv_index].nr_blocks = drvinfo->nr_blocks;
1503         h->drv[drv_index].heads = drvinfo->heads;
1504         h->drv[drv_index].sectors = drvinfo->sectors;
1505         h->drv[drv_index].cylinders = drvinfo->cylinders;
1506         h->drv[drv_index].raid_level = drvinfo->raid_level;
1507         memcpy(h->drv[drv_index].serial_no, drvinfo->serial_no, 16);
1508
1509         ++h->num_luns;
1510         disk = h->gendisk[drv_index];
1511         set_capacity(disk, h->drv[drv_index].nr_blocks);
1512
1513         /* If it's not disk 0 (drv_index != 0)
1514          * or if it was disk 0, but there was previously
1515          * no actual corresponding configured logical drive
1516          * (raid_leve == -1) then we want to update the
1517          * logical drive's information.
1518          */
1519         if (drv_index || first_time)
1520                 cciss_add_disk(h, disk, drv_index);
1521
1522 freeret:
1523         kfree(inq_buff);
1524         kfree(drvinfo);
1525         return;
1526 mem_msg:
1527         printk(KERN_ERR "cciss: out of memory\n");
1528         goto freeret;
1529 }
1530
1531 /* This function will find the first index of the controllers drive array
1532  * that has a -1 for the raid_level and will return that index.  This is
1533  * where new drives will be added.  If the index to be returned is greater
1534  * than the highest_lun index for the controller then highest_lun is set
1535  * to this new index.  If there are no available indexes then -1 is returned.
1536  */
1537 static int cciss_find_free_drive_index(int ctlr)
1538 {
1539         int i;
1540
1541         for (i = 0; i < CISS_MAX_LUN; i++) {
1542                 if (hba[ctlr]->drv[i].raid_level == -1) {
1543                         if (i > hba[ctlr]->highest_lun)
1544                                 hba[ctlr]->highest_lun = i;
1545                         return i;
1546                 }
1547         }
1548         return -1;
1549 }
1550
1551 /* cciss_add_gendisk finds a free hba[]->drv structure
1552  * and allocates a gendisk if needed, and sets the lunid
1553  * in the drvinfo structure.   It returns the index into
1554  * the ->drv[] array, or -1 if none are free.
1555  * is_controller_node indicates whether highest_lun should
1556  * count this disk, or if it's only being added to provide
1557  * a means to talk to the controller in case no logical
1558  * drives have yet been configured.
1559  */
1560 static int cciss_add_gendisk(ctlr_info_t *h, __u32 lunid)
1561 {
1562         int drv_index;
1563
1564         drv_index = cciss_find_free_drive_index(h->ctlr);
1565         if (drv_index == -1)
1566                 return -1;
1567         /*Check if the gendisk needs to be allocated */
1568         if (!h->gendisk[drv_index]) {
1569                 h->gendisk[drv_index] =
1570                         alloc_disk(1 << NWD_SHIFT);
1571                 if (!h->gendisk[drv_index]) {
1572                         printk(KERN_ERR "cciss%d: could not "
1573                                 "allocate a new disk %d\n",
1574                                 h->ctlr, drv_index);
1575                         return -1;
1576                 }
1577         }
1578         h->drv[drv_index].LunID = lunid;
1579
1580         /* Don't need to mark this busy because nobody */
1581         /* else knows about this disk yet to contend */
1582         /* for access to it. */
1583         h->drv[drv_index].busy_configuring = 0;
1584         wmb();
1585         return drv_index;
1586 }
1587
1588 /* This is for the special case of a controller which
1589  * has no logical drives.  In this case, we still need
1590  * to register a disk so the controller can be accessed
1591  * by the Array Config Utility.
1592  */
1593 static void cciss_add_controller_node(ctlr_info_t *h)
1594 {
1595         struct gendisk *disk;
1596         int drv_index;
1597
1598         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
1599                 return;
1600
1601         drv_index = cciss_add_gendisk(h, 0);
1602         if (drv_index == -1) {
1603                 printk(KERN_WARNING "cciss%d: could not "
1604                         "add disk 0.\n", h->ctlr);
1605                 return;
1606         }
1607         h->drv[drv_index].block_size = 512;
1608         h->drv[drv_index].nr_blocks = 0;
1609         h->drv[drv_index].heads = 0;
1610         h->drv[drv_index].sectors = 0;
1611         h->drv[drv_index].cylinders = 0;
1612         h->drv[drv_index].raid_level = -1;
1613         memset(h->drv[drv_index].serial_no, 0, 16);
1614         disk = h->gendisk[drv_index];
1615         cciss_add_disk(h, disk, drv_index);
1616 }
1617
1618 /* This function will add and remove logical drives from the Logical
1619  * drive array of the controller and maintain persistency of ordering
1620  * so that mount points are preserved until the next reboot.  This allows
1621  * for the removal of logical drives in the middle of the drive array
1622  * without a re-ordering of those drives.
1623  * INPUT
1624  * h            = The controller to perform the operations on
1625  */
1626 static int rebuild_lun_table(ctlr_info_t *h, int first_time)
1627 {
1628         int ctlr = h->ctlr;
1629         int num_luns;
1630         ReportLunData_struct *ld_buff = NULL;
1631         int return_code;
1632         int listlength = 0;
1633         int i;
1634         int drv_found;
1635         int drv_index = 0;
1636         __u32 lunid = 0;
1637         unsigned long flags;
1638
1639         if (!capable(CAP_SYS_RAWIO))
1640                 return -EPERM;
1641
1642         /* Set busy_configuring flag for this operation */
1643         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1644         if (h->busy_configuring) {
1645                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1646                 return -EBUSY;
1647         }
1648         h->busy_configuring = 1;
1649         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1650
1651         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1652         if (ld_buff == NULL)
1653                 goto mem_msg;
1654
1655         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
1656                                       sizeof(ReportLunData_struct), 0,
1657                                       0, 0, TYPE_CMD);
1658
1659         if (return_code == IO_OK)
1660                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
1661         else {  /* reading number of logical volumes failed */
1662                 printk(KERN_WARNING "cciss: report logical volume"
1663                        " command failed\n");
1664                 listlength = 0;
1665                 goto freeret;
1666         }
1667
1668         num_luns = listlength / 8;      /* 8 bytes per entry */
1669         if (num_luns > CISS_MAX_LUN) {
1670                 num_luns = CISS_MAX_LUN;
1671                 printk(KERN_WARNING "cciss: more luns configured"
1672                        " on controller than can be handled by"
1673                        " this driver.\n");
1674         }
1675
1676         if (num_luns == 0)
1677                 cciss_add_controller_node(h);
1678
1679         /* Compare controller drive array to driver's drive array
1680          * to see if any drives are missing on the controller due
1681          * to action of Array Config Utility (user deletes drive)
1682          * and deregister logical drives which have disappeared.
1683          */
1684         for (i = 0; i <= h->highest_lun; i++) {
1685                 int j;
1686                 drv_found = 0;
1687                 for (j = 0; j < num_luns; j++) {
1688                         memcpy(&lunid, &ld_buff->LUN[j][0], 4);
1689                         lunid = le32_to_cpu(lunid);
1690                         if (h->drv[i].LunID == lunid) {
1691                                 drv_found = 1;
1692                                 break;
1693                         }
1694                 }
1695                 if (!drv_found) {
1696                         /* Deregister it from the OS, it's gone. */
1697                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1698                         h->drv[i].busy_configuring = 1;
1699                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1700                         return_code = deregister_disk(h->gendisk[i],
1701                                 &h->drv[i], 1);
1702                         h->drv[i].busy_configuring = 0;
1703                 }
1704         }
1705
1706         /* Compare controller drive array to driver's drive array.
1707          * Check for updates in the drive information and any new drives
1708          * on the controller due to ACU adding logical drives, or changing
1709          * a logical drive's size, etc.  Reregister any new/changed drives
1710          */
1711         for (i = 0; i < num_luns; i++) {
1712                 int j;
1713
1714                 drv_found = 0;
1715
1716                 memcpy(&lunid, &ld_buff->LUN[i][0], 4);
1717                 lunid = le32_to_cpu(lunid);
1718
1719                 /* Find if the LUN is already in the drive array
1720                  * of the driver.  If so then update its info
1721                  * if not in use.  If it does not exist then find
1722                  * the first free index and add it.
1723                  */
1724                 for (j = 0; j <= h->highest_lun; j++) {
1725                         if (h->drv[j].raid_level != -1 &&
1726                                 h->drv[j].LunID == lunid) {
1727                                 drv_index = j;
1728                                 drv_found = 1;
1729                                 break;
1730                         }
1731                 }
1732
1733                 /* check if the drive was found already in the array */
1734                 if (!drv_found) {
1735                         drv_index = cciss_add_gendisk(h, lunid);
1736                         if (drv_index == -1)
1737                                 goto freeret;
1738                 }
1739                 cciss_update_drive_info(ctlr, drv_index, first_time);
1740         }               /* end for */
1741
1742 freeret:
1743         kfree(ld_buff);
1744         h->busy_configuring = 0;
1745         /* We return -1 here to tell the ACU that we have registered/updated
1746          * all of the drives that we can and to keep it from calling us
1747          * additional times.
1748          */
1749         return -1;
1750 mem_msg:
1751         printk(KERN_ERR "cciss: out of memory\n");
1752         h->busy_configuring = 0;
1753         goto freeret;
1754 }
1755
1756 /* This function will deregister the disk and it's queue from the
1757  * kernel.  It must be called with the controller lock held and the
1758  * drv structures busy_configuring flag set.  It's parameters are:
1759  *
1760  * disk = This is the disk to be deregistered
1761  * drv  = This is the drive_info_struct associated with the disk to be
1762  *        deregistered.  It contains information about the disk used
1763  *        by the driver.
1764  * clear_all = This flag determines whether or not the disk information
1765  *             is going to be completely cleared out and the highest_lun
1766  *             reset.  Sometimes we want to clear out information about
1767  *             the disk in preparation for re-adding it.  In this case
1768  *             the highest_lun should be left unchanged and the LunID
1769  *             should not be cleared.
1770 */
1771 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
1772                            int clear_all)
1773 {
1774         int i;
1775         ctlr_info_t *h = get_host(disk);
1776
1777         if (!capable(CAP_SYS_RAWIO))
1778                 return -EPERM;
1779
1780         /* make sure logical volume is NOT is use */
1781         if (clear_all || (h->gendisk[0] == disk)) {
1782                 if (drv->usage_count > 1)
1783                         return -EBUSY;
1784         } else if (drv->usage_count > 0)
1785                 return -EBUSY;
1786
1787         /* invalidate the devices and deregister the disk.  If it is disk
1788          * zero do not deregister it but just zero out it's values.  This
1789          * allows us to delete disk zero but keep the controller registered.
1790          */
1791         if (h->gendisk[0] != disk) {
1792                 struct request_queue *q = disk->queue;
1793                 if (disk->flags & GENHD_FL_UP)
1794                         del_gendisk(disk);
1795                 if (q) {
1796                         blk_cleanup_queue(q);
1797                         /* Set drv->queue to NULL so that we do not try
1798                          * to call blk_start_queue on this queue in the
1799                          * interrupt handler
1800                          */
1801                         drv->queue = NULL;
1802                 }
1803                 /* If clear_all is set then we are deleting the logical
1804                  * drive, not just refreshing its info.  For drives
1805                  * other than disk 0 we will call put_disk.  We do not
1806                  * do this for disk 0 as we need it to be able to
1807                  * configure the controller.
1808                  */
1809                 if (clear_all){
1810                         /* This isn't pretty, but we need to find the
1811                          * disk in our array and NULL our the pointer.
1812                          * This is so that we will call alloc_disk if
1813                          * this index is used again later.
1814                          */
1815                         for (i=0; i < CISS_MAX_LUN; i++){
1816                                 if (h->gendisk[i] == disk) {
1817                                         h->gendisk[i] = NULL;
1818                                         break;
1819                                 }
1820                         }
1821                         put_disk(disk);
1822                 }
1823         } else {
1824                 set_capacity(disk, 0);
1825         }
1826
1827         --h->num_luns;
1828         /* zero out the disk size info */
1829         drv->nr_blocks = 0;
1830         drv->block_size = 0;
1831         drv->heads = 0;
1832         drv->sectors = 0;
1833         drv->cylinders = 0;
1834         drv->raid_level = -1;   /* This can be used as a flag variable to
1835                                  * indicate that this element of the drive
1836                                  * array is free.
1837                                  */
1838
1839         if (clear_all) {
1840                 /* check to see if it was the last disk */
1841                 if (drv == h->drv + h->highest_lun) {
1842                         /* if so, find the new hightest lun */
1843                         int i, newhighest = -1;
1844                         for (i = 0; i <= h->highest_lun; i++) {
1845                                 /* if the disk has size > 0, it is available */
1846                                 if (h->drv[i].heads)
1847                                         newhighest = i;
1848                         }
1849                         h->highest_lun = newhighest;
1850                 }
1851
1852                 drv->LunID = 0;
1853         }
1854         return 0;
1855 }
1856
1857 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff, size_t size, unsigned int use_unit_num,      /* 0: address the controller,
1858                                                                                                                            1: address logical volume log_unit,
1859                                                                                                                            2: periph device address is scsi3addr */
1860                     unsigned int log_unit, __u8 page_code,
1861                     unsigned char *scsi3addr, int cmd_type)
1862 {
1863         ctlr_info_t *h = hba[ctlr];
1864         u64bit buff_dma_handle;
1865         int status = IO_OK;
1866
1867         c->cmd_type = CMD_IOCTL_PEND;
1868         c->Header.ReplyQueue = 0;
1869         if (buff != NULL) {
1870                 c->Header.SGList = 1;
1871                 c->Header.SGTotal = 1;
1872         } else {
1873                 c->Header.SGList = 0;
1874                 c->Header.SGTotal = 0;
1875         }
1876         c->Header.Tag.lower = c->busaddr;
1877
1878         c->Request.Type.Type = cmd_type;
1879         if (cmd_type == TYPE_CMD) {
1880                 switch (cmd) {
1881                 case CISS_INQUIRY:
1882                         /* If the logical unit number is 0 then, this is going
1883                            to controller so It's a physical command
1884                            mode = 0 target = 0.  So we have nothing to write.
1885                            otherwise, if use_unit_num == 1,
1886                            mode = 1(volume set addressing) target = LUNID
1887                            otherwise, if use_unit_num == 2,
1888                            mode = 0(periph dev addr) target = scsi3addr */
1889                         if (use_unit_num == 1) {
1890                                 c->Header.LUN.LogDev.VolId =
1891                                     h->drv[log_unit].LunID;
1892                                 c->Header.LUN.LogDev.Mode = 1;
1893                         } else if (use_unit_num == 2) {
1894                                 memcpy(c->Header.LUN.LunAddrBytes, scsi3addr,
1895                                        8);
1896                                 c->Header.LUN.LogDev.Mode = 0;
1897                         }
1898                         /* are we trying to read a vital product page */
1899                         if (page_code != 0) {
1900                                 c->Request.CDB[1] = 0x01;
1901                                 c->Request.CDB[2] = page_code;
1902                         }
1903                         c->Request.CDBLen = 6;
1904                         c->Request.Type.Attribute = ATTR_SIMPLE;
1905                         c->Request.Type.Direction = XFER_READ;
1906                         c->Request.Timeout = 0;
1907                         c->Request.CDB[0] = CISS_INQUIRY;
1908                         c->Request.CDB[4] = size & 0xFF;
1909                         break;
1910                 case CISS_REPORT_LOG:
1911                 case CISS_REPORT_PHYS:
1912                         /* Talking to controller so It's a physical command
1913                            mode = 00 target = 0.  Nothing to write.
1914                          */
1915                         c->Request.CDBLen = 12;
1916                         c->Request.Type.Attribute = ATTR_SIMPLE;
1917                         c->Request.Type.Direction = XFER_READ;
1918                         c->Request.Timeout = 0;
1919                         c->Request.CDB[0] = cmd;
1920                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
1921                         c->Request.CDB[7] = (size >> 16) & 0xFF;
1922                         c->Request.CDB[8] = (size >> 8) & 0xFF;
1923                         c->Request.CDB[9] = size & 0xFF;
1924                         break;
1925
1926                 case CCISS_READ_CAPACITY:
1927                         c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1928                         c->Header.LUN.LogDev.Mode = 1;
1929                         c->Request.CDBLen = 10;
1930                         c->Request.Type.Attribute = ATTR_SIMPLE;
1931                         c->Request.Type.Direction = XFER_READ;
1932                         c->Request.Timeout = 0;
1933                         c->Request.CDB[0] = cmd;
1934                         break;
1935                 case CCISS_READ_CAPACITY_16:
1936                         c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1937                         c->Header.LUN.LogDev.Mode = 1;
1938                         c->Request.CDBLen = 16;
1939                         c->Request.Type.Attribute = ATTR_SIMPLE;
1940                         c->Request.Type.Direction = XFER_READ;
1941                         c->Request.Timeout = 0;
1942                         c->Request.CDB[0] = cmd;
1943                         c->Request.CDB[1] = 0x10;
1944                         c->Request.CDB[10] = (size >> 24) & 0xFF;
1945                         c->Request.CDB[11] = (size >> 16) & 0xFF;
1946                         c->Request.CDB[12] = (size >> 8) & 0xFF;
1947                         c->Request.CDB[13] = size & 0xFF;
1948                         c->Request.Timeout = 0;
1949                         c->Request.CDB[0] = cmd;
1950                         break;
1951                 case CCISS_CACHE_FLUSH:
1952                         c->Request.CDBLen = 12;
1953                         c->Request.Type.Attribute = ATTR_SIMPLE;
1954                         c->Request.Type.Direction = XFER_WRITE;
1955                         c->Request.Timeout = 0;
1956                         c->Request.CDB[0] = BMIC_WRITE;
1957                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
1958                         break;
1959                 default:
1960                         printk(KERN_WARNING
1961                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
1962                         return IO_ERROR;
1963                 }
1964         } else if (cmd_type == TYPE_MSG) {
1965                 switch (cmd) {
1966                 case 0: /* ABORT message */
1967                         c->Request.CDBLen = 12;
1968                         c->Request.Type.Attribute = ATTR_SIMPLE;
1969                         c->Request.Type.Direction = XFER_WRITE;
1970                         c->Request.Timeout = 0;
1971                         c->Request.CDB[0] = cmd;        /* abort */
1972                         c->Request.CDB[1] = 0;  /* abort a command */
1973                         /* buff contains the tag of the command to abort */
1974                         memcpy(&c->Request.CDB[4], buff, 8);
1975                         break;
1976                 case 1: /* RESET message */
1977                         c->Request.CDBLen = 12;
1978                         c->Request.Type.Attribute = ATTR_SIMPLE;
1979                         c->Request.Type.Direction = XFER_WRITE;
1980                         c->Request.Timeout = 0;
1981                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
1982                         c->Request.CDB[0] = cmd;        /* reset */
1983                         c->Request.CDB[1] = 0x04;       /* reset a LUN */
1984                         break;
1985                 case 3: /* No-Op message */
1986                         c->Request.CDBLen = 1;
1987                         c->Request.Type.Attribute = ATTR_SIMPLE;
1988                         c->Request.Type.Direction = XFER_WRITE;
1989                         c->Request.Timeout = 0;
1990                         c->Request.CDB[0] = cmd;
1991                         break;
1992                 default:
1993                         printk(KERN_WARNING
1994                                "cciss%d: unknown message type %d\n", ctlr, cmd);
1995                         return IO_ERROR;
1996                 }
1997         } else {
1998                 printk(KERN_WARNING
1999                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2000                 return IO_ERROR;
2001         }
2002         /* Fill in the scatter gather information */
2003         if (size > 0) {
2004                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2005                                                              buff, size,
2006                                                              PCI_DMA_BIDIRECTIONAL);
2007                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2008                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2009                 c->SG[0].Len = size;
2010                 c->SG[0].Ext = 0;       /* we are not chaining */
2011         }
2012         return status;
2013 }
2014
2015 static int sendcmd_withirq(__u8 cmd,
2016                            int ctlr,
2017                            void *buff,
2018                            size_t size,
2019                            unsigned int use_unit_num,
2020                            unsigned int log_unit, __u8 page_code, int cmd_type)
2021 {
2022         ctlr_info_t *h = hba[ctlr];
2023         CommandList_struct *c;
2024         u64bit buff_dma_handle;
2025         unsigned long flags;
2026         int return_status;
2027         DECLARE_COMPLETION_ONSTACK(wait);
2028
2029         if ((c = cmd_alloc(h, 0)) == NULL)
2030                 return -ENOMEM;
2031         return_status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
2032                                  log_unit, page_code, NULL, cmd_type);
2033         if (return_status != IO_OK) {
2034                 cmd_free(h, c, 0);
2035                 return return_status;
2036         }
2037       resend_cmd2:
2038         c->waiting = &wait;
2039
2040         /* Put the request on the tail of the queue and send it */
2041         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
2042         addQ(&h->reqQ, c);
2043         h->Qdepth++;
2044         start_io(h);
2045         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
2046
2047         wait_for_completion(&wait);
2048
2049         if (c->err_info->CommandStatus != 0) {  /* an error has occurred */
2050                 switch (c->err_info->CommandStatus) {
2051                 case CMD_TARGET_STATUS:
2052                         printk(KERN_WARNING "cciss: cmd %p has "
2053                                " completed with errors\n", c);
2054                         if (c->err_info->ScsiStatus) {
2055                                 printk(KERN_WARNING "cciss: cmd %p "
2056                                        "has SCSI Status = %x\n",
2057                                        c, c->err_info->ScsiStatus);
2058                         }
2059
2060                         break;
2061                 case CMD_DATA_UNDERRUN:
2062                 case CMD_DATA_OVERRUN:
2063                         /* expected for inquire and report lun commands */
2064                         break;
2065                 case CMD_INVALID:
2066                         printk(KERN_WARNING "cciss: Cmd %p is "
2067                                "reported invalid\n", c);
2068                         return_status = IO_ERROR;
2069                         break;
2070                 case CMD_PROTOCOL_ERR:
2071                         printk(KERN_WARNING "cciss: cmd %p has "
2072                                "protocol error \n", c);
2073                         return_status = IO_ERROR;
2074                         break;
2075                 case CMD_HARDWARE_ERR:
2076                         printk(KERN_WARNING "cciss: cmd %p had "
2077                                " hardware error\n", c);
2078                         return_status = IO_ERROR;
2079                         break;
2080                 case CMD_CONNECTION_LOST:
2081                         printk(KERN_WARNING "cciss: cmd %p had "
2082                                "connection lost\n", c);
2083                         return_status = IO_ERROR;
2084                         break;
2085                 case CMD_ABORTED:
2086                         printk(KERN_WARNING "cciss: cmd %p was "
2087                                "aborted\n", c);
2088                         return_status = IO_ERROR;
2089                         break;
2090                 case CMD_ABORT_FAILED:
2091                         printk(KERN_WARNING "cciss: cmd %p reports "
2092                                "abort failed\n", c);
2093                         return_status = IO_ERROR;
2094                         break;
2095                 case CMD_UNSOLICITED_ABORT:
2096                         printk(KERN_WARNING
2097                                "cciss%d: unsolicited abort %p\n", ctlr, c);
2098                         if (c->retry_count < MAX_CMD_RETRIES) {
2099                                 printk(KERN_WARNING
2100                                        "cciss%d: retrying %p\n", ctlr, c);
2101                                 c->retry_count++;
2102                                 /* erase the old error information */
2103                                 memset(c->err_info, 0,
2104                                        sizeof(ErrorInfo_struct));
2105                                 return_status = IO_OK;
2106                                 INIT_COMPLETION(wait);
2107                                 goto resend_cmd2;
2108                         }
2109                         return_status = IO_ERROR;
2110                         break;
2111                 default:
2112                         printk(KERN_WARNING "cciss: cmd %p returned "
2113                                "unknown status %x\n", c,
2114                                c->err_info->CommandStatus);
2115                         return_status = IO_ERROR;
2116                 }
2117         }
2118         /* unlock the buffers from DMA */
2119         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2120         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2121         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2122                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2123         cmd_free(h, c, 0);
2124         return return_status;
2125 }
2126
2127 static void cciss_geometry_inquiry(int ctlr, int logvol,
2128                                    int withirq, sector_t total_size,
2129                                    unsigned int block_size,
2130                                    InquiryData_struct *inq_buff,
2131                                    drive_info_struct *drv)
2132 {
2133         int return_code;
2134         unsigned long t;
2135
2136         memset(inq_buff, 0, sizeof(InquiryData_struct));
2137         if (withirq)
2138                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
2139                                               inq_buff, sizeof(*inq_buff), 1,
2140                                               logvol, 0xC1, TYPE_CMD);
2141         else
2142                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
2143                                       sizeof(*inq_buff), 1, logvol, 0xC1, NULL,
2144                                       TYPE_CMD);
2145         if (return_code == IO_OK) {
2146                 if (inq_buff->data_byte[8] == 0xFF) {
2147                         printk(KERN_WARNING
2148                                "cciss: reading geometry failed, volume "
2149                                "does not support reading geometry\n");
2150                         drv->heads = 255;
2151                         drv->sectors = 32;      // Sectors per track
2152                         drv->cylinders = total_size + 1;
2153                         drv->raid_level = RAID_UNKNOWN;
2154                 } else {
2155                         drv->heads = inq_buff->data_byte[6];
2156                         drv->sectors = inq_buff->data_byte[7];
2157                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2158                         drv->cylinders += inq_buff->data_byte[5];
2159                         drv->raid_level = inq_buff->data_byte[8];
2160                 }
2161                 drv->block_size = block_size;
2162                 drv->nr_blocks = total_size + 1;
2163                 t = drv->heads * drv->sectors;
2164                 if (t > 1) {
2165                         sector_t real_size = total_size + 1;
2166                         unsigned long rem = sector_div(real_size, t);
2167                         if (rem)
2168                                 real_size++;
2169                         drv->cylinders = real_size;
2170                 }
2171         } else {                /* Get geometry failed */
2172                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2173         }
2174         printk(KERN_INFO "      heads=%d, sectors=%d, cylinders=%d\n\n",
2175                drv->heads, drv->sectors, drv->cylinders);
2176 }
2177
2178 static void
2179 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
2180                     unsigned int *block_size)
2181 {
2182         ReadCapdata_struct *buf;
2183         int return_code;
2184
2185         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2186         if (!buf) {
2187                 printk(KERN_WARNING "cciss: out of memory\n");
2188                 return;
2189         }
2190
2191         if (withirq)
2192                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
2193                                 ctlr, buf, sizeof(ReadCapdata_struct),
2194                                         1, logvol, 0, TYPE_CMD);
2195         else
2196                 return_code = sendcmd(CCISS_READ_CAPACITY,
2197                                 ctlr, buf, sizeof(ReadCapdata_struct),
2198                                         1, logvol, 0, NULL, TYPE_CMD);
2199         if (return_code == IO_OK) {
2200                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2201                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2202         } else {                /* read capacity command failed */
2203                 printk(KERN_WARNING "cciss: read capacity failed\n");
2204                 *total_size = 0;
2205                 *block_size = BLOCK_SIZE;
2206         }
2207         if (*total_size != 0)
2208                 printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2209                 (unsigned long long)*total_size+1, *block_size);
2210         kfree(buf);
2211 }
2212
2213 static void
2214 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size,                                 unsigned int *block_size)
2215 {
2216         ReadCapdata_struct_16 *buf;
2217         int return_code;
2218
2219         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2220         if (!buf) {
2221                 printk(KERN_WARNING "cciss: out of memory\n");
2222                 return;
2223         }
2224
2225         if (withirq) {
2226                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2227                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2228                                 1, logvol, 0, TYPE_CMD);
2229         }
2230         else {
2231                 return_code = sendcmd(CCISS_READ_CAPACITY_16,
2232                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2233                                 1, logvol, 0, NULL, TYPE_CMD);
2234         }
2235         if (return_code == IO_OK) {
2236                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2237                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2238         } else {                /* read capacity command failed */
2239                 printk(KERN_WARNING "cciss: read capacity failed\n");
2240                 *total_size = 0;
2241                 *block_size = BLOCK_SIZE;
2242         }
2243         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2244                (unsigned long long)*total_size+1, *block_size);
2245         kfree(buf);
2246 }
2247
2248 static int cciss_revalidate(struct gendisk *disk)
2249 {
2250         ctlr_info_t *h = get_host(disk);
2251         drive_info_struct *drv = get_drv(disk);
2252         int logvol;
2253         int FOUND = 0;
2254         unsigned int block_size;
2255         sector_t total_size;
2256         InquiryData_struct *inq_buff = NULL;
2257
2258         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2259                 if (h->drv[logvol].LunID == drv->LunID) {
2260                         FOUND = 1;
2261                         break;
2262                 }
2263         }
2264
2265         if (!FOUND)
2266                 return 1;
2267
2268         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2269         if (inq_buff == NULL) {
2270                 printk(KERN_WARNING "cciss: out of memory\n");
2271                 return 1;
2272         }
2273         if (h->cciss_read == CCISS_READ_10) {
2274                 cciss_read_capacity(h->ctlr, logvol, 1,
2275                                         &total_size, &block_size);
2276         } else {
2277                 cciss_read_capacity_16(h->ctlr, logvol, 1,
2278                                         &total_size, &block_size);
2279         }
2280         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2281                                inq_buff, drv);
2282
2283         blk_queue_hardsect_size(drv->queue, drv->block_size);
2284         set_capacity(disk, drv->nr_blocks);
2285
2286         kfree(inq_buff);
2287         return 0;
2288 }
2289
2290 /*
2291  *   Wait polling for a command to complete.
2292  *   The memory mapped FIFO is polled for the completion.
2293  *   Used only at init time, interrupts from the HBA are disabled.
2294  */
2295 static unsigned long pollcomplete(int ctlr)
2296 {
2297         unsigned long done;
2298         int i;
2299
2300         /* Wait (up to 20 seconds) for a command to complete */
2301
2302         for (i = 20 * HZ; i > 0; i--) {
2303                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2304                 if (done == FIFO_EMPTY)
2305                         schedule_timeout_uninterruptible(1);
2306                 else
2307                         return done;
2308         }
2309         /* Invalid address to tell caller we ran out of time */
2310         return 1;
2311 }
2312
2313 static int add_sendcmd_reject(__u8 cmd, int ctlr, unsigned long complete)
2314 {
2315         /* We get in here if sendcmd() is polling for completions
2316            and gets some command back that it wasn't expecting --
2317            something other than that which it just sent down.
2318            Ordinarily, that shouldn't happen, but it can happen when
2319            the scsi tape stuff gets into error handling mode, and
2320            starts using sendcmd() to try to abort commands and
2321            reset tape drives.  In that case, sendcmd may pick up
2322            completions of commands that were sent to logical drives
2323            through the block i/o system, or cciss ioctls completing, etc.
2324            In that case, we need to save those completions for later
2325            processing by the interrupt handler.
2326          */
2327
2328 #ifdef CONFIG_CISS_SCSI_TAPE
2329         struct sendcmd_reject_list *srl = &hba[ctlr]->scsi_rejects;
2330
2331         /* If it's not the scsi tape stuff doing error handling, (abort */
2332         /* or reset) then we don't expect anything weird. */
2333         if (cmd != CCISS_RESET_MSG && cmd != CCISS_ABORT_MSG) {
2334 #endif
2335                 printk(KERN_WARNING "cciss cciss%d: SendCmd "
2336                        "Invalid command list address returned! (%lx)\n",
2337                        ctlr, complete);
2338                 /* not much we can do. */
2339 #ifdef CONFIG_CISS_SCSI_TAPE
2340                 return 1;
2341         }
2342
2343         /* We've sent down an abort or reset, but something else
2344            has completed */
2345         if (srl->ncompletions >= (hba[ctlr]->nr_cmds + 2)) {
2346                 /* Uh oh.  No room to save it for later... */
2347                 printk(KERN_WARNING "cciss%d: Sendcmd: Invalid command addr, "
2348                        "reject list overflow, command lost!\n", ctlr);
2349                 return 1;
2350         }
2351         /* Save it for later */
2352         srl->complete[srl->ncompletions] = complete;
2353         srl->ncompletions++;
2354 #endif
2355         return 0;
2356 }
2357
2358 /*
2359  * Send a command to the controller, and wait for it to complete.
2360  * Only used at init time.
2361  */
2362 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size, unsigned int use_unit_num,      /* 0: address the controller,
2363                                                                                                    1: address logical volume log_unit,
2364                                                                                                    2: periph device address is scsi3addr */
2365                    unsigned int log_unit,
2366                    __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2367 {
2368         CommandList_struct *c;
2369         int i;
2370         unsigned long complete;
2371         ctlr_info_t *info_p = hba[ctlr];
2372         u64bit buff_dma_handle;
2373         int status, done = 0;
2374
2375         if ((c = cmd_alloc(info_p, 1)) == NULL) {
2376                 printk(KERN_WARNING "cciss: unable to get memory");
2377                 return IO_ERROR;
2378         }
2379         status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
2380                           log_unit, page_code, scsi3addr, cmd_type);
2381         if (status != IO_OK) {
2382                 cmd_free(info_p, c, 1);
2383                 return status;
2384         }
2385       resend_cmd1:
2386         /*
2387          * Disable interrupt
2388          */
2389 #ifdef CCISS_DEBUG
2390         printk(KERN_DEBUG "cciss: turning intr off\n");
2391 #endif                          /* CCISS_DEBUG */
2392         info_p->access.set_intr_mask(info_p, CCISS_INTR_OFF);
2393
2394         /* Make sure there is room in the command FIFO */
2395         /* Actually it should be completely empty at this time */
2396         /* unless we are in here doing error handling for the scsi */
2397         /* tape side of the driver. */
2398         for (i = 200000; i > 0; i--) {
2399                 /* if fifo isn't full go */
2400                 if (!(info_p->access.fifo_full(info_p))) {
2401
2402                         break;
2403                 }
2404                 udelay(10);
2405                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2406                        " waiting!\n", ctlr);
2407         }
2408         /*
2409          * Send the cmd
2410          */
2411         info_p->access.submit_command(info_p, c);
2412         done = 0;
2413         do {
2414                 complete = pollcomplete(ctlr);
2415
2416 #ifdef CCISS_DEBUG
2417                 printk(KERN_DEBUG "cciss: command completed\n");
2418 #endif                          /* CCISS_DEBUG */
2419
2420                 if (complete == 1) {
2421                         printk(KERN_WARNING
2422                                "cciss cciss%d: SendCmd Timeout out, "
2423                                "No command list address returned!\n", ctlr);
2424                         status = IO_ERROR;
2425                         done = 1;
2426                         break;
2427                 }
2428
2429                 /* This will need to change for direct lookup completions */
2430                 if ((complete & CISS_ERROR_BIT)
2431                     && (complete & ~CISS_ERROR_BIT) == c->busaddr) {
2432                         /* if data overrun or underun on Report command
2433                            ignore it
2434                          */
2435                         if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2436                              (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2437                              (c->Request.CDB[0] == CISS_INQUIRY)) &&
2438                             ((c->err_info->CommandStatus ==
2439                               CMD_DATA_OVERRUN) ||
2440                              (c->err_info->CommandStatus == CMD_DATA_UNDERRUN)
2441                             )) {
2442                                 complete = c->busaddr;
2443                         } else {
2444                                 if (c->err_info->CommandStatus ==
2445                                     CMD_UNSOLICITED_ABORT) {
2446                                         printk(KERN_WARNING "cciss%d: "
2447                                                "unsolicited abort %p\n",
2448                                                ctlr, c);
2449                                         if (c->retry_count < MAX_CMD_RETRIES) {
2450                                                 printk(KERN_WARNING
2451                                                        "cciss%d: retrying %p\n",
2452                                                        ctlr, c);
2453                                                 c->retry_count++;
2454                                                 /* erase the old error */
2455                                                 /* information */
2456                                                 memset(c->err_info, 0,
2457                                                        sizeof
2458                                                        (ErrorInfo_struct));
2459                                                 goto resend_cmd1;
2460                                         } else {
2461                                                 printk(KERN_WARNING
2462                                                        "cciss%d: retried %p too "
2463                                                        "many times\n", ctlr, c);
2464                                                 status = IO_ERROR;
2465                                                 goto cleanup1;
2466                                         }
2467                                 } else if (c->err_info->CommandStatus ==
2468                                            CMD_UNABORTABLE) {
2469                                         printk(KERN_WARNING
2470                                                "cciss%d: command could not be aborted.\n",
2471                                                ctlr);
2472                                         status = IO_ERROR;
2473                                         goto cleanup1;
2474                                 }
2475                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2476                                        " Error %x \n", ctlr,
2477                                        c->err_info->CommandStatus);
2478                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2479                                        " offensive info\n"
2480                                        "  size %x\n   num %x   value %x\n",
2481                                        ctlr,
2482                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2483                                        offense_size,
2484                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2485                                        offense_num,
2486                                        c->err_info->MoreErrInfo.Invalid_Cmd.
2487                                        offense_value);
2488                                 status = IO_ERROR;
2489                                 goto cleanup1;
2490                         }
2491                 }
2492                 /* This will need changing for direct lookup completions */
2493                 if (complete != c->busaddr) {
2494                         if (add_sendcmd_reject(cmd, ctlr, complete) != 0) {
2495                                 BUG();  /* we are pretty much hosed if we get here. */
2496                         }
2497                         continue;
2498                 } else
2499                         done = 1;
2500         } while (!done);
2501
2502       cleanup1:
2503         /* unlock the data buffer from DMA */
2504         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2505         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2506         pci_unmap_single(info_p->pdev, (dma_addr_t) buff_dma_handle.val,
2507                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2508 #ifdef CONFIG_CISS_SCSI_TAPE
2509         /* if we saved some commands for later, process them now. */
2510         if (info_p->scsi_rejects.ncompletions > 0)
2511                 do_cciss_intr(0, info_p);
2512 #endif
2513         cmd_free(info_p, c, 1);
2514         return status;
2515 }
2516
2517 /*
2518  * Map (physical) PCI mem into (virtual) kernel space
2519  */
2520 static void __iomem *remap_pci_mem(ulong base, ulong size)
2521 {
2522         ulong page_base = ((ulong) base) & PAGE_MASK;
2523         ulong page_offs = ((ulong) base) - page_base;
2524         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2525
2526         return page_remapped ? (page_remapped + page_offs) : NULL;
2527 }
2528
2529 /*
2530  * Takes jobs of the Q and sends them to the hardware, then puts it on
2531  * the Q to wait for completion.
2532  */
2533 static void start_io(ctlr_info_t *h)
2534 {
2535         CommandList_struct *c;
2536
2537         while ((c = h->reqQ) != NULL) {
2538                 /* can't do anything if fifo is full */
2539                 if ((h->access.fifo_full(h))) {
2540                         printk(KERN_WARNING "cciss: fifo full\n");
2541                         break;
2542                 }
2543
2544                 /* Get the first entry from the Request Q */
2545                 removeQ(&(h->reqQ), c);
2546                 h->Qdepth--;
2547
2548                 /* Tell the controller execute command */
2549                 h->access.submit_command(h, c);
2550
2551                 /* Put job onto the completed Q */
2552                 addQ(&(h->cmpQ), c);
2553         }
2554 }
2555
2556 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2557 /* Zeros out the error record and then resends the command back */
2558 /* to the controller */
2559 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2560 {
2561         /* erase the old error information */
2562         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2563
2564         /* add it to software queue and then send it to the controller */
2565         addQ(&(h->reqQ), c);
2566         h->Qdepth++;
2567         if (h->Qdepth > h->maxQsinceinit)
2568                 h->maxQsinceinit = h->Qdepth;
2569
2570         start_io(h);
2571 }
2572
2573 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2574         unsigned int msg_byte, unsigned int host_byte,
2575         unsigned int driver_byte)
2576 {
2577         /* inverse of macros in scsi.h */
2578         return (scsi_status_byte & 0xff) |
2579                 ((msg_byte & 0xff) << 8) |
2580                 ((host_byte & 0xff) << 16) |
2581                 ((driver_byte & 0xff) << 24);
2582 }
2583
2584 static inline int evaluate_target_status(CommandList_struct *cmd)
2585 {
2586         unsigned char sense_key;
2587         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2588         int error_value;
2589
2590         /* If we get in here, it means we got "target status", that is, scsi status */
2591         status_byte = cmd->err_info->ScsiStatus;
2592         driver_byte = DRIVER_OK;
2593         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2594
2595         if (blk_pc_request(cmd->rq))
2596                 host_byte = DID_PASSTHROUGH;
2597         else
2598                 host_byte = DID_OK;
2599
2600         error_value = make_status_bytes(status_byte, msg_byte,
2601                 host_byte, driver_byte);
2602
2603         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2604                 if (!blk_pc_request(cmd->rq))
2605                         printk(KERN_WARNING "cciss: cmd %p "
2606                                "has SCSI Status 0x%x\n",
2607                                cmd, cmd->err_info->ScsiStatus);
2608                 return error_value;
2609         }
2610
2611         /* check the sense key */
2612         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2613         /* no status or recovered error */
2614         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2615                 error_value = 0;
2616
2617         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2618                 if (error_value != 0)
2619                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2620                                " sense key = 0x%x\n", cmd, sense_key);
2621                 return error_value;
2622         }
2623
2624         /* SG_IO or similar, copy sense data back */
2625         if (cmd->rq->sense) {
2626                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2627                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2628                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2629                         cmd->rq->sense_len);
2630         } else
2631                 cmd->rq->sense_len = 0;
2632
2633         return error_value;
2634 }
2635
2636 /* checks the status of the job and calls complete buffers to mark all
2637  * buffers for the completed job. Note that this function does not need
2638  * to hold the hba/queue lock.
2639  */
2640 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2641                                     int timeout)
2642 {
2643         int retry_cmd = 0;
2644         struct request *rq = cmd->rq;
2645
2646         rq->errors = 0;
2647
2648         if (timeout)
2649                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
2650
2651         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
2652                 goto after_error_processing;
2653
2654         switch (cmd->err_info->CommandStatus) {
2655         case CMD_TARGET_STATUS:
2656                 rq->errors = evaluate_target_status(cmd);
2657                 break;
2658         case CMD_DATA_UNDERRUN:
2659                 if (blk_fs_request(cmd->rq)) {
2660                         printk(KERN_WARNING "cciss: cmd %p has"
2661                                " completed with data underrun "
2662                                "reported\n", cmd);
2663                         cmd->rq->data_len = cmd->err_info->ResidualCnt;
2664                 }
2665                 break;
2666         case CMD_DATA_OVERRUN:
2667                 if (blk_fs_request(cmd->rq))
2668                         printk(KERN_WARNING "cciss: cmd %p has"
2669                                " completed with data overrun "
2670                                "reported\n", cmd);
2671                 break;
2672         case CMD_INVALID:
2673                 printk(KERN_WARNING "cciss: cmd %p is "
2674                        "reported invalid\n", cmd);
2675                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2676                         cmd->err_info->CommandStatus, DRIVER_OK,
2677                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2678                 break;
2679         case CMD_PROTOCOL_ERR:
2680                 printk(KERN_WARNING "cciss: cmd %p has "
2681                        "protocol error \n", cmd);
2682                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2683                         cmd->err_info->CommandStatus, DRIVER_OK,
2684                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2685                 break;
2686         case CMD_HARDWARE_ERR:
2687                 printk(KERN_WARNING "cciss: cmd %p had "
2688                        " hardware error\n", cmd);
2689                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2690                         cmd->err_info->CommandStatus, DRIVER_OK,
2691                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2692                 break;
2693         case CMD_CONNECTION_LOST:
2694                 printk(KERN_WARNING "cciss: cmd %p had "
2695                        "connection lost\n", cmd);
2696                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2697                         cmd->err_info->CommandStatus, DRIVER_OK,
2698                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2699                 break;
2700         case CMD_ABORTED:
2701                 printk(KERN_WARNING "cciss: cmd %p was "
2702                        "aborted\n", cmd);
2703                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2704                         cmd->err_info->CommandStatus, DRIVER_OK,
2705                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
2706                 break;
2707         case CMD_ABORT_FAILED:
2708                 printk(KERN_WARNING "cciss: cmd %p reports "
2709                        "abort failed\n", cmd);
2710                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2711                         cmd->err_info->CommandStatus, DRIVER_OK,
2712                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2713                 break;
2714         case CMD_UNSOLICITED_ABORT:
2715                 printk(KERN_WARNING "cciss%d: unsolicited "
2716                        "abort %p\n", h->ctlr, cmd);
2717                 if (cmd->retry_count < MAX_CMD_RETRIES) {
2718                         retry_cmd = 1;
2719                         printk(KERN_WARNING
2720                                "cciss%d: retrying %p\n", h->ctlr, cmd);
2721                         cmd->retry_count++;
2722                 } else
2723                         printk(KERN_WARNING
2724                                "cciss%d: %p retried too "
2725                                "many times\n", h->ctlr, cmd);
2726                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2727                         cmd->err_info->CommandStatus, DRIVER_OK,
2728                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
2729                 break;
2730         case CMD_TIMEOUT:
2731                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
2732                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2733                         cmd->err_info->CommandStatus, DRIVER_OK,
2734                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2735                 break;
2736         default:
2737                 printk(KERN_WARNING "cciss: cmd %p returned "
2738                        "unknown status %x\n", cmd,
2739                        cmd->err_info->CommandStatus);
2740                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2741                         cmd->err_info->CommandStatus, DRIVER_OK,
2742                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2743         }
2744
2745 after_error_processing:
2746
2747         /* We need to return this command */
2748         if (retry_cmd) {
2749                 resend_cciss_cmd(h, cmd);
2750                 return;
2751         }
2752         cmd->rq->completion_data = cmd;
2753         blk_complete_request(cmd->rq);
2754 }
2755
2756 /*
2757  * Get a request and submit it to the controller.
2758  */
2759 static void do_cciss_request(struct request_queue *q)
2760 {
2761         ctlr_info_t *h = q->queuedata;
2762         CommandList_struct *c;
2763         sector_t start_blk;
2764         int seg;
2765         struct request *creq;
2766         u64bit temp64;
2767         struct scatterlist tmp_sg[MAXSGENTRIES];
2768         drive_info_struct *drv;
2769         int i, dir;
2770
2771         /* We call start_io here in case there is a command waiting on the
2772          * queue that has not been sent.
2773          */
2774         if (blk_queue_plugged(q))
2775                 goto startio;
2776
2777       queue:
2778         creq = elv_next_request(q);
2779         if (!creq)
2780                 goto startio;
2781
2782         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
2783
2784         if ((c = cmd_alloc(h, 1)) == NULL)
2785                 goto full;
2786
2787         blkdev_dequeue_request(creq);
2788
2789         spin_unlock_irq(q->queue_lock);
2790
2791         c->cmd_type = CMD_RWREQ;
2792         c->rq = creq;
2793
2794         /* fill in the request */
2795         drv = creq->rq_disk->private_data;
2796         c->Header.ReplyQueue = 0;       // unused in simple mode
2797         /* got command from pool, so use the command block index instead */
2798         /* for direct lookups. */
2799         /* The first 2 bits are reserved for controller error reporting. */
2800         c->Header.Tag.lower = (c->cmdindex << 3);
2801         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
2802         c->Header.LUN.LogDev.VolId = drv->LunID;
2803         c->Header.LUN.LogDev.Mode = 1;
2804         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
2805         c->Request.Type.Type = TYPE_CMD;        // It is a command.
2806         c->Request.Type.Attribute = ATTR_SIMPLE;
2807         c->Request.Type.Direction =
2808             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
2809         c->Request.Timeout = 0; // Don't time out
2810         c->Request.CDB[0] =
2811             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
2812         start_blk = creq->sector;
2813 #ifdef CCISS_DEBUG
2814         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n", (int)creq->sector,
2815                (int)creq->nr_sectors);
2816 #endif                          /* CCISS_DEBUG */
2817
2818         sg_init_table(tmp_sg, MAXSGENTRIES);
2819         seg = blk_rq_map_sg(q, creq, tmp_sg);
2820
2821         /* get the DMA records for the setup */
2822         if (c->Request.Type.Direction == XFER_READ)
2823                 dir = PCI_DMA_FROMDEVICE;
2824         else
2825                 dir = PCI_DMA_TODEVICE;
2826
2827         for (i = 0; i < seg; i++) {
2828                 c->SG[i].Len = tmp_sg[i].length;
2829                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
2830                                                   tmp_sg[i].offset,
2831                                                   tmp_sg[i].length, dir);
2832                 c->SG[i].Addr.lower = temp64.val32.lower;
2833                 c->SG[i].Addr.upper = temp64.val32.upper;
2834                 c->SG[i].Ext = 0;       // we are not chaining
2835         }
2836         /* track how many SG entries we are using */
2837         if (seg > h->maxSG)
2838                 h->maxSG = seg;
2839
2840 #ifdef CCISS_DEBUG
2841         printk(KERN_DEBUG "cciss: Submitting %d sectors in %d segments\n",
2842                creq->nr_sectors, seg);
2843 #endif                          /* CCISS_DEBUG */
2844
2845         c->Header.SGList = c->Header.SGTotal = seg;
2846         if (likely(blk_fs_request(creq))) {
2847                 if(h->cciss_read == CCISS_READ_10) {
2848                         c->Request.CDB[1] = 0;
2849                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
2850                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
2851                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
2852                         c->Request.CDB[5] = start_blk & 0xff;
2853                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
2854                         c->Request.CDB[7] = (creq->nr_sectors >> 8) & 0xff;
2855                         c->Request.CDB[8] = creq->nr_sectors & 0xff;
2856                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
2857                 } else {
2858                         u32 upper32 = upper_32_bits(start_blk);
2859
2860                         c->Request.CDBLen = 16;
2861                         c->Request.CDB[1]= 0;
2862                         c->Request.CDB[2]= (upper32 >> 24) & 0xff;      //MSB
2863                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
2864                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
2865                         c->Request.CDB[5]= upper32 & 0xff;
2866                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
2867                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
2868                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
2869                         c->Request.CDB[9]= start_blk & 0xff;
2870                         c->Request.CDB[10]= (creq->nr_sectors >>  24) & 0xff;
2871                         c->Request.CDB[11]= (creq->nr_sectors >>  16) & 0xff;
2872                         c->Request.CDB[12]= (creq->nr_sectors >>  8) & 0xff;
2873                         c->Request.CDB[13]= creq->nr_sectors & 0xff;
2874                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
2875                 }
2876         } else if (blk_pc_request(creq)) {
2877                 c->Request.CDBLen = creq->cmd_len;
2878                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
2879         } else {
2880                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
2881                 BUG();
2882         }
2883
2884         spin_lock_irq(q->queue_lock);
2885
2886         addQ(&(h->reqQ), c);
2887         h->Qdepth++;
2888         if (h->Qdepth > h->maxQsinceinit)
2889                 h->maxQsinceinit = h->Qdepth;
2890
2891         goto queue;
2892 full:
2893         blk_stop_queue(q);
2894 startio:
2895         /* We will already have the driver lock here so not need
2896          * to lock it.
2897          */
2898         start_io(h);
2899 }
2900
2901 static inline unsigned long get_next_completion(ctlr_info_t *h)
2902 {
2903 #ifdef CONFIG_CISS_SCSI_TAPE
2904         /* Any rejects from sendcmd() lying around? Process them first */
2905         if (h->scsi_rejects.ncompletions == 0)
2906                 return h->access.command_completed(h);
2907         else {
2908                 struct sendcmd_reject_list *srl;
2909                 int n;
2910                 srl = &h->scsi_rejects;
2911                 n = --srl->ncompletions;
2912                 /* printk("cciss%d: processing saved reject\n", h->ctlr); */
2913                 printk("p");
2914                 return srl->complete[n];
2915         }
2916 #else
2917         return h->access.command_completed(h);
2918 #endif
2919 }
2920
2921 static inline int interrupt_pending(ctlr_info_t *h)
2922 {
2923 #ifdef CONFIG_CISS_SCSI_TAPE
2924         return (h->access.intr_pending(h)
2925                 || (h->scsi_rejects.ncompletions > 0));
2926 #else
2927         return h->access.intr_pending(h);
2928 #endif
2929 }
2930
2931 static inline long interrupt_not_for_us(ctlr_info_t *h)
2932 {
2933 #ifdef CONFIG_CISS_SCSI_TAPE
2934         return (((h->access.intr_pending(h) == 0) ||
2935                  (h->interrupts_enabled == 0))
2936                 && (h->scsi_rejects.ncompletions == 0));
2937 #else
2938         return (((h->access.intr_pending(h) == 0) ||
2939                  (h->interrupts_enabled == 0)));
2940 #endif
2941 }
2942
2943 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
2944 {
2945         ctlr_info_t *h = dev_id;
2946         CommandList_struct *c;
2947         unsigned long flags;
2948         __u32 a, a1, a2;
2949
2950         if (interrupt_not_for_us(h))
2951                 return IRQ_NONE;
2952         /*
2953          * If there are completed commands in the completion queue,
2954          * we had better do something about it.
2955          */
2956         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2957         while (interrupt_pending(h)) {
2958                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
2959                         a1 = a;
2960                         if ((a & 0x04)) {
2961                                 a2 = (a >> 3);
2962                                 if (a2 >= h->nr_cmds) {
2963                                         printk(KERN_WARNING
2964                                                "cciss: controller cciss%d failed, stopping.\n",
2965                                                h->ctlr);
2966                                         fail_all_cmds(h->ctlr);
2967                                         return IRQ_HANDLED;
2968                                 }
2969
2970                                 c = h->cmd_pool + a2;
2971                                 a = c->busaddr;
2972
2973                         } else {
2974                                 a &= ~3;
2975                                 if ((c = h->cmpQ) == NULL) {
2976                                         printk(KERN_WARNING
2977                                                "cciss: Completion of %08x ignored\n",
2978                                                a1);
2979                                         continue;
2980                                 }
2981                                 while (c->busaddr != a) {
2982                                         c = c->next;
2983                                         if (c == h->cmpQ)
2984                                                 break;
2985                                 }
2986                         }
2987                         /*
2988                          * If we've found the command, take it off the
2989                          * completion Q and free it
2990                          */
2991                         if (c->busaddr == a) {
2992                                 removeQ(&h->cmpQ, c);
2993                                 if (c->cmd_type == CMD_RWREQ) {
2994                                         complete_command(h, c, 0);
2995                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
2996                                         complete(c->waiting);
2997                                 }
2998 #                               ifdef CONFIG_CISS_SCSI_TAPE
2999                                 else if (c->cmd_type == CMD_SCSI)
3000                                         complete_scsi_command(c, 0, a1);
3001 #                               endif
3002                                 continue;
3003                         }
3004                 }
3005         }
3006
3007         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3008         return IRQ_HANDLED;
3009 }
3010
3011 /*
3012  *  We cannot read the structure directly, for portability we must use
3013  *   the io functions.
3014  *   This is for debug only.
3015  */
3016 #ifdef CCISS_DEBUG
3017 static void print_cfg_table(CfgTable_struct *tb)
3018 {
3019         int i;
3020         char temp_name[17];
3021
3022         printk("Controller Configuration information\n");
3023         printk("------------------------------------\n");
3024         for (i = 0; i < 4; i++)
3025                 temp_name[i] = readb(&(tb->Signature[i]));
3026         temp_name[4] = '\0';
3027         printk("   Signature = %s\n", temp_name);
3028         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3029         printk("   Transport methods supported = 0x%x\n",
3030                readl(&(tb->TransportSupport)));
3031         printk("   Transport methods active = 0x%x\n",
3032                readl(&(tb->TransportActive)));
3033         printk("   Requested transport Method = 0x%x\n",
3034                readl(&(tb->HostWrite.TransportRequest)));
3035         printk("   Coalesce Interrupt Delay = 0x%x\n",
3036                readl(&(tb->HostWrite.CoalIntDelay)));
3037         printk("   Coalesce Interrupt Count = 0x%x\n",
3038                readl(&(tb->HostWrite.CoalIntCount)));
3039         printk("   Max outstanding commands = 0x%d\n",
3040                readl(&(tb->CmdsOutMax)));
3041         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3042         for (i = 0; i < 16; i++)
3043                 temp_name[i] = readb(&(tb->ServerName[i]));
3044         temp_name[16] = '\0';
3045         printk("   Server Name = %s\n", temp_name);
3046         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3047 }
3048 #endif                          /* CCISS_DEBUG */
3049
3050 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3051 {
3052         int i, offset, mem_type, bar_type;
3053         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3054                 return 0;
3055         offset = 0;
3056         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3057                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3058                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3059                         offset += 4;
3060                 else {
3061                         mem_type = pci_resource_flags(pdev, i) &
3062                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3063                         switch (mem_type) {
3064                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3065                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3066                                 offset += 4;    /* 32 bit */
3067                                 break;
3068                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3069                                 offset += 8;
3070                                 break;
3071                         default:        /* reserved in PCI 2.2 */
3072                                 printk(KERN_WARNING
3073                                        "Base address is invalid\n");
3074                                 return -1;
3075                                 break;
3076                         }
3077                 }
3078                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3079                         return i + 1;
3080         }
3081         return -1;
3082 }
3083
3084 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3085  * controllers that are capable. If not, we use IO-APIC mode.
3086  */
3087
3088 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3089                                            struct pci_dev *pdev, __u32 board_id)
3090 {
3091 #ifdef CONFIG_PCI_MSI
3092         int err;
3093         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3094         {0, 2}, {0, 3}
3095         };
3096
3097         /* Some boards advertise MSI but don't really support it */
3098         if ((board_id == 0x40700E11) ||
3099             (board_id == 0x40800E11) ||
3100             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3101                 goto default_int_mode;
3102
3103         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3104                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3105                 if (!err) {
3106                         c->intr[0] = cciss_msix_entries[0].vector;
3107                         c->intr[1] = cciss_msix_entries[1].vector;
3108                         c->intr[2] = cciss_msix_entries[2].vector;
3109                         c->intr[3] = cciss_msix_entries[3].vector;
3110                         c->msix_vector = 1;
3111                         return;
3112                 }
3113                 if (err > 0) {
3114                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3115                                "available\n", err);
3116                         goto default_int_mode;
3117                 } else {
3118                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3119                                err);
3120                         goto default_int_mode;
3121                 }
3122         }
3123         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3124                 if (!pci_enable_msi(pdev)) {
3125                         c->msi_vector = 1;
3126                 } else {
3127                         printk(KERN_WARNING "cciss: MSI init failed\n");
3128                 }
3129         }
3130 default_int_mode:
3131 #endif                          /* CONFIG_PCI_MSI */
3132         /* if we get here we're going to use the default interrupt mode */
3133         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3134         return;
3135 }
3136
3137 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3138 {
3139         ushort subsystem_vendor_id, subsystem_device_id, command;
3140         __u32 board_id, scratchpad = 0;
3141         __u64 cfg_offset;
3142         __u32 cfg_base_addr;
3143         __u64 cfg_base_addr_index;
3144         int i, err;
3145
3146         /* check to see if controller has been disabled */
3147         /* BEFORE trying to enable it */
3148         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3149         if (!(command & 0x02)) {
3150                 printk(KERN_WARNING
3151                        "cciss: controller appears to be disabled\n");
3152                 return -ENODEV;
3153         }
3154
3155         err = pci_enable_device(pdev);
3156         if (err) {
3157                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3158                 return err;
3159         }
3160
3161         err = pci_request_regions(pdev, "cciss");
3162         if (err) {
3163                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3164                        "aborting\n");
3165                 return err;
3166         }
3167
3168         subsystem_vendor_id = pdev->subsystem_vendor;
3169         subsystem_device_id = pdev->subsystem_device;
3170         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3171                     subsystem_vendor_id);
3172
3173 #ifdef CCISS_DEBUG
3174         printk("command = %x\n", command);
3175         printk("irq = %x\n", pdev->irq);
3176         printk("board_id = %x\n", board_id);
3177 #endif                          /* CCISS_DEBUG */
3178
3179 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3180  * else we use the IO-APIC interrupt assigned to us by system ROM.
3181  */
3182         cciss_interrupt_mode(c, pdev, board_id);
3183
3184         /*
3185          * Memory base addr is first addr , the second points to the config
3186          *   table
3187          */
3188
3189         c->paddr = pci_resource_start(pdev, 0); /* addressing mode bits already removed */
3190 #ifdef CCISS_DEBUG
3191         printk("address 0 = %x\n", c->paddr);
3192 #endif                          /* CCISS_DEBUG */
3193         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3194
3195         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3196          * We poll for up to 120 secs, once per 100ms. */
3197         for (i = 0; i < 1200; i++) {
3198                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3199                 if (scratchpad == CCISS_FIRMWARE_READY)
3200                         break;
3201                 set_current_state(TASK_INTERRUPTIBLE);
3202                 schedule_timeout(HZ / 10);      /* wait 100ms */
3203         }
3204         if (scratchpad != CCISS_FIRMWARE_READY) {
3205                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3206                 err = -ENODEV;
3207                 goto err_out_free_res;
3208         }
3209
3210         /* get the address index number */
3211         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3212         cfg_base_addr &= (__u32) 0x0000ffff;
3213 #ifdef CCISS_DEBUG
3214         printk("cfg base address = %x\n", cfg_base_addr);
3215 #endif                          /* CCISS_DEBUG */
3216         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3217 #ifdef CCISS_DEBUG
3218         printk("cfg base address index = %x\n", cfg_base_addr_index);
3219 #endif                          /* CCISS_DEBUG */
3220         if (cfg_base_addr_index == -1) {
3221                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3222                 err = -ENODEV;
3223                 goto err_out_free_res;
3224         }
3225
3226         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3227 #ifdef CCISS_DEBUG
3228         printk("cfg offset = %x\n", cfg_offset);
3229 #endif                          /* CCISS_DEBUG */
3230         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3231                                                        cfg_base_addr_index) +
3232                                     cfg_offset, sizeof(CfgTable_struct));
3233         c->board_id = board_id;
3234
3235 #ifdef CCISS_DEBUG
3236         print_cfg_table(c->cfgtable);
3237 #endif                          /* CCISS_DEBUG */
3238
3239         /* Some controllers support Zero Memory Raid (ZMR).
3240          * When configured in ZMR mode the number of supported
3241          * commands drops to 64. So instead of just setting an
3242          * arbitrary value we make the driver a little smarter.
3243          * We read the config table to tell us how many commands
3244          * are supported on the controller then subtract 4 to
3245          * leave a little room for ioctl calls.
3246          */
3247         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3248         for (i = 0; i < ARRAY_SIZE(products); i++) {
3249                 if (board_id == products[i].board_id) {
3250                         c->product_name = products[i].product_name;
3251                         c->access = *(products[i].access);
3252                         c->nr_cmds = c->max_commands - 4;
3253                         break;
3254                 }
3255         }
3256         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3257             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3258             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3259             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3260                 printk("Does not appear to be a valid CISS config table\n");
3261                 err = -ENODEV;
3262                 goto err_out_free_res;
3263         }
3264         /* We didn't find the controller in our list. We know the
3265          * signature is valid. If it's an HP device let's try to
3266          * bind to the device and fire it up. Otherwise we bail.
3267          */
3268         if (i == ARRAY_SIZE(products)) {
3269                 if (subsystem_vendor_id == PCI_VENDOR_ID_HP) {
3270                         c->product_name = products[i-1].product_name;
3271                         c->access = *(products[i-1].access);
3272                         c->nr_cmds = c->max_commands - 4;
3273                         printk(KERN_WARNING "cciss: This is an unknown "
3274                                 "Smart Array controller.\n"
3275                                 "cciss: Please update to the latest driver "
3276                                 "available from www.hp.com.\n");
3277                 } else {
3278                         printk(KERN_WARNING "cciss: Sorry, I don't know how"
3279                                 " to access the Smart Array controller %08lx\n"
3280                                         , (unsigned long)board_id);
3281                         err = -ENODEV;
3282                         goto err_out_free_res;
3283                 }
3284         }
3285 #ifdef CONFIG_X86
3286         {
3287                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3288                 __u32 prefetch;
3289                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3290                 prefetch |= 0x100;
3291                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3292         }
3293 #endif
3294
3295         /* Disabling DMA prefetch and refetch for the P600.
3296          * An ASIC bug may result in accesses to invalid memory addresses.
3297          * We've disabled prefetch for some time now. Testing with XEN
3298          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3299          */
3300         if(board_id == 0x3225103C) {
3301                 __u32 dma_prefetch;
3302                 __u32 dma_refetch;
3303                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3304                 dma_prefetch |= 0x8000;
3305                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3306                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3307                 dma_refetch |= 0x1;
3308                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3309         }
3310
3311 #ifdef CCISS_DEBUG
3312         printk("Trying to put board into Simple mode\n");
3313 #endif                          /* CCISS_DEBUG */
3314         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3315         /* Update the field, and then ring the doorbell */
3316         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3317         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3318
3319         /* under certain very rare conditions, this can take awhile.
3320          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3321          * as we enter this code.) */
3322         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3323                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3324                         break;
3325                 /* delay and try again */
3326                 set_current_state(TASK_INTERRUPTIBLE);
3327                 schedule_timeout(10);
3328         }
3329
3330 #ifdef CCISS_DEBUG
3331         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3332                readl(c->vaddr + SA5_DOORBELL));
3333 #endif                          /* CCISS_DEBUG */
3334 #ifdef CCISS_DEBUG
3335         print_cfg_table(c->cfgtable);
3336 #endif                          /* CCISS_DEBUG */
3337
3338         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3339                 printk(KERN_WARNING "cciss: unable to get board into"
3340                        " simple mode\n");
3341                 err = -ENODEV;
3342                 goto err_out_free_res;
3343         }
3344         return 0;
3345
3346 err_out_free_res:
3347         /*
3348          * Deliberately omit pci_disable_device(): it does something nasty to
3349          * Smart Array controllers that pci_enable_device does not undo
3350          */
3351         pci_release_regions(pdev);
3352         return err;
3353 }
3354
3355 /* Function to find the first free pointer into our hba[] array
3356  * Returns -1 if no free entries are left.
3357  */
3358 static int alloc_cciss_hba(void)
3359 {
3360         int i;
3361
3362         for (i = 0; i < MAX_CTLR; i++) {
3363                 if (!hba[i]) {
3364                         ctlr_info_t *p;
3365
3366                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3367                         if (!p)
3368                                 goto Enomem;
3369                         hba[i] = p;
3370                         return i;
3371                 }
3372         }
3373         printk(KERN_WARNING "cciss: This driver supports a maximum"
3374                " of %d controllers.\n", MAX_CTLR);
3375         return -1;
3376 Enomem:
3377         printk(KERN_ERR "cciss: out of memory.\n");
3378         return -1;
3379 }
3380
3381 static void free_hba(int i)
3382 {
3383         ctlr_info_t *p = hba[i];
3384         int n;
3385
3386         hba[i] = NULL;
3387         for (n = 0; n < CISS_MAX_LUN; n++)
3388                 put_disk(p->gendisk[n]);
3389         kfree(p);
3390 }
3391
3392 /*
3393  *  This is it.  Find all the controllers and register them.  I really hate
3394  *  stealing all these major device numbers.
3395  *  returns the number of block devices registered.
3396  */
3397 static int __devinit cciss_init_one(struct pci_dev *pdev,
3398                                     const struct pci_device_id *ent)
3399 {
3400         int i;
3401         int j = 0;
3402         int rc;
3403         int dac;
3404
3405         i = alloc_cciss_hba();
3406         if (i < 0)
3407                 return -1;
3408
3409         hba[i]->busy_initializing = 1;
3410
3411         if (cciss_pci_init(hba[i], pdev) != 0)
3412                 goto clean1;
3413
3414         sprintf(hba[i]->devname, "cciss%d", i);
3415         hba[i]->ctlr = i;
3416         hba[i]->pdev = pdev;
3417
3418         /* configure PCI DMA stuff */
3419         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK))
3420                 dac = 1;
3421         else if (!pci_set_dma_mask(pdev, DMA_32BIT_MASK))
3422                 dac = 0;
3423         else {
3424                 printk(KERN_ERR "cciss: no suitable DMA available\n");
3425                 goto clean1;
3426         }
3427
3428         /*
3429          * register with the major number, or get a dynamic major number
3430          * by passing 0 as argument.  This is done for greater than
3431          * 8 controller support.
3432          */
3433         if (i < MAX_CTLR_ORIG)
3434                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
3435         rc = register_blkdev(hba[i]->major, hba[i]->devname);
3436         if (rc == -EBUSY || rc == -EINVAL) {
3437                 printk(KERN_ERR
3438                        "cciss:  Unable to get major number %d for %s "
3439                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
3440                 goto clean1;
3441         } else {
3442                 if (i >= MAX_CTLR_ORIG)
3443                         hba[i]->major = rc;
3444         }
3445
3446         /* make sure the board interrupts are off */
3447         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
3448         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
3449                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
3450                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
3451                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
3452                 goto clean2;
3453         }
3454
3455         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
3456                hba[i]->devname, pdev->device, pci_name(pdev),
3457                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
3458
3459         hba[i]->cmd_pool_bits =
3460             kmalloc(((hba[i]->nr_cmds + BITS_PER_LONG -
3461                       1) / BITS_PER_LONG) * sizeof(unsigned long), GFP_KERNEL);
3462         hba[i]->cmd_pool = (CommandList_struct *)
3463             pci_alloc_consistent(hba[i]->pdev,
3464                     hba[i]->nr_cmds * sizeof(CommandList_struct),
3465                     &(hba[i]->cmd_pool_dhandle));
3466         hba[i]->errinfo_pool = (ErrorInfo_struct *)
3467             pci_alloc_consistent(hba[i]->pdev,
3468                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3469                     &(hba[i]->errinfo_pool_dhandle));
3470         if ((hba[i]->cmd_pool_bits == NULL)
3471             || (hba[i]->cmd_pool == NULL)
3472             || (hba[i]->errinfo_pool == NULL)) {
3473                 printk(KERN_ERR "cciss: out of memory");
3474                 goto clean4;
3475         }
3476 #ifdef CONFIG_CISS_SCSI_TAPE
3477         hba[i]->scsi_rejects.complete =
3478             kmalloc(sizeof(hba[i]->scsi_rejects.complete[0]) *
3479                     (hba[i]->nr_cmds + 5), GFP_KERNEL);
3480         if (hba[i]->scsi_rejects.complete == NULL) {
3481                 printk(KERN_ERR "cciss: out of memory");
3482                 goto clean4;
3483         }
3484 #endif
3485         spin_lock_init(&hba[i]->lock);
3486
3487         /* Initialize the pdev driver private data.
3488            have it point to hba[i].  */
3489         pci_set_drvdata(pdev, hba[i]);
3490         /* command and error info recs zeroed out before
3491            they are used */
3492         memset(hba[i]->cmd_pool_bits, 0,
3493                ((hba[i]->nr_cmds + BITS_PER_LONG -
3494                  1) / BITS_PER_LONG) * sizeof(unsigned long));
3495
3496         hba[i]->num_luns = 0;
3497         hba[i]->highest_lun = -1;
3498         for (j = 0; j < CISS_MAX_LUN; j++) {
3499                 hba[i]->drv[j].raid_level = -1;
3500                 hba[i]->drv[j].queue = NULL;
3501                 hba[i]->gendisk[j] = NULL;
3502         }
3503
3504         cciss_scsi_setup(i);
3505
3506         /* Turn the interrupts on so we can service requests */
3507         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
3508
3509         cciss_procinit(i);
3510
3511         hba[i]->cciss_max_sectors = 2048;
3512
3513         hba[i]->busy_initializing = 0;
3514
3515         rebuild_lun_table(hba[i], 1);
3516         return 1;
3517
3518 clean4:
3519 #ifdef CONFIG_CISS_SCSI_TAPE
3520         kfree(hba[i]->scsi_rejects.complete);
3521 #endif
3522         kfree(hba[i]->cmd_pool_bits);
3523         if (hba[i]->cmd_pool)
3524                 pci_free_consistent(hba[i]->pdev,
3525                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
3526                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3527         if (hba[i]->errinfo_pool)
3528                 pci_free_consistent(hba[i]->pdev,
3529                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3530                                     hba[i]->errinfo_pool,
3531                                     hba[i]->errinfo_pool_dhandle);
3532         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
3533 clean2:
3534         unregister_blkdev(hba[i]->major, hba[i]->devname);
3535 clean1:
3536         hba[i]->busy_initializing = 0;
3537         /* cleanup any queues that may have been initialized */
3538         for (j=0; j <= hba[i]->highest_lun; j++){
3539                 drive_info_struct *drv = &(hba[i]->drv[j]);
3540                 if (drv->queue)
3541                         blk_cleanup_queue(drv->queue);
3542         }
3543         /*
3544          * Deliberately omit pci_disable_device(): it does something nasty to
3545          * Smart Array controllers that pci_enable_device does not undo
3546          */
3547         pci_release_regions(pdev);
3548         pci_set_drvdata(pdev, NULL);
3549         free_hba(i);
3550         return -1;
3551 }
3552
3553 static void cciss_shutdown(struct pci_dev *pdev)
3554 {
3555         ctlr_info_t *tmp_ptr;
3556         int i;
3557         char flush_buf[4];
3558         int return_code;
3559
3560         tmp_ptr = pci_get_drvdata(pdev);
3561         if (tmp_ptr == NULL)
3562                 return;
3563         i = tmp_ptr->ctlr;
3564         if (hba[i] == NULL)
3565                 return;
3566
3567         /* Turn board interrupts off  and send the flush cache command */
3568         /* sendcmd will turn off interrupt, and send the flush...
3569          * To write all data in the battery backed cache to disks */
3570         memset(flush_buf, 0, 4);
3571         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0, 0, 0, NULL,
3572                               TYPE_CMD);
3573         if (return_code == IO_OK) {
3574                 printk(KERN_INFO "Completed flushing cache on controller %d\n", i);
3575         } else {
3576                 printk(KERN_WARNING "Error flushing cache on controller %d\n", i);
3577         }
3578         free_irq(hba[i]->intr[2], hba[i]);
3579 }
3580
3581 static void __devexit cciss_remove_one(struct pci_dev *pdev)
3582 {
3583         ctlr_info_t *tmp_ptr;
3584         int i, j;
3585
3586         if (pci_get_drvdata(pdev) == NULL) {
3587                 printk(KERN_ERR "cciss: Unable to remove device \n");
3588                 return;
3589         }
3590         tmp_ptr = pci_get_drvdata(pdev);
3591         i = tmp_ptr->ctlr;
3592         if (hba[i] == NULL) {
3593                 printk(KERN_ERR "cciss: device appears to "
3594                        "already be removed \n");
3595                 return;
3596         }
3597
3598         remove_proc_entry(hba[i]->devname, proc_cciss);
3599         unregister_blkdev(hba[i]->major, hba[i]->devname);
3600
3601         /* remove it from the disk list */
3602         for (j = 0; j < CISS_MAX_LUN; j++) {
3603                 struct gendisk *disk = hba[i]->gendisk[j];
3604                 if (disk) {
3605                         struct request_queue *q = disk->queue;
3606
3607                         if (disk->flags & GENHD_FL_UP)
3608                                 del_gendisk(disk);
3609                         if (q)
3610                                 blk_cleanup_queue(q);
3611                 }
3612         }
3613
3614         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
3615
3616         cciss_shutdown(pdev);
3617
3618 #ifdef CONFIG_PCI_MSI
3619         if (hba[i]->msix_vector)
3620                 pci_disable_msix(hba[i]->pdev);
3621         else if (hba[i]->msi_vector)
3622                 pci_disable_msi(hba[i]->pdev);
3623 #endif                          /* CONFIG_PCI_MSI */
3624
3625         iounmap(hba[i]->vaddr);
3626
3627         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
3628                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3629         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
3630                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
3631         kfree(hba[i]->cmd_pool_bits);
3632 #ifdef CONFIG_CISS_SCSI_TAPE
3633         kfree(hba[i]->scsi_rejects.complete);
3634 #endif
3635         /*
3636          * Deliberately omit pci_disable_device(): it does something nasty to
3637          * Smart Array controllers that pci_enable_device does not undo
3638          */
3639         pci_release_regions(pdev);
3640         pci_set_drvdata(pdev, NULL);
3641         free_hba(i);
3642 }
3643
3644 static struct pci_driver cciss_pci_driver = {
3645         .name = "cciss",
3646         .probe = cciss_init_one,
3647         .remove = __devexit_p(cciss_remove_one),
3648         .id_table = cciss_pci_device_id,        /* id_table */
3649         .shutdown = cciss_shutdown,
3650 };
3651
3652 /*
3653  *  This is it.  Register the PCI driver information for the cards we control
3654  *  the OS will call our registered routines when it finds one of our cards.
3655  */
3656 static int __init cciss_init(void)
3657 {
3658         printk(KERN_INFO DRIVER_NAME "\n");
3659
3660         /* Register for our PCI devices */
3661         return pci_register_driver(&cciss_pci_driver);
3662 }
3663
3664 static void __exit cciss_cleanup(void)
3665 {
3666         int i;
3667
3668         pci_unregister_driver(&cciss_pci_driver);
3669         /* double check that all controller entrys have been removed */
3670         for (i = 0; i < MAX_CTLR; i++) {
3671                 if (hba[i] != NULL) {
3672                         printk(KERN_WARNING "cciss: had to remove"
3673                                " controller %d\n", i);
3674                         cciss_remove_one(hba[i]->pdev);
3675                 }
3676         }
3677         remove_proc_entry("driver/cciss", NULL);
3678 }
3679
3680 static void fail_all_cmds(unsigned long ctlr)
3681 {
3682         /* If we get here, the board is apparently dead. */
3683         ctlr_info_t *h = hba[ctlr];
3684         CommandList_struct *c;
3685         unsigned long flags;
3686
3687         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
3688         h->alive = 0;           /* the controller apparently died... */
3689
3690         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
3691
3692         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
3693
3694         /* move everything off the request queue onto the completed queue */
3695         while ((c = h->reqQ) != NULL) {
3696                 removeQ(&(h->reqQ), c);
3697                 h->Qdepth--;
3698                 addQ(&(h->cmpQ), c);
3699         }
3700
3701         /* Now, fail everything on the completed queue with a HW error */
3702         while ((c = h->cmpQ) != NULL) {
3703                 removeQ(&h->cmpQ, c);
3704                 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
3705                 if (c->cmd_type == CMD_RWREQ) {
3706                         complete_command(h, c, 0);
3707                 } else if (c->cmd_type == CMD_IOCTL_PEND)
3708                         complete(c->waiting);
3709 #ifdef CONFIG_CISS_SCSI_TAPE
3710                 else if (c->cmd_type == CMD_SCSI)
3711                         complete_scsi_command(c, 0, 0);
3712 #endif
3713         }
3714         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
3715         return;
3716 }
3717
3718 module_init(cciss_init);
3719 module_exit(cciss_cleanup);