drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/compat.h>
  43 #include <linux/mutex.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/io.h>
  46
  47 #include <linux/dma-mapping.h>
  48 #include <linux/blkdev.h>
  49 #include <linux/genhd.h>
  50 #include <linux/completion.h>
  51 #include <scsi/scsi.h>
  52 #include <scsi/sg.h>
  53 #include <scsi/scsi_ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/scatterlist.h>
  56 #include <linux/kthread.h>
  57
  58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  59 #define DRIVER_NAME "HP CISS Driver (v 3.6.20)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 20)
  61
  62 /* Embedded module documentation macros - see modules.h */
  63 MODULE_AUTHOR("Hewlett-Packard Company");
  64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  65 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
  66                         " SA6i P600 P800 P400 P400i E200 E200i E500 P700m"
  67                         " Smart Array G2 Series SAS/SATA Controllers");
  68 MODULE_VERSION("3.6.20");
  69 MODULE_LICENSE("GPL");
  70
  71 static int cciss_allow_hpsa;
  72 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  73 MODULE_PARM_DESC(cciss_allow_hpsa,
  74         "Prevent cciss driver from accessing hardware known to be "
  75         " supported by the hpsa driver");
  76
  77 #include "cciss_cmd.h"
  78 #include "cciss.h"
  79 #include <linux/cciss_ioctl.h>
  80
  81 /* define the PCI info for the cards we can control */
  82 static const struct pci_device_id cciss_pci_device_id[] = {
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  90         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  91         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 110         {0,}
 111 };
 112
 113 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 114
 115 /*  board_id = Subsystem Device ID & Vendor ID
 116  *  product = Marketing Name for the board
 117  *  access = Address of the struct of function pointers
 118  */
 119 static struct board_type products[] = {
 120         {0x40700E11, "Smart Array 5300", &SA5_access},
 121         {0x40800E11, "Smart Array 5i", &SA5B_access},
 122         {0x40820E11, "Smart Array 532", &SA5B_access},
 123         {0x40830E11, "Smart Array 5312", &SA5B_access},
 124         {0x409A0E11, "Smart Array 641", &SA5_access},
 125         {0x409B0E11, "Smart Array 642", &SA5_access},
 126         {0x409C0E11, "Smart Array 6400", &SA5_access},
 127         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 128         {0x40910E11, "Smart Array 6i", &SA5_access},
 129         {0x3225103C, "Smart Array P600", &SA5_access},
 130         {0x3235103C, "Smart Array P400i", &SA5_access},
 131         {0x3211103C, "Smart Array E200i", &SA5_access},
 132         {0x3212103C, "Smart Array E200", &SA5_access},
 133         {0x3213103C, "Smart Array E200i", &SA5_access},
 134         {0x3214103C, "Smart Array E200i", &SA5_access},
 135         {0x3215103C, "Smart Array E200i", &SA5_access},
 136         {0x3237103C, "Smart Array E500", &SA5_access},
 137 /* controllers below this line are also supported by the hpsa driver. */
 138 #define HPSA_BOUNDARY 0x3223103C
 139         {0x3223103C, "Smart Array P800", &SA5_access},
 140         {0x3234103C, "Smart Array P400", &SA5_access},
 141         {0x323D103C, "Smart Array P700m", &SA5_access},
 142         {0x3241103C, "Smart Array P212", &SA5_access},
 143         {0x3243103C, "Smart Array P410", &SA5_access},
 144         {0x3245103C, "Smart Array P410i", &SA5_access},
 145         {0x3247103C, "Smart Array P411", &SA5_access},
 146         {0x3249103C, "Smart Array P812", &SA5_access},
 147         {0x324A103C, "Smart Array P712m", &SA5_access},
 148         {0x324B103C, "Smart Array P711m", &SA5_access},
 149 };
 150
 151 /* How long to wait (in milliseconds) for board to go into simple mode */
 152 #define MAX_CONFIG_WAIT 30000
 153 #define MAX_IOCTL_CONFIG_WAIT 1000
 154
 155 /*define how many times we will try a command because of bus resets */
 156 #define MAX_CMD_RETRIES 3
 157
 158 #define MAX_CTLR        32
 159
 160 /* Originally cciss driver only supports 8 major numbers */
 161 #define MAX_CTLR_ORIG   8
 162
 163 static ctlr_info_t *hba[MAX_CTLR];
 164
 165 static struct task_struct *cciss_scan_thread;
 166 static DEFINE_MUTEX(scan_mutex);
 167 static LIST_HEAD(scan_q);
 168
 169 static void do_cciss_request(struct request_queue *q);
 170 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 171 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 172 static int cciss_open(struct block_device *bdev, fmode_t mode);
 173 static int cciss_release(struct gendisk *disk, fmode_t mode);
 174 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 175                        unsigned int cmd, unsigned long arg);
 176 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 177
 178 static int cciss_revalidate(struct gendisk *disk);
 179 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 180 static int deregister_disk(ctlr_info_t *h, int drv_index,
 181                            int clear_all, int via_ioctl);
 182
 183 static void cciss_read_capacity(int ctlr, int logvol,
 184                         sector_t *total_size, unsigned int *block_size);
 185 static void cciss_read_capacity_16(int ctlr, int logvol,
 186                         sector_t *total_size, unsigned int *block_size);
 187 static void cciss_geometry_inquiry(int ctlr, int logvol,
 188                         sector_t total_size,
 189                         unsigned int block_size, InquiryData_struct *inq_buff,
 190                                    drive_info_struct *drv);
 191 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 192                                            __u32);
 193 static void start_io(ctlr_info_t *h);
 194 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 195                         __u8 page_code, unsigned char scsi3addr[],
 196                         int cmd_type);
 197 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 198         int attempt_retry);
 199 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 200
 201 static int add_to_scan_list(struct ctlr_info *h);
 202 static int scan_thread(void *data);
 203 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 204 static void cciss_hba_release(struct device *dev);
 205 static void cciss_device_release(struct device *dev);
 206 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 207 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 208
 209 #ifdef CONFIG_PROC_FS
 210 static void cciss_procinit(int i);
 211 #else
 212 static void cciss_procinit(int i)
 213 {
 214 }
 215 #endif                          /* CONFIG_PROC_FS */
 216
 217 #ifdef CONFIG_COMPAT
 218 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 219                               unsigned, unsigned long);
 220 #endif
 221
 222 static const struct block_device_operations cciss_fops = {
 223         .owner = THIS_MODULE,
 224         .open = cciss_open,
 225         .release = cciss_release,
 226         .locked_ioctl = cciss_ioctl,
 227         .getgeo = cciss_getgeo,
 228 #ifdef CONFIG_COMPAT
 229         .compat_ioctl = cciss_compat_ioctl,
 230 #endif
 231         .revalidate_disk = cciss_revalidate,
 232 };
 233
 234 /*
 235  * Enqueuing and dequeuing functions for cmdlists.
 236  */
 237 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 238 {
 239         hlist_add_head(&c->list, list);
 240 }
 241
 242 static inline void removeQ(CommandList_struct *c)
 243 {
 244         /*
 245          * After kexec/dump some commands might still
 246          * be in flight, which the firmware will try
 247          * to complete. Resetting the firmware doesn't work
 248          * with old fw revisions, so we have to mark
 249          * them off as 'stale' to prevent the driver from
 250          * falling over.
 251          */
 252         if (WARN_ON(hlist_unhashed(&c->list))) {
 253                 c->cmd_type = CMD_MSG_STALE;
 254                 return;
 255         }
 256
 257         hlist_del_init(&c->list);
 258 }
 259
 260 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 261         CommandList_struct *c)
 262 {
 263         unsigned long flags;
 264         spin_lock_irqsave(&h->lock, flags);
 265         addQ(&h->reqQ, c);
 266         h->Qdepth++;
 267         start_io(h);
 268         spin_unlock_irqrestore(&h->lock, flags);
 269 }
 270
 271 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 272         int nr_cmds)
 273 {
 274         int i;
 275
 276         if (!cmd_sg_list)
 277                 return;
 278         for (i = 0; i < nr_cmds; i++) {
 279                 kfree(cmd_sg_list[i]);
 280                 cmd_sg_list[i] = NULL;
 281         }
 282         kfree(cmd_sg_list);
 283 }
 284
 285 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 286         ctlr_info_t *h, int chainsize, int nr_cmds)
 287 {
 288         int j;
 289         SGDescriptor_struct **cmd_sg_list;
 290
 291         if (chainsize <= 0)
 292                 return NULL;
 293
 294         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 295         if (!cmd_sg_list)
 296                 return NULL;
 297
 298         /* Build up chain blocks for each command */
 299         for (j = 0; j < nr_cmds; j++) {
 300                 /* Need a block of chainsized s/g elements. */
 301                 cmd_sg_list[j] = kmalloc((chainsize *
 302                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 303                 if (!cmd_sg_list[j]) {
 304                         dev_err(&h->pdev->dev, "Cannot get memory "
 305                                 "for s/g chains.\n");
 306                         goto clean;
 307                 }
 308         }
 309         return cmd_sg_list;
 310 clean:
 311         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 312         return NULL;
 313 }
 314
 315 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 316 {
 317         SGDescriptor_struct *chain_sg;
 318         u64bit temp64;
 319
 320         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 321                 return;
 322
 323         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 324         temp64.val32.lower = chain_sg->Addr.lower;
 325         temp64.val32.upper = chain_sg->Addr.upper;
 326         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 327 }
 328
 329 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 330         SGDescriptor_struct *chain_block, int len)
 331 {
 332         SGDescriptor_struct *chain_sg;
 333         u64bit temp64;
 334
 335         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 336         chain_sg->Ext = CCISS_SG_CHAIN;
 337         chain_sg->Len = len;
 338         temp64.val = pci_map_single(h->pdev, chain_block, len,
 339                                 PCI_DMA_TODEVICE);
 340         chain_sg->Addr.lower = temp64.val32.lower;
 341         chain_sg->Addr.upper = temp64.val32.upper;
 342 }
 343
 344 #include "cciss_scsi.c"         /* For SCSI tape support */
 345
 346 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 347         "UNKNOWN"
 348 };
 349 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
 350
 351 #ifdef CONFIG_PROC_FS
 352
 353 /*
 354  * Report information about this controller.
 355  */
 356 #define ENG_GIG 1000000000
 357 #define ENG_GIG_FACTOR (ENG_GIG/512)
 358 #define ENGAGE_SCSI     "engage scsi"
 359
 360 static struct proc_dir_entry *proc_cciss;
 361
 362 static void cciss_seq_show_header(struct seq_file *seq)
 363 {
 364         ctlr_info_t *h = seq->private;
 365
 366         seq_printf(seq, "%s: HP %s Controller\n"
 367                 "Board ID: 0x%08lx\n"
 368                 "Firmware Version: %c%c%c%c\n"
 369                 "IRQ: %d\n"
 370                 "Logical drives: %d\n"
 371                 "Current Q depth: %d\n"
 372                 "Current # commands on controller: %d\n"
 373                 "Max Q depth since init: %d\n"
 374                 "Max # commands on controller since init: %d\n"
 375                 "Max SG entries since init: %d\n",
 376                 h->devname,
 377                 h->product_name,
 378                 (unsigned long)h->board_id,
 379                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 380                 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
 381                 h->num_luns,
 382                 h->Qdepth, h->commands_outstanding,
 383                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 384
 385 #ifdef CONFIG_CISS_SCSI_TAPE
 386         cciss_seq_tape_report(seq, h->ctlr);
 387 #endif /* CONFIG_CISS_SCSI_TAPE */
 388 }
 389
 390 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 391 {
 392         ctlr_info_t *h = seq->private;
 393         unsigned ctlr = h->ctlr;
 394         unsigned long flags;
 395
 396         /* prevent displaying bogus info during configuration
 397          * or deconfiguration of a logical volume
 398          */
 399         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 400         if (h->busy_configuring) {
 401                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 402                 return ERR_PTR(-EBUSY);
 403         }
 404         h->busy_configuring = 1;
 405         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 406
 407         if (*pos == 0)
 408                 cciss_seq_show_header(seq);
 409
 410         return pos;
 411 }
 412
 413 static int cciss_seq_show(struct seq_file *seq, void *v)
 414 {
 415         sector_t vol_sz, vol_sz_frac;
 416         ctlr_info_t *h = seq->private;
 417         unsigned ctlr = h->ctlr;
 418         loff_t *pos = v;
 419         drive_info_struct *drv = h->drv[*pos];
 420
 421         if (*pos > h->highest_lun)
 422                 return 0;
 423
 424         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 425                 return 0;
 426
 427         if (drv->heads == 0)
 428                 return 0;
 429
 430         vol_sz = drv->nr_blocks;
 431         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 432         vol_sz_frac *= 100;
 433         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 434
 435         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 436                 drv->raid_level = RAID_UNKNOWN;
 437         seq_printf(seq, "cciss/c%dd%d:"
 438                         "\t%4u.%02uGB\tRAID %s\n",
 439                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 440                         raid_label[drv->raid_level]);
 441         return 0;
 442 }
 443
 444 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 445 {
 446         ctlr_info_t *h = seq->private;
 447
 448         if (*pos > h->highest_lun)
 449                 return NULL;
 450         *pos += 1;
 451
 452         return pos;
 453 }
 454
 455 static void cciss_seq_stop(struct seq_file *seq, void *v)
 456 {
 457         ctlr_info_t *h = seq->private;
 458
 459         /* Only reset h->busy_configuring if we succeeded in setting
 460          * it during cciss_seq_start. */
 461         if (v == ERR_PTR(-EBUSY))
 462                 return;
 463
 464         h->busy_configuring = 0;
 465 }
 466
 467 static const struct seq_operations cciss_seq_ops = {
 468         .start = cciss_seq_start,
 469         .show  = cciss_seq_show,
 470         .next  = cciss_seq_next,
 471         .stop  = cciss_seq_stop,
 472 };
 473
 474 static int cciss_seq_open(struct inode *inode, struct file *file)
 475 {
 476         int ret = seq_open(file, &cciss_seq_ops);
 477         struct seq_file *seq = file->private_data;
 478
 479         if (!ret)
 480                 seq->private = PDE(inode)->data;
 481
 482         return ret;
 483 }
 484
 485 static ssize_t
 486 cciss_proc_write(struct file *file, const char __user *buf,
 487                  size_t length, loff_t *ppos)
 488 {
 489         int err;
 490         char *buffer;
 491
 492 #ifndef CONFIG_CISS_SCSI_TAPE
 493         return -EINVAL;
 494 #endif
 495
 496         if (!buf || length > PAGE_SIZE - 1)
 497                 return -EINVAL;
 498
 499         buffer = (char *)__get_free_page(GFP_KERNEL);
 500         if (!buffer)
 501                 return -ENOMEM;
 502
 503         err = -EFAULT;
 504         if (copy_from_user(buffer, buf, length))
 505                 goto out;
 506         buffer[length] = '\0';
 507
 508 #ifdef CONFIG_CISS_SCSI_TAPE
 509         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 510                 struct seq_file *seq = file->private_data;
 511                 ctlr_info_t *h = seq->private;
 512
 513                 err = cciss_engage_scsi(h->ctlr);
 514                 if (err == 0)
 515                         err = length;
 516         } else
 517 #endif /* CONFIG_CISS_SCSI_TAPE */
 518                 err = -EINVAL;
 519         /* might be nice to have "disengage" too, but it's not
 520            safely possible. (only 1 module use count, lock issues.) */
 521
 522 out:
 523         free_page((unsigned long)buffer);
 524         return err;
 525 }
 526
 527 static const struct file_operations cciss_proc_fops = {
 528         .owner   = THIS_MODULE,
 529         .open    = cciss_seq_open,
 530         .read    = seq_read,
 531         .llseek  = seq_lseek,
 532         .release = seq_release,
 533         .write   = cciss_proc_write,
 534 };
 535
 536 static void __devinit cciss_procinit(int i)
 537 {
 538         struct proc_dir_entry *pde;
 539
 540         if (proc_cciss == NULL)
 541                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 542         if (!proc_cciss)
 543                 return;
 544         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 545                                         S_IROTH, proc_cciss,
 546                                         &cciss_proc_fops, hba[i]);
 547 }
 548 #endif                          /* CONFIG_PROC_FS */
 549
 550 #define MAX_PRODUCT_NAME_LEN 19
 551
 552 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 553 #define to_drv(n) container_of(n, drive_info_struct, dev)
 554
 555 static ssize_t host_store_rescan(struct device *dev,
 556                                  struct device_attribute *attr,
 557                                  const char *buf, size_t count)
 558 {
 559         struct ctlr_info *h = to_hba(dev);
 560
 561         add_to_scan_list(h);
 562         wake_up_process(cciss_scan_thread);
 563         wait_for_completion_interruptible(&h->scan_wait);
 564
 565         return count;
 566 }
 567 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 568
 569 static ssize_t dev_show_unique_id(struct device *dev,
 570                                  struct device_attribute *attr,
 571                                  char *buf)
 572 {
 573         drive_info_struct *drv = to_drv(dev);
 574         struct ctlr_info *h = to_hba(drv->dev.parent);
 575         __u8 sn[16];
 576         unsigned long flags;
 577         int ret = 0;
 578
 579         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 580         if (h->busy_configuring)
 581                 ret = -EBUSY;
 582         else
 583                 memcpy(sn, drv->serial_no, sizeof(sn));
 584         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 585
 586         if (ret)
 587                 return ret;
 588         else
 589                 return snprintf(buf, 16 * 2 + 2,
 590                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 591                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 592                                 sn[0], sn[1], sn[2], sn[3],
 593                                 sn[4], sn[5], sn[6], sn[7],
 594                                 sn[8], sn[9], sn[10], sn[11],
 595                                 sn[12], sn[13], sn[14], sn[15]);
 596 }
 597 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 598
 599 static ssize_t dev_show_vendor(struct device *dev,
 600                                struct device_attribute *attr,
 601                                char *buf)
 602 {
 603         drive_info_struct *drv = to_drv(dev);
 604         struct ctlr_info *h = to_hba(drv->dev.parent);
 605         char vendor[VENDOR_LEN + 1];
 606         unsigned long flags;
 607         int ret = 0;
 608
 609         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 610         if (h->busy_configuring)
 611                 ret = -EBUSY;
 612         else
 613                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 614         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 615
 616         if (ret)
 617                 return ret;
 618         else
 619                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 620 }
 621 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 622
 623 static ssize_t dev_show_model(struct device *dev,
 624                               struct device_attribute *attr,
 625                               char *buf)
 626 {
 627         drive_info_struct *drv = to_drv(dev);
 628         struct ctlr_info *h = to_hba(drv->dev.parent);
 629         char model[MODEL_LEN + 1];
 630         unsigned long flags;
 631         int ret = 0;
 632
 633         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 634         if (h->busy_configuring)
 635                 ret = -EBUSY;
 636         else
 637                 memcpy(model, drv->model, MODEL_LEN + 1);
 638         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 639
 640         if (ret)
 641                 return ret;
 642         else
 643                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 644 }
 645 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 646
 647 static ssize_t dev_show_rev(struct device *dev,
 648                             struct device_attribute *attr,
 649                             char *buf)
 650 {
 651         drive_info_struct *drv = to_drv(dev);
 652         struct ctlr_info *h = to_hba(drv->dev.parent);
 653         char rev[REV_LEN + 1];
 654         unsigned long flags;
 655         int ret = 0;
 656
 657         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 658         if (h->busy_configuring)
 659                 ret = -EBUSY;
 660         else
 661                 memcpy(rev, drv->rev, REV_LEN + 1);
 662         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 663
 664         if (ret)
 665                 return ret;
 666         else
 667                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 668 }
 669 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 670
 671 static ssize_t cciss_show_lunid(struct device *dev,
 672                                 struct device_attribute *attr, char *buf)
 673 {
 674         drive_info_struct *drv = to_drv(dev);
 675         struct ctlr_info *h = to_hba(drv->dev.parent);
 676         unsigned long flags;
 677         unsigned char lunid[8];
 678
 679         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 680         if (h->busy_configuring) {
 681                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 682                 return -EBUSY;
 683         }
 684         if (!drv->heads) {
 685                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 686                 return -ENOTTY;
 687         }
 688         memcpy(lunid, drv->LunID, sizeof(lunid));
 689         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 690         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 691                 lunid[0], lunid[1], lunid[2], lunid[3],
 692                 lunid[4], lunid[5], lunid[6], lunid[7]);
 693 }
 694 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 695
 696 static ssize_t cciss_show_raid_level(struct device *dev,
 697                                      struct device_attribute *attr, char *buf)
 698 {
 699         drive_info_struct *drv = to_drv(dev);
 700         struct ctlr_info *h = to_hba(drv->dev.parent);
 701         int raid;
 702         unsigned long flags;
 703
 704         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 705         if (h->busy_configuring) {
 706                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 707                 return -EBUSY;
 708         }
 709         raid = drv->raid_level;
 710         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 711         if (raid < 0 || raid > RAID_UNKNOWN)
 712                 raid = RAID_UNKNOWN;
 713
 714         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 715                         raid_label[raid]);
 716 }
 717 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 718
 719 static ssize_t cciss_show_usage_count(struct device *dev,
 720                                       struct device_attribute *attr, char *buf)
 721 {
 722         drive_info_struct *drv = to_drv(dev);
 723         struct ctlr_info *h = to_hba(drv->dev.parent);
 724         unsigned long flags;
 725         int count;
 726
 727         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 728         if (h->busy_configuring) {
 729                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 730                 return -EBUSY;
 731         }
 732         count = drv->usage_count;
 733         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 734         return snprintf(buf, 20, "%d\n", count);
 735 }
 736 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 737
 738 static struct attribute *cciss_host_attrs[] = {
 739         &dev_attr_rescan.attr,
 740         NULL
 741 };
 742
 743 static struct attribute_group cciss_host_attr_group = {
 744         .attrs = cciss_host_attrs,
 745 };
 746
 747 static const struct attribute_group *cciss_host_attr_groups[] = {
 748         &cciss_host_attr_group,
 749         NULL
 750 };
 751
 752 static struct device_type cciss_host_type = {
 753         .name           = "cciss_host",
 754         .groups         = cciss_host_attr_groups,
 755         .release        = cciss_hba_release,
 756 };
 757
 758 static struct attribute *cciss_dev_attrs[] = {
 759         &dev_attr_unique_id.attr,
 760         &dev_attr_model.attr,
 761         &dev_attr_vendor.attr,
 762         &dev_attr_rev.attr,
 763         &dev_attr_lunid.attr,
 764         &dev_attr_raid_level.attr,
 765         &dev_attr_usage_count.attr,
 766         NULL
 767 };
 768
 769 static struct attribute_group cciss_dev_attr_group = {
 770         .attrs = cciss_dev_attrs,
 771 };
 772
 773 static const struct attribute_group *cciss_dev_attr_groups[] = {
 774         &cciss_dev_attr_group,
 775         NULL
 776 };
 777
 778 static struct device_type cciss_dev_type = {
 779         .name           = "cciss_device",
 780         .groups         = cciss_dev_attr_groups,
 781         .release        = cciss_device_release,
 782 };
 783
 784 static struct bus_type cciss_bus_type = {
 785         .name           = "cciss",
 786 };
 787
 788 /*
 789  * cciss_hba_release is called when the reference count
 790  * of h->dev goes to zero.
 791  */
 792 static void cciss_hba_release(struct device *dev)
 793 {
 794         /*
 795          * nothing to do, but need this to avoid a warning
 796          * about not having a release handler from lib/kref.c.
 797          */
 798 }
 799
 800 /*
 801  * Initialize sysfs entry for each controller.  This sets up and registers
 802  * the 'cciss#' directory for each individual controller under
 803  * /sys/bus/pci/devices/<dev>/.
 804  */
 805 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 806 {
 807         device_initialize(&h->dev);
 808         h->dev.type = &cciss_host_type;
 809         h->dev.bus = &cciss_bus_type;
 810         dev_set_name(&h->dev, "%s", h->devname);
 811         h->dev.parent = &h->pdev->dev;
 812
 813         return device_add(&h->dev);
 814 }
 815
 816 /*
 817  * Remove sysfs entries for an hba.
 818  */
 819 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 820 {
 821         device_del(&h->dev);
 822         put_device(&h->dev); /* final put. */
 823 }
 824
 825 /* cciss_device_release is called when the reference count
 826  * of h->drv[x]dev goes to zero.
 827  */
 828 static void cciss_device_release(struct device *dev)
 829 {
 830         drive_info_struct *drv = to_drv(dev);
 831         kfree(drv);
 832 }
 833
 834 /*
 835  * Initialize sysfs for each logical drive.  This sets up and registers
 836  * the 'c#d#' directory for each individual logical drive under
 837  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 838  * /sys/block/cciss!c#d# to this entry.
 839  */
 840 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 841                                        int drv_index)
 842 {
 843         struct device *dev;
 844
 845         if (h->drv[drv_index]->device_initialized)
 846                 return 0;
 847
 848         dev = &h->drv[drv_index]->dev;
 849         device_initialize(dev);
 850         dev->type = &cciss_dev_type;
 851         dev->bus = &cciss_bus_type;
 852         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 853         dev->parent = &h->dev;
 854         h->drv[drv_index]->device_initialized = 1;
 855         return device_add(dev);
 856 }
 857
 858 /*
 859  * Remove sysfs entries for a logical drive.
 860  */
 861 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 862         int ctlr_exiting)
 863 {
 864         struct device *dev = &h->drv[drv_index]->dev;
 865
 866         /* special case for c*d0, we only destroy it on controller exit */
 867         if (drv_index == 0 && !ctlr_exiting)
 868                 return;
 869
 870         device_del(dev);
 871         put_device(dev); /* the "final" put. */
 872         h->drv[drv_index] = NULL;
 873 }
 874
 875 /*
 876  * For operations that cannot sleep, a command block is allocated at init,
 877  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 878  * which ones are free or in use.  For operations that can wait for kmalloc
 879  * to possible sleep, this routine can be called with get_from_pool set to 0.
 880  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 881  */
 882 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 883 {
 884         CommandList_struct *c;
 885         int i;
 886         u64bit temp64;
 887         dma_addr_t cmd_dma_handle, err_dma_handle;
 888
 889         if (!get_from_pool) {
 890                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 891                         sizeof(CommandList_struct), &cmd_dma_handle);
 892                 if (c == NULL)
 893                         return NULL;
 894                 memset(c, 0, sizeof(CommandList_struct));
 895
 896                 c->cmdindex = -1;
 897
 898                 c->err_info = (ErrorInfo_struct *)
 899                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 900                             &err_dma_handle);
 901
 902                 if (c->err_info == NULL) {
 903                         pci_free_consistent(h->pdev,
 904                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 905                         return NULL;
 906                 }
 907                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 908         } else {                /* get it out of the controllers pool */
 909
 910                 do {
 911                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 912                         if (i == h->nr_cmds)
 913                                 return NULL;
 914                 } while (test_and_set_bit
 915                          (i & (BITS_PER_LONG - 1),
 916                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 917 #ifdef CCISS_DEBUG
 918                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 919 #endif
 920                 c = h->cmd_pool + i;
 921                 memset(c, 0, sizeof(CommandList_struct));
 922                 cmd_dma_handle = h->cmd_pool_dhandle
 923                     + i * sizeof(CommandList_struct);
 924                 c->err_info = h->errinfo_pool + i;
 925                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 926                 err_dma_handle = h->errinfo_pool_dhandle
 927                     + i * sizeof(ErrorInfo_struct);
 928                 h->nr_allocs++;
 929
 930                 c->cmdindex = i;
 931         }
 932
 933         INIT_HLIST_NODE(&c->list);
 934         c->busaddr = (__u32) cmd_dma_handle;
 935         temp64.val = (__u64) err_dma_handle;
 936         c->ErrDesc.Addr.lower = temp64.val32.lower;
 937         c->ErrDesc.Addr.upper = temp64.val32.upper;
 938         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 939
 940         c->ctlr = h->ctlr;
 941         return c;
 942 }
 943
 944 /*
 945  * Frees a command block that was previously allocated with cmd_alloc().
 946  */
 947 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 948 {
 949         int i;
 950         u64bit temp64;
 951
 952         if (!got_from_pool) {
 953                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 954                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 955                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 956                                     c->err_info, (dma_addr_t) temp64.val);
 957                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 958                                     c, (dma_addr_t) c->busaddr);
 959         } else {
 960                 i = c - h->cmd_pool;
 961                 clear_bit(i & (BITS_PER_LONG - 1),
 962                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 963                 h->nr_frees++;
 964         }
 965 }
 966
 967 static inline ctlr_info_t *get_host(struct gendisk *disk)
 968 {
 969         return disk->queue->queuedata;
 970 }
 971
 972 static inline drive_info_struct *get_drv(struct gendisk *disk)
 973 {
 974         return disk->private_data;
 975 }
 976
 977 /*
 978  * Open.  Make sure the device is really there.
 979  */
 980 static int cciss_open(struct block_device *bdev, fmode_t mode)
 981 {
 982         ctlr_info_t *host = get_host(bdev->bd_disk);
 983         drive_info_struct *drv = get_drv(bdev->bd_disk);
 984
 985 #ifdef CCISS_DEBUG
 986         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
 987 #endif                          /* CCISS_DEBUG */
 988
 989         if (drv->busy_configuring)
 990                 return -EBUSY;
 991         /*
 992          * Root is allowed to open raw volume zero even if it's not configured
 993          * so array config can still work. Root is also allowed to open any
 994          * volume that has a LUN ID, so it can issue IOCTL to reread the
 995          * disk information.  I don't think I really like this
 996          * but I'm already using way to many device nodes to claim another one
 997          * for "raw controller".
 998          */
 999         if (drv->heads == 0) {
1000                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1001                         /* if not node 0 make sure it is a partition = 0 */
1002                         if (MINOR(bdev->bd_dev) & 0x0f) {
1003                                 return -ENXIO;
1004                                 /* if it is, make sure we have a LUN ID */
1005                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1006                                 sizeof(drv->LunID))) {
1007                                 return -ENXIO;
1008                         }
1009                 }
1010                 if (!capable(CAP_SYS_ADMIN))
1011                         return -EPERM;
1012         }
1013         drv->usage_count++;
1014         host->usage_count++;
1015         return 0;
1016 }
1017
1018 /*
1019  * Close.  Sync first.
1020  */
1021 static int cciss_release(struct gendisk *disk, fmode_t mode)
1022 {
1023         ctlr_info_t *host = get_host(disk);
1024         drive_info_struct *drv = get_drv(disk);
1025
1026 #ifdef CCISS_DEBUG
1027         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
1028 #endif                          /* CCISS_DEBUG */
1029
1030         drv->usage_count--;
1031         host->usage_count--;
1032         return 0;
1033 }
1034
1035 #ifdef CONFIG_COMPAT
1036
1037 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1038                     unsigned cmd, unsigned long arg)
1039 {
1040         int ret;
1041         lock_kernel();
1042         ret = cciss_ioctl(bdev, mode, cmd, arg);
1043         unlock_kernel();
1044         return ret;
1045 }
1046
1047 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1048                                   unsigned cmd, unsigned long arg);
1049 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1050                                       unsigned cmd, unsigned long arg);
1051
1052 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1053                               unsigned cmd, unsigned long arg)
1054 {
1055         switch (cmd) {
1056         case CCISS_GETPCIINFO:
1057         case CCISS_GETINTINFO:
1058         case CCISS_SETINTINFO:
1059         case CCISS_GETNODENAME:
1060         case CCISS_SETNODENAME:
1061         case CCISS_GETHEARTBEAT:
1062         case CCISS_GETBUSTYPES:
1063         case CCISS_GETFIRMVER:
1064         case CCISS_GETDRIVVER:
1065         case CCISS_REVALIDVOLS:
1066         case CCISS_DEREGDISK:
1067         case CCISS_REGNEWDISK:
1068         case CCISS_REGNEWD:
1069         case CCISS_RESCANDISK:
1070         case CCISS_GETLUNINFO:
1071                 return do_ioctl(bdev, mode, cmd, arg);
1072
1073         case CCISS_PASSTHRU32:
1074                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1075         case CCISS_BIG_PASSTHRU32:
1076                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1077
1078         default:
1079                 return -ENOIOCTLCMD;
1080         }
1081 }
1082
1083 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1084                                   unsigned cmd, unsigned long arg)
1085 {
1086         IOCTL32_Command_struct __user *arg32 =
1087             (IOCTL32_Command_struct __user *) arg;
1088         IOCTL_Command_struct arg64;
1089         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1090         int err;
1091         u32 cp;
1092
1093         err = 0;
1094         err |=
1095             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1096                            sizeof(arg64.LUN_info));
1097         err |=
1098             copy_from_user(&arg64.Request, &arg32->Request,
1099                            sizeof(arg64.Request));
1100         err |=
1101             copy_from_user(&arg64.error_info, &arg32->error_info,
1102                            sizeof(arg64.error_info));
1103         err |= get_user(arg64.buf_size, &arg32->buf_size);
1104         err |= get_user(cp, &arg32->buf);
1105         arg64.buf = compat_ptr(cp);
1106         err |= copy_to_user(p, &arg64, sizeof(arg64));
1107
1108         if (err)
1109                 return -EFAULT;
1110
1111         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1112         if (err)
1113                 return err;
1114         err |=
1115             copy_in_user(&arg32->error_info, &p->error_info,
1116                          sizeof(arg32->error_info));
1117         if (err)
1118                 return -EFAULT;
1119         return err;
1120 }
1121
1122 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1123                                       unsigned cmd, unsigned long arg)
1124 {
1125         BIG_IOCTL32_Command_struct __user *arg32 =
1126             (BIG_IOCTL32_Command_struct __user *) arg;
1127         BIG_IOCTL_Command_struct arg64;
1128         BIG_IOCTL_Command_struct __user *p =
1129             compat_alloc_user_space(sizeof(arg64));
1130         int err;
1131         u32 cp;
1132
1133         err = 0;
1134         err |=
1135             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1136                            sizeof(arg64.LUN_info));
1137         err |=
1138             copy_from_user(&arg64.Request, &arg32->Request,
1139                            sizeof(arg64.Request));
1140         err |=
1141             copy_from_user(&arg64.error_info, &arg32->error_info,
1142                            sizeof(arg64.error_info));
1143         err |= get_user(arg64.buf_size, &arg32->buf_size);
1144         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1145         err |= get_user(cp, &arg32->buf);
1146         arg64.buf = compat_ptr(cp);
1147         err |= copy_to_user(p, &arg64, sizeof(arg64));
1148
1149         if (err)
1150                 return -EFAULT;
1151
1152         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1153         if (err)
1154                 return err;
1155         err |=
1156             copy_in_user(&arg32->error_info, &p->error_info,
1157                          sizeof(arg32->error_info));
1158         if (err)
1159                 return -EFAULT;
1160         return err;
1161 }
1162 #endif
1163
1164 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1165 {
1166         drive_info_struct *drv = get_drv(bdev->bd_disk);
1167
1168         if (!drv->cylinders)
1169                 return -ENXIO;
1170
1171         geo->heads = drv->heads;
1172         geo->sectors = drv->sectors;
1173         geo->cylinders = drv->cylinders;
1174         return 0;
1175 }
1176
1177 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1178 {
1179         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1180                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1181                 (void)check_for_unit_attention(host, c);
1182 }
1183 /*
1184  * ioctl
1185  */
1186 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1187                        unsigned int cmd, unsigned long arg)
1188 {
1189         struct gendisk *disk = bdev->bd_disk;
1190         ctlr_info_t *host = get_host(disk);
1191         drive_info_struct *drv = get_drv(disk);
1192         int ctlr = host->ctlr;
1193         void __user *argp = (void __user *)arg;
1194
1195 #ifdef CCISS_DEBUG
1196         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1197 #endif                          /* CCISS_DEBUG */
1198
1199         switch (cmd) {
1200         case CCISS_GETPCIINFO:
1201                 {
1202                         cciss_pci_info_struct pciinfo;
1203
1204                         if (!arg)
1205                                 return -EINVAL;
1206                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1207                         pciinfo.bus = host->pdev->bus->number;
1208                         pciinfo.dev_fn = host->pdev->devfn;
1209                         pciinfo.board_id = host->board_id;
1210                         if (copy_to_user
1211                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1212                                 return -EFAULT;
1213                         return 0;
1214                 }
1215         case CCISS_GETINTINFO:
1216                 {
1217                         cciss_coalint_struct intinfo;
1218                         if (!arg)
1219                                 return -EINVAL;
1220                         intinfo.delay =
1221                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1222                         intinfo.count =
1223                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1224                         if (copy_to_user
1225                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1226                                 return -EFAULT;
1227                         return 0;
1228                 }
1229         case CCISS_SETINTINFO:
1230                 {
1231                         cciss_coalint_struct intinfo;
1232                         unsigned long flags;
1233                         int i;
1234
1235                         if (!arg)
1236                                 return -EINVAL;
1237                         if (!capable(CAP_SYS_ADMIN))
1238                                 return -EPERM;
1239                         if (copy_from_user
1240                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1241                                 return -EFAULT;
1242                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1243                         {
1244 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1245                                 return -EINVAL;
1246                         }
1247                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1248                         /* Update the field, and then ring the doorbell */
1249                         writel(intinfo.delay,
1250                                &(host->cfgtable->HostWrite.CoalIntDelay));
1251                         writel(intinfo.count,
1252                                &(host->cfgtable->HostWrite.CoalIntCount));
1253                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1254
1255                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1256                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1257                                       & CFGTBL_ChangeReq))
1258                                         break;
1259                                 /* delay and try again */
1260                                 udelay(1000);
1261                         }
1262                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1263                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1264                                 return -EAGAIN;
1265                         return 0;
1266                 }
1267         case CCISS_GETNODENAME:
1268                 {
1269                         NodeName_type NodeName;
1270                         int i;
1271
1272                         if (!arg)
1273                                 return -EINVAL;
1274                         for (i = 0; i < 16; i++)
1275                                 NodeName[i] =
1276                                     readb(&host->cfgtable->ServerName[i]);
1277                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1278                                 return -EFAULT;
1279                         return 0;
1280                 }
1281         case CCISS_SETNODENAME:
1282                 {
1283                         NodeName_type NodeName;
1284                         unsigned long flags;
1285                         int i;
1286
1287                         if (!arg)
1288                                 return -EINVAL;
1289                         if (!capable(CAP_SYS_ADMIN))
1290                                 return -EPERM;
1291
1292                         if (copy_from_user
1293                             (NodeName, argp, sizeof(NodeName_type)))
1294                                 return -EFAULT;
1295
1296                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1297
1298                         /* Update the field, and then ring the doorbell */
1299                         for (i = 0; i < 16; i++)
1300                                 writeb(NodeName[i],
1301                                        &host->cfgtable->ServerName[i]);
1302
1303                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1304
1305                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1306                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1307                                       & CFGTBL_ChangeReq))
1308                                         break;
1309                                 /* delay and try again */
1310                                 udelay(1000);
1311                         }
1312                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1313                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1314                                 return -EAGAIN;
1315                         return 0;
1316                 }
1317
1318         case CCISS_GETHEARTBEAT:
1319                 {
1320                         Heartbeat_type heartbeat;
1321
1322                         if (!arg)
1323                                 return -EINVAL;
1324                         heartbeat = readl(&host->cfgtable->HeartBeat);
1325                         if (copy_to_user
1326                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1327                                 return -EFAULT;
1328                         return 0;
1329                 }
1330         case CCISS_GETBUSTYPES:
1331                 {
1332                         BusTypes_type BusTypes;
1333
1334                         if (!arg)
1335                                 return -EINVAL;
1336                         BusTypes = readl(&host->cfgtable->BusTypes);
1337                         if (copy_to_user
1338                             (argp, &BusTypes, sizeof(BusTypes_type)))
1339                                 return -EFAULT;
1340                         return 0;
1341                 }
1342         case CCISS_GETFIRMVER:
1343                 {
1344                         FirmwareVer_type firmware;
1345
1346                         if (!arg)
1347                                 return -EINVAL;
1348                         memcpy(firmware, host->firm_ver, 4);
1349
1350                         if (copy_to_user
1351                             (argp, firmware, sizeof(FirmwareVer_type)))
1352                                 return -EFAULT;
1353                         return 0;
1354                 }
1355         case CCISS_GETDRIVVER:
1356                 {
1357                         DriverVer_type DriverVer = DRIVER_VERSION;
1358
1359                         if (!arg)
1360                                 return -EINVAL;
1361
1362                         if (copy_to_user
1363                             (argp, &DriverVer, sizeof(DriverVer_type)))
1364                                 return -EFAULT;
1365                         return 0;
1366                 }
1367
1368         case CCISS_DEREGDISK:
1369         case CCISS_REGNEWD:
1370         case CCISS_REVALIDVOLS:
1371                 return rebuild_lun_table(host, 0, 1);
1372
1373         case CCISS_GETLUNINFO:{
1374                         LogvolInfo_struct luninfo;
1375
1376                         memcpy(&luninfo.LunID, drv->LunID,
1377                                 sizeof(luninfo.LunID));
1378                         luninfo.num_opens = drv->usage_count;
1379                         luninfo.num_parts = 0;
1380                         if (copy_to_user(argp, &luninfo,
1381                                          sizeof(LogvolInfo_struct)))
1382                                 return -EFAULT;
1383                         return 0;
1384                 }
1385         case CCISS_PASSTHRU:
1386                 {
1387                         IOCTL_Command_struct iocommand;
1388                         CommandList_struct *c;
1389                         char *buff = NULL;
1390                         u64bit temp64;
1391                         DECLARE_COMPLETION_ONSTACK(wait);
1392
1393                         if (!arg)
1394                                 return -EINVAL;
1395
1396                         if (!capable(CAP_SYS_RAWIO))
1397                                 return -EPERM;
1398
1399                         if (copy_from_user
1400                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1401                                 return -EFAULT;
1402                         if ((iocommand.buf_size < 1) &&
1403                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1404                                 return -EINVAL;
1405                         }
1406 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1407                         /* Check kmalloc limits */
1408                         if (iocommand.buf_size > 128000)
1409                                 return -EINVAL;
1410 #endif
1411                         if (iocommand.buf_size > 0) {
1412                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1413                                 if (buff == NULL)
1414                                         return -EFAULT;
1415                         }
1416                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1417                                 /* Copy the data into the buffer we created */
1418                                 if (copy_from_user
1419                                     (buff, iocommand.buf, iocommand.buf_size)) {
1420                                         kfree(buff);
1421                                         return -EFAULT;
1422                                 }
1423                         } else {
1424                                 memset(buff, 0, iocommand.buf_size);
1425                         }
1426                         if ((c = cmd_alloc(host, 0)) == NULL) {
1427                                 kfree(buff);
1428                                 return -ENOMEM;
1429                         }
1430                         /* Fill in the command type */
1431                         c->cmd_type = CMD_IOCTL_PEND;
1432                         /* Fill in Command Header */
1433                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1434                         if (iocommand.buf_size > 0) /* buffer to fill */
1435                         {
1436                                 c->Header.SGList = 1;
1437                                 c->Header.SGTotal = 1;
1438                         } else /* no buffers to fill */
1439                         {
1440                                 c->Header.SGList = 0;
1441                                 c->Header.SGTotal = 0;
1442                         }
1443                         c->Header.LUN = iocommand.LUN_info;
1444                         /* use the kernel address the cmd block for tag */
1445                         c->Header.Tag.lower = c->busaddr;
1446
1447                         /* Fill in Request block */
1448                         c->Request = iocommand.Request;
1449
1450                         /* Fill in the scatter gather information */
1451                         if (iocommand.buf_size > 0) {
1452                                 temp64.val = pci_map_single(host->pdev, buff,
1453                                         iocommand.buf_size,
1454                                         PCI_DMA_BIDIRECTIONAL);
1455                                 c->SG[0].Addr.lower = temp64.val32.lower;
1456                                 c->SG[0].Addr.upper = temp64.val32.upper;
1457                                 c->SG[0].Len = iocommand.buf_size;
1458                                 c->SG[0].Ext = 0;  /* we are not chaining */
1459                         }
1460                         c->waiting = &wait;
1461
1462                         enqueue_cmd_and_start_io(host, c);
1463                         wait_for_completion(&wait);
1464
1465                         /* unlock the buffers from DMA */
1466                         temp64.val32.lower = c->SG[0].Addr.lower;
1467                         temp64.val32.upper = c->SG[0].Addr.upper;
1468                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1469                                          iocommand.buf_size,
1470                                          PCI_DMA_BIDIRECTIONAL);
1471
1472                         check_ioctl_unit_attention(host, c);
1473
1474                         /* Copy the error information out */
1475                         iocommand.error_info = *(c->err_info);
1476                         if (copy_to_user
1477                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1478                                 kfree(buff);
1479                                 cmd_free(host, c, 0);
1480                                 return -EFAULT;
1481                         }
1482
1483                         if (iocommand.Request.Type.Direction == XFER_READ) {
1484                                 /* Copy the data out of the buffer we created */
1485                                 if (copy_to_user
1486                                     (iocommand.buf, buff, iocommand.buf_size)) {
1487                                         kfree(buff);
1488                                         cmd_free(host, c, 0);
1489                                         return -EFAULT;
1490                                 }
1491                         }
1492                         kfree(buff);
1493                         cmd_free(host, c, 0);
1494                         return 0;
1495                 }
1496         case CCISS_BIG_PASSTHRU:{
1497                         BIG_IOCTL_Command_struct *ioc;
1498                         CommandList_struct *c;
1499                         unsigned char **buff = NULL;
1500                         int *buff_size = NULL;
1501                         u64bit temp64;
1502                         BYTE sg_used = 0;
1503                         int status = 0;
1504                         int i;
1505                         DECLARE_COMPLETION_ONSTACK(wait);
1506                         __u32 left;
1507                         __u32 sz;
1508                         BYTE __user *data_ptr;
1509
1510                         if (!arg)
1511                                 return -EINVAL;
1512                         if (!capable(CAP_SYS_RAWIO))
1513                                 return -EPERM;
1514                         ioc = (BIG_IOCTL_Command_struct *)
1515                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1516                         if (!ioc) {
1517                                 status = -ENOMEM;
1518                                 goto cleanup1;
1519                         }
1520                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1521                                 status = -EFAULT;
1522                                 goto cleanup1;
1523                         }
1524                         if ((ioc->buf_size < 1) &&
1525                             (ioc->Request.Type.Direction != XFER_NONE)) {
1526                                 status = -EINVAL;
1527                                 goto cleanup1;
1528                         }
1529                         /* Check kmalloc limits  using all SGs */
1530                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1531                                 status = -EINVAL;
1532                                 goto cleanup1;
1533                         }
1534                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1535                                 status = -EINVAL;
1536                                 goto cleanup1;
1537                         }
1538                         buff =
1539                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1540                         if (!buff) {
1541                                 status = -ENOMEM;
1542                                 goto cleanup1;
1543                         }
1544                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1545                                                    GFP_KERNEL);
1546                         if (!buff_size) {
1547                                 status = -ENOMEM;
1548                                 goto cleanup1;
1549                         }
1550                         left = ioc->buf_size;
1551                         data_ptr = ioc->buf;
1552                         while (left) {
1553                                 sz = (left >
1554                                       ioc->malloc_size) ? ioc->
1555                                     malloc_size : left;
1556                                 buff_size[sg_used] = sz;
1557                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1558                                 if (buff[sg_used] == NULL) {
1559                                         status = -ENOMEM;
1560                                         goto cleanup1;
1561                                 }
1562                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1563                                         if (copy_from_user
1564                                             (buff[sg_used], data_ptr, sz)) {
1565                                                 status = -EFAULT;
1566                                                 goto cleanup1;
1567                                         }
1568                                 } else {
1569                                         memset(buff[sg_used], 0, sz);
1570                                 }
1571                                 left -= sz;
1572                                 data_ptr += sz;
1573                                 sg_used++;
1574                         }
1575                         if ((c = cmd_alloc(host, 0)) == NULL) {
1576                                 status = -ENOMEM;
1577                                 goto cleanup1;
1578                         }
1579                         c->cmd_type = CMD_IOCTL_PEND;
1580                         c->Header.ReplyQueue = 0;
1581
1582                         if (ioc->buf_size > 0) {
1583                                 c->Header.SGList = sg_used;
1584                                 c->Header.SGTotal = sg_used;
1585                         } else {
1586                                 c->Header.SGList = 0;
1587                                 c->Header.SGTotal = 0;
1588                         }
1589                         c->Header.LUN = ioc->LUN_info;
1590                         c->Header.Tag.lower = c->busaddr;
1591
1592                         c->Request = ioc->Request;
1593                         if (ioc->buf_size > 0) {
1594                                 for (i = 0; i < sg_used; i++) {
1595                                         temp64.val =
1596                                             pci_map_single(host->pdev, buff[i],
1597                                                     buff_size[i],
1598                                                     PCI_DMA_BIDIRECTIONAL);
1599                                         c->SG[i].Addr.lower =
1600                                             temp64.val32.lower;
1601                                         c->SG[i].Addr.upper =
1602                                             temp64.val32.upper;
1603                                         c->SG[i].Len = buff_size[i];
1604                                         c->SG[i].Ext = 0;       /* we are not chaining */
1605                                 }
1606                         }
1607                         c->waiting = &wait;
1608                         enqueue_cmd_and_start_io(host, c);
1609                         wait_for_completion(&wait);
1610                         /* unlock the buffers from DMA */
1611                         for (i = 0; i < sg_used; i++) {
1612                                 temp64.val32.lower = c->SG[i].Addr.lower;
1613                                 temp64.val32.upper = c->SG[i].Addr.upper;
1614                                 pci_unmap_single(host->pdev,
1615                                         (dma_addr_t) temp64.val, buff_size[i],
1616                                         PCI_DMA_BIDIRECTIONAL);
1617                         }
1618                         check_ioctl_unit_attention(host, c);
1619                         /* Copy the error information out */
1620                         ioc->error_info = *(c->err_info);
1621                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1622                                 cmd_free(host, c, 0);
1623                                 status = -EFAULT;
1624                                 goto cleanup1;
1625                         }
1626                         if (ioc->Request.Type.Direction == XFER_READ) {
1627                                 /* Copy the data out of the buffer we created */
1628                                 BYTE __user *ptr = ioc->buf;
1629                                 for (i = 0; i < sg_used; i++) {
1630                                         if (copy_to_user
1631                                             (ptr, buff[i], buff_size[i])) {
1632                                                 cmd_free(host, c, 0);
1633                                                 status = -EFAULT;
1634                                                 goto cleanup1;
1635                                         }
1636                                         ptr += buff_size[i];
1637                                 }
1638                         }
1639                         cmd_free(host, c, 0);
1640                         status = 0;
1641                       cleanup1:
1642                         if (buff) {
1643                                 for (i = 0; i < sg_used; i++)
1644                                         kfree(buff[i]);
1645                                 kfree(buff);
1646                         }
1647                         kfree(buff_size);
1648                         kfree(ioc);
1649                         return status;
1650                 }
1651
1652         /* scsi_cmd_ioctl handles these, below, though some are not */
1653         /* very meaningful for cciss.  SG_IO is the main one people want. */
1654
1655         case SG_GET_VERSION_NUM:
1656         case SG_SET_TIMEOUT:
1657         case SG_GET_TIMEOUT:
1658         case SG_GET_RESERVED_SIZE:
1659         case SG_SET_RESERVED_SIZE:
1660         case SG_EMULATED_HOST:
1661         case SG_IO:
1662         case SCSI_IOCTL_SEND_COMMAND:
1663                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1664
1665         /* scsi_cmd_ioctl would normally handle these, below, but */
1666         /* they aren't a good fit for cciss, as CD-ROMs are */
1667         /* not supported, and we don't have any bus/target/lun */
1668         /* which we present to the kernel. */
1669
1670         case CDROM_SEND_PACKET:
1671         case CDROMCLOSETRAY:
1672         case CDROMEJECT:
1673         case SCSI_IOCTL_GET_IDLUN:
1674         case SCSI_IOCTL_GET_BUS_NUMBER:
1675         default:
1676                 return -ENOTTY;
1677         }
1678 }
1679
1680 static void cciss_check_queues(ctlr_info_t *h)
1681 {
1682         int start_queue = h->next_to_run;
1683         int i;
1684
1685         /* check to see if we have maxed out the number of commands that can
1686          * be placed on the queue.  If so then exit.  We do this check here
1687          * in case the interrupt we serviced was from an ioctl and did not
1688          * free any new commands.
1689          */
1690         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1691                 return;
1692
1693         /* We have room on the queue for more commands.  Now we need to queue
1694          * them up.  We will also keep track of the next queue to run so
1695          * that every queue gets a chance to be started first.
1696          */
1697         for (i = 0; i < h->highest_lun + 1; i++) {
1698                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1699                 /* make sure the disk has been added and the drive is real
1700                  * because this can be called from the middle of init_one.
1701                  */
1702                 if (!h->drv[curr_queue])
1703                         continue;
1704                 if (!(h->drv[curr_queue]->queue) ||
1705                         !(h->drv[curr_queue]->heads))
1706                         continue;
1707                 blk_start_queue(h->gendisk[curr_queue]->queue);
1708
1709                 /* check to see if we have maxed out the number of commands
1710                  * that can be placed on the queue.
1711                  */
1712                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1713                         if (curr_queue == start_queue) {
1714                                 h->next_to_run =
1715                                     (start_queue + 1) % (h->highest_lun + 1);
1716                                 break;
1717                         } else {
1718                                 h->next_to_run = curr_queue;
1719                                 break;
1720                         }
1721                 }
1722         }
1723 }
1724
1725 static void cciss_softirq_done(struct request *rq)
1726 {
1727         CommandList_struct *cmd = rq->completion_data;
1728         ctlr_info_t *h = hba[cmd->ctlr];
1729         SGDescriptor_struct *curr_sg = cmd->SG;
1730         u64bit temp64;
1731         unsigned long flags;
1732         int i, ddir;
1733         int sg_index = 0;
1734
1735         if (cmd->Request.Type.Direction == XFER_READ)
1736                 ddir = PCI_DMA_FROMDEVICE;
1737         else
1738                 ddir = PCI_DMA_TODEVICE;
1739
1740         /* command did not need to be retried */
1741         /* unmap the DMA mapping for all the scatter gather elements */
1742         for (i = 0; i < cmd->Header.SGList; i++) {
1743                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1744                         cciss_unmap_sg_chain_block(h, cmd);
1745                         /* Point to the next block */
1746                         curr_sg = h->cmd_sg_list[cmd->cmdindex];
1747                         sg_index = 0;
1748                 }
1749                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1750                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1751                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1752                                 ddir);
1753                 ++sg_index;
1754         }
1755
1756 #ifdef CCISS_DEBUG
1757         printk("Done with %p\n", rq);
1758 #endif                          /* CCISS_DEBUG */
1759
1760         /* set the residual count for pc requests */
1761         if (blk_pc_request(rq))
1762                 rq->resid_len = cmd->err_info->ResidualCnt;
1763
1764         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1765
1766         spin_lock_irqsave(&h->lock, flags);
1767         cmd_free(h, cmd, 1);
1768         cciss_check_queues(h);
1769         spin_unlock_irqrestore(&h->lock, flags);
1770 }
1771
1772 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1773         unsigned char scsi3addr[], uint32_t log_unit)
1774 {
1775         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1776                 sizeof(h->drv[log_unit]->LunID));
1777 }
1778
1779 /* This function gets the SCSI vendor, model, and revision of a logical drive
1780  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1781  * they cannot be read.
1782  */
1783 static void cciss_get_device_descr(int ctlr, int logvol,
1784                                    char *vendor, char *model, char *rev)
1785 {
1786         int rc;
1787         InquiryData_struct *inq_buf;
1788         unsigned char scsi3addr[8];
1789
1790         *vendor = '\0';
1791         *model = '\0';
1792         *rev = '\0';
1793
1794         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1795         if (!inq_buf)
1796                 return;
1797
1798         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1799         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1800                         scsi3addr, TYPE_CMD);
1801         if (rc == IO_OK) {
1802                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1803                 vendor[VENDOR_LEN] = '\0';
1804                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1805                 model[MODEL_LEN] = '\0';
1806                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1807                 rev[REV_LEN] = '\0';
1808         }
1809
1810         kfree(inq_buf);
1811         return;
1812 }
1813
1814 /* This function gets the serial number of a logical drive via
1815  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1816  * number cannot be had, for whatever reason, 16 bytes of 0xff
1817  * are returned instead.
1818  */
1819 static void cciss_get_serial_no(int ctlr, int logvol,
1820                                 unsigned char *serial_no, int buflen)
1821 {
1822 #define PAGE_83_INQ_BYTES 64
1823         int rc;
1824         unsigned char *buf;
1825         unsigned char scsi3addr[8];
1826
1827         if (buflen > 16)
1828                 buflen = 16;
1829         memset(serial_no, 0xff, buflen);
1830         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1831         if (!buf)
1832                 return;
1833         memset(serial_no, 0, buflen);
1834         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1835         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1836                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1837         if (rc == IO_OK)
1838                 memcpy(serial_no, &buf[8], buflen);
1839         kfree(buf);
1840         return;
1841 }
1842
1843 /*
1844  * cciss_add_disk sets up the block device queue for a logical drive
1845  */
1846 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1847                                 int drv_index)
1848 {
1849         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1850         if (!disk->queue)
1851                 goto init_queue_failure;
1852         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1853         disk->major = h->major;
1854         disk->first_minor = drv_index << NWD_SHIFT;
1855         disk->fops = &cciss_fops;
1856         if (cciss_create_ld_sysfs_entry(h, drv_index))
1857                 goto cleanup_queue;
1858         disk->private_data = h->drv[drv_index];
1859         disk->driverfs_dev = &h->drv[drv_index]->dev;
1860
1861         /* Set up queue information */
1862         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1863
1864         /* This is a hardware imposed limit. */
1865         blk_queue_max_segments(disk->queue, h->maxsgentries);
1866
1867         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1868
1869         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1870
1871         disk->queue->queuedata = h;
1872
1873         blk_queue_logical_block_size(disk->queue,
1874                                      h->drv[drv_index]->block_size);
1875
1876         /* Make sure all queue data is written out before */
1877         /* setting h->drv[drv_index]->queue, as setting this */
1878         /* allows the interrupt handler to start the queue */
1879         wmb();
1880         h->drv[drv_index]->queue = disk->queue;
1881         add_disk(disk);
1882         return 0;
1883
1884 cleanup_queue:
1885         blk_cleanup_queue(disk->queue);
1886         disk->queue = NULL;
1887 init_queue_failure:
1888         return -1;
1889 }
1890
1891 /* This function will check the usage_count of the drive to be updated/added.
1892  * If the usage_count is zero and it is a heretofore unknown drive, or,
1893  * the drive's capacity, geometry, or serial number has changed,
1894  * then the drive information will be updated and the disk will be
1895  * re-registered with the kernel.  If these conditions don't hold,
1896  * then it will be left alone for the next reboot.  The exception to this
1897  * is disk 0 which will always be left registered with the kernel since it
1898  * is also the controller node.  Any changes to disk 0 will show up on
1899  * the next reboot.
1900  */
1901 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1902         int via_ioctl)
1903 {
1904         ctlr_info_t *h = hba[ctlr];
1905         struct gendisk *disk;
1906         InquiryData_struct *inq_buff = NULL;
1907         unsigned int block_size;
1908         sector_t total_size;
1909         unsigned long flags = 0;
1910         int ret = 0;
1911         drive_info_struct *drvinfo;
1912
1913         /* Get information about the disk and modify the driver structure */
1914         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1915         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1916         if (inq_buff == NULL || drvinfo == NULL)
1917                 goto mem_msg;
1918
1919         /* testing to see if 16-byte CDBs are already being used */
1920         if (h->cciss_read == CCISS_READ_16) {
1921                 cciss_read_capacity_16(h->ctlr, drv_index,
1922                         &total_size, &block_size);
1923
1924         } else {
1925                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1926                 /* if read_capacity returns all F's this volume is >2TB */
1927                 /* in size so we switch to 16-byte CDB's for all */
1928                 /* read/write ops */
1929                 if (total_size == 0xFFFFFFFFULL) {
1930                         cciss_read_capacity_16(ctlr, drv_index,
1931                         &total_size, &block_size);
1932                         h->cciss_read = CCISS_READ_16;
1933                         h->cciss_write = CCISS_WRITE_16;
1934                 } else {
1935                         h->cciss_read = CCISS_READ_10;
1936                         h->cciss_write = CCISS_WRITE_10;
1937                 }
1938         }
1939
1940         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1941                                inq_buff, drvinfo);
1942         drvinfo->block_size = block_size;
1943         drvinfo->nr_blocks = total_size + 1;
1944
1945         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1946                                 drvinfo->model, drvinfo->rev);
1947         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1948                         sizeof(drvinfo->serial_no));
1949         /* Save the lunid in case we deregister the disk, below. */
1950         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1951                 sizeof(drvinfo->LunID));
1952
1953         /* Is it the same disk we already know, and nothing's changed? */
1954         if (h->drv[drv_index]->raid_level != -1 &&
1955                 ((memcmp(drvinfo->serial_no,
1956                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1957                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1958                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1959                 drvinfo->heads == h->drv[drv_index]->heads &&
1960                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1961                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1962                         /* The disk is unchanged, nothing to update */
1963                         goto freeret;
1964
1965         /* If we get here it's not the same disk, or something's changed,
1966          * so we need to * deregister it, and re-register it, if it's not
1967          * in use.
1968          * If the disk already exists then deregister it before proceeding
1969          * (unless it's the first disk (for the controller node).
1970          */
1971         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1972                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1973                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1974                 h->drv[drv_index]->busy_configuring = 1;
1975                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1976
1977                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1978                  * which keeps the interrupt handler from starting
1979                  * the queue.
1980                  */
1981                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1982         }
1983
1984         /* If the disk is in use return */
1985         if (ret)
1986                 goto freeret;
1987
1988         /* Save the new information from cciss_geometry_inquiry
1989          * and serial number inquiry.  If the disk was deregistered
1990          * above, then h->drv[drv_index] will be NULL.
1991          */
1992         if (h->drv[drv_index] == NULL) {
1993                 drvinfo->device_initialized = 0;
1994                 h->drv[drv_index] = drvinfo;
1995                 drvinfo = NULL; /* so it won't be freed below. */
1996         } else {
1997                 /* special case for cxd0 */
1998                 h->drv[drv_index]->block_size = drvinfo->block_size;
1999                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2000                 h->drv[drv_index]->heads = drvinfo->heads;
2001                 h->drv[drv_index]->sectors = drvinfo->sectors;
2002                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2003                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2004                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2005                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2006                         VENDOR_LEN + 1);
2007                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2008                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2009         }
2010
2011         ++h->num_luns;
2012         disk = h->gendisk[drv_index];
2013         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2014
2015         /* If it's not disk 0 (drv_index != 0)
2016          * or if it was disk 0, but there was previously
2017          * no actual corresponding configured logical drive
2018          * (raid_leve == -1) then we want to update the
2019          * logical drive's information.
2020          */
2021         if (drv_index || first_time) {
2022                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2023                         cciss_free_gendisk(h, drv_index);
2024                         cciss_free_drive_info(h, drv_index);
2025                         printk(KERN_WARNING "cciss:%d could not update "
2026                                 "disk %d\n", h->ctlr, drv_index);
2027                         --h->num_luns;
2028                 }
2029         }
2030
2031 freeret:
2032         kfree(inq_buff);
2033         kfree(drvinfo);
2034         return;
2035 mem_msg:
2036         printk(KERN_ERR "cciss: out of memory\n");
2037         goto freeret;
2038 }
2039
2040 /* This function will find the first index of the controllers drive array
2041  * that has a null drv pointer and allocate the drive info struct and
2042  * will return that index   This is where new drives will be added.
2043  * If the index to be returned is greater than the highest_lun index for
2044  * the controller then highest_lun is set * to this new index.
2045  * If there are no available indexes or if tha allocation fails, then -1
2046  * is returned.  * "controller_node" is used to know if this is a real
2047  * logical drive, or just the controller node, which determines if this
2048  * counts towards highest_lun.
2049  */
2050 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2051 {
2052         int i;
2053         drive_info_struct *drv;
2054
2055         /* Search for an empty slot for our drive info */
2056         for (i = 0; i < CISS_MAX_LUN; i++) {
2057
2058                 /* if not cxd0 case, and it's occupied, skip it. */
2059                 if (h->drv[i] && i != 0)
2060                         continue;
2061                 /*
2062                  * If it's cxd0 case, and drv is alloc'ed already, and a
2063                  * disk is configured there, skip it.
2064                  */
2065                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2066                         continue;
2067
2068                 /*
2069                  * We've found an empty slot.  Update highest_lun
2070                  * provided this isn't just the fake cxd0 controller node.
2071                  */
2072                 if (i > h->highest_lun && !controller_node)
2073                         h->highest_lun = i;
2074
2075                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2076                 if (i == 0 && h->drv[i] != NULL)
2077                         return i;
2078
2079                 /*
2080                  * Found an empty slot, not already alloc'ed.  Allocate it.
2081                  * Mark it with raid_level == -1, so we know it's new later on.
2082                  */
2083                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2084                 if (!drv)
2085                         return -1;
2086                 drv->raid_level = -1; /* so we know it's new */
2087                 h->drv[i] = drv;
2088                 return i;
2089         }
2090         return -1;
2091 }
2092
2093 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2094 {
2095         kfree(h->drv[drv_index]);
2096         h->drv[drv_index] = NULL;
2097 }
2098
2099 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2100 {
2101         put_disk(h->gendisk[drv_index]);
2102         h->gendisk[drv_index] = NULL;
2103 }
2104
2105 /* cciss_add_gendisk finds a free hba[]->drv structure
2106  * and allocates a gendisk if needed, and sets the lunid
2107  * in the drvinfo structure.   It returns the index into
2108  * the ->drv[] array, or -1 if none are free.
2109  * is_controller_node indicates whether highest_lun should
2110  * count this disk, or if it's only being added to provide
2111  * a means to talk to the controller in case no logical
2112  * drives have yet been configured.
2113  */
2114 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2115         int controller_node)
2116 {
2117         int drv_index;
2118
2119         drv_index = cciss_alloc_drive_info(h, controller_node);
2120         if (drv_index == -1)
2121                 return -1;
2122
2123         /*Check if the gendisk needs to be allocated */
2124         if (!h->gendisk[drv_index]) {
2125                 h->gendisk[drv_index] =
2126                         alloc_disk(1 << NWD_SHIFT);
2127                 if (!h->gendisk[drv_index]) {
2128                         printk(KERN_ERR "cciss%d: could not "
2129                                 "allocate a new disk %d\n",
2130                                 h->ctlr, drv_index);
2131                         goto err_free_drive_info;
2132                 }
2133         }
2134         memcpy(h->drv[drv_index]->LunID, lunid,
2135                 sizeof(h->drv[drv_index]->LunID));
2136         if (cciss_create_ld_sysfs_entry(h, drv_index))
2137                 goto err_free_disk;
2138         /* Don't need to mark this busy because nobody */
2139         /* else knows about this disk yet to contend */
2140         /* for access to it. */
2141         h->drv[drv_index]->busy_configuring = 0;
2142         wmb();
2143         return drv_index;
2144
2145 err_free_disk:
2146         cciss_free_gendisk(h, drv_index);
2147 err_free_drive_info:
2148         cciss_free_drive_info(h, drv_index);
2149         return -1;
2150 }
2151
2152 /* This is for the special case of a controller which
2153  * has no logical drives.  In this case, we still need
2154  * to register a disk so the controller can be accessed
2155  * by the Array Config Utility.
2156  */
2157 static void cciss_add_controller_node(ctlr_info_t *h)
2158 {
2159         struct gendisk *disk;
2160         int drv_index;
2161
2162         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2163                 return;
2164
2165         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2166         if (drv_index == -1)
2167                 goto error;
2168         h->drv[drv_index]->block_size = 512;
2169         h->drv[drv_index]->nr_blocks = 0;
2170         h->drv[drv_index]->heads = 0;
2171         h->drv[drv_index]->sectors = 0;
2172         h->drv[drv_index]->cylinders = 0;
2173         h->drv[drv_index]->raid_level = -1;
2174         memset(h->drv[drv_index]->serial_no, 0, 16);
2175         disk = h->gendisk[drv_index];
2176         if (cciss_add_disk(h, disk, drv_index) == 0)
2177                 return;
2178         cciss_free_gendisk(h, drv_index);
2179         cciss_free_drive_info(h, drv_index);
2180 error:
2181         printk(KERN_WARNING "cciss%d: could not "
2182                 "add disk 0.\n", h->ctlr);
2183         return;
2184 }
2185
2186 /* This function will add and remove logical drives from the Logical
2187  * drive array of the controller and maintain persistency of ordering
2188  * so that mount points are preserved until the next reboot.  This allows
2189  * for the removal of logical drives in the middle of the drive array
2190  * without a re-ordering of those drives.
2191  * INPUT
2192  * h            = The controller to perform the operations on
2193  */
2194 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2195         int via_ioctl)
2196 {
2197         int ctlr = h->ctlr;
2198         int num_luns;
2199         ReportLunData_struct *ld_buff = NULL;
2200         int return_code;
2201         int listlength = 0;
2202         int i;
2203         int drv_found;
2204         int drv_index = 0;
2205         unsigned char lunid[8] = CTLR_LUNID;
2206         unsigned long flags;
2207
2208         if (!capable(CAP_SYS_RAWIO))
2209                 return -EPERM;
2210
2211         /* Set busy_configuring flag for this operation */
2212         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2213         if (h->busy_configuring) {
2214                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2215                 return -EBUSY;
2216         }
2217         h->busy_configuring = 1;
2218         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2219
2220         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2221         if (ld_buff == NULL)
2222                 goto mem_msg;
2223
2224         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2225                                       sizeof(ReportLunData_struct),
2226                                       0, CTLR_LUNID, TYPE_CMD);
2227
2228         if (return_code == IO_OK)
2229                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2230         else {  /* reading number of logical volumes failed */
2231                 printk(KERN_WARNING "cciss: report logical volume"
2232                        " command failed\n");
2233                 listlength = 0;
2234                 goto freeret;
2235         }
2236
2237         num_luns = listlength / 8;      /* 8 bytes per entry */
2238         if (num_luns > CISS_MAX_LUN) {
2239                 num_luns = CISS_MAX_LUN;
2240                 printk(KERN_WARNING "cciss: more luns configured"
2241                        " on controller than can be handled by"
2242                        " this driver.\n");
2243         }
2244
2245         if (num_luns == 0)
2246                 cciss_add_controller_node(h);
2247
2248         /* Compare controller drive array to driver's drive array
2249          * to see if any drives are missing on the controller due
2250          * to action of Array Config Utility (user deletes drive)
2251          * and deregister logical drives which have disappeared.
2252          */
2253         for (i = 0; i <= h->highest_lun; i++) {
2254                 int j;
2255                 drv_found = 0;
2256
2257                 /* skip holes in the array from already deleted drives */
2258                 if (h->drv[i] == NULL)
2259                         continue;
2260
2261                 for (j = 0; j < num_luns; j++) {
2262                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2263                         if (memcmp(h->drv[i]->LunID, lunid,
2264                                 sizeof(lunid)) == 0) {
2265                                 drv_found = 1;
2266                                 break;
2267                         }
2268                 }
2269                 if (!drv_found) {
2270                         /* Deregister it from the OS, it's gone. */
2271                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2272                         h->drv[i]->busy_configuring = 1;
2273                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2274                         return_code = deregister_disk(h, i, 1, via_ioctl);
2275                         if (h->drv[i] != NULL)
2276                                 h->drv[i]->busy_configuring = 0;
2277                 }
2278         }
2279
2280         /* Compare controller drive array to driver's drive array.
2281          * Check for updates in the drive information and any new drives
2282          * on the controller due to ACU adding logical drives, or changing
2283          * a logical drive's size, etc.  Reregister any new/changed drives
2284          */
2285         for (i = 0; i < num_luns; i++) {
2286                 int j;
2287
2288                 drv_found = 0;
2289
2290                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2291                 /* Find if the LUN is already in the drive array
2292                  * of the driver.  If so then update its info
2293                  * if not in use.  If it does not exist then find
2294                  * the first free index and add it.
2295                  */
2296                 for (j = 0; j <= h->highest_lun; j++) {
2297                         if (h->drv[j] != NULL &&
2298                                 memcmp(h->drv[j]->LunID, lunid,
2299                                         sizeof(h->drv[j]->LunID)) == 0) {
2300                                 drv_index = j;
2301                                 drv_found = 1;
2302                                 break;
2303                         }
2304                 }
2305
2306                 /* check if the drive was found already in the array */
2307                 if (!drv_found) {
2308                         drv_index = cciss_add_gendisk(h, lunid, 0);
2309                         if (drv_index == -1)
2310                                 goto freeret;
2311                 }
2312                 cciss_update_drive_info(ctlr, drv_index, first_time,
2313                         via_ioctl);
2314         }               /* end for */
2315
2316 freeret:
2317         kfree(ld_buff);
2318         h->busy_configuring = 0;
2319         /* We return -1 here to tell the ACU that we have registered/updated
2320          * all of the drives that we can and to keep it from calling us
2321          * additional times.
2322          */
2323         return -1;
2324 mem_msg:
2325         printk(KERN_ERR "cciss: out of memory\n");
2326         h->busy_configuring = 0;
2327         goto freeret;
2328 }
2329
2330 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2331 {
2332         /* zero out the disk size info */
2333         drive_info->nr_blocks = 0;
2334         drive_info->block_size = 0;
2335         drive_info->heads = 0;
2336         drive_info->sectors = 0;
2337         drive_info->cylinders = 0;
2338         drive_info->raid_level = -1;
2339         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2340         memset(drive_info->model, 0, sizeof(drive_info->model));
2341         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2342         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2343         /*
2344          * don't clear the LUNID though, we need to remember which
2345          * one this one is.
2346          */
2347 }
2348
2349 /* This function will deregister the disk and it's queue from the
2350  * kernel.  It must be called with the controller lock held and the
2351  * drv structures busy_configuring flag set.  It's parameters are:
2352  *
2353  * disk = This is the disk to be deregistered
2354  * drv  = This is the drive_info_struct associated with the disk to be
2355  *        deregistered.  It contains information about the disk used
2356  *        by the driver.
2357  * clear_all = This flag determines whether or not the disk information
2358  *             is going to be completely cleared out and the highest_lun
2359  *             reset.  Sometimes we want to clear out information about
2360  *             the disk in preparation for re-adding it.  In this case
2361  *             the highest_lun should be left unchanged and the LunID
2362  *             should not be cleared.
2363  * via_ioctl
2364  *    This indicates whether we've reached this path via ioctl.
2365  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2366  *    If this path is reached via ioctl(), then the max_usage_count will
2367  *    be 1, as the process calling ioctl() has got to have the device open.
2368  *    If we get here via sysfs, then the max usage count will be zero.
2369 */
2370 static int deregister_disk(ctlr_info_t *h, int drv_index,
2371                            int clear_all, int via_ioctl)
2372 {
2373         int i;
2374         struct gendisk *disk;
2375         drive_info_struct *drv;
2376         int recalculate_highest_lun;
2377
2378         if (!capable(CAP_SYS_RAWIO))
2379                 return -EPERM;
2380
2381         drv = h->drv[drv_index];
2382         disk = h->gendisk[drv_index];
2383
2384         /* make sure logical volume is NOT is use */
2385         if (clear_all || (h->gendisk[0] == disk)) {
2386                 if (drv->usage_count > via_ioctl)
2387                         return -EBUSY;
2388         } else if (drv->usage_count > 0)
2389                 return -EBUSY;
2390
2391         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2392
2393         /* invalidate the devices and deregister the disk.  If it is disk
2394          * zero do not deregister it but just zero out it's values.  This
2395          * allows us to delete disk zero but keep the controller registered.
2396          */
2397         if (h->gendisk[0] != disk) {
2398                 struct request_queue *q = disk->queue;
2399                 if (disk->flags & GENHD_FL_UP) {
2400                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2401                         del_gendisk(disk);
2402                 }
2403                 if (q)
2404                         blk_cleanup_queue(q);
2405                 /* If clear_all is set then we are deleting the logical
2406                  * drive, not just refreshing its info.  For drives
2407                  * other than disk 0 we will call put_disk.  We do not
2408                  * do this for disk 0 as we need it to be able to
2409                  * configure the controller.
2410                  */
2411                 if (clear_all){
2412                         /* This isn't pretty, but we need to find the
2413                          * disk in our array and NULL our the pointer.
2414                          * This is so that we will call alloc_disk if
2415                          * this index is used again later.
2416                          */
2417                         for (i=0; i < CISS_MAX_LUN; i++){
2418                                 if (h->gendisk[i] == disk) {
2419                                         h->gendisk[i] = NULL;
2420                                         break;
2421                                 }
2422                         }
2423                         put_disk(disk);
2424                 }
2425         } else {
2426                 set_capacity(disk, 0);
2427                 cciss_clear_drive_info(drv);
2428         }
2429
2430         --h->num_luns;
2431
2432         /* if it was the last disk, find the new hightest lun */
2433         if (clear_all && recalculate_highest_lun) {
2434                 int newhighest = -1;
2435                 for (i = 0; i <= h->highest_lun; i++) {
2436                         /* if the disk has size > 0, it is available */
2437                         if (h->drv[i] && h->drv[i]->heads)
2438                                 newhighest = i;
2439                 }
2440                 h->highest_lun = newhighest;
2441         }
2442         return 0;
2443 }
2444
2445 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2446                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2447                 int cmd_type)
2448 {
2449         ctlr_info_t *h = hba[ctlr];
2450         u64bit buff_dma_handle;
2451         int status = IO_OK;
2452
2453         c->cmd_type = CMD_IOCTL_PEND;
2454         c->Header.ReplyQueue = 0;
2455         if (buff != NULL) {
2456                 c->Header.SGList = 1;
2457                 c->Header.SGTotal = 1;
2458         } else {
2459                 c->Header.SGList = 0;
2460                 c->Header.SGTotal = 0;
2461         }
2462         c->Header.Tag.lower = c->busaddr;
2463         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2464
2465         c->Request.Type.Type = cmd_type;
2466         if (cmd_type == TYPE_CMD) {
2467                 switch (cmd) {
2468                 case CISS_INQUIRY:
2469                         /* are we trying to read a vital product page */
2470                         if (page_code != 0) {
2471                                 c->Request.CDB[1] = 0x01;
2472                                 c->Request.CDB[2] = page_code;
2473                         }
2474                         c->Request.CDBLen = 6;
2475                         c->Request.Type.Attribute = ATTR_SIMPLE;
2476                         c->Request.Type.Direction = XFER_READ;
2477                         c->Request.Timeout = 0;
2478                         c->Request.CDB[0] = CISS_INQUIRY;
2479                         c->Request.CDB[4] = size & 0xFF;
2480                         break;
2481                 case CISS_REPORT_LOG:
2482                 case CISS_REPORT_PHYS:
2483                         /* Talking to controller so It's a physical command
2484                            mode = 00 target = 0.  Nothing to write.
2485                          */
2486                         c->Request.CDBLen = 12;
2487                         c->Request.Type.Attribute = ATTR_SIMPLE;
2488                         c->Request.Type.Direction = XFER_READ;
2489                         c->Request.Timeout = 0;
2490                         c->Request.CDB[0] = cmd;
2491                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2492                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2493                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2494                         c->Request.CDB[9] = size & 0xFF;
2495                         break;
2496
2497                 case CCISS_READ_CAPACITY:
2498                         c->Request.CDBLen = 10;
2499                         c->Request.Type.Attribute = ATTR_SIMPLE;
2500                         c->Request.Type.Direction = XFER_READ;
2501                         c->Request.Timeout = 0;
2502                         c->Request.CDB[0] = cmd;
2503                         break;
2504                 case CCISS_READ_CAPACITY_16:
2505                         c->Request.CDBLen = 16;
2506                         c->Request.Type.Attribute = ATTR_SIMPLE;
2507                         c->Request.Type.Direction = XFER_READ;
2508                         c->Request.Timeout = 0;
2509                         c->Request.CDB[0] = cmd;
2510                         c->Request.CDB[1] = 0x10;
2511                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2512                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2513                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2514                         c->Request.CDB[13] = size & 0xFF;
2515                         c->Request.Timeout = 0;
2516                         c->Request.CDB[0] = cmd;
2517                         break;
2518                 case CCISS_CACHE_FLUSH:
2519                         c->Request.CDBLen = 12;
2520                         c->Request.Type.Attribute = ATTR_SIMPLE;
2521                         c->Request.Type.Direction = XFER_WRITE;
2522                         c->Request.Timeout = 0;
2523                         c->Request.CDB[0] = BMIC_WRITE;
2524                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2525                         break;
2526                 case TEST_UNIT_READY:
2527                         c->Request.CDBLen = 6;
2528                         c->Request.Type.Attribute = ATTR_SIMPLE;
2529                         c->Request.Type.Direction = XFER_NONE;
2530                         c->Request.Timeout = 0;
2531                         break;
2532                 default:
2533                         printk(KERN_WARNING
2534                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2535                         return IO_ERROR;
2536                 }
2537         } else if (cmd_type == TYPE_MSG) {
2538                 switch (cmd) {
2539                 case 0: /* ABORT message */
2540                         c->Request.CDBLen = 12;
2541                         c->Request.Type.Attribute = ATTR_SIMPLE;
2542                         c->Request.Type.Direction = XFER_WRITE;
2543                         c->Request.Timeout = 0;
2544                         c->Request.CDB[0] = cmd;        /* abort */
2545                         c->Request.CDB[1] = 0;  /* abort a command */
2546                         /* buff contains the tag of the command to abort */
2547                         memcpy(&c->Request.CDB[4], buff, 8);
2548                         break;
2549                 case 1: /* RESET message */
2550                         c->Request.CDBLen = 16;
2551                         c->Request.Type.Attribute = ATTR_SIMPLE;
2552                         c->Request.Type.Direction = XFER_NONE;
2553                         c->Request.Timeout = 0;
2554                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2555                         c->Request.CDB[0] = cmd;        /* reset */
2556                         c->Request.CDB[1] = 0x03;       /* reset a target */
2557                         break;
2558                 case 3: /* No-Op message */
2559                         c->Request.CDBLen = 1;
2560                         c->Request.Type.Attribute = ATTR_SIMPLE;
2561                         c->Request.Type.Direction = XFER_WRITE;
2562                         c->Request.Timeout = 0;
2563                         c->Request.CDB[0] = cmd;
2564                         break;
2565                 default:
2566                         printk(KERN_WARNING
2567                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2568                         return IO_ERROR;
2569                 }
2570         } else {
2571                 printk(KERN_WARNING
2572                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2573                 return IO_ERROR;
2574         }
2575         /* Fill in the scatter gather information */
2576         if (size > 0) {
2577                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2578                                                              buff, size,
2579                                                              PCI_DMA_BIDIRECTIONAL);
2580                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2581                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2582                 c->SG[0].Len = size;
2583                 c->SG[0].Ext = 0;       /* we are not chaining */
2584         }
2585         return status;
2586 }
2587
2588 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2589 {
2590         switch (c->err_info->ScsiStatus) {
2591         case SAM_STAT_GOOD:
2592                 return IO_OK;
2593         case SAM_STAT_CHECK_CONDITION:
2594                 switch (0xf & c->err_info->SenseInfo[2]) {
2595                 case 0: return IO_OK; /* no sense */
2596                 case 1: return IO_OK; /* recovered error */
2597                 default:
2598                         if (check_for_unit_attention(h, c))
2599                                 return IO_NEEDS_RETRY;
2600                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2601                                 "check condition, sense key = 0x%02x\n",
2602                                 h->ctlr, c->Request.CDB[0],
2603                                 c->err_info->SenseInfo[2]);
2604                 }
2605                 break;
2606         default:
2607                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2608                         "scsi status = 0x%02x\n", h->ctlr,
2609                         c->Request.CDB[0], c->err_info->ScsiStatus);
2610                 break;
2611         }
2612         return IO_ERROR;
2613 }
2614
2615 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2616 {
2617         int return_status = IO_OK;
2618
2619         if (c->err_info->CommandStatus == CMD_SUCCESS)
2620                 return IO_OK;
2621
2622         switch (c->err_info->CommandStatus) {
2623         case CMD_TARGET_STATUS:
2624                 return_status = check_target_status(h, c);
2625                 break;
2626         case CMD_DATA_UNDERRUN:
2627         case CMD_DATA_OVERRUN:
2628                 /* expected for inquiry and report lun commands */
2629                 break;
2630         case CMD_INVALID:
2631                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2632                        "reported invalid\n", c->Request.CDB[0]);
2633                 return_status = IO_ERROR;
2634                 break;
2635         case CMD_PROTOCOL_ERR:
2636                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2637                        "protocol error \n", c->Request.CDB[0]);
2638                 return_status = IO_ERROR;
2639                 break;
2640         case CMD_HARDWARE_ERR:
2641                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2642                        " hardware error\n", c->Request.CDB[0]);
2643                 return_status = IO_ERROR;
2644                 break;
2645         case CMD_CONNECTION_LOST:
2646                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2647                        "connection lost\n", c->Request.CDB[0]);
2648                 return_status = IO_ERROR;
2649                 break;
2650         case CMD_ABORTED:
2651                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2652                        "aborted\n", c->Request.CDB[0]);
2653                 return_status = IO_ERROR;
2654                 break;
2655         case CMD_ABORT_FAILED:
2656                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2657                        "abort failed\n", c->Request.CDB[0]);
2658                 return_status = IO_ERROR;
2659                 break;
2660         case CMD_UNSOLICITED_ABORT:
2661                 printk(KERN_WARNING
2662                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2663                         c->Request.CDB[0]);
2664                 return_status = IO_NEEDS_RETRY;
2665                 break;
2666         default:
2667                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2668                        "unknown status %x\n", c->Request.CDB[0],
2669                        c->err_info->CommandStatus);
2670                 return_status = IO_ERROR;
2671         }
2672         return return_status;
2673 }
2674
2675 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2676         int attempt_retry)
2677 {
2678         DECLARE_COMPLETION_ONSTACK(wait);
2679         u64bit buff_dma_handle;
2680         int return_status = IO_OK;
2681
2682 resend_cmd2:
2683         c->waiting = &wait;
2684         enqueue_cmd_and_start_io(h, c);
2685
2686         wait_for_completion(&wait);
2687
2688         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2689                 goto command_done;
2690
2691         return_status = process_sendcmd_error(h, c);
2692
2693         if (return_status == IO_NEEDS_RETRY &&
2694                 c->retry_count < MAX_CMD_RETRIES) {
2695                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2696                         c->Request.CDB[0]);
2697                 c->retry_count++;
2698                 /* erase the old error information */
2699                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2700                 return_status = IO_OK;
2701                 INIT_COMPLETION(wait);
2702                 goto resend_cmd2;
2703         }
2704
2705 command_done:
2706         /* unlock the buffers from DMA */
2707         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2708         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2709         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2710                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2711         return return_status;
2712 }
2713
2714 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2715                            __u8 page_code, unsigned char scsi3addr[],
2716                         int cmd_type)
2717 {
2718         ctlr_info_t *h = hba[ctlr];
2719         CommandList_struct *c;
2720         int return_status;
2721
2722         c = cmd_alloc(h, 0);
2723         if (!c)
2724                 return -ENOMEM;
2725         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2726                 scsi3addr, cmd_type);
2727         if (return_status == IO_OK)
2728                 return_status = sendcmd_withirq_core(h, c, 1);
2729
2730         cmd_free(h, c, 0);
2731         return return_status;
2732 }
2733
2734 static void cciss_geometry_inquiry(int ctlr, int logvol,
2735                                    sector_t total_size,
2736                                    unsigned int block_size,
2737                                    InquiryData_struct *inq_buff,
2738                                    drive_info_struct *drv)
2739 {
2740         int return_code;
2741         unsigned long t;
2742         unsigned char scsi3addr[8];
2743
2744         memset(inq_buff, 0, sizeof(InquiryData_struct));
2745         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2746         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2747                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2748         if (return_code == IO_OK) {
2749                 if (inq_buff->data_byte[8] == 0xFF) {
2750                         printk(KERN_WARNING
2751                                "cciss: reading geometry failed, volume "
2752                                "does not support reading geometry\n");
2753                         drv->heads = 255;
2754                         drv->sectors = 32;      /* Sectors per track */
2755                         drv->cylinders = total_size + 1;
2756                         drv->raid_level = RAID_UNKNOWN;
2757                 } else {
2758                         drv->heads = inq_buff->data_byte[6];
2759                         drv->sectors = inq_buff->data_byte[7];
2760                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2761                         drv->cylinders += inq_buff->data_byte[5];
2762                         drv->raid_level = inq_buff->data_byte[8];
2763                 }
2764                 drv->block_size = block_size;
2765                 drv->nr_blocks = total_size + 1;
2766                 t = drv->heads * drv->sectors;
2767                 if (t > 1) {
2768                         sector_t real_size = total_size + 1;
2769                         unsigned long rem = sector_div(real_size, t);
2770                         if (rem)
2771                                 real_size++;
2772                         drv->cylinders = real_size;
2773                 }
2774         } else {                /* Get geometry failed */
2775                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2776         }
2777 }
2778
2779 static void
2780 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2781                     unsigned int *block_size)
2782 {
2783         ReadCapdata_struct *buf;
2784         int return_code;
2785         unsigned char scsi3addr[8];
2786
2787         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2788         if (!buf) {
2789                 printk(KERN_WARNING "cciss: out of memory\n");
2790                 return;
2791         }
2792
2793         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2794         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2795                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2796         if (return_code == IO_OK) {
2797                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2798                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2799         } else {                /* read capacity command failed */
2800                 printk(KERN_WARNING "cciss: read capacity failed\n");
2801                 *total_size = 0;
2802                 *block_size = BLOCK_SIZE;
2803         }
2804         kfree(buf);
2805 }
2806
2807 static void cciss_read_capacity_16(int ctlr, int logvol,
2808         sector_t *total_size, unsigned int *block_size)
2809 {
2810         ReadCapdata_struct_16 *buf;
2811         int return_code;
2812         unsigned char scsi3addr[8];
2813
2814         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2815         if (!buf) {
2816                 printk(KERN_WARNING "cciss: out of memory\n");
2817                 return;
2818         }
2819
2820         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2821         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2822                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2823                         0, scsi3addr, TYPE_CMD);
2824         if (return_code == IO_OK) {
2825                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2826                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2827         } else {                /* read capacity command failed */
2828                 printk(KERN_WARNING "cciss: read capacity failed\n");
2829                 *total_size = 0;
2830                 *block_size = BLOCK_SIZE;
2831         }
2832         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2833                (unsigned long long)*total_size+1, *block_size);
2834         kfree(buf);
2835 }
2836
2837 static int cciss_revalidate(struct gendisk *disk)
2838 {
2839         ctlr_info_t *h = get_host(disk);
2840         drive_info_struct *drv = get_drv(disk);
2841         int logvol;
2842         int FOUND = 0;
2843         unsigned int block_size;
2844         sector_t total_size;
2845         InquiryData_struct *inq_buff = NULL;
2846
2847         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2848                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2849                         sizeof(drv->LunID)) == 0) {
2850                         FOUND = 1;
2851                         break;
2852                 }
2853         }
2854
2855         if (!FOUND)
2856                 return 1;
2857
2858         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2859         if (inq_buff == NULL) {
2860                 printk(KERN_WARNING "cciss: out of memory\n");
2861                 return 1;
2862         }
2863         if (h->cciss_read == CCISS_READ_10) {
2864                 cciss_read_capacity(h->ctlr, logvol,
2865                                         &total_size, &block_size);
2866         } else {
2867                 cciss_read_capacity_16(h->ctlr, logvol,
2868                                         &total_size, &block_size);
2869         }
2870         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2871                                inq_buff, drv);
2872
2873         blk_queue_logical_block_size(drv->queue, drv->block_size);
2874         set_capacity(disk, drv->nr_blocks);
2875
2876         kfree(inq_buff);
2877         return 0;
2878 }
2879
2880 /*
2881  * Map (physical) PCI mem into (virtual) kernel space
2882  */
2883 static void __iomem *remap_pci_mem(ulong base, ulong size)
2884 {
2885         ulong page_base = ((ulong) base) & PAGE_MASK;
2886         ulong page_offs = ((ulong) base) - page_base;
2887         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2888
2889         return page_remapped ? (page_remapped + page_offs) : NULL;
2890 }
2891
2892 /*
2893  * Takes jobs of the Q and sends them to the hardware, then puts it on
2894  * the Q to wait for completion.
2895  */
2896 static void start_io(ctlr_info_t *h)
2897 {
2898         CommandList_struct *c;
2899
2900         while (!hlist_empty(&h->reqQ)) {
2901                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2902                 /* can't do anything if fifo is full */
2903                 if ((h->access.fifo_full(h))) {
2904                         printk(KERN_WARNING "cciss: fifo full\n");
2905                         break;
2906                 }
2907
2908                 /* Get the first entry from the Request Q */
2909                 removeQ(c);
2910                 h->Qdepth--;
2911
2912                 /* Tell the controller execute command */
2913                 h->access.submit_command(h, c);
2914
2915                 /* Put job onto the completed Q */
2916                 addQ(&h->cmpQ, c);
2917         }
2918 }
2919
2920 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2921 /* Zeros out the error record and then resends the command back */
2922 /* to the controller */
2923 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2924 {
2925         /* erase the old error information */
2926         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2927
2928         /* add it to software queue and then send it to the controller */
2929         addQ(&h->reqQ, c);
2930         h->Qdepth++;
2931         if (h->Qdepth > h->maxQsinceinit)
2932                 h->maxQsinceinit = h->Qdepth;
2933
2934         start_io(h);
2935 }
2936
2937 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2938         unsigned int msg_byte, unsigned int host_byte,
2939         unsigned int driver_byte)
2940 {
2941         /* inverse of macros in scsi.h */
2942         return (scsi_status_byte & 0xff) |
2943                 ((msg_byte & 0xff) << 8) |
2944                 ((host_byte & 0xff) << 16) |
2945                 ((driver_byte & 0xff) << 24);
2946 }
2947
2948 static inline int evaluate_target_status(ctlr_info_t *h,
2949                         CommandList_struct *cmd, int *retry_cmd)
2950 {
2951         unsigned char sense_key;
2952         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2953         int error_value;
2954
2955         *retry_cmd = 0;
2956         /* If we get in here, it means we got "target status", that is, scsi status */
2957         status_byte = cmd->err_info->ScsiStatus;
2958         driver_byte = DRIVER_OK;
2959         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2960
2961         if (blk_pc_request(cmd->rq))
2962                 host_byte = DID_PASSTHROUGH;
2963         else
2964                 host_byte = DID_OK;
2965
2966         error_value = make_status_bytes(status_byte, msg_byte,
2967                 host_byte, driver_byte);
2968
2969         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2970                 if (!blk_pc_request(cmd->rq))
2971                         printk(KERN_WARNING "cciss: cmd %p "
2972                                "has SCSI Status 0x%x\n",
2973                                cmd, cmd->err_info->ScsiStatus);
2974                 return error_value;
2975         }
2976
2977         /* check the sense key */
2978         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2979         /* no status or recovered error */
2980         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2981                 error_value = 0;
2982
2983         if (check_for_unit_attention(h, cmd)) {
2984                 *retry_cmd = !blk_pc_request(cmd->rq);
2985                 return 0;
2986         }
2987
2988         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2989                 if (error_value != 0)
2990                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2991                                " sense key = 0x%x\n", cmd, sense_key);
2992                 return error_value;
2993         }
2994
2995         /* SG_IO or similar, copy sense data back */
2996         if (cmd->rq->sense) {
2997                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2998                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2999                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3000                         cmd->rq->sense_len);
3001         } else
3002                 cmd->rq->sense_len = 0;
3003
3004         return error_value;
3005 }
3006
3007 /* checks the status of the job and calls complete buffers to mark all
3008  * buffers for the completed job. Note that this function does not need
3009  * to hold the hba/queue lock.
3010  */
3011 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3012                                     int timeout)
3013 {
3014         int retry_cmd = 0;
3015         struct request *rq = cmd->rq;
3016
3017         rq->errors = 0;
3018
3019         if (timeout)
3020                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3021
3022         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3023                 goto after_error_processing;
3024
3025         switch (cmd->err_info->CommandStatus) {
3026         case CMD_TARGET_STATUS:
3027                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3028                 break;
3029         case CMD_DATA_UNDERRUN:
3030                 if (blk_fs_request(cmd->rq)) {
3031                         printk(KERN_WARNING "cciss: cmd %p has"
3032                                " completed with data underrun "
3033                                "reported\n", cmd);
3034                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3035                 }
3036                 break;
3037         case CMD_DATA_OVERRUN:
3038                 if (blk_fs_request(cmd->rq))
3039                         printk(KERN_WARNING "cciss: cmd %p has"
3040                                " completed with data overrun "
3041                                "reported\n", cmd);
3042                 break;
3043         case CMD_INVALID:
3044                 printk(KERN_WARNING "cciss: cmd %p is "
3045                        "reported invalid\n", cmd);
3046                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3047                         cmd->err_info->CommandStatus, DRIVER_OK,
3048                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3049                 break;
3050         case CMD_PROTOCOL_ERR:
3051                 printk(KERN_WARNING "cciss: cmd %p has "
3052                        "protocol error \n", cmd);
3053                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3054                         cmd->err_info->CommandStatus, DRIVER_OK,
3055                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3056                 break;
3057         case CMD_HARDWARE_ERR:
3058                 printk(KERN_WARNING "cciss: cmd %p had "
3059                        " hardware error\n", cmd);
3060                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3061                         cmd->err_info->CommandStatus, DRIVER_OK,
3062                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3063                 break;
3064         case CMD_CONNECTION_LOST:
3065                 printk(KERN_WARNING "cciss: cmd %p had "
3066                        "connection lost\n", cmd);
3067                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3068                         cmd->err_info->CommandStatus, DRIVER_OK,
3069                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3070                 break;
3071         case CMD_ABORTED:
3072                 printk(KERN_WARNING "cciss: cmd %p was "
3073                        "aborted\n", cmd);
3074                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3075                         cmd->err_info->CommandStatus, DRIVER_OK,
3076                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3077                 break;
3078         case CMD_ABORT_FAILED:
3079                 printk(KERN_WARNING "cciss: cmd %p reports "
3080                        "abort failed\n", cmd);
3081                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3082                         cmd->err_info->CommandStatus, DRIVER_OK,
3083                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3084                 break;
3085         case CMD_UNSOLICITED_ABORT:
3086                 printk(KERN_WARNING "cciss%d: unsolicited "
3087                        "abort %p\n", h->ctlr, cmd);
3088                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3089                         retry_cmd = 1;
3090                         printk(KERN_WARNING
3091                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3092                         cmd->retry_count++;
3093                 } else
3094                         printk(KERN_WARNING
3095                                "cciss%d: %p retried too "
3096                                "many times\n", h->ctlr, cmd);
3097                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3098                         cmd->err_info->CommandStatus, DRIVER_OK,
3099                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3100                 break;
3101         case CMD_TIMEOUT:
3102                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3103                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3104                         cmd->err_info->CommandStatus, DRIVER_OK,
3105                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3106                 break;
3107         default:
3108                 printk(KERN_WARNING "cciss: cmd %p returned "
3109                        "unknown status %x\n", cmd,
3110                        cmd->err_info->CommandStatus);
3111                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3112                         cmd->err_info->CommandStatus, DRIVER_OK,
3113                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3114         }
3115
3116 after_error_processing:
3117
3118         /* We need to return this command */
3119         if (retry_cmd) {
3120                 resend_cciss_cmd(h, cmd);
3121                 return;
3122         }
3123         cmd->rq->completion_data = cmd;
3124         blk_complete_request(cmd->rq);
3125 }
3126
3127 static inline u32 cciss_tag_contains_index(u32 tag)
3128 {
3129 #define DIRECT_LOOKUP_BIT 0x04
3130         return tag & DIRECT_LOOKUP_BIT;
3131 }
3132
3133 static inline u32 cciss_tag_to_index(u32 tag)
3134 {
3135 #define DIRECT_LOOKUP_SHIFT 3
3136         return tag >> DIRECT_LOOKUP_SHIFT;
3137 }
3138
3139 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3140 {
3141 #define CCISS_ERROR_BITS 0x03
3142         return tag & ~CCISS_ERROR_BITS;
3143 }
3144
3145 static inline void cciss_mark_tag_indexed(u32 *tag)
3146 {
3147         *tag |= DIRECT_LOOKUP_BIT;
3148 }
3149
3150 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3151 {
3152         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3153 }
3154
3155 /*
3156  * Get a request and submit it to the controller.
3157  */
3158 static void do_cciss_request(struct request_queue *q)
3159 {
3160         ctlr_info_t *h = q->queuedata;
3161         CommandList_struct *c;
3162         sector_t start_blk;
3163         int seg;
3164         struct request *creq;
3165         u64bit temp64;
3166         struct scatterlist *tmp_sg;
3167         SGDescriptor_struct *curr_sg;
3168         drive_info_struct *drv;
3169         int i, dir;
3170         int sg_index = 0;
3171         int chained = 0;
3172
3173         /* We call start_io here in case there is a command waiting on the
3174          * queue that has not been sent.
3175          */
3176         if (blk_queue_plugged(q))
3177                 goto startio;
3178
3179       queue:
3180         creq = blk_peek_request(q);
3181         if (!creq)
3182                 goto startio;
3183
3184         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3185
3186         if ((c = cmd_alloc(h, 1)) == NULL)
3187                 goto full;
3188
3189         blk_start_request(creq);
3190
3191         tmp_sg = h->scatter_list[c->cmdindex];
3192         spin_unlock_irq(q->queue_lock);
3193
3194         c->cmd_type = CMD_RWREQ;
3195         c->rq = creq;
3196
3197         /* fill in the request */
3198         drv = creq->rq_disk->private_data;
3199         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3200         /* got command from pool, so use the command block index instead */
3201         /* for direct lookups. */
3202         /* The first 2 bits are reserved for controller error reporting. */
3203         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3204         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3205         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3206         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3207         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3208         c->Request.Type.Attribute = ATTR_SIMPLE;
3209         c->Request.Type.Direction =
3210             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3211         c->Request.Timeout = 0; /* Don't time out */
3212         c->Request.CDB[0] =
3213             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3214         start_blk = blk_rq_pos(creq);
3215 #ifdef CCISS_DEBUG
3216         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3217                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3218 #endif                          /* CCISS_DEBUG */
3219
3220         sg_init_table(tmp_sg, h->maxsgentries);
3221         seg = blk_rq_map_sg(q, creq, tmp_sg);
3222
3223         /* get the DMA records for the setup */
3224         if (c->Request.Type.Direction == XFER_READ)
3225                 dir = PCI_DMA_FROMDEVICE;
3226         else
3227                 dir = PCI_DMA_TODEVICE;
3228
3229         curr_sg = c->SG;
3230         sg_index = 0;
3231         chained = 0;
3232
3233         for (i = 0; i < seg; i++) {
3234                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3235                         !chained && ((seg - i) > 1)) {
3236                         /* Point to next chain block. */
3237                         curr_sg = h->cmd_sg_list[c->cmdindex];
3238                         sg_index = 0;
3239                         chained = 1;
3240                 }
3241                 curr_sg[sg_index].Len = tmp_sg[i].length;
3242                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3243                                                 tmp_sg[i].offset,
3244                                                 tmp_sg[i].length, dir);
3245                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3246                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3247                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3248                 ++sg_index;
3249         }
3250         if (chained)
3251                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3252                         (seg - (h->max_cmd_sgentries - 1)) *
3253                                 sizeof(SGDescriptor_struct));
3254
3255         /* track how many SG entries we are using */
3256         if (seg > h->maxSG)
3257                 h->maxSG = seg;
3258
3259 #ifdef CCISS_DEBUG
3260         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3261                         "chained[%d]\n",
3262                         blk_rq_sectors(creq), seg, chained);
3263 #endif                          /* CCISS_DEBUG */
3264
3265         c->Header.SGList = c->Header.SGTotal = seg + chained;
3266         if (seg > h->max_cmd_sgentries)
3267                 c->Header.SGList = h->max_cmd_sgentries;
3268
3269         if (likely(blk_fs_request(creq))) {
3270                 if(h->cciss_read == CCISS_READ_10) {
3271                         c->Request.CDB[1] = 0;
3272                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3273                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3274                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3275                         c->Request.CDB[5] = start_blk & 0xff;
3276                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3277                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3278                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3279                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3280                 } else {
3281                         u32 upper32 = upper_32_bits(start_blk);
3282
3283                         c->Request.CDBLen = 16;
3284                         c->Request.CDB[1]= 0;
3285                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3286                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3287                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3288                         c->Request.CDB[5]= upper32 & 0xff;
3289                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3290                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3291                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3292                         c->Request.CDB[9]= start_blk & 0xff;
3293                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3294                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3295                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3296                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3297                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3298                 }
3299         } else if (blk_pc_request(creq)) {
3300                 c->Request.CDBLen = creq->cmd_len;
3301                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3302         } else {
3303                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3304                 BUG();
3305         }
3306
3307         spin_lock_irq(q->queue_lock);
3308
3309         addQ(&h->reqQ, c);
3310         h->Qdepth++;
3311         if (h->Qdepth > h->maxQsinceinit)
3312                 h->maxQsinceinit = h->Qdepth;
3313
3314         goto queue;
3315 full:
3316         blk_stop_queue(q);
3317 startio:
3318         /* We will already have the driver lock here so not need
3319          * to lock it.
3320          */
3321         start_io(h);
3322 }
3323
3324 static inline unsigned long get_next_completion(ctlr_info_t *h)
3325 {
3326         return h->access.command_completed(h);
3327 }
3328
3329 static inline int interrupt_pending(ctlr_info_t *h)
3330 {
3331         return h->access.intr_pending(h);
3332 }
3333
3334 static inline long interrupt_not_for_us(ctlr_info_t *h)
3335 {
3336         return (((h->access.intr_pending(h) == 0) ||
3337                 (h->interrupts_enabled == 0)));
3338 }
3339
3340 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3341                         u32 raw_tag)
3342 {
3343         if (unlikely(tag_index >= h->nr_cmds)) {
3344                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3345                 return 1;
3346         }
3347         return 0;
3348 }
3349
3350 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3351                                 u32 raw_tag)
3352 {
3353         removeQ(c);
3354         if (likely(c->cmd_type == CMD_RWREQ))
3355                 complete_command(h, c, 0);
3356         else if (c->cmd_type == CMD_IOCTL_PEND)
3357                 complete(c->waiting);
3358 #ifdef CONFIG_CISS_SCSI_TAPE
3359         else if (c->cmd_type == CMD_SCSI)
3360                 complete_scsi_command(c, 0, raw_tag);
3361 #endif
3362 }
3363
3364 /* process completion of an indexed ("direct lookup") command */
3365 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3366 {
3367         u32 tag_index;
3368         CommandList_struct *c;
3369
3370         tag_index = cciss_tag_to_index(raw_tag);
3371         if (bad_tag(h, tag_index, raw_tag))
3372                 return get_next_completion(h);
3373         c = h->cmd_pool + tag_index;
3374         finish_cmd(h, c, raw_tag);
3375         return get_next_completion(h);
3376 }
3377
3378 /* process completion of a non-indexed command */
3379 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3380 {
3381         u32 tag;
3382         CommandList_struct *c = NULL;
3383         struct hlist_node *tmp;
3384         __u32 busaddr_masked, tag_masked;
3385
3386         tag = cciss_tag_discard_error_bits(raw_tag);
3387         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3388                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3389                 tag_masked = cciss_tag_discard_error_bits(tag);
3390                 if (busaddr_masked == tag_masked) {
3391                         finish_cmd(h, c, raw_tag);
3392                         return get_next_completion(h);
3393                 }
3394         }
3395         bad_tag(h, h->nr_cmds + 1, raw_tag);
3396         return get_next_completion(h);
3397 }
3398
3399 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3400 {
3401         ctlr_info_t *h = dev_id;
3402         unsigned long flags;
3403         u32 raw_tag;
3404
3405         if (interrupt_not_for_us(h))
3406                 return IRQ_NONE;
3407         /*
3408          * If there are completed commands in the completion queue,
3409          * we had better do something about it.
3410          */
3411         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3412         while (interrupt_pending(h)) {
3413                 raw_tag = get_next_completion(h);
3414                 while (raw_tag != FIFO_EMPTY) {
3415                         if (cciss_tag_contains_index(raw_tag))
3416                                 raw_tag = process_indexed_cmd(h, raw_tag);
3417                         else
3418                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3419                 }
3420         }
3421
3422         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3423         return IRQ_HANDLED;
3424 }
3425
3426 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3427  * check the interrupt pending register because it is not set.
3428  */
3429 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3430 {
3431         ctlr_info_t *h = dev_id;
3432         unsigned long flags;
3433         u32 raw_tag;
3434
3435         if (interrupt_not_for_us(h))
3436                 return IRQ_NONE;
3437         /*
3438          * If there are completed commands in the completion queue,
3439          * we had better do something about it.
3440          */
3441         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3442         raw_tag = get_next_completion(h);
3443         while (raw_tag != FIFO_EMPTY) {
3444                 if (cciss_tag_contains_index(raw_tag))
3445                         raw_tag = process_indexed_cmd(h, raw_tag);
3446                 else
3447                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3448         }
3449
3450         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3451         return IRQ_HANDLED;
3452 }
3453
3454 /**
3455  * add_to_scan_list() - add controller to rescan queue
3456  * @h:                Pointer to the controller.
3457  *
3458  * Adds the controller to the rescan queue if not already on the queue.
3459  *
3460  * returns 1 if added to the queue, 0 if skipped (could be on the
3461  * queue already, or the controller could be initializing or shutting
3462  * down).
3463  **/
3464 static int add_to_scan_list(struct ctlr_info *h)
3465 {
3466         struct ctlr_info *test_h;
3467         int found = 0;
3468         int ret = 0;
3469
3470         if (h->busy_initializing)
3471                 return 0;
3472
3473         if (!mutex_trylock(&h->busy_shutting_down))
3474                 return 0;
3475
3476         mutex_lock(&scan_mutex);
3477         list_for_each_entry(test_h, &scan_q, scan_list) {
3478                 if (test_h == h) {
3479                         found = 1;
3480                         break;
3481                 }
3482         }
3483         if (!found && !h->busy_scanning) {
3484                 INIT_COMPLETION(h->scan_wait);
3485                 list_add_tail(&h->scan_list, &scan_q);
3486                 ret = 1;
3487         }
3488         mutex_unlock(&scan_mutex);
3489         mutex_unlock(&h->busy_shutting_down);
3490
3491         return ret;
3492 }
3493
3494 /**
3495  * remove_from_scan_list() - remove controller from rescan queue
3496  * @h:                     Pointer to the controller.
3497  *
3498  * Removes the controller from the rescan queue if present. Blocks if
3499  * the controller is currently conducting a rescan.  The controller
3500  * can be in one of three states:
3501  * 1. Doesn't need a scan
3502  * 2. On the scan list, but not scanning yet (we remove it)
3503  * 3. Busy scanning (and not on the list). In this case we want to wait for
3504  *    the scan to complete to make sure the scanning thread for this
3505  *    controller is completely idle.
3506  **/
3507 static void remove_from_scan_list(struct ctlr_info *h)
3508 {
3509         struct ctlr_info *test_h, *tmp_h;
3510
3511         mutex_lock(&scan_mutex);
3512         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3513                 if (test_h == h) { /* state 2. */
3514                         list_del(&h->scan_list);
3515                         complete_all(&h->scan_wait);
3516                         mutex_unlock(&scan_mutex);
3517                         return;
3518                 }
3519         }
3520         if (h->busy_scanning) { /* state 3. */
3521                 mutex_unlock(&scan_mutex);
3522                 wait_for_completion(&h->scan_wait);
3523         } else { /* state 1, nothing to do. */
3524                 mutex_unlock(&scan_mutex);
3525         }
3526 }
3527
3528 /**
3529  * scan_thread() - kernel thread used to rescan controllers
3530  * @data:        Ignored.
3531  *
3532  * A kernel thread used scan for drive topology changes on
3533  * controllers. The thread processes only one controller at a time
3534  * using a queue.  Controllers are added to the queue using
3535  * add_to_scan_list() and removed from the queue either after done
3536  * processing or using remove_from_scan_list().
3537  *
3538  * returns 0.
3539  **/
3540 static int scan_thread(void *data)
3541 {
3542         struct ctlr_info *h;
3543
3544         while (1) {
3545                 set_current_state(TASK_INTERRUPTIBLE);
3546                 schedule();
3547                 if (kthread_should_stop())
3548                         break;
3549
3550                 while (1) {
3551                         mutex_lock(&scan_mutex);
3552                         if (list_empty(&scan_q)) {
3553                                 mutex_unlock(&scan_mutex);
3554                                 break;
3555                         }
3556
3557                         h = list_entry(scan_q.next,
3558                                        struct ctlr_info,
3559                                        scan_list);
3560                         list_del(&h->scan_list);
3561                         h->busy_scanning = 1;
3562                         mutex_unlock(&scan_mutex);
3563
3564                         rebuild_lun_table(h, 0, 0);
3565                         complete_all(&h->scan_wait);
3566                         mutex_lock(&scan_mutex);
3567                         h->busy_scanning = 0;
3568                         mutex_unlock(&scan_mutex);
3569                 }
3570         }
3571
3572         return 0;
3573 }
3574
3575 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3576 {
3577         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3578                 return 0;
3579
3580         switch (c->err_info->SenseInfo[12]) {
3581         case STATE_CHANGED:
3582                 printk(KERN_WARNING "cciss%d: a state change "
3583                         "detected, command retried\n", h->ctlr);
3584                 return 1;
3585         break;
3586         case LUN_FAILED:
3587                 printk(KERN_WARNING "cciss%d: LUN failure "
3588                         "detected, action required\n", h->ctlr);
3589                 return 1;
3590         break;
3591         case REPORT_LUNS_CHANGED:
3592                 printk(KERN_WARNING "cciss%d: report LUN data "
3593                         "changed\n", h->ctlr);
3594         /*
3595          * Here, we could call add_to_scan_list and wake up the scan thread,
3596          * except that it's quite likely that we will get more than one
3597          * REPORT_LUNS_CHANGED condition in quick succession, which means
3598          * that those which occur after the first one will likely happen
3599          * *during* the scan_thread's rescan.  And the rescan code is not
3600          * robust enough to restart in the middle, undoing what it has already
3601          * done, and it's not clear that it's even possible to do this, since
3602          * part of what it does is notify the block layer, which starts
3603          * doing it's own i/o to read partition tables and so on, and the
3604          * driver doesn't have visibility to know what might need undoing.
3605          * In any event, if possible, it is horribly complicated to get right
3606          * so we just don't do it for now.
3607          *
3608          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3609          */
3610                 return 1;
3611         break;
3612         case POWER_OR_RESET:
3613                 printk(KERN_WARNING "cciss%d: a power on "
3614                         "or device reset detected\n", h->ctlr);
3615                 return 1;
3616         break;
3617         case UNIT_ATTENTION_CLEARED:
3618                 printk(KERN_WARNING "cciss%d: unit attention "
3619                     "cleared by another initiator\n", h->ctlr);
3620                 return 1;
3621         break;
3622         default:
3623                 printk(KERN_WARNING "cciss%d: unknown "
3624                         "unit attention detected\n", h->ctlr);
3625                                 return 1;
3626         }
3627 }
3628
3629 /*
3630  *  We cannot read the structure directly, for portability we must use
3631  *   the io functions.
3632  *   This is for debug only.
3633  */
3634 #ifdef CCISS_DEBUG
3635 static void print_cfg_table(CfgTable_struct *tb)
3636 {
3637         int i;
3638         char temp_name[17];
3639
3640         printk("Controller Configuration information\n");
3641         printk("------------------------------------\n");
3642         for (i = 0; i < 4; i++)
3643                 temp_name[i] = readb(&(tb->Signature[i]));
3644         temp_name[4] = '\0';
3645         printk("   Signature = %s\n", temp_name);
3646         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3647         printk("   Transport methods supported = 0x%x\n",
3648                readl(&(tb->TransportSupport)));
3649         printk("   Transport methods active = 0x%x\n",
3650                readl(&(tb->TransportActive)));
3651         printk("   Requested transport Method = 0x%x\n",
3652                readl(&(tb->HostWrite.TransportRequest)));
3653         printk("   Coalesce Interrupt Delay = 0x%x\n",
3654                readl(&(tb->HostWrite.CoalIntDelay)));
3655         printk("   Coalesce Interrupt Count = 0x%x\n",
3656                readl(&(tb->HostWrite.CoalIntCount)));
3657         printk("   Max outstanding commands = 0x%d\n",
3658                readl(&(tb->CmdsOutMax)));
3659         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3660         for (i = 0; i < 16; i++)
3661                 temp_name[i] = readb(&(tb->ServerName[i]));
3662         temp_name[16] = '\0';
3663         printk("   Server Name = %s\n", temp_name);
3664         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3665 }
3666 #endif                          /* CCISS_DEBUG */
3667
3668 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3669 {
3670         int i, offset, mem_type, bar_type;
3671         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3672                 return 0;
3673         offset = 0;
3674         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3675                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3676                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3677                         offset += 4;
3678                 else {
3679                         mem_type = pci_resource_flags(pdev, i) &
3680                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3681                         switch (mem_type) {
3682                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3683                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3684                                 offset += 4;    /* 32 bit */
3685                                 break;
3686                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3687                                 offset += 8;
3688                                 break;
3689                         default:        /* reserved in PCI 2.2 */
3690                                 printk(KERN_WARNING
3691                                        "Base address is invalid\n");
3692                                 return -1;
3693                                 break;
3694                         }
3695                 }
3696                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3697                         return i + 1;
3698         }
3699         return -1;
3700 }
3701
3702 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3703  * controllers that are capable. If not, we use IO-APIC mode.
3704  */
3705
3706 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3707                                            struct pci_dev *pdev, __u32 board_id)
3708 {
3709 #ifdef CONFIG_PCI_MSI
3710         int err;
3711         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3712         {0, 2}, {0, 3}
3713         };
3714
3715         /* Some boards advertise MSI but don't really support it */
3716         if ((board_id == 0x40700E11) ||
3717             (board_id == 0x40800E11) ||
3718             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3719                 goto default_int_mode;
3720
3721         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3722                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3723                 if (!err) {
3724                         c->intr[0] = cciss_msix_entries[0].vector;
3725                         c->intr[1] = cciss_msix_entries[1].vector;
3726                         c->intr[2] = cciss_msix_entries[2].vector;
3727                         c->intr[3] = cciss_msix_entries[3].vector;
3728                         c->msix_vector = 1;
3729                         return;
3730                 }
3731                 if (err > 0) {
3732                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3733                                "available\n", err);
3734                         goto default_int_mode;
3735                 } else {
3736                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3737                                err);
3738                         goto default_int_mode;
3739                 }
3740         }
3741         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3742                 if (!pci_enable_msi(pdev)) {
3743                         c->msi_vector = 1;
3744                 } else {
3745                         printk(KERN_WARNING "cciss: MSI init failed\n");
3746                 }
3747         }
3748 default_int_mode:
3749 #endif                          /* CONFIG_PCI_MSI */
3750         /* if we get here we're going to use the default interrupt mode */
3751         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3752         return;
3753 }
3754
3755 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3756 {
3757         ushort subsystem_vendor_id, subsystem_device_id, command;
3758         __u32 board_id, scratchpad = 0;
3759         __u64 cfg_offset;
3760         __u32 cfg_base_addr;
3761         __u64 cfg_base_addr_index;
3762         int i, prod_index, err;
3763
3764         subsystem_vendor_id = pdev->subsystem_vendor;
3765         subsystem_device_id = pdev->subsystem_device;
3766         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3767                     subsystem_vendor_id);
3768
3769         for (i = 0; i < ARRAY_SIZE(products); i++) {
3770                 /* Stand aside for hpsa driver on request */
3771                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3772                         return -ENODEV;
3773                 if (board_id == products[i].board_id)
3774                         break;
3775         }
3776         prod_index = i;
3777         if (prod_index == ARRAY_SIZE(products)) {
3778                 dev_warn(&pdev->dev,
3779                         "unrecognized board ID: 0x%08lx, ignoring.\n",
3780                         (unsigned long) board_id);
3781                 return -ENODEV;
3782         }
3783
3784         /* check to see if controller has been disabled */
3785         /* BEFORE trying to enable it */
3786         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3787         if (!(command & 0x02)) {
3788                 printk(KERN_WARNING
3789                        "cciss: controller appears to be disabled\n");
3790                 return -ENODEV;
3791         }
3792
3793         err = pci_enable_device(pdev);
3794         if (err) {
3795                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3796                 return err;
3797         }
3798
3799         err = pci_request_regions(pdev, "cciss");
3800         if (err) {
3801                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3802                        "aborting\n");
3803                 return err;
3804         }
3805
3806 #ifdef CCISS_DEBUG
3807         printk("command = %x\n", command);
3808         printk("irq = %x\n", pdev->irq);
3809         printk("board_id = %x\n", board_id);
3810 #endif                          /* CCISS_DEBUG */
3811
3812 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3813  * else we use the IO-APIC interrupt assigned to us by system ROM.
3814  */
3815         cciss_interrupt_mode(c, pdev, board_id);
3816
3817         /* find the memory BAR */
3818         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3819                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3820                         break;
3821         }
3822         if (i == DEVICE_COUNT_RESOURCE) {
3823                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3824                 err = -ENODEV;
3825                 goto err_out_free_res;
3826         }
3827
3828         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3829                                                  * already removed
3830                                                  */
3831
3832 #ifdef CCISS_DEBUG
3833         printk("address 0 = %lx\n", c->paddr);
3834 #endif                          /* CCISS_DEBUG */
3835         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3836
3837         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3838          * We poll for up to 120 secs, once per 100ms. */
3839         for (i = 0; i < 1200; i++) {
3840                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3841                 if (scratchpad == CCISS_FIRMWARE_READY)
3842                         break;
3843                 set_current_state(TASK_INTERRUPTIBLE);
3844                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
3845         }
3846         if (scratchpad != CCISS_FIRMWARE_READY) {
3847                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3848                 err = -ENODEV;
3849                 goto err_out_free_res;
3850         }
3851
3852         /* get the address index number */
3853         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3854         cfg_base_addr &= (__u32) 0x0000ffff;
3855 #ifdef CCISS_DEBUG
3856         printk("cfg base address = %x\n", cfg_base_addr);
3857 #endif                          /* CCISS_DEBUG */
3858         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3859 #ifdef CCISS_DEBUG
3860         printk("cfg base address index = %llx\n",
3861                 (unsigned long long)cfg_base_addr_index);
3862 #endif                          /* CCISS_DEBUG */
3863         if (cfg_base_addr_index == -1) {
3864                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3865                 err = -ENODEV;
3866                 goto err_out_free_res;
3867         }
3868
3869         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3870 #ifdef CCISS_DEBUG
3871         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3872 #endif                          /* CCISS_DEBUG */
3873         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3874                                                        cfg_base_addr_index) +
3875                                     cfg_offset, sizeof(CfgTable_struct));
3876         c->board_id = board_id;
3877
3878 #ifdef CCISS_DEBUG
3879         print_cfg_table(c->cfgtable);
3880 #endif                          /* CCISS_DEBUG */
3881
3882         /* Some controllers support Zero Memory Raid (ZMR).
3883          * When configured in ZMR mode the number of supported
3884          * commands drops to 64. So instead of just setting an
3885          * arbitrary value we make the driver a little smarter.
3886          * We read the config table to tell us how many commands
3887          * are supported on the controller then subtract 4 to
3888          * leave a little room for ioctl calls.
3889          */
3890         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3891         c->maxsgentries = readl(&(c->cfgtable->MaxSGElements));
3892
3893         /*
3894          * Limit native command to 32 s/g elements to save dma'able memory.
3895          * Howvever spec says if 0, use 31
3896          */
3897
3898         c->max_cmd_sgentries = 31;
3899         if (c->maxsgentries > 512) {
3900                 c->max_cmd_sgentries = 32;
3901                 c->chainsize = c->maxsgentries - c->max_cmd_sgentries + 1;
3902                 c->maxsgentries -= 1;   /* account for chain pointer */
3903         } else {
3904                 c->maxsgentries = 31;   /* Default to traditional value */
3905                 c->chainsize = 0;       /* traditional */
3906         }
3907
3908         c->product_name = products[prod_index].product_name;
3909         c->access = *(products[prod_index].access);
3910         c->nr_cmds = c->max_commands - 4;
3911         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3912             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3913             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3914             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3915                 printk("Does not appear to be a valid CISS config table\n");
3916                 err = -ENODEV;
3917                 goto err_out_free_res;
3918         }
3919 #ifdef CONFIG_X86
3920         {
3921                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3922                 __u32 prefetch;
3923                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3924                 prefetch |= 0x100;
3925                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3926         }
3927 #endif
3928
3929         /* Disabling DMA prefetch and refetch for the P600.
3930          * An ASIC bug may result in accesses to invalid memory addresses.
3931          * We've disabled prefetch for some time now. Testing with XEN
3932          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3933          */
3934         if(board_id == 0x3225103C) {
3935                 __u32 dma_prefetch;
3936                 __u32 dma_refetch;
3937                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3938                 dma_prefetch |= 0x8000;
3939                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3940                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3941                 dma_refetch |= 0x1;
3942                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3943         }
3944
3945 #ifdef CCISS_DEBUG
3946         printk("Trying to put board into Simple mode\n");
3947 #endif                          /* CCISS_DEBUG */
3948         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3949         /* Update the field, and then ring the doorbell */
3950         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3951         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3952
3953         /* under certain very rare conditions, this can take awhile.
3954          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3955          * as we enter this code.) */
3956         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3957                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3958                         break;
3959                 /* delay and try again */
3960                 set_current_state(TASK_INTERRUPTIBLE);
3961                 schedule_timeout(msecs_to_jiffies(1));
3962         }
3963
3964 #ifdef CCISS_DEBUG
3965         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3966                readl(c->vaddr + SA5_DOORBELL));
3967 #endif                          /* CCISS_DEBUG */
3968 #ifdef CCISS_DEBUG
3969         print_cfg_table(c->cfgtable);
3970 #endif                          /* CCISS_DEBUG */
3971
3972         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3973                 printk(KERN_WARNING "cciss: unable to get board into"
3974                        " simple mode\n");
3975                 err = -ENODEV;
3976                 goto err_out_free_res;
3977         }
3978         return 0;
3979
3980 err_out_free_res:
3981         /*
3982          * Deliberately omit pci_disable_device(): it does something nasty to
3983          * Smart Array controllers that pci_enable_device does not undo
3984          */
3985         pci_release_regions(pdev);
3986         return err;
3987 }
3988
3989 /* Function to find the first free pointer into our hba[] array
3990  * Returns -1 if no free entries are left.
3991  */
3992 static int alloc_cciss_hba(void)
3993 {
3994         int i;
3995
3996         for (i = 0; i < MAX_CTLR; i++) {
3997                 if (!hba[i]) {
3998                         ctlr_info_t *p;
3999
4000                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4001                         if (!p)
4002                                 goto Enomem;
4003                         hba[i] = p;
4004                         return i;
4005                 }
4006         }
4007         printk(KERN_WARNING "cciss: This driver supports a maximum"
4008                " of %d controllers.\n", MAX_CTLR);
4009         return -1;
4010 Enomem:
4011         printk(KERN_ERR "cciss: out of memory.\n");
4012         return -1;
4013 }
4014
4015 static void free_hba(int n)
4016 {
4017         ctlr_info_t *h = hba[n];
4018         int i;
4019
4020         hba[n] = NULL;
4021         for (i = 0; i < h->highest_lun + 1; i++)
4022                 if (h->gendisk[i] != NULL)
4023                         put_disk(h->gendisk[i]);
4024         kfree(h);
4025 }
4026
4027 /* Send a message CDB to the firmware. */
4028 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4029 {
4030         typedef struct {
4031                 CommandListHeader_struct CommandHeader;
4032                 RequestBlock_struct Request;
4033                 ErrDescriptor_struct ErrorDescriptor;
4034         } Command;
4035         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4036         Command *cmd;
4037         dma_addr_t paddr64;
4038         uint32_t paddr32, tag;
4039         void __iomem *vaddr;
4040         int i, err;
4041
4042         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4043         if (vaddr == NULL)
4044                 return -ENOMEM;
4045
4046         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4047            CCISS commands, so they must be allocated from the lower 4GiB of
4048            memory. */
4049         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4050         if (err) {
4051                 iounmap(vaddr);
4052                 return -ENOMEM;
4053         }
4054
4055         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4056         if (cmd == NULL) {
4057                 iounmap(vaddr);
4058                 return -ENOMEM;
4059         }
4060
4061         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4062            although there's no guarantee, we assume that the address is at
4063            least 4-byte aligned (most likely, it's page-aligned). */
4064         paddr32 = paddr64;
4065
4066         cmd->CommandHeader.ReplyQueue = 0;
4067         cmd->CommandHeader.SGList = 0;
4068         cmd->CommandHeader.SGTotal = 0;
4069         cmd->CommandHeader.Tag.lower = paddr32;
4070         cmd->CommandHeader.Tag.upper = 0;
4071         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4072
4073         cmd->Request.CDBLen = 16;
4074         cmd->Request.Type.Type = TYPE_MSG;
4075         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4076         cmd->Request.Type.Direction = XFER_NONE;
4077         cmd->Request.Timeout = 0; /* Don't time out */
4078         cmd->Request.CDB[0] = opcode;
4079         cmd->Request.CDB[1] = type;
4080         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4081
4082         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4083         cmd->ErrorDescriptor.Addr.upper = 0;
4084         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4085
4086         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4087
4088         for (i = 0; i < 10; i++) {
4089                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4090                 if ((tag & ~3) == paddr32)
4091                         break;
4092                 schedule_timeout_uninterruptible(HZ);
4093         }
4094
4095         iounmap(vaddr);
4096
4097         /* we leak the DMA buffer here ... no choice since the controller could
4098            still complete the command. */
4099         if (i == 10) {
4100                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4101                         opcode, type);
4102                 return -ETIMEDOUT;
4103         }
4104
4105         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4106
4107         if (tag & 2) {
4108                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4109                         opcode, type);
4110                 return -EIO;
4111         }
4112
4113         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4114                 opcode, type);
4115         return 0;
4116 }
4117
4118 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4119 #define cciss_noop(p) cciss_message(p, 3, 0)
4120
4121 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4122 {
4123 /* the #defines are stolen from drivers/pci/msi.h. */
4124 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4125 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4126
4127         int pos;
4128         u16 control = 0;
4129
4130         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4131         if (pos) {
4132                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4133                 if (control & PCI_MSI_FLAGS_ENABLE) {
4134                         printk(KERN_INFO "cciss: resetting MSI\n");
4135                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4136                 }
4137         }
4138
4139         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4140         if (pos) {
4141                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4142                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4143                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4144                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4145                 }
4146         }
4147
4148         return 0;
4149 }
4150
4151 /* This does a hard reset of the controller using PCI power management
4152  * states. */
4153 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4154 {
4155         u16 pmcsr, saved_config_space[32];
4156         int i, pos;
4157
4158         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4159
4160         /* This is very nearly the same thing as
4161
4162            pci_save_state(pci_dev);
4163            pci_set_power_state(pci_dev, PCI_D3hot);
4164            pci_set_power_state(pci_dev, PCI_D0);
4165            pci_restore_state(pci_dev);
4166
4167            but we can't use these nice canned kernel routines on
4168            kexec, because they also check the MSI/MSI-X state in PCI
4169            configuration space and do the wrong thing when it is
4170            set/cleared.  Also, the pci_save/restore_state functions
4171            violate the ordering requirements for restoring the
4172            configuration space from the CCISS document (see the
4173            comment below).  So we roll our own .... */
4174
4175         for (i = 0; i < 32; i++)
4176                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4177
4178         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4179         if (pos == 0) {
4180                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4181                 return -ENODEV;
4182         }
4183
4184         /* Quoting from the Open CISS Specification: "The Power
4185          * Management Control/Status Register (CSR) controls the power
4186          * state of the device.  The normal operating state is D0,
4187          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4188          * the controller, place the interface device in D3 then to
4189          * D0, this causes a secondary PCI reset which will reset the
4190          * controller." */
4191
4192         /* enter the D3hot power management state */
4193         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4194         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4195         pmcsr |= PCI_D3hot;
4196         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4197
4198         schedule_timeout_uninterruptible(HZ >> 1);
4199
4200         /* enter the D0 power management state */
4201         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4202         pmcsr |= PCI_D0;
4203         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4204
4205         schedule_timeout_uninterruptible(HZ >> 1);
4206
4207         /* Restore the PCI configuration space.  The Open CISS
4208          * Specification says, "Restore the PCI Configuration
4209          * Registers, offsets 00h through 60h. It is important to
4210          * restore the command register, 16-bits at offset 04h,
4211          * last. Do not restore the configuration status register,
4212          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4213         for (i = 0; i < 32; i++) {
4214                 if (i == 2 || i == 3)
4215                         continue;
4216                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4217         }
4218         wmb();
4219         pci_write_config_word(pdev, 4, saved_config_space[2]);
4220
4221         return 0;
4222 }
4223
4224 /*
4225  *  This is it.  Find all the controllers and register them.  I really hate
4226  *  stealing all these major device numbers.
4227  *  returns the number of block devices registered.
4228  */
4229 static int __devinit cciss_init_one(struct pci_dev *pdev,
4230                                     const struct pci_device_id *ent)
4231 {
4232         int i;
4233         int j = 0;
4234         int k = 0;
4235         int rc;
4236         int dac, return_code;
4237         InquiryData_struct *inq_buff;
4238
4239         if (reset_devices) {
4240                 /* Reset the controller with a PCI power-cycle */
4241                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4242                         return -ENODEV;
4243
4244                 /* Now try to get the controller to respond to a no-op. Some
4245                    devices (notably the HP Smart Array 5i Controller) need
4246                    up to 30 seconds to respond. */
4247                 for (i=0; i<30; i++) {
4248                         if (cciss_noop(pdev) == 0)
4249                                 break;
4250
4251                         schedule_timeout_uninterruptible(HZ);
4252                 }
4253                 if (i == 30) {
4254                         printk(KERN_ERR "cciss: controller seems dead\n");
4255                         return -EBUSY;
4256                 }
4257         }
4258
4259         i = alloc_cciss_hba();
4260         if (i < 0)
4261                 return -1;
4262
4263         hba[i]->busy_initializing = 1;
4264         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4265         INIT_HLIST_HEAD(&hba[i]->reqQ);
4266         mutex_init(&hba[i]->busy_shutting_down);
4267
4268         if (cciss_pci_init(hba[i], pdev) != 0)
4269                 goto clean_no_release_regions;
4270
4271         sprintf(hba[i]->devname, "cciss%d", i);
4272         hba[i]->ctlr = i;
4273         hba[i]->pdev = pdev;
4274
4275         init_completion(&hba[i]->scan_wait);
4276
4277         if (cciss_create_hba_sysfs_entry(hba[i]))
4278                 goto clean0;
4279
4280         /* configure PCI DMA stuff */
4281         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4282                 dac = 1;
4283         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4284                 dac = 0;
4285         else {
4286                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4287                 goto clean1;
4288         }
4289
4290         /*
4291          * register with the major number, or get a dynamic major number
4292          * by passing 0 as argument.  This is done for greater than
4293          * 8 controller support.
4294          */
4295         if (i < MAX_CTLR_ORIG)
4296                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4297         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4298         if (rc == -EBUSY || rc == -EINVAL) {
4299                 printk(KERN_ERR
4300                        "cciss:  Unable to get major number %d for %s "
4301                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4302                 goto clean1;
4303         } else {
4304                 if (i >= MAX_CTLR_ORIG)
4305                         hba[i]->major = rc;
4306         }
4307
4308         /* make sure the board interrupts are off */
4309         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4310         if (hba[i]->msi_vector || hba[i]->msix_vector) {
4311                 if (request_irq(hba[i]->intr[SIMPLE_MODE_INT],
4312                                 do_cciss_msix_intr,
4313                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4314                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4315                                hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4316                         goto clean2;
4317                 }
4318         } else {
4319                 if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intx,
4320                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4321                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4322                                hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4323                         goto clean2;
4324                 }
4325         }
4326
4327         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4328                hba[i]->devname, pdev->device, pci_name(pdev),
4329                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4330
4331         hba[i]->cmd_pool_bits =
4332             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4333                         * sizeof(unsigned long), GFP_KERNEL);
4334         hba[i]->cmd_pool = (CommandList_struct *)
4335             pci_alloc_consistent(hba[i]->pdev,
4336                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4337                     &(hba[i]->cmd_pool_dhandle));
4338         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4339             pci_alloc_consistent(hba[i]->pdev,
4340                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4341                     &(hba[i]->errinfo_pool_dhandle));
4342         if ((hba[i]->cmd_pool_bits == NULL)
4343             || (hba[i]->cmd_pool == NULL)
4344             || (hba[i]->errinfo_pool == NULL)) {
4345                 printk(KERN_ERR "cciss: out of memory");
4346                 goto clean4;
4347         }
4348
4349         /* Need space for temp scatter list */
4350         hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4351                                                 sizeof(struct scatterlist *),
4352                                                 GFP_KERNEL);
4353         for (k = 0; k < hba[i]->nr_cmds; k++) {
4354                 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4355                                                         hba[i]->maxsgentries,
4356                                                         GFP_KERNEL);
4357                 if (hba[i]->scatter_list[k] == NULL) {
4358                         printk(KERN_ERR "cciss%d: could not allocate "
4359                                 "s/g lists\n", i);
4360                         goto clean4;
4361                 }
4362         }
4363         hba[i]->cmd_sg_list = cciss_allocate_sg_chain_blocks(hba[i],
4364                 hba[i]->chainsize, hba[i]->nr_cmds);
4365         if (!hba[i]->cmd_sg_list && hba[i]->chainsize > 0)
4366                 goto clean4;
4367
4368         spin_lock_init(&hba[i]->lock);
4369
4370         /* Initialize the pdev driver private data.
4371            have it point to hba[i].  */
4372         pci_set_drvdata(pdev, hba[i]);
4373         /* command and error info recs zeroed out before
4374            they are used */
4375         memset(hba[i]->cmd_pool_bits, 0,
4376                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4377                         * sizeof(unsigned long));
4378
4379         hba[i]->num_luns = 0;
4380         hba[i]->highest_lun = -1;
4381         for (j = 0; j < CISS_MAX_LUN; j++) {
4382                 hba[i]->drv[j] = NULL;
4383                 hba[i]->gendisk[j] = NULL;
4384         }
4385
4386         cciss_scsi_setup(i);
4387
4388         /* Turn the interrupts on so we can service requests */
4389         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4390
4391         /* Get the firmware version */
4392         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4393         if (inq_buff == NULL) {
4394                 printk(KERN_ERR "cciss: out of memory\n");
4395                 goto clean4;
4396         }
4397
4398         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4399                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4400         if (return_code == IO_OK) {
4401                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4402                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4403                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4404                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4405         } else {         /* send command failed */
4406                 printk(KERN_WARNING "cciss: unable to determine firmware"
4407                         " version of controller\n");
4408         }
4409         kfree(inq_buff);
4410
4411         cciss_procinit(i);
4412
4413         hba[i]->cciss_max_sectors = 8192;
4414
4415         rebuild_lun_table(hba[i], 1, 0);
4416         hba[i]->busy_initializing = 0;
4417         return 1;
4418
4419 clean4:
4420         kfree(hba[i]->cmd_pool_bits);
4421         /* Free up sg elements */
4422         for (k = 0; k < hba[i]->nr_cmds; k++)
4423                 kfree(hba[i]->scatter_list[k]);
4424         kfree(hba[i]->scatter_list);
4425         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4426         if (hba[i]->cmd_pool)
4427                 pci_free_consistent(hba[i]->pdev,
4428                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4429                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4430         if (hba[i]->errinfo_pool)
4431                 pci_free_consistent(hba[i]->pdev,
4432                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4433                                     hba[i]->errinfo_pool,
4434                                     hba[i]->errinfo_pool_dhandle);
4435         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4436 clean2:
4437         unregister_blkdev(hba[i]->major, hba[i]->devname);
4438 clean1:
4439         cciss_destroy_hba_sysfs_entry(hba[i]);
4440 clean0:
4441         pci_release_regions(pdev);
4442 clean_no_release_regions:
4443         hba[i]->busy_initializing = 0;
4444
4445         /*
4446          * Deliberately omit pci_disable_device(): it does something nasty to
4447          * Smart Array controllers that pci_enable_device does not undo
4448          */
4449         pci_set_drvdata(pdev, NULL);
4450         free_hba(i);
4451         return -1;
4452 }
4453
4454 static void cciss_shutdown(struct pci_dev *pdev)
4455 {
4456         ctlr_info_t *h;
4457         char *flush_buf;
4458         int return_code;
4459
4460         h = pci_get_drvdata(pdev);
4461         flush_buf = kzalloc(4, GFP_KERNEL);
4462         if (!flush_buf) {
4463                 printk(KERN_WARNING
4464                         "cciss:%d cache not flushed, out of memory.\n",
4465                         h->ctlr);
4466                 return;
4467         }
4468         /* write all data in the battery backed cache to disk */
4469         memset(flush_buf, 0, 4);
4470         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4471                 4, 0, CTLR_LUNID, TYPE_CMD);
4472         kfree(flush_buf);
4473         if (return_code != IO_OK)
4474                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4475                         h->ctlr);
4476         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4477         free_irq(h->intr[2], h);
4478 }
4479
4480 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4481 {
4482         ctlr_info_t *tmp_ptr;
4483         int i, j;
4484
4485         if (pci_get_drvdata(pdev) == NULL) {
4486                 printk(KERN_ERR "cciss: Unable to remove device \n");
4487                 return;
4488         }
4489
4490         tmp_ptr = pci_get_drvdata(pdev);
4491         i = tmp_ptr->ctlr;
4492         if (hba[i] == NULL) {
4493                 printk(KERN_ERR "cciss: device appears to "
4494                        "already be removed \n");
4495                 return;
4496         }
4497
4498         mutex_lock(&hba[i]->busy_shutting_down);
4499
4500         remove_from_scan_list(hba[i]);
4501         remove_proc_entry(hba[i]->devname, proc_cciss);
4502         unregister_blkdev(hba[i]->major, hba[i]->devname);
4503
4504         /* remove it from the disk list */
4505         for (j = 0; j < CISS_MAX_LUN; j++) {
4506                 struct gendisk *disk = hba[i]->gendisk[j];
4507                 if (disk) {
4508                         struct request_queue *q = disk->queue;
4509
4510                         if (disk->flags & GENHD_FL_UP) {
4511                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4512                                 del_gendisk(disk);
4513                         }
4514                         if (q)
4515                                 blk_cleanup_queue(q);
4516                 }
4517         }
4518
4519 #ifdef CONFIG_CISS_SCSI_TAPE
4520         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4521 #endif
4522
4523         cciss_shutdown(pdev);
4524
4525 #ifdef CONFIG_PCI_MSI
4526         if (hba[i]->msix_vector)
4527                 pci_disable_msix(hba[i]->pdev);
4528         else if (hba[i]->msi_vector)
4529                 pci_disable_msi(hba[i]->pdev);
4530 #endif                          /* CONFIG_PCI_MSI */
4531
4532         iounmap(hba[i]->vaddr);
4533
4534         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4535                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4536         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4537                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4538         kfree(hba[i]->cmd_pool_bits);
4539         /* Free up sg elements */
4540         for (j = 0; j < hba[i]->nr_cmds; j++)
4541                 kfree(hba[i]->scatter_list[j]);
4542         kfree(hba[i]->scatter_list);
4543         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4544         /*
4545          * Deliberately omit pci_disable_device(): it does something nasty to
4546          * Smart Array controllers that pci_enable_device does not undo
4547          */
4548         pci_release_regions(pdev);
4549         pci_set_drvdata(pdev, NULL);
4550         cciss_destroy_hba_sysfs_entry(hba[i]);
4551         mutex_unlock(&hba[i]->busy_shutting_down);
4552         free_hba(i);
4553 }
4554
4555 static struct pci_driver cciss_pci_driver = {
4556         .name = "cciss",
4557         .probe = cciss_init_one,
4558         .remove = __devexit_p(cciss_remove_one),
4559         .id_table = cciss_pci_device_id,        /* id_table */
4560         .shutdown = cciss_shutdown,
4561 };
4562
4563 /*
4564  *  This is it.  Register the PCI driver information for the cards we control
4565  *  the OS will call our registered routines when it finds one of our cards.
4566  */
4567 static int __init cciss_init(void)
4568 {
4569         int err;
4570
4571         /*
4572          * The hardware requires that commands are aligned on a 64-bit
4573          * boundary. Given that we use pci_alloc_consistent() to allocate an
4574          * array of them, the size must be a multiple of 8 bytes.
4575          */
4576         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4577
4578         printk(KERN_INFO DRIVER_NAME "\n");
4579
4580         err = bus_register(&cciss_bus_type);
4581         if (err)
4582                 return err;
4583
4584         /* Start the scan thread */
4585         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4586         if (IS_ERR(cciss_scan_thread)) {
4587                 err = PTR_ERR(cciss_scan_thread);
4588                 goto err_bus_unregister;
4589         }
4590
4591         /* Register for our PCI devices */
4592         err = pci_register_driver(&cciss_pci_driver);
4593         if (err)
4594                 goto err_thread_stop;
4595
4596         return err;
4597
4598 err_thread_stop:
4599         kthread_stop(cciss_scan_thread);
4600 err_bus_unregister:
4601         bus_unregister(&cciss_bus_type);
4602
4603         return err;
4604 }
4605
4606 static void __exit cciss_cleanup(void)
4607 {
4608         int i;
4609
4610         pci_unregister_driver(&cciss_pci_driver);
4611         /* double check that all controller entrys have been removed */
4612         for (i = 0; i < MAX_CTLR; i++) {
4613                 if (hba[i] != NULL) {
4614                         printk(KERN_WARNING "cciss: had to remove"
4615                                " controller %d\n", i);
4616                         cciss_remove_one(hba[i]->pdev);
4617                 }
4618         }
4619         kthread_stop(cciss_scan_thread);
4620         remove_proc_entry("driver/cciss", NULL);
4621         bus_unregister(&cciss_bus_type);
4622 }
4623
4624 module_init(cciss_init);
4625 module_exit(cciss_cleanup);