drivers/block/cciss.c

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