drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/compat.h>
  43 #include <linux/mutex.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/io.h>
  46
  47 #include <linux/dma-mapping.h>
  48 #include <linux/blkdev.h>
  49 #include <linux/genhd.h>
  50 #include <linux/completion.h>
  51 #include <scsi/scsi.h>
  52 #include <scsi/sg.h>
  53 #include <scsi/scsi_ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/scatterlist.h>
  56 #include <linux/kthread.h>
  57
  58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  59 #define DRIVER_NAME "HP CISS Driver (v 3.6.20)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 20)
  61
  62 /* Embedded module documentation macros - see modules.h */
  63 MODULE_AUTHOR("Hewlett-Packard Company");
  64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  65 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
  66                         " SA6i P600 P800 P400 P400i E200 E200i E500 P700m"
  67                         " Smart Array G2 Series SAS/SATA Controllers");
  68 MODULE_VERSION("3.6.20");
  69 MODULE_LICENSE("GPL");
  70
  71 static int cciss_allow_hpsa;
  72 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  73 MODULE_PARM_DESC(cciss_allow_hpsa,
  74         "Prevent cciss driver from accessing hardware known to be "
  75         " supported by the hpsa driver");
  76
  77 #include "cciss_cmd.h"
  78 #include "cciss.h"
  79 #include <linux/cciss_ioctl.h>
  80
  81 /* define the PCI info for the cards we can control */
  82 static const struct pci_device_id cciss_pci_device_id[] = {
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  90         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  91         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 110         {0,}
 111 };
 112
 113 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 114
 115 /*  board_id = Subsystem Device ID & Vendor ID
 116  *  product = Marketing Name for the board
 117  *  access = Address of the struct of function pointers
 118  */
 119 static struct board_type products[] = {
 120         {0x40700E11, "Smart Array 5300", &SA5_access},
 121         {0x40800E11, "Smart Array 5i", &SA5B_access},
 122         {0x40820E11, "Smart Array 532", &SA5B_access},
 123         {0x40830E11, "Smart Array 5312", &SA5B_access},
 124         {0x409A0E11, "Smart Array 641", &SA5_access},
 125         {0x409B0E11, "Smart Array 642", &SA5_access},
 126         {0x409C0E11, "Smart Array 6400", &SA5_access},
 127         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 128         {0x40910E11, "Smart Array 6i", &SA5_access},
 129         {0x3225103C, "Smart Array P600", &SA5_access},
 130         {0x3235103C, "Smart Array P400i", &SA5_access},
 131         {0x3211103C, "Smart Array E200i", &SA5_access},
 132         {0x3212103C, "Smart Array E200", &SA5_access},
 133         {0x3213103C, "Smart Array E200i", &SA5_access},
 134         {0x3214103C, "Smart Array E200i", &SA5_access},
 135         {0x3215103C, "Smart Array E200i", &SA5_access},
 136         {0x3237103C, "Smart Array E500", &SA5_access},
 137 /* controllers below this line are also supported by the hpsa driver. */
 138 #define HPSA_BOUNDARY 0x3223103C
 139         {0x3223103C, "Smart Array P800", &SA5_access},
 140         {0x3234103C, "Smart Array P400", &SA5_access},
 141         {0x323D103C, "Smart Array P700m", &SA5_access},
 142         {0x3241103C, "Smart Array P212", &SA5_access},
 143         {0x3243103C, "Smart Array P410", &SA5_access},
 144         {0x3245103C, "Smart Array P410i", &SA5_access},
 145         {0x3247103C, "Smart Array P411", &SA5_access},
 146         {0x3249103C, "Smart Array P812", &SA5_access},
 147         {0x324A103C, "Smart Array P712m", &SA5_access},
 148         {0x324B103C, "Smart Array P711m", &SA5_access},
 149 };
 150
 151 /* How long to wait (in milliseconds) for board to go into simple mode */
 152 #define MAX_CONFIG_WAIT 30000
 153 #define MAX_IOCTL_CONFIG_WAIT 1000
 154
 155 /*define how many times we will try a command because of bus resets */
 156 #define MAX_CMD_RETRIES 3
 157
 158 #define MAX_CTLR        32
 159
 160 /* Originally cciss driver only supports 8 major numbers */
 161 #define MAX_CTLR_ORIG   8
 162
 163 static ctlr_info_t *hba[MAX_CTLR];
 164
 165 static struct task_struct *cciss_scan_thread;
 166 static DEFINE_MUTEX(scan_mutex);
 167 static LIST_HEAD(scan_q);
 168
 169 static void do_cciss_request(struct request_queue *q);
 170 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
 171 static int cciss_open(struct block_device *bdev, fmode_t mode);
 172 static int cciss_release(struct gendisk *disk, fmode_t mode);
 173 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 174                        unsigned int cmd, unsigned long arg);
 175 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 176
 177 static int cciss_revalidate(struct gendisk *disk);
 178 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 179 static int deregister_disk(ctlr_info_t *h, int drv_index,
 180                            int clear_all, int via_ioctl);
 181
 182 static void cciss_read_capacity(int ctlr, int logvol,
 183                         sector_t *total_size, unsigned int *block_size);
 184 static void cciss_read_capacity_16(int ctlr, int logvol,
 185                         sector_t *total_size, unsigned int *block_size);
 186 static void cciss_geometry_inquiry(int ctlr, int logvol,
 187                         sector_t total_size,
 188                         unsigned int block_size, InquiryData_struct *inq_buff,
 189                                    drive_info_struct *drv);
 190 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 191                                            __u32);
 192 static void start_io(ctlr_info_t *h);
 193 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 194                         __u8 page_code, unsigned char scsi3addr[],
 195                         int cmd_type);
 196 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 197         int attempt_retry);
 198 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 199
 200 static void fail_all_cmds(unsigned long ctlr);
 201 static int add_to_scan_list(struct ctlr_info *h);
 202 static int scan_thread(void *data);
 203 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 204 static void cciss_hba_release(struct device *dev);
 205 static void cciss_device_release(struct device *dev);
 206 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 207 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 208
 209 #ifdef CONFIG_PROC_FS
 210 static void cciss_procinit(int i);
 211 #else
 212 static void cciss_procinit(int i)
 213 {
 214 }
 215 #endif                          /* CONFIG_PROC_FS */
 216
 217 #ifdef CONFIG_COMPAT
 218 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 219                               unsigned, unsigned long);
 220 #endif
 221
 222 static const struct block_device_operations cciss_fops = {
 223         .owner = THIS_MODULE,
 224         .open = cciss_open,
 225         .release = cciss_release,
 226         .locked_ioctl = cciss_ioctl,
 227         .getgeo = cciss_getgeo,
 228 #ifdef CONFIG_COMPAT
 229         .compat_ioctl = cciss_compat_ioctl,
 230 #endif
 231         .revalidate_disk = cciss_revalidate,
 232 };
 233
 234 /*
 235  * Enqueuing and dequeuing functions for cmdlists.
 236  */
 237 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 238 {
 239         hlist_add_head(&c->list, list);
 240 }
 241
 242 static inline void removeQ(CommandList_struct *c)
 243 {
 244         /*
 245          * After kexec/dump some commands might still
 246          * be in flight, which the firmware will try
 247          * to complete. Resetting the firmware doesn't work
 248          * with old fw revisions, so we have to mark
 249          * them off as 'stale' to prevent the driver from
 250          * falling over.
 251          */
 252         if (WARN_ON(hlist_unhashed(&c->list))) {
 253                 c->cmd_type = CMD_MSG_STALE;
 254                 return;
 255         }
 256
 257         hlist_del_init(&c->list);
 258 }
 259
 260 #include "cciss_scsi.c"         /* For SCSI tape support */
 261
 262 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 263         "UNKNOWN"
 264 };
 265 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
 266
 267 #ifdef CONFIG_PROC_FS
 268
 269 /*
 270  * Report information about this controller.
 271  */
 272 #define ENG_GIG 1000000000
 273 #define ENG_GIG_FACTOR (ENG_GIG/512)
 274 #define ENGAGE_SCSI     "engage scsi"
 275
 276 static struct proc_dir_entry *proc_cciss;
 277
 278 static void cciss_seq_show_header(struct seq_file *seq)
 279 {
 280         ctlr_info_t *h = seq->private;
 281
 282         seq_printf(seq, "%s: HP %s Controller\n"
 283                 "Board ID: 0x%08lx\n"
 284                 "Firmware Version: %c%c%c%c\n"
 285                 "IRQ: %d\n"
 286                 "Logical drives: %d\n"
 287                 "Current Q depth: %d\n"
 288                 "Current # commands on controller: %d\n"
 289                 "Max Q depth since init: %d\n"
 290                 "Max # commands on controller since init: %d\n"
 291                 "Max SG entries since init: %d\n",
 292                 h->devname,
 293                 h->product_name,
 294                 (unsigned long)h->board_id,
 295                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 296                 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
 297                 h->num_luns,
 298                 h->Qdepth, h->commands_outstanding,
 299                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 300
 301 #ifdef CONFIG_CISS_SCSI_TAPE
 302         cciss_seq_tape_report(seq, h->ctlr);
 303 #endif /* CONFIG_CISS_SCSI_TAPE */
 304 }
 305
 306 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 307 {
 308         ctlr_info_t *h = seq->private;
 309         unsigned ctlr = h->ctlr;
 310         unsigned long flags;
 311
 312         /* prevent displaying bogus info during configuration
 313          * or deconfiguration of a logical volume
 314          */
 315         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 316         if (h->busy_configuring) {
 317                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 318                 return ERR_PTR(-EBUSY);
 319         }
 320         h->busy_configuring = 1;
 321         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 322
 323         if (*pos == 0)
 324                 cciss_seq_show_header(seq);
 325
 326         return pos;
 327 }
 328
 329 static int cciss_seq_show(struct seq_file *seq, void *v)
 330 {
 331         sector_t vol_sz, vol_sz_frac;
 332         ctlr_info_t *h = seq->private;
 333         unsigned ctlr = h->ctlr;
 334         loff_t *pos = v;
 335         drive_info_struct *drv = h->drv[*pos];
 336
 337         if (*pos > h->highest_lun)
 338                 return 0;
 339
 340         if (drv->heads == 0)
 341                 return 0;
 342
 343         vol_sz = drv->nr_blocks;
 344         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 345         vol_sz_frac *= 100;
 346         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 347
 348         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 349                 drv->raid_level = RAID_UNKNOWN;
 350         seq_printf(seq, "cciss/c%dd%d:"
 351                         "\t%4u.%02uGB\tRAID %s\n",
 352                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 353                         raid_label[drv->raid_level]);
 354         return 0;
 355 }
 356
 357 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 358 {
 359         ctlr_info_t *h = seq->private;
 360
 361         if (*pos > h->highest_lun)
 362                 return NULL;
 363         *pos += 1;
 364
 365         return pos;
 366 }
 367
 368 static void cciss_seq_stop(struct seq_file *seq, void *v)
 369 {
 370         ctlr_info_t *h = seq->private;
 371
 372         /* Only reset h->busy_configuring if we succeeded in setting
 373          * it during cciss_seq_start. */
 374         if (v == ERR_PTR(-EBUSY))
 375                 return;
 376
 377         h->busy_configuring = 0;
 378 }
 379
 380 static const struct seq_operations cciss_seq_ops = {
 381         .start = cciss_seq_start,
 382         .show  = cciss_seq_show,
 383         .next  = cciss_seq_next,
 384         .stop  = cciss_seq_stop,
 385 };
 386
 387 static int cciss_seq_open(struct inode *inode, struct file *file)
 388 {
 389         int ret = seq_open(file, &cciss_seq_ops);
 390         struct seq_file *seq = file->private_data;
 391
 392         if (!ret)
 393                 seq->private = PDE(inode)->data;
 394
 395         return ret;
 396 }
 397
 398 static ssize_t
 399 cciss_proc_write(struct file *file, const char __user *buf,
 400                  size_t length, loff_t *ppos)
 401 {
 402         int err;
 403         char *buffer;
 404
 405 #ifndef CONFIG_CISS_SCSI_TAPE
 406         return -EINVAL;
 407 #endif
 408
 409         if (!buf || length > PAGE_SIZE - 1)
 410                 return -EINVAL;
 411
 412         buffer = (char *)__get_free_page(GFP_KERNEL);
 413         if (!buffer)
 414                 return -ENOMEM;
 415
 416         err = -EFAULT;
 417         if (copy_from_user(buffer, buf, length))
 418                 goto out;
 419         buffer[length] = '\0';
 420
 421 #ifdef CONFIG_CISS_SCSI_TAPE
 422         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 423                 struct seq_file *seq = file->private_data;
 424                 ctlr_info_t *h = seq->private;
 425                 int rc;
 426
 427                 rc = cciss_engage_scsi(h->ctlr);
 428                 if (rc != 0)
 429                         err = -rc;
 430                 else
 431                         err = length;
 432         } else
 433 #endif /* CONFIG_CISS_SCSI_TAPE */
 434                 err = -EINVAL;
 435         /* might be nice to have "disengage" too, but it's not
 436            safely possible. (only 1 module use count, lock issues.) */
 437
 438 out:
 439         free_page((unsigned long)buffer);
 440         return err;
 441 }
 442
 443 static const struct file_operations cciss_proc_fops = {
 444         .owner   = THIS_MODULE,
 445         .open    = cciss_seq_open,
 446         .read    = seq_read,
 447         .llseek  = seq_lseek,
 448         .release = seq_release,
 449         .write   = cciss_proc_write,
 450 };
 451
 452 static void __devinit cciss_procinit(int i)
 453 {
 454         struct proc_dir_entry *pde;
 455
 456         if (proc_cciss == NULL)
 457                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 458         if (!proc_cciss)
 459                 return;
 460         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 461                                         S_IROTH, proc_cciss,
 462                                         &cciss_proc_fops, hba[i]);
 463 }
 464 #endif                          /* CONFIG_PROC_FS */
 465
 466 #define MAX_PRODUCT_NAME_LEN 19
 467
 468 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 469 #define to_drv(n) container_of(n, drive_info_struct, dev)
 470
 471 static ssize_t host_store_rescan(struct device *dev,
 472                                  struct device_attribute *attr,
 473                                  const char *buf, size_t count)
 474 {
 475         struct ctlr_info *h = to_hba(dev);
 476
 477         add_to_scan_list(h);
 478         wake_up_process(cciss_scan_thread);
 479         wait_for_completion_interruptible(&h->scan_wait);
 480
 481         return count;
 482 }
 483 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 484
 485 static ssize_t dev_show_unique_id(struct device *dev,
 486                                  struct device_attribute *attr,
 487                                  char *buf)
 488 {
 489         drive_info_struct *drv = to_drv(dev);
 490         struct ctlr_info *h = to_hba(drv->dev.parent);
 491         __u8 sn[16];
 492         unsigned long flags;
 493         int ret = 0;
 494
 495         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 496         if (h->busy_configuring)
 497                 ret = -EBUSY;
 498         else
 499                 memcpy(sn, drv->serial_no, sizeof(sn));
 500         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 501
 502         if (ret)
 503                 return ret;
 504         else
 505                 return snprintf(buf, 16 * 2 + 2,
 506                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 507                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 508                                 sn[0], sn[1], sn[2], sn[3],
 509                                 sn[4], sn[5], sn[6], sn[7],
 510                                 sn[8], sn[9], sn[10], sn[11],
 511                                 sn[12], sn[13], sn[14], sn[15]);
 512 }
 513 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 514
 515 static ssize_t dev_show_vendor(struct device *dev,
 516                                struct device_attribute *attr,
 517                                char *buf)
 518 {
 519         drive_info_struct *drv = to_drv(dev);
 520         struct ctlr_info *h = to_hba(drv->dev.parent);
 521         char vendor[VENDOR_LEN + 1];
 522         unsigned long flags;
 523         int ret = 0;
 524
 525         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 526         if (h->busy_configuring)
 527                 ret = -EBUSY;
 528         else
 529                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 530         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 531
 532         if (ret)
 533                 return ret;
 534         else
 535                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 536 }
 537 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 538
 539 static ssize_t dev_show_model(struct device *dev,
 540                               struct device_attribute *attr,
 541                               char *buf)
 542 {
 543         drive_info_struct *drv = to_drv(dev);
 544         struct ctlr_info *h = to_hba(drv->dev.parent);
 545         char model[MODEL_LEN + 1];
 546         unsigned long flags;
 547         int ret = 0;
 548
 549         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 550         if (h->busy_configuring)
 551                 ret = -EBUSY;
 552         else
 553                 memcpy(model, drv->model, MODEL_LEN + 1);
 554         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 555
 556         if (ret)
 557                 return ret;
 558         else
 559                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 560 }
 561 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 562
 563 static ssize_t dev_show_rev(struct device *dev,
 564                             struct device_attribute *attr,
 565                             char *buf)
 566 {
 567         drive_info_struct *drv = to_drv(dev);
 568         struct ctlr_info *h = to_hba(drv->dev.parent);
 569         char rev[REV_LEN + 1];
 570         unsigned long flags;
 571         int ret = 0;
 572
 573         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 574         if (h->busy_configuring)
 575                 ret = -EBUSY;
 576         else
 577                 memcpy(rev, drv->rev, REV_LEN + 1);
 578         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 579
 580         if (ret)
 581                 return ret;
 582         else
 583                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 584 }
 585 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 586
 587 static ssize_t cciss_show_lunid(struct device *dev,
 588                                 struct device_attribute *attr, char *buf)
 589 {
 590         drive_info_struct *drv = to_drv(dev);
 591         struct ctlr_info *h = to_hba(drv->dev.parent);
 592         unsigned long flags;
 593         unsigned char lunid[8];
 594
 595         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 596         if (h->busy_configuring) {
 597                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 598                 return -EBUSY;
 599         }
 600         if (!drv->heads) {
 601                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 602                 return -ENOTTY;
 603         }
 604         memcpy(lunid, drv->LunID, sizeof(lunid));
 605         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 606         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 607                 lunid[0], lunid[1], lunid[2], lunid[3],
 608                 lunid[4], lunid[5], lunid[6], lunid[7]);
 609 }
 610 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 611
 612 static ssize_t cciss_show_raid_level(struct device *dev,
 613                                      struct device_attribute *attr, char *buf)
 614 {
 615         drive_info_struct *drv = to_drv(dev);
 616         struct ctlr_info *h = to_hba(drv->dev.parent);
 617         int raid;
 618         unsigned long flags;
 619
 620         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 621         if (h->busy_configuring) {
 622                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 623                 return -EBUSY;
 624         }
 625         raid = drv->raid_level;
 626         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 627         if (raid < 0 || raid > RAID_UNKNOWN)
 628                 raid = RAID_UNKNOWN;
 629
 630         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 631                         raid_label[raid]);
 632 }
 633 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 634
 635 static ssize_t cciss_show_usage_count(struct device *dev,
 636                                       struct device_attribute *attr, char *buf)
 637 {
 638         drive_info_struct *drv = to_drv(dev);
 639         struct ctlr_info *h = to_hba(drv->dev.parent);
 640         unsigned long flags;
 641         int count;
 642
 643         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 644         if (h->busy_configuring) {
 645                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 646                 return -EBUSY;
 647         }
 648         count = drv->usage_count;
 649         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 650         return snprintf(buf, 20, "%d\n", count);
 651 }
 652 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 653
 654 static struct attribute *cciss_host_attrs[] = {
 655         &dev_attr_rescan.attr,
 656         NULL
 657 };
 658
 659 static struct attribute_group cciss_host_attr_group = {
 660         .attrs = cciss_host_attrs,
 661 };
 662
 663 static const struct attribute_group *cciss_host_attr_groups[] = {
 664         &cciss_host_attr_group,
 665         NULL
 666 };
 667
 668 static struct device_type cciss_host_type = {
 669         .name           = "cciss_host",
 670         .groups         = cciss_host_attr_groups,
 671         .release        = cciss_hba_release,
 672 };
 673
 674 static struct attribute *cciss_dev_attrs[] = {
 675         &dev_attr_unique_id.attr,
 676         &dev_attr_model.attr,
 677         &dev_attr_vendor.attr,
 678         &dev_attr_rev.attr,
 679         &dev_attr_lunid.attr,
 680         &dev_attr_raid_level.attr,
 681         &dev_attr_usage_count.attr,
 682         NULL
 683 };
 684
 685 static struct attribute_group cciss_dev_attr_group = {
 686         .attrs = cciss_dev_attrs,
 687 };
 688
 689 static const struct attribute_group *cciss_dev_attr_groups[] = {
 690         &cciss_dev_attr_group,
 691         NULL
 692 };
 693
 694 static struct device_type cciss_dev_type = {
 695         .name           = "cciss_device",
 696         .groups         = cciss_dev_attr_groups,
 697         .release        = cciss_device_release,
 698 };
 699
 700 static struct bus_type cciss_bus_type = {
 701         .name           = "cciss",
 702 };
 703
 704 /*
 705  * cciss_hba_release is called when the reference count
 706  * of h->dev goes to zero.
 707  */
 708 static void cciss_hba_release(struct device *dev)
 709 {
 710         /*
 711          * nothing to do, but need this to avoid a warning
 712          * about not having a release handler from lib/kref.c.
 713          */
 714 }
 715
 716 /*
 717  * Initialize sysfs entry for each controller.  This sets up and registers
 718  * the 'cciss#' directory for each individual controller under
 719  * /sys/bus/pci/devices/<dev>/.
 720  */
 721 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 722 {
 723         device_initialize(&h->dev);
 724         h->dev.type = &cciss_host_type;
 725         h->dev.bus = &cciss_bus_type;
 726         dev_set_name(&h->dev, "%s", h->devname);
 727         h->dev.parent = &h->pdev->dev;
 728
 729         return device_add(&h->dev);
 730 }
 731
 732 /*
 733  * Remove sysfs entries for an hba.
 734  */
 735 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 736 {
 737         device_del(&h->dev);
 738         put_device(&h->dev); /* final put. */
 739 }
 740
 741 /* cciss_device_release is called when the reference count
 742  * of h->drv[x]dev goes to zero.
 743  */
 744 static void cciss_device_release(struct device *dev)
 745 {
 746         drive_info_struct *drv = to_drv(dev);
 747         kfree(drv);
 748 }
 749
 750 /*
 751  * Initialize sysfs for each logical drive.  This sets up and registers
 752  * the 'c#d#' directory for each individual logical drive under
 753  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 754  * /sys/block/cciss!c#d# to this entry.
 755  */
 756 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 757                                        int drv_index)
 758 {
 759         struct device *dev;
 760
 761         if (h->drv[drv_index]->device_initialized)
 762                 return 0;
 763
 764         dev = &h->drv[drv_index]->dev;
 765         device_initialize(dev);
 766         dev->type = &cciss_dev_type;
 767         dev->bus = &cciss_bus_type;
 768         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 769         dev->parent = &h->dev;
 770         h->drv[drv_index]->device_initialized = 1;
 771         return device_add(dev);
 772 }
 773
 774 /*
 775  * Remove sysfs entries for a logical drive.
 776  */
 777 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 778         int ctlr_exiting)
 779 {
 780         struct device *dev = &h->drv[drv_index]->dev;
 781
 782         /* special case for c*d0, we only destroy it on controller exit */
 783         if (drv_index == 0 && !ctlr_exiting)
 784                 return;
 785
 786         device_del(dev);
 787         put_device(dev); /* the "final" put. */
 788         h->drv[drv_index] = NULL;
 789 }
 790
 791 /*
 792  * For operations that cannot sleep, a command block is allocated at init,
 793  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 794  * which ones are free or in use.  For operations that can wait for kmalloc
 795  * to possible sleep, this routine can be called with get_from_pool set to 0.
 796  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 797  */
 798 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 799 {
 800         CommandList_struct *c;
 801         int i;
 802         u64bit temp64;
 803         dma_addr_t cmd_dma_handle, err_dma_handle;
 804
 805         if (!get_from_pool) {
 806                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 807                         sizeof(CommandList_struct), &cmd_dma_handle);
 808                 if (c == NULL)
 809                         return NULL;
 810                 memset(c, 0, sizeof(CommandList_struct));
 811
 812                 c->cmdindex = -1;
 813
 814                 c->err_info = (ErrorInfo_struct *)
 815                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 816                             &err_dma_handle);
 817
 818                 if (c->err_info == NULL) {
 819                         pci_free_consistent(h->pdev,
 820                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 821                         return NULL;
 822                 }
 823                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 824         } else {                /* get it out of the controllers pool */
 825
 826                 do {
 827                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 828                         if (i == h->nr_cmds)
 829                                 return NULL;
 830                 } while (test_and_set_bit
 831                          (i & (BITS_PER_LONG - 1),
 832                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 833 #ifdef CCISS_DEBUG
 834                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 835 #endif
 836                 c = h->cmd_pool + i;
 837                 memset(c, 0, sizeof(CommandList_struct));
 838                 cmd_dma_handle = h->cmd_pool_dhandle
 839                     + i * sizeof(CommandList_struct);
 840                 c->err_info = h->errinfo_pool + i;
 841                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 842                 err_dma_handle = h->errinfo_pool_dhandle
 843                     + i * sizeof(ErrorInfo_struct);
 844                 h->nr_allocs++;
 845
 846                 c->cmdindex = i;
 847         }
 848
 849         INIT_HLIST_NODE(&c->list);
 850         c->busaddr = (__u32) cmd_dma_handle;
 851         temp64.val = (__u64) err_dma_handle;
 852         c->ErrDesc.Addr.lower = temp64.val32.lower;
 853         c->ErrDesc.Addr.upper = temp64.val32.upper;
 854         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 855
 856         c->ctlr = h->ctlr;
 857         return c;
 858 }
 859
 860 /*
 861  * Frees a command block that was previously allocated with cmd_alloc().
 862  */
 863 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 864 {
 865         int i;
 866         u64bit temp64;
 867
 868         if (!got_from_pool) {
 869                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 870                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 871                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 872                                     c->err_info, (dma_addr_t) temp64.val);
 873                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 874                                     c, (dma_addr_t) c->busaddr);
 875         } else {
 876                 i = c - h->cmd_pool;
 877                 clear_bit(i & (BITS_PER_LONG - 1),
 878                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 879                 h->nr_frees++;
 880         }
 881 }
 882
 883 static inline ctlr_info_t *get_host(struct gendisk *disk)
 884 {
 885         return disk->queue->queuedata;
 886 }
 887
 888 static inline drive_info_struct *get_drv(struct gendisk *disk)
 889 {
 890         return disk->private_data;
 891 }
 892
 893 /*
 894  * Open.  Make sure the device is really there.
 895  */
 896 static int cciss_open(struct block_device *bdev, fmode_t mode)
 897 {
 898         ctlr_info_t *host = get_host(bdev->bd_disk);
 899         drive_info_struct *drv = get_drv(bdev->bd_disk);
 900
 901 #ifdef CCISS_DEBUG
 902         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
 903 #endif                          /* CCISS_DEBUG */
 904
 905         if (drv->busy_configuring)
 906                 return -EBUSY;
 907         /*
 908          * Root is allowed to open raw volume zero even if it's not configured
 909          * so array config can still work. Root is also allowed to open any
 910          * volume that has a LUN ID, so it can issue IOCTL to reread the
 911          * disk information.  I don't think I really like this
 912          * but I'm already using way to many device nodes to claim another one
 913          * for "raw controller".
 914          */
 915         if (drv->heads == 0) {
 916                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
 917                         /* if not node 0 make sure it is a partition = 0 */
 918                         if (MINOR(bdev->bd_dev) & 0x0f) {
 919                                 return -ENXIO;
 920                                 /* if it is, make sure we have a LUN ID */
 921                         } else if (memcmp(drv->LunID, CTLR_LUNID,
 922                                 sizeof(drv->LunID))) {
 923                                 return -ENXIO;
 924                         }
 925                 }
 926                 if (!capable(CAP_SYS_ADMIN))
 927                         return -EPERM;
 928         }
 929         drv->usage_count++;
 930         host->usage_count++;
 931         return 0;
 932 }
 933
 934 /*
 935  * Close.  Sync first.
 936  */
 937 static int cciss_release(struct gendisk *disk, fmode_t mode)
 938 {
 939         ctlr_info_t *host = get_host(disk);
 940         drive_info_struct *drv = get_drv(disk);
 941
 942 #ifdef CCISS_DEBUG
 943         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
 944 #endif                          /* CCISS_DEBUG */
 945
 946         drv->usage_count--;
 947         host->usage_count--;
 948         return 0;
 949 }
 950
 951 #ifdef CONFIG_COMPAT
 952
 953 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 954                     unsigned cmd, unsigned long arg)
 955 {
 956         int ret;
 957         lock_kernel();
 958         ret = cciss_ioctl(bdev, mode, cmd, arg);
 959         unlock_kernel();
 960         return ret;
 961 }
 962
 963 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 964                                   unsigned cmd, unsigned long arg);
 965 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
 966                                       unsigned cmd, unsigned long arg);
 967
 968 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
 969                               unsigned cmd, unsigned long arg)
 970 {
 971         switch (cmd) {
 972         case CCISS_GETPCIINFO:
 973         case CCISS_GETINTINFO:
 974         case CCISS_SETINTINFO:
 975         case CCISS_GETNODENAME:
 976         case CCISS_SETNODENAME:
 977         case CCISS_GETHEARTBEAT:
 978         case CCISS_GETBUSTYPES:
 979         case CCISS_GETFIRMVER:
 980         case CCISS_GETDRIVVER:
 981         case CCISS_REVALIDVOLS:
 982         case CCISS_DEREGDISK:
 983         case CCISS_REGNEWDISK:
 984         case CCISS_REGNEWD:
 985         case CCISS_RESCANDISK:
 986         case CCISS_GETLUNINFO:
 987                 return do_ioctl(bdev, mode, cmd, arg);
 988
 989         case CCISS_PASSTHRU32:
 990                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
 991         case CCISS_BIG_PASSTHRU32:
 992                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
 993
 994         default:
 995                 return -ENOIOCTLCMD;
 996         }
 997 }
 998
 999 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1000                                   unsigned cmd, unsigned long arg)
1001 {
1002         IOCTL32_Command_struct __user *arg32 =
1003             (IOCTL32_Command_struct __user *) arg;
1004         IOCTL_Command_struct arg64;
1005         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1006         int err;
1007         u32 cp;
1008
1009         err = 0;
1010         err |=
1011             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1012                            sizeof(arg64.LUN_info));
1013         err |=
1014             copy_from_user(&arg64.Request, &arg32->Request,
1015                            sizeof(arg64.Request));
1016         err |=
1017             copy_from_user(&arg64.error_info, &arg32->error_info,
1018                            sizeof(arg64.error_info));
1019         err |= get_user(arg64.buf_size, &arg32->buf_size);
1020         err |= get_user(cp, &arg32->buf);
1021         arg64.buf = compat_ptr(cp);
1022         err |= copy_to_user(p, &arg64, sizeof(arg64));
1023
1024         if (err)
1025                 return -EFAULT;
1026
1027         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1028         if (err)
1029                 return err;
1030         err |=
1031             copy_in_user(&arg32->error_info, &p->error_info,
1032                          sizeof(arg32->error_info));
1033         if (err)
1034                 return -EFAULT;
1035         return err;
1036 }
1037
1038 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1039                                       unsigned cmd, unsigned long arg)
1040 {
1041         BIG_IOCTL32_Command_struct __user *arg32 =
1042             (BIG_IOCTL32_Command_struct __user *) arg;
1043         BIG_IOCTL_Command_struct arg64;
1044         BIG_IOCTL_Command_struct __user *p =
1045             compat_alloc_user_space(sizeof(arg64));
1046         int err;
1047         u32 cp;
1048
1049         err = 0;
1050         err |=
1051             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1052                            sizeof(arg64.LUN_info));
1053         err |=
1054             copy_from_user(&arg64.Request, &arg32->Request,
1055                            sizeof(arg64.Request));
1056         err |=
1057             copy_from_user(&arg64.error_info, &arg32->error_info,
1058                            sizeof(arg64.error_info));
1059         err |= get_user(arg64.buf_size, &arg32->buf_size);
1060         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1061         err |= get_user(cp, &arg32->buf);
1062         arg64.buf = compat_ptr(cp);
1063         err |= copy_to_user(p, &arg64, sizeof(arg64));
1064
1065         if (err)
1066                 return -EFAULT;
1067
1068         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1069         if (err)
1070                 return err;
1071         err |=
1072             copy_in_user(&arg32->error_info, &p->error_info,
1073                          sizeof(arg32->error_info));
1074         if (err)
1075                 return -EFAULT;
1076         return err;
1077 }
1078 #endif
1079
1080 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1081 {
1082         drive_info_struct *drv = get_drv(bdev->bd_disk);
1083
1084         if (!drv->cylinders)
1085                 return -ENXIO;
1086
1087         geo->heads = drv->heads;
1088         geo->sectors = drv->sectors;
1089         geo->cylinders = drv->cylinders;
1090         return 0;
1091 }
1092
1093 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1094 {
1095         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1096                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1097                 (void)check_for_unit_attention(host, c);
1098 }
1099 /*
1100  * ioctl
1101  */
1102 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1103                        unsigned int cmd, unsigned long arg)
1104 {
1105         struct gendisk *disk = bdev->bd_disk;
1106         ctlr_info_t *host = get_host(disk);
1107         drive_info_struct *drv = get_drv(disk);
1108         int ctlr = host->ctlr;
1109         void __user *argp = (void __user *)arg;
1110
1111 #ifdef CCISS_DEBUG
1112         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1113 #endif                          /* CCISS_DEBUG */
1114
1115         switch (cmd) {
1116         case CCISS_GETPCIINFO:
1117                 {
1118                         cciss_pci_info_struct pciinfo;
1119
1120                         if (!arg)
1121                                 return -EINVAL;
1122                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1123                         pciinfo.bus = host->pdev->bus->number;
1124                         pciinfo.dev_fn = host->pdev->devfn;
1125                         pciinfo.board_id = host->board_id;
1126                         if (copy_to_user
1127                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1128                                 return -EFAULT;
1129                         return 0;
1130                 }
1131         case CCISS_GETINTINFO:
1132                 {
1133                         cciss_coalint_struct intinfo;
1134                         if (!arg)
1135                                 return -EINVAL;
1136                         intinfo.delay =
1137                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1138                         intinfo.count =
1139                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1140                         if (copy_to_user
1141                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1142                                 return -EFAULT;
1143                         return 0;
1144                 }
1145         case CCISS_SETINTINFO:
1146                 {
1147                         cciss_coalint_struct intinfo;
1148                         unsigned long flags;
1149                         int i;
1150
1151                         if (!arg)
1152                                 return -EINVAL;
1153                         if (!capable(CAP_SYS_ADMIN))
1154                                 return -EPERM;
1155                         if (copy_from_user
1156                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1157                                 return -EFAULT;
1158                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1159                         {
1160 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1161                                 return -EINVAL;
1162                         }
1163                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1164                         /* Update the field, and then ring the doorbell */
1165                         writel(intinfo.delay,
1166                                &(host->cfgtable->HostWrite.CoalIntDelay));
1167                         writel(intinfo.count,
1168                                &(host->cfgtable->HostWrite.CoalIntCount));
1169                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1170
1171                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1172                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1173                                       & CFGTBL_ChangeReq))
1174                                         break;
1175                                 /* delay and try again */
1176                                 udelay(1000);
1177                         }
1178                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1179                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1180                                 return -EAGAIN;
1181                         return 0;
1182                 }
1183         case CCISS_GETNODENAME:
1184                 {
1185                         NodeName_type NodeName;
1186                         int i;
1187
1188                         if (!arg)
1189                                 return -EINVAL;
1190                         for (i = 0; i < 16; i++)
1191                                 NodeName[i] =
1192                                     readb(&host->cfgtable->ServerName[i]);
1193                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1194                                 return -EFAULT;
1195                         return 0;
1196                 }
1197         case CCISS_SETNODENAME:
1198                 {
1199                         NodeName_type NodeName;
1200                         unsigned long flags;
1201                         int i;
1202
1203                         if (!arg)
1204                                 return -EINVAL;
1205                         if (!capable(CAP_SYS_ADMIN))
1206                                 return -EPERM;
1207
1208                         if (copy_from_user
1209                             (NodeName, argp, sizeof(NodeName_type)))
1210                                 return -EFAULT;
1211
1212                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1213
1214                         /* Update the field, and then ring the doorbell */
1215                         for (i = 0; i < 16; i++)
1216                                 writeb(NodeName[i],
1217                                        &host->cfgtable->ServerName[i]);
1218
1219                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1220
1221                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1222                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1223                                       & CFGTBL_ChangeReq))
1224                                         break;
1225                                 /* delay and try again */
1226                                 udelay(1000);
1227                         }
1228                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1229                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1230                                 return -EAGAIN;
1231                         return 0;
1232                 }
1233
1234         case CCISS_GETHEARTBEAT:
1235                 {
1236                         Heartbeat_type heartbeat;
1237
1238                         if (!arg)
1239                                 return -EINVAL;
1240                         heartbeat = readl(&host->cfgtable->HeartBeat);
1241                         if (copy_to_user
1242                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1243                                 return -EFAULT;
1244                         return 0;
1245                 }
1246         case CCISS_GETBUSTYPES:
1247                 {
1248                         BusTypes_type BusTypes;
1249
1250                         if (!arg)
1251                                 return -EINVAL;
1252                         BusTypes = readl(&host->cfgtable->BusTypes);
1253                         if (copy_to_user
1254                             (argp, &BusTypes, sizeof(BusTypes_type)))
1255                                 return -EFAULT;
1256                         return 0;
1257                 }
1258         case CCISS_GETFIRMVER:
1259                 {
1260                         FirmwareVer_type firmware;
1261
1262                         if (!arg)
1263                                 return -EINVAL;
1264                         memcpy(firmware, host->firm_ver, 4);
1265
1266                         if (copy_to_user
1267                             (argp, firmware, sizeof(FirmwareVer_type)))
1268                                 return -EFAULT;
1269                         return 0;
1270                 }
1271         case CCISS_GETDRIVVER:
1272                 {
1273                         DriverVer_type DriverVer = DRIVER_VERSION;
1274
1275                         if (!arg)
1276                                 return -EINVAL;
1277
1278                         if (copy_to_user
1279                             (argp, &DriverVer, sizeof(DriverVer_type)))
1280                                 return -EFAULT;
1281                         return 0;
1282                 }
1283
1284         case CCISS_DEREGDISK:
1285         case CCISS_REGNEWD:
1286         case CCISS_REVALIDVOLS:
1287                 return rebuild_lun_table(host, 0, 1);
1288
1289         case CCISS_GETLUNINFO:{
1290                         LogvolInfo_struct luninfo;
1291
1292                         memcpy(&luninfo.LunID, drv->LunID,
1293                                 sizeof(luninfo.LunID));
1294                         luninfo.num_opens = drv->usage_count;
1295                         luninfo.num_parts = 0;
1296                         if (copy_to_user(argp, &luninfo,
1297                                          sizeof(LogvolInfo_struct)))
1298                                 return -EFAULT;
1299                         return 0;
1300                 }
1301         case CCISS_PASSTHRU:
1302                 {
1303                         IOCTL_Command_struct iocommand;
1304                         CommandList_struct *c;
1305                         char *buff = NULL;
1306                         u64bit temp64;
1307                         unsigned long flags;
1308                         DECLARE_COMPLETION_ONSTACK(wait);
1309
1310                         if (!arg)
1311                                 return -EINVAL;
1312
1313                         if (!capable(CAP_SYS_RAWIO))
1314                                 return -EPERM;
1315
1316                         if (copy_from_user
1317                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1318                                 return -EFAULT;
1319                         if ((iocommand.buf_size < 1) &&
1320                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1321                                 return -EINVAL;
1322                         }
1323 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1324                         /* Check kmalloc limits */
1325                         if (iocommand.buf_size > 128000)
1326                                 return -EINVAL;
1327 #endif
1328                         if (iocommand.buf_size > 0) {
1329                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1330                                 if (buff == NULL)
1331                                         return -EFAULT;
1332                         }
1333                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1334                                 /* Copy the data into the buffer we created */
1335                                 if (copy_from_user
1336                                     (buff, iocommand.buf, iocommand.buf_size)) {
1337                                         kfree(buff);
1338                                         return -EFAULT;
1339                                 }
1340                         } else {
1341                                 memset(buff, 0, iocommand.buf_size);
1342                         }
1343                         if ((c = cmd_alloc(host, 0)) == NULL) {
1344                                 kfree(buff);
1345                                 return -ENOMEM;
1346                         }
1347                         // Fill in the command type
1348                         c->cmd_type = CMD_IOCTL_PEND;
1349                         // Fill in Command Header
1350                         c->Header.ReplyQueue = 0;       // unused in simple mode
1351                         if (iocommand.buf_size > 0)     // buffer to fill
1352                         {
1353                                 c->Header.SGList = 1;
1354                                 c->Header.SGTotal = 1;
1355                         } else  // no buffers to fill
1356                         {
1357                                 c->Header.SGList = 0;
1358                                 c->Header.SGTotal = 0;
1359                         }
1360                         c->Header.LUN = iocommand.LUN_info;
1361                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
1362
1363                         // Fill in Request block
1364                         c->Request = iocommand.Request;
1365
1366                         // Fill in the scatter gather information
1367                         if (iocommand.buf_size > 0) {
1368                                 temp64.val = pci_map_single(host->pdev, buff,
1369                                         iocommand.buf_size,
1370                                         PCI_DMA_BIDIRECTIONAL);
1371                                 c->SG[0].Addr.lower = temp64.val32.lower;
1372                                 c->SG[0].Addr.upper = temp64.val32.upper;
1373                                 c->SG[0].Len = iocommand.buf_size;
1374                                 c->SG[0].Ext = 0;       // we are not chaining
1375                         }
1376                         c->waiting = &wait;
1377
1378                         /* Put the request on the tail of the request queue */
1379                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1380                         addQ(&host->reqQ, c);
1381                         host->Qdepth++;
1382                         start_io(host);
1383                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1384
1385                         wait_for_completion(&wait);
1386
1387                         /* unlock the buffers from DMA */
1388                         temp64.val32.lower = c->SG[0].Addr.lower;
1389                         temp64.val32.upper = c->SG[0].Addr.upper;
1390                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1391                                          iocommand.buf_size,
1392                                          PCI_DMA_BIDIRECTIONAL);
1393
1394                         check_ioctl_unit_attention(host, c);
1395
1396                         /* Copy the error information out */
1397                         iocommand.error_info = *(c->err_info);
1398                         if (copy_to_user
1399                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1400                                 kfree(buff);
1401                                 cmd_free(host, c, 0);
1402                                 return -EFAULT;
1403                         }
1404
1405                         if (iocommand.Request.Type.Direction == XFER_READ) {
1406                                 /* Copy the data out of the buffer we created */
1407                                 if (copy_to_user
1408                                     (iocommand.buf, buff, iocommand.buf_size)) {
1409                                         kfree(buff);
1410                                         cmd_free(host, c, 0);
1411                                         return -EFAULT;
1412                                 }
1413                         }
1414                         kfree(buff);
1415                         cmd_free(host, c, 0);
1416                         return 0;
1417                 }
1418         case CCISS_BIG_PASSTHRU:{
1419                         BIG_IOCTL_Command_struct *ioc;
1420                         CommandList_struct *c;
1421                         unsigned char **buff = NULL;
1422                         int *buff_size = NULL;
1423                         u64bit temp64;
1424                         unsigned long flags;
1425                         BYTE sg_used = 0;
1426                         int status = 0;
1427                         int i;
1428                         DECLARE_COMPLETION_ONSTACK(wait);
1429                         __u32 left;
1430                         __u32 sz;
1431                         BYTE __user *data_ptr;
1432
1433                         if (!arg)
1434                                 return -EINVAL;
1435                         if (!capable(CAP_SYS_RAWIO))
1436                                 return -EPERM;
1437                         ioc = (BIG_IOCTL_Command_struct *)
1438                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1439                         if (!ioc) {
1440                                 status = -ENOMEM;
1441                                 goto cleanup1;
1442                         }
1443                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1444                                 status = -EFAULT;
1445                                 goto cleanup1;
1446                         }
1447                         if ((ioc->buf_size < 1) &&
1448                             (ioc->Request.Type.Direction != XFER_NONE)) {
1449                                 status = -EINVAL;
1450                                 goto cleanup1;
1451                         }
1452                         /* Check kmalloc limits  using all SGs */
1453                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1454                                 status = -EINVAL;
1455                                 goto cleanup1;
1456                         }
1457                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1458                                 status = -EINVAL;
1459                                 goto cleanup1;
1460                         }
1461                         buff =
1462                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1463                         if (!buff) {
1464                                 status = -ENOMEM;
1465                                 goto cleanup1;
1466                         }
1467                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1468                                                    GFP_KERNEL);
1469                         if (!buff_size) {
1470                                 status = -ENOMEM;
1471                                 goto cleanup1;
1472                         }
1473                         left = ioc->buf_size;
1474                         data_ptr = ioc->buf;
1475                         while (left) {
1476                                 sz = (left >
1477                                       ioc->malloc_size) ? ioc->
1478                                     malloc_size : left;
1479                                 buff_size[sg_used] = sz;
1480                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1481                                 if (buff[sg_used] == NULL) {
1482                                         status = -ENOMEM;
1483                                         goto cleanup1;
1484                                 }
1485                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1486                                         if (copy_from_user
1487                                             (buff[sg_used], data_ptr, sz)) {
1488                                                 status = -EFAULT;
1489                                                 goto cleanup1;
1490                                         }
1491                                 } else {
1492                                         memset(buff[sg_used], 0, sz);
1493                                 }
1494                                 left -= sz;
1495                                 data_ptr += sz;
1496                                 sg_used++;
1497                         }
1498                         if ((c = cmd_alloc(host, 0)) == NULL) {
1499                                 status = -ENOMEM;
1500                                 goto cleanup1;
1501                         }
1502                         c->cmd_type = CMD_IOCTL_PEND;
1503                         c->Header.ReplyQueue = 0;
1504
1505                         if (ioc->buf_size > 0) {
1506                                 c->Header.SGList = sg_used;
1507                                 c->Header.SGTotal = sg_used;
1508                         } else {
1509                                 c->Header.SGList = 0;
1510                                 c->Header.SGTotal = 0;
1511                         }
1512                         c->Header.LUN = ioc->LUN_info;
1513                         c->Header.Tag.lower = c->busaddr;
1514
1515                         c->Request = ioc->Request;
1516                         if (ioc->buf_size > 0) {
1517                                 int i;
1518                                 for (i = 0; i < sg_used; i++) {
1519                                         temp64.val =
1520                                             pci_map_single(host->pdev, buff[i],
1521                                                     buff_size[i],
1522                                                     PCI_DMA_BIDIRECTIONAL);
1523                                         c->SG[i].Addr.lower =
1524                                             temp64.val32.lower;
1525                                         c->SG[i].Addr.upper =
1526                                             temp64.val32.upper;
1527                                         c->SG[i].Len = buff_size[i];
1528                                         c->SG[i].Ext = 0;       /* we are not chaining */
1529                                 }
1530                         }
1531                         c->waiting = &wait;
1532                         /* Put the request on the tail of the request queue */
1533                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1534                         addQ(&host->reqQ, c);
1535                         host->Qdepth++;
1536                         start_io(host);
1537                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1538                         wait_for_completion(&wait);
1539                         /* unlock the buffers from DMA */
1540                         for (i = 0; i < sg_used; i++) {
1541                                 temp64.val32.lower = c->SG[i].Addr.lower;
1542                                 temp64.val32.upper = c->SG[i].Addr.upper;
1543                                 pci_unmap_single(host->pdev,
1544                                         (dma_addr_t) temp64.val, buff_size[i],
1545                                         PCI_DMA_BIDIRECTIONAL);
1546                         }
1547                         check_ioctl_unit_attention(host, c);
1548                         /* Copy the error information out */
1549                         ioc->error_info = *(c->err_info);
1550                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1551                                 cmd_free(host, c, 0);
1552                                 status = -EFAULT;
1553                                 goto cleanup1;
1554                         }
1555                         if (ioc->Request.Type.Direction == XFER_READ) {
1556                                 /* Copy the data out of the buffer we created */
1557                                 BYTE __user *ptr = ioc->buf;
1558                                 for (i = 0; i < sg_used; i++) {
1559                                         if (copy_to_user
1560                                             (ptr, buff[i], buff_size[i])) {
1561                                                 cmd_free(host, c, 0);
1562                                                 status = -EFAULT;
1563                                                 goto cleanup1;
1564                                         }
1565                                         ptr += buff_size[i];
1566                                 }
1567                         }
1568                         cmd_free(host, c, 0);
1569                         status = 0;
1570                       cleanup1:
1571                         if (buff) {
1572                                 for (i = 0; i < sg_used; i++)
1573                                         kfree(buff[i]);
1574                                 kfree(buff);
1575                         }
1576                         kfree(buff_size);
1577                         kfree(ioc);
1578                         return status;
1579                 }
1580
1581         /* scsi_cmd_ioctl handles these, below, though some are not */
1582         /* very meaningful for cciss.  SG_IO is the main one people want. */
1583
1584         case SG_GET_VERSION_NUM:
1585         case SG_SET_TIMEOUT:
1586         case SG_GET_TIMEOUT:
1587         case SG_GET_RESERVED_SIZE:
1588         case SG_SET_RESERVED_SIZE:
1589         case SG_EMULATED_HOST:
1590         case SG_IO:
1591         case SCSI_IOCTL_SEND_COMMAND:
1592                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1593
1594         /* scsi_cmd_ioctl would normally handle these, below, but */
1595         /* they aren't a good fit for cciss, as CD-ROMs are */
1596         /* not supported, and we don't have any bus/target/lun */
1597         /* which we present to the kernel. */
1598
1599         case CDROM_SEND_PACKET:
1600         case CDROMCLOSETRAY:
1601         case CDROMEJECT:
1602         case SCSI_IOCTL_GET_IDLUN:
1603         case SCSI_IOCTL_GET_BUS_NUMBER:
1604         default:
1605                 return -ENOTTY;
1606         }
1607 }
1608
1609 static void cciss_check_queues(ctlr_info_t *h)
1610 {
1611         int start_queue = h->next_to_run;
1612         int i;
1613
1614         /* check to see if we have maxed out the number of commands that can
1615          * be placed on the queue.  If so then exit.  We do this check here
1616          * in case the interrupt we serviced was from an ioctl and did not
1617          * free any new commands.
1618          */
1619         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1620                 return;
1621
1622         /* We have room on the queue for more commands.  Now we need to queue
1623          * them up.  We will also keep track of the next queue to run so
1624          * that every queue gets a chance to be started first.
1625          */
1626         for (i = 0; i < h->highest_lun + 1; i++) {
1627                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1628                 /* make sure the disk has been added and the drive is real
1629                  * because this can be called from the middle of init_one.
1630                  */
1631                 if (!h->drv[curr_queue])
1632                         continue;
1633                 if (!(h->drv[curr_queue]->queue) ||
1634                         !(h->drv[curr_queue]->heads))
1635                         continue;
1636                 blk_start_queue(h->gendisk[curr_queue]->queue);
1637
1638                 /* check to see if we have maxed out the number of commands
1639                  * that can be placed on the queue.
1640                  */
1641                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1642                         if (curr_queue == start_queue) {
1643                                 h->next_to_run =
1644                                     (start_queue + 1) % (h->highest_lun + 1);
1645                                 break;
1646                         } else {
1647                                 h->next_to_run = curr_queue;
1648                                 break;
1649                         }
1650                 }
1651         }
1652 }
1653
1654 static void cciss_softirq_done(struct request *rq)
1655 {
1656         CommandList_struct *cmd = rq->completion_data;
1657         ctlr_info_t *h = hba[cmd->ctlr];
1658         unsigned long flags;
1659         u64bit temp64;
1660         int i, ddir;
1661
1662         if (cmd->Request.Type.Direction == XFER_READ)
1663                 ddir = PCI_DMA_FROMDEVICE;
1664         else
1665                 ddir = PCI_DMA_TODEVICE;
1666
1667         /* command did not need to be retried */
1668         /* unmap the DMA mapping for all the scatter gather elements */
1669         for (i = 0; i < cmd->Header.SGList; i++) {
1670                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1671                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1672                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1673         }
1674
1675 #ifdef CCISS_DEBUG
1676         printk("Done with %p\n", rq);
1677 #endif                          /* CCISS_DEBUG */
1678
1679         /* set the residual count for pc requests */
1680         if (blk_pc_request(rq))
1681                 rq->resid_len = cmd->err_info->ResidualCnt;
1682
1683         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1684
1685         spin_lock_irqsave(&h->lock, flags);
1686         cmd_free(h, cmd, 1);
1687         cciss_check_queues(h);
1688         spin_unlock_irqrestore(&h->lock, flags);
1689 }
1690
1691 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1692         unsigned char scsi3addr[], uint32_t log_unit)
1693 {
1694         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1695                 sizeof(h->drv[log_unit]->LunID));
1696 }
1697
1698 /* This function gets the SCSI vendor, model, and revision of a logical drive
1699  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1700  * they cannot be read.
1701  */
1702 static void cciss_get_device_descr(int ctlr, int logvol,
1703                                    char *vendor, char *model, char *rev)
1704 {
1705         int rc;
1706         InquiryData_struct *inq_buf;
1707         unsigned char scsi3addr[8];
1708
1709         *vendor = '\0';
1710         *model = '\0';
1711         *rev = '\0';
1712
1713         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1714         if (!inq_buf)
1715                 return;
1716
1717         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1718         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1719                         scsi3addr, TYPE_CMD);
1720         if (rc == IO_OK) {
1721                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1722                 vendor[VENDOR_LEN] = '\0';
1723                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1724                 model[MODEL_LEN] = '\0';
1725                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1726                 rev[REV_LEN] = '\0';
1727         }
1728
1729         kfree(inq_buf);
1730         return;
1731 }
1732
1733 /* This function gets the serial number of a logical drive via
1734  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1735  * number cannot be had, for whatever reason, 16 bytes of 0xff
1736  * are returned instead.
1737  */
1738 static void cciss_get_serial_no(int ctlr, int logvol,
1739                                 unsigned char *serial_no, int buflen)
1740 {
1741 #define PAGE_83_INQ_BYTES 64
1742         int rc;
1743         unsigned char *buf;
1744         unsigned char scsi3addr[8];
1745
1746         if (buflen > 16)
1747                 buflen = 16;
1748         memset(serial_no, 0xff, buflen);
1749         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1750         if (!buf)
1751                 return;
1752         memset(serial_no, 0, buflen);
1753         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1754         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1755                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1756         if (rc == IO_OK)
1757                 memcpy(serial_no, &buf[8], buflen);
1758         kfree(buf);
1759         return;
1760 }
1761
1762 /*
1763  * cciss_add_disk sets up the block device queue for a logical drive
1764  */
1765 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1766                                 int drv_index)
1767 {
1768         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1769         if (!disk->queue)
1770                 goto init_queue_failure;
1771         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1772         disk->major = h->major;
1773         disk->first_minor = drv_index << NWD_SHIFT;
1774         disk->fops = &cciss_fops;
1775         if (cciss_create_ld_sysfs_entry(h, drv_index))
1776                 goto cleanup_queue;
1777         disk->private_data = h->drv[drv_index];
1778         disk->driverfs_dev = &h->drv[drv_index]->dev;
1779
1780         /* Set up queue information */
1781         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1782
1783         /* This is a hardware imposed limit. */
1784         blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1785
1786         /* This is a limit in the driver and could be eliminated. */
1787         blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1788
1789         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1790
1791         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1792
1793         disk->queue->queuedata = h;
1794
1795         blk_queue_logical_block_size(disk->queue,
1796                                      h->drv[drv_index]->block_size);
1797
1798         /* Make sure all queue data is written out before */
1799         /* setting h->drv[drv_index]->queue, as setting this */
1800         /* allows the interrupt handler to start the queue */
1801         wmb();
1802         h->drv[drv_index]->queue = disk->queue;
1803         add_disk(disk);
1804         return 0;
1805
1806 cleanup_queue:
1807         blk_cleanup_queue(disk->queue);
1808         disk->queue = NULL;
1809 init_queue_failure:
1810         return -1;
1811 }
1812
1813 /* This function will check the usage_count of the drive to be updated/added.
1814  * If the usage_count is zero and it is a heretofore unknown drive, or,
1815  * the drive's capacity, geometry, or serial number has changed,
1816  * then the drive information will be updated and the disk will be
1817  * re-registered with the kernel.  If these conditions don't hold,
1818  * then it will be left alone for the next reboot.  The exception to this
1819  * is disk 0 which will always be left registered with the kernel since it
1820  * is also the controller node.  Any changes to disk 0 will show up on
1821  * the next reboot.
1822  */
1823 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1824         int via_ioctl)
1825 {
1826         ctlr_info_t *h = hba[ctlr];
1827         struct gendisk *disk;
1828         InquiryData_struct *inq_buff = NULL;
1829         unsigned int block_size;
1830         sector_t total_size;
1831         unsigned long flags = 0;
1832         int ret = 0;
1833         drive_info_struct *drvinfo;
1834
1835         /* Get information about the disk and modify the driver structure */
1836         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1837         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1838         if (inq_buff == NULL || drvinfo == NULL)
1839                 goto mem_msg;
1840
1841         /* testing to see if 16-byte CDBs are already being used */
1842         if (h->cciss_read == CCISS_READ_16) {
1843                 cciss_read_capacity_16(h->ctlr, drv_index,
1844                         &total_size, &block_size);
1845
1846         } else {
1847                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1848                 /* if read_capacity returns all F's this volume is >2TB */
1849                 /* in size so we switch to 16-byte CDB's for all */
1850                 /* read/write ops */
1851                 if (total_size == 0xFFFFFFFFULL) {
1852                         cciss_read_capacity_16(ctlr, drv_index,
1853                         &total_size, &block_size);
1854                         h->cciss_read = CCISS_READ_16;
1855                         h->cciss_write = CCISS_WRITE_16;
1856                 } else {
1857                         h->cciss_read = CCISS_READ_10;
1858                         h->cciss_write = CCISS_WRITE_10;
1859                 }
1860         }
1861
1862         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1863                                inq_buff, drvinfo);
1864         drvinfo->block_size = block_size;
1865         drvinfo->nr_blocks = total_size + 1;
1866
1867         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1868                                 drvinfo->model, drvinfo->rev);
1869         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1870                         sizeof(drvinfo->serial_no));
1871         /* Save the lunid in case we deregister the disk, below. */
1872         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1873                 sizeof(drvinfo->LunID));
1874
1875         /* Is it the same disk we already know, and nothing's changed? */
1876         if (h->drv[drv_index]->raid_level != -1 &&
1877                 ((memcmp(drvinfo->serial_no,
1878                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1879                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1880                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1881                 drvinfo->heads == h->drv[drv_index]->heads &&
1882                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1883                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1884                         /* The disk is unchanged, nothing to update */
1885                         goto freeret;
1886
1887         /* If we get here it's not the same disk, or something's changed,
1888          * so we need to * deregister it, and re-register it, if it's not
1889          * in use.
1890          * If the disk already exists then deregister it before proceeding
1891          * (unless it's the first disk (for the controller node).
1892          */
1893         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1894                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1895                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1896                 h->drv[drv_index]->busy_configuring = 1;
1897                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1898
1899                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1900                  * which keeps the interrupt handler from starting
1901                  * the queue.
1902                  */
1903                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1904         }
1905
1906         /* If the disk is in use return */
1907         if (ret)
1908                 goto freeret;
1909
1910         /* Save the new information from cciss_geometry_inquiry
1911          * and serial number inquiry.  If the disk was deregistered
1912          * above, then h->drv[drv_index] will be NULL.
1913          */
1914         if (h->drv[drv_index] == NULL) {
1915                 drvinfo->device_initialized = 0;
1916                 h->drv[drv_index] = drvinfo;
1917                 drvinfo = NULL; /* so it won't be freed below. */
1918         } else {
1919                 /* special case for cxd0 */
1920                 h->drv[drv_index]->block_size = drvinfo->block_size;
1921                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
1922                 h->drv[drv_index]->heads = drvinfo->heads;
1923                 h->drv[drv_index]->sectors = drvinfo->sectors;
1924                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
1925                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
1926                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
1927                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
1928                         VENDOR_LEN + 1);
1929                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
1930                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
1931         }
1932
1933         ++h->num_luns;
1934         disk = h->gendisk[drv_index];
1935         set_capacity(disk, h->drv[drv_index]->nr_blocks);
1936
1937         /* If it's not disk 0 (drv_index != 0)
1938          * or if it was disk 0, but there was previously
1939          * no actual corresponding configured logical drive
1940          * (raid_leve == -1) then we want to update the
1941          * logical drive's information.
1942          */
1943         if (drv_index || first_time) {
1944                 if (cciss_add_disk(h, disk, drv_index) != 0) {
1945                         cciss_free_gendisk(h, drv_index);
1946                         cciss_free_drive_info(h, drv_index);
1947                         printk(KERN_WARNING "cciss:%d could not update "
1948                                 "disk %d\n", h->ctlr, drv_index);
1949                         --h->num_luns;
1950                 }
1951         }
1952
1953 freeret:
1954         kfree(inq_buff);
1955         kfree(drvinfo);
1956         return;
1957 mem_msg:
1958         printk(KERN_ERR "cciss: out of memory\n");
1959         goto freeret;
1960 }
1961
1962 /* This function will find the first index of the controllers drive array
1963  * that has a null drv pointer and allocate the drive info struct and
1964  * will return that index   This is where new drives will be added.
1965  * If the index to be returned is greater than the highest_lun index for
1966  * the controller then highest_lun is set * to this new index.
1967  * If there are no available indexes or if tha allocation fails, then -1
1968  * is returned.  * "controller_node" is used to know if this is a real
1969  * logical drive, or just the controller node, which determines if this
1970  * counts towards highest_lun.
1971  */
1972 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
1973 {
1974         int i;
1975         drive_info_struct *drv;
1976
1977         /* Search for an empty slot for our drive info */
1978         for (i = 0; i < CISS_MAX_LUN; i++) {
1979
1980                 /* if not cxd0 case, and it's occupied, skip it. */
1981                 if (h->drv[i] && i != 0)
1982                         continue;
1983                 /*
1984                  * If it's cxd0 case, and drv is alloc'ed already, and a
1985                  * disk is configured there, skip it.
1986                  */
1987                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
1988                         continue;
1989
1990                 /*
1991                  * We've found an empty slot.  Update highest_lun
1992                  * provided this isn't just the fake cxd0 controller node.
1993                  */
1994                 if (i > h->highest_lun && !controller_node)
1995                         h->highest_lun = i;
1996
1997                 /* If adding a real disk at cxd0, and it's already alloc'ed */
1998                 if (i == 0 && h->drv[i] != NULL)
1999                         return i;
2000
2001                 /*
2002                  * Found an empty slot, not already alloc'ed.  Allocate it.
2003                  * Mark it with raid_level == -1, so we know it's new later on.
2004                  */
2005                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2006                 if (!drv)
2007                         return -1;
2008                 drv->raid_level = -1; /* so we know it's new */
2009                 h->drv[i] = drv;
2010                 return i;
2011         }
2012         return -1;
2013 }
2014
2015 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2016 {
2017         kfree(h->drv[drv_index]);
2018         h->drv[drv_index] = NULL;
2019 }
2020
2021 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2022 {
2023         put_disk(h->gendisk[drv_index]);
2024         h->gendisk[drv_index] = NULL;
2025 }
2026
2027 /* cciss_add_gendisk finds a free hba[]->drv structure
2028  * and allocates a gendisk if needed, and sets the lunid
2029  * in the drvinfo structure.   It returns the index into
2030  * the ->drv[] array, or -1 if none are free.
2031  * is_controller_node indicates whether highest_lun should
2032  * count this disk, or if it's only being added to provide
2033  * a means to talk to the controller in case no logical
2034  * drives have yet been configured.
2035  */
2036 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2037         int controller_node)
2038 {
2039         int drv_index;
2040
2041         drv_index = cciss_alloc_drive_info(h, controller_node);
2042         if (drv_index == -1)
2043                 return -1;
2044
2045         /*Check if the gendisk needs to be allocated */
2046         if (!h->gendisk[drv_index]) {
2047                 h->gendisk[drv_index] =
2048                         alloc_disk(1 << NWD_SHIFT);
2049                 if (!h->gendisk[drv_index]) {
2050                         printk(KERN_ERR "cciss%d: could not "
2051                                 "allocate a new disk %d\n",
2052                                 h->ctlr, drv_index);
2053                         goto err_free_drive_info;
2054                 }
2055         }
2056         memcpy(h->drv[drv_index]->LunID, lunid,
2057                 sizeof(h->drv[drv_index]->LunID));
2058         if (cciss_create_ld_sysfs_entry(h, drv_index))
2059                 goto err_free_disk;
2060         /* Don't need to mark this busy because nobody */
2061         /* else knows about this disk yet to contend */
2062         /* for access to it. */
2063         h->drv[drv_index]->busy_configuring = 0;
2064         wmb();
2065         return drv_index;
2066
2067 err_free_disk:
2068         cciss_free_gendisk(h, drv_index);
2069 err_free_drive_info:
2070         cciss_free_drive_info(h, drv_index);
2071         return -1;
2072 }
2073
2074 /* This is for the special case of a controller which
2075  * has no logical drives.  In this case, we still need
2076  * to register a disk so the controller can be accessed
2077  * by the Array Config Utility.
2078  */
2079 static void cciss_add_controller_node(ctlr_info_t *h)
2080 {
2081         struct gendisk *disk;
2082         int drv_index;
2083
2084         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2085                 return;
2086
2087         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2088         if (drv_index == -1)
2089                 goto error;
2090         h->drv[drv_index]->block_size = 512;
2091         h->drv[drv_index]->nr_blocks = 0;
2092         h->drv[drv_index]->heads = 0;
2093         h->drv[drv_index]->sectors = 0;
2094         h->drv[drv_index]->cylinders = 0;
2095         h->drv[drv_index]->raid_level = -1;
2096         memset(h->drv[drv_index]->serial_no, 0, 16);
2097         disk = h->gendisk[drv_index];
2098         if (cciss_add_disk(h, disk, drv_index) == 0)
2099                 return;
2100         cciss_free_gendisk(h, drv_index);
2101         cciss_free_drive_info(h, drv_index);
2102 error:
2103         printk(KERN_WARNING "cciss%d: could not "
2104                 "add disk 0.\n", h->ctlr);
2105         return;
2106 }
2107
2108 /* This function will add and remove logical drives from the Logical
2109  * drive array of the controller and maintain persistency of ordering
2110  * so that mount points are preserved until the next reboot.  This allows
2111  * for the removal of logical drives in the middle of the drive array
2112  * without a re-ordering of those drives.
2113  * INPUT
2114  * h            = The controller to perform the operations on
2115  */
2116 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2117         int via_ioctl)
2118 {
2119         int ctlr = h->ctlr;
2120         int num_luns;
2121         ReportLunData_struct *ld_buff = NULL;
2122         int return_code;
2123         int listlength = 0;
2124         int i;
2125         int drv_found;
2126         int drv_index = 0;
2127         unsigned char lunid[8] = CTLR_LUNID;
2128         unsigned long flags;
2129
2130         if (!capable(CAP_SYS_RAWIO))
2131                 return -EPERM;
2132
2133         /* Set busy_configuring flag for this operation */
2134         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2135         if (h->busy_configuring) {
2136                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2137                 return -EBUSY;
2138         }
2139         h->busy_configuring = 1;
2140         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2141
2142         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2143         if (ld_buff == NULL)
2144                 goto mem_msg;
2145
2146         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2147                                       sizeof(ReportLunData_struct),
2148                                       0, CTLR_LUNID, TYPE_CMD);
2149
2150         if (return_code == IO_OK)
2151                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2152         else {  /* reading number of logical volumes failed */
2153                 printk(KERN_WARNING "cciss: report logical volume"
2154                        " command failed\n");
2155                 listlength = 0;
2156                 goto freeret;
2157         }
2158
2159         num_luns = listlength / 8;      /* 8 bytes per entry */
2160         if (num_luns > CISS_MAX_LUN) {
2161                 num_luns = CISS_MAX_LUN;
2162                 printk(KERN_WARNING "cciss: more luns configured"
2163                        " on controller than can be handled by"
2164                        " this driver.\n");
2165         }
2166
2167         if (num_luns == 0)
2168                 cciss_add_controller_node(h);
2169
2170         /* Compare controller drive array to driver's drive array
2171          * to see if any drives are missing on the controller due
2172          * to action of Array Config Utility (user deletes drive)
2173          * and deregister logical drives which have disappeared.
2174          */
2175         for (i = 0; i <= h->highest_lun; i++) {
2176                 int j;
2177                 drv_found = 0;
2178
2179                 /* skip holes in the array from already deleted drives */
2180                 if (h->drv[i] == NULL)
2181                         continue;
2182
2183                 for (j = 0; j < num_luns; j++) {
2184                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2185                         if (memcmp(h->drv[i]->LunID, lunid,
2186                                 sizeof(lunid)) == 0) {
2187                                 drv_found = 1;
2188                                 break;
2189                         }
2190                 }
2191                 if (!drv_found) {
2192                         /* Deregister it from the OS, it's gone. */
2193                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2194                         h->drv[i]->busy_configuring = 1;
2195                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2196                         return_code = deregister_disk(h, i, 1, via_ioctl);
2197                         if (h->drv[i] != NULL)
2198                                 h->drv[i]->busy_configuring = 0;
2199                 }
2200         }
2201
2202         /* Compare controller drive array to driver's drive array.
2203          * Check for updates in the drive information and any new drives
2204          * on the controller due to ACU adding logical drives, or changing
2205          * a logical drive's size, etc.  Reregister any new/changed drives
2206          */
2207         for (i = 0; i < num_luns; i++) {
2208                 int j;
2209
2210                 drv_found = 0;
2211
2212                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2213                 /* Find if the LUN is already in the drive array
2214                  * of the driver.  If so then update its info
2215                  * if not in use.  If it does not exist then find
2216                  * the first free index and add it.
2217                  */
2218                 for (j = 0; j <= h->highest_lun; j++) {
2219                         if (h->drv[j] != NULL &&
2220                                 memcmp(h->drv[j]->LunID, lunid,
2221                                         sizeof(h->drv[j]->LunID)) == 0) {
2222                                 drv_index = j;
2223                                 drv_found = 1;
2224                                 break;
2225                         }
2226                 }
2227
2228                 /* check if the drive was found already in the array */
2229                 if (!drv_found) {
2230                         drv_index = cciss_add_gendisk(h, lunid, 0);
2231                         if (drv_index == -1)
2232                                 goto freeret;
2233                 }
2234                 cciss_update_drive_info(ctlr, drv_index, first_time,
2235                         via_ioctl);
2236         }               /* end for */
2237
2238 freeret:
2239         kfree(ld_buff);
2240         h->busy_configuring = 0;
2241         /* We return -1 here to tell the ACU that we have registered/updated
2242          * all of the drives that we can and to keep it from calling us
2243          * additional times.
2244          */
2245         return -1;
2246 mem_msg:
2247         printk(KERN_ERR "cciss: out of memory\n");
2248         h->busy_configuring = 0;
2249         goto freeret;
2250 }
2251
2252 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2253 {
2254         /* zero out the disk size info */
2255         drive_info->nr_blocks = 0;
2256         drive_info->block_size = 0;
2257         drive_info->heads = 0;
2258         drive_info->sectors = 0;
2259         drive_info->cylinders = 0;
2260         drive_info->raid_level = -1;
2261         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2262         memset(drive_info->model, 0, sizeof(drive_info->model));
2263         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2264         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2265         /*
2266          * don't clear the LUNID though, we need to remember which
2267          * one this one is.
2268          */
2269 }
2270
2271 /* This function will deregister the disk and it's queue from the
2272  * kernel.  It must be called with the controller lock held and the
2273  * drv structures busy_configuring flag set.  It's parameters are:
2274  *
2275  * disk = This is the disk to be deregistered
2276  * drv  = This is the drive_info_struct associated with the disk to be
2277  *        deregistered.  It contains information about the disk used
2278  *        by the driver.
2279  * clear_all = This flag determines whether or not the disk information
2280  *             is going to be completely cleared out and the highest_lun
2281  *             reset.  Sometimes we want to clear out information about
2282  *             the disk in preparation for re-adding it.  In this case
2283  *             the highest_lun should be left unchanged and the LunID
2284  *             should not be cleared.
2285  * via_ioctl
2286  *    This indicates whether we've reached this path via ioctl.
2287  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2288  *    If this path is reached via ioctl(), then the max_usage_count will
2289  *    be 1, as the process calling ioctl() has got to have the device open.
2290  *    If we get here via sysfs, then the max usage count will be zero.
2291 */
2292 static int deregister_disk(ctlr_info_t *h, int drv_index,
2293                            int clear_all, int via_ioctl)
2294 {
2295         int i;
2296         struct gendisk *disk;
2297         drive_info_struct *drv;
2298         int recalculate_highest_lun;
2299
2300         if (!capable(CAP_SYS_RAWIO))
2301                 return -EPERM;
2302
2303         drv = h->drv[drv_index];
2304         disk = h->gendisk[drv_index];
2305
2306         /* make sure logical volume is NOT is use */
2307         if (clear_all || (h->gendisk[0] == disk)) {
2308                 if (drv->usage_count > via_ioctl)
2309                         return -EBUSY;
2310         } else if (drv->usage_count > 0)
2311                 return -EBUSY;
2312
2313         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2314
2315         /* invalidate the devices and deregister the disk.  If it is disk
2316          * zero do not deregister it but just zero out it's values.  This
2317          * allows us to delete disk zero but keep the controller registered.
2318          */
2319         if (h->gendisk[0] != disk) {
2320                 struct request_queue *q = disk->queue;
2321                 if (disk->flags & GENHD_FL_UP) {
2322                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2323                         del_gendisk(disk);
2324                 }
2325                 if (q)
2326                         blk_cleanup_queue(q);
2327                 /* If clear_all is set then we are deleting the logical
2328                  * drive, not just refreshing its info.  For drives
2329                  * other than disk 0 we will call put_disk.  We do not
2330                  * do this for disk 0 as we need it to be able to
2331                  * configure the controller.
2332                  */
2333                 if (clear_all){
2334                         /* This isn't pretty, but we need to find the
2335                          * disk in our array and NULL our the pointer.
2336                          * This is so that we will call alloc_disk if
2337                          * this index is used again later.
2338                          */
2339                         for (i=0; i < CISS_MAX_LUN; i++){
2340                                 if (h->gendisk[i] == disk) {
2341                                         h->gendisk[i] = NULL;
2342                                         break;
2343                                 }
2344                         }
2345                         put_disk(disk);
2346                 }
2347         } else {
2348                 set_capacity(disk, 0);
2349                 cciss_clear_drive_info(drv);
2350         }
2351
2352         --h->num_luns;
2353
2354         /* if it was the last disk, find the new hightest lun */
2355         if (clear_all && recalculate_highest_lun) {
2356                 int i, newhighest = -1;
2357                 for (i = 0; i <= h->highest_lun; i++) {
2358                         /* if the disk has size > 0, it is available */
2359                         if (h->drv[i] && h->drv[i]->heads)
2360                                 newhighest = i;
2361                 }
2362                 h->highest_lun = newhighest;
2363         }
2364         return 0;
2365 }
2366
2367 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2368                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2369                 int cmd_type)
2370 {
2371         ctlr_info_t *h = hba[ctlr];
2372         u64bit buff_dma_handle;
2373         int status = IO_OK;
2374
2375         c->cmd_type = CMD_IOCTL_PEND;
2376         c->Header.ReplyQueue = 0;
2377         if (buff != NULL) {
2378                 c->Header.SGList = 1;
2379                 c->Header.SGTotal = 1;
2380         } else {
2381                 c->Header.SGList = 0;
2382                 c->Header.SGTotal = 0;
2383         }
2384         c->Header.Tag.lower = c->busaddr;
2385         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2386
2387         c->Request.Type.Type = cmd_type;
2388         if (cmd_type == TYPE_CMD) {
2389                 switch (cmd) {
2390                 case CISS_INQUIRY:
2391                         /* are we trying to read a vital product page */
2392                         if (page_code != 0) {
2393                                 c->Request.CDB[1] = 0x01;
2394                                 c->Request.CDB[2] = page_code;
2395                         }
2396                         c->Request.CDBLen = 6;
2397                         c->Request.Type.Attribute = ATTR_SIMPLE;
2398                         c->Request.Type.Direction = XFER_READ;
2399                         c->Request.Timeout = 0;
2400                         c->Request.CDB[0] = CISS_INQUIRY;
2401                         c->Request.CDB[4] = size & 0xFF;
2402                         break;
2403                 case CISS_REPORT_LOG:
2404                 case CISS_REPORT_PHYS:
2405                         /* Talking to controller so It's a physical command
2406                            mode = 00 target = 0.  Nothing to write.
2407                          */
2408                         c->Request.CDBLen = 12;
2409                         c->Request.Type.Attribute = ATTR_SIMPLE;
2410                         c->Request.Type.Direction = XFER_READ;
2411                         c->Request.Timeout = 0;
2412                         c->Request.CDB[0] = cmd;
2413                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
2414                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2415                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2416                         c->Request.CDB[9] = size & 0xFF;
2417                         break;
2418
2419                 case CCISS_READ_CAPACITY:
2420                         c->Request.CDBLen = 10;
2421                         c->Request.Type.Attribute = ATTR_SIMPLE;
2422                         c->Request.Type.Direction = XFER_READ;
2423                         c->Request.Timeout = 0;
2424                         c->Request.CDB[0] = cmd;
2425                         break;
2426                 case CCISS_READ_CAPACITY_16:
2427                         c->Request.CDBLen = 16;
2428                         c->Request.Type.Attribute = ATTR_SIMPLE;
2429                         c->Request.Type.Direction = XFER_READ;
2430                         c->Request.Timeout = 0;
2431                         c->Request.CDB[0] = cmd;
2432                         c->Request.CDB[1] = 0x10;
2433                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2434                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2435                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2436                         c->Request.CDB[13] = size & 0xFF;
2437                         c->Request.Timeout = 0;
2438                         c->Request.CDB[0] = cmd;
2439                         break;
2440                 case CCISS_CACHE_FLUSH:
2441                         c->Request.CDBLen = 12;
2442                         c->Request.Type.Attribute = ATTR_SIMPLE;
2443                         c->Request.Type.Direction = XFER_WRITE;
2444                         c->Request.Timeout = 0;
2445                         c->Request.CDB[0] = BMIC_WRITE;
2446                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2447                         break;
2448                 case TEST_UNIT_READY:
2449                         c->Request.CDBLen = 6;
2450                         c->Request.Type.Attribute = ATTR_SIMPLE;
2451                         c->Request.Type.Direction = XFER_NONE;
2452                         c->Request.Timeout = 0;
2453                         break;
2454                 default:
2455                         printk(KERN_WARNING
2456                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2457                         return IO_ERROR;
2458                 }
2459         } else if (cmd_type == TYPE_MSG) {
2460                 switch (cmd) {
2461                 case 0: /* ABORT message */
2462                         c->Request.CDBLen = 12;
2463                         c->Request.Type.Attribute = ATTR_SIMPLE;
2464                         c->Request.Type.Direction = XFER_WRITE;
2465                         c->Request.Timeout = 0;
2466                         c->Request.CDB[0] = cmd;        /* abort */
2467                         c->Request.CDB[1] = 0;  /* abort a command */
2468                         /* buff contains the tag of the command to abort */
2469                         memcpy(&c->Request.CDB[4], buff, 8);
2470                         break;
2471                 case 1: /* RESET message */
2472                         c->Request.CDBLen = 16;
2473                         c->Request.Type.Attribute = ATTR_SIMPLE;
2474                         c->Request.Type.Direction = XFER_NONE;
2475                         c->Request.Timeout = 0;
2476                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2477                         c->Request.CDB[0] = cmd;        /* reset */
2478                         c->Request.CDB[1] = 0x03;       /* reset a target */
2479                         break;
2480                 case 3: /* No-Op message */
2481                         c->Request.CDBLen = 1;
2482                         c->Request.Type.Attribute = ATTR_SIMPLE;
2483                         c->Request.Type.Direction = XFER_WRITE;
2484                         c->Request.Timeout = 0;
2485                         c->Request.CDB[0] = cmd;
2486                         break;
2487                 default:
2488                         printk(KERN_WARNING
2489                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2490                         return IO_ERROR;
2491                 }
2492         } else {
2493                 printk(KERN_WARNING
2494                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2495                 return IO_ERROR;
2496         }
2497         /* Fill in the scatter gather information */
2498         if (size > 0) {
2499                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2500                                                              buff, size,
2501                                                              PCI_DMA_BIDIRECTIONAL);
2502                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2503                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2504                 c->SG[0].Len = size;
2505                 c->SG[0].Ext = 0;       /* we are not chaining */
2506         }
2507         return status;
2508 }
2509
2510 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2511 {
2512         switch (c->err_info->ScsiStatus) {
2513         case SAM_STAT_GOOD:
2514                 return IO_OK;
2515         case SAM_STAT_CHECK_CONDITION:
2516                 switch (0xf & c->err_info->SenseInfo[2]) {
2517                 case 0: return IO_OK; /* no sense */
2518                 case 1: return IO_OK; /* recovered error */
2519                 default:
2520                         if (check_for_unit_attention(h, c))
2521                                 return IO_NEEDS_RETRY;
2522                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2523                                 "check condition, sense key = 0x%02x\n",
2524                                 h->ctlr, c->Request.CDB[0],
2525                                 c->err_info->SenseInfo[2]);
2526                 }
2527                 break;
2528         default:
2529                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2530                         "scsi status = 0x%02x\n", h->ctlr,
2531                         c->Request.CDB[0], c->err_info->ScsiStatus);
2532                 break;
2533         }
2534         return IO_ERROR;
2535 }
2536
2537 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2538 {
2539         int return_status = IO_OK;
2540
2541         if (c->err_info->CommandStatus == CMD_SUCCESS)
2542                 return IO_OK;
2543
2544         switch (c->err_info->CommandStatus) {
2545         case CMD_TARGET_STATUS:
2546                 return_status = check_target_status(h, c);
2547                 break;
2548         case CMD_DATA_UNDERRUN:
2549         case CMD_DATA_OVERRUN:
2550                 /* expected for inquiry and report lun commands */
2551                 break;
2552         case CMD_INVALID:
2553                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2554                        "reported invalid\n", c->Request.CDB[0]);
2555                 return_status = IO_ERROR;
2556                 break;
2557         case CMD_PROTOCOL_ERR:
2558                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2559                        "protocol error \n", c->Request.CDB[0]);
2560                 return_status = IO_ERROR;
2561                 break;
2562         case CMD_HARDWARE_ERR:
2563                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2564                        " hardware error\n", c->Request.CDB[0]);
2565                 return_status = IO_ERROR;
2566                 break;
2567         case CMD_CONNECTION_LOST:
2568                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2569                        "connection lost\n", c->Request.CDB[0]);
2570                 return_status = IO_ERROR;
2571                 break;
2572         case CMD_ABORTED:
2573                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2574                        "aborted\n", c->Request.CDB[0]);
2575                 return_status = IO_ERROR;
2576                 break;
2577         case CMD_ABORT_FAILED:
2578                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2579                        "abort failed\n", c->Request.CDB[0]);
2580                 return_status = IO_ERROR;
2581                 break;
2582         case CMD_UNSOLICITED_ABORT:
2583                 printk(KERN_WARNING
2584                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2585                         c->Request.CDB[0]);
2586                 return_status = IO_NEEDS_RETRY;
2587                 break;
2588         default:
2589                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2590                        "unknown status %x\n", c->Request.CDB[0],
2591                        c->err_info->CommandStatus);
2592                 return_status = IO_ERROR;
2593         }
2594         return return_status;
2595 }
2596
2597 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2598         int attempt_retry)
2599 {
2600         DECLARE_COMPLETION_ONSTACK(wait);
2601         u64bit buff_dma_handle;
2602         unsigned long flags;
2603         int return_status = IO_OK;
2604
2605 resend_cmd2:
2606         c->waiting = &wait;
2607         /* Put the request on the tail of the queue and send it */
2608         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2609         addQ(&h->reqQ, c);
2610         h->Qdepth++;
2611         start_io(h);
2612         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2613
2614         wait_for_completion(&wait);
2615
2616         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2617                 goto command_done;
2618
2619         return_status = process_sendcmd_error(h, c);
2620
2621         if (return_status == IO_NEEDS_RETRY &&
2622                 c->retry_count < MAX_CMD_RETRIES) {
2623                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2624                         c->Request.CDB[0]);
2625                 c->retry_count++;
2626                 /* erase the old error information */
2627                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2628                 return_status = IO_OK;
2629                 INIT_COMPLETION(wait);
2630                 goto resend_cmd2;
2631         }
2632
2633 command_done:
2634         /* unlock the buffers from DMA */
2635         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2636         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2637         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2638                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2639         return return_status;
2640 }
2641
2642 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2643                            __u8 page_code, unsigned char scsi3addr[],
2644                         int cmd_type)
2645 {
2646         ctlr_info_t *h = hba[ctlr];
2647         CommandList_struct *c;
2648         int return_status;
2649
2650         c = cmd_alloc(h, 0);
2651         if (!c)
2652                 return -ENOMEM;
2653         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2654                 scsi3addr, cmd_type);
2655         if (return_status == IO_OK)
2656                 return_status = sendcmd_withirq_core(h, c, 1);
2657
2658         cmd_free(h, c, 0);
2659         return return_status;
2660 }
2661
2662 static void cciss_geometry_inquiry(int ctlr, int logvol,
2663                                    sector_t total_size,
2664                                    unsigned int block_size,
2665                                    InquiryData_struct *inq_buff,
2666                                    drive_info_struct *drv)
2667 {
2668         int return_code;
2669         unsigned long t;
2670         unsigned char scsi3addr[8];
2671
2672         memset(inq_buff, 0, sizeof(InquiryData_struct));
2673         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2674         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2675                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2676         if (return_code == IO_OK) {
2677                 if (inq_buff->data_byte[8] == 0xFF) {
2678                         printk(KERN_WARNING
2679                                "cciss: reading geometry failed, volume "
2680                                "does not support reading geometry\n");
2681                         drv->heads = 255;
2682                         drv->sectors = 32;      // Sectors per track
2683                         drv->cylinders = total_size + 1;
2684                         drv->raid_level = RAID_UNKNOWN;
2685                 } else {
2686                         drv->heads = inq_buff->data_byte[6];
2687                         drv->sectors = inq_buff->data_byte[7];
2688                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2689                         drv->cylinders += inq_buff->data_byte[5];
2690                         drv->raid_level = inq_buff->data_byte[8];
2691                 }
2692                 drv->block_size = block_size;
2693                 drv->nr_blocks = total_size + 1;
2694                 t = drv->heads * drv->sectors;
2695                 if (t > 1) {
2696                         sector_t real_size = total_size + 1;
2697                         unsigned long rem = sector_div(real_size, t);
2698                         if (rem)
2699                                 real_size++;
2700                         drv->cylinders = real_size;
2701                 }
2702         } else {                /* Get geometry failed */
2703                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2704         }
2705 }
2706
2707 static void
2708 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2709                     unsigned int *block_size)
2710 {
2711         ReadCapdata_struct *buf;
2712         int return_code;
2713         unsigned char scsi3addr[8];
2714
2715         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2716         if (!buf) {
2717                 printk(KERN_WARNING "cciss: out of memory\n");
2718                 return;
2719         }
2720
2721         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2722         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2723                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2724         if (return_code == IO_OK) {
2725                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2726                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2727         } else {                /* read capacity command failed */
2728                 printk(KERN_WARNING "cciss: read capacity failed\n");
2729                 *total_size = 0;
2730                 *block_size = BLOCK_SIZE;
2731         }
2732         kfree(buf);
2733 }
2734
2735 static void cciss_read_capacity_16(int ctlr, int logvol,
2736         sector_t *total_size, unsigned int *block_size)
2737 {
2738         ReadCapdata_struct_16 *buf;
2739         int return_code;
2740         unsigned char scsi3addr[8];
2741
2742         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2743         if (!buf) {
2744                 printk(KERN_WARNING "cciss: out of memory\n");
2745                 return;
2746         }
2747
2748         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2749         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2750                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2751                         0, scsi3addr, TYPE_CMD);
2752         if (return_code == IO_OK) {
2753                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2754                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2755         } else {                /* read capacity command failed */
2756                 printk(KERN_WARNING "cciss: read capacity failed\n");
2757                 *total_size = 0;
2758                 *block_size = BLOCK_SIZE;
2759         }
2760         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2761                (unsigned long long)*total_size+1, *block_size);
2762         kfree(buf);
2763 }
2764
2765 static int cciss_revalidate(struct gendisk *disk)
2766 {
2767         ctlr_info_t *h = get_host(disk);
2768         drive_info_struct *drv = get_drv(disk);
2769         int logvol;
2770         int FOUND = 0;
2771         unsigned int block_size;
2772         sector_t total_size;
2773         InquiryData_struct *inq_buff = NULL;
2774
2775         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2776                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2777                         sizeof(drv->LunID)) == 0) {
2778                         FOUND = 1;
2779                         break;
2780                 }
2781         }
2782
2783         if (!FOUND)
2784                 return 1;
2785
2786         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2787         if (inq_buff == NULL) {
2788                 printk(KERN_WARNING "cciss: out of memory\n");
2789                 return 1;
2790         }
2791         if (h->cciss_read == CCISS_READ_10) {
2792                 cciss_read_capacity(h->ctlr, logvol,
2793                                         &total_size, &block_size);
2794         } else {
2795                 cciss_read_capacity_16(h->ctlr, logvol,
2796                                         &total_size, &block_size);
2797         }
2798         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2799                                inq_buff, drv);
2800
2801         blk_queue_logical_block_size(drv->queue, drv->block_size);
2802         set_capacity(disk, drv->nr_blocks);
2803
2804         kfree(inq_buff);
2805         return 0;
2806 }
2807
2808 /*
2809  * Map (physical) PCI mem into (virtual) kernel space
2810  */
2811 static void __iomem *remap_pci_mem(ulong base, ulong size)
2812 {
2813         ulong page_base = ((ulong) base) & PAGE_MASK;
2814         ulong page_offs = ((ulong) base) - page_base;
2815         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2816
2817         return page_remapped ? (page_remapped + page_offs) : NULL;
2818 }
2819
2820 /*
2821  * Takes jobs of the Q and sends them to the hardware, then puts it on
2822  * the Q to wait for completion.
2823  */
2824 static void start_io(ctlr_info_t *h)
2825 {
2826         CommandList_struct *c;
2827
2828         while (!hlist_empty(&h->reqQ)) {
2829                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2830                 /* can't do anything if fifo is full */
2831                 if ((h->access.fifo_full(h))) {
2832                         printk(KERN_WARNING "cciss: fifo full\n");
2833                         break;
2834                 }
2835
2836                 /* Get the first entry from the Request Q */
2837                 removeQ(c);
2838                 h->Qdepth--;
2839
2840                 /* Tell the controller execute command */
2841                 h->access.submit_command(h, c);
2842
2843                 /* Put job onto the completed Q */
2844                 addQ(&h->cmpQ, c);
2845         }
2846 }
2847
2848 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2849 /* Zeros out the error record and then resends the command back */
2850 /* to the controller */
2851 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2852 {
2853         /* erase the old error information */
2854         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2855
2856         /* add it to software queue and then send it to the controller */
2857         addQ(&h->reqQ, c);
2858         h->Qdepth++;
2859         if (h->Qdepth > h->maxQsinceinit)
2860                 h->maxQsinceinit = h->Qdepth;
2861
2862         start_io(h);
2863 }
2864
2865 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2866         unsigned int msg_byte, unsigned int host_byte,
2867         unsigned int driver_byte)
2868 {
2869         /* inverse of macros in scsi.h */
2870         return (scsi_status_byte & 0xff) |
2871                 ((msg_byte & 0xff) << 8) |
2872                 ((host_byte & 0xff) << 16) |
2873                 ((driver_byte & 0xff) << 24);
2874 }
2875
2876 static inline int evaluate_target_status(ctlr_info_t *h,
2877                         CommandList_struct *cmd, int *retry_cmd)
2878 {
2879         unsigned char sense_key;
2880         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2881         int error_value;
2882
2883         *retry_cmd = 0;
2884         /* If we get in here, it means we got "target status", that is, scsi status */
2885         status_byte = cmd->err_info->ScsiStatus;
2886         driver_byte = DRIVER_OK;
2887         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2888
2889         if (blk_pc_request(cmd->rq))
2890                 host_byte = DID_PASSTHROUGH;
2891         else
2892                 host_byte = DID_OK;
2893
2894         error_value = make_status_bytes(status_byte, msg_byte,
2895                 host_byte, driver_byte);
2896
2897         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2898                 if (!blk_pc_request(cmd->rq))
2899                         printk(KERN_WARNING "cciss: cmd %p "
2900                                "has SCSI Status 0x%x\n",
2901                                cmd, cmd->err_info->ScsiStatus);
2902                 return error_value;
2903         }
2904
2905         /* check the sense key */
2906         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2907         /* no status or recovered error */
2908         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2909                 error_value = 0;
2910
2911         if (check_for_unit_attention(h, cmd)) {
2912                 *retry_cmd = !blk_pc_request(cmd->rq);
2913                 return 0;
2914         }
2915
2916         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2917                 if (error_value != 0)
2918                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2919                                " sense key = 0x%x\n", cmd, sense_key);
2920                 return error_value;
2921         }
2922
2923         /* SG_IO or similar, copy sense data back */
2924         if (cmd->rq->sense) {
2925                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2926                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2927                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2928                         cmd->rq->sense_len);
2929         } else
2930                 cmd->rq->sense_len = 0;
2931
2932         return error_value;
2933 }
2934
2935 /* checks the status of the job and calls complete buffers to mark all
2936  * buffers for the completed job. Note that this function does not need
2937  * to hold the hba/queue lock.
2938  */
2939 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2940                                     int timeout)
2941 {
2942         int retry_cmd = 0;
2943         struct request *rq = cmd->rq;
2944
2945         rq->errors = 0;
2946
2947         if (timeout)
2948                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
2949
2950         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
2951                 goto after_error_processing;
2952
2953         switch (cmd->err_info->CommandStatus) {
2954         case CMD_TARGET_STATUS:
2955                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
2956                 break;
2957         case CMD_DATA_UNDERRUN:
2958                 if (blk_fs_request(cmd->rq)) {
2959                         printk(KERN_WARNING "cciss: cmd %p has"
2960                                " completed with data underrun "
2961                                "reported\n", cmd);
2962                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
2963                 }
2964                 break;
2965         case CMD_DATA_OVERRUN:
2966                 if (blk_fs_request(cmd->rq))
2967                         printk(KERN_WARNING "cciss: cmd %p has"
2968                                " completed with data overrun "
2969                                "reported\n", cmd);
2970                 break;
2971         case CMD_INVALID:
2972                 printk(KERN_WARNING "cciss: cmd %p is "
2973                        "reported invalid\n", cmd);
2974                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2975                         cmd->err_info->CommandStatus, DRIVER_OK,
2976                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2977                 break;
2978         case CMD_PROTOCOL_ERR:
2979                 printk(KERN_WARNING "cciss: cmd %p has "
2980                        "protocol error \n", cmd);
2981                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2982                         cmd->err_info->CommandStatus, DRIVER_OK,
2983                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2984                 break;
2985         case CMD_HARDWARE_ERR:
2986                 printk(KERN_WARNING "cciss: cmd %p had "
2987                        " hardware error\n", cmd);
2988                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2989                         cmd->err_info->CommandStatus, DRIVER_OK,
2990                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2991                 break;
2992         case CMD_CONNECTION_LOST:
2993                 printk(KERN_WARNING "cciss: cmd %p had "
2994                        "connection lost\n", cmd);
2995                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2996                         cmd->err_info->CommandStatus, DRIVER_OK,
2997                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2998                 break;
2999         case CMD_ABORTED:
3000                 printk(KERN_WARNING "cciss: cmd %p was "
3001                        "aborted\n", cmd);
3002                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3003                         cmd->err_info->CommandStatus, DRIVER_OK,
3004                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3005                 break;
3006         case CMD_ABORT_FAILED:
3007                 printk(KERN_WARNING "cciss: cmd %p reports "
3008                        "abort failed\n", cmd);
3009                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3010                         cmd->err_info->CommandStatus, DRIVER_OK,
3011                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3012                 break;
3013         case CMD_UNSOLICITED_ABORT:
3014                 printk(KERN_WARNING "cciss%d: unsolicited "
3015                        "abort %p\n", h->ctlr, cmd);
3016                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3017                         retry_cmd = 1;
3018                         printk(KERN_WARNING
3019                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3020                         cmd->retry_count++;
3021                 } else
3022                         printk(KERN_WARNING
3023                                "cciss%d: %p retried too "
3024                                "many times\n", h->ctlr, cmd);
3025                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3026                         cmd->err_info->CommandStatus, DRIVER_OK,
3027                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3028                 break;
3029         case CMD_TIMEOUT:
3030                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3031                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3032                         cmd->err_info->CommandStatus, DRIVER_OK,
3033                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3034                 break;
3035         default:
3036                 printk(KERN_WARNING "cciss: cmd %p returned "
3037                        "unknown status %x\n", cmd,
3038                        cmd->err_info->CommandStatus);
3039                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3040                         cmd->err_info->CommandStatus, DRIVER_OK,
3041                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3042         }
3043
3044 after_error_processing:
3045
3046         /* We need to return this command */
3047         if (retry_cmd) {
3048                 resend_cciss_cmd(h, cmd);
3049                 return;
3050         }
3051         cmd->rq->completion_data = cmd;
3052         blk_complete_request(cmd->rq);
3053 }
3054
3055 /*
3056  * Get a request and submit it to the controller.
3057  */
3058 static void do_cciss_request(struct request_queue *q)
3059 {
3060         ctlr_info_t *h = q->queuedata;
3061         CommandList_struct *c;
3062         sector_t start_blk;
3063         int seg;
3064         struct request *creq;
3065         u64bit temp64;
3066         struct scatterlist tmp_sg[MAXSGENTRIES];
3067         drive_info_struct *drv;
3068         int i, dir;
3069
3070         /* We call start_io here in case there is a command waiting on the
3071          * queue that has not been sent.
3072          */
3073         if (blk_queue_plugged(q))
3074                 goto startio;
3075
3076       queue:
3077         creq = blk_peek_request(q);
3078         if (!creq)
3079                 goto startio;
3080
3081         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
3082
3083         if ((c = cmd_alloc(h, 1)) == NULL)
3084                 goto full;
3085
3086         blk_start_request(creq);
3087
3088         spin_unlock_irq(q->queue_lock);
3089
3090         c->cmd_type = CMD_RWREQ;
3091         c->rq = creq;
3092
3093         /* fill in the request */
3094         drv = creq->rq_disk->private_data;
3095         c->Header.ReplyQueue = 0;       // unused in simple mode
3096         /* got command from pool, so use the command block index instead */
3097         /* for direct lookups. */
3098         /* The first 2 bits are reserved for controller error reporting. */
3099         c->Header.Tag.lower = (c->cmdindex << 3);
3100         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
3101         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3102         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
3103         c->Request.Type.Type = TYPE_CMD;        // It is a command.
3104         c->Request.Type.Attribute = ATTR_SIMPLE;
3105         c->Request.Type.Direction =
3106             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3107         c->Request.Timeout = 0; // Don't time out
3108         c->Request.CDB[0] =
3109             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3110         start_blk = blk_rq_pos(creq);
3111 #ifdef CCISS_DEBUG
3112         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3113                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3114 #endif                          /* CCISS_DEBUG */
3115
3116         sg_init_table(tmp_sg, MAXSGENTRIES);
3117         seg = blk_rq_map_sg(q, creq, tmp_sg);
3118
3119         /* get the DMA records for the setup */
3120         if (c->Request.Type.Direction == XFER_READ)
3121                 dir = PCI_DMA_FROMDEVICE;
3122         else
3123                 dir = PCI_DMA_TODEVICE;
3124
3125         for (i = 0; i < seg; i++) {
3126                 c->SG[i].Len = tmp_sg[i].length;
3127                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3128                                                   tmp_sg[i].offset,
3129                                                   tmp_sg[i].length, dir);
3130                 c->SG[i].Addr.lower = temp64.val32.lower;
3131                 c->SG[i].Addr.upper = temp64.val32.upper;
3132                 c->SG[i].Ext = 0;       // we are not chaining
3133         }
3134         /* track how many SG entries we are using */
3135         if (seg > h->maxSG)
3136                 h->maxSG = seg;
3137
3138 #ifdef CCISS_DEBUG
3139         printk(KERN_DEBUG "cciss: Submitting %u sectors in %d segments\n",
3140                blk_rq_sectors(creq), seg);
3141 #endif                          /* CCISS_DEBUG */
3142
3143         c->Header.SGList = c->Header.SGTotal = seg;
3144         if (likely(blk_fs_request(creq))) {
3145                 if(h->cciss_read == CCISS_READ_10) {
3146                         c->Request.CDB[1] = 0;
3147                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
3148                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3149                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3150                         c->Request.CDB[5] = start_blk & 0xff;
3151                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
3152                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3153                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3154                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3155                 } else {
3156                         u32 upper32 = upper_32_bits(start_blk);
3157
3158                         c->Request.CDBLen = 16;
3159                         c->Request.CDB[1]= 0;
3160                         c->Request.CDB[2]= (upper32 >> 24) & 0xff;      //MSB
3161                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3162                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3163                         c->Request.CDB[5]= upper32 & 0xff;
3164                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3165                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3166                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3167                         c->Request.CDB[9]= start_blk & 0xff;
3168                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3169                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3170                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3171                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3172                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3173                 }
3174         } else if (blk_pc_request(creq)) {
3175                 c->Request.CDBLen = creq->cmd_len;
3176                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3177         } else {
3178                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3179                 BUG();
3180         }
3181
3182         spin_lock_irq(q->queue_lock);
3183
3184         addQ(&h->reqQ, c);
3185         h->Qdepth++;
3186         if (h->Qdepth > h->maxQsinceinit)
3187                 h->maxQsinceinit = h->Qdepth;
3188
3189         goto queue;
3190 full:
3191         blk_stop_queue(q);
3192 startio:
3193         /* We will already have the driver lock here so not need
3194          * to lock it.
3195          */
3196         start_io(h);
3197 }
3198
3199 static inline unsigned long get_next_completion(ctlr_info_t *h)
3200 {
3201         return h->access.command_completed(h);
3202 }
3203
3204 static inline int interrupt_pending(ctlr_info_t *h)
3205 {
3206         return h->access.intr_pending(h);
3207 }
3208
3209 static inline long interrupt_not_for_us(ctlr_info_t *h)
3210 {
3211         return (((h->access.intr_pending(h) == 0) ||
3212                  (h->interrupts_enabled == 0)));
3213 }
3214
3215 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3216 {
3217         ctlr_info_t *h = dev_id;
3218         CommandList_struct *c;
3219         unsigned long flags;
3220         __u32 a, a1, a2;
3221
3222         if (interrupt_not_for_us(h))
3223                 return IRQ_NONE;
3224         /*
3225          * If there are completed commands in the completion queue,
3226          * we had better do something about it.
3227          */
3228         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3229         while (interrupt_pending(h)) {
3230                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3231                         a1 = a;
3232                         if ((a & 0x04)) {
3233                                 a2 = (a >> 3);
3234                                 if (a2 >= h->nr_cmds) {
3235                                         printk(KERN_WARNING
3236                                                "cciss: controller cciss%d failed, stopping.\n",
3237                                                h->ctlr);
3238                                         fail_all_cmds(h->ctlr);
3239                                         return IRQ_HANDLED;
3240                                 }
3241
3242                                 c = h->cmd_pool + a2;
3243                                 a = c->busaddr;
3244
3245                         } else {
3246                                 struct hlist_node *tmp;
3247
3248                                 a &= ~3;
3249                                 c = NULL;
3250                                 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3251                                         if (c->busaddr == a)
3252                                                 break;
3253                                 }
3254                         }
3255                         /*
3256                          * If we've found the command, take it off the
3257                          * completion Q and free it
3258                          */
3259                         if (c && c->busaddr == a) {
3260                                 removeQ(c);
3261                                 if (c->cmd_type == CMD_RWREQ) {
3262                                         complete_command(h, c, 0);
3263                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3264                                         complete(c->waiting);
3265                                 }
3266 #                               ifdef CONFIG_CISS_SCSI_TAPE
3267                                 else if (c->cmd_type == CMD_SCSI)
3268                                         complete_scsi_command(c, 0, a1);
3269 #                               endif
3270                                 continue;
3271                         }
3272                 }
3273         }
3274
3275         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3276         return IRQ_HANDLED;
3277 }
3278
3279 /**
3280  * add_to_scan_list() - add controller to rescan queue
3281  * @h:                Pointer to the controller.
3282  *
3283  * Adds the controller to the rescan queue if not already on the queue.
3284  *
3285  * returns 1 if added to the queue, 0 if skipped (could be on the
3286  * queue already, or the controller could be initializing or shutting
3287  * down).
3288  **/
3289 static int add_to_scan_list(struct ctlr_info *h)
3290 {
3291         struct ctlr_info *test_h;
3292         int found = 0;
3293         int ret = 0;
3294
3295         if (h->busy_initializing)
3296                 return 0;
3297
3298         if (!mutex_trylock(&h->busy_shutting_down))
3299                 return 0;
3300
3301         mutex_lock(&scan_mutex);
3302         list_for_each_entry(test_h, &scan_q, scan_list) {
3303                 if (test_h == h) {
3304                         found = 1;
3305                         break;
3306                 }
3307         }
3308         if (!found && !h->busy_scanning) {
3309                 INIT_COMPLETION(h->scan_wait);
3310                 list_add_tail(&h->scan_list, &scan_q);
3311                 ret = 1;
3312         }
3313         mutex_unlock(&scan_mutex);
3314         mutex_unlock(&h->busy_shutting_down);
3315
3316         return ret;
3317 }
3318
3319 /**
3320  * remove_from_scan_list() - remove controller from rescan queue
3321  * @h:                     Pointer to the controller.
3322  *
3323  * Removes the controller from the rescan queue if present. Blocks if
3324  * the controller is currently conducting a rescan.  The controller
3325  * can be in one of three states:
3326  * 1. Doesn't need a scan
3327  * 2. On the scan list, but not scanning yet (we remove it)
3328  * 3. Busy scanning (and not on the list). In this case we want to wait for
3329  *    the scan to complete to make sure the scanning thread for this
3330  *    controller is completely idle.
3331  **/
3332 static void remove_from_scan_list(struct ctlr_info *h)
3333 {
3334         struct ctlr_info *test_h, *tmp_h;
3335
3336         mutex_lock(&scan_mutex);
3337         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3338                 if (test_h == h) { /* state 2. */
3339                         list_del(&h->scan_list);
3340                         complete_all(&h->scan_wait);
3341                         mutex_unlock(&scan_mutex);
3342                         return;
3343                 }
3344         }
3345         if (h->busy_scanning) { /* state 3. */
3346                 mutex_unlock(&scan_mutex);
3347                 wait_for_completion(&h->scan_wait);
3348         } else { /* state 1, nothing to do. */
3349                 mutex_unlock(&scan_mutex);
3350         }
3351 }
3352
3353 /**
3354  * scan_thread() - kernel thread used to rescan controllers
3355  * @data:        Ignored.
3356  *
3357  * A kernel thread used scan for drive topology changes on
3358  * controllers. The thread processes only one controller at a time
3359  * using a queue.  Controllers are added to the queue using
3360  * add_to_scan_list() and removed from the queue either after done
3361  * processing or using remove_from_scan_list().
3362  *
3363  * returns 0.
3364  **/
3365 static int scan_thread(void *data)
3366 {
3367         struct ctlr_info *h;
3368
3369         while (1) {
3370                 set_current_state(TASK_INTERRUPTIBLE);
3371                 schedule();
3372                 if (kthread_should_stop())
3373                         break;
3374
3375                 while (1) {
3376                         mutex_lock(&scan_mutex);
3377                         if (list_empty(&scan_q)) {
3378                                 mutex_unlock(&scan_mutex);
3379                                 break;
3380                         }
3381
3382                         h = list_entry(scan_q.next,
3383                                        struct ctlr_info,
3384                                        scan_list);
3385                         list_del(&h->scan_list);
3386                         h->busy_scanning = 1;
3387                         mutex_unlock(&scan_mutex);
3388
3389                         rebuild_lun_table(h, 0, 0);
3390                         complete_all(&h->scan_wait);
3391                         mutex_lock(&scan_mutex);
3392                         h->busy_scanning = 0;
3393                         mutex_unlock(&scan_mutex);
3394                 }
3395         }
3396
3397         return 0;
3398 }
3399
3400 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3401 {
3402         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3403                 return 0;
3404
3405         switch (c->err_info->SenseInfo[12]) {
3406         case STATE_CHANGED:
3407                 printk(KERN_WARNING "cciss%d: a state change "
3408                         "detected, command retried\n", h->ctlr);
3409                 return 1;
3410         break;
3411         case LUN_FAILED:
3412                 printk(KERN_WARNING "cciss%d: LUN failure "
3413                         "detected, action required\n", h->ctlr);
3414                 return 1;
3415         break;
3416         case REPORT_LUNS_CHANGED:
3417                 printk(KERN_WARNING "cciss%d: report LUN data "
3418                         "changed\n", h->ctlr);
3419                 add_to_scan_list(h);
3420                 wake_up_process(cciss_scan_thread);
3421                 return 1;
3422         break;
3423         case POWER_OR_RESET:
3424                 printk(KERN_WARNING "cciss%d: a power on "
3425                         "or device reset detected\n", h->ctlr);
3426                 return 1;
3427         break;
3428         case UNIT_ATTENTION_CLEARED:
3429                 printk(KERN_WARNING "cciss%d: unit attention "
3430                     "cleared by another initiator\n", h->ctlr);
3431                 return 1;
3432         break;
3433         default:
3434                 printk(KERN_WARNING "cciss%d: unknown "
3435                         "unit attention detected\n", h->ctlr);
3436                                 return 1;
3437         }
3438 }
3439
3440 /*
3441  *  We cannot read the structure directly, for portability we must use
3442  *   the io functions.
3443  *   This is for debug only.
3444  */
3445 #ifdef CCISS_DEBUG
3446 static void print_cfg_table(CfgTable_struct *tb)
3447 {
3448         int i;
3449         char temp_name[17];
3450
3451         printk("Controller Configuration information\n");
3452         printk("------------------------------------\n");
3453         for (i = 0; i < 4; i++)
3454                 temp_name[i] = readb(&(tb->Signature[i]));
3455         temp_name[4] = '\0';
3456         printk("   Signature = %s\n", temp_name);
3457         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3458         printk("   Transport methods supported = 0x%x\n",
3459                readl(&(tb->TransportSupport)));
3460         printk("   Transport methods active = 0x%x\n",
3461                readl(&(tb->TransportActive)));
3462         printk("   Requested transport Method = 0x%x\n",
3463                readl(&(tb->HostWrite.TransportRequest)));
3464         printk("   Coalesce Interrupt Delay = 0x%x\n",
3465                readl(&(tb->HostWrite.CoalIntDelay)));
3466         printk("   Coalesce Interrupt Count = 0x%x\n",
3467                readl(&(tb->HostWrite.CoalIntCount)));
3468         printk("   Max outstanding commands = 0x%d\n",
3469                readl(&(tb->CmdsOutMax)));
3470         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3471         for (i = 0; i < 16; i++)
3472                 temp_name[i] = readb(&(tb->ServerName[i]));
3473         temp_name[16] = '\0';
3474         printk("   Server Name = %s\n", temp_name);
3475         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3476 }
3477 #endif                          /* CCISS_DEBUG */
3478
3479 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3480 {
3481         int i, offset, mem_type, bar_type;
3482         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3483                 return 0;
3484         offset = 0;
3485         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3486                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3487                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3488                         offset += 4;
3489                 else {
3490                         mem_type = pci_resource_flags(pdev, i) &
3491                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3492                         switch (mem_type) {
3493                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3494                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3495                                 offset += 4;    /* 32 bit */
3496                                 break;
3497                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3498                                 offset += 8;
3499                                 break;
3500                         default:        /* reserved in PCI 2.2 */
3501                                 printk(KERN_WARNING
3502                                        "Base address is invalid\n");
3503                                 return -1;
3504                                 break;
3505                         }
3506                 }
3507                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3508                         return i + 1;
3509         }
3510         return -1;
3511 }
3512
3513 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3514  * controllers that are capable. If not, we use IO-APIC mode.
3515  */
3516
3517 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3518                                            struct pci_dev *pdev, __u32 board_id)
3519 {
3520 #ifdef CONFIG_PCI_MSI
3521         int err;
3522         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3523         {0, 2}, {0, 3}
3524         };
3525
3526         /* Some boards advertise MSI but don't really support it */
3527         if ((board_id == 0x40700E11) ||
3528             (board_id == 0x40800E11) ||
3529             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3530                 goto default_int_mode;
3531
3532         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3533                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3534                 if (!err) {
3535                         c->intr[0] = cciss_msix_entries[0].vector;
3536                         c->intr[1] = cciss_msix_entries[1].vector;
3537                         c->intr[2] = cciss_msix_entries[2].vector;
3538                         c->intr[3] = cciss_msix_entries[3].vector;
3539                         c->msix_vector = 1;
3540                         return;
3541                 }
3542                 if (err > 0) {
3543                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3544                                "available\n", err);
3545                         goto default_int_mode;
3546                 } else {
3547                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3548                                err);
3549                         goto default_int_mode;
3550                 }
3551         }
3552         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3553                 if (!pci_enable_msi(pdev)) {
3554                         c->msi_vector = 1;
3555                 } else {
3556                         printk(KERN_WARNING "cciss: MSI init failed\n");
3557                 }
3558         }
3559 default_int_mode:
3560 #endif                          /* CONFIG_PCI_MSI */
3561         /* if we get here we're going to use the default interrupt mode */
3562         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3563         return;
3564 }
3565
3566 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3567 {
3568         ushort subsystem_vendor_id, subsystem_device_id, command;
3569         __u32 board_id, scratchpad = 0;
3570         __u64 cfg_offset;
3571         __u32 cfg_base_addr;
3572         __u64 cfg_base_addr_index;
3573         int i, prod_index, err;
3574
3575         subsystem_vendor_id = pdev->subsystem_vendor;
3576         subsystem_device_id = pdev->subsystem_device;
3577         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3578                     subsystem_vendor_id);
3579
3580         for (i = 0; i < ARRAY_SIZE(products); i++) {
3581                 /* Stand aside for hpsa driver on request */
3582                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3583                         return -ENODEV;
3584                 if (board_id == products[i].board_id)
3585                         break;
3586         }
3587         prod_index = i;
3588         if (prod_index == ARRAY_SIZE(products)) {
3589                 dev_warn(&pdev->dev,
3590                         "unrecognized board ID: 0x%08lx, ignoring.\n",
3591                         (unsigned long) board_id);
3592                 return -ENODEV;
3593         }
3594
3595         /* check to see if controller has been disabled */
3596         /* BEFORE trying to enable it */
3597         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3598         if (!(command & 0x02)) {
3599                 printk(KERN_WARNING
3600                        "cciss: controller appears to be disabled\n");
3601                 return -ENODEV;
3602         }
3603
3604         err = pci_enable_device(pdev);
3605         if (err) {
3606                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3607                 return err;
3608         }
3609
3610         err = pci_request_regions(pdev, "cciss");
3611         if (err) {
3612                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3613                        "aborting\n");
3614                 return err;
3615         }
3616
3617 #ifdef CCISS_DEBUG
3618         printk("command = %x\n", command);
3619         printk("irq = %x\n", pdev->irq);
3620         printk("board_id = %x\n", board_id);
3621 #endif                          /* CCISS_DEBUG */
3622
3623 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3624  * else we use the IO-APIC interrupt assigned to us by system ROM.
3625  */
3626         cciss_interrupt_mode(c, pdev, board_id);
3627
3628         /* find the memory BAR */
3629         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3630                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3631                         break;
3632         }
3633         if (i == DEVICE_COUNT_RESOURCE) {
3634                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3635                 err = -ENODEV;
3636                 goto err_out_free_res;
3637         }
3638
3639         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3640                                                  * already removed
3641                                                  */
3642
3643 #ifdef CCISS_DEBUG
3644         printk("address 0 = %lx\n", c->paddr);
3645 #endif                          /* CCISS_DEBUG */
3646         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3647
3648         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3649          * We poll for up to 120 secs, once per 100ms. */
3650         for (i = 0; i < 1200; i++) {
3651                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3652                 if (scratchpad == CCISS_FIRMWARE_READY)
3653                         break;
3654                 set_current_state(TASK_INTERRUPTIBLE);
3655                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
3656         }
3657         if (scratchpad != CCISS_FIRMWARE_READY) {
3658                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3659                 err = -ENODEV;
3660                 goto err_out_free_res;
3661         }
3662
3663         /* get the address index number */
3664         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3665         cfg_base_addr &= (__u32) 0x0000ffff;
3666 #ifdef CCISS_DEBUG
3667         printk("cfg base address = %x\n", cfg_base_addr);
3668 #endif                          /* CCISS_DEBUG */
3669         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3670 #ifdef CCISS_DEBUG
3671         printk("cfg base address index = %llx\n",
3672                 (unsigned long long)cfg_base_addr_index);
3673 #endif                          /* CCISS_DEBUG */
3674         if (cfg_base_addr_index == -1) {
3675                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3676                 err = -ENODEV;
3677                 goto err_out_free_res;
3678         }
3679
3680         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3681 #ifdef CCISS_DEBUG
3682         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3683 #endif                          /* CCISS_DEBUG */
3684         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3685                                                        cfg_base_addr_index) +
3686                                     cfg_offset, sizeof(CfgTable_struct));
3687         c->board_id = board_id;
3688
3689 #ifdef CCISS_DEBUG
3690         print_cfg_table(c->cfgtable);
3691 #endif                          /* CCISS_DEBUG */
3692
3693         /* Some controllers support Zero Memory Raid (ZMR).
3694          * When configured in ZMR mode the number of supported
3695          * commands drops to 64. So instead of just setting an
3696          * arbitrary value we make the driver a little smarter.
3697          * We read the config table to tell us how many commands
3698          * are supported on the controller then subtract 4 to
3699          * leave a little room for ioctl calls.
3700          */
3701         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3702         c->product_name = products[prod_index].product_name;
3703         c->access = *(products[prod_index].access);
3704         c->nr_cmds = c->max_commands - 4;
3705         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3706             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3707             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3708             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3709                 printk("Does not appear to be a valid CISS config table\n");
3710                 err = -ENODEV;
3711                 goto err_out_free_res;
3712         }
3713 #ifdef CONFIG_X86
3714         {
3715                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3716                 __u32 prefetch;
3717                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3718                 prefetch |= 0x100;
3719                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3720         }
3721 #endif
3722
3723         /* Disabling DMA prefetch and refetch for the P600.
3724          * An ASIC bug may result in accesses to invalid memory addresses.
3725          * We've disabled prefetch for some time now. Testing with XEN
3726          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3727          */
3728         if(board_id == 0x3225103C) {
3729                 __u32 dma_prefetch;
3730                 __u32 dma_refetch;
3731                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3732                 dma_prefetch |= 0x8000;
3733                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3734                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3735                 dma_refetch |= 0x1;
3736                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3737         }
3738
3739 #ifdef CCISS_DEBUG
3740         printk("Trying to put board into Simple mode\n");
3741 #endif                          /* CCISS_DEBUG */
3742         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3743         /* Update the field, and then ring the doorbell */
3744         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3745         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3746
3747         /* under certain very rare conditions, this can take awhile.
3748          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3749          * as we enter this code.) */
3750         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3751                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3752                         break;
3753                 /* delay and try again */
3754                 set_current_state(TASK_INTERRUPTIBLE);
3755                 schedule_timeout(msecs_to_jiffies(1));
3756         }
3757
3758 #ifdef CCISS_DEBUG
3759         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3760                readl(c->vaddr + SA5_DOORBELL));
3761 #endif                          /* CCISS_DEBUG */
3762 #ifdef CCISS_DEBUG
3763         print_cfg_table(c->cfgtable);
3764 #endif                          /* CCISS_DEBUG */
3765
3766         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3767                 printk(KERN_WARNING "cciss: unable to get board into"
3768                        " simple mode\n");
3769                 err = -ENODEV;
3770                 goto err_out_free_res;
3771         }
3772         return 0;
3773
3774 err_out_free_res:
3775         /*
3776          * Deliberately omit pci_disable_device(): it does something nasty to
3777          * Smart Array controllers that pci_enable_device does not undo
3778          */
3779         pci_release_regions(pdev);
3780         return err;
3781 }
3782
3783 /* Function to find the first free pointer into our hba[] array
3784  * Returns -1 if no free entries are left.
3785  */
3786 static int alloc_cciss_hba(void)
3787 {
3788         int i;
3789
3790         for (i = 0; i < MAX_CTLR; i++) {
3791                 if (!hba[i]) {
3792                         ctlr_info_t *p;
3793
3794                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3795                         if (!p)
3796                                 goto Enomem;
3797                         hba[i] = p;
3798                         return i;
3799                 }
3800         }
3801         printk(KERN_WARNING "cciss: This driver supports a maximum"
3802                " of %d controllers.\n", MAX_CTLR);
3803         return -1;
3804 Enomem:
3805         printk(KERN_ERR "cciss: out of memory.\n");
3806         return -1;
3807 }
3808
3809 static void free_hba(int n)
3810 {
3811         ctlr_info_t *h = hba[n];
3812         int i;
3813
3814         hba[n] = NULL;
3815         for (i = 0; i < h->highest_lun + 1; i++)
3816                 if (h->gendisk[i] != NULL)
3817                         put_disk(h->gendisk[i]);
3818         kfree(h);
3819 }
3820
3821 /* Send a message CDB to the firmware. */
3822 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3823 {
3824         typedef struct {
3825                 CommandListHeader_struct CommandHeader;
3826                 RequestBlock_struct Request;
3827                 ErrDescriptor_struct ErrorDescriptor;
3828         } Command;
3829         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3830         Command *cmd;
3831         dma_addr_t paddr64;
3832         uint32_t paddr32, tag;
3833         void __iomem *vaddr;
3834         int i, err;
3835
3836         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3837         if (vaddr == NULL)
3838                 return -ENOMEM;
3839
3840         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3841            CCISS commands, so they must be allocated from the lower 4GiB of
3842            memory. */
3843         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3844         if (err) {
3845                 iounmap(vaddr);
3846                 return -ENOMEM;
3847         }
3848
3849         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3850         if (cmd == NULL) {
3851                 iounmap(vaddr);
3852                 return -ENOMEM;
3853         }
3854
3855         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
3856            although there's no guarantee, we assume that the address is at
3857            least 4-byte aligned (most likely, it's page-aligned). */
3858         paddr32 = paddr64;
3859
3860         cmd->CommandHeader.ReplyQueue = 0;
3861         cmd->CommandHeader.SGList = 0;
3862         cmd->CommandHeader.SGTotal = 0;
3863         cmd->CommandHeader.Tag.lower = paddr32;
3864         cmd->CommandHeader.Tag.upper = 0;
3865         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
3866
3867         cmd->Request.CDBLen = 16;
3868         cmd->Request.Type.Type = TYPE_MSG;
3869         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
3870         cmd->Request.Type.Direction = XFER_NONE;
3871         cmd->Request.Timeout = 0; /* Don't time out */
3872         cmd->Request.CDB[0] = opcode;
3873         cmd->Request.CDB[1] = type;
3874         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
3875
3876         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
3877         cmd->ErrorDescriptor.Addr.upper = 0;
3878         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
3879
3880         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
3881
3882         for (i = 0; i < 10; i++) {
3883                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
3884                 if ((tag & ~3) == paddr32)
3885                         break;
3886                 schedule_timeout_uninterruptible(HZ);
3887         }
3888
3889         iounmap(vaddr);
3890
3891         /* we leak the DMA buffer here ... no choice since the controller could
3892            still complete the command. */
3893         if (i == 10) {
3894                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
3895                         opcode, type);
3896                 return -ETIMEDOUT;
3897         }
3898
3899         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
3900
3901         if (tag & 2) {
3902                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
3903                         opcode, type);
3904                 return -EIO;
3905         }
3906
3907         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
3908                 opcode, type);
3909         return 0;
3910 }
3911
3912 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
3913 #define cciss_noop(p) cciss_message(p, 3, 0)
3914
3915 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
3916 {
3917 /* the #defines are stolen from drivers/pci/msi.h. */
3918 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
3919 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
3920
3921         int pos;
3922         u16 control = 0;
3923
3924         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
3925         if (pos) {
3926                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
3927                 if (control & PCI_MSI_FLAGS_ENABLE) {
3928                         printk(KERN_INFO "cciss: resetting MSI\n");
3929                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
3930                 }
3931         }
3932
3933         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
3934         if (pos) {
3935                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
3936                 if (control & PCI_MSIX_FLAGS_ENABLE) {
3937                         printk(KERN_INFO "cciss: resetting MSI-X\n");
3938                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
3939                 }
3940         }
3941
3942         return 0;
3943 }
3944
3945 /* This does a hard reset of the controller using PCI power management
3946  * states. */
3947 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
3948 {
3949         u16 pmcsr, saved_config_space[32];
3950         int i, pos;
3951
3952         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
3953
3954         /* This is very nearly the same thing as
3955
3956            pci_save_state(pci_dev);
3957            pci_set_power_state(pci_dev, PCI_D3hot);
3958            pci_set_power_state(pci_dev, PCI_D0);
3959            pci_restore_state(pci_dev);
3960
3961            but we can't use these nice canned kernel routines on
3962            kexec, because they also check the MSI/MSI-X state in PCI
3963            configuration space and do the wrong thing when it is
3964            set/cleared.  Also, the pci_save/restore_state functions
3965            violate the ordering requirements for restoring the
3966            configuration space from the CCISS document (see the
3967            comment below).  So we roll our own .... */
3968
3969         for (i = 0; i < 32; i++)
3970                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
3971
3972         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
3973         if (pos == 0) {
3974                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
3975                 return -ENODEV;
3976         }
3977
3978         /* Quoting from the Open CISS Specification: "The Power
3979          * Management Control/Status Register (CSR) controls the power
3980          * state of the device.  The normal operating state is D0,
3981          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
3982          * the controller, place the interface device in D3 then to
3983          * D0, this causes a secondary PCI reset which will reset the
3984          * controller." */
3985
3986         /* enter the D3hot power management state */
3987         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
3988         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3989         pmcsr |= PCI_D3hot;
3990         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
3991
3992         schedule_timeout_uninterruptible(HZ >> 1);
3993
3994         /* enter the D0 power management state */
3995         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
3996         pmcsr |= PCI_D0;
3997         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
3998
3999         schedule_timeout_uninterruptible(HZ >> 1);
4000
4001         /* Restore the PCI configuration space.  The Open CISS
4002          * Specification says, "Restore the PCI Configuration
4003          * Registers, offsets 00h through 60h. It is important to
4004          * restore the command register, 16-bits at offset 04h,
4005          * last. Do not restore the configuration status register,
4006          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4007         for (i = 0; i < 32; i++) {
4008                 if (i == 2 || i == 3)
4009                         continue;
4010                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4011         }
4012         wmb();
4013         pci_write_config_word(pdev, 4, saved_config_space[2]);
4014
4015         return 0;
4016 }
4017
4018 /*
4019  *  This is it.  Find all the controllers and register them.  I really hate
4020  *  stealing all these major device numbers.
4021  *  returns the number of block devices registered.
4022  */
4023 static int __devinit cciss_init_one(struct pci_dev *pdev,
4024                                     const struct pci_device_id *ent)
4025 {
4026         int i;
4027         int j = 0;
4028         int rc;
4029         int dac, return_code;
4030         InquiryData_struct *inq_buff;
4031
4032         if (reset_devices) {
4033                 /* Reset the controller with a PCI power-cycle */
4034                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4035                         return -ENODEV;
4036
4037                 /* Now try to get the controller to respond to a no-op. Some
4038                    devices (notably the HP Smart Array 5i Controller) need
4039                    up to 30 seconds to respond. */
4040                 for (i=0; i<30; i++) {
4041                         if (cciss_noop(pdev) == 0)
4042                                 break;
4043
4044                         schedule_timeout_uninterruptible(HZ);
4045                 }
4046                 if (i == 30) {
4047                         printk(KERN_ERR "cciss: controller seems dead\n");
4048                         return -EBUSY;
4049                 }
4050         }
4051
4052         i = alloc_cciss_hba();
4053         if (i < 0)
4054                 return -1;
4055
4056         hba[i]->busy_initializing = 1;
4057         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4058         INIT_HLIST_HEAD(&hba[i]->reqQ);
4059         mutex_init(&hba[i]->busy_shutting_down);
4060
4061         if (cciss_pci_init(hba[i], pdev) != 0)
4062                 goto clean_no_release_regions;
4063
4064         sprintf(hba[i]->devname, "cciss%d", i);
4065         hba[i]->ctlr = i;
4066         hba[i]->pdev = pdev;
4067
4068         init_completion(&hba[i]->scan_wait);
4069
4070         if (cciss_create_hba_sysfs_entry(hba[i]))
4071                 goto clean0;
4072
4073         /* configure PCI DMA stuff */
4074         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4075                 dac = 1;
4076         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4077                 dac = 0;
4078         else {
4079                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4080                 goto clean1;
4081         }
4082
4083         /*
4084          * register with the major number, or get a dynamic major number
4085          * by passing 0 as argument.  This is done for greater than
4086          * 8 controller support.
4087          */
4088         if (i < MAX_CTLR_ORIG)
4089                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4090         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4091         if (rc == -EBUSY || rc == -EINVAL) {
4092                 printk(KERN_ERR
4093                        "cciss:  Unable to get major number %d for %s "
4094                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4095                 goto clean1;
4096         } else {
4097                 if (i >= MAX_CTLR_ORIG)
4098                         hba[i]->major = rc;
4099         }
4100
4101         /* make sure the board interrupts are off */
4102         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4103         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
4104                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
4105                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4106                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4107                 goto clean2;
4108         }
4109
4110         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4111                hba[i]->devname, pdev->device, pci_name(pdev),
4112                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4113
4114         hba[i]->cmd_pool_bits =
4115             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4116                         * sizeof(unsigned long), GFP_KERNEL);
4117         hba[i]->cmd_pool = (CommandList_struct *)
4118             pci_alloc_consistent(hba[i]->pdev,
4119                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4120                     &(hba[i]->cmd_pool_dhandle));
4121         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4122             pci_alloc_consistent(hba[i]->pdev,
4123                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4124                     &(hba[i]->errinfo_pool_dhandle));
4125         if ((hba[i]->cmd_pool_bits == NULL)
4126             || (hba[i]->cmd_pool == NULL)
4127             || (hba[i]->errinfo_pool == NULL)) {
4128                 printk(KERN_ERR "cciss: out of memory");
4129                 goto clean4;
4130         }
4131         spin_lock_init(&hba[i]->lock);
4132
4133         /* Initialize the pdev driver private data.
4134            have it point to hba[i].  */
4135         pci_set_drvdata(pdev, hba[i]);
4136         /* command and error info recs zeroed out before
4137            they are used */
4138         memset(hba[i]->cmd_pool_bits, 0,
4139                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4140                         * sizeof(unsigned long));
4141
4142         hba[i]->num_luns = 0;
4143         hba[i]->highest_lun = -1;
4144         for (j = 0; j < CISS_MAX_LUN; j++) {
4145                 hba[i]->drv[j] = NULL;
4146                 hba[i]->gendisk[j] = NULL;
4147         }
4148
4149         cciss_scsi_setup(i);
4150
4151         /* Turn the interrupts on so we can service requests */
4152         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4153
4154         /* Get the firmware version */
4155         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4156         if (inq_buff == NULL) {
4157                 printk(KERN_ERR "cciss: out of memory\n");
4158                 goto clean4;
4159         }
4160
4161         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4162                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4163         if (return_code == IO_OK) {
4164                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4165                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4166                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4167                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4168         } else {         /* send command failed */
4169                 printk(KERN_WARNING "cciss: unable to determine firmware"
4170                         " version of controller\n");
4171         }
4172         kfree(inq_buff);
4173
4174         cciss_procinit(i);
4175
4176         hba[i]->cciss_max_sectors = 2048;
4177
4178         rebuild_lun_table(hba[i], 1, 0);
4179         hba[i]->busy_initializing = 0;
4180         return 1;
4181
4182 clean4:
4183         kfree(hba[i]->cmd_pool_bits);
4184         if (hba[i]->cmd_pool)
4185                 pci_free_consistent(hba[i]->pdev,
4186                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4187                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4188         if (hba[i]->errinfo_pool)
4189                 pci_free_consistent(hba[i]->pdev,
4190                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4191                                     hba[i]->errinfo_pool,
4192                                     hba[i]->errinfo_pool_dhandle);
4193         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4194 clean2:
4195         unregister_blkdev(hba[i]->major, hba[i]->devname);
4196 clean1:
4197         cciss_destroy_hba_sysfs_entry(hba[i]);
4198 clean0:
4199         pci_release_regions(pdev);
4200 clean_no_release_regions:
4201         hba[i]->busy_initializing = 0;
4202
4203         /*
4204          * Deliberately omit pci_disable_device(): it does something nasty to
4205          * Smart Array controllers that pci_enable_device does not undo
4206          */
4207         pci_set_drvdata(pdev, NULL);
4208         free_hba(i);
4209         return -1;
4210 }
4211
4212 static void cciss_shutdown(struct pci_dev *pdev)
4213 {
4214         ctlr_info_t *h;
4215         char *flush_buf;
4216         int return_code;
4217
4218         h = pci_get_drvdata(pdev);
4219         flush_buf = kzalloc(4, GFP_KERNEL);
4220         if (!flush_buf) {
4221                 printk(KERN_WARNING
4222                         "cciss:%d cache not flushed, out of memory.\n",
4223                         h->ctlr);
4224                 return;
4225         }
4226         /* write all data in the battery backed cache to disk */
4227         memset(flush_buf, 0, 4);
4228         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4229                 4, 0, CTLR_LUNID, TYPE_CMD);
4230         kfree(flush_buf);
4231         if (return_code != IO_OK)
4232                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4233                         h->ctlr);
4234         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4235         free_irq(h->intr[2], h);
4236 }
4237
4238 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4239 {
4240         ctlr_info_t *tmp_ptr;
4241         int i, j;
4242
4243         if (pci_get_drvdata(pdev) == NULL) {
4244                 printk(KERN_ERR "cciss: Unable to remove device \n");
4245                 return;
4246         }
4247
4248         tmp_ptr = pci_get_drvdata(pdev);
4249         i = tmp_ptr->ctlr;
4250         if (hba[i] == NULL) {
4251                 printk(KERN_ERR "cciss: device appears to "
4252                        "already be removed \n");
4253                 return;
4254         }
4255
4256         mutex_lock(&hba[i]->busy_shutting_down);
4257
4258         remove_from_scan_list(hba[i]);
4259         remove_proc_entry(hba[i]->devname, proc_cciss);
4260         unregister_blkdev(hba[i]->major, hba[i]->devname);
4261
4262         /* remove it from the disk list */
4263         for (j = 0; j < CISS_MAX_LUN; j++) {
4264                 struct gendisk *disk = hba[i]->gendisk[j];
4265                 if (disk) {
4266                         struct request_queue *q = disk->queue;
4267
4268                         if (disk->flags & GENHD_FL_UP) {
4269                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4270                                 del_gendisk(disk);
4271                         }
4272                         if (q)
4273                                 blk_cleanup_queue(q);
4274                 }
4275         }
4276
4277 #ifdef CONFIG_CISS_SCSI_TAPE
4278         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4279 #endif
4280
4281         cciss_shutdown(pdev);
4282
4283 #ifdef CONFIG_PCI_MSI
4284         if (hba[i]->msix_vector)
4285                 pci_disable_msix(hba[i]->pdev);
4286         else if (hba[i]->msi_vector)
4287                 pci_disable_msi(hba[i]->pdev);
4288 #endif                          /* CONFIG_PCI_MSI */
4289
4290         iounmap(hba[i]->vaddr);
4291
4292         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4293                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4294         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4295                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4296         kfree(hba[i]->cmd_pool_bits);
4297         /*
4298          * Deliberately omit pci_disable_device(): it does something nasty to
4299          * Smart Array controllers that pci_enable_device does not undo
4300          */
4301         pci_release_regions(pdev);
4302         pci_set_drvdata(pdev, NULL);
4303         cciss_destroy_hba_sysfs_entry(hba[i]);
4304         mutex_unlock(&hba[i]->busy_shutting_down);
4305         free_hba(i);
4306 }
4307
4308 static struct pci_driver cciss_pci_driver = {
4309         .name = "cciss",
4310         .probe = cciss_init_one,
4311         .remove = __devexit_p(cciss_remove_one),
4312         .id_table = cciss_pci_device_id,        /* id_table */
4313         .shutdown = cciss_shutdown,
4314 };
4315
4316 /*
4317  *  This is it.  Register the PCI driver information for the cards we control
4318  *  the OS will call our registered routines when it finds one of our cards.
4319  */
4320 static int __init cciss_init(void)
4321 {
4322         int err;
4323
4324         /*
4325          * The hardware requires that commands are aligned on a 64-bit
4326          * boundary. Given that we use pci_alloc_consistent() to allocate an
4327          * array of them, the size must be a multiple of 8 bytes.
4328          */
4329         BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4330
4331         printk(KERN_INFO DRIVER_NAME "\n");
4332
4333         err = bus_register(&cciss_bus_type);
4334         if (err)
4335                 return err;
4336
4337         /* Start the scan thread */
4338         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4339         if (IS_ERR(cciss_scan_thread)) {
4340                 err = PTR_ERR(cciss_scan_thread);
4341                 goto err_bus_unregister;
4342         }
4343
4344         /* Register for our PCI devices */
4345         err = pci_register_driver(&cciss_pci_driver);
4346         if (err)
4347                 goto err_thread_stop;
4348
4349         return err;
4350
4351 err_thread_stop:
4352         kthread_stop(cciss_scan_thread);
4353 err_bus_unregister:
4354         bus_unregister(&cciss_bus_type);
4355
4356         return err;
4357 }
4358
4359 static void __exit cciss_cleanup(void)
4360 {
4361         int i;
4362
4363         pci_unregister_driver(&cciss_pci_driver);
4364         /* double check that all controller entrys have been removed */
4365         for (i = 0; i < MAX_CTLR; i++) {
4366                 if (hba[i] != NULL) {
4367                         printk(KERN_WARNING "cciss: had to remove"
4368                                " controller %d\n", i);
4369                         cciss_remove_one(hba[i]->pdev);
4370                 }
4371         }
4372         kthread_stop(cciss_scan_thread);
4373         remove_proc_entry("driver/cciss", NULL);
4374         bus_unregister(&cciss_bus_type);
4375 }
4376
4377 static void fail_all_cmds(unsigned long ctlr)
4378 {
4379         /* If we get here, the board is apparently dead. */
4380         ctlr_info_t *h = hba[ctlr];
4381         CommandList_struct *c;
4382         unsigned long flags;
4383
4384         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4385         h->alive = 0;           /* the controller apparently died... */
4386
4387         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4388
4389         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
4390
4391         /* move everything off the request queue onto the completed queue */
4392         while (!hlist_empty(&h->reqQ)) {
4393                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4394                 removeQ(c);
4395                 h->Qdepth--;
4396                 addQ(&h->cmpQ, c);
4397         }
4398
4399         /* Now, fail everything on the completed queue with a HW error */
4400         while (!hlist_empty(&h->cmpQ)) {
4401                 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4402                 removeQ(c);
4403                 if (c->cmd_type != CMD_MSG_STALE)
4404                         c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4405                 if (c->cmd_type == CMD_RWREQ) {
4406                         complete_command(h, c, 0);
4407                 } else if (c->cmd_type == CMD_IOCTL_PEND)
4408                         complete(c->waiting);
4409 #ifdef CONFIG_CISS_SCSI_TAPE
4410                 else if (c->cmd_type == CMD_SCSI)
4411                         complete_scsi_command(c, 0, 0);
4412 #endif
4413         }
4414         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4415         return;
4416 }
4417
4418 module_init(cciss_init);
4419 module_exit(cciss_cleanup);