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, int withirq,
 183                         sector_t *total_size, unsigned int *block_size);
 184 static void cciss_read_capacity_16(int ctlr, int logvol, int withirq,
 185                         sector_t *total_size, unsigned int *block_size);
 186 static void cciss_geometry_inquiry(int ctlr, int logvol,
 187                         int withirq, 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(__u8 cmd, int ctlr, void *buff, size_t size,
 194                    __u8 page_code, unsigned char *scsi3addr, int cmd_type);
 195 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 196                         __u8 page_code, unsigned char scsi3addr[],
 197                         int cmd_type);
 198 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 199         int attempt_retry);
 200 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 201
 202 static void fail_all_cmds(unsigned long ctlr);
 203 static int add_to_scan_list(struct ctlr_info *h);
 204 static int scan_thread(void *data);
 205 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 206 static void cciss_hba_release(struct device *dev);
 207 static void cciss_device_release(struct device *dev);
 208 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 209 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 210
 211 #ifdef CONFIG_PROC_FS
 212 static void cciss_procinit(int i);
 213 #else
 214 static void cciss_procinit(int i)
 215 {
 216 }
 217 #endif                          /* CONFIG_PROC_FS */
 218
 219 #ifdef CONFIG_COMPAT
 220 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 221                               unsigned, unsigned long);
 222 #endif
 223
 224 static const struct block_device_operations cciss_fops = {
 225         .owner = THIS_MODULE,
 226         .open = cciss_open,
 227         .release = cciss_release,
 228         .locked_ioctl = cciss_ioctl,
 229         .getgeo = cciss_getgeo,
 230 #ifdef CONFIG_COMPAT
 231         .compat_ioctl = cciss_compat_ioctl,
 232 #endif
 233         .revalidate_disk = cciss_revalidate,
 234 };
 235
 236 /*
 237  * Enqueuing and dequeuing functions for cmdlists.
 238  */
 239 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 240 {
 241         hlist_add_head(&c->list, list);
 242 }
 243
 244 static inline void removeQ(CommandList_struct *c)
 245 {
 246         /*
 247          * After kexec/dump some commands might still
 248          * be in flight, which the firmware will try
 249          * to complete. Resetting the firmware doesn't work
 250          * with old fw revisions, so we have to mark
 251          * them off as 'stale' to prevent the driver from
 252          * falling over.
 253          */
 254         if (WARN_ON(hlist_unhashed(&c->list))) {
 255                 c->cmd_type = CMD_MSG_STALE;
 256                 return;
 257         }
 258
 259         hlist_del_init(&c->list);
 260 }
 261
 262 #include "cciss_scsi.c"         /* For SCSI tape support */
 263
 264 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 265         "UNKNOWN"
 266 };
 267 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
 268
 269 #ifdef CONFIG_PROC_FS
 270
 271 /*
 272  * Report information about this controller.
 273  */
 274 #define ENG_GIG 1000000000
 275 #define ENG_GIG_FACTOR (ENG_GIG/512)
 276 #define ENGAGE_SCSI     "engage scsi"
 277
 278 static struct proc_dir_entry *proc_cciss;
 279
 280 static void cciss_seq_show_header(struct seq_file *seq)
 281 {
 282         ctlr_info_t *h = seq->private;
 283
 284         seq_printf(seq, "%s: HP %s Controller\n"
 285                 "Board ID: 0x%08lx\n"
 286                 "Firmware Version: %c%c%c%c\n"
 287                 "IRQ: %d\n"
 288                 "Logical drives: %d\n"
 289                 "Current Q depth: %d\n"
 290                 "Current # commands on controller: %d\n"
 291                 "Max Q depth since init: %d\n"
 292                 "Max # commands on controller since init: %d\n"
 293                 "Max SG entries since init: %d\n",
 294                 h->devname,
 295                 h->product_name,
 296                 (unsigned long)h->board_id,
 297                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 298                 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
 299                 h->num_luns,
 300                 h->Qdepth, h->commands_outstanding,
 301                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 302
 303 #ifdef CONFIG_CISS_SCSI_TAPE
 304         cciss_seq_tape_report(seq, h->ctlr);
 305 #endif /* CONFIG_CISS_SCSI_TAPE */
 306 }
 307
 308 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 309 {
 310         ctlr_info_t *h = seq->private;
 311         unsigned ctlr = h->ctlr;
 312         unsigned long flags;
 313
 314         /* prevent displaying bogus info during configuration
 315          * or deconfiguration of a logical volume
 316          */
 317         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 318         if (h->busy_configuring) {
 319                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 320                 return ERR_PTR(-EBUSY);
 321         }
 322         h->busy_configuring = 1;
 323         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 324
 325         if (*pos == 0)
 326                 cciss_seq_show_header(seq);
 327
 328         return pos;
 329 }
 330
 331 static int cciss_seq_show(struct seq_file *seq, void *v)
 332 {
 333         sector_t vol_sz, vol_sz_frac;
 334         ctlr_info_t *h = seq->private;
 335         unsigned ctlr = h->ctlr;
 336         loff_t *pos = v;
 337         drive_info_struct *drv = h->drv[*pos];
 338
 339         if (*pos > h->highest_lun)
 340                 return 0;
 341
 342         if (drv->heads == 0)
 343                 return 0;
 344
 345         vol_sz = drv->nr_blocks;
 346         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 347         vol_sz_frac *= 100;
 348         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 349
 350         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 351                 drv->raid_level = RAID_UNKNOWN;
 352         seq_printf(seq, "cciss/c%dd%d:"
 353                         "\t%4u.%02uGB\tRAID %s\n",
 354                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 355                         raid_label[drv->raid_level]);
 356         return 0;
 357 }
 358
 359 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 360 {
 361         ctlr_info_t *h = seq->private;
 362
 363         if (*pos > h->highest_lun)
 364                 return NULL;
 365         *pos += 1;
 366
 367         return pos;
 368 }
 369
 370 static void cciss_seq_stop(struct seq_file *seq, void *v)
 371 {
 372         ctlr_info_t *h = seq->private;
 373
 374         /* Only reset h->busy_configuring if we succeeded in setting
 375          * it during cciss_seq_start. */
 376         if (v == ERR_PTR(-EBUSY))
 377                 return;
 378
 379         h->busy_configuring = 0;
 380 }
 381
 382 static const struct seq_operations cciss_seq_ops = {
 383         .start = cciss_seq_start,
 384         .show  = cciss_seq_show,
 385         .next  = cciss_seq_next,
 386         .stop  = cciss_seq_stop,
 387 };
 388
 389 static int cciss_seq_open(struct inode *inode, struct file *file)
 390 {
 391         int ret = seq_open(file, &cciss_seq_ops);
 392         struct seq_file *seq = file->private_data;
 393
 394         if (!ret)
 395                 seq->private = PDE(inode)->data;
 396
 397         return ret;
 398 }
 399
 400 static ssize_t
 401 cciss_proc_write(struct file *file, const char __user *buf,
 402                  size_t length, loff_t *ppos)
 403 {
 404         int err;
 405         char *buffer;
 406
 407 #ifndef CONFIG_CISS_SCSI_TAPE
 408         return -EINVAL;
 409 #endif
 410
 411         if (!buf || length > PAGE_SIZE - 1)
 412                 return -EINVAL;
 413
 414         buffer = (char *)__get_free_page(GFP_KERNEL);
 415         if (!buffer)
 416                 return -ENOMEM;
 417
 418         err = -EFAULT;
 419         if (copy_from_user(buffer, buf, length))
 420                 goto out;
 421         buffer[length] = '\0';
 422
 423 #ifdef CONFIG_CISS_SCSI_TAPE
 424         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 425                 struct seq_file *seq = file->private_data;
 426                 ctlr_info_t *h = seq->private;
 427                 int rc;
 428
 429                 rc = cciss_engage_scsi(h->ctlr);
 430                 if (rc != 0)
 431                         err = -rc;
 432                 else
 433                         err = length;
 434         } else
 435 #endif /* CONFIG_CISS_SCSI_TAPE */
 436                 err = -EINVAL;
 437         /* might be nice to have "disengage" too, but it's not
 438            safely possible. (only 1 module use count, lock issues.) */
 439
 440 out:
 441         free_page((unsigned long)buffer);
 442         return err;
 443 }
 444
 445 static const struct file_operations cciss_proc_fops = {
 446         .owner   = THIS_MODULE,
 447         .open    = cciss_seq_open,
 448         .read    = seq_read,
 449         .llseek  = seq_lseek,
 450         .release = seq_release,
 451         .write   = cciss_proc_write,
 452 };
 453
 454 static void __devinit cciss_procinit(int i)
 455 {
 456         struct proc_dir_entry *pde;
 457
 458         if (proc_cciss == NULL)
 459                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 460         if (!proc_cciss)
 461                 return;
 462         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 463                                         S_IROTH, proc_cciss,
 464                                         &cciss_proc_fops, hba[i]);
 465 }
 466 #endif                          /* CONFIG_PROC_FS */
 467
 468 #define MAX_PRODUCT_NAME_LEN 19
 469
 470 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 471 #define to_drv(n) container_of(n, drive_info_struct, dev)
 472
 473 static ssize_t host_store_rescan(struct device *dev,
 474                                  struct device_attribute *attr,
 475                                  const char *buf, size_t count)
 476 {
 477         struct ctlr_info *h = to_hba(dev);
 478
 479         add_to_scan_list(h);
 480         wake_up_process(cciss_scan_thread);
 481         wait_for_completion_interruptible(&h->scan_wait);
 482
 483         return count;
 484 }
 485 DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 486
 487 static ssize_t dev_show_unique_id(struct device *dev,
 488                                  struct device_attribute *attr,
 489                                  char *buf)
 490 {
 491         drive_info_struct *drv = to_drv(dev);
 492         struct ctlr_info *h = to_hba(drv->dev.parent);
 493         __u8 sn[16];
 494         unsigned long flags;
 495         int ret = 0;
 496
 497         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 498         if (h->busy_configuring)
 499                 ret = -EBUSY;
 500         else
 501                 memcpy(sn, drv->serial_no, sizeof(sn));
 502         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 503
 504         if (ret)
 505                 return ret;
 506         else
 507                 return snprintf(buf, 16 * 2 + 2,
 508                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 509                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 510                                 sn[0], sn[1], sn[2], sn[3],
 511                                 sn[4], sn[5], sn[6], sn[7],
 512                                 sn[8], sn[9], sn[10], sn[11],
 513                                 sn[12], sn[13], sn[14], sn[15]);
 514 }
 515 DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 516
 517 static ssize_t dev_show_vendor(struct device *dev,
 518                                struct device_attribute *attr,
 519                                char *buf)
 520 {
 521         drive_info_struct *drv = to_drv(dev);
 522         struct ctlr_info *h = to_hba(drv->dev.parent);
 523         char vendor[VENDOR_LEN + 1];
 524         unsigned long flags;
 525         int ret = 0;
 526
 527         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 528         if (h->busy_configuring)
 529                 ret = -EBUSY;
 530         else
 531                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 532         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 533
 534         if (ret)
 535                 return ret;
 536         else
 537                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 538 }
 539 DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 540
 541 static ssize_t dev_show_model(struct device *dev,
 542                               struct device_attribute *attr,
 543                               char *buf)
 544 {
 545         drive_info_struct *drv = to_drv(dev);
 546         struct ctlr_info *h = to_hba(drv->dev.parent);
 547         char model[MODEL_LEN + 1];
 548         unsigned long flags;
 549         int ret = 0;
 550
 551         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 552         if (h->busy_configuring)
 553                 ret = -EBUSY;
 554         else
 555                 memcpy(model, drv->model, MODEL_LEN + 1);
 556         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 557
 558         if (ret)
 559                 return ret;
 560         else
 561                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 562 }
 563 DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 564
 565 static ssize_t dev_show_rev(struct device *dev,
 566                             struct device_attribute *attr,
 567                             char *buf)
 568 {
 569         drive_info_struct *drv = to_drv(dev);
 570         struct ctlr_info *h = to_hba(drv->dev.parent);
 571         char rev[REV_LEN + 1];
 572         unsigned long flags;
 573         int ret = 0;
 574
 575         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 576         if (h->busy_configuring)
 577                 ret = -EBUSY;
 578         else
 579                 memcpy(rev, drv->rev, REV_LEN + 1);
 580         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 581
 582         if (ret)
 583                 return ret;
 584         else
 585                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 586 }
 587 DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 588
 589 static ssize_t cciss_show_lunid(struct device *dev,
 590                                 struct device_attribute *attr, char *buf)
 591 {
 592         drive_info_struct *drv = to_drv(dev);
 593         struct ctlr_info *h = to_hba(drv->dev.parent);
 594         unsigned long flags;
 595         unsigned char lunid[8];
 596
 597         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 598         if (h->busy_configuring) {
 599                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 600                 return -EBUSY;
 601         }
 602         if (!drv->heads) {
 603                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 604                 return -ENOTTY;
 605         }
 606         memcpy(lunid, drv->LunID, sizeof(lunid));
 607         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 608         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 609                 lunid[0], lunid[1], lunid[2], lunid[3],
 610                 lunid[4], lunid[5], lunid[6], lunid[7]);
 611 }
 612 DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 613
 614 static ssize_t cciss_show_raid_level(struct device *dev,
 615                                      struct device_attribute *attr, char *buf)
 616 {
 617         drive_info_struct *drv = to_drv(dev);
 618         struct ctlr_info *h = to_hba(drv->dev.parent);
 619         int raid;
 620         unsigned long flags;
 621
 622         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 623         if (h->busy_configuring) {
 624                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 625                 return -EBUSY;
 626         }
 627         raid = drv->raid_level;
 628         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 629         if (raid < 0 || raid > RAID_UNKNOWN)
 630                 raid = RAID_UNKNOWN;
 631
 632         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 633                         raid_label[raid]);
 634 }
 635 DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 636
 637 static ssize_t cciss_show_usage_count(struct device *dev,
 638                                       struct device_attribute *attr, char *buf)
 639 {
 640         drive_info_struct *drv = to_drv(dev);
 641         struct ctlr_info *h = to_hba(drv->dev.parent);
 642         unsigned long flags;
 643         int count;
 644
 645         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 646         if (h->busy_configuring) {
 647                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 648                 return -EBUSY;
 649         }
 650         count = drv->usage_count;
 651         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 652         return snprintf(buf, 20, "%d\n", count);
 653 }
 654 DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 655
 656 static struct attribute *cciss_host_attrs[] = {
 657         &dev_attr_rescan.attr,
 658         NULL
 659 };
 660
 661 static struct attribute_group cciss_host_attr_group = {
 662         .attrs = cciss_host_attrs,
 663 };
 664
 665 static const struct attribute_group *cciss_host_attr_groups[] = {
 666         &cciss_host_attr_group,
 667         NULL
 668 };
 669
 670 static struct device_type cciss_host_type = {
 671         .name           = "cciss_host",
 672         .groups         = cciss_host_attr_groups,
 673         .release        = cciss_hba_release,
 674 };
 675
 676 static struct attribute *cciss_dev_attrs[] = {
 677         &dev_attr_unique_id.attr,
 678         &dev_attr_model.attr,
 679         &dev_attr_vendor.attr,
 680         &dev_attr_rev.attr,
 681         &dev_attr_lunid.attr,
 682         &dev_attr_raid_level.attr,
 683         &dev_attr_usage_count.attr,
 684         NULL
 685 };
 686
 687 static struct attribute_group cciss_dev_attr_group = {
 688         .attrs = cciss_dev_attrs,
 689 };
 690
 691 static const struct attribute_group *cciss_dev_attr_groups[] = {
 692         &cciss_dev_attr_group,
 693         NULL
 694 };
 695
 696 static struct device_type cciss_dev_type = {
 697         .name           = "cciss_device",
 698         .groups         = cciss_dev_attr_groups,
 699         .release        = cciss_device_release,
 700 };
 701
 702 static struct bus_type cciss_bus_type = {
 703         .name           = "cciss",
 704 };
 705
 706 /*
 707  * cciss_hba_release is called when the reference count
 708  * of h->dev goes to zero.
 709  */
 710 static void cciss_hba_release(struct device *dev)
 711 {
 712         /*
 713          * nothing to do, but need this to avoid a warning
 714          * about not having a release handler from lib/kref.c.
 715          */
 716 }
 717
 718 /*
 719  * Initialize sysfs entry for each controller.  This sets up and registers
 720  * the 'cciss#' directory for each individual controller under
 721  * /sys/bus/pci/devices/<dev>/.
 722  */
 723 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 724 {
 725         device_initialize(&h->dev);
 726         h->dev.type = &cciss_host_type;
 727         h->dev.bus = &cciss_bus_type;
 728         dev_set_name(&h->dev, "%s", h->devname);
 729         h->dev.parent = &h->pdev->dev;
 730
 731         return device_add(&h->dev);
 732 }
 733
 734 /*
 735  * Remove sysfs entries for an hba.
 736  */
 737 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 738 {
 739         device_del(&h->dev);
 740         put_device(&h->dev); /* final put. */
 741 }
 742
 743 /* cciss_device_release is called when the reference count
 744  * of h->drv[x]dev goes to zero.
 745  */
 746 static void cciss_device_release(struct device *dev)
 747 {
 748         drive_info_struct *drv = to_drv(dev);
 749         kfree(drv);
 750 }
 751
 752 /*
 753  * Initialize sysfs for each logical drive.  This sets up and registers
 754  * the 'c#d#' directory for each individual logical drive under
 755  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 756  * /sys/block/cciss!c#d# to this entry.
 757  */
 758 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 759                                        int drv_index)
 760 {
 761         struct device *dev;
 762
 763         if (h->drv[drv_index]->device_initialized)
 764                 return 0;
 765
 766         dev = &h->drv[drv_index]->dev;
 767         device_initialize(dev);
 768         dev->type = &cciss_dev_type;
 769         dev->bus = &cciss_bus_type;
 770         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 771         dev->parent = &h->dev;
 772         h->drv[drv_index]->device_initialized = 1;
 773         return device_add(dev);
 774 }
 775
 776 /*
 777  * Remove sysfs entries for a logical drive.
 778  */
 779 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 780         int ctlr_exiting)
 781 {
 782         struct device *dev = &h->drv[drv_index]->dev;
 783
 784         /* special case for c*d0, we only destroy it on controller exit */
 785         if (drv_index == 0 && !ctlr_exiting)
 786                 return;
 787
 788         device_del(dev);
 789         put_device(dev); /* the "final" put. */
 790         h->drv[drv_index] = NULL;
 791 }
 792
 793 /*
 794  * For operations that cannot sleep, a command block is allocated at init,
 795  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 796  * which ones are free or in use.  For operations that can wait for kmalloc
 797  * to possible sleep, this routine can be called with get_from_pool set to 0.
 798  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 799  */
 800 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 801 {
 802         CommandList_struct *c;
 803         int i;
 804         u64bit temp64;
 805         dma_addr_t cmd_dma_handle, err_dma_handle;
 806
 807         if (!get_from_pool) {
 808                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 809                         sizeof(CommandList_struct), &cmd_dma_handle);
 810                 if (c == NULL)
 811                         return NULL;
 812                 memset(c, 0, sizeof(CommandList_struct));
 813
 814                 c->cmdindex = -1;
 815
 816                 c->err_info = (ErrorInfo_struct *)
 817                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 818                             &err_dma_handle);
 819
 820                 if (c->err_info == NULL) {
 821                         pci_free_consistent(h->pdev,
 822                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 823                         return NULL;
 824                 }
 825                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 826         } else {                /* get it out of the controllers pool */
 827
 828                 do {
 829                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 830                         if (i == h->nr_cmds)
 831                                 return NULL;
 832                 } while (test_and_set_bit
 833                          (i & (BITS_PER_LONG - 1),
 834                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 835 #ifdef CCISS_DEBUG
 836                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 837 #endif
 838                 c = h->cmd_pool + i;
 839                 memset(c, 0, sizeof(CommandList_struct));
 840                 cmd_dma_handle = h->cmd_pool_dhandle
 841                     + i * sizeof(CommandList_struct);
 842                 c->err_info = h->errinfo_pool + i;
 843                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 844                 err_dma_handle = h->errinfo_pool_dhandle
 845                     + i * sizeof(ErrorInfo_struct);
 846                 h->nr_allocs++;
 847
 848                 c->cmdindex = i;
 849         }
 850
 851         INIT_HLIST_NODE(&c->list);
 852         c->busaddr = (__u32) cmd_dma_handle;
 853         temp64.val = (__u64) err_dma_handle;
 854         c->ErrDesc.Addr.lower = temp64.val32.lower;
 855         c->ErrDesc.Addr.upper = temp64.val32.upper;
 856         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 857
 858         c->ctlr = h->ctlr;
 859         return c;
 860 }
 861
 862 /*
 863  * Frees a command block that was previously allocated with cmd_alloc().
 864  */
 865 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 866 {
 867         int i;
 868         u64bit temp64;
 869
 870         if (!got_from_pool) {
 871                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 872                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 873                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 874                                     c->err_info, (dma_addr_t) temp64.val);
 875                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 876                                     c, (dma_addr_t) c->busaddr);
 877         } else {
 878                 i = c - h->cmd_pool;
 879                 clear_bit(i & (BITS_PER_LONG - 1),
 880                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 881                 h->nr_frees++;
 882         }
 883 }
 884
 885 static inline ctlr_info_t *get_host(struct gendisk *disk)
 886 {
 887         return disk->queue->queuedata;
 888 }
 889
 890 static inline drive_info_struct *get_drv(struct gendisk *disk)
 891 {
 892         return disk->private_data;
 893 }
 894
 895 /*
 896  * Open.  Make sure the device is really there.
 897  */
 898 static int cciss_open(struct block_device *bdev, fmode_t mode)
 899 {
 900         ctlr_info_t *host = get_host(bdev->bd_disk);
 901         drive_info_struct *drv = get_drv(bdev->bd_disk);
 902
 903 #ifdef CCISS_DEBUG
 904         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
 905 #endif                          /* CCISS_DEBUG */
 906
 907         if (drv->busy_configuring)
 908                 return -EBUSY;
 909         /*
 910          * Root is allowed to open raw volume zero even if it's not configured
 911          * so array config can still work. Root is also allowed to open any
 912          * volume that has a LUN ID, so it can issue IOCTL to reread the
 913          * disk information.  I don't think I really like this
 914          * but I'm already using way to many device nodes to claim another one
 915          * for "raw controller".
 916          */
 917         if (drv->heads == 0) {
 918                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
 919                         /* if not node 0 make sure it is a partition = 0 */
 920                         if (MINOR(bdev->bd_dev) & 0x0f) {
 921                                 return -ENXIO;
 922                                 /* if it is, make sure we have a LUN ID */
 923                         } else if (memcmp(drv->LunID, CTLR_LUNID,
 924                                 sizeof(drv->LunID))) {
 925                                 return -ENXIO;
 926                         }
 927                 }
 928                 if (!capable(CAP_SYS_ADMIN))
 929                         return -EPERM;
 930         }
 931         drv->usage_count++;
 932         host->usage_count++;
 933         return 0;
 934 }
 935
 936 /*
 937  * Close.  Sync first.
 938  */
 939 static int cciss_release(struct gendisk *disk, fmode_t mode)
 940 {
 941         ctlr_info_t *host = get_host(disk);
 942         drive_info_struct *drv = get_drv(disk);
 943
 944 #ifdef CCISS_DEBUG
 945         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
 946 #endif                          /* CCISS_DEBUG */
 947
 948         drv->usage_count--;
 949         host->usage_count--;
 950         return 0;
 951 }
 952
 953 #ifdef CONFIG_COMPAT
 954
 955 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 956                     unsigned cmd, unsigned long arg)
 957 {
 958         int ret;
 959         lock_kernel();
 960         ret = cciss_ioctl(bdev, mode, cmd, arg);
 961         unlock_kernel();
 962         return ret;
 963 }
 964
 965 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 966                                   unsigned cmd, unsigned long arg);
 967 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
 968                                       unsigned cmd, unsigned long arg);
 969
 970 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
 971                               unsigned cmd, unsigned long arg)
 972 {
 973         switch (cmd) {
 974         case CCISS_GETPCIINFO:
 975         case CCISS_GETINTINFO:
 976         case CCISS_SETINTINFO:
 977         case CCISS_GETNODENAME:
 978         case CCISS_SETNODENAME:
 979         case CCISS_GETHEARTBEAT:
 980         case CCISS_GETBUSTYPES:
 981         case CCISS_GETFIRMVER:
 982         case CCISS_GETDRIVVER:
 983         case CCISS_REVALIDVOLS:
 984         case CCISS_DEREGDISK:
 985         case CCISS_REGNEWDISK:
 986         case CCISS_REGNEWD:
 987         case CCISS_RESCANDISK:
 988         case CCISS_GETLUNINFO:
 989                 return do_ioctl(bdev, mode, cmd, arg);
 990
 991         case CCISS_PASSTHRU32:
 992                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
 993         case CCISS_BIG_PASSTHRU32:
 994                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
 995
 996         default:
 997                 return -ENOIOCTLCMD;
 998         }
 999 }
1000
1001 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1002                                   unsigned cmd, unsigned long arg)
1003 {
1004         IOCTL32_Command_struct __user *arg32 =
1005             (IOCTL32_Command_struct __user *) arg;
1006         IOCTL_Command_struct arg64;
1007         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1008         int err;
1009         u32 cp;
1010
1011         err = 0;
1012         err |=
1013             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1014                            sizeof(arg64.LUN_info));
1015         err |=
1016             copy_from_user(&arg64.Request, &arg32->Request,
1017                            sizeof(arg64.Request));
1018         err |=
1019             copy_from_user(&arg64.error_info, &arg32->error_info,
1020                            sizeof(arg64.error_info));
1021         err |= get_user(arg64.buf_size, &arg32->buf_size);
1022         err |= get_user(cp, &arg32->buf);
1023         arg64.buf = compat_ptr(cp);
1024         err |= copy_to_user(p, &arg64, sizeof(arg64));
1025
1026         if (err)
1027                 return -EFAULT;
1028
1029         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1030         if (err)
1031                 return err;
1032         err |=
1033             copy_in_user(&arg32->error_info, &p->error_info,
1034                          sizeof(arg32->error_info));
1035         if (err)
1036                 return -EFAULT;
1037         return err;
1038 }
1039
1040 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1041                                       unsigned cmd, unsigned long arg)
1042 {
1043         BIG_IOCTL32_Command_struct __user *arg32 =
1044             (BIG_IOCTL32_Command_struct __user *) arg;
1045         BIG_IOCTL_Command_struct arg64;
1046         BIG_IOCTL_Command_struct __user *p =
1047             compat_alloc_user_space(sizeof(arg64));
1048         int err;
1049         u32 cp;
1050
1051         err = 0;
1052         err |=
1053             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1054                            sizeof(arg64.LUN_info));
1055         err |=
1056             copy_from_user(&arg64.Request, &arg32->Request,
1057                            sizeof(arg64.Request));
1058         err |=
1059             copy_from_user(&arg64.error_info, &arg32->error_info,
1060                            sizeof(arg64.error_info));
1061         err |= get_user(arg64.buf_size, &arg32->buf_size);
1062         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1063         err |= get_user(cp, &arg32->buf);
1064         arg64.buf = compat_ptr(cp);
1065         err |= copy_to_user(p, &arg64, sizeof(arg64));
1066
1067         if (err)
1068                 return -EFAULT;
1069
1070         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1071         if (err)
1072                 return err;
1073         err |=
1074             copy_in_user(&arg32->error_info, &p->error_info,
1075                          sizeof(arg32->error_info));
1076         if (err)
1077                 return -EFAULT;
1078         return err;
1079 }
1080 #endif
1081
1082 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1083 {
1084         drive_info_struct *drv = get_drv(bdev->bd_disk);
1085
1086         if (!drv->cylinders)
1087                 return -ENXIO;
1088
1089         geo->heads = drv->heads;
1090         geo->sectors = drv->sectors;
1091         geo->cylinders = drv->cylinders;
1092         return 0;
1093 }
1094
1095 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1096 {
1097         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1098                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1099                 (void)check_for_unit_attention(host, c);
1100 }
1101 /*
1102  * ioctl
1103  */
1104 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1105                        unsigned int cmd, unsigned long arg)
1106 {
1107         struct gendisk *disk = bdev->bd_disk;
1108         ctlr_info_t *host = get_host(disk);
1109         drive_info_struct *drv = get_drv(disk);
1110         int ctlr = host->ctlr;
1111         void __user *argp = (void __user *)arg;
1112
1113 #ifdef CCISS_DEBUG
1114         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1115 #endif                          /* CCISS_DEBUG */
1116
1117         switch (cmd) {
1118         case CCISS_GETPCIINFO:
1119                 {
1120                         cciss_pci_info_struct pciinfo;
1121
1122                         if (!arg)
1123                                 return -EINVAL;
1124                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1125                         pciinfo.bus = host->pdev->bus->number;
1126                         pciinfo.dev_fn = host->pdev->devfn;
1127                         pciinfo.board_id = host->board_id;
1128                         if (copy_to_user
1129                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1130                                 return -EFAULT;
1131                         return 0;
1132                 }
1133         case CCISS_GETINTINFO:
1134                 {
1135                         cciss_coalint_struct intinfo;
1136                         if (!arg)
1137                                 return -EINVAL;
1138                         intinfo.delay =
1139                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1140                         intinfo.count =
1141                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1142                         if (copy_to_user
1143                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1144                                 return -EFAULT;
1145                         return 0;
1146                 }
1147         case CCISS_SETINTINFO:
1148                 {
1149                         cciss_coalint_struct intinfo;
1150                         unsigned long flags;
1151                         int i;
1152
1153                         if (!arg)
1154                                 return -EINVAL;
1155                         if (!capable(CAP_SYS_ADMIN))
1156                                 return -EPERM;
1157                         if (copy_from_user
1158                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1159                                 return -EFAULT;
1160                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1161                         {
1162 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1163                                 return -EINVAL;
1164                         }
1165                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1166                         /* Update the field, and then ring the doorbell */
1167                         writel(intinfo.delay,
1168                                &(host->cfgtable->HostWrite.CoalIntDelay));
1169                         writel(intinfo.count,
1170                                &(host->cfgtable->HostWrite.CoalIntCount));
1171                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1172
1173                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1174                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1175                                       & CFGTBL_ChangeReq))
1176                                         break;
1177                                 /* delay and try again */
1178                                 udelay(1000);
1179                         }
1180                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1181                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1182                                 return -EAGAIN;
1183                         return 0;
1184                 }
1185         case CCISS_GETNODENAME:
1186                 {
1187                         NodeName_type NodeName;
1188                         int i;
1189
1190                         if (!arg)
1191                                 return -EINVAL;
1192                         for (i = 0; i < 16; i++)
1193                                 NodeName[i] =
1194                                     readb(&host->cfgtable->ServerName[i]);
1195                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1196                                 return -EFAULT;
1197                         return 0;
1198                 }
1199         case CCISS_SETNODENAME:
1200                 {
1201                         NodeName_type NodeName;
1202                         unsigned long flags;
1203                         int i;
1204
1205                         if (!arg)
1206                                 return -EINVAL;
1207                         if (!capable(CAP_SYS_ADMIN))
1208                                 return -EPERM;
1209
1210                         if (copy_from_user
1211                             (NodeName, argp, sizeof(NodeName_type)))
1212                                 return -EFAULT;
1213
1214                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1215
1216                         /* Update the field, and then ring the doorbell */
1217                         for (i = 0; i < 16; i++)
1218                                 writeb(NodeName[i],
1219                                        &host->cfgtable->ServerName[i]);
1220
1221                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1222
1223                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1224                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1225                                       & CFGTBL_ChangeReq))
1226                                         break;
1227                                 /* delay and try again */
1228                                 udelay(1000);
1229                         }
1230                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1231                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1232                                 return -EAGAIN;
1233                         return 0;
1234                 }
1235
1236         case CCISS_GETHEARTBEAT:
1237                 {
1238                         Heartbeat_type heartbeat;
1239
1240                         if (!arg)
1241                                 return -EINVAL;
1242                         heartbeat = readl(&host->cfgtable->HeartBeat);
1243                         if (copy_to_user
1244                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1245                                 return -EFAULT;
1246                         return 0;
1247                 }
1248         case CCISS_GETBUSTYPES:
1249                 {
1250                         BusTypes_type BusTypes;
1251
1252                         if (!arg)
1253                                 return -EINVAL;
1254                         BusTypes = readl(&host->cfgtable->BusTypes);
1255                         if (copy_to_user
1256                             (argp, &BusTypes, sizeof(BusTypes_type)))
1257                                 return -EFAULT;
1258                         return 0;
1259                 }
1260         case CCISS_GETFIRMVER:
1261                 {
1262                         FirmwareVer_type firmware;
1263
1264                         if (!arg)
1265                                 return -EINVAL;
1266                         memcpy(firmware, host->firm_ver, 4);
1267
1268                         if (copy_to_user
1269                             (argp, firmware, sizeof(FirmwareVer_type)))
1270                                 return -EFAULT;
1271                         return 0;
1272                 }
1273         case CCISS_GETDRIVVER:
1274                 {
1275                         DriverVer_type DriverVer = DRIVER_VERSION;
1276
1277                         if (!arg)
1278                                 return -EINVAL;
1279
1280                         if (copy_to_user
1281                             (argp, &DriverVer, sizeof(DriverVer_type)))
1282                                 return -EFAULT;
1283                         return 0;
1284                 }
1285
1286         case CCISS_DEREGDISK:
1287         case CCISS_REGNEWD:
1288         case CCISS_REVALIDVOLS:
1289                 return rebuild_lun_table(host, 0, 1);
1290
1291         case CCISS_GETLUNINFO:{
1292                         LogvolInfo_struct luninfo;
1293
1294                         memcpy(&luninfo.LunID, drv->LunID,
1295                                 sizeof(luninfo.LunID));
1296                         luninfo.num_opens = drv->usage_count;
1297                         luninfo.num_parts = 0;
1298                         if (copy_to_user(argp, &luninfo,
1299                                          sizeof(LogvolInfo_struct)))
1300                                 return -EFAULT;
1301                         return 0;
1302                 }
1303         case CCISS_PASSTHRU:
1304                 {
1305                         IOCTL_Command_struct iocommand;
1306                         CommandList_struct *c;
1307                         char *buff = NULL;
1308                         u64bit temp64;
1309                         unsigned long flags;
1310                         DECLARE_COMPLETION_ONSTACK(wait);
1311
1312                         if (!arg)
1313                                 return -EINVAL;
1314
1315                         if (!capable(CAP_SYS_RAWIO))
1316                                 return -EPERM;
1317
1318                         if (copy_from_user
1319                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1320                                 return -EFAULT;
1321                         if ((iocommand.buf_size < 1) &&
1322                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1323                                 return -EINVAL;
1324                         }
1325 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1326                         /* Check kmalloc limits */
1327                         if (iocommand.buf_size > 128000)
1328                                 return -EINVAL;
1329 #endif
1330                         if (iocommand.buf_size > 0) {
1331                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1332                                 if (buff == NULL)
1333                                         return -EFAULT;
1334                         }
1335                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1336                                 /* Copy the data into the buffer we created */
1337                                 if (copy_from_user
1338                                     (buff, iocommand.buf, iocommand.buf_size)) {
1339                                         kfree(buff);
1340                                         return -EFAULT;
1341                                 }
1342                         } else {
1343                                 memset(buff, 0, iocommand.buf_size);
1344                         }
1345                         if ((c = cmd_alloc(host, 0)) == NULL) {
1346                                 kfree(buff);
1347                                 return -ENOMEM;
1348                         }
1349                         // Fill in the command type
1350                         c->cmd_type = CMD_IOCTL_PEND;
1351                         // Fill in Command Header
1352                         c->Header.ReplyQueue = 0;       // unused in simple mode
1353                         if (iocommand.buf_size > 0)     // buffer to fill
1354                         {
1355                                 c->Header.SGList = 1;
1356                                 c->Header.SGTotal = 1;
1357                         } else  // no buffers to fill
1358                         {
1359                                 c->Header.SGList = 0;
1360                                 c->Header.SGTotal = 0;
1361                         }
1362                         c->Header.LUN = iocommand.LUN_info;
1363                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
1364
1365                         // Fill in Request block
1366                         c->Request = iocommand.Request;
1367
1368                         // Fill in the scatter gather information
1369                         if (iocommand.buf_size > 0) {
1370                                 temp64.val = pci_map_single(host->pdev, buff,
1371                                         iocommand.buf_size,
1372                                         PCI_DMA_BIDIRECTIONAL);
1373                                 c->SG[0].Addr.lower = temp64.val32.lower;
1374                                 c->SG[0].Addr.upper = temp64.val32.upper;
1375                                 c->SG[0].Len = iocommand.buf_size;
1376                                 c->SG[0].Ext = 0;       // we are not chaining
1377                         }
1378                         c->waiting = &wait;
1379
1380                         /* Put the request on the tail of the request queue */
1381                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1382                         addQ(&host->reqQ, c);
1383                         host->Qdepth++;
1384                         start_io(host);
1385                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1386
1387                         wait_for_completion(&wait);
1388
1389                         /* unlock the buffers from DMA */
1390                         temp64.val32.lower = c->SG[0].Addr.lower;
1391                         temp64.val32.upper = c->SG[0].Addr.upper;
1392                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1393                                          iocommand.buf_size,
1394                                          PCI_DMA_BIDIRECTIONAL);
1395
1396                         check_ioctl_unit_attention(host, c);
1397
1398                         /* Copy the error information out */
1399                         iocommand.error_info = *(c->err_info);
1400                         if (copy_to_user
1401                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1402                                 kfree(buff);
1403                                 cmd_free(host, c, 0);
1404                                 return -EFAULT;
1405                         }
1406
1407                         if (iocommand.Request.Type.Direction == XFER_READ) {
1408                                 /* Copy the data out of the buffer we created */
1409                                 if (copy_to_user
1410                                     (iocommand.buf, buff, iocommand.buf_size)) {
1411                                         kfree(buff);
1412                                         cmd_free(host, c, 0);
1413                                         return -EFAULT;
1414                                 }
1415                         }
1416                         kfree(buff);
1417                         cmd_free(host, c, 0);
1418                         return 0;
1419                 }
1420         case CCISS_BIG_PASSTHRU:{
1421                         BIG_IOCTL_Command_struct *ioc;
1422                         CommandList_struct *c;
1423                         unsigned char **buff = NULL;
1424                         int *buff_size = NULL;
1425                         u64bit temp64;
1426                         unsigned long flags;
1427                         BYTE sg_used = 0;
1428                         int status = 0;
1429                         int i;
1430                         DECLARE_COMPLETION_ONSTACK(wait);
1431                         __u32 left;
1432                         __u32 sz;
1433                         BYTE __user *data_ptr;
1434
1435                         if (!arg)
1436                                 return -EINVAL;
1437                         if (!capable(CAP_SYS_RAWIO))
1438                                 return -EPERM;
1439                         ioc = (BIG_IOCTL_Command_struct *)
1440                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1441                         if (!ioc) {
1442                                 status = -ENOMEM;
1443                                 goto cleanup1;
1444                         }
1445                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1446                                 status = -EFAULT;
1447                                 goto cleanup1;
1448                         }
1449                         if ((ioc->buf_size < 1) &&
1450                             (ioc->Request.Type.Direction != XFER_NONE)) {
1451                                 status = -EINVAL;
1452                                 goto cleanup1;
1453                         }
1454                         /* Check kmalloc limits  using all SGs */
1455                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1456                                 status = -EINVAL;
1457                                 goto cleanup1;
1458                         }
1459                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1460                                 status = -EINVAL;
1461                                 goto cleanup1;
1462                         }
1463                         buff =
1464                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1465                         if (!buff) {
1466                                 status = -ENOMEM;
1467                                 goto cleanup1;
1468                         }
1469                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1470                                                    GFP_KERNEL);
1471                         if (!buff_size) {
1472                                 status = -ENOMEM;
1473                                 goto cleanup1;
1474                         }
1475                         left = ioc->buf_size;
1476                         data_ptr = ioc->buf;
1477                         while (left) {
1478                                 sz = (left >
1479                                       ioc->malloc_size) ? ioc->
1480                                     malloc_size : left;
1481                                 buff_size[sg_used] = sz;
1482                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1483                                 if (buff[sg_used] == NULL) {
1484                                         status = -ENOMEM;
1485                                         goto cleanup1;
1486                                 }
1487                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1488                                         if (copy_from_user
1489                                             (buff[sg_used], data_ptr, sz)) {
1490                                                 status = -EFAULT;
1491                                                 goto cleanup1;
1492                                         }
1493                                 } else {
1494                                         memset(buff[sg_used], 0, sz);
1495                                 }
1496                                 left -= sz;
1497                                 data_ptr += sz;
1498                                 sg_used++;
1499                         }
1500                         if ((c = cmd_alloc(host, 0)) == NULL) {
1501                                 status = -ENOMEM;
1502                                 goto cleanup1;
1503                         }
1504                         c->cmd_type = CMD_IOCTL_PEND;
1505                         c->Header.ReplyQueue = 0;
1506
1507                         if (ioc->buf_size > 0) {
1508                                 c->Header.SGList = sg_used;
1509                                 c->Header.SGTotal = sg_used;
1510                         } else {
1511                                 c->Header.SGList = 0;
1512                                 c->Header.SGTotal = 0;
1513                         }
1514                         c->Header.LUN = ioc->LUN_info;
1515                         c->Header.Tag.lower = c->busaddr;
1516
1517                         c->Request = ioc->Request;
1518                         if (ioc->buf_size > 0) {
1519                                 int i;
1520                                 for (i = 0; i < sg_used; i++) {
1521                                         temp64.val =
1522                                             pci_map_single(host->pdev, buff[i],
1523                                                     buff_size[i],
1524                                                     PCI_DMA_BIDIRECTIONAL);
1525                                         c->SG[i].Addr.lower =
1526                                             temp64.val32.lower;
1527                                         c->SG[i].Addr.upper =
1528                                             temp64.val32.upper;
1529                                         c->SG[i].Len = buff_size[i];
1530                                         c->SG[i].Ext = 0;       /* we are not chaining */
1531                                 }
1532                         }
1533                         c->waiting = &wait;
1534                         /* Put the request on the tail of the request queue */
1535                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1536                         addQ(&host->reqQ, c);
1537                         host->Qdepth++;
1538                         start_io(host);
1539                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1540                         wait_for_completion(&wait);
1541                         /* unlock the buffers from DMA */
1542                         for (i = 0; i < sg_used; i++) {
1543                                 temp64.val32.lower = c->SG[i].Addr.lower;
1544                                 temp64.val32.upper = c->SG[i].Addr.upper;
1545                                 pci_unmap_single(host->pdev,
1546                                         (dma_addr_t) temp64.val, buff_size[i],
1547                                         PCI_DMA_BIDIRECTIONAL);
1548                         }
1549                         check_ioctl_unit_attention(host, c);
1550                         /* Copy the error information out */
1551                         ioc->error_info = *(c->err_info);
1552                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1553                                 cmd_free(host, c, 0);
1554                                 status = -EFAULT;
1555                                 goto cleanup1;
1556                         }
1557                         if (ioc->Request.Type.Direction == XFER_READ) {
1558                                 /* Copy the data out of the buffer we created */
1559                                 BYTE __user *ptr = ioc->buf;
1560                                 for (i = 0; i < sg_used; i++) {
1561                                         if (copy_to_user
1562                                             (ptr, buff[i], buff_size[i])) {
1563                                                 cmd_free(host, c, 0);
1564                                                 status = -EFAULT;
1565                                                 goto cleanup1;
1566                                         }
1567                                         ptr += buff_size[i];
1568                                 }
1569                         }
1570                         cmd_free(host, c, 0);
1571                         status = 0;
1572                       cleanup1:
1573                         if (buff) {
1574                                 for (i = 0; i < sg_used; i++)
1575                                         kfree(buff[i]);
1576                                 kfree(buff);
1577                         }
1578                         kfree(buff_size);
1579                         kfree(ioc);
1580                         return status;
1581                 }
1582
1583         /* scsi_cmd_ioctl handles these, below, though some are not */
1584         /* very meaningful for cciss.  SG_IO is the main one people want. */
1585
1586         case SG_GET_VERSION_NUM:
1587         case SG_SET_TIMEOUT:
1588         case SG_GET_TIMEOUT:
1589         case SG_GET_RESERVED_SIZE:
1590         case SG_SET_RESERVED_SIZE:
1591         case SG_EMULATED_HOST:
1592         case SG_IO:
1593         case SCSI_IOCTL_SEND_COMMAND:
1594                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1595
1596         /* scsi_cmd_ioctl would normally handle these, below, but */
1597         /* they aren't a good fit for cciss, as CD-ROMs are */
1598         /* not supported, and we don't have any bus/target/lun */
1599         /* which we present to the kernel. */
1600
1601         case CDROM_SEND_PACKET:
1602         case CDROMCLOSETRAY:
1603         case CDROMEJECT:
1604         case SCSI_IOCTL_GET_IDLUN:
1605         case SCSI_IOCTL_GET_BUS_NUMBER:
1606         default:
1607                 return -ENOTTY;
1608         }
1609 }
1610
1611 static void cciss_check_queues(ctlr_info_t *h)
1612 {
1613         int start_queue = h->next_to_run;
1614         int i;
1615
1616         /* check to see if we have maxed out the number of commands that can
1617          * be placed on the queue.  If so then exit.  We do this check here
1618          * in case the interrupt we serviced was from an ioctl and did not
1619          * free any new commands.
1620          */
1621         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1622                 return;
1623
1624         /* We have room on the queue for more commands.  Now we need to queue
1625          * them up.  We will also keep track of the next queue to run so
1626          * that every queue gets a chance to be started first.
1627          */
1628         for (i = 0; i < h->highest_lun + 1; i++) {
1629                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1630                 /* make sure the disk has been added and the drive is real
1631                  * because this can be called from the middle of init_one.
1632                  */
1633                 if (!h->drv[curr_queue])
1634                         continue;
1635                 if (!(h->drv[curr_queue]->queue) ||
1636                         !(h->drv[curr_queue]->heads))
1637                         continue;
1638                 blk_start_queue(h->gendisk[curr_queue]->queue);
1639
1640                 /* check to see if we have maxed out the number of commands
1641                  * that can be placed on the queue.
1642                  */
1643                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1644                         if (curr_queue == start_queue) {
1645                                 h->next_to_run =
1646                                     (start_queue + 1) % (h->highest_lun + 1);
1647                                 break;
1648                         } else {
1649                                 h->next_to_run = curr_queue;
1650                                 break;
1651                         }
1652                 }
1653         }
1654 }
1655
1656 static void cciss_softirq_done(struct request *rq)
1657 {
1658         CommandList_struct *cmd = rq->completion_data;
1659         ctlr_info_t *h = hba[cmd->ctlr];
1660         unsigned long flags;
1661         u64bit temp64;
1662         int i, ddir;
1663
1664         if (cmd->Request.Type.Direction == XFER_READ)
1665                 ddir = PCI_DMA_FROMDEVICE;
1666         else
1667                 ddir = PCI_DMA_TODEVICE;
1668
1669         /* command did not need to be retried */
1670         /* unmap the DMA mapping for all the scatter gather elements */
1671         for (i = 0; i < cmd->Header.SGList; i++) {
1672                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1673                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1674                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1675         }
1676
1677 #ifdef CCISS_DEBUG
1678         printk("Done with %p\n", rq);
1679 #endif                          /* CCISS_DEBUG */
1680
1681         /* set the residual count for pc requests */
1682         if (blk_pc_request(rq))
1683                 rq->resid_len = cmd->err_info->ResidualCnt;
1684
1685         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1686
1687         spin_lock_irqsave(&h->lock, flags);
1688         cmd_free(h, cmd, 1);
1689         cciss_check_queues(h);
1690         spin_unlock_irqrestore(&h->lock, flags);
1691 }
1692
1693 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1694         unsigned char scsi3addr[], uint32_t log_unit)
1695 {
1696         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1697                 sizeof(h->drv[log_unit]->LunID));
1698 }
1699
1700 /* This function gets the SCSI vendor, model, and revision of a logical drive
1701  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1702  * they cannot be read.
1703  */
1704 static void cciss_get_device_descr(int ctlr, int logvol, int withirq,
1705                                    char *vendor, char *model, char *rev)
1706 {
1707         int rc;
1708         InquiryData_struct *inq_buf;
1709         unsigned char scsi3addr[8];
1710
1711         *vendor = '\0';
1712         *model = '\0';
1713         *rev = '\0';
1714
1715         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1716         if (!inq_buf)
1717                 return;
1718
1719         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1720         if (withirq)
1721                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf,
1722                              sizeof(InquiryData_struct), 0,
1723                                 scsi3addr, TYPE_CMD);
1724         else
1725                 rc = sendcmd(CISS_INQUIRY, ctlr, inq_buf,
1726                              sizeof(InquiryData_struct), 0,
1727                                 scsi3addr, TYPE_CMD);
1728         if (rc == IO_OK) {
1729                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1730                 vendor[VENDOR_LEN] = '\0';
1731                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1732                 model[MODEL_LEN] = '\0';
1733                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1734                 rev[REV_LEN] = '\0';
1735         }
1736
1737         kfree(inq_buf);
1738         return;
1739 }
1740
1741 /* This function gets the serial number of a logical drive via
1742  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1743  * number cannot be had, for whatever reason, 16 bytes of 0xff
1744  * are returned instead.
1745  */
1746 static void cciss_get_serial_no(int ctlr, int logvol, int withirq,
1747                                 unsigned char *serial_no, int buflen)
1748 {
1749 #define PAGE_83_INQ_BYTES 64
1750         int rc;
1751         unsigned char *buf;
1752         unsigned char scsi3addr[8];
1753
1754         if (buflen > 16)
1755                 buflen = 16;
1756         memset(serial_no, 0xff, buflen);
1757         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1758         if (!buf)
1759                 return;
1760         memset(serial_no, 0, buflen);
1761         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1762         if (withirq)
1763                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1764                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1765         else
1766                 rc = sendcmd(CISS_INQUIRY, ctlr, buf,
1767                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1768         if (rc == IO_OK)
1769                 memcpy(serial_no, &buf[8], buflen);
1770         kfree(buf);
1771         return;
1772 }
1773
1774 /*
1775  * cciss_add_disk sets up the block device queue for a logical drive
1776  */
1777 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1778                                 int drv_index)
1779 {
1780         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1781         if (!disk->queue)
1782                 goto init_queue_failure;
1783         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1784         disk->major = h->major;
1785         disk->first_minor = drv_index << NWD_SHIFT;
1786         disk->fops = &cciss_fops;
1787         if (cciss_create_ld_sysfs_entry(h, drv_index))
1788                 goto cleanup_queue;
1789         disk->private_data = h->drv[drv_index];
1790         disk->driverfs_dev = &h->drv[drv_index]->dev;
1791
1792         /* Set up queue information */
1793         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1794
1795         /* This is a hardware imposed limit. */
1796         blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1797
1798         /* This is a limit in the driver and could be eliminated. */
1799         blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1800
1801         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1802
1803         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1804
1805         disk->queue->queuedata = h;
1806
1807         blk_queue_logical_block_size(disk->queue,
1808                                      h->drv[drv_index]->block_size);
1809
1810         /* Make sure all queue data is written out before */
1811         /* setting h->drv[drv_index]->queue, as setting this */
1812         /* allows the interrupt handler to start the queue */
1813         wmb();
1814         h->drv[drv_index]->queue = disk->queue;
1815         add_disk(disk);
1816         return 0;
1817
1818 cleanup_queue:
1819         blk_cleanup_queue(disk->queue);
1820         disk->queue = NULL;
1821 init_queue_failure:
1822         return -1;
1823 }
1824
1825 /* This function will check the usage_count of the drive to be updated/added.
1826  * If the usage_count is zero and it is a heretofore unknown drive, or,
1827  * the drive's capacity, geometry, or serial number has changed,
1828  * then the drive information will be updated and the disk will be
1829  * re-registered with the kernel.  If these conditions don't hold,
1830  * then it will be left alone for the next reboot.  The exception to this
1831  * is disk 0 which will always be left registered with the kernel since it
1832  * is also the controller node.  Any changes to disk 0 will show up on
1833  * the next reboot.
1834  */
1835 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1836         int via_ioctl)
1837 {
1838         ctlr_info_t *h = hba[ctlr];
1839         struct gendisk *disk;
1840         InquiryData_struct *inq_buff = NULL;
1841         unsigned int block_size;
1842         sector_t total_size;
1843         unsigned long flags = 0;
1844         int ret = 0;
1845         drive_info_struct *drvinfo;
1846
1847         /* Get information about the disk and modify the driver structure */
1848         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1849         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1850         if (inq_buff == NULL || drvinfo == NULL)
1851                 goto mem_msg;
1852
1853         /* testing to see if 16-byte CDBs are already being used */
1854         if (h->cciss_read == CCISS_READ_16) {
1855                 cciss_read_capacity_16(h->ctlr, drv_index, 1,
1856                         &total_size, &block_size);
1857
1858         } else {
1859                 cciss_read_capacity(ctlr, drv_index, 1,
1860                                     &total_size, &block_size);
1861
1862                 /* if read_capacity returns all F's this volume is >2TB */
1863                 /* in size so we switch to 16-byte CDB's for all */
1864                 /* read/write ops */
1865                 if (total_size == 0xFFFFFFFFULL) {
1866                         cciss_read_capacity_16(ctlr, drv_index, 1,
1867                         &total_size, &block_size);
1868                         h->cciss_read = CCISS_READ_16;
1869                         h->cciss_write = CCISS_WRITE_16;
1870                 } else {
1871                         h->cciss_read = CCISS_READ_10;
1872                         h->cciss_write = CCISS_WRITE_10;
1873                 }
1874         }
1875
1876         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1877                                inq_buff, drvinfo);
1878         drvinfo->block_size = block_size;
1879         drvinfo->nr_blocks = total_size + 1;
1880
1881         cciss_get_device_descr(ctlr, drv_index, 1, drvinfo->vendor,
1882                                 drvinfo->model, drvinfo->rev);
1883         cciss_get_serial_no(ctlr, drv_index, 1, drvinfo->serial_no,
1884                         sizeof(drvinfo->serial_no));
1885         /* Save the lunid in case we deregister the disk, below. */
1886         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1887                 sizeof(drvinfo->LunID));
1888
1889         /* Is it the same disk we already know, and nothing's changed? */
1890         if (h->drv[drv_index]->raid_level != -1 &&
1891                 ((memcmp(drvinfo->serial_no,
1892                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1893                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1894                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1895                 drvinfo->heads == h->drv[drv_index]->heads &&
1896                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1897                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1898                         /* The disk is unchanged, nothing to update */
1899                         goto freeret;
1900
1901         /* If we get here it's not the same disk, or something's changed,
1902          * so we need to * deregister it, and re-register it, if it's not
1903          * in use.
1904          * If the disk already exists then deregister it before proceeding
1905          * (unless it's the first disk (for the controller node).
1906          */
1907         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1908                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1909                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1910                 h->drv[drv_index]->busy_configuring = 1;
1911                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1912
1913                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1914                  * which keeps the interrupt handler from starting
1915                  * the queue.
1916                  */
1917                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1918         }
1919
1920         /* If the disk is in use return */
1921         if (ret)
1922                 goto freeret;
1923
1924         /* Save the new information from cciss_geometry_inquiry
1925          * and serial number inquiry.  If the disk was deregistered
1926          * above, then h->drv[drv_index] will be NULL.
1927          */
1928         if (h->drv[drv_index] == NULL) {
1929                 drvinfo->device_initialized = 0;
1930                 h->drv[drv_index] = drvinfo;
1931                 drvinfo = NULL; /* so it won't be freed below. */
1932         } else {
1933                 /* special case for cxd0 */
1934                 h->drv[drv_index]->block_size = drvinfo->block_size;
1935                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
1936                 h->drv[drv_index]->heads = drvinfo->heads;
1937                 h->drv[drv_index]->sectors = drvinfo->sectors;
1938                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
1939                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
1940                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
1941                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
1942                         VENDOR_LEN + 1);
1943                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
1944                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
1945         }
1946
1947         ++h->num_luns;
1948         disk = h->gendisk[drv_index];
1949         set_capacity(disk, h->drv[drv_index]->nr_blocks);
1950
1951         /* If it's not disk 0 (drv_index != 0)
1952          * or if it was disk 0, but there was previously
1953          * no actual corresponding configured logical drive
1954          * (raid_leve == -1) then we want to update the
1955          * logical drive's information.
1956          */
1957         if (drv_index || first_time) {
1958                 if (cciss_add_disk(h, disk, drv_index) != 0) {
1959                         cciss_free_gendisk(h, drv_index);
1960                         cciss_free_drive_info(h, drv_index);
1961                         printk(KERN_WARNING "cciss:%d could not update "
1962                                 "disk %d\n", h->ctlr, drv_index);
1963                         --h->num_luns;
1964                 }
1965         }
1966
1967 freeret:
1968         kfree(inq_buff);
1969         kfree(drvinfo);
1970         return;
1971 mem_msg:
1972         printk(KERN_ERR "cciss: out of memory\n");
1973         goto freeret;
1974 }
1975
1976 /* This function will find the first index of the controllers drive array
1977  * that has a null drv pointer and allocate the drive info struct and
1978  * will return that index   This is where new drives will be added.
1979  * If the index to be returned is greater than the highest_lun index for
1980  * the controller then highest_lun is set * to this new index.
1981  * If there are no available indexes or if tha allocation fails, then -1
1982  * is returned.  * "controller_node" is used to know if this is a real
1983  * logical drive, or just the controller node, which determines if this
1984  * counts towards highest_lun.
1985  */
1986 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
1987 {
1988         int i;
1989         drive_info_struct *drv;
1990
1991         /* Search for an empty slot for our drive info */
1992         for (i = 0; i < CISS_MAX_LUN; i++) {
1993
1994                 /* if not cxd0 case, and it's occupied, skip it. */
1995                 if (h->drv[i] && i != 0)
1996                         continue;
1997                 /*
1998                  * If it's cxd0 case, and drv is alloc'ed already, and a
1999                  * disk is configured there, skip it.
2000                  */
2001                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2002                         continue;
2003
2004                 /*
2005                  * We've found an empty slot.  Update highest_lun
2006                  * provided this isn't just the fake cxd0 controller node.
2007                  */
2008                 if (i > h->highest_lun && !controller_node)
2009                         h->highest_lun = i;
2010
2011                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2012                 if (i == 0 && h->drv[i] != NULL)
2013                         return i;
2014
2015                 /*
2016                  * Found an empty slot, not already alloc'ed.  Allocate it.
2017                  * Mark it with raid_level == -1, so we know it's new later on.
2018                  */
2019                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2020                 if (!drv)
2021                         return -1;
2022                 drv->raid_level = -1; /* so we know it's new */
2023                 h->drv[i] = drv;
2024                 return i;
2025         }
2026         return -1;
2027 }
2028
2029 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2030 {
2031         kfree(h->drv[drv_index]);
2032         h->drv[drv_index] = NULL;
2033 }
2034
2035 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2036 {
2037         put_disk(h->gendisk[drv_index]);
2038         h->gendisk[drv_index] = NULL;
2039 }
2040
2041 /* cciss_add_gendisk finds a free hba[]->drv structure
2042  * and allocates a gendisk if needed, and sets the lunid
2043  * in the drvinfo structure.   It returns the index into
2044  * the ->drv[] array, or -1 if none are free.
2045  * is_controller_node indicates whether highest_lun should
2046  * count this disk, or if it's only being added to provide
2047  * a means to talk to the controller in case no logical
2048  * drives have yet been configured.
2049  */
2050 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2051         int controller_node)
2052 {
2053         int drv_index;
2054
2055         drv_index = cciss_alloc_drive_info(h, controller_node);
2056         if (drv_index == -1)
2057                 return -1;
2058
2059         /*Check if the gendisk needs to be allocated */
2060         if (!h->gendisk[drv_index]) {
2061                 h->gendisk[drv_index] =
2062                         alloc_disk(1 << NWD_SHIFT);
2063                 if (!h->gendisk[drv_index]) {
2064                         printk(KERN_ERR "cciss%d: could not "
2065                                 "allocate a new disk %d\n",
2066                                 h->ctlr, drv_index);
2067                         goto err_free_drive_info;
2068                 }
2069         }
2070         memcpy(h->drv[drv_index]->LunID, lunid,
2071                 sizeof(h->drv[drv_index]->LunID));
2072         if (cciss_create_ld_sysfs_entry(h, drv_index))
2073                 goto err_free_disk;
2074         /* Don't need to mark this busy because nobody */
2075         /* else knows about this disk yet to contend */
2076         /* for access to it. */
2077         h->drv[drv_index]->busy_configuring = 0;
2078         wmb();
2079         return drv_index;
2080
2081 err_free_disk:
2082         cciss_free_gendisk(h, drv_index);
2083 err_free_drive_info:
2084         cciss_free_drive_info(h, drv_index);
2085         return -1;
2086 }
2087
2088 /* This is for the special case of a controller which
2089  * has no logical drives.  In this case, we still need
2090  * to register a disk so the controller can be accessed
2091  * by the Array Config Utility.
2092  */
2093 static void cciss_add_controller_node(ctlr_info_t *h)
2094 {
2095         struct gendisk *disk;
2096         int drv_index;
2097
2098         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2099                 return;
2100
2101         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2102         if (drv_index == -1)
2103                 goto error;
2104         h->drv[drv_index]->block_size = 512;
2105         h->drv[drv_index]->nr_blocks = 0;
2106         h->drv[drv_index]->heads = 0;
2107         h->drv[drv_index]->sectors = 0;
2108         h->drv[drv_index]->cylinders = 0;
2109         h->drv[drv_index]->raid_level = -1;
2110         memset(h->drv[drv_index]->serial_no, 0, 16);
2111         disk = h->gendisk[drv_index];
2112         if (cciss_add_disk(h, disk, drv_index) == 0)
2113                 return;
2114         cciss_free_gendisk(h, drv_index);
2115         cciss_free_drive_info(h, drv_index);
2116 error:
2117         printk(KERN_WARNING "cciss%d: could not "
2118                 "add disk 0.\n", h->ctlr);
2119         return;
2120 }
2121
2122 /* This function will add and remove logical drives from the Logical
2123  * drive array of the controller and maintain persistency of ordering
2124  * so that mount points are preserved until the next reboot.  This allows
2125  * for the removal of logical drives in the middle of the drive array
2126  * without a re-ordering of those drives.
2127  * INPUT
2128  * h            = The controller to perform the operations on
2129  */
2130 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2131         int via_ioctl)
2132 {
2133         int ctlr = h->ctlr;
2134         int num_luns;
2135         ReportLunData_struct *ld_buff = NULL;
2136         int return_code;
2137         int listlength = 0;
2138         int i;
2139         int drv_found;
2140         int drv_index = 0;
2141         unsigned char lunid[8] = CTLR_LUNID;
2142         unsigned long flags;
2143
2144         if (!capable(CAP_SYS_RAWIO))
2145                 return -EPERM;
2146
2147         /* Set busy_configuring flag for this operation */
2148         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2149         if (h->busy_configuring) {
2150                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2151                 return -EBUSY;
2152         }
2153         h->busy_configuring = 1;
2154         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2155
2156         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2157         if (ld_buff == NULL)
2158                 goto mem_msg;
2159
2160         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2161                                       sizeof(ReportLunData_struct),
2162                                       0, CTLR_LUNID, TYPE_CMD);
2163
2164         if (return_code == IO_OK)
2165                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2166         else {  /* reading number of logical volumes failed */
2167                 printk(KERN_WARNING "cciss: report logical volume"
2168                        " command failed\n");
2169                 listlength = 0;
2170                 goto freeret;
2171         }
2172
2173         num_luns = listlength / 8;      /* 8 bytes per entry */
2174         if (num_luns > CISS_MAX_LUN) {
2175                 num_luns = CISS_MAX_LUN;
2176                 printk(KERN_WARNING "cciss: more luns configured"
2177                        " on controller than can be handled by"
2178                        " this driver.\n");
2179         }
2180
2181         if (num_luns == 0)
2182                 cciss_add_controller_node(h);
2183
2184         /* Compare controller drive array to driver's drive array
2185          * to see if any drives are missing on the controller due
2186          * to action of Array Config Utility (user deletes drive)
2187          * and deregister logical drives which have disappeared.
2188          */
2189         for (i = 0; i <= h->highest_lun; i++) {
2190                 int j;
2191                 drv_found = 0;
2192
2193                 /* skip holes in the array from already deleted drives */
2194                 if (h->drv[i] == NULL)
2195                         continue;
2196
2197                 for (j = 0; j < num_luns; j++) {
2198                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2199                         if (memcmp(h->drv[i]->LunID, lunid,
2200                                 sizeof(lunid)) == 0) {
2201                                 drv_found = 1;
2202                                 break;
2203                         }
2204                 }
2205                 if (!drv_found) {
2206                         /* Deregister it from the OS, it's gone. */
2207                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2208                         h->drv[i]->busy_configuring = 1;
2209                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2210                         return_code = deregister_disk(h, i, 1, via_ioctl);
2211                         if (h->drv[i] != NULL)
2212                                 h->drv[i]->busy_configuring = 0;
2213                 }
2214         }
2215
2216         /* Compare controller drive array to driver's drive array.
2217          * Check for updates in the drive information and any new drives
2218          * on the controller due to ACU adding logical drives, or changing
2219          * a logical drive's size, etc.  Reregister any new/changed drives
2220          */
2221         for (i = 0; i < num_luns; i++) {
2222                 int j;
2223
2224                 drv_found = 0;
2225
2226                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2227                 /* Find if the LUN is already in the drive array
2228                  * of the driver.  If so then update its info
2229                  * if not in use.  If it does not exist then find
2230                  * the first free index and add it.
2231                  */
2232                 for (j = 0; j <= h->highest_lun; j++) {
2233                         if (h->drv[j] != NULL &&
2234                                 memcmp(h->drv[j]->LunID, lunid,
2235                                         sizeof(h->drv[j]->LunID)) == 0) {
2236                                 drv_index = j;
2237                                 drv_found = 1;
2238                                 break;
2239                         }
2240                 }
2241
2242                 /* check if the drive was found already in the array */
2243                 if (!drv_found) {
2244                         drv_index = cciss_add_gendisk(h, lunid, 0);
2245                         if (drv_index == -1)
2246                                 goto freeret;
2247                 }
2248                 cciss_update_drive_info(ctlr, drv_index, first_time,
2249                         via_ioctl);
2250         }               /* end for */
2251
2252 freeret:
2253         kfree(ld_buff);
2254         h->busy_configuring = 0;
2255         /* We return -1 here to tell the ACU that we have registered/updated
2256          * all of the drives that we can and to keep it from calling us
2257          * additional times.
2258          */
2259         return -1;
2260 mem_msg:
2261         printk(KERN_ERR "cciss: out of memory\n");
2262         h->busy_configuring = 0;
2263         goto freeret;
2264 }
2265
2266 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2267 {
2268         /* zero out the disk size info */
2269         drive_info->nr_blocks = 0;
2270         drive_info->block_size = 0;
2271         drive_info->heads = 0;
2272         drive_info->sectors = 0;
2273         drive_info->cylinders = 0;
2274         drive_info->raid_level = -1;
2275         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2276         memset(drive_info->model, 0, sizeof(drive_info->model));
2277         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2278         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2279         /*
2280          * don't clear the LUNID though, we need to remember which
2281          * one this one is.
2282          */
2283 }
2284
2285 /* This function will deregister the disk and it's queue from the
2286  * kernel.  It must be called with the controller lock held and the
2287  * drv structures busy_configuring flag set.  It's parameters are:
2288  *
2289  * disk = This is the disk to be deregistered
2290  * drv  = This is the drive_info_struct associated with the disk to be
2291  *        deregistered.  It contains information about the disk used
2292  *        by the driver.
2293  * clear_all = This flag determines whether or not the disk information
2294  *             is going to be completely cleared out and the highest_lun
2295  *             reset.  Sometimes we want to clear out information about
2296  *             the disk in preparation for re-adding it.  In this case
2297  *             the highest_lun should be left unchanged and the LunID
2298  *             should not be cleared.
2299  * via_ioctl
2300  *    This indicates whether we've reached this path via ioctl.
2301  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2302  *    If this path is reached via ioctl(), then the max_usage_count will
2303  *    be 1, as the process calling ioctl() has got to have the device open.
2304  *    If we get here via sysfs, then the max usage count will be zero.
2305 */
2306 static int deregister_disk(ctlr_info_t *h, int drv_index,
2307                            int clear_all, int via_ioctl)
2308 {
2309         int i;
2310         struct gendisk *disk;
2311         drive_info_struct *drv;
2312         int recalculate_highest_lun;
2313
2314         if (!capable(CAP_SYS_RAWIO))
2315                 return -EPERM;
2316
2317         drv = h->drv[drv_index];
2318         disk = h->gendisk[drv_index];
2319
2320         /* make sure logical volume is NOT is use */
2321         if (clear_all || (h->gendisk[0] == disk)) {
2322                 if (drv->usage_count > via_ioctl)
2323                         return -EBUSY;
2324         } else if (drv->usage_count > 0)
2325                 return -EBUSY;
2326
2327         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2328
2329         /* invalidate the devices and deregister the disk.  If it is disk
2330          * zero do not deregister it but just zero out it's values.  This
2331          * allows us to delete disk zero but keep the controller registered.
2332          */
2333         if (h->gendisk[0] != disk) {
2334                 struct request_queue *q = disk->queue;
2335                 if (disk->flags & GENHD_FL_UP) {
2336                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2337                         del_gendisk(disk);
2338                 }
2339                 if (q)
2340                         blk_cleanup_queue(q);
2341                 /* If clear_all is set then we are deleting the logical
2342                  * drive, not just refreshing its info.  For drives
2343                  * other than disk 0 we will call put_disk.  We do not
2344                  * do this for disk 0 as we need it to be able to
2345                  * configure the controller.
2346                  */
2347                 if (clear_all){
2348                         /* This isn't pretty, but we need to find the
2349                          * disk in our array and NULL our the pointer.
2350                          * This is so that we will call alloc_disk if
2351                          * this index is used again later.
2352                          */
2353                         for (i=0; i < CISS_MAX_LUN; i++){
2354                                 if (h->gendisk[i] == disk) {
2355                                         h->gendisk[i] = NULL;
2356                                         break;
2357                                 }
2358                         }
2359                         put_disk(disk);
2360                 }
2361         } else {
2362                 set_capacity(disk, 0);
2363                 cciss_clear_drive_info(drv);
2364         }
2365
2366         --h->num_luns;
2367
2368         /* if it was the last disk, find the new hightest lun */
2369         if (clear_all && recalculate_highest_lun) {
2370                 int i, newhighest = -1;
2371                 for (i = 0; i <= h->highest_lun; i++) {
2372                         /* if the disk has size > 0, it is available */
2373                         if (h->drv[i] && h->drv[i]->heads)
2374                                 newhighest = i;
2375                 }
2376                 h->highest_lun = newhighest;
2377         }
2378         return 0;
2379 }
2380
2381 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2382                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2383                 int cmd_type)
2384 {
2385         ctlr_info_t *h = hba[ctlr];
2386         u64bit buff_dma_handle;
2387         int status = IO_OK;
2388
2389         c->cmd_type = CMD_IOCTL_PEND;
2390         c->Header.ReplyQueue = 0;
2391         if (buff != NULL) {
2392                 c->Header.SGList = 1;
2393                 c->Header.SGTotal = 1;
2394         } else {
2395                 c->Header.SGList = 0;
2396                 c->Header.SGTotal = 0;
2397         }
2398         c->Header.Tag.lower = c->busaddr;
2399         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2400
2401         c->Request.Type.Type = cmd_type;
2402         if (cmd_type == TYPE_CMD) {
2403                 switch (cmd) {
2404                 case CISS_INQUIRY:
2405                         /* are we trying to read a vital product page */
2406                         if (page_code != 0) {
2407                                 c->Request.CDB[1] = 0x01;
2408                                 c->Request.CDB[2] = page_code;
2409                         }
2410                         c->Request.CDBLen = 6;
2411                         c->Request.Type.Attribute = ATTR_SIMPLE;
2412                         c->Request.Type.Direction = XFER_READ;
2413                         c->Request.Timeout = 0;
2414                         c->Request.CDB[0] = CISS_INQUIRY;
2415                         c->Request.CDB[4] = size & 0xFF;
2416                         break;
2417                 case CISS_REPORT_LOG:
2418                 case CISS_REPORT_PHYS:
2419                         /* Talking to controller so It's a physical command
2420                            mode = 00 target = 0.  Nothing to write.
2421                          */
2422                         c->Request.CDBLen = 12;
2423                         c->Request.Type.Attribute = ATTR_SIMPLE;
2424                         c->Request.Type.Direction = XFER_READ;
2425                         c->Request.Timeout = 0;
2426                         c->Request.CDB[0] = cmd;
2427                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
2428                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2429                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2430                         c->Request.CDB[9] = size & 0xFF;
2431                         break;
2432
2433                 case CCISS_READ_CAPACITY:
2434                         c->Request.CDBLen = 10;
2435                         c->Request.Type.Attribute = ATTR_SIMPLE;
2436                         c->Request.Type.Direction = XFER_READ;
2437                         c->Request.Timeout = 0;
2438                         c->Request.CDB[0] = cmd;
2439                         break;
2440                 case CCISS_READ_CAPACITY_16:
2441                         c->Request.CDBLen = 16;
2442                         c->Request.Type.Attribute = ATTR_SIMPLE;
2443                         c->Request.Type.Direction = XFER_READ;
2444                         c->Request.Timeout = 0;
2445                         c->Request.CDB[0] = cmd;
2446                         c->Request.CDB[1] = 0x10;
2447                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2448                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2449                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2450                         c->Request.CDB[13] = size & 0xFF;
2451                         c->Request.Timeout = 0;
2452                         c->Request.CDB[0] = cmd;
2453                         break;
2454                 case CCISS_CACHE_FLUSH:
2455                         c->Request.CDBLen = 12;
2456                         c->Request.Type.Attribute = ATTR_SIMPLE;
2457                         c->Request.Type.Direction = XFER_WRITE;
2458                         c->Request.Timeout = 0;
2459                         c->Request.CDB[0] = BMIC_WRITE;
2460                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2461                         break;
2462                 case TEST_UNIT_READY:
2463                         c->Request.CDBLen = 6;
2464                         c->Request.Type.Attribute = ATTR_SIMPLE;
2465                         c->Request.Type.Direction = XFER_NONE;
2466                         c->Request.Timeout = 0;
2467                         break;
2468                 default:
2469                         printk(KERN_WARNING
2470                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2471                         return IO_ERROR;
2472                 }
2473         } else if (cmd_type == TYPE_MSG) {
2474                 switch (cmd) {
2475                 case 0: /* ABORT message */
2476                         c->Request.CDBLen = 12;
2477                         c->Request.Type.Attribute = ATTR_SIMPLE;
2478                         c->Request.Type.Direction = XFER_WRITE;
2479                         c->Request.Timeout = 0;
2480                         c->Request.CDB[0] = cmd;        /* abort */
2481                         c->Request.CDB[1] = 0;  /* abort a command */
2482                         /* buff contains the tag of the command to abort */
2483                         memcpy(&c->Request.CDB[4], buff, 8);
2484                         break;
2485                 case 1: /* RESET message */
2486                         c->Request.CDBLen = 16;
2487                         c->Request.Type.Attribute = ATTR_SIMPLE;
2488                         c->Request.Type.Direction = XFER_NONE;
2489                         c->Request.Timeout = 0;
2490                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2491                         c->Request.CDB[0] = cmd;        /* reset */
2492                         c->Request.CDB[1] = 0x03;       /* reset a target */
2493                         break;
2494                 case 3: /* No-Op message */
2495                         c->Request.CDBLen = 1;
2496                         c->Request.Type.Attribute = ATTR_SIMPLE;
2497                         c->Request.Type.Direction = XFER_WRITE;
2498                         c->Request.Timeout = 0;
2499                         c->Request.CDB[0] = cmd;
2500                         break;
2501                 default:
2502                         printk(KERN_WARNING
2503                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2504                         return IO_ERROR;
2505                 }
2506         } else {
2507                 printk(KERN_WARNING
2508                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2509                 return IO_ERROR;
2510         }
2511         /* Fill in the scatter gather information */
2512         if (size > 0) {
2513                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2514                                                              buff, size,
2515                                                              PCI_DMA_BIDIRECTIONAL);
2516                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2517                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2518                 c->SG[0].Len = size;
2519                 c->SG[0].Ext = 0;       /* we are not chaining */
2520         }
2521         return status;
2522 }
2523
2524 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2525 {
2526         switch (c->err_info->ScsiStatus) {
2527         case SAM_STAT_GOOD:
2528                 return IO_OK;
2529         case SAM_STAT_CHECK_CONDITION:
2530                 switch (0xf & c->err_info->SenseInfo[2]) {
2531                 case 0: return IO_OK; /* no sense */
2532                 case 1: return IO_OK; /* recovered error */
2533                 default:
2534                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2535                                 "check condition, sense key = 0x%02x\n",
2536                                 h->ctlr, c->Request.CDB[0],
2537                                 c->err_info->SenseInfo[2]);
2538                 }
2539                 break;
2540         default:
2541                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2542                         "scsi status = 0x%02x\n", h->ctlr,
2543                         c->Request.CDB[0], c->err_info->ScsiStatus);
2544                 break;
2545         }
2546         return IO_ERROR;
2547 }
2548
2549 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2550 {
2551         int return_status = IO_OK;
2552
2553         if (c->err_info->CommandStatus == CMD_SUCCESS)
2554                 return IO_OK;
2555
2556         switch (c->err_info->CommandStatus) {
2557         case CMD_TARGET_STATUS:
2558                 return_status = check_target_status(h, c);
2559                 break;
2560         case CMD_DATA_UNDERRUN:
2561         case CMD_DATA_OVERRUN:
2562                 /* expected for inquiry and report lun commands */
2563                 break;
2564         case CMD_INVALID:
2565                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2566                        "reported invalid\n", c->Request.CDB[0]);
2567                 return_status = IO_ERROR;
2568                 break;
2569         case CMD_PROTOCOL_ERR:
2570                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2571                        "protocol error \n", c->Request.CDB[0]);
2572                 return_status = IO_ERROR;
2573                 break;
2574         case CMD_HARDWARE_ERR:
2575                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2576                        " hardware error\n", c->Request.CDB[0]);
2577                 return_status = IO_ERROR;
2578                 break;
2579         case CMD_CONNECTION_LOST:
2580                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2581                        "connection lost\n", c->Request.CDB[0]);
2582                 return_status = IO_ERROR;
2583                 break;
2584         case CMD_ABORTED:
2585                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2586                        "aborted\n", c->Request.CDB[0]);
2587                 return_status = IO_ERROR;
2588                 break;
2589         case CMD_ABORT_FAILED:
2590                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2591                        "abort failed\n", c->Request.CDB[0]);
2592                 return_status = IO_ERROR;
2593                 break;
2594         case CMD_UNSOLICITED_ABORT:
2595                 printk(KERN_WARNING
2596                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2597                         c->Request.CDB[0]);
2598                 return_status = IO_NEEDS_RETRY;
2599                 break;
2600         default:
2601                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2602                        "unknown status %x\n", c->Request.CDB[0],
2603                        c->err_info->CommandStatus);
2604                 return_status = IO_ERROR;
2605         }
2606         return return_status;
2607 }
2608
2609 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2610         int attempt_retry)
2611 {
2612         DECLARE_COMPLETION_ONSTACK(wait);
2613         u64bit buff_dma_handle;
2614         unsigned long flags;
2615         int return_status = IO_OK;
2616
2617 resend_cmd2:
2618         c->waiting = &wait;
2619         /* Put the request on the tail of the queue and send it */
2620         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2621         addQ(&h->reqQ, c);
2622         h->Qdepth++;
2623         start_io(h);
2624         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2625
2626         wait_for_completion(&wait);
2627
2628         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2629                 goto command_done;
2630
2631         return_status = process_sendcmd_error(h, c);
2632
2633         if (return_status == IO_NEEDS_RETRY &&
2634                 c->retry_count < MAX_CMD_RETRIES) {
2635                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2636                         c->Request.CDB[0]);
2637                 c->retry_count++;
2638                 /* erase the old error information */
2639                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2640                 return_status = IO_OK;
2641                 INIT_COMPLETION(wait);
2642                 goto resend_cmd2;
2643         }
2644
2645 command_done:
2646         /* unlock the buffers from DMA */
2647         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2648         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2649         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2650                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2651         return return_status;
2652 }
2653
2654 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2655                            __u8 page_code, unsigned char scsi3addr[],
2656                         int cmd_type)
2657 {
2658         ctlr_info_t *h = hba[ctlr];
2659         CommandList_struct *c;
2660         int return_status;
2661
2662         c = cmd_alloc(h, 0);
2663         if (!c)
2664                 return -ENOMEM;
2665         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2666                 scsi3addr, cmd_type);
2667         if (return_status == IO_OK)
2668                 return_status = sendcmd_withirq_core(h, c, 1);
2669
2670         cmd_free(h, c, 0);
2671         return return_status;
2672 }
2673
2674 static void cciss_geometry_inquiry(int ctlr, int logvol,
2675                                    int withirq, sector_t total_size,
2676                                    unsigned int block_size,
2677                                    InquiryData_struct *inq_buff,
2678                                    drive_info_struct *drv)
2679 {
2680         int return_code;
2681         unsigned long t;
2682         unsigned char scsi3addr[8];
2683
2684         memset(inq_buff, 0, sizeof(InquiryData_struct));
2685         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2686         if (withirq)
2687                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
2688                                               inq_buff, sizeof(*inq_buff),
2689                                               0xC1, scsi3addr, TYPE_CMD);
2690         else
2691                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
2692                                       sizeof(*inq_buff), 0xC1, scsi3addr,
2693                                       TYPE_CMD);
2694         if (return_code == IO_OK) {
2695                 if (inq_buff->data_byte[8] == 0xFF) {
2696                         printk(KERN_WARNING
2697                                "cciss: reading geometry failed, volume "
2698                                "does not support reading geometry\n");
2699                         drv->heads = 255;
2700                         drv->sectors = 32;      // Sectors per track
2701                         drv->cylinders = total_size + 1;
2702                         drv->raid_level = RAID_UNKNOWN;
2703                 } else {
2704                         drv->heads = inq_buff->data_byte[6];
2705                         drv->sectors = inq_buff->data_byte[7];
2706                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2707                         drv->cylinders += inq_buff->data_byte[5];
2708                         drv->raid_level = inq_buff->data_byte[8];
2709                 }
2710                 drv->block_size = block_size;
2711                 drv->nr_blocks = total_size + 1;
2712                 t = drv->heads * drv->sectors;
2713                 if (t > 1) {
2714                         sector_t real_size = total_size + 1;
2715                         unsigned long rem = sector_div(real_size, t);
2716                         if (rem)
2717                                 real_size++;
2718                         drv->cylinders = real_size;
2719                 }
2720         } else {                /* Get geometry failed */
2721                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2722         }
2723 }
2724
2725 static void
2726 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
2727                     unsigned int *block_size)
2728 {
2729         ReadCapdata_struct *buf;
2730         int return_code;
2731         unsigned char scsi3addr[8];
2732
2733         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2734         if (!buf) {
2735                 printk(KERN_WARNING "cciss: out of memory\n");
2736                 return;
2737         }
2738
2739         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2740         if (withirq)
2741                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
2742                                 ctlr, buf, sizeof(ReadCapdata_struct),
2743                                         0, scsi3addr, TYPE_CMD);
2744         else
2745                 return_code = sendcmd(CCISS_READ_CAPACITY,
2746                                 ctlr, buf, sizeof(ReadCapdata_struct),
2747                                         0, scsi3addr, TYPE_CMD);
2748         if (return_code == IO_OK) {
2749                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2750                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2751         } else {                /* read capacity command failed */
2752                 printk(KERN_WARNING "cciss: read capacity failed\n");
2753                 *total_size = 0;
2754                 *block_size = BLOCK_SIZE;
2755         }
2756         kfree(buf);
2757 }
2758
2759 static void
2760 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size,                                 unsigned int *block_size)
2761 {
2762         ReadCapdata_struct_16 *buf;
2763         int return_code;
2764         unsigned char scsi3addr[8];
2765
2766         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2767         if (!buf) {
2768                 printk(KERN_WARNING "cciss: out of memory\n");
2769                 return;
2770         }
2771
2772         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2773         if (withirq) {
2774                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2775                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2776                                 0, scsi3addr, TYPE_CMD);
2777         }
2778         else {
2779                 return_code = sendcmd(CCISS_READ_CAPACITY_16,
2780                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2781                                 0, scsi3addr, TYPE_CMD);
2782         }
2783         if (return_code == IO_OK) {
2784                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2785                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2786         } else {                /* read capacity command failed */
2787                 printk(KERN_WARNING "cciss: read capacity failed\n");
2788                 *total_size = 0;
2789                 *block_size = BLOCK_SIZE;
2790         }
2791         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2792                (unsigned long long)*total_size+1, *block_size);
2793         kfree(buf);
2794 }
2795
2796 static int cciss_revalidate(struct gendisk *disk)
2797 {
2798         ctlr_info_t *h = get_host(disk);
2799         drive_info_struct *drv = get_drv(disk);
2800         int logvol;
2801         int FOUND = 0;
2802         unsigned int block_size;
2803         sector_t total_size;
2804         InquiryData_struct *inq_buff = NULL;
2805
2806         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2807                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2808                         sizeof(drv->LunID)) == 0) {
2809                         FOUND = 1;
2810                         break;
2811                 }
2812         }
2813
2814         if (!FOUND)
2815                 return 1;
2816
2817         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2818         if (inq_buff == NULL) {
2819                 printk(KERN_WARNING "cciss: out of memory\n");
2820                 return 1;
2821         }
2822         if (h->cciss_read == CCISS_READ_10) {
2823                 cciss_read_capacity(h->ctlr, logvol, 1,
2824                                         &total_size, &block_size);
2825         } else {
2826                 cciss_read_capacity_16(h->ctlr, logvol, 1,
2827                                         &total_size, &block_size);
2828         }
2829         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2830                                inq_buff, drv);
2831
2832         blk_queue_logical_block_size(drv->queue, drv->block_size);
2833         set_capacity(disk, drv->nr_blocks);
2834
2835         kfree(inq_buff);
2836         return 0;
2837 }
2838
2839 /*
2840  *   Wait polling for a command to complete.
2841  *   The memory mapped FIFO is polled for the completion.
2842  *   Used only at init time, interrupts from the HBA are disabled.
2843  */
2844 static unsigned long pollcomplete(int ctlr)
2845 {
2846         unsigned long done;
2847         int i;
2848
2849         /* Wait (up to 20 seconds) for a command to complete */
2850
2851         for (i = 20 * HZ; i > 0; i--) {
2852                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2853                 if (done == FIFO_EMPTY)
2854                         schedule_timeout_uninterruptible(1);
2855                 else
2856                         return done;
2857         }
2858         /* Invalid address to tell caller we ran out of time */
2859         return 1;
2860 }
2861
2862 /* Send command c to controller h and poll for it to complete.
2863  * Turns interrupts off on the board.  Used at driver init time
2864  * and during SCSI error recovery.
2865  */
2866 static int sendcmd_core(ctlr_info_t *h, CommandList_struct *c)
2867 {
2868         int i;
2869         unsigned long complete;
2870         int status = IO_ERROR;
2871         u64bit buff_dma_handle;
2872
2873 resend_cmd1:
2874
2875         /* Disable interrupt on the board. */
2876         h->access.set_intr_mask(h, CCISS_INTR_OFF);
2877
2878         /* Make sure there is room in the command FIFO */
2879         /* Actually it should be completely empty at this time */
2880         /* unless we are in here doing error handling for the scsi */
2881         /* tape side of the driver. */
2882         for (i = 200000; i > 0; i--) {
2883                 /* if fifo isn't full go */
2884                 if (!(h->access.fifo_full(h)))
2885                         break;
2886                 udelay(10);
2887                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2888                        " waiting!\n", h->ctlr);
2889         }
2890         h->access.submit_command(h, c); /* Send the cmd */
2891         do {
2892                 complete = pollcomplete(h->ctlr);
2893
2894 #ifdef CCISS_DEBUG
2895                 printk(KERN_DEBUG "cciss: command completed\n");
2896 #endif                          /* CCISS_DEBUG */
2897
2898                 if (complete == 1) {
2899                         printk(KERN_WARNING
2900                                "cciss cciss%d: SendCmd Timeout out, "
2901                                "No command list address returned!\n", h->ctlr);
2902                         status = IO_ERROR;
2903                         break;
2904                 }
2905
2906                 /* Make sure it's the command we're expecting. */
2907                 if ((complete & ~CISS_ERROR_BIT) != c->busaddr) {
2908                         printk(KERN_WARNING "cciss%d: Unexpected command "
2909                                 "completion.\n", h->ctlr);
2910                         continue;
2911                 }
2912
2913                 /* It is our command.  If no error, we're done. */
2914                 if (!(complete & CISS_ERROR_BIT)) {
2915                         status = IO_OK;
2916                         break;
2917                 }
2918
2919                 /* There is an error... */
2920
2921                 /* if data overrun or underun on Report command ignore it */
2922                 if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2923                      (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2924                      (c->Request.CDB[0] == CISS_INQUIRY)) &&
2925                         ((c->err_info->CommandStatus == CMD_DATA_OVERRUN) ||
2926                          (c->err_info->CommandStatus == CMD_DATA_UNDERRUN))) {
2927                         complete = c->busaddr;
2928                         status = IO_OK;
2929                         break;
2930                 }
2931
2932                 if (c->err_info->CommandStatus == CMD_UNSOLICITED_ABORT) {
2933                         printk(KERN_WARNING "cciss%d: unsolicited abort %p\n",
2934                                 h->ctlr, c);
2935                         if (c->retry_count < MAX_CMD_RETRIES) {
2936                                 printk(KERN_WARNING "cciss%d: retrying %p\n",
2937                                    h->ctlr, c);
2938                                 c->retry_count++;
2939                                 /* erase the old error information */
2940                                 memset(c->err_info, 0, sizeof(c->err_info));
2941                                 goto resend_cmd1;
2942                         }
2943                         printk(KERN_WARNING "cciss%d: retried %p too many "
2944                                 "times\n", h->ctlr, c);
2945                         status = IO_ERROR;
2946                         break;
2947                 }
2948
2949                 if (c->err_info->CommandStatus == CMD_UNABORTABLE) {
2950                         printk(KERN_WARNING "cciss%d: command could not be "
2951                                 "aborted.\n", h->ctlr);
2952                         status = IO_ERROR;
2953                         break;
2954                 }
2955
2956                 if (c->err_info->CommandStatus == CMD_TARGET_STATUS) {
2957                         status = check_target_status(h, c);
2958                         break;
2959                 }
2960
2961                 printk(KERN_WARNING "cciss%d: sendcmd error\n", h->ctlr);
2962                 printk(KERN_WARNING "cmd = 0x%02x, CommandStatus = 0x%02x\n",
2963                         c->Request.CDB[0], c->err_info->CommandStatus);
2964                 status = IO_ERROR;
2965                 break;
2966
2967         } while (1);
2968
2969         /* unlock the data buffer from DMA */
2970         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2971         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2972         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2973                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2974         return status;
2975 }
2976
2977 /*
2978  * Send a command to the controller, and wait for it to complete.
2979  * Used at init time, and during SCSI error recovery.
2980  */
2981 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
2982         __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2983 {
2984         CommandList_struct *c;
2985         int status;
2986
2987         c = cmd_alloc(hba[ctlr], 1);
2988         if (!c) {
2989                 printk(KERN_WARNING "cciss: unable to get memory");
2990                 return IO_ERROR;
2991         }
2992         status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2993                 scsi3addr, cmd_type);
2994         if (status == IO_OK)
2995                 status = sendcmd_core(hba[ctlr], c);
2996         cmd_free(hba[ctlr], c, 1);
2997         return status;
2998 }
2999
3000 /*
3001  * Map (physical) PCI mem into (virtual) kernel space
3002  */
3003 static void __iomem *remap_pci_mem(ulong base, ulong size)
3004 {
3005         ulong page_base = ((ulong) base) & PAGE_MASK;
3006         ulong page_offs = ((ulong) base) - page_base;
3007         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
3008
3009         return page_remapped ? (page_remapped + page_offs) : NULL;
3010 }
3011
3012 /*
3013  * Takes jobs of the Q and sends them to the hardware, then puts it on
3014  * the Q to wait for completion.
3015  */
3016 static void start_io(ctlr_info_t *h)
3017 {
3018         CommandList_struct *c;
3019
3020         while (!hlist_empty(&h->reqQ)) {
3021                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
3022                 /* can't do anything if fifo is full */
3023                 if ((h->access.fifo_full(h))) {
3024                         printk(KERN_WARNING "cciss: fifo full\n");
3025                         break;
3026                 }
3027
3028                 /* Get the first entry from the Request Q */
3029                 removeQ(c);
3030                 h->Qdepth--;
3031
3032                 /* Tell the controller execute command */
3033                 h->access.submit_command(h, c);
3034
3035                 /* Put job onto the completed Q */
3036                 addQ(&h->cmpQ, c);
3037         }
3038 }
3039
3040 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
3041 /* Zeros out the error record and then resends the command back */
3042 /* to the controller */
3043 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
3044 {
3045         /* erase the old error information */
3046         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
3047
3048         /* add it to software queue and then send it to the controller */
3049         addQ(&h->reqQ, c);
3050         h->Qdepth++;
3051         if (h->Qdepth > h->maxQsinceinit)
3052                 h->maxQsinceinit = h->Qdepth;
3053
3054         start_io(h);
3055 }
3056
3057 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
3058         unsigned int msg_byte, unsigned int host_byte,
3059         unsigned int driver_byte)
3060 {
3061         /* inverse of macros in scsi.h */
3062         return (scsi_status_byte & 0xff) |
3063                 ((msg_byte & 0xff) << 8) |
3064                 ((host_byte & 0xff) << 16) |
3065                 ((driver_byte & 0xff) << 24);
3066 }
3067
3068 static inline int evaluate_target_status(ctlr_info_t *h,
3069                         CommandList_struct *cmd, int *retry_cmd)
3070 {
3071         unsigned char sense_key;
3072         unsigned char status_byte, msg_byte, host_byte, driver_byte;
3073         int error_value;
3074
3075         *retry_cmd = 0;
3076         /* If we get in here, it means we got "target status", that is, scsi status */
3077         status_byte = cmd->err_info->ScsiStatus;
3078         driver_byte = DRIVER_OK;
3079         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3080
3081         if (blk_pc_request(cmd->rq))
3082                 host_byte = DID_PASSTHROUGH;
3083         else
3084                 host_byte = DID_OK;
3085
3086         error_value = make_status_bytes(status_byte, msg_byte,
3087                 host_byte, driver_byte);
3088
3089         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3090                 if (!blk_pc_request(cmd->rq))
3091                         printk(KERN_WARNING "cciss: cmd %p "
3092                                "has SCSI Status 0x%x\n",
3093                                cmd, cmd->err_info->ScsiStatus);
3094                 return error_value;
3095         }
3096
3097         /* check the sense key */
3098         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3099         /* no status or recovered error */
3100         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
3101                 error_value = 0;
3102
3103         if (check_for_unit_attention(h, cmd)) {
3104                 *retry_cmd = !blk_pc_request(cmd->rq);
3105                 return 0;
3106         }
3107
3108         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
3109                 if (error_value != 0)
3110                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3111                                " sense key = 0x%x\n", cmd, sense_key);
3112                 return error_value;
3113         }
3114
3115         /* SG_IO or similar, copy sense data back */
3116         if (cmd->rq->sense) {
3117                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3118                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3119                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3120                         cmd->rq->sense_len);
3121         } else
3122                 cmd->rq->sense_len = 0;
3123
3124         return error_value;
3125 }
3126
3127 /* checks the status of the job and calls complete buffers to mark all
3128  * buffers for the completed job. Note that this function does not need
3129  * to hold the hba/queue lock.
3130  */
3131 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3132                                     int timeout)
3133 {
3134         int retry_cmd = 0;
3135         struct request *rq = cmd->rq;
3136
3137         rq->errors = 0;
3138
3139         if (timeout)
3140                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3141
3142         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3143                 goto after_error_processing;
3144
3145         switch (cmd->err_info->CommandStatus) {
3146         case CMD_TARGET_STATUS:
3147                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3148                 break;
3149         case CMD_DATA_UNDERRUN:
3150                 if (blk_fs_request(cmd->rq)) {
3151                         printk(KERN_WARNING "cciss: cmd %p has"
3152                                " completed with data underrun "
3153                                "reported\n", cmd);
3154                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3155                 }
3156                 break;
3157         case CMD_DATA_OVERRUN:
3158                 if (blk_fs_request(cmd->rq))
3159                         printk(KERN_WARNING "cciss: cmd %p has"
3160                                " completed with data overrun "
3161                                "reported\n", cmd);
3162                 break;
3163         case CMD_INVALID:
3164                 printk(KERN_WARNING "cciss: cmd %p is "
3165                        "reported invalid\n", cmd);
3166                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3167                         cmd->err_info->CommandStatus, DRIVER_OK,
3168                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3169                 break;
3170         case CMD_PROTOCOL_ERR:
3171                 printk(KERN_WARNING "cciss: cmd %p has "
3172                        "protocol error \n", cmd);
3173                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3174                         cmd->err_info->CommandStatus, DRIVER_OK,
3175                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3176                 break;
3177         case CMD_HARDWARE_ERR:
3178                 printk(KERN_WARNING "cciss: cmd %p had "
3179                        " hardware error\n", cmd);
3180                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3181                         cmd->err_info->CommandStatus, DRIVER_OK,
3182                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3183                 break;
3184         case CMD_CONNECTION_LOST:
3185                 printk(KERN_WARNING "cciss: cmd %p had "
3186                        "connection lost\n", cmd);
3187                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3188                         cmd->err_info->CommandStatus, DRIVER_OK,
3189                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3190                 break;
3191         case CMD_ABORTED:
3192                 printk(KERN_WARNING "cciss: cmd %p was "
3193                        "aborted\n", cmd);
3194                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3195                         cmd->err_info->CommandStatus, DRIVER_OK,
3196                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3197                 break;
3198         case CMD_ABORT_FAILED:
3199                 printk(KERN_WARNING "cciss: cmd %p reports "
3200                        "abort failed\n", cmd);
3201                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3202                         cmd->err_info->CommandStatus, DRIVER_OK,
3203                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3204                 break;
3205         case CMD_UNSOLICITED_ABORT:
3206                 printk(KERN_WARNING "cciss%d: unsolicited "
3207                        "abort %p\n", h->ctlr, cmd);
3208                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3209                         retry_cmd = 1;
3210                         printk(KERN_WARNING
3211                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3212                         cmd->retry_count++;
3213                 } else
3214                         printk(KERN_WARNING
3215                                "cciss%d: %p retried too "
3216                                "many times\n", h->ctlr, cmd);
3217                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3218                         cmd->err_info->CommandStatus, DRIVER_OK,
3219                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3220                 break;
3221         case CMD_TIMEOUT:
3222                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3223                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3224                         cmd->err_info->CommandStatus, DRIVER_OK,
3225                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3226                 break;
3227         default:
3228                 printk(KERN_WARNING "cciss: cmd %p returned "
3229                        "unknown status %x\n", cmd,
3230                        cmd->err_info->CommandStatus);
3231                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3232                         cmd->err_info->CommandStatus, DRIVER_OK,
3233                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3234         }
3235
3236 after_error_processing:
3237
3238         /* We need to return this command */
3239         if (retry_cmd) {
3240                 resend_cciss_cmd(h, cmd);
3241                 return;
3242         }
3243         cmd->rq->completion_data = cmd;
3244         blk_complete_request(cmd->rq);
3245 }
3246
3247 /*
3248  * Get a request and submit it to the controller.
3249  */
3250 static void do_cciss_request(struct request_queue *q)
3251 {
3252         ctlr_info_t *h = q->queuedata;
3253         CommandList_struct *c;
3254         sector_t start_blk;
3255         int seg;
3256         struct request *creq;
3257         u64bit temp64;
3258         struct scatterlist tmp_sg[MAXSGENTRIES];
3259         drive_info_struct *drv;
3260         int i, dir;
3261
3262         /* We call start_io here in case there is a command waiting on the
3263          * queue that has not been sent.
3264          */
3265         if (blk_queue_plugged(q))
3266                 goto startio;
3267
3268       queue:
3269         creq = blk_peek_request(q);
3270         if (!creq)
3271                 goto startio;
3272
3273         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
3274
3275         if ((c = cmd_alloc(h, 1)) == NULL)
3276                 goto full;
3277
3278         blk_start_request(creq);
3279
3280         spin_unlock_irq(q->queue_lock);
3281
3282         c->cmd_type = CMD_RWREQ;
3283         c->rq = creq;
3284
3285         /* fill in the request */
3286         drv = creq->rq_disk->private_data;
3287         c->Header.ReplyQueue = 0;       // unused in simple mode
3288         /* got command from pool, so use the command block index instead */
3289         /* for direct lookups. */
3290         /* The first 2 bits are reserved for controller error reporting. */
3291         c->Header.Tag.lower = (c->cmdindex << 3);
3292         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
3293         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3294         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
3295         c->Request.Type.Type = TYPE_CMD;        // It is a command.
3296         c->Request.Type.Attribute = ATTR_SIMPLE;
3297         c->Request.Type.Direction =
3298             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3299         c->Request.Timeout = 0; // Don't time out
3300         c->Request.CDB[0] =
3301             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3302         start_blk = blk_rq_pos(creq);
3303 #ifdef CCISS_DEBUG
3304         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3305                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3306 #endif                          /* CCISS_DEBUG */
3307
3308         sg_init_table(tmp_sg, MAXSGENTRIES);
3309         seg = blk_rq_map_sg(q, creq, tmp_sg);
3310
3311         /* get the DMA records for the setup */
3312         if (c->Request.Type.Direction == XFER_READ)
3313                 dir = PCI_DMA_FROMDEVICE;
3314         else
3315                 dir = PCI_DMA_TODEVICE;
3316
3317         for (i = 0; i < seg; i++) {
3318                 c->SG[i].Len = tmp_sg[i].length;
3319                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3320                                                   tmp_sg[i].offset,
3321                                                   tmp_sg[i].length, dir);
3322                 c->SG[i].Addr.lower = temp64.val32.lower;
3323                 c->SG[i].Addr.upper = temp64.val32.upper;
3324                 c->SG[i].Ext = 0;       // we are not chaining
3325         }
3326         /* track how many SG entries we are using */
3327         if (seg > h->maxSG)
3328                 h->maxSG = seg;
3329
3330 #ifdef CCISS_DEBUG
3331         printk(KERN_DEBUG "cciss: Submitting %u sectors in %d segments\n",
3332                blk_rq_sectors(creq), seg);
3333 #endif                          /* CCISS_DEBUG */
3334
3335         c->Header.SGList = c->Header.SGTotal = seg;
3336         if (likely(blk_fs_request(creq))) {
3337                 if(h->cciss_read == CCISS_READ_10) {
3338                         c->Request.CDB[1] = 0;
3339                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
3340                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3341                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3342                         c->Request.CDB[5] = start_blk & 0xff;
3343                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
3344                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3345                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3346                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3347                 } else {
3348                         u32 upper32 = upper_32_bits(start_blk);
3349
3350                         c->Request.CDBLen = 16;
3351                         c->Request.CDB[1]= 0;
3352                         c->Request.CDB[2]= (upper32 >> 24) & 0xff;      //MSB
3353                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3354                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3355                         c->Request.CDB[5]= upper32 & 0xff;
3356                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3357                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3358                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3359                         c->Request.CDB[9]= start_blk & 0xff;
3360                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3361                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3362                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3363                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3364                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3365                 }
3366         } else if (blk_pc_request(creq)) {
3367                 c->Request.CDBLen = creq->cmd_len;
3368                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3369         } else {
3370                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3371                 BUG();
3372         }
3373
3374         spin_lock_irq(q->queue_lock);
3375
3376         addQ(&h->reqQ, c);
3377         h->Qdepth++;
3378         if (h->Qdepth > h->maxQsinceinit)
3379                 h->maxQsinceinit = h->Qdepth;
3380
3381         goto queue;
3382 full:
3383         blk_stop_queue(q);
3384 startio:
3385         /* We will already have the driver lock here so not need
3386          * to lock it.
3387          */
3388         start_io(h);
3389 }
3390
3391 static inline unsigned long get_next_completion(ctlr_info_t *h)
3392 {
3393         return h->access.command_completed(h);
3394 }
3395
3396 static inline int interrupt_pending(ctlr_info_t *h)
3397 {
3398         return h->access.intr_pending(h);
3399 }
3400
3401 static inline long interrupt_not_for_us(ctlr_info_t *h)
3402 {
3403         return (((h->access.intr_pending(h) == 0) ||
3404                  (h->interrupts_enabled == 0)));
3405 }
3406
3407 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3408 {
3409         ctlr_info_t *h = dev_id;
3410         CommandList_struct *c;
3411         unsigned long flags;
3412         __u32 a, a1, a2;
3413
3414         if (interrupt_not_for_us(h))
3415                 return IRQ_NONE;
3416         /*
3417          * If there are completed commands in the completion queue,
3418          * we had better do something about it.
3419          */
3420         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3421         while (interrupt_pending(h)) {
3422                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3423                         a1 = a;
3424                         if ((a & 0x04)) {
3425                                 a2 = (a >> 3);
3426                                 if (a2 >= h->nr_cmds) {
3427                                         printk(KERN_WARNING
3428                                                "cciss: controller cciss%d failed, stopping.\n",
3429                                                h->ctlr);
3430                                         fail_all_cmds(h->ctlr);
3431                                         return IRQ_HANDLED;
3432                                 }
3433
3434                                 c = h->cmd_pool + a2;
3435                                 a = c->busaddr;
3436
3437                         } else {
3438                                 struct hlist_node *tmp;
3439
3440                                 a &= ~3;
3441                                 c = NULL;
3442                                 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3443                                         if (c->busaddr == a)
3444                                                 break;
3445                                 }
3446                         }
3447                         /*
3448                          * If we've found the command, take it off the
3449                          * completion Q and free it
3450                          */
3451                         if (c && c->busaddr == a) {
3452                                 removeQ(c);
3453                                 if (c->cmd_type == CMD_RWREQ) {
3454                                         complete_command(h, c, 0);
3455                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3456                                         complete(c->waiting);
3457                                 }
3458 #                               ifdef CONFIG_CISS_SCSI_TAPE
3459                                 else if (c->cmd_type == CMD_SCSI)
3460                                         complete_scsi_command(c, 0, a1);
3461 #                               endif
3462                                 continue;
3463                         }
3464                 }
3465         }
3466
3467         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3468         return IRQ_HANDLED;
3469 }
3470
3471 /**
3472  * add_to_scan_list() - add controller to rescan queue
3473  * @h:                Pointer to the controller.
3474  *
3475  * Adds the controller to the rescan queue if not already on the queue.
3476  *
3477  * returns 1 if added to the queue, 0 if skipped (could be on the
3478  * queue already, or the controller could be initializing or shutting
3479  * down).
3480  **/
3481 static int add_to_scan_list(struct ctlr_info *h)
3482 {
3483         struct ctlr_info *test_h;
3484         int found = 0;
3485         int ret = 0;
3486
3487         if (h->busy_initializing)
3488                 return 0;
3489
3490         if (!mutex_trylock(&h->busy_shutting_down))
3491                 return 0;
3492
3493         mutex_lock(&scan_mutex);
3494         list_for_each_entry(test_h, &scan_q, scan_list) {
3495                 if (test_h == h) {
3496                         found = 1;
3497                         break;
3498                 }
3499         }
3500         if (!found && !h->busy_scanning) {
3501                 INIT_COMPLETION(h->scan_wait);
3502                 list_add_tail(&h->scan_list, &scan_q);
3503                 ret = 1;
3504         }
3505         mutex_unlock(&scan_mutex);
3506         mutex_unlock(&h->busy_shutting_down);
3507
3508         return ret;
3509 }
3510
3511 /**
3512  * remove_from_scan_list() - remove controller from rescan queue
3513  * @h:                     Pointer to the controller.
3514  *
3515  * Removes the controller from the rescan queue if present. Blocks if
3516  * the controller is currently conducting a rescan.
3517  **/
3518 static void remove_from_scan_list(struct ctlr_info *h)
3519 {
3520         struct ctlr_info *test_h, *tmp_h;
3521         int scanning = 0;
3522
3523         mutex_lock(&scan_mutex);
3524         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3525                 if (test_h == h) {
3526                         list_del(&h->scan_list);
3527                         complete_all(&h->scan_wait);
3528                         mutex_unlock(&scan_mutex);
3529                         return;
3530                 }
3531         }
3532         if (&h->busy_scanning)
3533                 scanning = 0;
3534         mutex_unlock(&scan_mutex);
3535
3536         if (scanning)
3537                 wait_for_completion(&h->scan_wait);
3538 }
3539
3540 /**
3541  * scan_thread() - kernel thread used to rescan controllers
3542  * @data:        Ignored.
3543  *
3544  * A kernel thread used scan for drive topology changes on
3545  * controllers. The thread processes only one controller at a time
3546  * using a queue.  Controllers are added to the queue using
3547  * add_to_scan_list() and removed from the queue either after done
3548  * processing or using remove_from_scan_list().
3549  *
3550  * returns 0.
3551  **/
3552 static int scan_thread(void *data)
3553 {
3554         struct ctlr_info *h;
3555
3556         while (1) {
3557                 set_current_state(TASK_INTERRUPTIBLE);
3558                 schedule();
3559                 if (kthread_should_stop())
3560                         break;
3561
3562                 while (1) {
3563                         mutex_lock(&scan_mutex);
3564                         if (list_empty(&scan_q)) {
3565                                 mutex_unlock(&scan_mutex);
3566                                 break;
3567                         }
3568
3569                         h = list_entry(scan_q.next,
3570                                        struct ctlr_info,
3571                                        scan_list);
3572                         list_del(&h->scan_list);
3573                         h->busy_scanning = 1;
3574                         mutex_unlock(&scan_mutex);
3575
3576                         if (h) {
3577                                 rebuild_lun_table(h, 0, 0);
3578                                 complete_all(&h->scan_wait);
3579                                 mutex_lock(&scan_mutex);
3580                                 h->busy_scanning = 0;
3581                                 mutex_unlock(&scan_mutex);
3582                         }
3583                 }
3584         }
3585
3586         return 0;
3587 }
3588
3589 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3590 {
3591         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3592                 return 0;
3593
3594         switch (c->err_info->SenseInfo[12]) {
3595         case STATE_CHANGED:
3596                 printk(KERN_WARNING "cciss%d: a state change "
3597                         "detected, command retried\n", h->ctlr);
3598                 return 1;
3599         break;
3600         case LUN_FAILED:
3601                 printk(KERN_WARNING "cciss%d: LUN failure "
3602                         "detected, action required\n", h->ctlr);
3603                 return 1;
3604         break;
3605         case REPORT_LUNS_CHANGED:
3606                 printk(KERN_WARNING "cciss%d: report LUN data "
3607                         "changed\n", h->ctlr);
3608                 add_to_scan_list(h);
3609                 wake_up_process(cciss_scan_thread);
3610                 return 1;
3611         break;
3612         case POWER_OR_RESET:
3613                 printk(KERN_WARNING "cciss%d: a power on "
3614                         "or device reset detected\n", h->ctlr);
3615                 return 1;
3616         break;
3617         case UNIT_ATTENTION_CLEARED:
3618                 printk(KERN_WARNING "cciss%d: unit attention "
3619                     "cleared by another initiator\n", h->ctlr);
3620                 return 1;
3621         break;
3622         default:
3623                 printk(KERN_WARNING "cciss%d: unknown "
3624                         "unit attention detected\n", h->ctlr);
3625                                 return 1;
3626         }
3627 }
3628
3629 /*
3630  *  We cannot read the structure directly, for portability we must use
3631  *   the io functions.
3632  *   This is for debug only.
3633  */
3634 #ifdef CCISS_DEBUG
3635 static void print_cfg_table(CfgTable_struct *tb)
3636 {
3637         int i;
3638         char temp_name[17];
3639
3640         printk("Controller Configuration information\n");
3641         printk("------------------------------------\n");
3642         for (i = 0; i < 4; i++)
3643                 temp_name[i] = readb(&(tb->Signature[i]));
3644         temp_name[4] = '\0';
3645         printk("   Signature = %s\n", temp_name);
3646         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3647         printk("   Transport methods supported = 0x%x\n",
3648                readl(&(tb->TransportSupport)));
3649         printk("   Transport methods active = 0x%x\n",
3650                readl(&(tb->TransportActive)));
3651         printk("   Requested transport Method = 0x%x\n",
3652                readl(&(tb->HostWrite.TransportRequest)));
3653         printk("   Coalesce Interrupt Delay = 0x%x\n",
3654                readl(&(tb->HostWrite.CoalIntDelay)));
3655         printk("   Coalesce Interrupt Count = 0x%x\n",
3656                readl(&(tb->HostWrite.CoalIntCount)));
3657         printk("   Max outstanding commands = 0x%d\n",
3658                readl(&(tb->CmdsOutMax)));
3659         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3660         for (i = 0; i < 16; i++)
3661                 temp_name[i] = readb(&(tb->ServerName[i]));
3662         temp_name[16] = '\0';
3663         printk("   Server Name = %s\n", temp_name);
3664         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3665 }
3666 #endif                          /* CCISS_DEBUG */
3667
3668 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3669 {
3670         int i, offset, mem_type, bar_type;
3671         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3672                 return 0;
3673         offset = 0;
3674         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3675                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3676                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3677                         offset += 4;
3678                 else {
3679                         mem_type = pci_resource_flags(pdev, i) &
3680                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3681                         switch (mem_type) {
3682                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3683                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3684                                 offset += 4;    /* 32 bit */
3685                                 break;
3686                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3687                                 offset += 8;
3688                                 break;
3689                         default:        /* reserved in PCI 2.2 */
3690                                 printk(KERN_WARNING
3691                                        "Base address is invalid\n");
3692                                 return -1;
3693                                 break;
3694                         }
3695                 }
3696                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3697                         return i + 1;
3698         }
3699         return -1;
3700 }
3701
3702 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3703  * controllers that are capable. If not, we use IO-APIC mode.
3704  */
3705
3706 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3707                                            struct pci_dev *pdev, __u32 board_id)
3708 {
3709 #ifdef CONFIG_PCI_MSI
3710         int err;
3711         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3712         {0, 2}, {0, 3}
3713         };
3714
3715         /* Some boards advertise MSI but don't really support it */
3716         if ((board_id == 0x40700E11) ||
3717             (board_id == 0x40800E11) ||
3718             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3719                 goto default_int_mode;
3720
3721         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3722                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3723                 if (!err) {
3724                         c->intr[0] = cciss_msix_entries[0].vector;
3725                         c->intr[1] = cciss_msix_entries[1].vector;
3726                         c->intr[2] = cciss_msix_entries[2].vector;
3727                         c->intr[3] = cciss_msix_entries[3].vector;
3728                         c->msix_vector = 1;
3729                         return;
3730                 }
3731                 if (err > 0) {
3732                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3733                                "available\n", err);
3734                         goto default_int_mode;
3735                 } else {
3736                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3737                                err);
3738                         goto default_int_mode;
3739                 }
3740         }
3741         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3742                 if (!pci_enable_msi(pdev)) {
3743                         c->msi_vector = 1;
3744                 } else {
3745                         printk(KERN_WARNING "cciss: MSI init failed\n");
3746                 }
3747         }
3748 default_int_mode:
3749 #endif                          /* CONFIG_PCI_MSI */
3750         /* if we get here we're going to use the default interrupt mode */
3751         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3752         return;
3753 }
3754
3755 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3756 {
3757         ushort subsystem_vendor_id, subsystem_device_id, command;
3758         __u32 board_id, scratchpad = 0;
3759         __u64 cfg_offset;
3760         __u32 cfg_base_addr;
3761         __u64 cfg_base_addr_index;
3762         int i, prod_index, err;
3763
3764         subsystem_vendor_id = pdev->subsystem_vendor;
3765         subsystem_device_id = pdev->subsystem_device;
3766         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3767                     subsystem_vendor_id);
3768
3769         for (i = 0; i < ARRAY_SIZE(products); i++) {
3770                 /* Stand aside for hpsa driver on request */
3771                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3772                         return -ENODEV;
3773                 if (board_id == products[i].board_id)
3774                         break;
3775         }
3776         prod_index = i;
3777         if (prod_index == ARRAY_SIZE(products)) {
3778                 dev_warn(&pdev->dev,
3779                         "unrecognized board ID: 0x%08lx, ignoring.\n",
3780                         (unsigned long) board_id);
3781                 return -ENODEV;
3782         }
3783
3784         /* check to see if controller has been disabled */
3785         /* BEFORE trying to enable it */
3786         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3787         if (!(command & 0x02)) {
3788                 printk(KERN_WARNING
3789                        "cciss: controller appears to be disabled\n");
3790                 return -ENODEV;
3791         }
3792
3793         err = pci_enable_device(pdev);
3794         if (err) {
3795                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3796                 return err;
3797         }
3798
3799         err = pci_request_regions(pdev, "cciss");
3800         if (err) {
3801                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3802                        "aborting\n");
3803                 return err;
3804         }
3805
3806 #ifdef CCISS_DEBUG
3807         printk("command = %x\n", command);
3808         printk("irq = %x\n", pdev->irq);
3809         printk("board_id = %x\n", board_id);
3810 #endif                          /* CCISS_DEBUG */
3811
3812 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3813  * else we use the IO-APIC interrupt assigned to us by system ROM.
3814  */
3815         cciss_interrupt_mode(c, pdev, board_id);
3816
3817         /* find the memory BAR */
3818         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3819                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3820                         break;
3821         }
3822         if (i == DEVICE_COUNT_RESOURCE) {
3823                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3824                 err = -ENODEV;
3825                 goto err_out_free_res;
3826         }
3827
3828         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3829                                                  * already removed
3830                                                  */
3831
3832 #ifdef CCISS_DEBUG
3833         printk("address 0 = %lx\n", c->paddr);
3834 #endif                          /* CCISS_DEBUG */
3835         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3836
3837         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3838          * We poll for up to 120 secs, once per 100ms. */
3839         for (i = 0; i < 1200; i++) {
3840                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3841                 if (scratchpad == CCISS_FIRMWARE_READY)
3842                         break;
3843                 set_current_state(TASK_INTERRUPTIBLE);
3844                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
3845         }
3846         if (scratchpad != CCISS_FIRMWARE_READY) {
3847                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3848                 err = -ENODEV;
3849                 goto err_out_free_res;
3850         }
3851
3852         /* get the address index number */
3853         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3854         cfg_base_addr &= (__u32) 0x0000ffff;
3855 #ifdef CCISS_DEBUG
3856         printk("cfg base address = %x\n", cfg_base_addr);
3857 #endif                          /* CCISS_DEBUG */
3858         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3859 #ifdef CCISS_DEBUG
3860         printk("cfg base address index = %llx\n",
3861                 (unsigned long long)cfg_base_addr_index);
3862 #endif                          /* CCISS_DEBUG */
3863         if (cfg_base_addr_index == -1) {
3864                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3865                 err = -ENODEV;
3866                 goto err_out_free_res;
3867         }
3868
3869         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3870 #ifdef CCISS_DEBUG
3871         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3872 #endif                          /* CCISS_DEBUG */
3873         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3874                                                        cfg_base_addr_index) +
3875                                     cfg_offset, sizeof(CfgTable_struct));
3876         c->board_id = board_id;
3877
3878 #ifdef CCISS_DEBUG
3879         print_cfg_table(c->cfgtable);
3880 #endif                          /* CCISS_DEBUG */
3881
3882         /* Some controllers support Zero Memory Raid (ZMR).
3883          * When configured in ZMR mode the number of supported
3884          * commands drops to 64. So instead of just setting an
3885          * arbitrary value we make the driver a little smarter.
3886          * We read the config table to tell us how many commands
3887          * are supported on the controller then subtract 4 to
3888          * leave a little room for ioctl calls.
3889          */
3890         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3891         c->product_name = products[prod_index].product_name;
3892         c->access = *(products[prod_index].access);
3893         c->nr_cmds = c->max_commands - 4;
3894         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3895             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3896             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3897             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3898                 printk("Does not appear to be a valid CISS config table\n");
3899                 err = -ENODEV;
3900                 goto err_out_free_res;
3901         }
3902 #ifdef CONFIG_X86
3903         {
3904                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3905                 __u32 prefetch;
3906                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3907                 prefetch |= 0x100;
3908                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3909         }
3910 #endif
3911
3912         /* Disabling DMA prefetch and refetch for the P600.
3913          * An ASIC bug may result in accesses to invalid memory addresses.
3914          * We've disabled prefetch for some time now. Testing with XEN
3915          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3916          */
3917         if(board_id == 0x3225103C) {
3918                 __u32 dma_prefetch;
3919                 __u32 dma_refetch;
3920                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3921                 dma_prefetch |= 0x8000;
3922                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3923                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3924                 dma_refetch |= 0x1;
3925                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3926         }
3927
3928 #ifdef CCISS_DEBUG
3929         printk("Trying to put board into Simple mode\n");
3930 #endif                          /* CCISS_DEBUG */
3931         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3932         /* Update the field, and then ring the doorbell */
3933         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3934         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3935
3936         /* under certain very rare conditions, this can take awhile.
3937          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3938          * as we enter this code.) */
3939         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3940                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3941                         break;
3942                 /* delay and try again */
3943                 set_current_state(TASK_INTERRUPTIBLE);
3944                 schedule_timeout(msecs_to_jiffies(1));
3945         }
3946
3947 #ifdef CCISS_DEBUG
3948         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3949                readl(c->vaddr + SA5_DOORBELL));
3950 #endif                          /* CCISS_DEBUG */
3951 #ifdef CCISS_DEBUG
3952         print_cfg_table(c->cfgtable);
3953 #endif                          /* CCISS_DEBUG */
3954
3955         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3956                 printk(KERN_WARNING "cciss: unable to get board into"
3957                        " simple mode\n");
3958                 err = -ENODEV;
3959                 goto err_out_free_res;
3960         }
3961         return 0;
3962
3963 err_out_free_res:
3964         /*
3965          * Deliberately omit pci_disable_device(): it does something nasty to
3966          * Smart Array controllers that pci_enable_device does not undo
3967          */
3968         pci_release_regions(pdev);
3969         return err;
3970 }
3971
3972 /* Function to find the first free pointer into our hba[] array
3973  * Returns -1 if no free entries are left.
3974  */
3975 static int alloc_cciss_hba(void)
3976 {
3977         int i;
3978
3979         for (i = 0; i < MAX_CTLR; i++) {
3980                 if (!hba[i]) {
3981                         ctlr_info_t *p;
3982
3983                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3984                         if (!p)
3985                                 goto Enomem;
3986                         hba[i] = p;
3987                         return i;
3988                 }
3989         }
3990         printk(KERN_WARNING "cciss: This driver supports a maximum"
3991                " of %d controllers.\n", MAX_CTLR);
3992         return -1;
3993 Enomem:
3994         printk(KERN_ERR "cciss: out of memory.\n");
3995         return -1;
3996 }
3997
3998 static void free_hba(int n)
3999 {
4000         ctlr_info_t *h = hba[n];
4001         int i;
4002
4003         hba[n] = NULL;
4004         for (i = 0; i < h->highest_lun + 1; i++)
4005                 if (h->gendisk[i] != NULL)
4006                         put_disk(h->gendisk[i]);
4007         kfree(h);
4008 }
4009
4010 /* Send a message CDB to the firmware. */
4011 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4012 {
4013         typedef struct {
4014                 CommandListHeader_struct CommandHeader;
4015                 RequestBlock_struct Request;
4016                 ErrDescriptor_struct ErrorDescriptor;
4017         } Command;
4018         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4019         Command *cmd;
4020         dma_addr_t paddr64;
4021         uint32_t paddr32, tag;
4022         void __iomem *vaddr;
4023         int i, err;
4024
4025         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4026         if (vaddr == NULL)
4027                 return -ENOMEM;
4028
4029         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4030            CCISS commands, so they must be allocated from the lower 4GiB of
4031            memory. */
4032         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4033         if (err) {
4034                 iounmap(vaddr);
4035                 return -ENOMEM;
4036         }
4037
4038         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4039         if (cmd == NULL) {
4040                 iounmap(vaddr);
4041                 return -ENOMEM;
4042         }
4043
4044         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4045            although there's no guarantee, we assume that the address is at
4046            least 4-byte aligned (most likely, it's page-aligned). */
4047         paddr32 = paddr64;
4048
4049         cmd->CommandHeader.ReplyQueue = 0;
4050         cmd->CommandHeader.SGList = 0;
4051         cmd->CommandHeader.SGTotal = 0;
4052         cmd->CommandHeader.Tag.lower = paddr32;
4053         cmd->CommandHeader.Tag.upper = 0;
4054         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4055
4056         cmd->Request.CDBLen = 16;
4057         cmd->Request.Type.Type = TYPE_MSG;
4058         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4059         cmd->Request.Type.Direction = XFER_NONE;
4060         cmd->Request.Timeout = 0; /* Don't time out */
4061         cmd->Request.CDB[0] = opcode;
4062         cmd->Request.CDB[1] = type;
4063         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4064
4065         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4066         cmd->ErrorDescriptor.Addr.upper = 0;
4067         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4068
4069         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4070
4071         for (i = 0; i < 10; i++) {
4072                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4073                 if ((tag & ~3) == paddr32)
4074                         break;
4075                 schedule_timeout_uninterruptible(HZ);
4076         }
4077
4078         iounmap(vaddr);
4079
4080         /* we leak the DMA buffer here ... no choice since the controller could
4081            still complete the command. */
4082         if (i == 10) {
4083                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4084                         opcode, type);
4085                 return -ETIMEDOUT;
4086         }
4087
4088         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4089
4090         if (tag & 2) {
4091                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4092                         opcode, type);
4093                 return -EIO;
4094         }
4095
4096         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4097                 opcode, type);
4098         return 0;
4099 }
4100
4101 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4102 #define cciss_noop(p) cciss_message(p, 3, 0)
4103
4104 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4105 {
4106 /* the #defines are stolen from drivers/pci/msi.h. */
4107 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4108 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4109
4110         int pos;
4111         u16 control = 0;
4112
4113         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4114         if (pos) {
4115                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4116                 if (control & PCI_MSI_FLAGS_ENABLE) {
4117                         printk(KERN_INFO "cciss: resetting MSI\n");
4118                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4119                 }
4120         }
4121
4122         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4123         if (pos) {
4124                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4125                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4126                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4127                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4128                 }
4129         }
4130
4131         return 0;
4132 }
4133
4134 /* This does a hard reset of the controller using PCI power management
4135  * states. */
4136 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4137 {
4138         u16 pmcsr, saved_config_space[32];
4139         int i, pos;
4140
4141         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4142
4143         /* This is very nearly the same thing as
4144
4145            pci_save_state(pci_dev);
4146            pci_set_power_state(pci_dev, PCI_D3hot);
4147            pci_set_power_state(pci_dev, PCI_D0);
4148            pci_restore_state(pci_dev);
4149
4150            but we can't use these nice canned kernel routines on
4151            kexec, because they also check the MSI/MSI-X state in PCI
4152            configuration space and do the wrong thing when it is
4153            set/cleared.  Also, the pci_save/restore_state functions
4154            violate the ordering requirements for restoring the
4155            configuration space from the CCISS document (see the
4156            comment below).  So we roll our own .... */
4157
4158         for (i = 0; i < 32; i++)
4159                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4160
4161         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4162         if (pos == 0) {
4163                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4164                 return -ENODEV;
4165         }
4166
4167         /* Quoting from the Open CISS Specification: "The Power
4168          * Management Control/Status Register (CSR) controls the power
4169          * state of the device.  The normal operating state is D0,
4170          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4171          * the controller, place the interface device in D3 then to
4172          * D0, this causes a secondary PCI reset which will reset the
4173          * controller." */
4174
4175         /* enter the D3hot power management state */
4176         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4177         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4178         pmcsr |= PCI_D3hot;
4179         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4180
4181         schedule_timeout_uninterruptible(HZ >> 1);
4182
4183         /* enter the D0 power management state */
4184         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4185         pmcsr |= PCI_D0;
4186         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4187
4188         schedule_timeout_uninterruptible(HZ >> 1);
4189
4190         /* Restore the PCI configuration space.  The Open CISS
4191          * Specification says, "Restore the PCI Configuration
4192          * Registers, offsets 00h through 60h. It is important to
4193          * restore the command register, 16-bits at offset 04h,
4194          * last. Do not restore the configuration status register,
4195          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4196         for (i = 0; i < 32; i++) {
4197                 if (i == 2 || i == 3)
4198                         continue;
4199                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4200         }
4201         wmb();
4202         pci_write_config_word(pdev, 4, saved_config_space[2]);
4203
4204         return 0;
4205 }
4206
4207 /*
4208  *  This is it.  Find all the controllers and register them.  I really hate
4209  *  stealing all these major device numbers.
4210  *  returns the number of block devices registered.
4211  */
4212 static int __devinit cciss_init_one(struct pci_dev *pdev,
4213                                     const struct pci_device_id *ent)
4214 {
4215         int i;
4216         int j = 0;
4217         int rc;
4218         int dac, return_code;
4219         InquiryData_struct *inq_buff;
4220
4221         if (reset_devices) {
4222                 /* Reset the controller with a PCI power-cycle */
4223                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4224                         return -ENODEV;
4225
4226                 /* Now try to get the controller to respond to a no-op. Some
4227                    devices (notably the HP Smart Array 5i Controller) need
4228                    up to 30 seconds to respond. */
4229                 for (i=0; i<30; i++) {
4230                         if (cciss_noop(pdev) == 0)
4231                                 break;
4232
4233                         schedule_timeout_uninterruptible(HZ);
4234                 }
4235                 if (i == 30) {
4236                         printk(KERN_ERR "cciss: controller seems dead\n");
4237                         return -EBUSY;
4238                 }
4239         }
4240
4241         i = alloc_cciss_hba();
4242         if (i < 0)
4243                 return -1;
4244
4245         hba[i]->busy_initializing = 1;
4246         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4247         INIT_HLIST_HEAD(&hba[i]->reqQ);
4248         mutex_init(&hba[i]->busy_shutting_down);
4249
4250         if (cciss_pci_init(hba[i], pdev) != 0)
4251                 goto clean_no_release_regions;
4252
4253         sprintf(hba[i]->devname, "cciss%d", i);
4254         hba[i]->ctlr = i;
4255         hba[i]->pdev = pdev;
4256
4257         init_completion(&hba[i]->scan_wait);
4258
4259         if (cciss_create_hba_sysfs_entry(hba[i]))
4260                 goto clean0;
4261
4262         /* configure PCI DMA stuff */
4263         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4264                 dac = 1;
4265         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4266                 dac = 0;
4267         else {
4268                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4269                 goto clean1;
4270         }
4271
4272         /*
4273          * register with the major number, or get a dynamic major number
4274          * by passing 0 as argument.  This is done for greater than
4275          * 8 controller support.
4276          */
4277         if (i < MAX_CTLR_ORIG)
4278                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4279         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4280         if (rc == -EBUSY || rc == -EINVAL) {
4281                 printk(KERN_ERR
4282                        "cciss:  Unable to get major number %d for %s "
4283                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4284                 goto clean1;
4285         } else {
4286                 if (i >= MAX_CTLR_ORIG)
4287                         hba[i]->major = rc;
4288         }
4289
4290         /* make sure the board interrupts are off */
4291         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4292         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
4293                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
4294                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4295                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4296                 goto clean2;
4297         }
4298
4299         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4300                hba[i]->devname, pdev->device, pci_name(pdev),
4301                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4302
4303         hba[i]->cmd_pool_bits =
4304             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4305                         * sizeof(unsigned long), GFP_KERNEL);
4306         hba[i]->cmd_pool = (CommandList_struct *)
4307             pci_alloc_consistent(hba[i]->pdev,
4308                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4309                     &(hba[i]->cmd_pool_dhandle));
4310         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4311             pci_alloc_consistent(hba[i]->pdev,
4312                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4313                     &(hba[i]->errinfo_pool_dhandle));
4314         if ((hba[i]->cmd_pool_bits == NULL)
4315             || (hba[i]->cmd_pool == NULL)
4316             || (hba[i]->errinfo_pool == NULL)) {
4317                 printk(KERN_ERR "cciss: out of memory");
4318                 goto clean4;
4319         }
4320         spin_lock_init(&hba[i]->lock);
4321
4322         /* Initialize the pdev driver private data.
4323            have it point to hba[i].  */
4324         pci_set_drvdata(pdev, hba[i]);
4325         /* command and error info recs zeroed out before
4326            they are used */
4327         memset(hba[i]->cmd_pool_bits, 0,
4328                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4329                         * sizeof(unsigned long));
4330
4331         hba[i]->num_luns = 0;
4332         hba[i]->highest_lun = -1;
4333         for (j = 0; j < CISS_MAX_LUN; j++) {
4334                 hba[i]->drv[j] = NULL;
4335                 hba[i]->gendisk[j] = NULL;
4336         }
4337
4338         cciss_scsi_setup(i);
4339
4340         /* Turn the interrupts on so we can service requests */
4341         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4342
4343         /* Get the firmware version */
4344         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4345         if (inq_buff == NULL) {
4346                 printk(KERN_ERR "cciss: out of memory\n");
4347                 goto clean4;
4348         }
4349
4350         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4351                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4352         if (return_code == IO_OK) {
4353                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4354                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4355                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4356                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4357         } else {         /* send command failed */
4358                 printk(KERN_WARNING "cciss: unable to determine firmware"
4359                         " version of controller\n");
4360         }
4361         kfree(inq_buff);
4362
4363         cciss_procinit(i);
4364
4365         hba[i]->cciss_max_sectors = 2048;
4366
4367         rebuild_lun_table(hba[i], 1, 0);
4368         hba[i]->busy_initializing = 0;
4369         return 1;
4370
4371 clean4:
4372         kfree(hba[i]->cmd_pool_bits);
4373         if (hba[i]->cmd_pool)
4374                 pci_free_consistent(hba[i]->pdev,
4375                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4376                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4377         if (hba[i]->errinfo_pool)
4378                 pci_free_consistent(hba[i]->pdev,
4379                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4380                                     hba[i]->errinfo_pool,
4381                                     hba[i]->errinfo_pool_dhandle);
4382         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4383 clean2:
4384         unregister_blkdev(hba[i]->major, hba[i]->devname);
4385 clean1:
4386         cciss_destroy_hba_sysfs_entry(hba[i]);
4387 clean0:
4388         pci_release_regions(pdev);
4389 clean_no_release_regions:
4390         hba[i]->busy_initializing = 0;
4391
4392         /*
4393          * Deliberately omit pci_disable_device(): it does something nasty to
4394          * Smart Array controllers that pci_enable_device does not undo
4395          */
4396         pci_set_drvdata(pdev, NULL);
4397         free_hba(i);
4398         return -1;
4399 }
4400
4401 static void cciss_shutdown(struct pci_dev *pdev)
4402 {
4403         ctlr_info_t *tmp_ptr;
4404         int i;
4405         char flush_buf[4];
4406         int return_code;
4407
4408         tmp_ptr = pci_get_drvdata(pdev);
4409         if (tmp_ptr == NULL)
4410                 return;
4411         i = tmp_ptr->ctlr;
4412         if (hba[i] == NULL)
4413                 return;
4414
4415         /* Turn board interrupts off  and send the flush cache command */
4416         /* sendcmd will turn off interrupt, and send the flush...
4417          * To write all data in the battery backed cache to disks */
4418         memset(flush_buf, 0, 4);
4419         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0,
4420                 CTLR_LUNID, TYPE_CMD);
4421         if (return_code == IO_OK) {
4422                 printk(KERN_INFO "Completed flushing cache on controller %d\n", i);
4423         } else {
4424                 printk(KERN_WARNING "Error flushing cache on controller %d\n", i);
4425         }
4426         free_irq(hba[i]->intr[2], hba[i]);
4427 }
4428
4429 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4430 {
4431         ctlr_info_t *tmp_ptr;
4432         int i, j;
4433
4434         if (pci_get_drvdata(pdev) == NULL) {
4435                 printk(KERN_ERR "cciss: Unable to remove device \n");
4436                 return;
4437         }
4438
4439         tmp_ptr = pci_get_drvdata(pdev);
4440         i = tmp_ptr->ctlr;
4441         if (hba[i] == NULL) {
4442                 printk(KERN_ERR "cciss: device appears to "
4443                        "already be removed \n");
4444                 return;
4445         }
4446
4447         mutex_lock(&hba[i]->busy_shutting_down);
4448
4449         remove_from_scan_list(hba[i]);
4450         remove_proc_entry(hba[i]->devname, proc_cciss);
4451         unregister_blkdev(hba[i]->major, hba[i]->devname);
4452
4453         /* remove it from the disk list */
4454         for (j = 0; j < CISS_MAX_LUN; j++) {
4455                 struct gendisk *disk = hba[i]->gendisk[j];
4456                 if (disk) {
4457                         struct request_queue *q = disk->queue;
4458
4459                         if (disk->flags & GENHD_FL_UP) {
4460                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4461                                 del_gendisk(disk);
4462                         }
4463                         if (q)
4464                                 blk_cleanup_queue(q);
4465                 }
4466         }
4467
4468 #ifdef CONFIG_CISS_SCSI_TAPE
4469         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4470 #endif
4471
4472         cciss_shutdown(pdev);
4473
4474 #ifdef CONFIG_PCI_MSI
4475         if (hba[i]->msix_vector)
4476                 pci_disable_msix(hba[i]->pdev);
4477         else if (hba[i]->msi_vector)
4478                 pci_disable_msi(hba[i]->pdev);
4479 #endif                          /* CONFIG_PCI_MSI */
4480
4481         iounmap(hba[i]->vaddr);
4482
4483         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4484                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4485         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4486                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4487         kfree(hba[i]->cmd_pool_bits);
4488         /*
4489          * Deliberately omit pci_disable_device(): it does something nasty to
4490          * Smart Array controllers that pci_enable_device does not undo
4491          */
4492         pci_release_regions(pdev);
4493         pci_set_drvdata(pdev, NULL);
4494         cciss_destroy_hba_sysfs_entry(hba[i]);
4495         mutex_unlock(&hba[i]->busy_shutting_down);
4496         free_hba(i);
4497 }
4498
4499 static struct pci_driver cciss_pci_driver = {
4500         .name = "cciss",
4501         .probe = cciss_init_one,
4502         .remove = __devexit_p(cciss_remove_one),
4503         .id_table = cciss_pci_device_id,        /* id_table */
4504         .shutdown = cciss_shutdown,
4505 };
4506
4507 /*
4508  *  This is it.  Register the PCI driver information for the cards we control
4509  *  the OS will call our registered routines when it finds one of our cards.
4510  */
4511 static int __init cciss_init(void)
4512 {
4513         int err;
4514
4515         /*
4516          * The hardware requires that commands are aligned on a 64-bit
4517          * boundary. Given that we use pci_alloc_consistent() to allocate an
4518          * array of them, the size must be a multiple of 8 bytes.
4519          */
4520         BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4521
4522         printk(KERN_INFO DRIVER_NAME "\n");
4523
4524         err = bus_register(&cciss_bus_type);
4525         if (err)
4526                 return err;
4527
4528         /* Start the scan thread */
4529         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4530         if (IS_ERR(cciss_scan_thread)) {
4531                 err = PTR_ERR(cciss_scan_thread);
4532                 goto err_bus_unregister;
4533         }
4534
4535         /* Register for our PCI devices */
4536         err = pci_register_driver(&cciss_pci_driver);
4537         if (err)
4538                 goto err_thread_stop;
4539
4540         return err;
4541
4542 err_thread_stop:
4543         kthread_stop(cciss_scan_thread);
4544 err_bus_unregister:
4545         bus_unregister(&cciss_bus_type);
4546
4547         return err;
4548 }
4549
4550 static void __exit cciss_cleanup(void)
4551 {
4552         int i;
4553
4554         pci_unregister_driver(&cciss_pci_driver);
4555         /* double check that all controller entrys have been removed */
4556         for (i = 0; i < MAX_CTLR; i++) {
4557                 if (hba[i] != NULL) {
4558                         printk(KERN_WARNING "cciss: had to remove"
4559                                " controller %d\n", i);
4560                         cciss_remove_one(hba[i]->pdev);
4561                 }
4562         }
4563         kthread_stop(cciss_scan_thread);
4564         remove_proc_entry("driver/cciss", NULL);
4565         bus_unregister(&cciss_bus_type);
4566 }
4567
4568 static void fail_all_cmds(unsigned long ctlr)
4569 {
4570         /* If we get here, the board is apparently dead. */
4571         ctlr_info_t *h = hba[ctlr];
4572         CommandList_struct *c;
4573         unsigned long flags;
4574
4575         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4576         h->alive = 0;           /* the controller apparently died... */
4577
4578         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4579
4580         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
4581
4582         /* move everything off the request queue onto the completed queue */
4583         while (!hlist_empty(&h->reqQ)) {
4584                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4585                 removeQ(c);
4586                 h->Qdepth--;
4587                 addQ(&h->cmpQ, c);
4588         }
4589
4590         /* Now, fail everything on the completed queue with a HW error */
4591         while (!hlist_empty(&h->cmpQ)) {
4592                 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4593                 removeQ(c);
4594                 if (c->cmd_type != CMD_MSG_STALE)
4595                         c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4596                 if (c->cmd_type == CMD_RWREQ) {
4597                         complete_command(h, c, 0);
4598                 } else if (c->cmd_type == CMD_IOCTL_PEND)
4599                         complete(c->waiting);
4600 #ifdef CONFIG_CISS_SCSI_TAPE
4601                 else if (c->cmd_type == CMD_SCSI)
4602                         complete_scsi_command(c, 0, 0);
4603 #endif
4604         }
4605         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4606         return;
4607 }
4608
4609 module_init(cciss_init);
4610 module_exit(cciss_cleanup);