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 == NULL) /* it's possible for h->drv[] to have holes. */
 343                 return 0;
 344
 345         if (drv->heads == 0)
 346                 return 0;
 347
 348         vol_sz = drv->nr_blocks;
 349         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 350         vol_sz_frac *= 100;
 351         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 352
 353         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 354                 drv->raid_level = RAID_UNKNOWN;
 355         seq_printf(seq, "cciss/c%dd%d:"
 356                         "\t%4u.%02uGB\tRAID %s\n",
 357                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 358                         raid_label[drv->raid_level]);
 359         return 0;
 360 }
 361
 362 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 363 {
 364         ctlr_info_t *h = seq->private;
 365
 366         if (*pos > h->highest_lun)
 367                 return NULL;
 368         *pos += 1;
 369
 370         return pos;
 371 }
 372
 373 static void cciss_seq_stop(struct seq_file *seq, void *v)
 374 {
 375         ctlr_info_t *h = seq->private;
 376
 377         /* Only reset h->busy_configuring if we succeeded in setting
 378          * it during cciss_seq_start. */
 379         if (v == ERR_PTR(-EBUSY))
 380                 return;
 381
 382         h->busy_configuring = 0;
 383 }
 384
 385 static const struct seq_operations cciss_seq_ops = {
 386         .start = cciss_seq_start,
 387         .show  = cciss_seq_show,
 388         .next  = cciss_seq_next,
 389         .stop  = cciss_seq_stop,
 390 };
 391
 392 static int cciss_seq_open(struct inode *inode, struct file *file)
 393 {
 394         int ret = seq_open(file, &cciss_seq_ops);
 395         struct seq_file *seq = file->private_data;
 396
 397         if (!ret)
 398                 seq->private = PDE(inode)->data;
 399
 400         return ret;
 401 }
 402
 403 static ssize_t
 404 cciss_proc_write(struct file *file, const char __user *buf,
 405                  size_t length, loff_t *ppos)
 406 {
 407         int err;
 408         char *buffer;
 409
 410 #ifndef CONFIG_CISS_SCSI_TAPE
 411         return -EINVAL;
 412 #endif
 413
 414         if (!buf || length > PAGE_SIZE - 1)
 415                 return -EINVAL;
 416
 417         buffer = (char *)__get_free_page(GFP_KERNEL);
 418         if (!buffer)
 419                 return -ENOMEM;
 420
 421         err = -EFAULT;
 422         if (copy_from_user(buffer, buf, length))
 423                 goto out;
 424         buffer[length] = '\0';
 425
 426 #ifdef CONFIG_CISS_SCSI_TAPE
 427         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 428                 struct seq_file *seq = file->private_data;
 429                 ctlr_info_t *h = seq->private;
 430                 int rc;
 431
 432                 rc = cciss_engage_scsi(h->ctlr);
 433                 if (rc != 0)
 434                         err = -rc;
 435                 else
 436                         err = length;
 437         } else
 438 #endif /* CONFIG_CISS_SCSI_TAPE */
 439                 err = -EINVAL;
 440         /* might be nice to have "disengage" too, but it's not
 441            safely possible. (only 1 module use count, lock issues.) */
 442
 443 out:
 444         free_page((unsigned long)buffer);
 445         return err;
 446 }
 447
 448 static const struct file_operations cciss_proc_fops = {
 449         .owner   = THIS_MODULE,
 450         .open    = cciss_seq_open,
 451         .read    = seq_read,
 452         .llseek  = seq_lseek,
 453         .release = seq_release,
 454         .write   = cciss_proc_write,
 455 };
 456
 457 static void __devinit cciss_procinit(int i)
 458 {
 459         struct proc_dir_entry *pde;
 460
 461         if (proc_cciss == NULL)
 462                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 463         if (!proc_cciss)
 464                 return;
 465         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 466                                         S_IROTH, proc_cciss,
 467                                         &cciss_proc_fops, hba[i]);
 468 }
 469 #endif                          /* CONFIG_PROC_FS */
 470
 471 #define MAX_PRODUCT_NAME_LEN 19
 472
 473 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 474 #define to_drv(n) container_of(n, drive_info_struct, dev)
 475
 476 static ssize_t host_store_rescan(struct device *dev,
 477                                  struct device_attribute *attr,
 478                                  const char *buf, size_t count)
 479 {
 480         struct ctlr_info *h = to_hba(dev);
 481
 482         add_to_scan_list(h);
 483         wake_up_process(cciss_scan_thread);
 484         wait_for_completion_interruptible(&h->scan_wait);
 485
 486         return count;
 487 }
 488 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 489
 490 static ssize_t dev_show_unique_id(struct device *dev,
 491                                  struct device_attribute *attr,
 492                                  char *buf)
 493 {
 494         drive_info_struct *drv = to_drv(dev);
 495         struct ctlr_info *h = to_hba(drv->dev.parent);
 496         __u8 sn[16];
 497         unsigned long flags;
 498         int ret = 0;
 499
 500         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 501         if (h->busy_configuring)
 502                 ret = -EBUSY;
 503         else
 504                 memcpy(sn, drv->serial_no, sizeof(sn));
 505         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 506
 507         if (ret)
 508                 return ret;
 509         else
 510                 return snprintf(buf, 16 * 2 + 2,
 511                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 512                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 513                                 sn[0], sn[1], sn[2], sn[3],
 514                                 sn[4], sn[5], sn[6], sn[7],
 515                                 sn[8], sn[9], sn[10], sn[11],
 516                                 sn[12], sn[13], sn[14], sn[15]);
 517 }
 518 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 519
 520 static ssize_t dev_show_vendor(struct device *dev,
 521                                struct device_attribute *attr,
 522                                char *buf)
 523 {
 524         drive_info_struct *drv = to_drv(dev);
 525         struct ctlr_info *h = to_hba(drv->dev.parent);
 526         char vendor[VENDOR_LEN + 1];
 527         unsigned long flags;
 528         int ret = 0;
 529
 530         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 531         if (h->busy_configuring)
 532                 ret = -EBUSY;
 533         else
 534                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 535         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 536
 537         if (ret)
 538                 return ret;
 539         else
 540                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 541 }
 542 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 543
 544 static ssize_t dev_show_model(struct device *dev,
 545                               struct device_attribute *attr,
 546                               char *buf)
 547 {
 548         drive_info_struct *drv = to_drv(dev);
 549         struct ctlr_info *h = to_hba(drv->dev.parent);
 550         char model[MODEL_LEN + 1];
 551         unsigned long flags;
 552         int ret = 0;
 553
 554         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 555         if (h->busy_configuring)
 556                 ret = -EBUSY;
 557         else
 558                 memcpy(model, drv->model, MODEL_LEN + 1);
 559         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 560
 561         if (ret)
 562                 return ret;
 563         else
 564                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 565 }
 566 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 567
 568 static ssize_t dev_show_rev(struct device *dev,
 569                             struct device_attribute *attr,
 570                             char *buf)
 571 {
 572         drive_info_struct *drv = to_drv(dev);
 573         struct ctlr_info *h = to_hba(drv->dev.parent);
 574         char rev[REV_LEN + 1];
 575         unsigned long flags;
 576         int ret = 0;
 577
 578         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 579         if (h->busy_configuring)
 580                 ret = -EBUSY;
 581         else
 582                 memcpy(rev, drv->rev, REV_LEN + 1);
 583         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 584
 585         if (ret)
 586                 return ret;
 587         else
 588                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 589 }
 590 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 591
 592 static ssize_t cciss_show_lunid(struct device *dev,
 593                                 struct device_attribute *attr, char *buf)
 594 {
 595         drive_info_struct *drv = to_drv(dev);
 596         struct ctlr_info *h = to_hba(drv->dev.parent);
 597         unsigned long flags;
 598         unsigned char lunid[8];
 599
 600         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 601         if (h->busy_configuring) {
 602                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 603                 return -EBUSY;
 604         }
 605         if (!drv->heads) {
 606                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 607                 return -ENOTTY;
 608         }
 609         memcpy(lunid, drv->LunID, sizeof(lunid));
 610         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 611         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 612                 lunid[0], lunid[1], lunid[2], lunid[3],
 613                 lunid[4], lunid[5], lunid[6], lunid[7]);
 614 }
 615 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 616
 617 static ssize_t cciss_show_raid_level(struct device *dev,
 618                                      struct device_attribute *attr, char *buf)
 619 {
 620         drive_info_struct *drv = to_drv(dev);
 621         struct ctlr_info *h = to_hba(drv->dev.parent);
 622         int raid;
 623         unsigned long flags;
 624
 625         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 626         if (h->busy_configuring) {
 627                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 628                 return -EBUSY;
 629         }
 630         raid = drv->raid_level;
 631         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 632         if (raid < 0 || raid > RAID_UNKNOWN)
 633                 raid = RAID_UNKNOWN;
 634
 635         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 636                         raid_label[raid]);
 637 }
 638 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 639
 640 static ssize_t cciss_show_usage_count(struct device *dev,
 641                                       struct device_attribute *attr, char *buf)
 642 {
 643         drive_info_struct *drv = to_drv(dev);
 644         struct ctlr_info *h = to_hba(drv->dev.parent);
 645         unsigned long flags;
 646         int count;
 647
 648         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 649         if (h->busy_configuring) {
 650                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 651                 return -EBUSY;
 652         }
 653         count = drv->usage_count;
 654         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 655         return snprintf(buf, 20, "%d\n", count);
 656 }
 657 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 658
 659 static struct attribute *cciss_host_attrs[] = {
 660         &dev_attr_rescan.attr,
 661         NULL
 662 };
 663
 664 static struct attribute_group cciss_host_attr_group = {
 665         .attrs = cciss_host_attrs,
 666 };
 667
 668 static const struct attribute_group *cciss_host_attr_groups[] = {
 669         &cciss_host_attr_group,
 670         NULL
 671 };
 672
 673 static struct device_type cciss_host_type = {
 674         .name           = "cciss_host",
 675         .groups         = cciss_host_attr_groups,
 676         .release        = cciss_hba_release,
 677 };
 678
 679 static struct attribute *cciss_dev_attrs[] = {
 680         &dev_attr_unique_id.attr,
 681         &dev_attr_model.attr,
 682         &dev_attr_vendor.attr,
 683         &dev_attr_rev.attr,
 684         &dev_attr_lunid.attr,
 685         &dev_attr_raid_level.attr,
 686         &dev_attr_usage_count.attr,
 687         NULL
 688 };
 689
 690 static struct attribute_group cciss_dev_attr_group = {
 691         .attrs = cciss_dev_attrs,
 692 };
 693
 694 static const struct attribute_group *cciss_dev_attr_groups[] = {
 695         &cciss_dev_attr_group,
 696         NULL
 697 };
 698
 699 static struct device_type cciss_dev_type = {
 700         .name           = "cciss_device",
 701         .groups         = cciss_dev_attr_groups,
 702         .release        = cciss_device_release,
 703 };
 704
 705 static struct bus_type cciss_bus_type = {
 706         .name           = "cciss",
 707 };
 708
 709 /*
 710  * cciss_hba_release is called when the reference count
 711  * of h->dev goes to zero.
 712  */
 713 static void cciss_hba_release(struct device *dev)
 714 {
 715         /*
 716          * nothing to do, but need this to avoid a warning
 717          * about not having a release handler from lib/kref.c.
 718          */
 719 }
 720
 721 /*
 722  * Initialize sysfs entry for each controller.  This sets up and registers
 723  * the 'cciss#' directory for each individual controller under
 724  * /sys/bus/pci/devices/<dev>/.
 725  */
 726 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 727 {
 728         device_initialize(&h->dev);
 729         h->dev.type = &cciss_host_type;
 730         h->dev.bus = &cciss_bus_type;
 731         dev_set_name(&h->dev, "%s", h->devname);
 732         h->dev.parent = &h->pdev->dev;
 733
 734         return device_add(&h->dev);
 735 }
 736
 737 /*
 738  * Remove sysfs entries for an hba.
 739  */
 740 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 741 {
 742         device_del(&h->dev);
 743         put_device(&h->dev); /* final put. */
 744 }
 745
 746 /* cciss_device_release is called when the reference count
 747  * of h->drv[x]dev goes to zero.
 748  */
 749 static void cciss_device_release(struct device *dev)
 750 {
 751         drive_info_struct *drv = to_drv(dev);
 752         kfree(drv);
 753 }
 754
 755 /*
 756  * Initialize sysfs for each logical drive.  This sets up and registers
 757  * the 'c#d#' directory for each individual logical drive under
 758  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 759  * /sys/block/cciss!c#d# to this entry.
 760  */
 761 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 762                                        int drv_index)
 763 {
 764         struct device *dev;
 765
 766         if (h->drv[drv_index]->device_initialized)
 767                 return 0;
 768
 769         dev = &h->drv[drv_index]->dev;
 770         device_initialize(dev);
 771         dev->type = &cciss_dev_type;
 772         dev->bus = &cciss_bus_type;
 773         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 774         dev->parent = &h->dev;
 775         h->drv[drv_index]->device_initialized = 1;
 776         return device_add(dev);
 777 }
 778
 779 /*
 780  * Remove sysfs entries for a logical drive.
 781  */
 782 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 783         int ctlr_exiting)
 784 {
 785         struct device *dev = &h->drv[drv_index]->dev;
 786
 787         /* special case for c*d0, we only destroy it on controller exit */
 788         if (drv_index == 0 && !ctlr_exiting)
 789                 return;
 790
 791         device_del(dev);
 792         put_device(dev); /* the "final" put. */
 793         h->drv[drv_index] = NULL;
 794 }
 795
 796 /*
 797  * For operations that cannot sleep, a command block is allocated at init,
 798  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 799  * which ones are free or in use.  For operations that can wait for kmalloc
 800  * to possible sleep, this routine can be called with get_from_pool set to 0.
 801  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 802  */
 803 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 804 {
 805         CommandList_struct *c;
 806         int i;
 807         u64bit temp64;
 808         dma_addr_t cmd_dma_handle, err_dma_handle;
 809
 810         if (!get_from_pool) {
 811                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 812                         sizeof(CommandList_struct), &cmd_dma_handle);
 813                 if (c == NULL)
 814                         return NULL;
 815                 memset(c, 0, sizeof(CommandList_struct));
 816
 817                 c->cmdindex = -1;
 818
 819                 c->err_info = (ErrorInfo_struct *)
 820                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 821                             &err_dma_handle);
 822
 823                 if (c->err_info == NULL) {
 824                         pci_free_consistent(h->pdev,
 825                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 826                         return NULL;
 827                 }
 828                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 829         } else {                /* get it out of the controllers pool */
 830
 831                 do {
 832                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 833                         if (i == h->nr_cmds)
 834                                 return NULL;
 835                 } while (test_and_set_bit
 836                          (i & (BITS_PER_LONG - 1),
 837                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 838 #ifdef CCISS_DEBUG
 839                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 840 #endif
 841                 c = h->cmd_pool + i;
 842                 memset(c, 0, sizeof(CommandList_struct));
 843                 cmd_dma_handle = h->cmd_pool_dhandle
 844                     + i * sizeof(CommandList_struct);
 845                 c->err_info = h->errinfo_pool + i;
 846                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 847                 err_dma_handle = h->errinfo_pool_dhandle
 848                     + i * sizeof(ErrorInfo_struct);
 849                 h->nr_allocs++;
 850
 851                 c->cmdindex = i;
 852         }
 853
 854         INIT_HLIST_NODE(&c->list);
 855         c->busaddr = (__u32) cmd_dma_handle;
 856         temp64.val = (__u64) err_dma_handle;
 857         c->ErrDesc.Addr.lower = temp64.val32.lower;
 858         c->ErrDesc.Addr.upper = temp64.val32.upper;
 859         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 860
 861         c->ctlr = h->ctlr;
 862         return c;
 863 }
 864
 865 /*
 866  * Frees a command block that was previously allocated with cmd_alloc().
 867  */
 868 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 869 {
 870         int i;
 871         u64bit temp64;
 872
 873         if (!got_from_pool) {
 874                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 875                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 876                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 877                                     c->err_info, (dma_addr_t) temp64.val);
 878                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 879                                     c, (dma_addr_t) c->busaddr);
 880         } else {
 881                 i = c - h->cmd_pool;
 882                 clear_bit(i & (BITS_PER_LONG - 1),
 883                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 884                 h->nr_frees++;
 885         }
 886 }
 887
 888 static inline ctlr_info_t *get_host(struct gendisk *disk)
 889 {
 890         return disk->queue->queuedata;
 891 }
 892
 893 static inline drive_info_struct *get_drv(struct gendisk *disk)
 894 {
 895         return disk->private_data;
 896 }
 897
 898 /*
 899  * Open.  Make sure the device is really there.
 900  */
 901 static int cciss_open(struct block_device *bdev, fmode_t mode)
 902 {
 903         ctlr_info_t *host = get_host(bdev->bd_disk);
 904         drive_info_struct *drv = get_drv(bdev->bd_disk);
 905
 906 #ifdef CCISS_DEBUG
 907         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
 908 #endif                          /* CCISS_DEBUG */
 909
 910         if (drv->busy_configuring)
 911                 return -EBUSY;
 912         /*
 913          * Root is allowed to open raw volume zero even if it's not configured
 914          * so array config can still work. Root is also allowed to open any
 915          * volume that has a LUN ID, so it can issue IOCTL to reread the
 916          * disk information.  I don't think I really like this
 917          * but I'm already using way to many device nodes to claim another one
 918          * for "raw controller".
 919          */
 920         if (drv->heads == 0) {
 921                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
 922                         /* if not node 0 make sure it is a partition = 0 */
 923                         if (MINOR(bdev->bd_dev) & 0x0f) {
 924                                 return -ENXIO;
 925                                 /* if it is, make sure we have a LUN ID */
 926                         } else if (memcmp(drv->LunID, CTLR_LUNID,
 927                                 sizeof(drv->LunID))) {
 928                                 return -ENXIO;
 929                         }
 930                 }
 931                 if (!capable(CAP_SYS_ADMIN))
 932                         return -EPERM;
 933         }
 934         drv->usage_count++;
 935         host->usage_count++;
 936         return 0;
 937 }
 938
 939 /*
 940  * Close.  Sync first.
 941  */
 942 static int cciss_release(struct gendisk *disk, fmode_t mode)
 943 {
 944         ctlr_info_t *host = get_host(disk);
 945         drive_info_struct *drv = get_drv(disk);
 946
 947 #ifdef CCISS_DEBUG
 948         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
 949 #endif                          /* CCISS_DEBUG */
 950
 951         drv->usage_count--;
 952         host->usage_count--;
 953         return 0;
 954 }
 955
 956 #ifdef CONFIG_COMPAT
 957
 958 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 959                     unsigned cmd, unsigned long arg)
 960 {
 961         int ret;
 962         lock_kernel();
 963         ret = cciss_ioctl(bdev, mode, cmd, arg);
 964         unlock_kernel();
 965         return ret;
 966 }
 967
 968 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 969                                   unsigned cmd, unsigned long arg);
 970 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
 971                                       unsigned cmd, unsigned long arg);
 972
 973 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
 974                               unsigned cmd, unsigned long arg)
 975 {
 976         switch (cmd) {
 977         case CCISS_GETPCIINFO:
 978         case CCISS_GETINTINFO:
 979         case CCISS_SETINTINFO:
 980         case CCISS_GETNODENAME:
 981         case CCISS_SETNODENAME:
 982         case CCISS_GETHEARTBEAT:
 983         case CCISS_GETBUSTYPES:
 984         case CCISS_GETFIRMVER:
 985         case CCISS_GETDRIVVER:
 986         case CCISS_REVALIDVOLS:
 987         case CCISS_DEREGDISK:
 988         case CCISS_REGNEWDISK:
 989         case CCISS_REGNEWD:
 990         case CCISS_RESCANDISK:
 991         case CCISS_GETLUNINFO:
 992                 return do_ioctl(bdev, mode, cmd, arg);
 993
 994         case CCISS_PASSTHRU32:
 995                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
 996         case CCISS_BIG_PASSTHRU32:
 997                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
 998
 999         default:
1000                 return -ENOIOCTLCMD;
1001         }
1002 }
1003
1004 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1005                                   unsigned cmd, unsigned long arg)
1006 {
1007         IOCTL32_Command_struct __user *arg32 =
1008             (IOCTL32_Command_struct __user *) arg;
1009         IOCTL_Command_struct arg64;
1010         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1011         int err;
1012         u32 cp;
1013
1014         err = 0;
1015         err |=
1016             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1017                            sizeof(arg64.LUN_info));
1018         err |=
1019             copy_from_user(&arg64.Request, &arg32->Request,
1020                            sizeof(arg64.Request));
1021         err |=
1022             copy_from_user(&arg64.error_info, &arg32->error_info,
1023                            sizeof(arg64.error_info));
1024         err |= get_user(arg64.buf_size, &arg32->buf_size);
1025         err |= get_user(cp, &arg32->buf);
1026         arg64.buf = compat_ptr(cp);
1027         err |= copy_to_user(p, &arg64, sizeof(arg64));
1028
1029         if (err)
1030                 return -EFAULT;
1031
1032         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1033         if (err)
1034                 return err;
1035         err |=
1036             copy_in_user(&arg32->error_info, &p->error_info,
1037                          sizeof(arg32->error_info));
1038         if (err)
1039                 return -EFAULT;
1040         return err;
1041 }
1042
1043 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1044                                       unsigned cmd, unsigned long arg)
1045 {
1046         BIG_IOCTL32_Command_struct __user *arg32 =
1047             (BIG_IOCTL32_Command_struct __user *) arg;
1048         BIG_IOCTL_Command_struct arg64;
1049         BIG_IOCTL_Command_struct __user *p =
1050             compat_alloc_user_space(sizeof(arg64));
1051         int err;
1052         u32 cp;
1053
1054         err = 0;
1055         err |=
1056             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1057                            sizeof(arg64.LUN_info));
1058         err |=
1059             copy_from_user(&arg64.Request, &arg32->Request,
1060                            sizeof(arg64.Request));
1061         err |=
1062             copy_from_user(&arg64.error_info, &arg32->error_info,
1063                            sizeof(arg64.error_info));
1064         err |= get_user(arg64.buf_size, &arg32->buf_size);
1065         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1066         err |= get_user(cp, &arg32->buf);
1067         arg64.buf = compat_ptr(cp);
1068         err |= copy_to_user(p, &arg64, sizeof(arg64));
1069
1070         if (err)
1071                 return -EFAULT;
1072
1073         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1074         if (err)
1075                 return err;
1076         err |=
1077             copy_in_user(&arg32->error_info, &p->error_info,
1078                          sizeof(arg32->error_info));
1079         if (err)
1080                 return -EFAULT;
1081         return err;
1082 }
1083 #endif
1084
1085 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1086 {
1087         drive_info_struct *drv = get_drv(bdev->bd_disk);
1088
1089         if (!drv->cylinders)
1090                 return -ENXIO;
1091
1092         geo->heads = drv->heads;
1093         geo->sectors = drv->sectors;
1094         geo->cylinders = drv->cylinders;
1095         return 0;
1096 }
1097
1098 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1099 {
1100         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1101                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1102                 (void)check_for_unit_attention(host, c);
1103 }
1104 /*
1105  * ioctl
1106  */
1107 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1108                        unsigned int cmd, unsigned long arg)
1109 {
1110         struct gendisk *disk = bdev->bd_disk;
1111         ctlr_info_t *host = get_host(disk);
1112         drive_info_struct *drv = get_drv(disk);
1113         int ctlr = host->ctlr;
1114         void __user *argp = (void __user *)arg;
1115
1116 #ifdef CCISS_DEBUG
1117         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1118 #endif                          /* CCISS_DEBUG */
1119
1120         switch (cmd) {
1121         case CCISS_GETPCIINFO:
1122                 {
1123                         cciss_pci_info_struct pciinfo;
1124
1125                         if (!arg)
1126                                 return -EINVAL;
1127                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1128                         pciinfo.bus = host->pdev->bus->number;
1129                         pciinfo.dev_fn = host->pdev->devfn;
1130                         pciinfo.board_id = host->board_id;
1131                         if (copy_to_user
1132                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1133                                 return -EFAULT;
1134                         return 0;
1135                 }
1136         case CCISS_GETINTINFO:
1137                 {
1138                         cciss_coalint_struct intinfo;
1139                         if (!arg)
1140                                 return -EINVAL;
1141                         intinfo.delay =
1142                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1143                         intinfo.count =
1144                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1145                         if (copy_to_user
1146                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1147                                 return -EFAULT;
1148                         return 0;
1149                 }
1150         case CCISS_SETINTINFO:
1151                 {
1152                         cciss_coalint_struct intinfo;
1153                         unsigned long flags;
1154                         int i;
1155
1156                         if (!arg)
1157                                 return -EINVAL;
1158                         if (!capable(CAP_SYS_ADMIN))
1159                                 return -EPERM;
1160                         if (copy_from_user
1161                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1162                                 return -EFAULT;
1163                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1164                         {
1165 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1166                                 return -EINVAL;
1167                         }
1168                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1169                         /* Update the field, and then ring the doorbell */
1170                         writel(intinfo.delay,
1171                                &(host->cfgtable->HostWrite.CoalIntDelay));
1172                         writel(intinfo.count,
1173                                &(host->cfgtable->HostWrite.CoalIntCount));
1174                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1175
1176                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1177                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1178                                       & CFGTBL_ChangeReq))
1179                                         break;
1180                                 /* delay and try again */
1181                                 udelay(1000);
1182                         }
1183                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1184                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1185                                 return -EAGAIN;
1186                         return 0;
1187                 }
1188         case CCISS_GETNODENAME:
1189                 {
1190                         NodeName_type NodeName;
1191                         int i;
1192
1193                         if (!arg)
1194                                 return -EINVAL;
1195                         for (i = 0; i < 16; i++)
1196                                 NodeName[i] =
1197                                     readb(&host->cfgtable->ServerName[i]);
1198                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1199                                 return -EFAULT;
1200                         return 0;
1201                 }
1202         case CCISS_SETNODENAME:
1203                 {
1204                         NodeName_type NodeName;
1205                         unsigned long flags;
1206                         int i;
1207
1208                         if (!arg)
1209                                 return -EINVAL;
1210                         if (!capable(CAP_SYS_ADMIN))
1211                                 return -EPERM;
1212
1213                         if (copy_from_user
1214                             (NodeName, argp, sizeof(NodeName_type)))
1215                                 return -EFAULT;
1216
1217                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1218
1219                         /* Update the field, and then ring the doorbell */
1220                         for (i = 0; i < 16; i++)
1221                                 writeb(NodeName[i],
1222                                        &host->cfgtable->ServerName[i]);
1223
1224                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1225
1226                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1227                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1228                                       & CFGTBL_ChangeReq))
1229                                         break;
1230                                 /* delay and try again */
1231                                 udelay(1000);
1232                         }
1233                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1234                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1235                                 return -EAGAIN;
1236                         return 0;
1237                 }
1238
1239         case CCISS_GETHEARTBEAT:
1240                 {
1241                         Heartbeat_type heartbeat;
1242
1243                         if (!arg)
1244                                 return -EINVAL;
1245                         heartbeat = readl(&host->cfgtable->HeartBeat);
1246                         if (copy_to_user
1247                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1248                                 return -EFAULT;
1249                         return 0;
1250                 }
1251         case CCISS_GETBUSTYPES:
1252                 {
1253                         BusTypes_type BusTypes;
1254
1255                         if (!arg)
1256                                 return -EINVAL;
1257                         BusTypes = readl(&host->cfgtable->BusTypes);
1258                         if (copy_to_user
1259                             (argp, &BusTypes, sizeof(BusTypes_type)))
1260                                 return -EFAULT;
1261                         return 0;
1262                 }
1263         case CCISS_GETFIRMVER:
1264                 {
1265                         FirmwareVer_type firmware;
1266
1267                         if (!arg)
1268                                 return -EINVAL;
1269                         memcpy(firmware, host->firm_ver, 4);
1270
1271                         if (copy_to_user
1272                             (argp, firmware, sizeof(FirmwareVer_type)))
1273                                 return -EFAULT;
1274                         return 0;
1275                 }
1276         case CCISS_GETDRIVVER:
1277                 {
1278                         DriverVer_type DriverVer = DRIVER_VERSION;
1279
1280                         if (!arg)
1281                                 return -EINVAL;
1282
1283                         if (copy_to_user
1284                             (argp, &DriverVer, sizeof(DriverVer_type)))
1285                                 return -EFAULT;
1286                         return 0;
1287                 }
1288
1289         case CCISS_DEREGDISK:
1290         case CCISS_REGNEWD:
1291         case CCISS_REVALIDVOLS:
1292                 return rebuild_lun_table(host, 0, 1);
1293
1294         case CCISS_GETLUNINFO:{
1295                         LogvolInfo_struct luninfo;
1296
1297                         memcpy(&luninfo.LunID, drv->LunID,
1298                                 sizeof(luninfo.LunID));
1299                         luninfo.num_opens = drv->usage_count;
1300                         luninfo.num_parts = 0;
1301                         if (copy_to_user(argp, &luninfo,
1302                                          sizeof(LogvolInfo_struct)))
1303                                 return -EFAULT;
1304                         return 0;
1305                 }
1306         case CCISS_PASSTHRU:
1307                 {
1308                         IOCTL_Command_struct iocommand;
1309                         CommandList_struct *c;
1310                         char *buff = NULL;
1311                         u64bit temp64;
1312                         unsigned long flags;
1313                         DECLARE_COMPLETION_ONSTACK(wait);
1314
1315                         if (!arg)
1316                                 return -EINVAL;
1317
1318                         if (!capable(CAP_SYS_RAWIO))
1319                                 return -EPERM;
1320
1321                         if (copy_from_user
1322                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1323                                 return -EFAULT;
1324                         if ((iocommand.buf_size < 1) &&
1325                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1326                                 return -EINVAL;
1327                         }
1328 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1329                         /* Check kmalloc limits */
1330                         if (iocommand.buf_size > 128000)
1331                                 return -EINVAL;
1332 #endif
1333                         if (iocommand.buf_size > 0) {
1334                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1335                                 if (buff == NULL)
1336                                         return -EFAULT;
1337                         }
1338                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1339                                 /* Copy the data into the buffer we created */
1340                                 if (copy_from_user
1341                                     (buff, iocommand.buf, iocommand.buf_size)) {
1342                                         kfree(buff);
1343                                         return -EFAULT;
1344                                 }
1345                         } else {
1346                                 memset(buff, 0, iocommand.buf_size);
1347                         }
1348                         if ((c = cmd_alloc(host, 0)) == NULL) {
1349                                 kfree(buff);
1350                                 return -ENOMEM;
1351                         }
1352                         // Fill in the command type
1353                         c->cmd_type = CMD_IOCTL_PEND;
1354                         // Fill in Command Header
1355                         c->Header.ReplyQueue = 0;       // unused in simple mode
1356                         if (iocommand.buf_size > 0)     // buffer to fill
1357                         {
1358                                 c->Header.SGList = 1;
1359                                 c->Header.SGTotal = 1;
1360                         } else  // no buffers to fill
1361                         {
1362                                 c->Header.SGList = 0;
1363                                 c->Header.SGTotal = 0;
1364                         }
1365                         c->Header.LUN = iocommand.LUN_info;
1366                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
1367
1368                         // Fill in Request block
1369                         c->Request = iocommand.Request;
1370
1371                         // Fill in the scatter gather information
1372                         if (iocommand.buf_size > 0) {
1373                                 temp64.val = pci_map_single(host->pdev, buff,
1374                                         iocommand.buf_size,
1375                                         PCI_DMA_BIDIRECTIONAL);
1376                                 c->SG[0].Addr.lower = temp64.val32.lower;
1377                                 c->SG[0].Addr.upper = temp64.val32.upper;
1378                                 c->SG[0].Len = iocommand.buf_size;
1379                                 c->SG[0].Ext = 0;       // we are not chaining
1380                         }
1381                         c->waiting = &wait;
1382
1383                         /* Put the request on the tail of the request queue */
1384                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1385                         addQ(&host->reqQ, c);
1386                         host->Qdepth++;
1387                         start_io(host);
1388                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1389
1390                         wait_for_completion(&wait);
1391
1392                         /* unlock the buffers from DMA */
1393                         temp64.val32.lower = c->SG[0].Addr.lower;
1394                         temp64.val32.upper = c->SG[0].Addr.upper;
1395                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1396                                          iocommand.buf_size,
1397                                          PCI_DMA_BIDIRECTIONAL);
1398
1399                         check_ioctl_unit_attention(host, c);
1400
1401                         /* Copy the error information out */
1402                         iocommand.error_info = *(c->err_info);
1403                         if (copy_to_user
1404                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1405                                 kfree(buff);
1406                                 cmd_free(host, c, 0);
1407                                 return -EFAULT;
1408                         }
1409
1410                         if (iocommand.Request.Type.Direction == XFER_READ) {
1411                                 /* Copy the data out of the buffer we created */
1412                                 if (copy_to_user
1413                                     (iocommand.buf, buff, iocommand.buf_size)) {
1414                                         kfree(buff);
1415                                         cmd_free(host, c, 0);
1416                                         return -EFAULT;
1417                                 }
1418                         }
1419                         kfree(buff);
1420                         cmd_free(host, c, 0);
1421                         return 0;
1422                 }
1423         case CCISS_BIG_PASSTHRU:{
1424                         BIG_IOCTL_Command_struct *ioc;
1425                         CommandList_struct *c;
1426                         unsigned char **buff = NULL;
1427                         int *buff_size = NULL;
1428                         u64bit temp64;
1429                         unsigned long flags;
1430                         BYTE sg_used = 0;
1431                         int status = 0;
1432                         int i;
1433                         DECLARE_COMPLETION_ONSTACK(wait);
1434                         __u32 left;
1435                         __u32 sz;
1436                         BYTE __user *data_ptr;
1437
1438                         if (!arg)
1439                                 return -EINVAL;
1440                         if (!capable(CAP_SYS_RAWIO))
1441                                 return -EPERM;
1442                         ioc = (BIG_IOCTL_Command_struct *)
1443                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1444                         if (!ioc) {
1445                                 status = -ENOMEM;
1446                                 goto cleanup1;
1447                         }
1448                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1449                                 status = -EFAULT;
1450                                 goto cleanup1;
1451                         }
1452                         if ((ioc->buf_size < 1) &&
1453                             (ioc->Request.Type.Direction != XFER_NONE)) {
1454                                 status = -EINVAL;
1455                                 goto cleanup1;
1456                         }
1457                         /* Check kmalloc limits  using all SGs */
1458                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1459                                 status = -EINVAL;
1460                                 goto cleanup1;
1461                         }
1462                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1463                                 status = -EINVAL;
1464                                 goto cleanup1;
1465                         }
1466                         buff =
1467                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1468                         if (!buff) {
1469                                 status = -ENOMEM;
1470                                 goto cleanup1;
1471                         }
1472                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1473                                                    GFP_KERNEL);
1474                         if (!buff_size) {
1475                                 status = -ENOMEM;
1476                                 goto cleanup1;
1477                         }
1478                         left = ioc->buf_size;
1479                         data_ptr = ioc->buf;
1480                         while (left) {
1481                                 sz = (left >
1482                                       ioc->malloc_size) ? ioc->
1483                                     malloc_size : left;
1484                                 buff_size[sg_used] = sz;
1485                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1486                                 if (buff[sg_used] == NULL) {
1487                                         status = -ENOMEM;
1488                                         goto cleanup1;
1489                                 }
1490                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1491                                         if (copy_from_user
1492                                             (buff[sg_used], data_ptr, sz)) {
1493                                                 status = -EFAULT;
1494                                                 goto cleanup1;
1495                                         }
1496                                 } else {
1497                                         memset(buff[sg_used], 0, sz);
1498                                 }
1499                                 left -= sz;
1500                                 data_ptr += sz;
1501                                 sg_used++;
1502                         }
1503                         if ((c = cmd_alloc(host, 0)) == NULL) {
1504                                 status = -ENOMEM;
1505                                 goto cleanup1;
1506                         }
1507                         c->cmd_type = CMD_IOCTL_PEND;
1508                         c->Header.ReplyQueue = 0;
1509
1510                         if (ioc->buf_size > 0) {
1511                                 c->Header.SGList = sg_used;
1512                                 c->Header.SGTotal = sg_used;
1513                         } else {
1514                                 c->Header.SGList = 0;
1515                                 c->Header.SGTotal = 0;
1516                         }
1517                         c->Header.LUN = ioc->LUN_info;
1518                         c->Header.Tag.lower = c->busaddr;
1519
1520                         c->Request = ioc->Request;
1521                         if (ioc->buf_size > 0) {
1522                                 int i;
1523                                 for (i = 0; i < sg_used; i++) {
1524                                         temp64.val =
1525                                             pci_map_single(host->pdev, buff[i],
1526                                                     buff_size[i],
1527                                                     PCI_DMA_BIDIRECTIONAL);
1528                                         c->SG[i].Addr.lower =
1529                                             temp64.val32.lower;
1530                                         c->SG[i].Addr.upper =
1531                                             temp64.val32.upper;
1532                                         c->SG[i].Len = buff_size[i];
1533                                         c->SG[i].Ext = 0;       /* we are not chaining */
1534                                 }
1535                         }
1536                         c->waiting = &wait;
1537                         /* Put the request on the tail of the request queue */
1538                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1539                         addQ(&host->reqQ, c);
1540                         host->Qdepth++;
1541                         start_io(host);
1542                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1543                         wait_for_completion(&wait);
1544                         /* unlock the buffers from DMA */
1545                         for (i = 0; i < sg_used; i++) {
1546                                 temp64.val32.lower = c->SG[i].Addr.lower;
1547                                 temp64.val32.upper = c->SG[i].Addr.upper;
1548                                 pci_unmap_single(host->pdev,
1549                                         (dma_addr_t) temp64.val, buff_size[i],
1550                                         PCI_DMA_BIDIRECTIONAL);
1551                         }
1552                         check_ioctl_unit_attention(host, c);
1553                         /* Copy the error information out */
1554                         ioc->error_info = *(c->err_info);
1555                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1556                                 cmd_free(host, c, 0);
1557                                 status = -EFAULT;
1558                                 goto cleanup1;
1559                         }
1560                         if (ioc->Request.Type.Direction == XFER_READ) {
1561                                 /* Copy the data out of the buffer we created */
1562                                 BYTE __user *ptr = ioc->buf;
1563                                 for (i = 0; i < sg_used; i++) {
1564                                         if (copy_to_user
1565                                             (ptr, buff[i], buff_size[i])) {
1566                                                 cmd_free(host, c, 0);
1567                                                 status = -EFAULT;
1568                                                 goto cleanup1;
1569                                         }
1570                                         ptr += buff_size[i];
1571                                 }
1572                         }
1573                         cmd_free(host, c, 0);
1574                         status = 0;
1575                       cleanup1:
1576                         if (buff) {
1577                                 for (i = 0; i < sg_used; i++)
1578                                         kfree(buff[i]);
1579                                 kfree(buff);
1580                         }
1581                         kfree(buff_size);
1582                         kfree(ioc);
1583                         return status;
1584                 }
1585
1586         /* scsi_cmd_ioctl handles these, below, though some are not */
1587         /* very meaningful for cciss.  SG_IO is the main one people want. */
1588
1589         case SG_GET_VERSION_NUM:
1590         case SG_SET_TIMEOUT:
1591         case SG_GET_TIMEOUT:
1592         case SG_GET_RESERVED_SIZE:
1593         case SG_SET_RESERVED_SIZE:
1594         case SG_EMULATED_HOST:
1595         case SG_IO:
1596         case SCSI_IOCTL_SEND_COMMAND:
1597                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1598
1599         /* scsi_cmd_ioctl would normally handle these, below, but */
1600         /* they aren't a good fit for cciss, as CD-ROMs are */
1601         /* not supported, and we don't have any bus/target/lun */
1602         /* which we present to the kernel. */
1603
1604         case CDROM_SEND_PACKET:
1605         case CDROMCLOSETRAY:
1606         case CDROMEJECT:
1607         case SCSI_IOCTL_GET_IDLUN:
1608         case SCSI_IOCTL_GET_BUS_NUMBER:
1609         default:
1610                 return -ENOTTY;
1611         }
1612 }
1613
1614 static void cciss_check_queues(ctlr_info_t *h)
1615 {
1616         int start_queue = h->next_to_run;
1617         int i;
1618
1619         /* check to see if we have maxed out the number of commands that can
1620          * be placed on the queue.  If so then exit.  We do this check here
1621          * in case the interrupt we serviced was from an ioctl and did not
1622          * free any new commands.
1623          */
1624         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1625                 return;
1626
1627         /* We have room on the queue for more commands.  Now we need to queue
1628          * them up.  We will also keep track of the next queue to run so
1629          * that every queue gets a chance to be started first.
1630          */
1631         for (i = 0; i < h->highest_lun + 1; i++) {
1632                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1633                 /* make sure the disk has been added and the drive is real
1634                  * because this can be called from the middle of init_one.
1635                  */
1636                 if (!h->drv[curr_queue])
1637                         continue;
1638                 if (!(h->drv[curr_queue]->queue) ||
1639                         !(h->drv[curr_queue]->heads))
1640                         continue;
1641                 blk_start_queue(h->gendisk[curr_queue]->queue);
1642
1643                 /* check to see if we have maxed out the number of commands
1644                  * that can be placed on the queue.
1645                  */
1646                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1647                         if (curr_queue == start_queue) {
1648                                 h->next_to_run =
1649                                     (start_queue + 1) % (h->highest_lun + 1);
1650                                 break;
1651                         } else {
1652                                 h->next_to_run = curr_queue;
1653                                 break;
1654                         }
1655                 }
1656         }
1657 }
1658
1659 static void cciss_softirq_done(struct request *rq)
1660 {
1661         CommandList_struct *cmd = rq->completion_data;
1662         ctlr_info_t *h = hba[cmd->ctlr];
1663         unsigned long flags;
1664         u64bit temp64;
1665         int i, ddir;
1666
1667         if (cmd->Request.Type.Direction == XFER_READ)
1668                 ddir = PCI_DMA_FROMDEVICE;
1669         else
1670                 ddir = PCI_DMA_TODEVICE;
1671
1672         /* command did not need to be retried */
1673         /* unmap the DMA mapping for all the scatter gather elements */
1674         for (i = 0; i < cmd->Header.SGList; i++) {
1675                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1676                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1677                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1678         }
1679
1680 #ifdef CCISS_DEBUG
1681         printk("Done with %p\n", rq);
1682 #endif                          /* CCISS_DEBUG */
1683
1684         /* set the residual count for pc requests */
1685         if (blk_pc_request(rq))
1686                 rq->resid_len = cmd->err_info->ResidualCnt;
1687
1688         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1689
1690         spin_lock_irqsave(&h->lock, flags);
1691         cmd_free(h, cmd, 1);
1692         cciss_check_queues(h);
1693         spin_unlock_irqrestore(&h->lock, flags);
1694 }
1695
1696 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1697         unsigned char scsi3addr[], uint32_t log_unit)
1698 {
1699         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1700                 sizeof(h->drv[log_unit]->LunID));
1701 }
1702
1703 /* This function gets the SCSI vendor, model, and revision of a logical drive
1704  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1705  * they cannot be read.
1706  */
1707 static void cciss_get_device_descr(int ctlr, int logvol, int withirq,
1708                                    char *vendor, char *model, char *rev)
1709 {
1710         int rc;
1711         InquiryData_struct *inq_buf;
1712         unsigned char scsi3addr[8];
1713
1714         *vendor = '\0';
1715         *model = '\0';
1716         *rev = '\0';
1717
1718         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1719         if (!inq_buf)
1720                 return;
1721
1722         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1723         if (withirq)
1724                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf,
1725                              sizeof(InquiryData_struct), 0,
1726                                 scsi3addr, TYPE_CMD);
1727         else
1728                 rc = sendcmd(CISS_INQUIRY, ctlr, inq_buf,
1729                              sizeof(InquiryData_struct), 0,
1730                                 scsi3addr, TYPE_CMD);
1731         if (rc == IO_OK) {
1732                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1733                 vendor[VENDOR_LEN] = '\0';
1734                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1735                 model[MODEL_LEN] = '\0';
1736                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1737                 rev[REV_LEN] = '\0';
1738         }
1739
1740         kfree(inq_buf);
1741         return;
1742 }
1743
1744 /* This function gets the serial number of a logical drive via
1745  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1746  * number cannot be had, for whatever reason, 16 bytes of 0xff
1747  * are returned instead.
1748  */
1749 static void cciss_get_serial_no(int ctlr, int logvol, int withirq,
1750                                 unsigned char *serial_no, int buflen)
1751 {
1752 #define PAGE_83_INQ_BYTES 64
1753         int rc;
1754         unsigned char *buf;
1755         unsigned char scsi3addr[8];
1756
1757         if (buflen > 16)
1758                 buflen = 16;
1759         memset(serial_no, 0xff, buflen);
1760         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1761         if (!buf)
1762                 return;
1763         memset(serial_no, 0, buflen);
1764         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1765         if (withirq)
1766                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1767                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1768         else
1769                 rc = sendcmd(CISS_INQUIRY, ctlr, buf,
1770                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1771         if (rc == IO_OK)
1772                 memcpy(serial_no, &buf[8], buflen);
1773         kfree(buf);
1774         return;
1775 }
1776
1777 /*
1778  * cciss_add_disk sets up the block device queue for a logical drive
1779  */
1780 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1781                                 int drv_index)
1782 {
1783         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1784         if (!disk->queue)
1785                 goto init_queue_failure;
1786         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1787         disk->major = h->major;
1788         disk->first_minor = drv_index << NWD_SHIFT;
1789         disk->fops = &cciss_fops;
1790         if (cciss_create_ld_sysfs_entry(h, drv_index))
1791                 goto cleanup_queue;
1792         disk->private_data = h->drv[drv_index];
1793         disk->driverfs_dev = &h->drv[drv_index]->dev;
1794
1795         /* Set up queue information */
1796         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1797
1798         /* This is a hardware imposed limit. */
1799         blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1800
1801         /* This is a limit in the driver and could be eliminated. */
1802         blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1803
1804         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1805
1806         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1807
1808         disk->queue->queuedata = h;
1809
1810         blk_queue_logical_block_size(disk->queue,
1811                                      h->drv[drv_index]->block_size);
1812
1813         /* Make sure all queue data is written out before */
1814         /* setting h->drv[drv_index]->queue, as setting this */
1815         /* allows the interrupt handler to start the queue */
1816         wmb();
1817         h->drv[drv_index]->queue = disk->queue;
1818         add_disk(disk);
1819         return 0;
1820
1821 cleanup_queue:
1822         blk_cleanup_queue(disk->queue);
1823         disk->queue = NULL;
1824 init_queue_failure:
1825         return -1;
1826 }
1827
1828 /* This function will check the usage_count of the drive to be updated/added.
1829  * If the usage_count is zero and it is a heretofore unknown drive, or,
1830  * the drive's capacity, geometry, or serial number has changed,
1831  * then the drive information will be updated and the disk will be
1832  * re-registered with the kernel.  If these conditions don't hold,
1833  * then it will be left alone for the next reboot.  The exception to this
1834  * is disk 0 which will always be left registered with the kernel since it
1835  * is also the controller node.  Any changes to disk 0 will show up on
1836  * the next reboot.
1837  */
1838 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1839         int via_ioctl)
1840 {
1841         ctlr_info_t *h = hba[ctlr];
1842         struct gendisk *disk;
1843         InquiryData_struct *inq_buff = NULL;
1844         unsigned int block_size;
1845         sector_t total_size;
1846         unsigned long flags = 0;
1847         int ret = 0;
1848         drive_info_struct *drvinfo;
1849
1850         /* Get information about the disk and modify the driver structure */
1851         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1852         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1853         if (inq_buff == NULL || drvinfo == NULL)
1854                 goto mem_msg;
1855
1856         /* testing to see if 16-byte CDBs are already being used */
1857         if (h->cciss_read == CCISS_READ_16) {
1858                 cciss_read_capacity_16(h->ctlr, drv_index, 1,
1859                         &total_size, &block_size);
1860
1861         } else {
1862                 cciss_read_capacity(ctlr, drv_index, 1,
1863                                     &total_size, &block_size);
1864
1865                 /* if read_capacity returns all F's this volume is >2TB */
1866                 /* in size so we switch to 16-byte CDB's for all */
1867                 /* read/write ops */
1868                 if (total_size == 0xFFFFFFFFULL) {
1869                         cciss_read_capacity_16(ctlr, drv_index, 1,
1870                         &total_size, &block_size);
1871                         h->cciss_read = CCISS_READ_16;
1872                         h->cciss_write = CCISS_WRITE_16;
1873                 } else {
1874                         h->cciss_read = CCISS_READ_10;
1875                         h->cciss_write = CCISS_WRITE_10;
1876                 }
1877         }
1878
1879         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1880                                inq_buff, drvinfo);
1881         drvinfo->block_size = block_size;
1882         drvinfo->nr_blocks = total_size + 1;
1883
1884         cciss_get_device_descr(ctlr, drv_index, 1, drvinfo->vendor,
1885                                 drvinfo->model, drvinfo->rev);
1886         cciss_get_serial_no(ctlr, drv_index, 1, drvinfo->serial_no,
1887                         sizeof(drvinfo->serial_no));
1888         /* Save the lunid in case we deregister the disk, below. */
1889         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1890                 sizeof(drvinfo->LunID));
1891
1892         /* Is it the same disk we already know, and nothing's changed? */
1893         if (h->drv[drv_index]->raid_level != -1 &&
1894                 ((memcmp(drvinfo->serial_no,
1895                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1896                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1897                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1898                 drvinfo->heads == h->drv[drv_index]->heads &&
1899                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1900                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1901                         /* The disk is unchanged, nothing to update */
1902                         goto freeret;
1903
1904         /* If we get here it's not the same disk, or something's changed,
1905          * so we need to * deregister it, and re-register it, if it's not
1906          * in use.
1907          * If the disk already exists then deregister it before proceeding
1908          * (unless it's the first disk (for the controller node).
1909          */
1910         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1911                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1912                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1913                 h->drv[drv_index]->busy_configuring = 1;
1914                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1915
1916                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1917                  * which keeps the interrupt handler from starting
1918                  * the queue.
1919                  */
1920                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1921         }
1922
1923         /* If the disk is in use return */
1924         if (ret)
1925                 goto freeret;
1926
1927         /* Save the new information from cciss_geometry_inquiry
1928          * and serial number inquiry.  If the disk was deregistered
1929          * above, then h->drv[drv_index] will be NULL.
1930          */
1931         if (h->drv[drv_index] == NULL) {
1932                 drvinfo->device_initialized = 0;
1933                 h->drv[drv_index] = drvinfo;
1934                 drvinfo = NULL; /* so it won't be freed below. */
1935         } else {
1936                 /* special case for cxd0 */
1937                 h->drv[drv_index]->block_size = drvinfo->block_size;
1938                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
1939                 h->drv[drv_index]->heads = drvinfo->heads;
1940                 h->drv[drv_index]->sectors = drvinfo->sectors;
1941                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
1942                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
1943                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
1944                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
1945                         VENDOR_LEN + 1);
1946                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
1947                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
1948         }
1949
1950         ++h->num_luns;
1951         disk = h->gendisk[drv_index];
1952         set_capacity(disk, h->drv[drv_index]->nr_blocks);
1953
1954         /* If it's not disk 0 (drv_index != 0)
1955          * or if it was disk 0, but there was previously
1956          * no actual corresponding configured logical drive
1957          * (raid_leve == -1) then we want to update the
1958          * logical drive's information.
1959          */
1960         if (drv_index || first_time) {
1961                 if (cciss_add_disk(h, disk, drv_index) != 0) {
1962                         cciss_free_gendisk(h, drv_index);
1963                         cciss_free_drive_info(h, drv_index);
1964                         printk(KERN_WARNING "cciss:%d could not update "
1965                                 "disk %d\n", h->ctlr, drv_index);
1966                         --h->num_luns;
1967                 }
1968         }
1969
1970 freeret:
1971         kfree(inq_buff);
1972         kfree(drvinfo);
1973         return;
1974 mem_msg:
1975         printk(KERN_ERR "cciss: out of memory\n");
1976         goto freeret;
1977 }
1978
1979 /* This function will find the first index of the controllers drive array
1980  * that has a null drv pointer and allocate the drive info struct and
1981  * will return that index   This is where new drives will be added.
1982  * If the index to be returned is greater than the highest_lun index for
1983  * the controller then highest_lun is set * to this new index.
1984  * If there are no available indexes or if tha allocation fails, then -1
1985  * is returned.  * "controller_node" is used to know if this is a real
1986  * logical drive, or just the controller node, which determines if this
1987  * counts towards highest_lun.
1988  */
1989 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
1990 {
1991         int i;
1992         drive_info_struct *drv;
1993
1994         /* Search for an empty slot for our drive info */
1995         for (i = 0; i < CISS_MAX_LUN; i++) {
1996
1997                 /* if not cxd0 case, and it's occupied, skip it. */
1998                 if (h->drv[i] && i != 0)
1999                         continue;
2000                 /*
2001                  * If it's cxd0 case, and drv is alloc'ed already, and a
2002                  * disk is configured there, skip it.
2003                  */
2004                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2005                         continue;
2006
2007                 /*
2008                  * We've found an empty slot.  Update highest_lun
2009                  * provided this isn't just the fake cxd0 controller node.
2010                  */
2011                 if (i > h->highest_lun && !controller_node)
2012                         h->highest_lun = i;
2013
2014                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2015                 if (i == 0 && h->drv[i] != NULL)
2016                         return i;
2017
2018                 /*
2019                  * Found an empty slot, not already alloc'ed.  Allocate it.
2020                  * Mark it with raid_level == -1, so we know it's new later on.
2021                  */
2022                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2023                 if (!drv)
2024                         return -1;
2025                 drv->raid_level = -1; /* so we know it's new */
2026                 h->drv[i] = drv;
2027                 return i;
2028         }
2029         return -1;
2030 }
2031
2032 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2033 {
2034         kfree(h->drv[drv_index]);
2035         h->drv[drv_index] = NULL;
2036 }
2037
2038 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2039 {
2040         put_disk(h->gendisk[drv_index]);
2041         h->gendisk[drv_index] = NULL;
2042 }
2043
2044 /* cciss_add_gendisk finds a free hba[]->drv structure
2045  * and allocates a gendisk if needed, and sets the lunid
2046  * in the drvinfo structure.   It returns the index into
2047  * the ->drv[] array, or -1 if none are free.
2048  * is_controller_node indicates whether highest_lun should
2049  * count this disk, or if it's only being added to provide
2050  * a means to talk to the controller in case no logical
2051  * drives have yet been configured.
2052  */
2053 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2054         int controller_node)
2055 {
2056         int drv_index;
2057
2058         drv_index = cciss_alloc_drive_info(h, controller_node);
2059         if (drv_index == -1)
2060                 return -1;
2061
2062         /*Check if the gendisk needs to be allocated */
2063         if (!h->gendisk[drv_index]) {
2064                 h->gendisk[drv_index] =
2065                         alloc_disk(1 << NWD_SHIFT);
2066                 if (!h->gendisk[drv_index]) {
2067                         printk(KERN_ERR "cciss%d: could not "
2068                                 "allocate a new disk %d\n",
2069                                 h->ctlr, drv_index);
2070                         goto err_free_drive_info;
2071                 }
2072         }
2073         memcpy(h->drv[drv_index]->LunID, lunid,
2074                 sizeof(h->drv[drv_index]->LunID));
2075         if (cciss_create_ld_sysfs_entry(h, drv_index))
2076                 goto err_free_disk;
2077         /* Don't need to mark this busy because nobody */
2078         /* else knows about this disk yet to contend */
2079         /* for access to it. */
2080         h->drv[drv_index]->busy_configuring = 0;
2081         wmb();
2082         return drv_index;
2083
2084 err_free_disk:
2085         cciss_free_gendisk(h, drv_index);
2086 err_free_drive_info:
2087         cciss_free_drive_info(h, drv_index);
2088         return -1;
2089 }
2090
2091 /* This is for the special case of a controller which
2092  * has no logical drives.  In this case, we still need
2093  * to register a disk so the controller can be accessed
2094  * by the Array Config Utility.
2095  */
2096 static void cciss_add_controller_node(ctlr_info_t *h)
2097 {
2098         struct gendisk *disk;
2099         int drv_index;
2100
2101         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2102                 return;
2103
2104         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2105         if (drv_index == -1)
2106                 goto error;
2107         h->drv[drv_index]->block_size = 512;
2108         h->drv[drv_index]->nr_blocks = 0;
2109         h->drv[drv_index]->heads = 0;
2110         h->drv[drv_index]->sectors = 0;
2111         h->drv[drv_index]->cylinders = 0;
2112         h->drv[drv_index]->raid_level = -1;
2113         memset(h->drv[drv_index]->serial_no, 0, 16);
2114         disk = h->gendisk[drv_index];
2115         if (cciss_add_disk(h, disk, drv_index) == 0)
2116                 return;
2117         cciss_free_gendisk(h, drv_index);
2118         cciss_free_drive_info(h, drv_index);
2119 error:
2120         printk(KERN_WARNING "cciss%d: could not "
2121                 "add disk 0.\n", h->ctlr);
2122         return;
2123 }
2124
2125 /* This function will add and remove logical drives from the Logical
2126  * drive array of the controller and maintain persistency of ordering
2127  * so that mount points are preserved until the next reboot.  This allows
2128  * for the removal of logical drives in the middle of the drive array
2129  * without a re-ordering of those drives.
2130  * INPUT
2131  * h            = The controller to perform the operations on
2132  */
2133 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2134         int via_ioctl)
2135 {
2136         int ctlr = h->ctlr;
2137         int num_luns;
2138         ReportLunData_struct *ld_buff = NULL;
2139         int return_code;
2140         int listlength = 0;
2141         int i;
2142         int drv_found;
2143         int drv_index = 0;
2144         unsigned char lunid[8] = CTLR_LUNID;
2145         unsigned long flags;
2146
2147         if (!capable(CAP_SYS_RAWIO))
2148                 return -EPERM;
2149
2150         /* Set busy_configuring flag for this operation */
2151         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2152         if (h->busy_configuring) {
2153                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2154                 return -EBUSY;
2155         }
2156         h->busy_configuring = 1;
2157         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2158
2159         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2160         if (ld_buff == NULL)
2161                 goto mem_msg;
2162
2163         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2164                                       sizeof(ReportLunData_struct),
2165                                       0, CTLR_LUNID, TYPE_CMD);
2166
2167         if (return_code == IO_OK)
2168                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2169         else {  /* reading number of logical volumes failed */
2170                 printk(KERN_WARNING "cciss: report logical volume"
2171                        " command failed\n");
2172                 listlength = 0;
2173                 goto freeret;
2174         }
2175
2176         num_luns = listlength / 8;      /* 8 bytes per entry */
2177         if (num_luns > CISS_MAX_LUN) {
2178                 num_luns = CISS_MAX_LUN;
2179                 printk(KERN_WARNING "cciss: more luns configured"
2180                        " on controller than can be handled by"
2181                        " this driver.\n");
2182         }
2183
2184         if (num_luns == 0)
2185                 cciss_add_controller_node(h);
2186
2187         /* Compare controller drive array to driver's drive array
2188          * to see if any drives are missing on the controller due
2189          * to action of Array Config Utility (user deletes drive)
2190          * and deregister logical drives which have disappeared.
2191          */
2192         for (i = 0; i <= h->highest_lun; i++) {
2193                 int j;
2194                 drv_found = 0;
2195
2196                 /* skip holes in the array from already deleted drives */
2197                 if (h->drv[i] == NULL)
2198                         continue;
2199
2200                 for (j = 0; j < num_luns; j++) {
2201                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2202                         if (memcmp(h->drv[i]->LunID, lunid,
2203                                 sizeof(lunid)) == 0) {
2204                                 drv_found = 1;
2205                                 break;
2206                         }
2207                 }
2208                 if (!drv_found) {
2209                         /* Deregister it from the OS, it's gone. */
2210                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2211                         h->drv[i]->busy_configuring = 1;
2212                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2213                         return_code = deregister_disk(h, i, 1, via_ioctl);
2214                         if (h->drv[i] != NULL)
2215                                 h->drv[i]->busy_configuring = 0;
2216                 }
2217         }
2218
2219         /* Compare controller drive array to driver's drive array.
2220          * Check for updates in the drive information and any new drives
2221          * on the controller due to ACU adding logical drives, or changing
2222          * a logical drive's size, etc.  Reregister any new/changed drives
2223          */
2224         for (i = 0; i < num_luns; i++) {
2225                 int j;
2226
2227                 drv_found = 0;
2228
2229                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2230                 /* Find if the LUN is already in the drive array
2231                  * of the driver.  If so then update its info
2232                  * if not in use.  If it does not exist then find
2233                  * the first free index and add it.
2234                  */
2235                 for (j = 0; j <= h->highest_lun; j++) {
2236                         if (h->drv[j] != NULL &&
2237                                 memcmp(h->drv[j]->LunID, lunid,
2238                                         sizeof(h->drv[j]->LunID)) == 0) {
2239                                 drv_index = j;
2240                                 drv_found = 1;
2241                                 break;
2242                         }
2243                 }
2244
2245                 /* check if the drive was found already in the array */
2246                 if (!drv_found) {
2247                         drv_index = cciss_add_gendisk(h, lunid, 0);
2248                         if (drv_index == -1)
2249                                 goto freeret;
2250                 }
2251                 cciss_update_drive_info(ctlr, drv_index, first_time,
2252                         via_ioctl);
2253         }               /* end for */
2254
2255 freeret:
2256         kfree(ld_buff);
2257         h->busy_configuring = 0;
2258         /* We return -1 here to tell the ACU that we have registered/updated
2259          * all of the drives that we can and to keep it from calling us
2260          * additional times.
2261          */
2262         return -1;
2263 mem_msg:
2264         printk(KERN_ERR "cciss: out of memory\n");
2265         h->busy_configuring = 0;
2266         goto freeret;
2267 }
2268
2269 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2270 {
2271         /* zero out the disk size info */
2272         drive_info->nr_blocks = 0;
2273         drive_info->block_size = 0;
2274         drive_info->heads = 0;
2275         drive_info->sectors = 0;
2276         drive_info->cylinders = 0;
2277         drive_info->raid_level = -1;
2278         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2279         memset(drive_info->model, 0, sizeof(drive_info->model));
2280         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2281         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2282         /*
2283          * don't clear the LUNID though, we need to remember which
2284          * one this one is.
2285          */
2286 }
2287
2288 /* This function will deregister the disk and it's queue from the
2289  * kernel.  It must be called with the controller lock held and the
2290  * drv structures busy_configuring flag set.  It's parameters are:
2291  *
2292  * disk = This is the disk to be deregistered
2293  * drv  = This is the drive_info_struct associated with the disk to be
2294  *        deregistered.  It contains information about the disk used
2295  *        by the driver.
2296  * clear_all = This flag determines whether or not the disk information
2297  *             is going to be completely cleared out and the highest_lun
2298  *             reset.  Sometimes we want to clear out information about
2299  *             the disk in preparation for re-adding it.  In this case
2300  *             the highest_lun should be left unchanged and the LunID
2301  *             should not be cleared.
2302  * via_ioctl
2303  *    This indicates whether we've reached this path via ioctl.
2304  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2305  *    If this path is reached via ioctl(), then the max_usage_count will
2306  *    be 1, as the process calling ioctl() has got to have the device open.
2307  *    If we get here via sysfs, then the max usage count will be zero.
2308 */
2309 static int deregister_disk(ctlr_info_t *h, int drv_index,
2310                            int clear_all, int via_ioctl)
2311 {
2312         int i;
2313         struct gendisk *disk;
2314         drive_info_struct *drv;
2315         int recalculate_highest_lun;
2316
2317         if (!capable(CAP_SYS_RAWIO))
2318                 return -EPERM;
2319
2320         drv = h->drv[drv_index];
2321         disk = h->gendisk[drv_index];
2322
2323         /* make sure logical volume is NOT is use */
2324         if (clear_all || (h->gendisk[0] == disk)) {
2325                 if (drv->usage_count > via_ioctl)
2326                         return -EBUSY;
2327         } else if (drv->usage_count > 0)
2328                 return -EBUSY;
2329
2330         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2331
2332         /* invalidate the devices and deregister the disk.  If it is disk
2333          * zero do not deregister it but just zero out it's values.  This
2334          * allows us to delete disk zero but keep the controller registered.
2335          */
2336         if (h->gendisk[0] != disk) {
2337                 struct request_queue *q = disk->queue;
2338                 if (disk->flags & GENHD_FL_UP) {
2339                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2340                         del_gendisk(disk);
2341                 }
2342                 if (q)
2343                         blk_cleanup_queue(q);
2344                 /* If clear_all is set then we are deleting the logical
2345                  * drive, not just refreshing its info.  For drives
2346                  * other than disk 0 we will call put_disk.  We do not
2347                  * do this for disk 0 as we need it to be able to
2348                  * configure the controller.
2349                  */
2350                 if (clear_all){
2351                         /* This isn't pretty, but we need to find the
2352                          * disk in our array and NULL our the pointer.
2353                          * This is so that we will call alloc_disk if
2354                          * this index is used again later.
2355                          */
2356                         for (i=0; i < CISS_MAX_LUN; i++){
2357                                 if (h->gendisk[i] == disk) {
2358                                         h->gendisk[i] = NULL;
2359                                         break;
2360                                 }
2361                         }
2362                         put_disk(disk);
2363                 }
2364         } else {
2365                 set_capacity(disk, 0);
2366                 cciss_clear_drive_info(drv);
2367         }
2368
2369         --h->num_luns;
2370
2371         /* if it was the last disk, find the new hightest lun */
2372         if (clear_all && recalculate_highest_lun) {
2373                 int i, newhighest = -1;
2374                 for (i = 0; i <= h->highest_lun; i++) {
2375                         /* if the disk has size > 0, it is available */
2376                         if (h->drv[i] && h->drv[i]->heads)
2377                                 newhighest = i;
2378                 }
2379                 h->highest_lun = newhighest;
2380         }
2381         return 0;
2382 }
2383
2384 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2385                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2386                 int cmd_type)
2387 {
2388         ctlr_info_t *h = hba[ctlr];
2389         u64bit buff_dma_handle;
2390         int status = IO_OK;
2391
2392         c->cmd_type = CMD_IOCTL_PEND;
2393         c->Header.ReplyQueue = 0;
2394         if (buff != NULL) {
2395                 c->Header.SGList = 1;
2396                 c->Header.SGTotal = 1;
2397         } else {
2398                 c->Header.SGList = 0;
2399                 c->Header.SGTotal = 0;
2400         }
2401         c->Header.Tag.lower = c->busaddr;
2402         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2403
2404         c->Request.Type.Type = cmd_type;
2405         if (cmd_type == TYPE_CMD) {
2406                 switch (cmd) {
2407                 case CISS_INQUIRY:
2408                         /* are we trying to read a vital product page */
2409                         if (page_code != 0) {
2410                                 c->Request.CDB[1] = 0x01;
2411                                 c->Request.CDB[2] = page_code;
2412                         }
2413                         c->Request.CDBLen = 6;
2414                         c->Request.Type.Attribute = ATTR_SIMPLE;
2415                         c->Request.Type.Direction = XFER_READ;
2416                         c->Request.Timeout = 0;
2417                         c->Request.CDB[0] = CISS_INQUIRY;
2418                         c->Request.CDB[4] = size & 0xFF;
2419                         break;
2420                 case CISS_REPORT_LOG:
2421                 case CISS_REPORT_PHYS:
2422                         /* Talking to controller so It's a physical command
2423                            mode = 00 target = 0.  Nothing to write.
2424                          */
2425                         c->Request.CDBLen = 12;
2426                         c->Request.Type.Attribute = ATTR_SIMPLE;
2427                         c->Request.Type.Direction = XFER_READ;
2428                         c->Request.Timeout = 0;
2429                         c->Request.CDB[0] = cmd;
2430                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
2431                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2432                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2433                         c->Request.CDB[9] = size & 0xFF;
2434                         break;
2435
2436                 case CCISS_READ_CAPACITY:
2437                         c->Request.CDBLen = 10;
2438                         c->Request.Type.Attribute = ATTR_SIMPLE;
2439                         c->Request.Type.Direction = XFER_READ;
2440                         c->Request.Timeout = 0;
2441                         c->Request.CDB[0] = cmd;
2442                         break;
2443                 case CCISS_READ_CAPACITY_16:
2444                         c->Request.CDBLen = 16;
2445                         c->Request.Type.Attribute = ATTR_SIMPLE;
2446                         c->Request.Type.Direction = XFER_READ;
2447                         c->Request.Timeout = 0;
2448                         c->Request.CDB[0] = cmd;
2449                         c->Request.CDB[1] = 0x10;
2450                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2451                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2452                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2453                         c->Request.CDB[13] = size & 0xFF;
2454                         c->Request.Timeout = 0;
2455                         c->Request.CDB[0] = cmd;
2456                         break;
2457                 case CCISS_CACHE_FLUSH:
2458                         c->Request.CDBLen = 12;
2459                         c->Request.Type.Attribute = ATTR_SIMPLE;
2460                         c->Request.Type.Direction = XFER_WRITE;
2461                         c->Request.Timeout = 0;
2462                         c->Request.CDB[0] = BMIC_WRITE;
2463                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2464                         break;
2465                 case TEST_UNIT_READY:
2466                         c->Request.CDBLen = 6;
2467                         c->Request.Type.Attribute = ATTR_SIMPLE;
2468                         c->Request.Type.Direction = XFER_NONE;
2469                         c->Request.Timeout = 0;
2470                         break;
2471                 default:
2472                         printk(KERN_WARNING
2473                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2474                         return IO_ERROR;
2475                 }
2476         } else if (cmd_type == TYPE_MSG) {
2477                 switch (cmd) {
2478                 case 0: /* ABORT message */
2479                         c->Request.CDBLen = 12;
2480                         c->Request.Type.Attribute = ATTR_SIMPLE;
2481                         c->Request.Type.Direction = XFER_WRITE;
2482                         c->Request.Timeout = 0;
2483                         c->Request.CDB[0] = cmd;        /* abort */
2484                         c->Request.CDB[1] = 0;  /* abort a command */
2485                         /* buff contains the tag of the command to abort */
2486                         memcpy(&c->Request.CDB[4], buff, 8);
2487                         break;
2488                 case 1: /* RESET message */
2489                         c->Request.CDBLen = 16;
2490                         c->Request.Type.Attribute = ATTR_SIMPLE;
2491                         c->Request.Type.Direction = XFER_NONE;
2492                         c->Request.Timeout = 0;
2493                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2494                         c->Request.CDB[0] = cmd;        /* reset */
2495                         c->Request.CDB[1] = 0x03;       /* reset a target */
2496                         break;
2497                 case 3: /* No-Op message */
2498                         c->Request.CDBLen = 1;
2499                         c->Request.Type.Attribute = ATTR_SIMPLE;
2500                         c->Request.Type.Direction = XFER_WRITE;
2501                         c->Request.Timeout = 0;
2502                         c->Request.CDB[0] = cmd;
2503                         break;
2504                 default:
2505                         printk(KERN_WARNING
2506                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2507                         return IO_ERROR;
2508                 }
2509         } else {
2510                 printk(KERN_WARNING
2511                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2512                 return IO_ERROR;
2513         }
2514         /* Fill in the scatter gather information */
2515         if (size > 0) {
2516                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2517                                                              buff, size,
2518                                                              PCI_DMA_BIDIRECTIONAL);
2519                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2520                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2521                 c->SG[0].Len = size;
2522                 c->SG[0].Ext = 0;       /* we are not chaining */
2523         }
2524         return status;
2525 }
2526
2527 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2528 {
2529         switch (c->err_info->ScsiStatus) {
2530         case SAM_STAT_GOOD:
2531                 return IO_OK;
2532         case SAM_STAT_CHECK_CONDITION:
2533                 switch (0xf & c->err_info->SenseInfo[2]) {
2534                 case 0: return IO_OK; /* no sense */
2535                 case 1: return IO_OK; /* recovered error */
2536                 default:
2537                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2538                                 "check condition, sense key = 0x%02x\n",
2539                                 h->ctlr, c->Request.CDB[0],
2540                                 c->err_info->SenseInfo[2]);
2541                 }
2542                 break;
2543         default:
2544                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2545                         "scsi status = 0x%02x\n", h->ctlr,
2546                         c->Request.CDB[0], c->err_info->ScsiStatus);
2547                 break;
2548         }
2549         return IO_ERROR;
2550 }
2551
2552 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2553 {
2554         int return_status = IO_OK;
2555
2556         if (c->err_info->CommandStatus == CMD_SUCCESS)
2557                 return IO_OK;
2558
2559         switch (c->err_info->CommandStatus) {
2560         case CMD_TARGET_STATUS:
2561                 return_status = check_target_status(h, c);
2562                 break;
2563         case CMD_DATA_UNDERRUN:
2564         case CMD_DATA_OVERRUN:
2565                 /* expected for inquiry and report lun commands */
2566                 break;
2567         case CMD_INVALID:
2568                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2569                        "reported invalid\n", c->Request.CDB[0]);
2570                 return_status = IO_ERROR;
2571                 break;
2572         case CMD_PROTOCOL_ERR:
2573                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2574                        "protocol error \n", c->Request.CDB[0]);
2575                 return_status = IO_ERROR;
2576                 break;
2577         case CMD_HARDWARE_ERR:
2578                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2579                        " hardware error\n", c->Request.CDB[0]);
2580                 return_status = IO_ERROR;
2581                 break;
2582         case CMD_CONNECTION_LOST:
2583                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2584                        "connection lost\n", c->Request.CDB[0]);
2585                 return_status = IO_ERROR;
2586                 break;
2587         case CMD_ABORTED:
2588                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2589                        "aborted\n", c->Request.CDB[0]);
2590                 return_status = IO_ERROR;
2591                 break;
2592         case CMD_ABORT_FAILED:
2593                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2594                        "abort failed\n", c->Request.CDB[0]);
2595                 return_status = IO_ERROR;
2596                 break;
2597         case CMD_UNSOLICITED_ABORT:
2598                 printk(KERN_WARNING
2599                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2600                         c->Request.CDB[0]);
2601                 return_status = IO_NEEDS_RETRY;
2602                 break;
2603         default:
2604                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2605                        "unknown status %x\n", c->Request.CDB[0],
2606                        c->err_info->CommandStatus);
2607                 return_status = IO_ERROR;
2608         }
2609         return return_status;
2610 }
2611
2612 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2613         int attempt_retry)
2614 {
2615         DECLARE_COMPLETION_ONSTACK(wait);
2616         u64bit buff_dma_handle;
2617         unsigned long flags;
2618         int return_status = IO_OK;
2619
2620 resend_cmd2:
2621         c->waiting = &wait;
2622         /* Put the request on the tail of the queue and send it */
2623         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2624         addQ(&h->reqQ, c);
2625         h->Qdepth++;
2626         start_io(h);
2627         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2628
2629         wait_for_completion(&wait);
2630
2631         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2632                 goto command_done;
2633
2634         return_status = process_sendcmd_error(h, c);
2635
2636         if (return_status == IO_NEEDS_RETRY &&
2637                 c->retry_count < MAX_CMD_RETRIES) {
2638                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2639                         c->Request.CDB[0]);
2640                 c->retry_count++;
2641                 /* erase the old error information */
2642                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2643                 return_status = IO_OK;
2644                 INIT_COMPLETION(wait);
2645                 goto resend_cmd2;
2646         }
2647
2648 command_done:
2649         /* unlock the buffers from DMA */
2650         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2651         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2652         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2653                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2654         return return_status;
2655 }
2656
2657 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2658                            __u8 page_code, unsigned char scsi3addr[],
2659                         int cmd_type)
2660 {
2661         ctlr_info_t *h = hba[ctlr];
2662         CommandList_struct *c;
2663         int return_status;
2664
2665         c = cmd_alloc(h, 0);
2666         if (!c)
2667                 return -ENOMEM;
2668         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2669                 scsi3addr, cmd_type);
2670         if (return_status == IO_OK)
2671                 return_status = sendcmd_withirq_core(h, c, 1);
2672
2673         cmd_free(h, c, 0);
2674         return return_status;
2675 }
2676
2677 static void cciss_geometry_inquiry(int ctlr, int logvol,
2678                                    int withirq, sector_t total_size,
2679                                    unsigned int block_size,
2680                                    InquiryData_struct *inq_buff,
2681                                    drive_info_struct *drv)
2682 {
2683         int return_code;
2684         unsigned long t;
2685         unsigned char scsi3addr[8];
2686
2687         memset(inq_buff, 0, sizeof(InquiryData_struct));
2688         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2689         if (withirq)
2690                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
2691                                               inq_buff, sizeof(*inq_buff),
2692                                               0xC1, scsi3addr, TYPE_CMD);
2693         else
2694                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
2695                                       sizeof(*inq_buff), 0xC1, scsi3addr,
2696                                       TYPE_CMD);
2697         if (return_code == IO_OK) {
2698                 if (inq_buff->data_byte[8] == 0xFF) {
2699                         printk(KERN_WARNING
2700                                "cciss: reading geometry failed, volume "
2701                                "does not support reading geometry\n");
2702                         drv->heads = 255;
2703                         drv->sectors = 32;      // Sectors per track
2704                         drv->cylinders = total_size + 1;
2705                         drv->raid_level = RAID_UNKNOWN;
2706                 } else {
2707                         drv->heads = inq_buff->data_byte[6];
2708                         drv->sectors = inq_buff->data_byte[7];
2709                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2710                         drv->cylinders += inq_buff->data_byte[5];
2711                         drv->raid_level = inq_buff->data_byte[8];
2712                 }
2713                 drv->block_size = block_size;
2714                 drv->nr_blocks = total_size + 1;
2715                 t = drv->heads * drv->sectors;
2716                 if (t > 1) {
2717                         sector_t real_size = total_size + 1;
2718                         unsigned long rem = sector_div(real_size, t);
2719                         if (rem)
2720                                 real_size++;
2721                         drv->cylinders = real_size;
2722                 }
2723         } else {                /* Get geometry failed */
2724                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2725         }
2726 }
2727
2728 static void
2729 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
2730                     unsigned int *block_size)
2731 {
2732         ReadCapdata_struct *buf;
2733         int return_code;
2734         unsigned char scsi3addr[8];
2735
2736         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2737         if (!buf) {
2738                 printk(KERN_WARNING "cciss: out of memory\n");
2739                 return;
2740         }
2741
2742         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2743         if (withirq)
2744                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
2745                                 ctlr, buf, sizeof(ReadCapdata_struct),
2746                                         0, scsi3addr, TYPE_CMD);
2747         else
2748                 return_code = sendcmd(CCISS_READ_CAPACITY,
2749                                 ctlr, buf, sizeof(ReadCapdata_struct),
2750                                         0, scsi3addr, TYPE_CMD);
2751         if (return_code == IO_OK) {
2752                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2753                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2754         } else {                /* read capacity command failed */
2755                 printk(KERN_WARNING "cciss: read capacity failed\n");
2756                 *total_size = 0;
2757                 *block_size = BLOCK_SIZE;
2758         }
2759         kfree(buf);
2760 }
2761
2762 static void
2763 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size,                                 unsigned int *block_size)
2764 {
2765         ReadCapdata_struct_16 *buf;
2766         int return_code;
2767         unsigned char scsi3addr[8];
2768
2769         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2770         if (!buf) {
2771                 printk(KERN_WARNING "cciss: out of memory\n");
2772                 return;
2773         }
2774
2775         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2776         if (withirq) {
2777                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2778                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2779                                 0, scsi3addr, TYPE_CMD);
2780         }
2781         else {
2782                 return_code = sendcmd(CCISS_READ_CAPACITY_16,
2783                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2784                                 0, scsi3addr, TYPE_CMD);
2785         }
2786         if (return_code == IO_OK) {
2787                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2788                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2789         } else {                /* read capacity command failed */
2790                 printk(KERN_WARNING "cciss: read capacity failed\n");
2791                 *total_size = 0;
2792                 *block_size = BLOCK_SIZE;
2793         }
2794         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2795                (unsigned long long)*total_size+1, *block_size);
2796         kfree(buf);
2797 }
2798
2799 static int cciss_revalidate(struct gendisk *disk)
2800 {
2801         ctlr_info_t *h = get_host(disk);
2802         drive_info_struct *drv = get_drv(disk);
2803         int logvol;
2804         int FOUND = 0;
2805         unsigned int block_size;
2806         sector_t total_size;
2807         InquiryData_struct *inq_buff = NULL;
2808
2809         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2810                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2811                         sizeof(drv->LunID)) == 0) {
2812                         FOUND = 1;
2813                         break;
2814                 }
2815         }
2816
2817         if (!FOUND)
2818                 return 1;
2819
2820         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2821         if (inq_buff == NULL) {
2822                 printk(KERN_WARNING "cciss: out of memory\n");
2823                 return 1;
2824         }
2825         if (h->cciss_read == CCISS_READ_10) {
2826                 cciss_read_capacity(h->ctlr, logvol, 1,
2827                                         &total_size, &block_size);
2828         } else {
2829                 cciss_read_capacity_16(h->ctlr, logvol, 1,
2830                                         &total_size, &block_size);
2831         }
2832         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2833                                inq_buff, drv);
2834
2835         blk_queue_logical_block_size(drv->queue, drv->block_size);
2836         set_capacity(disk, drv->nr_blocks);
2837
2838         kfree(inq_buff);
2839         return 0;
2840 }
2841
2842 /*
2843  *   Wait polling for a command to complete.
2844  *   The memory mapped FIFO is polled for the completion.
2845  *   Used only at init time, interrupts from the HBA are disabled.
2846  */
2847 static unsigned long pollcomplete(int ctlr)
2848 {
2849         unsigned long done;
2850         int i;
2851
2852         /* Wait (up to 20 seconds) for a command to complete */
2853
2854         for (i = 20 * HZ; i > 0; i--) {
2855                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2856                 if (done == FIFO_EMPTY)
2857                         schedule_timeout_uninterruptible(1);
2858                 else
2859                         return done;
2860         }
2861         /* Invalid address to tell caller we ran out of time */
2862         return 1;
2863 }
2864
2865 /* Send command c to controller h and poll for it to complete.
2866  * Turns interrupts off on the board.  Used at driver init time
2867  * and during SCSI error recovery.
2868  */
2869 static int sendcmd_core(ctlr_info_t *h, CommandList_struct *c)
2870 {
2871         int i;
2872         unsigned long complete;
2873         int status = IO_ERROR;
2874         u64bit buff_dma_handle;
2875
2876 resend_cmd1:
2877
2878         /* Disable interrupt on the board. */
2879         h->access.set_intr_mask(h, CCISS_INTR_OFF);
2880
2881         /* Make sure there is room in the command FIFO */
2882         /* Actually it should be completely empty at this time */
2883         /* unless we are in here doing error handling for the scsi */
2884         /* tape side of the driver. */
2885         for (i = 200000; i > 0; i--) {
2886                 /* if fifo isn't full go */
2887                 if (!(h->access.fifo_full(h)))
2888                         break;
2889                 udelay(10);
2890                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2891                        " waiting!\n", h->ctlr);
2892         }
2893         h->access.submit_command(h, c); /* Send the cmd */
2894         do {
2895                 complete = pollcomplete(h->ctlr);
2896
2897 #ifdef CCISS_DEBUG
2898                 printk(KERN_DEBUG "cciss: command completed\n");
2899 #endif                          /* CCISS_DEBUG */
2900
2901                 if (complete == 1) {
2902                         printk(KERN_WARNING
2903                                "cciss cciss%d: SendCmd Timeout out, "
2904                                "No command list address returned!\n", h->ctlr);
2905                         status = IO_ERROR;
2906                         break;
2907                 }
2908
2909                 /* Make sure it's the command we're expecting. */
2910                 if ((complete & ~CISS_ERROR_BIT) != c->busaddr) {
2911                         printk(KERN_WARNING "cciss%d: Unexpected command "
2912                                 "completion.\n", h->ctlr);
2913                         continue;
2914                 }
2915
2916                 /* It is our command.  If no error, we're done. */
2917                 if (!(complete & CISS_ERROR_BIT)) {
2918                         status = IO_OK;
2919                         break;
2920                 }
2921
2922                 /* There is an error... */
2923
2924                 /* if data overrun or underun on Report command ignore it */
2925                 if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2926                      (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2927                      (c->Request.CDB[0] == CISS_INQUIRY)) &&
2928                         ((c->err_info->CommandStatus == CMD_DATA_OVERRUN) ||
2929                          (c->err_info->CommandStatus == CMD_DATA_UNDERRUN))) {
2930                         complete = c->busaddr;
2931                         status = IO_OK;
2932                         break;
2933                 }
2934
2935                 if (c->err_info->CommandStatus == CMD_UNSOLICITED_ABORT) {
2936                         printk(KERN_WARNING "cciss%d: unsolicited abort %p\n",
2937                                 h->ctlr, c);
2938                         if (c->retry_count < MAX_CMD_RETRIES) {
2939                                 printk(KERN_WARNING "cciss%d: retrying %p\n",
2940                                    h->ctlr, c);
2941                                 c->retry_count++;
2942                                 /* erase the old error information */
2943                                 memset(c->err_info, 0, sizeof(c->err_info));
2944                                 goto resend_cmd1;
2945                         }
2946                         printk(KERN_WARNING "cciss%d: retried %p too many "
2947                                 "times\n", h->ctlr, c);
2948                         status = IO_ERROR;
2949                         break;
2950                 }
2951
2952                 if (c->err_info->CommandStatus == CMD_UNABORTABLE) {
2953                         printk(KERN_WARNING "cciss%d: command could not be "
2954                                 "aborted.\n", h->ctlr);
2955                         status = IO_ERROR;
2956                         break;
2957                 }
2958
2959                 if (c->err_info->CommandStatus == CMD_TARGET_STATUS) {
2960                         status = check_target_status(h, c);
2961                         break;
2962                 }
2963
2964                 printk(KERN_WARNING "cciss%d: sendcmd error\n", h->ctlr);
2965                 printk(KERN_WARNING "cmd = 0x%02x, CommandStatus = 0x%02x\n",
2966                         c->Request.CDB[0], c->err_info->CommandStatus);
2967                 status = IO_ERROR;
2968                 break;
2969
2970         } while (1);
2971
2972         /* unlock the data buffer from DMA */
2973         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2974         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2975         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2976                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2977         return status;
2978 }
2979
2980 /*
2981  * Send a command to the controller, and wait for it to complete.
2982  * Used at init time, and during SCSI error recovery.
2983  */
2984 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
2985         __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2986 {
2987         CommandList_struct *c;
2988         int status;
2989
2990         c = cmd_alloc(hba[ctlr], 1);
2991         if (!c) {
2992                 printk(KERN_WARNING "cciss: unable to get memory");
2993                 return IO_ERROR;
2994         }
2995         status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2996                 scsi3addr, cmd_type);
2997         if (status == IO_OK)
2998                 status = sendcmd_core(hba[ctlr], c);
2999         cmd_free(hba[ctlr], c, 1);
3000         return status;
3001 }
3002
3003 /*
3004  * Map (physical) PCI mem into (virtual) kernel space
3005  */
3006 static void __iomem *remap_pci_mem(ulong base, ulong size)
3007 {
3008         ulong page_base = ((ulong) base) & PAGE_MASK;
3009         ulong page_offs = ((ulong) base) - page_base;
3010         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
3011
3012         return page_remapped ? (page_remapped + page_offs) : NULL;
3013 }
3014
3015 /*
3016  * Takes jobs of the Q and sends them to the hardware, then puts it on
3017  * the Q to wait for completion.
3018  */
3019 static void start_io(ctlr_info_t *h)
3020 {
3021         CommandList_struct *c;
3022
3023         while (!hlist_empty(&h->reqQ)) {
3024                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
3025                 /* can't do anything if fifo is full */
3026                 if ((h->access.fifo_full(h))) {
3027                         printk(KERN_WARNING "cciss: fifo full\n");
3028                         break;
3029                 }
3030
3031                 /* Get the first entry from the Request Q */
3032                 removeQ(c);
3033                 h->Qdepth--;
3034
3035                 /* Tell the controller execute command */
3036                 h->access.submit_command(h, c);
3037
3038                 /* Put job onto the completed Q */
3039                 addQ(&h->cmpQ, c);
3040         }
3041 }
3042
3043 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
3044 /* Zeros out the error record and then resends the command back */
3045 /* to the controller */
3046 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
3047 {
3048         /* erase the old error information */
3049         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
3050
3051         /* add it to software queue and then send it to the controller */
3052         addQ(&h->reqQ, c);
3053         h->Qdepth++;
3054         if (h->Qdepth > h->maxQsinceinit)
3055                 h->maxQsinceinit = h->Qdepth;
3056
3057         start_io(h);
3058 }
3059
3060 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
3061         unsigned int msg_byte, unsigned int host_byte,
3062         unsigned int driver_byte)
3063 {
3064         /* inverse of macros in scsi.h */
3065         return (scsi_status_byte & 0xff) |
3066                 ((msg_byte & 0xff) << 8) |
3067                 ((host_byte & 0xff) << 16) |
3068                 ((driver_byte & 0xff) << 24);
3069 }
3070
3071 static inline int evaluate_target_status(ctlr_info_t *h,
3072                         CommandList_struct *cmd, int *retry_cmd)
3073 {
3074         unsigned char sense_key;
3075         unsigned char status_byte, msg_byte, host_byte, driver_byte;
3076         int error_value;
3077
3078         *retry_cmd = 0;
3079         /* If we get in here, it means we got "target status", that is, scsi status */
3080         status_byte = cmd->err_info->ScsiStatus;
3081         driver_byte = DRIVER_OK;
3082         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3083
3084         if (blk_pc_request(cmd->rq))
3085                 host_byte = DID_PASSTHROUGH;
3086         else
3087                 host_byte = DID_OK;
3088
3089         error_value = make_status_bytes(status_byte, msg_byte,
3090                 host_byte, driver_byte);
3091
3092         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3093                 if (!blk_pc_request(cmd->rq))
3094                         printk(KERN_WARNING "cciss: cmd %p "
3095                                "has SCSI Status 0x%x\n",
3096                                cmd, cmd->err_info->ScsiStatus);
3097                 return error_value;
3098         }
3099
3100         /* check the sense key */
3101         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3102         /* no status or recovered error */
3103         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
3104                 error_value = 0;
3105
3106         if (check_for_unit_attention(h, cmd)) {
3107                 *retry_cmd = !blk_pc_request(cmd->rq);
3108                 return 0;
3109         }
3110
3111         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
3112                 if (error_value != 0)
3113                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3114                                " sense key = 0x%x\n", cmd, sense_key);
3115                 return error_value;
3116         }
3117
3118         /* SG_IO or similar, copy sense data back */
3119         if (cmd->rq->sense) {
3120                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3121                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3122                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3123                         cmd->rq->sense_len);
3124         } else
3125                 cmd->rq->sense_len = 0;
3126
3127         return error_value;
3128 }
3129
3130 /* checks the status of the job and calls complete buffers to mark all
3131  * buffers for the completed job. Note that this function does not need
3132  * to hold the hba/queue lock.
3133  */
3134 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3135                                     int timeout)
3136 {
3137         int retry_cmd = 0;
3138         struct request *rq = cmd->rq;
3139
3140         rq->errors = 0;
3141
3142         if (timeout)
3143                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3144
3145         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3146                 goto after_error_processing;
3147
3148         switch (cmd->err_info->CommandStatus) {
3149         case CMD_TARGET_STATUS:
3150                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3151                 break;
3152         case CMD_DATA_UNDERRUN:
3153                 if (blk_fs_request(cmd->rq)) {
3154                         printk(KERN_WARNING "cciss: cmd %p has"
3155                                " completed with data underrun "
3156                                "reported\n", cmd);
3157                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3158                 }
3159                 break;
3160         case CMD_DATA_OVERRUN:
3161                 if (blk_fs_request(cmd->rq))
3162                         printk(KERN_WARNING "cciss: cmd %p has"
3163                                " completed with data overrun "
3164                                "reported\n", cmd);
3165                 break;
3166         case CMD_INVALID:
3167                 printk(KERN_WARNING "cciss: cmd %p is "
3168                        "reported invalid\n", cmd);
3169                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3170                         cmd->err_info->CommandStatus, DRIVER_OK,
3171                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3172                 break;
3173         case CMD_PROTOCOL_ERR:
3174                 printk(KERN_WARNING "cciss: cmd %p has "
3175                        "protocol error \n", cmd);
3176                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3177                         cmd->err_info->CommandStatus, DRIVER_OK,
3178                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3179                 break;
3180         case CMD_HARDWARE_ERR:
3181                 printk(KERN_WARNING "cciss: cmd %p had "
3182                        " hardware error\n", cmd);
3183                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3184                         cmd->err_info->CommandStatus, DRIVER_OK,
3185                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3186                 break;
3187         case CMD_CONNECTION_LOST:
3188                 printk(KERN_WARNING "cciss: cmd %p had "
3189                        "connection lost\n", cmd);
3190                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3191                         cmd->err_info->CommandStatus, DRIVER_OK,
3192                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3193                 break;
3194         case CMD_ABORTED:
3195                 printk(KERN_WARNING "cciss: cmd %p was "
3196                        "aborted\n", cmd);
3197                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3198                         cmd->err_info->CommandStatus, DRIVER_OK,
3199                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3200                 break;
3201         case CMD_ABORT_FAILED:
3202                 printk(KERN_WARNING "cciss: cmd %p reports "
3203                        "abort failed\n", cmd);
3204                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3205                         cmd->err_info->CommandStatus, DRIVER_OK,
3206                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3207                 break;
3208         case CMD_UNSOLICITED_ABORT:
3209                 printk(KERN_WARNING "cciss%d: unsolicited "
3210                        "abort %p\n", h->ctlr, cmd);
3211                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3212                         retry_cmd = 1;
3213                         printk(KERN_WARNING
3214                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3215                         cmd->retry_count++;
3216                 } else
3217                         printk(KERN_WARNING
3218                                "cciss%d: %p retried too "
3219                                "many times\n", h->ctlr, cmd);
3220                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3221                         cmd->err_info->CommandStatus, DRIVER_OK,
3222                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3223                 break;
3224         case CMD_TIMEOUT:
3225                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3226                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3227                         cmd->err_info->CommandStatus, DRIVER_OK,
3228                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3229                 break;
3230         default:
3231                 printk(KERN_WARNING "cciss: cmd %p returned "
3232                        "unknown status %x\n", cmd,
3233                        cmd->err_info->CommandStatus);
3234                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3235                         cmd->err_info->CommandStatus, DRIVER_OK,
3236                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3237         }
3238
3239 after_error_processing:
3240
3241         /* We need to return this command */
3242         if (retry_cmd) {
3243                 resend_cciss_cmd(h, cmd);
3244                 return;
3245         }
3246         cmd->rq->completion_data = cmd;
3247         blk_complete_request(cmd->rq);
3248 }
3249
3250 /*
3251  * Get a request and submit it to the controller.
3252  */
3253 static void do_cciss_request(struct request_queue *q)
3254 {
3255         ctlr_info_t *h = q->queuedata;
3256         CommandList_struct *c;
3257         sector_t start_blk;
3258         int seg;
3259         struct request *creq;
3260         u64bit temp64;
3261         struct scatterlist tmp_sg[MAXSGENTRIES];
3262         drive_info_struct *drv;
3263         int i, dir;
3264
3265         /* We call start_io here in case there is a command waiting on the
3266          * queue that has not been sent.
3267          */
3268         if (blk_queue_plugged(q))
3269                 goto startio;
3270
3271       queue:
3272         creq = blk_peek_request(q);
3273         if (!creq)
3274                 goto startio;
3275
3276         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
3277
3278         if ((c = cmd_alloc(h, 1)) == NULL)
3279                 goto full;
3280
3281         blk_start_request(creq);
3282
3283         spin_unlock_irq(q->queue_lock);
3284
3285         c->cmd_type = CMD_RWREQ;
3286         c->rq = creq;
3287
3288         /* fill in the request */
3289         drv = creq->rq_disk->private_data;
3290         c->Header.ReplyQueue = 0;       // unused in simple mode
3291         /* got command from pool, so use the command block index instead */
3292         /* for direct lookups. */
3293         /* The first 2 bits are reserved for controller error reporting. */
3294         c->Header.Tag.lower = (c->cmdindex << 3);
3295         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
3296         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3297         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
3298         c->Request.Type.Type = TYPE_CMD;        // It is a command.
3299         c->Request.Type.Attribute = ATTR_SIMPLE;
3300         c->Request.Type.Direction =
3301             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3302         c->Request.Timeout = 0; // Don't time out
3303         c->Request.CDB[0] =
3304             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3305         start_blk = blk_rq_pos(creq);
3306 #ifdef CCISS_DEBUG
3307         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3308                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3309 #endif                          /* CCISS_DEBUG */
3310
3311         sg_init_table(tmp_sg, MAXSGENTRIES);
3312         seg = blk_rq_map_sg(q, creq, tmp_sg);
3313
3314         /* get the DMA records for the setup */
3315         if (c->Request.Type.Direction == XFER_READ)
3316                 dir = PCI_DMA_FROMDEVICE;
3317         else
3318                 dir = PCI_DMA_TODEVICE;
3319
3320         for (i = 0; i < seg; i++) {
3321                 c->SG[i].Len = tmp_sg[i].length;
3322                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3323                                                   tmp_sg[i].offset,
3324                                                   tmp_sg[i].length, dir);
3325                 c->SG[i].Addr.lower = temp64.val32.lower;
3326                 c->SG[i].Addr.upper = temp64.val32.upper;
3327                 c->SG[i].Ext = 0;       // we are not chaining
3328         }
3329         /* track how many SG entries we are using */
3330         if (seg > h->maxSG)
3331                 h->maxSG = seg;
3332
3333 #ifdef CCISS_DEBUG
3334         printk(KERN_DEBUG "cciss: Submitting %u sectors in %d segments\n",
3335                blk_rq_sectors(creq), seg);
3336 #endif                          /* CCISS_DEBUG */
3337
3338         c->Header.SGList = c->Header.SGTotal = seg;
3339         if (likely(blk_fs_request(creq))) {
3340                 if(h->cciss_read == CCISS_READ_10) {
3341                         c->Request.CDB[1] = 0;
3342                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
3343                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3344                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3345                         c->Request.CDB[5] = start_blk & 0xff;
3346                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
3347                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3348                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3349                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3350                 } else {
3351                         u32 upper32 = upper_32_bits(start_blk);
3352
3353                         c->Request.CDBLen = 16;
3354                         c->Request.CDB[1]= 0;
3355                         c->Request.CDB[2]= (upper32 >> 24) & 0xff;      //MSB
3356                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3357                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3358                         c->Request.CDB[5]= upper32 & 0xff;
3359                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3360                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3361                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3362                         c->Request.CDB[9]= start_blk & 0xff;
3363                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3364                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3365                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3366                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3367                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3368                 }
3369         } else if (blk_pc_request(creq)) {
3370                 c->Request.CDBLen = creq->cmd_len;
3371                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3372         } else {
3373                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3374                 BUG();
3375         }
3376
3377         spin_lock_irq(q->queue_lock);
3378
3379         addQ(&h->reqQ, c);
3380         h->Qdepth++;
3381         if (h->Qdepth > h->maxQsinceinit)
3382                 h->maxQsinceinit = h->Qdepth;
3383
3384         goto queue;
3385 full:
3386         blk_stop_queue(q);
3387 startio:
3388         /* We will already have the driver lock here so not need
3389          * to lock it.
3390          */
3391         start_io(h);
3392 }
3393
3394 static inline unsigned long get_next_completion(ctlr_info_t *h)
3395 {
3396         return h->access.command_completed(h);
3397 }
3398
3399 static inline int interrupt_pending(ctlr_info_t *h)
3400 {
3401         return h->access.intr_pending(h);
3402 }
3403
3404 static inline long interrupt_not_for_us(ctlr_info_t *h)
3405 {
3406         return (((h->access.intr_pending(h) == 0) ||
3407                  (h->interrupts_enabled == 0)));
3408 }
3409
3410 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3411 {
3412         ctlr_info_t *h = dev_id;
3413         CommandList_struct *c;
3414         unsigned long flags;
3415         __u32 a, a1, a2;
3416
3417         if (interrupt_not_for_us(h))
3418                 return IRQ_NONE;
3419         /*
3420          * If there are completed commands in the completion queue,
3421          * we had better do something about it.
3422          */
3423         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3424         while (interrupt_pending(h)) {
3425                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3426                         a1 = a;
3427                         if ((a & 0x04)) {
3428                                 a2 = (a >> 3);
3429                                 if (a2 >= h->nr_cmds) {
3430                                         printk(KERN_WARNING
3431                                                "cciss: controller cciss%d failed, stopping.\n",
3432                                                h->ctlr);
3433                                         fail_all_cmds(h->ctlr);
3434                                         return IRQ_HANDLED;
3435                                 }
3436
3437                                 c = h->cmd_pool + a2;
3438                                 a = c->busaddr;
3439
3440                         } else {
3441                                 struct hlist_node *tmp;
3442
3443                                 a &= ~3;
3444                                 c = NULL;
3445                                 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3446                                         if (c->busaddr == a)
3447                                                 break;
3448                                 }
3449                         }
3450                         /*
3451                          * If we've found the command, take it off the
3452                          * completion Q and free it
3453                          */
3454                         if (c && c->busaddr == a) {
3455                                 removeQ(c);
3456                                 if (c->cmd_type == CMD_RWREQ) {
3457                                         complete_command(h, c, 0);
3458                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3459                                         complete(c->waiting);
3460                                 }
3461 #                               ifdef CONFIG_CISS_SCSI_TAPE
3462                                 else if (c->cmd_type == CMD_SCSI)
3463                                         complete_scsi_command(c, 0, a1);
3464 #                               endif
3465                                 continue;
3466                         }
3467                 }
3468         }
3469
3470         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3471         return IRQ_HANDLED;
3472 }
3473
3474 /**
3475  * add_to_scan_list() - add controller to rescan queue
3476  * @h:                Pointer to the controller.
3477  *
3478  * Adds the controller to the rescan queue if not already on the queue.
3479  *
3480  * returns 1 if added to the queue, 0 if skipped (could be on the
3481  * queue already, or the controller could be initializing or shutting
3482  * down).
3483  **/
3484 static int add_to_scan_list(struct ctlr_info *h)
3485 {
3486         struct ctlr_info *test_h;
3487         int found = 0;
3488         int ret = 0;
3489
3490         if (h->busy_initializing)
3491                 return 0;
3492
3493         if (!mutex_trylock(&h->busy_shutting_down))
3494                 return 0;
3495
3496         mutex_lock(&scan_mutex);
3497         list_for_each_entry(test_h, &scan_q, scan_list) {
3498                 if (test_h == h) {
3499                         found = 1;
3500                         break;
3501                 }
3502         }
3503         if (!found && !h->busy_scanning) {
3504                 INIT_COMPLETION(h->scan_wait);
3505                 list_add_tail(&h->scan_list, &scan_q);
3506                 ret = 1;
3507         }
3508         mutex_unlock(&scan_mutex);
3509         mutex_unlock(&h->busy_shutting_down);
3510
3511         return ret;
3512 }
3513
3514 /**
3515  * remove_from_scan_list() - remove controller from rescan queue
3516  * @h:                     Pointer to the controller.
3517  *
3518  * Removes the controller from the rescan queue if present. Blocks if
3519  * the controller is currently conducting a rescan.
3520  **/
3521 static void remove_from_scan_list(struct ctlr_info *h)
3522 {
3523         struct ctlr_info *test_h, *tmp_h;
3524         int scanning = 0;
3525
3526         mutex_lock(&scan_mutex);
3527         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3528                 if (test_h == h) {
3529                         list_del(&h->scan_list);
3530                         complete_all(&h->scan_wait);
3531                         mutex_unlock(&scan_mutex);
3532                         return;
3533                 }
3534         }
3535         if (&h->busy_scanning)
3536                 scanning = 0;
3537         mutex_unlock(&scan_mutex);
3538
3539         if (scanning)
3540                 wait_for_completion(&h->scan_wait);
3541 }
3542
3543 /**
3544  * scan_thread() - kernel thread used to rescan controllers
3545  * @data:        Ignored.
3546  *
3547  * A kernel thread used scan for drive topology changes on
3548  * controllers. The thread processes only one controller at a time
3549  * using a queue.  Controllers are added to the queue using
3550  * add_to_scan_list() and removed from the queue either after done
3551  * processing or using remove_from_scan_list().
3552  *
3553  * returns 0.
3554  **/
3555 static int scan_thread(void *data)
3556 {
3557         struct ctlr_info *h;
3558
3559         while (1) {
3560                 set_current_state(TASK_INTERRUPTIBLE);
3561                 schedule();
3562                 if (kthread_should_stop())
3563                         break;
3564
3565                 while (1) {
3566                         mutex_lock(&scan_mutex);
3567                         if (list_empty(&scan_q)) {
3568                                 mutex_unlock(&scan_mutex);
3569                                 break;
3570                         }
3571
3572                         h = list_entry(scan_q.next,
3573                                        struct ctlr_info,
3574                                        scan_list);
3575                         list_del(&h->scan_list);
3576                         h->busy_scanning = 1;
3577                         mutex_unlock(&scan_mutex);
3578
3579                         if (h) {
3580                                 rebuild_lun_table(h, 0, 0);
3581                                 complete_all(&h->scan_wait);
3582                                 mutex_lock(&scan_mutex);
3583                                 h->busy_scanning = 0;
3584                                 mutex_unlock(&scan_mutex);
3585                         }
3586                 }
3587         }
3588
3589         return 0;
3590 }
3591
3592 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3593 {
3594         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3595                 return 0;
3596
3597         switch (c->err_info->SenseInfo[12]) {
3598         case STATE_CHANGED:
3599                 printk(KERN_WARNING "cciss%d: a state change "
3600                         "detected, command retried\n", h->ctlr);
3601                 return 1;
3602         break;
3603         case LUN_FAILED:
3604                 printk(KERN_WARNING "cciss%d: LUN failure "
3605                         "detected, action required\n", h->ctlr);
3606                 return 1;
3607         break;
3608         case REPORT_LUNS_CHANGED:
3609                 printk(KERN_WARNING "cciss%d: report LUN data "
3610                         "changed\n", h->ctlr);
3611                 add_to_scan_list(h);
3612                 wake_up_process(cciss_scan_thread);
3613                 return 1;
3614         break;
3615         case POWER_OR_RESET:
3616                 printk(KERN_WARNING "cciss%d: a power on "
3617                         "or device reset detected\n", h->ctlr);
3618                 return 1;
3619         break;
3620         case UNIT_ATTENTION_CLEARED:
3621                 printk(KERN_WARNING "cciss%d: unit attention "
3622                     "cleared by another initiator\n", h->ctlr);
3623                 return 1;
3624         break;
3625         default:
3626                 printk(KERN_WARNING "cciss%d: unknown "
3627                         "unit attention detected\n", h->ctlr);
3628                                 return 1;
3629         }
3630 }
3631
3632 /*
3633  *  We cannot read the structure directly, for portability we must use
3634  *   the io functions.
3635  *   This is for debug only.
3636  */
3637 #ifdef CCISS_DEBUG
3638 static void print_cfg_table(CfgTable_struct *tb)
3639 {
3640         int i;
3641         char temp_name[17];
3642
3643         printk("Controller Configuration information\n");
3644         printk("------------------------------------\n");
3645         for (i = 0; i < 4; i++)
3646                 temp_name[i] = readb(&(tb->Signature[i]));
3647         temp_name[4] = '\0';
3648         printk("   Signature = %s\n", temp_name);
3649         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3650         printk("   Transport methods supported = 0x%x\n",
3651                readl(&(tb->TransportSupport)));
3652         printk("   Transport methods active = 0x%x\n",
3653                readl(&(tb->TransportActive)));
3654         printk("   Requested transport Method = 0x%x\n",
3655                readl(&(tb->HostWrite.TransportRequest)));
3656         printk("   Coalesce Interrupt Delay = 0x%x\n",
3657                readl(&(tb->HostWrite.CoalIntDelay)));
3658         printk("   Coalesce Interrupt Count = 0x%x\n",
3659                readl(&(tb->HostWrite.CoalIntCount)));
3660         printk("   Max outstanding commands = 0x%d\n",
3661                readl(&(tb->CmdsOutMax)));
3662         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3663         for (i = 0; i < 16; i++)
3664                 temp_name[i] = readb(&(tb->ServerName[i]));
3665         temp_name[16] = '\0';
3666         printk("   Server Name = %s\n", temp_name);
3667         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3668 }
3669 #endif                          /* CCISS_DEBUG */
3670
3671 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3672 {
3673         int i, offset, mem_type, bar_type;
3674         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3675                 return 0;
3676         offset = 0;
3677         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3678                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3679                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3680                         offset += 4;
3681                 else {
3682                         mem_type = pci_resource_flags(pdev, i) &
3683                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3684                         switch (mem_type) {
3685                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3686                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3687                                 offset += 4;    /* 32 bit */
3688                                 break;
3689                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3690                                 offset += 8;
3691                                 break;
3692                         default:        /* reserved in PCI 2.2 */
3693                                 printk(KERN_WARNING
3694                                        "Base address is invalid\n");
3695                                 return -1;
3696                                 break;
3697                         }
3698                 }
3699                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3700                         return i + 1;
3701         }
3702         return -1;
3703 }
3704
3705 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3706  * controllers that are capable. If not, we use IO-APIC mode.
3707  */
3708
3709 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3710                                            struct pci_dev *pdev, __u32 board_id)
3711 {
3712 #ifdef CONFIG_PCI_MSI
3713         int err;
3714         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3715         {0, 2}, {0, 3}
3716         };
3717
3718         /* Some boards advertise MSI but don't really support it */
3719         if ((board_id == 0x40700E11) ||
3720             (board_id == 0x40800E11) ||
3721             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3722                 goto default_int_mode;
3723
3724         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3725                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3726                 if (!err) {
3727                         c->intr[0] = cciss_msix_entries[0].vector;
3728                         c->intr[1] = cciss_msix_entries[1].vector;
3729                         c->intr[2] = cciss_msix_entries[2].vector;
3730                         c->intr[3] = cciss_msix_entries[3].vector;
3731                         c->msix_vector = 1;
3732                         return;
3733                 }
3734                 if (err > 0) {
3735                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3736                                "available\n", err);
3737                         goto default_int_mode;
3738                 } else {
3739                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3740                                err);
3741                         goto default_int_mode;
3742                 }
3743         }
3744         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3745                 if (!pci_enable_msi(pdev)) {
3746                         c->msi_vector = 1;
3747                 } else {
3748                         printk(KERN_WARNING "cciss: MSI init failed\n");
3749                 }
3750         }
3751 default_int_mode:
3752 #endif                          /* CONFIG_PCI_MSI */
3753         /* if we get here we're going to use the default interrupt mode */
3754         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3755         return;
3756 }
3757
3758 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3759 {
3760         ushort subsystem_vendor_id, subsystem_device_id, command;
3761         __u32 board_id, scratchpad = 0;
3762         __u64 cfg_offset;
3763         __u32 cfg_base_addr;
3764         __u64 cfg_base_addr_index;
3765         int i, prod_index, err;
3766
3767         subsystem_vendor_id = pdev->subsystem_vendor;
3768         subsystem_device_id = pdev->subsystem_device;
3769         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3770                     subsystem_vendor_id);
3771
3772         for (i = 0; i < ARRAY_SIZE(products); i++) {
3773                 /* Stand aside for hpsa driver on request */
3774                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3775                         return -ENODEV;
3776                 if (board_id == products[i].board_id)
3777                         break;
3778         }
3779         prod_index = i;
3780         if (prod_index == ARRAY_SIZE(products)) {
3781                 dev_warn(&pdev->dev,
3782                         "unrecognized board ID: 0x%08lx, ignoring.\n",
3783                         (unsigned long) board_id);
3784                 return -ENODEV;
3785         }
3786
3787         /* check to see if controller has been disabled */
3788         /* BEFORE trying to enable it */
3789         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3790         if (!(command & 0x02)) {
3791                 printk(KERN_WARNING
3792                        "cciss: controller appears to be disabled\n");
3793                 return -ENODEV;
3794         }
3795
3796         err = pci_enable_device(pdev);
3797         if (err) {
3798                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3799                 return err;
3800         }
3801
3802         err = pci_request_regions(pdev, "cciss");
3803         if (err) {
3804                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3805                        "aborting\n");
3806                 return err;
3807         }
3808
3809 #ifdef CCISS_DEBUG
3810         printk("command = %x\n", command);
3811         printk("irq = %x\n", pdev->irq);
3812         printk("board_id = %x\n", board_id);
3813 #endif                          /* CCISS_DEBUG */
3814
3815 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3816  * else we use the IO-APIC interrupt assigned to us by system ROM.
3817  */
3818         cciss_interrupt_mode(c, pdev, board_id);
3819
3820         /* find the memory BAR */
3821         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3822                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3823                         break;
3824         }
3825         if (i == DEVICE_COUNT_RESOURCE) {
3826                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3827                 err = -ENODEV;
3828                 goto err_out_free_res;
3829         }
3830
3831         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3832                                                  * already removed
3833                                                  */
3834
3835 #ifdef CCISS_DEBUG
3836         printk("address 0 = %lx\n", c->paddr);
3837 #endif                          /* CCISS_DEBUG */
3838         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3839
3840         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3841          * We poll for up to 120 secs, once per 100ms. */
3842         for (i = 0; i < 1200; i++) {
3843                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3844                 if (scratchpad == CCISS_FIRMWARE_READY)
3845                         break;
3846                 set_current_state(TASK_INTERRUPTIBLE);
3847                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
3848         }
3849         if (scratchpad != CCISS_FIRMWARE_READY) {
3850                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3851                 err = -ENODEV;
3852                 goto err_out_free_res;
3853         }
3854
3855         /* get the address index number */
3856         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3857         cfg_base_addr &= (__u32) 0x0000ffff;
3858 #ifdef CCISS_DEBUG
3859         printk("cfg base address = %x\n", cfg_base_addr);
3860 #endif                          /* CCISS_DEBUG */
3861         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3862 #ifdef CCISS_DEBUG
3863         printk("cfg base address index = %llx\n",
3864                 (unsigned long long)cfg_base_addr_index);
3865 #endif                          /* CCISS_DEBUG */
3866         if (cfg_base_addr_index == -1) {
3867                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3868                 err = -ENODEV;
3869                 goto err_out_free_res;
3870         }
3871
3872         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3873 #ifdef CCISS_DEBUG
3874         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3875 #endif                          /* CCISS_DEBUG */
3876         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3877                                                        cfg_base_addr_index) +
3878                                     cfg_offset, sizeof(CfgTable_struct));
3879         c->board_id = board_id;
3880
3881 #ifdef CCISS_DEBUG
3882         print_cfg_table(c->cfgtable);
3883 #endif                          /* CCISS_DEBUG */
3884
3885         /* Some controllers support Zero Memory Raid (ZMR).
3886          * When configured in ZMR mode the number of supported
3887          * commands drops to 64. So instead of just setting an
3888          * arbitrary value we make the driver a little smarter.
3889          * We read the config table to tell us how many commands
3890          * are supported on the controller then subtract 4 to
3891          * leave a little room for ioctl calls.
3892          */
3893         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3894         c->product_name = products[prod_index].product_name;
3895         c->access = *(products[prod_index].access);
3896         c->nr_cmds = c->max_commands - 4;
3897         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3898             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3899             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3900             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3901                 printk("Does not appear to be a valid CISS config table\n");
3902                 err = -ENODEV;
3903                 goto err_out_free_res;
3904         }
3905 #ifdef CONFIG_X86
3906         {
3907                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3908                 __u32 prefetch;
3909                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3910                 prefetch |= 0x100;
3911                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3912         }
3913 #endif
3914
3915         /* Disabling DMA prefetch and refetch for the P600.
3916          * An ASIC bug may result in accesses to invalid memory addresses.
3917          * We've disabled prefetch for some time now. Testing with XEN
3918          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3919          */
3920         if(board_id == 0x3225103C) {
3921                 __u32 dma_prefetch;
3922                 __u32 dma_refetch;
3923                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3924                 dma_prefetch |= 0x8000;
3925                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3926                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3927                 dma_refetch |= 0x1;
3928                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3929         }
3930
3931 #ifdef CCISS_DEBUG
3932         printk("Trying to put board into Simple mode\n");
3933 #endif                          /* CCISS_DEBUG */
3934         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3935         /* Update the field, and then ring the doorbell */
3936         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3937         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3938
3939         /* under certain very rare conditions, this can take awhile.
3940          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3941          * as we enter this code.) */
3942         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3943                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3944                         break;
3945                 /* delay and try again */
3946                 set_current_state(TASK_INTERRUPTIBLE);
3947                 schedule_timeout(msecs_to_jiffies(1));
3948         }
3949
3950 #ifdef CCISS_DEBUG
3951         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3952                readl(c->vaddr + SA5_DOORBELL));
3953 #endif                          /* CCISS_DEBUG */
3954 #ifdef CCISS_DEBUG
3955         print_cfg_table(c->cfgtable);
3956 #endif                          /* CCISS_DEBUG */
3957
3958         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3959                 printk(KERN_WARNING "cciss: unable to get board into"
3960                        " simple mode\n");
3961                 err = -ENODEV;
3962                 goto err_out_free_res;
3963         }
3964         return 0;
3965
3966 err_out_free_res:
3967         /*
3968          * Deliberately omit pci_disable_device(): it does something nasty to
3969          * Smart Array controllers that pci_enable_device does not undo
3970          */
3971         pci_release_regions(pdev);
3972         return err;
3973 }
3974
3975 /* Function to find the first free pointer into our hba[] array
3976  * Returns -1 if no free entries are left.
3977  */
3978 static int alloc_cciss_hba(void)
3979 {
3980         int i;
3981
3982         for (i = 0; i < MAX_CTLR; i++) {
3983                 if (!hba[i]) {
3984                         ctlr_info_t *p;
3985
3986                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3987                         if (!p)
3988                                 goto Enomem;
3989                         hba[i] = p;
3990                         return i;
3991                 }
3992         }
3993         printk(KERN_WARNING "cciss: This driver supports a maximum"
3994                " of %d controllers.\n", MAX_CTLR);
3995         return -1;
3996 Enomem:
3997         printk(KERN_ERR "cciss: out of memory.\n");
3998         return -1;
3999 }
4000
4001 static void free_hba(int n)
4002 {
4003         ctlr_info_t *h = hba[n];
4004         int i;
4005
4006         hba[n] = NULL;
4007         for (i = 0; i < h->highest_lun + 1; i++)
4008                 if (h->gendisk[i] != NULL)
4009                         put_disk(h->gendisk[i]);
4010         kfree(h);
4011 }
4012
4013 /* Send a message CDB to the firmware. */
4014 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4015 {
4016         typedef struct {
4017                 CommandListHeader_struct CommandHeader;
4018                 RequestBlock_struct Request;
4019                 ErrDescriptor_struct ErrorDescriptor;
4020         } Command;
4021         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4022         Command *cmd;
4023         dma_addr_t paddr64;
4024         uint32_t paddr32, tag;
4025         void __iomem *vaddr;
4026         int i, err;
4027
4028         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4029         if (vaddr == NULL)
4030                 return -ENOMEM;
4031
4032         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4033            CCISS commands, so they must be allocated from the lower 4GiB of
4034            memory. */
4035         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4036         if (err) {
4037                 iounmap(vaddr);
4038                 return -ENOMEM;
4039         }
4040
4041         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4042         if (cmd == NULL) {
4043                 iounmap(vaddr);
4044                 return -ENOMEM;
4045         }
4046
4047         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4048            although there's no guarantee, we assume that the address is at
4049            least 4-byte aligned (most likely, it's page-aligned). */
4050         paddr32 = paddr64;
4051
4052         cmd->CommandHeader.ReplyQueue = 0;
4053         cmd->CommandHeader.SGList = 0;
4054         cmd->CommandHeader.SGTotal = 0;
4055         cmd->CommandHeader.Tag.lower = paddr32;
4056         cmd->CommandHeader.Tag.upper = 0;
4057         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4058
4059         cmd->Request.CDBLen = 16;
4060         cmd->Request.Type.Type = TYPE_MSG;
4061         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4062         cmd->Request.Type.Direction = XFER_NONE;
4063         cmd->Request.Timeout = 0; /* Don't time out */
4064         cmd->Request.CDB[0] = opcode;
4065         cmd->Request.CDB[1] = type;
4066         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4067
4068         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4069         cmd->ErrorDescriptor.Addr.upper = 0;
4070         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4071
4072         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4073
4074         for (i = 0; i < 10; i++) {
4075                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4076                 if ((tag & ~3) == paddr32)
4077                         break;
4078                 schedule_timeout_uninterruptible(HZ);
4079         }
4080
4081         iounmap(vaddr);
4082
4083         /* we leak the DMA buffer here ... no choice since the controller could
4084            still complete the command. */
4085         if (i == 10) {
4086                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4087                         opcode, type);
4088                 return -ETIMEDOUT;
4089         }
4090
4091         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4092
4093         if (tag & 2) {
4094                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4095                         opcode, type);
4096                 return -EIO;
4097         }
4098
4099         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4100                 opcode, type);
4101         return 0;
4102 }
4103
4104 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4105 #define cciss_noop(p) cciss_message(p, 3, 0)
4106
4107 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4108 {
4109 /* the #defines are stolen from drivers/pci/msi.h. */
4110 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4111 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4112
4113         int pos;
4114         u16 control = 0;
4115
4116         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4117         if (pos) {
4118                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4119                 if (control & PCI_MSI_FLAGS_ENABLE) {
4120                         printk(KERN_INFO "cciss: resetting MSI\n");
4121                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4122                 }
4123         }
4124
4125         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4126         if (pos) {
4127                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4128                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4129                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4130                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4131                 }
4132         }
4133
4134         return 0;
4135 }
4136
4137 /* This does a hard reset of the controller using PCI power management
4138  * states. */
4139 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4140 {
4141         u16 pmcsr, saved_config_space[32];
4142         int i, pos;
4143
4144         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4145
4146         /* This is very nearly the same thing as
4147
4148            pci_save_state(pci_dev);
4149            pci_set_power_state(pci_dev, PCI_D3hot);
4150            pci_set_power_state(pci_dev, PCI_D0);
4151            pci_restore_state(pci_dev);
4152
4153            but we can't use these nice canned kernel routines on
4154            kexec, because they also check the MSI/MSI-X state in PCI
4155            configuration space and do the wrong thing when it is
4156            set/cleared.  Also, the pci_save/restore_state functions
4157            violate the ordering requirements for restoring the
4158            configuration space from the CCISS document (see the
4159            comment below).  So we roll our own .... */
4160
4161         for (i = 0; i < 32; i++)
4162                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4163
4164         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4165         if (pos == 0) {
4166                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4167                 return -ENODEV;
4168         }
4169
4170         /* Quoting from the Open CISS Specification: "The Power
4171          * Management Control/Status Register (CSR) controls the power
4172          * state of the device.  The normal operating state is D0,
4173          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4174          * the controller, place the interface device in D3 then to
4175          * D0, this causes a secondary PCI reset which will reset the
4176          * controller." */
4177
4178         /* enter the D3hot power management state */
4179         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4180         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4181         pmcsr |= PCI_D3hot;
4182         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4183
4184         schedule_timeout_uninterruptible(HZ >> 1);
4185
4186         /* enter the D0 power management state */
4187         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4188         pmcsr |= PCI_D0;
4189         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4190
4191         schedule_timeout_uninterruptible(HZ >> 1);
4192
4193         /* Restore the PCI configuration space.  The Open CISS
4194          * Specification says, "Restore the PCI Configuration
4195          * Registers, offsets 00h through 60h. It is important to
4196          * restore the command register, 16-bits at offset 04h,
4197          * last. Do not restore the configuration status register,
4198          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4199         for (i = 0; i < 32; i++) {
4200                 if (i == 2 || i == 3)
4201                         continue;
4202                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4203         }
4204         wmb();
4205         pci_write_config_word(pdev, 4, saved_config_space[2]);
4206
4207         return 0;
4208 }
4209
4210 /*
4211  *  This is it.  Find all the controllers and register them.  I really hate
4212  *  stealing all these major device numbers.
4213  *  returns the number of block devices registered.
4214  */
4215 static int __devinit cciss_init_one(struct pci_dev *pdev,
4216                                     const struct pci_device_id *ent)
4217 {
4218         int i;
4219         int j = 0;
4220         int rc;
4221         int dac, return_code;
4222         InquiryData_struct *inq_buff;
4223
4224         if (reset_devices) {
4225                 /* Reset the controller with a PCI power-cycle */
4226                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4227                         return -ENODEV;
4228
4229                 /* Now try to get the controller to respond to a no-op. Some
4230                    devices (notably the HP Smart Array 5i Controller) need
4231                    up to 30 seconds to respond. */
4232                 for (i=0; i<30; i++) {
4233                         if (cciss_noop(pdev) == 0)
4234                                 break;
4235
4236                         schedule_timeout_uninterruptible(HZ);
4237                 }
4238                 if (i == 30) {
4239                         printk(KERN_ERR "cciss: controller seems dead\n");
4240                         return -EBUSY;
4241                 }
4242         }
4243
4244         i = alloc_cciss_hba();
4245         if (i < 0)
4246                 return -1;
4247
4248         hba[i]->busy_initializing = 1;
4249         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4250         INIT_HLIST_HEAD(&hba[i]->reqQ);
4251         mutex_init(&hba[i]->busy_shutting_down);
4252
4253         if (cciss_pci_init(hba[i], pdev) != 0)
4254                 goto clean_no_release_regions;
4255
4256         sprintf(hba[i]->devname, "cciss%d", i);
4257         hba[i]->ctlr = i;
4258         hba[i]->pdev = pdev;
4259
4260         init_completion(&hba[i]->scan_wait);
4261
4262         if (cciss_create_hba_sysfs_entry(hba[i]))
4263                 goto clean0;
4264
4265         /* configure PCI DMA stuff */
4266         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4267                 dac = 1;
4268         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4269                 dac = 0;
4270         else {
4271                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4272                 goto clean1;
4273         }
4274
4275         /*
4276          * register with the major number, or get a dynamic major number
4277          * by passing 0 as argument.  This is done for greater than
4278          * 8 controller support.
4279          */
4280         if (i < MAX_CTLR_ORIG)
4281                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4282         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4283         if (rc == -EBUSY || rc == -EINVAL) {
4284                 printk(KERN_ERR
4285                        "cciss:  Unable to get major number %d for %s "
4286                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4287                 goto clean1;
4288         } else {
4289                 if (i >= MAX_CTLR_ORIG)
4290                         hba[i]->major = rc;
4291         }
4292
4293         /* make sure the board interrupts are off */
4294         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4295         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
4296                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
4297                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4298                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4299                 goto clean2;
4300         }
4301
4302         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4303                hba[i]->devname, pdev->device, pci_name(pdev),
4304                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4305
4306         hba[i]->cmd_pool_bits =
4307             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4308                         * sizeof(unsigned long), GFP_KERNEL);
4309         hba[i]->cmd_pool = (CommandList_struct *)
4310             pci_alloc_consistent(hba[i]->pdev,
4311                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4312                     &(hba[i]->cmd_pool_dhandle));
4313         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4314             pci_alloc_consistent(hba[i]->pdev,
4315                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4316                     &(hba[i]->errinfo_pool_dhandle));
4317         if ((hba[i]->cmd_pool_bits == NULL)
4318             || (hba[i]->cmd_pool == NULL)
4319             || (hba[i]->errinfo_pool == NULL)) {
4320                 printk(KERN_ERR "cciss: out of memory");
4321                 goto clean4;
4322         }
4323         spin_lock_init(&hba[i]->lock);
4324
4325         /* Initialize the pdev driver private data.
4326            have it point to hba[i].  */
4327         pci_set_drvdata(pdev, hba[i]);
4328         /* command and error info recs zeroed out before
4329            they are used */
4330         memset(hba[i]->cmd_pool_bits, 0,
4331                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4332                         * sizeof(unsigned long));
4333
4334         hba[i]->num_luns = 0;
4335         hba[i]->highest_lun = -1;
4336         for (j = 0; j < CISS_MAX_LUN; j++) {
4337                 hba[i]->drv[j] = NULL;
4338                 hba[i]->gendisk[j] = NULL;
4339         }
4340
4341         cciss_scsi_setup(i);
4342
4343         /* Turn the interrupts on so we can service requests */
4344         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4345
4346         /* Get the firmware version */
4347         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4348         if (inq_buff == NULL) {
4349                 printk(KERN_ERR "cciss: out of memory\n");
4350                 goto clean4;
4351         }
4352
4353         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4354                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4355         if (return_code == IO_OK) {
4356                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4357                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4358                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4359                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4360         } else {         /* send command failed */
4361                 printk(KERN_WARNING "cciss: unable to determine firmware"
4362                         " version of controller\n");
4363         }
4364         kfree(inq_buff);
4365
4366         cciss_procinit(i);
4367
4368         hba[i]->cciss_max_sectors = 2048;
4369
4370         rebuild_lun_table(hba[i], 1, 0);
4371         hba[i]->busy_initializing = 0;
4372         return 1;
4373
4374 clean4:
4375         kfree(hba[i]->cmd_pool_bits);
4376         if (hba[i]->cmd_pool)
4377                 pci_free_consistent(hba[i]->pdev,
4378                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4379                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4380         if (hba[i]->errinfo_pool)
4381                 pci_free_consistent(hba[i]->pdev,
4382                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4383                                     hba[i]->errinfo_pool,
4384                                     hba[i]->errinfo_pool_dhandle);
4385         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4386 clean2:
4387         unregister_blkdev(hba[i]->major, hba[i]->devname);
4388 clean1:
4389         cciss_destroy_hba_sysfs_entry(hba[i]);
4390 clean0:
4391         pci_release_regions(pdev);
4392 clean_no_release_regions:
4393         hba[i]->busy_initializing = 0;
4394
4395         /*
4396          * Deliberately omit pci_disable_device(): it does something nasty to
4397          * Smart Array controllers that pci_enable_device does not undo
4398          */
4399         pci_set_drvdata(pdev, NULL);
4400         free_hba(i);
4401         return -1;
4402 }
4403
4404 static void cciss_shutdown(struct pci_dev *pdev)
4405 {
4406         ctlr_info_t *tmp_ptr;
4407         int i;
4408         char flush_buf[4];
4409         int return_code;
4410
4411         tmp_ptr = pci_get_drvdata(pdev);
4412         if (tmp_ptr == NULL)
4413                 return;
4414         i = tmp_ptr->ctlr;
4415         if (hba[i] == NULL)
4416                 return;
4417
4418         /* Turn board interrupts off  and send the flush cache command */
4419         /* sendcmd will turn off interrupt, and send the flush...
4420          * To write all data in the battery backed cache to disks */
4421         memset(flush_buf, 0, 4);
4422         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0,
4423                 CTLR_LUNID, TYPE_CMD);
4424         if (return_code == IO_OK) {
4425                 printk(KERN_INFO "Completed flushing cache on controller %d\n", i);
4426         } else {
4427                 printk(KERN_WARNING "Error flushing cache on controller %d\n", i);
4428         }
4429         free_irq(hba[i]->intr[2], hba[i]);
4430 }
4431
4432 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4433 {
4434         ctlr_info_t *tmp_ptr;
4435         int i, j;
4436
4437         if (pci_get_drvdata(pdev) == NULL) {
4438                 printk(KERN_ERR "cciss: Unable to remove device \n");
4439                 return;
4440         }
4441
4442         tmp_ptr = pci_get_drvdata(pdev);
4443         i = tmp_ptr->ctlr;
4444         if (hba[i] == NULL) {
4445                 printk(KERN_ERR "cciss: device appears to "
4446                        "already be removed \n");
4447                 return;
4448         }
4449
4450         mutex_lock(&hba[i]->busy_shutting_down);
4451
4452         remove_from_scan_list(hba[i]);
4453         remove_proc_entry(hba[i]->devname, proc_cciss);
4454         unregister_blkdev(hba[i]->major, hba[i]->devname);
4455
4456         /* remove it from the disk list */
4457         for (j = 0; j < CISS_MAX_LUN; j++) {
4458                 struct gendisk *disk = hba[i]->gendisk[j];
4459                 if (disk) {
4460                         struct request_queue *q = disk->queue;
4461
4462                         if (disk->flags & GENHD_FL_UP) {
4463                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4464                                 del_gendisk(disk);
4465                         }
4466                         if (q)
4467                                 blk_cleanup_queue(q);
4468                 }
4469         }
4470
4471 #ifdef CONFIG_CISS_SCSI_TAPE
4472         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4473 #endif
4474
4475         cciss_shutdown(pdev);
4476
4477 #ifdef CONFIG_PCI_MSI
4478         if (hba[i]->msix_vector)
4479                 pci_disable_msix(hba[i]->pdev);
4480         else if (hba[i]->msi_vector)
4481                 pci_disable_msi(hba[i]->pdev);
4482 #endif                          /* CONFIG_PCI_MSI */
4483
4484         iounmap(hba[i]->vaddr);
4485
4486         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4487                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4488         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4489                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4490         kfree(hba[i]->cmd_pool_bits);
4491         /*
4492          * Deliberately omit pci_disable_device(): it does something nasty to
4493          * Smart Array controllers that pci_enable_device does not undo
4494          */
4495         pci_release_regions(pdev);
4496         pci_set_drvdata(pdev, NULL);
4497         cciss_destroy_hba_sysfs_entry(hba[i]);
4498         mutex_unlock(&hba[i]->busy_shutting_down);
4499         free_hba(i);
4500 }
4501
4502 static struct pci_driver cciss_pci_driver = {
4503         .name = "cciss",
4504         .probe = cciss_init_one,
4505         .remove = __devexit_p(cciss_remove_one),
4506         .id_table = cciss_pci_device_id,        /* id_table */
4507         .shutdown = cciss_shutdown,
4508 };
4509
4510 /*
4511  *  This is it.  Register the PCI driver information for the cards we control
4512  *  the OS will call our registered routines when it finds one of our cards.
4513  */
4514 static int __init cciss_init(void)
4515 {
4516         int err;
4517
4518         /*
4519          * The hardware requires that commands are aligned on a 64-bit
4520          * boundary. Given that we use pci_alloc_consistent() to allocate an
4521          * array of them, the size must be a multiple of 8 bytes.
4522          */
4523         BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4524
4525         printk(KERN_INFO DRIVER_NAME "\n");
4526
4527         err = bus_register(&cciss_bus_type);
4528         if (err)
4529                 return err;
4530
4531         /* Start the scan thread */
4532         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4533         if (IS_ERR(cciss_scan_thread)) {
4534                 err = PTR_ERR(cciss_scan_thread);
4535                 goto err_bus_unregister;
4536         }
4537
4538         /* Register for our PCI devices */
4539         err = pci_register_driver(&cciss_pci_driver);
4540         if (err)
4541                 goto err_thread_stop;
4542
4543         return err;
4544
4545 err_thread_stop:
4546         kthread_stop(cciss_scan_thread);
4547 err_bus_unregister:
4548         bus_unregister(&cciss_bus_type);
4549
4550         return err;
4551 }
4552
4553 static void __exit cciss_cleanup(void)
4554 {
4555         int i;
4556
4557         pci_unregister_driver(&cciss_pci_driver);
4558         /* double check that all controller entrys have been removed */
4559         for (i = 0; i < MAX_CTLR; i++) {
4560                 if (hba[i] != NULL) {
4561                         printk(KERN_WARNING "cciss: had to remove"
4562                                " controller %d\n", i);
4563                         cciss_remove_one(hba[i]->pdev);
4564                 }
4565         }
4566         kthread_stop(cciss_scan_thread);
4567         remove_proc_entry("driver/cciss", NULL);
4568         bus_unregister(&cciss_bus_type);
4569 }
4570
4571 static void fail_all_cmds(unsigned long ctlr)
4572 {
4573         /* If we get here, the board is apparently dead. */
4574         ctlr_info_t *h = hba[ctlr];
4575         CommandList_struct *c;
4576         unsigned long flags;
4577
4578         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4579         h->alive = 0;           /* the controller apparently died... */
4580
4581         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4582
4583         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
4584
4585         /* move everything off the request queue onto the completed queue */
4586         while (!hlist_empty(&h->reqQ)) {
4587                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4588                 removeQ(c);
4589                 h->Qdepth--;
4590                 addQ(&h->cmpQ, c);
4591         }
4592
4593         /* Now, fail everything on the completed queue with a HW error */
4594         while (!hlist_empty(&h->cmpQ)) {
4595                 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4596                 removeQ(c);
4597                 if (c->cmd_type != CMD_MSG_STALE)
4598                         c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4599                 if (c->cmd_type == CMD_RWREQ) {
4600                         complete_command(h, c, 0);
4601                 } else if (c->cmd_type == CMD_IOCTL_PEND)
4602                         complete(c->waiting);
4603 #ifdef CONFIG_CISS_SCSI_TAPE
4604                 else if (c->cmd_type == CMD_SCSI)
4605                         complete_scsi_command(c, 0, 0);
4606 #endif
4607         }
4608         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4609         return;
4610 }
4611
4612 module_init(cciss_init);
4613 module_exit(cciss_cleanup);