SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/compat.h>
  43 #include <linux/mutex.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/io.h>
  46
  47 #include <linux/dma-mapping.h>
  48 #include <linux/blkdev.h>
  49 #include <linux/genhd.h>
  50 #include <linux/completion.h>
  51 #include <scsi/scsi.h>
  52 #include <scsi/sg.h>
  53 #include <scsi/scsi_ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/scatterlist.h>
  56 #include <linux/kthread.h>
  57
  58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  59 #define DRIVER_NAME "HP CISS Driver (v 3.6.20)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 20)
  61
  62 /* Embedded module documentation macros - see modules.h */
  63 MODULE_AUTHOR("Hewlett-Packard Company");
  64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  65 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
  66                         " SA6i P600 P800 P400 P400i E200 E200i E500 P700m"
  67                         " Smart Array G2 Series SAS/SATA Controllers");
  68 MODULE_VERSION("3.6.20");
  69 MODULE_LICENSE("GPL");
  70
  71 static int cciss_allow_hpsa;
  72 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  73 MODULE_PARM_DESC(cciss_allow_hpsa,
  74         "Prevent cciss driver from accessing hardware known to be "
  75         " supported by the hpsa driver");
  76
  77 #include "cciss_cmd.h"
  78 #include "cciss.h"
  79 #include <linux/cciss_ioctl.h>
  80
  81 /* define the PCI info for the cards we can control */
  82 static const struct pci_device_id cciss_pci_device_id[] = {
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  90         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  91         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 110         {0,}
 111 };
 112
 113 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 114
 115 /*  board_id = Subsystem Device ID & Vendor ID
 116  *  product = Marketing Name for the board
 117  *  access = Address of the struct of function pointers
 118  */
 119 static struct board_type products[] = {
 120         {0x40700E11, "Smart Array 5300", &SA5_access},
 121         {0x40800E11, "Smart Array 5i", &SA5B_access},
 122         {0x40820E11, "Smart Array 532", &SA5B_access},
 123         {0x40830E11, "Smart Array 5312", &SA5B_access},
 124         {0x409A0E11, "Smart Array 641", &SA5_access},
 125         {0x409B0E11, "Smart Array 642", &SA5_access},
 126         {0x409C0E11, "Smart Array 6400", &SA5_access},
 127         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 128         {0x40910E11, "Smart Array 6i", &SA5_access},
 129         {0x3225103C, "Smart Array P600", &SA5_access},
 130         {0x3235103C, "Smart Array P400i", &SA5_access},
 131         {0x3211103C, "Smart Array E200i", &SA5_access},
 132         {0x3212103C, "Smart Array E200", &SA5_access},
 133         {0x3213103C, "Smart Array E200i", &SA5_access},
 134         {0x3214103C, "Smart Array E200i", &SA5_access},
 135         {0x3215103C, "Smart Array E200i", &SA5_access},
 136         {0x3237103C, "Smart Array E500", &SA5_access},
 137 /* controllers below this line are also supported by the hpsa driver. */
 138 #define HPSA_BOUNDARY 0x3223103C
 139         {0x3223103C, "Smart Array P800", &SA5_access},
 140         {0x3234103C, "Smart Array P400", &SA5_access},
 141         {0x323D103C, "Smart Array P700m", &SA5_access},
 142         {0x3241103C, "Smart Array P212", &SA5_access},
 143         {0x3243103C, "Smart Array P410", &SA5_access},
 144         {0x3245103C, "Smart Array P410i", &SA5_access},
 145         {0x3247103C, "Smart Array P411", &SA5_access},
 146         {0x3249103C, "Smart Array P812", &SA5_access},
 147         {0x324A103C, "Smart Array P712m", &SA5_access},
 148         {0x324B103C, "Smart Array P711m", &SA5_access},
 149 };
 150
 151 /* How long to wait (in milliseconds) for board to go into simple mode */
 152 #define MAX_CONFIG_WAIT 30000
 153 #define MAX_IOCTL_CONFIG_WAIT 1000
 154
 155 /*define how many times we will try a command because of bus resets */
 156 #define MAX_CMD_RETRIES 3
 157
 158 #define MAX_CTLR        32
 159
 160 /* Originally cciss driver only supports 8 major numbers */
 161 #define MAX_CTLR_ORIG   8
 162
 163 static ctlr_info_t *hba[MAX_CTLR];
 164
 165 static struct task_struct *cciss_scan_thread;
 166 static DEFINE_MUTEX(scan_mutex);
 167 static LIST_HEAD(scan_q);
 168
 169 static void do_cciss_request(struct request_queue *q);
 170 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
 171 static int cciss_open(struct block_device *bdev, fmode_t mode);
 172 static int cciss_release(struct gendisk *disk, fmode_t mode);
 173 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 174                        unsigned int cmd, unsigned long arg);
 175 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 176
 177 static int cciss_revalidate(struct gendisk *disk);
 178 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 179 static int deregister_disk(ctlr_info_t *h, int drv_index,
 180                            int clear_all, int via_ioctl);
 181
 182 static void cciss_read_capacity(int ctlr, int logvol,
 183                         sector_t *total_size, unsigned int *block_size);
 184 static void cciss_read_capacity_16(int ctlr, int logvol,
 185                         sector_t *total_size, unsigned int *block_size);
 186 static void cciss_geometry_inquiry(int ctlr, int logvol,
 187                         sector_t total_size,
 188                         unsigned int block_size, InquiryData_struct *inq_buff,
 189                                    drive_info_struct *drv);
 190 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 191                                            __u32);
 192 static void start_io(ctlr_info_t *h);
 193 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 194                         __u8 page_code, unsigned char scsi3addr[],
 195                         int cmd_type);
 196 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 197         int attempt_retry);
 198 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 199
 200 static void fail_all_cmds(unsigned long ctlr);
 201 static int add_to_scan_list(struct ctlr_info *h);
 202 static int scan_thread(void *data);
 203 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 204 static void cciss_hba_release(struct device *dev);
 205 static void cciss_device_release(struct device *dev);
 206 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 207 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 208
 209 #ifdef CONFIG_PROC_FS
 210 static void cciss_procinit(int i);
 211 #else
 212 static void cciss_procinit(int i)
 213 {
 214 }
 215 #endif                          /* CONFIG_PROC_FS */
 216
 217 #ifdef CONFIG_COMPAT
 218 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 219                               unsigned, unsigned long);
 220 #endif
 221
 222 static const struct block_device_operations cciss_fops = {
 223         .owner = THIS_MODULE,
 224         .open = cciss_open,
 225         .release = cciss_release,
 226         .locked_ioctl = cciss_ioctl,
 227         .getgeo = cciss_getgeo,
 228 #ifdef CONFIG_COMPAT
 229         .compat_ioctl = cciss_compat_ioctl,
 230 #endif
 231         .revalidate_disk = cciss_revalidate,
 232 };
 233
 234 /*
 235  * Enqueuing and dequeuing functions for cmdlists.
 236  */
 237 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 238 {
 239         hlist_add_head(&c->list, list);
 240 }
 241
 242 static inline void removeQ(CommandList_struct *c)
 243 {
 244         /*
 245          * After kexec/dump some commands might still
 246          * be in flight, which the firmware will try
 247          * to complete. Resetting the firmware doesn't work
 248          * with old fw revisions, so we have to mark
 249          * them off as 'stale' to prevent the driver from
 250          * falling over.
 251          */
 252         if (WARN_ON(hlist_unhashed(&c->list))) {
 253                 c->cmd_type = CMD_MSG_STALE;
 254                 return;
 255         }
 256
 257         hlist_del_init(&c->list);
 258 }
 259
 260 #include "cciss_scsi.c"         /* For SCSI tape support */
 261
 262 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 263         "UNKNOWN"
 264 };
 265 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
 266
 267 #ifdef CONFIG_PROC_FS
 268
 269 /*
 270  * Report information about this controller.
 271  */
 272 #define ENG_GIG 1000000000
 273 #define ENG_GIG_FACTOR (ENG_GIG/512)
 274 #define ENGAGE_SCSI     "engage scsi"
 275
 276 static struct proc_dir_entry *proc_cciss;
 277
 278 static void cciss_seq_show_header(struct seq_file *seq)
 279 {
 280         ctlr_info_t *h = seq->private;
 281
 282         seq_printf(seq, "%s: HP %s Controller\n"
 283                 "Board ID: 0x%08lx\n"
 284                 "Firmware Version: %c%c%c%c\n"
 285                 "IRQ: %d\n"
 286                 "Logical drives: %d\n"
 287                 "Current Q depth: %d\n"
 288                 "Current # commands on controller: %d\n"
 289                 "Max Q depth since init: %d\n"
 290                 "Max # commands on controller since init: %d\n"
 291                 "Max SG entries since init: %d\n",
 292                 h->devname,
 293                 h->product_name,
 294                 (unsigned long)h->board_id,
 295                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 296                 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
 297                 h->num_luns,
 298                 h->Qdepth, h->commands_outstanding,
 299                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 300
 301 #ifdef CONFIG_CISS_SCSI_TAPE
 302         cciss_seq_tape_report(seq, h->ctlr);
 303 #endif /* CONFIG_CISS_SCSI_TAPE */
 304 }
 305
 306 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 307 {
 308         ctlr_info_t *h = seq->private;
 309         unsigned ctlr = h->ctlr;
 310         unsigned long flags;
 311
 312         /* prevent displaying bogus info during configuration
 313          * or deconfiguration of a logical volume
 314          */
 315         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 316         if (h->busy_configuring) {
 317                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 318                 return ERR_PTR(-EBUSY);
 319         }
 320         h->busy_configuring = 1;
 321         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 322
 323         if (*pos == 0)
 324                 cciss_seq_show_header(seq);
 325
 326         return pos;
 327 }
 328
 329 static int cciss_seq_show(struct seq_file *seq, void *v)
 330 {
 331         sector_t vol_sz, vol_sz_frac;
 332         ctlr_info_t *h = seq->private;
 333         unsigned ctlr = h->ctlr;
 334         loff_t *pos = v;
 335         drive_info_struct *drv = h->drv[*pos];
 336
 337         if (*pos > h->highest_lun)
 338                 return 0;
 339
 340         if (drv->heads == 0)
 341                 return 0;
 342
 343         vol_sz = drv->nr_blocks;
 344         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 345         vol_sz_frac *= 100;
 346         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 347
 348         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 349                 drv->raid_level = RAID_UNKNOWN;
 350         seq_printf(seq, "cciss/c%dd%d:"
 351                         "\t%4u.%02uGB\tRAID %s\n",
 352                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 353                         raid_label[drv->raid_level]);
 354         return 0;
 355 }
 356
 357 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 358 {
 359         ctlr_info_t *h = seq->private;
 360
 361         if (*pos > h->highest_lun)
 362                 return NULL;
 363         *pos += 1;
 364
 365         return pos;
 366 }
 367
 368 static void cciss_seq_stop(struct seq_file *seq, void *v)
 369 {
 370         ctlr_info_t *h = seq->private;
 371
 372         /* Only reset h->busy_configuring if we succeeded in setting
 373          * it during cciss_seq_start. */
 374         if (v == ERR_PTR(-EBUSY))
 375                 return;
 376
 377         h->busy_configuring = 0;
 378 }
 379
 380 static const struct seq_operations cciss_seq_ops = {
 381         .start = cciss_seq_start,
 382         .show  = cciss_seq_show,
 383         .next  = cciss_seq_next,
 384         .stop  = cciss_seq_stop,
 385 };
 386
 387 static int cciss_seq_open(struct inode *inode, struct file *file)
 388 {
 389         int ret = seq_open(file, &cciss_seq_ops);
 390         struct seq_file *seq = file->private_data;
 391
 392         if (!ret)
 393                 seq->private = PDE(inode)->data;
 394
 395         return ret;
 396 }
 397
 398 static ssize_t
 399 cciss_proc_write(struct file *file, const char __user *buf,
 400                  size_t length, loff_t *ppos)
 401 {
 402         int err;
 403         char *buffer;
 404
 405 #ifndef CONFIG_CISS_SCSI_TAPE
 406         return -EINVAL;
 407 #endif
 408
 409         if (!buf || length > PAGE_SIZE - 1)
 410                 return -EINVAL;
 411
 412         buffer = (char *)__get_free_page(GFP_KERNEL);
 413         if (!buffer)
 414                 return -ENOMEM;
 415
 416         err = -EFAULT;
 417         if (copy_from_user(buffer, buf, length))
 418                 goto out;
 419         buffer[length] = '\0';
 420
 421 #ifdef CONFIG_CISS_SCSI_TAPE
 422         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 423                 struct seq_file *seq = file->private_data;
 424                 ctlr_info_t *h = seq->private;
 425
 426                 err = cciss_engage_scsi(h->ctlr);
 427                 if (err == 0)
 428                         err = length;
 429         } else
 430 #endif /* CONFIG_CISS_SCSI_TAPE */
 431                 err = -EINVAL;
 432         /* might be nice to have "disengage" too, but it's not
 433            safely possible. (only 1 module use count, lock issues.) */
 434
 435 out:
 436         free_page((unsigned long)buffer);
 437         return err;
 438 }
 439
 440 static const struct file_operations cciss_proc_fops = {
 441         .owner   = THIS_MODULE,
 442         .open    = cciss_seq_open,
 443         .read    = seq_read,
 444         .llseek  = seq_lseek,
 445         .release = seq_release,
 446         .write   = cciss_proc_write,
 447 };
 448
 449 static void __devinit cciss_procinit(int i)
 450 {
 451         struct proc_dir_entry *pde;
 452
 453         if (proc_cciss == NULL)
 454                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 455         if (!proc_cciss)
 456                 return;
 457         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 458                                         S_IROTH, proc_cciss,
 459                                         &cciss_proc_fops, hba[i]);
 460 }
 461 #endif                          /* CONFIG_PROC_FS */
 462
 463 #define MAX_PRODUCT_NAME_LEN 19
 464
 465 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 466 #define to_drv(n) container_of(n, drive_info_struct, dev)
 467
 468 static ssize_t host_store_rescan(struct device *dev,
 469                                  struct device_attribute *attr,
 470                                  const char *buf, size_t count)
 471 {
 472         struct ctlr_info *h = to_hba(dev);
 473
 474         add_to_scan_list(h);
 475         wake_up_process(cciss_scan_thread);
 476         wait_for_completion_interruptible(&h->scan_wait);
 477
 478         return count;
 479 }
 480 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 481
 482 static ssize_t dev_show_unique_id(struct device *dev,
 483                                  struct device_attribute *attr,
 484                                  char *buf)
 485 {
 486         drive_info_struct *drv = to_drv(dev);
 487         struct ctlr_info *h = to_hba(drv->dev.parent);
 488         __u8 sn[16];
 489         unsigned long flags;
 490         int ret = 0;
 491
 492         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 493         if (h->busy_configuring)
 494                 ret = -EBUSY;
 495         else
 496                 memcpy(sn, drv->serial_no, sizeof(sn));
 497         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 498
 499         if (ret)
 500                 return ret;
 501         else
 502                 return snprintf(buf, 16 * 2 + 2,
 503                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 504                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 505                                 sn[0], sn[1], sn[2], sn[3],
 506                                 sn[4], sn[5], sn[6], sn[7],
 507                                 sn[8], sn[9], sn[10], sn[11],
 508                                 sn[12], sn[13], sn[14], sn[15]);
 509 }
 510 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 511
 512 static ssize_t dev_show_vendor(struct device *dev,
 513                                struct device_attribute *attr,
 514                                char *buf)
 515 {
 516         drive_info_struct *drv = to_drv(dev);
 517         struct ctlr_info *h = to_hba(drv->dev.parent);
 518         char vendor[VENDOR_LEN + 1];
 519         unsigned long flags;
 520         int ret = 0;
 521
 522         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 523         if (h->busy_configuring)
 524                 ret = -EBUSY;
 525         else
 526                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 527         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 528
 529         if (ret)
 530                 return ret;
 531         else
 532                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 533 }
 534 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 535
 536 static ssize_t dev_show_model(struct device *dev,
 537                               struct device_attribute *attr,
 538                               char *buf)
 539 {
 540         drive_info_struct *drv = to_drv(dev);
 541         struct ctlr_info *h = to_hba(drv->dev.parent);
 542         char model[MODEL_LEN + 1];
 543         unsigned long flags;
 544         int ret = 0;
 545
 546         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 547         if (h->busy_configuring)
 548                 ret = -EBUSY;
 549         else
 550                 memcpy(model, drv->model, MODEL_LEN + 1);
 551         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 552
 553         if (ret)
 554                 return ret;
 555         else
 556                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 557 }
 558 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 559
 560 static ssize_t dev_show_rev(struct device *dev,
 561                             struct device_attribute *attr,
 562                             char *buf)
 563 {
 564         drive_info_struct *drv = to_drv(dev);
 565         struct ctlr_info *h = to_hba(drv->dev.parent);
 566         char rev[REV_LEN + 1];
 567         unsigned long flags;
 568         int ret = 0;
 569
 570         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 571         if (h->busy_configuring)
 572                 ret = -EBUSY;
 573         else
 574                 memcpy(rev, drv->rev, REV_LEN + 1);
 575         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 576
 577         if (ret)
 578                 return ret;
 579         else
 580                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 581 }
 582 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 583
 584 static ssize_t cciss_show_lunid(struct device *dev,
 585                                 struct device_attribute *attr, char *buf)
 586 {
 587         drive_info_struct *drv = to_drv(dev);
 588         struct ctlr_info *h = to_hba(drv->dev.parent);
 589         unsigned long flags;
 590         unsigned char lunid[8];
 591
 592         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 593         if (h->busy_configuring) {
 594                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 595                 return -EBUSY;
 596         }
 597         if (!drv->heads) {
 598                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 599                 return -ENOTTY;
 600         }
 601         memcpy(lunid, drv->LunID, sizeof(lunid));
 602         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 603         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 604                 lunid[0], lunid[1], lunid[2], lunid[3],
 605                 lunid[4], lunid[5], lunid[6], lunid[7]);
 606 }
 607 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 608
 609 static ssize_t cciss_show_raid_level(struct device *dev,
 610                                      struct device_attribute *attr, char *buf)
 611 {
 612         drive_info_struct *drv = to_drv(dev);
 613         struct ctlr_info *h = to_hba(drv->dev.parent);
 614         int raid;
 615         unsigned long flags;
 616
 617         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 618         if (h->busy_configuring) {
 619                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 620                 return -EBUSY;
 621         }
 622         raid = drv->raid_level;
 623         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 624         if (raid < 0 || raid > RAID_UNKNOWN)
 625                 raid = RAID_UNKNOWN;
 626
 627         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 628                         raid_label[raid]);
 629 }
 630 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 631
 632 static ssize_t cciss_show_usage_count(struct device *dev,
 633                                       struct device_attribute *attr, char *buf)
 634 {
 635         drive_info_struct *drv = to_drv(dev);
 636         struct ctlr_info *h = to_hba(drv->dev.parent);
 637         unsigned long flags;
 638         int count;
 639
 640         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 641         if (h->busy_configuring) {
 642                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 643                 return -EBUSY;
 644         }
 645         count = drv->usage_count;
 646         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 647         return snprintf(buf, 20, "%d\n", count);
 648 }
 649 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 650
 651 static struct attribute *cciss_host_attrs[] = {
 652         &dev_attr_rescan.attr,
 653         NULL
 654 };
 655
 656 static struct attribute_group cciss_host_attr_group = {
 657         .attrs = cciss_host_attrs,
 658 };
 659
 660 static const struct attribute_group *cciss_host_attr_groups[] = {
 661         &cciss_host_attr_group,
 662         NULL
 663 };
 664
 665 static struct device_type cciss_host_type = {
 666         .name           = "cciss_host",
 667         .groups         = cciss_host_attr_groups,
 668         .release        = cciss_hba_release,
 669 };
 670
 671 static struct attribute *cciss_dev_attrs[] = {
 672         &dev_attr_unique_id.attr,
 673         &dev_attr_model.attr,
 674         &dev_attr_vendor.attr,
 675         &dev_attr_rev.attr,
 676         &dev_attr_lunid.attr,
 677         &dev_attr_raid_level.attr,
 678         &dev_attr_usage_count.attr,
 679         NULL
 680 };
 681
 682 static struct attribute_group cciss_dev_attr_group = {
 683         .attrs = cciss_dev_attrs,
 684 };
 685
 686 static const struct attribute_group *cciss_dev_attr_groups[] = {
 687         &cciss_dev_attr_group,
 688         NULL
 689 };
 690
 691 static struct device_type cciss_dev_type = {
 692         .name           = "cciss_device",
 693         .groups         = cciss_dev_attr_groups,
 694         .release        = cciss_device_release,
 695 };
 696
 697 static struct bus_type cciss_bus_type = {
 698         .name           = "cciss",
 699 };
 700
 701 /*
 702  * cciss_hba_release is called when the reference count
 703  * of h->dev goes to zero.
 704  */
 705 static void cciss_hba_release(struct device *dev)
 706 {
 707         /*
 708          * nothing to do, but need this to avoid a warning
 709          * about not having a release handler from lib/kref.c.
 710          */
 711 }
 712
 713 /*
 714  * Initialize sysfs entry for each controller.  This sets up and registers
 715  * the 'cciss#' directory for each individual controller under
 716  * /sys/bus/pci/devices/<dev>/.
 717  */
 718 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 719 {
 720         device_initialize(&h->dev);
 721         h->dev.type = &cciss_host_type;
 722         h->dev.bus = &cciss_bus_type;
 723         dev_set_name(&h->dev, "%s", h->devname);
 724         h->dev.parent = &h->pdev->dev;
 725
 726         return device_add(&h->dev);
 727 }
 728
 729 /*
 730  * Remove sysfs entries for an hba.
 731  */
 732 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 733 {
 734         device_del(&h->dev);
 735         put_device(&h->dev); /* final put. */
 736 }
 737
 738 /* cciss_device_release is called when the reference count
 739  * of h->drv[x]dev goes to zero.
 740  */
 741 static void cciss_device_release(struct device *dev)
 742 {
 743         drive_info_struct *drv = to_drv(dev);
 744         kfree(drv);
 745 }
 746
 747 /*
 748  * Initialize sysfs for each logical drive.  This sets up and registers
 749  * the 'c#d#' directory for each individual logical drive under
 750  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 751  * /sys/block/cciss!c#d# to this entry.
 752  */
 753 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 754                                        int drv_index)
 755 {
 756         struct device *dev;
 757
 758         if (h->drv[drv_index]->device_initialized)
 759                 return 0;
 760
 761         dev = &h->drv[drv_index]->dev;
 762         device_initialize(dev);
 763         dev->type = &cciss_dev_type;
 764         dev->bus = &cciss_bus_type;
 765         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 766         dev->parent = &h->dev;
 767         h->drv[drv_index]->device_initialized = 1;
 768         return device_add(dev);
 769 }
 770
 771 /*
 772  * Remove sysfs entries for a logical drive.
 773  */
 774 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 775         int ctlr_exiting)
 776 {
 777         struct device *dev = &h->drv[drv_index]->dev;
 778
 779         /* special case for c*d0, we only destroy it on controller exit */
 780         if (drv_index == 0 && !ctlr_exiting)
 781                 return;
 782
 783         device_del(dev);
 784         put_device(dev); /* the "final" put. */
 785         h->drv[drv_index] = NULL;
 786 }
 787
 788 /*
 789  * For operations that cannot sleep, a command block is allocated at init,
 790  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 791  * which ones are free or in use.  For operations that can wait for kmalloc
 792  * to possible sleep, this routine can be called with get_from_pool set to 0.
 793  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 794  */
 795 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 796 {
 797         CommandList_struct *c;
 798         int i;
 799         u64bit temp64;
 800         dma_addr_t cmd_dma_handle, err_dma_handle;
 801
 802         if (!get_from_pool) {
 803                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 804                         sizeof(CommandList_struct), &cmd_dma_handle);
 805                 if (c == NULL)
 806                         return NULL;
 807                 memset(c, 0, sizeof(CommandList_struct));
 808
 809                 c->cmdindex = -1;
 810
 811                 c->err_info = (ErrorInfo_struct *)
 812                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 813                             &err_dma_handle);
 814
 815                 if (c->err_info == NULL) {
 816                         pci_free_consistent(h->pdev,
 817                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 818                         return NULL;
 819                 }
 820                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 821         } else {                /* get it out of the controllers pool */
 822
 823                 do {
 824                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 825                         if (i == h->nr_cmds)
 826                                 return NULL;
 827                 } while (test_and_set_bit
 828                          (i & (BITS_PER_LONG - 1),
 829                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 830 #ifdef CCISS_DEBUG
 831                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 832 #endif
 833                 c = h->cmd_pool + i;
 834                 memset(c, 0, sizeof(CommandList_struct));
 835                 cmd_dma_handle = h->cmd_pool_dhandle
 836                     + i * sizeof(CommandList_struct);
 837                 c->err_info = h->errinfo_pool + i;
 838                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 839                 err_dma_handle = h->errinfo_pool_dhandle
 840                     + i * sizeof(ErrorInfo_struct);
 841                 h->nr_allocs++;
 842
 843                 c->cmdindex = i;
 844         }
 845
 846         INIT_HLIST_NODE(&c->list);
 847         c->busaddr = (__u32) cmd_dma_handle;
 848         temp64.val = (__u64) err_dma_handle;
 849         c->ErrDesc.Addr.lower = temp64.val32.lower;
 850         c->ErrDesc.Addr.upper = temp64.val32.upper;
 851         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 852
 853         c->ctlr = h->ctlr;
 854         return c;
 855 }
 856
 857 /*
 858  * Frees a command block that was previously allocated with cmd_alloc().
 859  */
 860 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 861 {
 862         int i;
 863         u64bit temp64;
 864
 865         if (!got_from_pool) {
 866                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 867                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 868                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 869                                     c->err_info, (dma_addr_t) temp64.val);
 870                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 871                                     c, (dma_addr_t) c->busaddr);
 872         } else {
 873                 i = c - h->cmd_pool;
 874                 clear_bit(i & (BITS_PER_LONG - 1),
 875                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 876                 h->nr_frees++;
 877         }
 878 }
 879
 880 static inline ctlr_info_t *get_host(struct gendisk *disk)
 881 {
 882         return disk->queue->queuedata;
 883 }
 884
 885 static inline drive_info_struct *get_drv(struct gendisk *disk)
 886 {
 887         return disk->private_data;
 888 }
 889
 890 /*
 891  * Open.  Make sure the device is really there.
 892  */
 893 static int cciss_open(struct block_device *bdev, fmode_t mode)
 894 {
 895         ctlr_info_t *host = get_host(bdev->bd_disk);
 896         drive_info_struct *drv = get_drv(bdev->bd_disk);
 897
 898 #ifdef CCISS_DEBUG
 899         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
 900 #endif                          /* CCISS_DEBUG */
 901
 902         if (drv->busy_configuring)
 903                 return -EBUSY;
 904         /*
 905          * Root is allowed to open raw volume zero even if it's not configured
 906          * so array config can still work. Root is also allowed to open any
 907          * volume that has a LUN ID, so it can issue IOCTL to reread the
 908          * disk information.  I don't think I really like this
 909          * but I'm already using way to many device nodes to claim another one
 910          * for "raw controller".
 911          */
 912         if (drv->heads == 0) {
 913                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
 914                         /* if not node 0 make sure it is a partition = 0 */
 915                         if (MINOR(bdev->bd_dev) & 0x0f) {
 916                                 return -ENXIO;
 917                                 /* if it is, make sure we have a LUN ID */
 918                         } else if (memcmp(drv->LunID, CTLR_LUNID,
 919                                 sizeof(drv->LunID))) {
 920                                 return -ENXIO;
 921                         }
 922                 }
 923                 if (!capable(CAP_SYS_ADMIN))
 924                         return -EPERM;
 925         }
 926         drv->usage_count++;
 927         host->usage_count++;
 928         return 0;
 929 }
 930
 931 /*
 932  * Close.  Sync first.
 933  */
 934 static int cciss_release(struct gendisk *disk, fmode_t mode)
 935 {
 936         ctlr_info_t *host = get_host(disk);
 937         drive_info_struct *drv = get_drv(disk);
 938
 939 #ifdef CCISS_DEBUG
 940         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
 941 #endif                          /* CCISS_DEBUG */
 942
 943         drv->usage_count--;
 944         host->usage_count--;
 945         return 0;
 946 }
 947
 948 #ifdef CONFIG_COMPAT
 949
 950 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 951                     unsigned cmd, unsigned long arg)
 952 {
 953         int ret;
 954         lock_kernel();
 955         ret = cciss_ioctl(bdev, mode, cmd, arg);
 956         unlock_kernel();
 957         return ret;
 958 }
 959
 960 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 961                                   unsigned cmd, unsigned long arg);
 962 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
 963                                       unsigned cmd, unsigned long arg);
 964
 965 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
 966                               unsigned cmd, unsigned long arg)
 967 {
 968         switch (cmd) {
 969         case CCISS_GETPCIINFO:
 970         case CCISS_GETINTINFO:
 971         case CCISS_SETINTINFO:
 972         case CCISS_GETNODENAME:
 973         case CCISS_SETNODENAME:
 974         case CCISS_GETHEARTBEAT:
 975         case CCISS_GETBUSTYPES:
 976         case CCISS_GETFIRMVER:
 977         case CCISS_GETDRIVVER:
 978         case CCISS_REVALIDVOLS:
 979         case CCISS_DEREGDISK:
 980         case CCISS_REGNEWDISK:
 981         case CCISS_REGNEWD:
 982         case CCISS_RESCANDISK:
 983         case CCISS_GETLUNINFO:
 984                 return do_ioctl(bdev, mode, cmd, arg);
 985
 986         case CCISS_PASSTHRU32:
 987                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
 988         case CCISS_BIG_PASSTHRU32:
 989                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
 990
 991         default:
 992                 return -ENOIOCTLCMD;
 993         }
 994 }
 995
 996 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 997                                   unsigned cmd, unsigned long arg)
 998 {
 999         IOCTL32_Command_struct __user *arg32 =
1000             (IOCTL32_Command_struct __user *) arg;
1001         IOCTL_Command_struct arg64;
1002         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1003         int err;
1004         u32 cp;
1005
1006         err = 0;
1007         err |=
1008             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1009                            sizeof(arg64.LUN_info));
1010         err |=
1011             copy_from_user(&arg64.Request, &arg32->Request,
1012                            sizeof(arg64.Request));
1013         err |=
1014             copy_from_user(&arg64.error_info, &arg32->error_info,
1015                            sizeof(arg64.error_info));
1016         err |= get_user(arg64.buf_size, &arg32->buf_size);
1017         err |= get_user(cp, &arg32->buf);
1018         arg64.buf = compat_ptr(cp);
1019         err |= copy_to_user(p, &arg64, sizeof(arg64));
1020
1021         if (err)
1022                 return -EFAULT;
1023
1024         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1025         if (err)
1026                 return err;
1027         err |=
1028             copy_in_user(&arg32->error_info, &p->error_info,
1029                          sizeof(arg32->error_info));
1030         if (err)
1031                 return -EFAULT;
1032         return err;
1033 }
1034
1035 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1036                                       unsigned cmd, unsigned long arg)
1037 {
1038         BIG_IOCTL32_Command_struct __user *arg32 =
1039             (BIG_IOCTL32_Command_struct __user *) arg;
1040         BIG_IOCTL_Command_struct arg64;
1041         BIG_IOCTL_Command_struct __user *p =
1042             compat_alloc_user_space(sizeof(arg64));
1043         int err;
1044         u32 cp;
1045
1046         err = 0;
1047         err |=
1048             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1049                            sizeof(arg64.LUN_info));
1050         err |=
1051             copy_from_user(&arg64.Request, &arg32->Request,
1052                            sizeof(arg64.Request));
1053         err |=
1054             copy_from_user(&arg64.error_info, &arg32->error_info,
1055                            sizeof(arg64.error_info));
1056         err |= get_user(arg64.buf_size, &arg32->buf_size);
1057         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1058         err |= get_user(cp, &arg32->buf);
1059         arg64.buf = compat_ptr(cp);
1060         err |= copy_to_user(p, &arg64, sizeof(arg64));
1061
1062         if (err)
1063                 return -EFAULT;
1064
1065         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1066         if (err)
1067                 return err;
1068         err |=
1069             copy_in_user(&arg32->error_info, &p->error_info,
1070                          sizeof(arg32->error_info));
1071         if (err)
1072                 return -EFAULT;
1073         return err;
1074 }
1075 #endif
1076
1077 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1078 {
1079         drive_info_struct *drv = get_drv(bdev->bd_disk);
1080
1081         if (!drv->cylinders)
1082                 return -ENXIO;
1083
1084         geo->heads = drv->heads;
1085         geo->sectors = drv->sectors;
1086         geo->cylinders = drv->cylinders;
1087         return 0;
1088 }
1089
1090 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1091 {
1092         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1093                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1094                 (void)check_for_unit_attention(host, c);
1095 }
1096 /*
1097  * ioctl
1098  */
1099 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1100                        unsigned int cmd, unsigned long arg)
1101 {
1102         struct gendisk *disk = bdev->bd_disk;
1103         ctlr_info_t *host = get_host(disk);
1104         drive_info_struct *drv = get_drv(disk);
1105         int ctlr = host->ctlr;
1106         void __user *argp = (void __user *)arg;
1107
1108 #ifdef CCISS_DEBUG
1109         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1110 #endif                          /* CCISS_DEBUG */
1111
1112         switch (cmd) {
1113         case CCISS_GETPCIINFO:
1114                 {
1115                         cciss_pci_info_struct pciinfo;
1116
1117                         if (!arg)
1118                                 return -EINVAL;
1119                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1120                         pciinfo.bus = host->pdev->bus->number;
1121                         pciinfo.dev_fn = host->pdev->devfn;
1122                         pciinfo.board_id = host->board_id;
1123                         if (copy_to_user
1124                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1125                                 return -EFAULT;
1126                         return 0;
1127                 }
1128         case CCISS_GETINTINFO:
1129                 {
1130                         cciss_coalint_struct intinfo;
1131                         if (!arg)
1132                                 return -EINVAL;
1133                         intinfo.delay =
1134                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1135                         intinfo.count =
1136                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1137                         if (copy_to_user
1138                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1139                                 return -EFAULT;
1140                         return 0;
1141                 }
1142         case CCISS_SETINTINFO:
1143                 {
1144                         cciss_coalint_struct intinfo;
1145                         unsigned long flags;
1146                         int i;
1147
1148                         if (!arg)
1149                                 return -EINVAL;
1150                         if (!capable(CAP_SYS_ADMIN))
1151                                 return -EPERM;
1152                         if (copy_from_user
1153                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1154                                 return -EFAULT;
1155                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1156                         {
1157 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1158                                 return -EINVAL;
1159                         }
1160                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1161                         /* Update the field, and then ring the doorbell */
1162                         writel(intinfo.delay,
1163                                &(host->cfgtable->HostWrite.CoalIntDelay));
1164                         writel(intinfo.count,
1165                                &(host->cfgtable->HostWrite.CoalIntCount));
1166                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1167
1168                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1169                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1170                                       & CFGTBL_ChangeReq))
1171                                         break;
1172                                 /* delay and try again */
1173                                 udelay(1000);
1174                         }
1175                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1176                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1177                                 return -EAGAIN;
1178                         return 0;
1179                 }
1180         case CCISS_GETNODENAME:
1181                 {
1182                         NodeName_type NodeName;
1183                         int i;
1184
1185                         if (!arg)
1186                                 return -EINVAL;
1187                         for (i = 0; i < 16; i++)
1188                                 NodeName[i] =
1189                                     readb(&host->cfgtable->ServerName[i]);
1190                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1191                                 return -EFAULT;
1192                         return 0;
1193                 }
1194         case CCISS_SETNODENAME:
1195                 {
1196                         NodeName_type NodeName;
1197                         unsigned long flags;
1198                         int i;
1199
1200                         if (!arg)
1201                                 return -EINVAL;
1202                         if (!capable(CAP_SYS_ADMIN))
1203                                 return -EPERM;
1204
1205                         if (copy_from_user
1206                             (NodeName, argp, sizeof(NodeName_type)))
1207                                 return -EFAULT;
1208
1209                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1210
1211                         /* Update the field, and then ring the doorbell */
1212                         for (i = 0; i < 16; i++)
1213                                 writeb(NodeName[i],
1214                                        &host->cfgtable->ServerName[i]);
1215
1216                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1217
1218                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1219                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1220                                       & CFGTBL_ChangeReq))
1221                                         break;
1222                                 /* delay and try again */
1223                                 udelay(1000);
1224                         }
1225                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1226                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1227                                 return -EAGAIN;
1228                         return 0;
1229                 }
1230
1231         case CCISS_GETHEARTBEAT:
1232                 {
1233                         Heartbeat_type heartbeat;
1234
1235                         if (!arg)
1236                                 return -EINVAL;
1237                         heartbeat = readl(&host->cfgtable->HeartBeat);
1238                         if (copy_to_user
1239                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1240                                 return -EFAULT;
1241                         return 0;
1242                 }
1243         case CCISS_GETBUSTYPES:
1244                 {
1245                         BusTypes_type BusTypes;
1246
1247                         if (!arg)
1248                                 return -EINVAL;
1249                         BusTypes = readl(&host->cfgtable->BusTypes);
1250                         if (copy_to_user
1251                             (argp, &BusTypes, sizeof(BusTypes_type)))
1252                                 return -EFAULT;
1253                         return 0;
1254                 }
1255         case CCISS_GETFIRMVER:
1256                 {
1257                         FirmwareVer_type firmware;
1258
1259                         if (!arg)
1260                                 return -EINVAL;
1261                         memcpy(firmware, host->firm_ver, 4);
1262
1263                         if (copy_to_user
1264                             (argp, firmware, sizeof(FirmwareVer_type)))
1265                                 return -EFAULT;
1266                         return 0;
1267                 }
1268         case CCISS_GETDRIVVER:
1269                 {
1270                         DriverVer_type DriverVer = DRIVER_VERSION;
1271
1272                         if (!arg)
1273                                 return -EINVAL;
1274
1275                         if (copy_to_user
1276                             (argp, &DriverVer, sizeof(DriverVer_type)))
1277                                 return -EFAULT;
1278                         return 0;
1279                 }
1280
1281         case CCISS_DEREGDISK:
1282         case CCISS_REGNEWD:
1283         case CCISS_REVALIDVOLS:
1284                 return rebuild_lun_table(host, 0, 1);
1285
1286         case CCISS_GETLUNINFO:{
1287                         LogvolInfo_struct luninfo;
1288
1289                         memcpy(&luninfo.LunID, drv->LunID,
1290                                 sizeof(luninfo.LunID));
1291                         luninfo.num_opens = drv->usage_count;
1292                         luninfo.num_parts = 0;
1293                         if (copy_to_user(argp, &luninfo,
1294                                          sizeof(LogvolInfo_struct)))
1295                                 return -EFAULT;
1296                         return 0;
1297                 }
1298         case CCISS_PASSTHRU:
1299                 {
1300                         IOCTL_Command_struct iocommand;
1301                         CommandList_struct *c;
1302                         char *buff = NULL;
1303                         u64bit temp64;
1304                         unsigned long flags;
1305                         DECLARE_COMPLETION_ONSTACK(wait);
1306
1307                         if (!arg)
1308                                 return -EINVAL;
1309
1310                         if (!capable(CAP_SYS_RAWIO))
1311                                 return -EPERM;
1312
1313                         if (copy_from_user
1314                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1315                                 return -EFAULT;
1316                         if ((iocommand.buf_size < 1) &&
1317                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1318                                 return -EINVAL;
1319                         }
1320 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1321                         /* Check kmalloc limits */
1322                         if (iocommand.buf_size > 128000)
1323                                 return -EINVAL;
1324 #endif
1325                         if (iocommand.buf_size > 0) {
1326                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1327                                 if (buff == NULL)
1328                                         return -EFAULT;
1329                         }
1330                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1331                                 /* Copy the data into the buffer we created */
1332                                 if (copy_from_user
1333                                     (buff, iocommand.buf, iocommand.buf_size)) {
1334                                         kfree(buff);
1335                                         return -EFAULT;
1336                                 }
1337                         } else {
1338                                 memset(buff, 0, iocommand.buf_size);
1339                         }
1340                         if ((c = cmd_alloc(host, 0)) == NULL) {
1341                                 kfree(buff);
1342                                 return -ENOMEM;
1343                         }
1344                         // Fill in the command type
1345                         c->cmd_type = CMD_IOCTL_PEND;
1346                         // Fill in Command Header
1347                         c->Header.ReplyQueue = 0;       // unused in simple mode
1348                         if (iocommand.buf_size > 0)     // buffer to fill
1349                         {
1350                                 c->Header.SGList = 1;
1351                                 c->Header.SGTotal = 1;
1352                         } else  // no buffers to fill
1353                         {
1354                                 c->Header.SGList = 0;
1355                                 c->Header.SGTotal = 0;
1356                         }
1357                         c->Header.LUN = iocommand.LUN_info;
1358                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
1359
1360                         // Fill in Request block
1361                         c->Request = iocommand.Request;
1362
1363                         // Fill in the scatter gather information
1364                         if (iocommand.buf_size > 0) {
1365                                 temp64.val = pci_map_single(host->pdev, buff,
1366                                         iocommand.buf_size,
1367                                         PCI_DMA_BIDIRECTIONAL);
1368                                 c->SG[0].Addr.lower = temp64.val32.lower;
1369                                 c->SG[0].Addr.upper = temp64.val32.upper;
1370                                 c->SG[0].Len = iocommand.buf_size;
1371                                 c->SG[0].Ext = 0;       // we are not chaining
1372                         }
1373                         c->waiting = &wait;
1374
1375                         /* Put the request on the tail of the request queue */
1376                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1377                         addQ(&host->reqQ, c);
1378                         host->Qdepth++;
1379                         start_io(host);
1380                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1381
1382                         wait_for_completion(&wait);
1383
1384                         /* unlock the buffers from DMA */
1385                         temp64.val32.lower = c->SG[0].Addr.lower;
1386                         temp64.val32.upper = c->SG[0].Addr.upper;
1387                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1388                                          iocommand.buf_size,
1389                                          PCI_DMA_BIDIRECTIONAL);
1390
1391                         check_ioctl_unit_attention(host, c);
1392
1393                         /* Copy the error information out */
1394                         iocommand.error_info = *(c->err_info);
1395                         if (copy_to_user
1396                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1397                                 kfree(buff);
1398                                 cmd_free(host, c, 0);
1399                                 return -EFAULT;
1400                         }
1401
1402                         if (iocommand.Request.Type.Direction == XFER_READ) {
1403                                 /* Copy the data out of the buffer we created */
1404                                 if (copy_to_user
1405                                     (iocommand.buf, buff, iocommand.buf_size)) {
1406                                         kfree(buff);
1407                                         cmd_free(host, c, 0);
1408                                         return -EFAULT;
1409                                 }
1410                         }
1411                         kfree(buff);
1412                         cmd_free(host, c, 0);
1413                         return 0;
1414                 }
1415         case CCISS_BIG_PASSTHRU:{
1416                         BIG_IOCTL_Command_struct *ioc;
1417                         CommandList_struct *c;
1418                         unsigned char **buff = NULL;
1419                         int *buff_size = NULL;
1420                         u64bit temp64;
1421                         unsigned long flags;
1422                         BYTE sg_used = 0;
1423                         int status = 0;
1424                         int i;
1425                         DECLARE_COMPLETION_ONSTACK(wait);
1426                         __u32 left;
1427                         __u32 sz;
1428                         BYTE __user *data_ptr;
1429
1430                         if (!arg)
1431                                 return -EINVAL;
1432                         if (!capable(CAP_SYS_RAWIO))
1433                                 return -EPERM;
1434                         ioc = (BIG_IOCTL_Command_struct *)
1435                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1436                         if (!ioc) {
1437                                 status = -ENOMEM;
1438                                 goto cleanup1;
1439                         }
1440                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1441                                 status = -EFAULT;
1442                                 goto cleanup1;
1443                         }
1444                         if ((ioc->buf_size < 1) &&
1445                             (ioc->Request.Type.Direction != XFER_NONE)) {
1446                                 status = -EINVAL;
1447                                 goto cleanup1;
1448                         }
1449                         /* Check kmalloc limits  using all SGs */
1450                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1451                                 status = -EINVAL;
1452                                 goto cleanup1;
1453                         }
1454                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1455                                 status = -EINVAL;
1456                                 goto cleanup1;
1457                         }
1458                         buff =
1459                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1460                         if (!buff) {
1461                                 status = -ENOMEM;
1462                                 goto cleanup1;
1463                         }
1464                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1465                                                    GFP_KERNEL);
1466                         if (!buff_size) {
1467                                 status = -ENOMEM;
1468                                 goto cleanup1;
1469                         }
1470                         left = ioc->buf_size;
1471                         data_ptr = ioc->buf;
1472                         while (left) {
1473                                 sz = (left >
1474                                       ioc->malloc_size) ? ioc->
1475                                     malloc_size : left;
1476                                 buff_size[sg_used] = sz;
1477                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1478                                 if (buff[sg_used] == NULL) {
1479                                         status = -ENOMEM;
1480                                         goto cleanup1;
1481                                 }
1482                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1483                                         if (copy_from_user
1484                                             (buff[sg_used], data_ptr, sz)) {
1485                                                 status = -EFAULT;
1486                                                 goto cleanup1;
1487                                         }
1488                                 } else {
1489                                         memset(buff[sg_used], 0, sz);
1490                                 }
1491                                 left -= sz;
1492                                 data_ptr += sz;
1493                                 sg_used++;
1494                         }
1495                         if ((c = cmd_alloc(host, 0)) == NULL) {
1496                                 status = -ENOMEM;
1497                                 goto cleanup1;
1498                         }
1499                         c->cmd_type = CMD_IOCTL_PEND;
1500                         c->Header.ReplyQueue = 0;
1501
1502                         if (ioc->buf_size > 0) {
1503                                 c->Header.SGList = sg_used;
1504                                 c->Header.SGTotal = sg_used;
1505                         } else {
1506                                 c->Header.SGList = 0;
1507                                 c->Header.SGTotal = 0;
1508                         }
1509                         c->Header.LUN = ioc->LUN_info;
1510                         c->Header.Tag.lower = c->busaddr;
1511
1512                         c->Request = ioc->Request;
1513                         if (ioc->buf_size > 0) {
1514                                 int i;
1515                                 for (i = 0; i < sg_used; i++) {
1516                                         temp64.val =
1517                                             pci_map_single(host->pdev, buff[i],
1518                                                     buff_size[i],
1519                                                     PCI_DMA_BIDIRECTIONAL);
1520                                         c->SG[i].Addr.lower =
1521                                             temp64.val32.lower;
1522                                         c->SG[i].Addr.upper =
1523                                             temp64.val32.upper;
1524                                         c->SG[i].Len = buff_size[i];
1525                                         c->SG[i].Ext = 0;       /* we are not chaining */
1526                                 }
1527                         }
1528                         c->waiting = &wait;
1529                         /* Put the request on the tail of the request queue */
1530                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1531                         addQ(&host->reqQ, c);
1532                         host->Qdepth++;
1533                         start_io(host);
1534                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1535                         wait_for_completion(&wait);
1536                         /* unlock the buffers from DMA */
1537                         for (i = 0; i < sg_used; i++) {
1538                                 temp64.val32.lower = c->SG[i].Addr.lower;
1539                                 temp64.val32.upper = c->SG[i].Addr.upper;
1540                                 pci_unmap_single(host->pdev,
1541                                         (dma_addr_t) temp64.val, buff_size[i],
1542                                         PCI_DMA_BIDIRECTIONAL);
1543                         }
1544                         check_ioctl_unit_attention(host, c);
1545                         /* Copy the error information out */
1546                         ioc->error_info = *(c->err_info);
1547                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1548                                 cmd_free(host, c, 0);
1549                                 status = -EFAULT;
1550                                 goto cleanup1;
1551                         }
1552                         if (ioc->Request.Type.Direction == XFER_READ) {
1553                                 /* Copy the data out of the buffer we created */
1554                                 BYTE __user *ptr = ioc->buf;
1555                                 for (i = 0; i < sg_used; i++) {
1556                                         if (copy_to_user
1557                                             (ptr, buff[i], buff_size[i])) {
1558                                                 cmd_free(host, c, 0);
1559                                                 status = -EFAULT;
1560                                                 goto cleanup1;
1561                                         }
1562                                         ptr += buff_size[i];
1563                                 }
1564                         }
1565                         cmd_free(host, c, 0);
1566                         status = 0;
1567                       cleanup1:
1568                         if (buff) {
1569                                 for (i = 0; i < sg_used; i++)
1570                                         kfree(buff[i]);
1571                                 kfree(buff);
1572                         }
1573                         kfree(buff_size);
1574                         kfree(ioc);
1575                         return status;
1576                 }
1577
1578         /* scsi_cmd_ioctl handles these, below, though some are not */
1579         /* very meaningful for cciss.  SG_IO is the main one people want. */
1580
1581         case SG_GET_VERSION_NUM:
1582         case SG_SET_TIMEOUT:
1583         case SG_GET_TIMEOUT:
1584         case SG_GET_RESERVED_SIZE:
1585         case SG_SET_RESERVED_SIZE:
1586         case SG_EMULATED_HOST:
1587         case SG_IO:
1588         case SCSI_IOCTL_SEND_COMMAND:
1589                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1590
1591         /* scsi_cmd_ioctl would normally handle these, below, but */
1592         /* they aren't a good fit for cciss, as CD-ROMs are */
1593         /* not supported, and we don't have any bus/target/lun */
1594         /* which we present to the kernel. */
1595
1596         case CDROM_SEND_PACKET:
1597         case CDROMCLOSETRAY:
1598         case CDROMEJECT:
1599         case SCSI_IOCTL_GET_IDLUN:
1600         case SCSI_IOCTL_GET_BUS_NUMBER:
1601         default:
1602                 return -ENOTTY;
1603         }
1604 }
1605
1606 static void cciss_check_queues(ctlr_info_t *h)
1607 {
1608         int start_queue = h->next_to_run;
1609         int i;
1610
1611         /* check to see if we have maxed out the number of commands that can
1612          * be placed on the queue.  If so then exit.  We do this check here
1613          * in case the interrupt we serviced was from an ioctl and did not
1614          * free any new commands.
1615          */
1616         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1617                 return;
1618
1619         /* We have room on the queue for more commands.  Now we need to queue
1620          * them up.  We will also keep track of the next queue to run so
1621          * that every queue gets a chance to be started first.
1622          */
1623         for (i = 0; i < h->highest_lun + 1; i++) {
1624                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1625                 /* make sure the disk has been added and the drive is real
1626                  * because this can be called from the middle of init_one.
1627                  */
1628                 if (!h->drv[curr_queue])
1629                         continue;
1630                 if (!(h->drv[curr_queue]->queue) ||
1631                         !(h->drv[curr_queue]->heads))
1632                         continue;
1633                 blk_start_queue(h->gendisk[curr_queue]->queue);
1634
1635                 /* check to see if we have maxed out the number of commands
1636                  * that can be placed on the queue.
1637                  */
1638                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1639                         if (curr_queue == start_queue) {
1640                                 h->next_to_run =
1641                                     (start_queue + 1) % (h->highest_lun + 1);
1642                                 break;
1643                         } else {
1644                                 h->next_to_run = curr_queue;
1645                                 break;
1646                         }
1647                 }
1648         }
1649 }
1650
1651 static void cciss_softirq_done(struct request *rq)
1652 {
1653         CommandList_struct *cmd = rq->completion_data;
1654         ctlr_info_t *h = hba[cmd->ctlr];
1655         SGDescriptor_struct *curr_sg = cmd->SG;
1656         unsigned long flags;
1657         u64bit temp64;
1658         int i, ddir;
1659         int sg_index = 0;
1660
1661         if (cmd->Request.Type.Direction == XFER_READ)
1662                 ddir = PCI_DMA_FROMDEVICE;
1663         else
1664                 ddir = PCI_DMA_TODEVICE;
1665
1666         /* command did not need to be retried */
1667         /* unmap the DMA mapping for all the scatter gather elements */
1668         for (i = 0; i < cmd->Header.SGList; i++) {
1669                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1670                         temp64.val32.lower = cmd->SG[i].Addr.lower;
1671                         temp64.val32.upper = cmd->SG[i].Addr.upper;
1672                         pci_dma_sync_single_for_cpu(h->pdev, temp64.val,
1673                                                 cmd->SG[i].Len, ddir);
1674                         pci_unmap_single(h->pdev, temp64.val,
1675                                                 cmd->SG[i].Len, ddir);
1676                         /* Point to the next block */
1677                         curr_sg = h->cmd_sg_list[cmd->cmdindex]->sgchain;
1678                         sg_index = 0;
1679                 }
1680                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1681                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1682                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1683                                 ddir);
1684                 ++sg_index;
1685         }
1686
1687 #ifdef CCISS_DEBUG
1688         printk("Done with %p\n", rq);
1689 #endif                          /* CCISS_DEBUG */
1690
1691         /* set the residual count for pc requests */
1692         if (blk_pc_request(rq))
1693                 rq->resid_len = cmd->err_info->ResidualCnt;
1694
1695         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1696
1697         spin_lock_irqsave(&h->lock, flags);
1698         cmd_free(h, cmd, 1);
1699         cciss_check_queues(h);
1700         spin_unlock_irqrestore(&h->lock, flags);
1701 }
1702
1703 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1704         unsigned char scsi3addr[], uint32_t log_unit)
1705 {
1706         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1707                 sizeof(h->drv[log_unit]->LunID));
1708 }
1709
1710 /* This function gets the SCSI vendor, model, and revision of a logical drive
1711  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1712  * they cannot be read.
1713  */
1714 static void cciss_get_device_descr(int ctlr, int logvol,
1715                                    char *vendor, char *model, char *rev)
1716 {
1717         int rc;
1718         InquiryData_struct *inq_buf;
1719         unsigned char scsi3addr[8];
1720
1721         *vendor = '\0';
1722         *model = '\0';
1723         *rev = '\0';
1724
1725         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1726         if (!inq_buf)
1727                 return;
1728
1729         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1730         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1731                         scsi3addr, TYPE_CMD);
1732         if (rc == IO_OK) {
1733                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1734                 vendor[VENDOR_LEN] = '\0';
1735                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1736                 model[MODEL_LEN] = '\0';
1737                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1738                 rev[REV_LEN] = '\0';
1739         }
1740
1741         kfree(inq_buf);
1742         return;
1743 }
1744
1745 /* This function gets the serial number of a logical drive via
1746  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1747  * number cannot be had, for whatever reason, 16 bytes of 0xff
1748  * are returned instead.
1749  */
1750 static void cciss_get_serial_no(int ctlr, int logvol,
1751                                 unsigned char *serial_no, int buflen)
1752 {
1753 #define PAGE_83_INQ_BYTES 64
1754         int rc;
1755         unsigned char *buf;
1756         unsigned char scsi3addr[8];
1757
1758         if (buflen > 16)
1759                 buflen = 16;
1760         memset(serial_no, 0xff, buflen);
1761         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1762         if (!buf)
1763                 return;
1764         memset(serial_no, 0, buflen);
1765         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1766         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1767                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1768         if (rc == IO_OK)
1769                 memcpy(serial_no, &buf[8], buflen);
1770         kfree(buf);
1771         return;
1772 }
1773
1774 /*
1775  * cciss_add_disk sets up the block device queue for a logical drive
1776  */
1777 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1778                                 int drv_index)
1779 {
1780         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1781         if (!disk->queue)
1782                 goto init_queue_failure;
1783         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1784         disk->major = h->major;
1785         disk->first_minor = drv_index << NWD_SHIFT;
1786         disk->fops = &cciss_fops;
1787         if (cciss_create_ld_sysfs_entry(h, drv_index))
1788                 goto cleanup_queue;
1789         disk->private_data = h->drv[drv_index];
1790         disk->driverfs_dev = &h->drv[drv_index]->dev;
1791
1792         /* Set up queue information */
1793         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1794
1795         /* This is a hardware imposed limit. */
1796         blk_queue_max_hw_segments(disk->queue, h->maxsgentries);
1797
1798         /* This is a limit in the driver and could be eliminated. */
1799         blk_queue_max_phys_segments(disk->queue, h->maxsgentries);
1800
1801         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1802
1803         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1804
1805         disk->queue->queuedata = h;
1806
1807         blk_queue_logical_block_size(disk->queue,
1808                                      h->drv[drv_index]->block_size);
1809
1810         /* Make sure all queue data is written out before */
1811         /* setting h->drv[drv_index]->queue, as setting this */
1812         /* allows the interrupt handler to start the queue */
1813         wmb();
1814         h->drv[drv_index]->queue = disk->queue;
1815         add_disk(disk);
1816         return 0;
1817
1818 cleanup_queue:
1819         blk_cleanup_queue(disk->queue);
1820         disk->queue = NULL;
1821 init_queue_failure:
1822         return -1;
1823 }
1824
1825 /* This function will check the usage_count of the drive to be updated/added.
1826  * If the usage_count is zero and it is a heretofore unknown drive, or,
1827  * the drive's capacity, geometry, or serial number has changed,
1828  * then the drive information will be updated and the disk will be
1829  * re-registered with the kernel.  If these conditions don't hold,
1830  * then it will be left alone for the next reboot.  The exception to this
1831  * is disk 0 which will always be left registered with the kernel since it
1832  * is also the controller node.  Any changes to disk 0 will show up on
1833  * the next reboot.
1834  */
1835 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1836         int via_ioctl)
1837 {
1838         ctlr_info_t *h = hba[ctlr];
1839         struct gendisk *disk;
1840         InquiryData_struct *inq_buff = NULL;
1841         unsigned int block_size;
1842         sector_t total_size;
1843         unsigned long flags = 0;
1844         int ret = 0;
1845         drive_info_struct *drvinfo;
1846
1847         /* Get information about the disk and modify the driver structure */
1848         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1849         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1850         if (inq_buff == NULL || drvinfo == NULL)
1851                 goto mem_msg;
1852
1853         /* testing to see if 16-byte CDBs are already being used */
1854         if (h->cciss_read == CCISS_READ_16) {
1855                 cciss_read_capacity_16(h->ctlr, drv_index,
1856                         &total_size, &block_size);
1857
1858         } else {
1859                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1860                 /* if read_capacity returns all F's this volume is >2TB */
1861                 /* in size so we switch to 16-byte CDB's for all */
1862                 /* read/write ops */
1863                 if (total_size == 0xFFFFFFFFULL) {
1864                         cciss_read_capacity_16(ctlr, drv_index,
1865                         &total_size, &block_size);
1866                         h->cciss_read = CCISS_READ_16;
1867                         h->cciss_write = CCISS_WRITE_16;
1868                 } else {
1869                         h->cciss_read = CCISS_READ_10;
1870                         h->cciss_write = CCISS_WRITE_10;
1871                 }
1872         }
1873
1874         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1875                                inq_buff, drvinfo);
1876         drvinfo->block_size = block_size;
1877         drvinfo->nr_blocks = total_size + 1;
1878
1879         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1880                                 drvinfo->model, drvinfo->rev);
1881         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1882                         sizeof(drvinfo->serial_no));
1883         /* Save the lunid in case we deregister the disk, below. */
1884         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1885                 sizeof(drvinfo->LunID));
1886
1887         /* Is it the same disk we already know, and nothing's changed? */
1888         if (h->drv[drv_index]->raid_level != -1 &&
1889                 ((memcmp(drvinfo->serial_no,
1890                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1891                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1892                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1893                 drvinfo->heads == h->drv[drv_index]->heads &&
1894                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1895                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1896                         /* The disk is unchanged, nothing to update */
1897                         goto freeret;
1898
1899         /* If we get here it's not the same disk, or something's changed,
1900          * so we need to * deregister it, and re-register it, if it's not
1901          * in use.
1902          * If the disk already exists then deregister it before proceeding
1903          * (unless it's the first disk (for the controller node).
1904          */
1905         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1906                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1907                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1908                 h->drv[drv_index]->busy_configuring = 1;
1909                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1910
1911                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1912                  * which keeps the interrupt handler from starting
1913                  * the queue.
1914                  */
1915                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1916         }
1917
1918         /* If the disk is in use return */
1919         if (ret)
1920                 goto freeret;
1921
1922         /* Save the new information from cciss_geometry_inquiry
1923          * and serial number inquiry.  If the disk was deregistered
1924          * above, then h->drv[drv_index] will be NULL.
1925          */
1926         if (h->drv[drv_index] == NULL) {
1927                 drvinfo->device_initialized = 0;
1928                 h->drv[drv_index] = drvinfo;
1929                 drvinfo = NULL; /* so it won't be freed below. */
1930         } else {
1931                 /* special case for cxd0 */
1932                 h->drv[drv_index]->block_size = drvinfo->block_size;
1933                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
1934                 h->drv[drv_index]->heads = drvinfo->heads;
1935                 h->drv[drv_index]->sectors = drvinfo->sectors;
1936                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
1937                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
1938                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
1939                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
1940                         VENDOR_LEN + 1);
1941                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
1942                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
1943         }
1944
1945         ++h->num_luns;
1946         disk = h->gendisk[drv_index];
1947         set_capacity(disk, h->drv[drv_index]->nr_blocks);
1948
1949         /* If it's not disk 0 (drv_index != 0)
1950          * or if it was disk 0, but there was previously
1951          * no actual corresponding configured logical drive
1952          * (raid_leve == -1) then we want to update the
1953          * logical drive's information.
1954          */
1955         if (drv_index || first_time) {
1956                 if (cciss_add_disk(h, disk, drv_index) != 0) {
1957                         cciss_free_gendisk(h, drv_index);
1958                         cciss_free_drive_info(h, drv_index);
1959                         printk(KERN_WARNING "cciss:%d could not update "
1960                                 "disk %d\n", h->ctlr, drv_index);
1961                         --h->num_luns;
1962                 }
1963         }
1964
1965 freeret:
1966         kfree(inq_buff);
1967         kfree(drvinfo);
1968         return;
1969 mem_msg:
1970         printk(KERN_ERR "cciss: out of memory\n");
1971         goto freeret;
1972 }
1973
1974 /* This function will find the first index of the controllers drive array
1975  * that has a null drv pointer and allocate the drive info struct and
1976  * will return that index   This is where new drives will be added.
1977  * If the index to be returned is greater than the highest_lun index for
1978  * the controller then highest_lun is set * to this new index.
1979  * If there are no available indexes or if tha allocation fails, then -1
1980  * is returned.  * "controller_node" is used to know if this is a real
1981  * logical drive, or just the controller node, which determines if this
1982  * counts towards highest_lun.
1983  */
1984 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
1985 {
1986         int i;
1987         drive_info_struct *drv;
1988
1989         /* Search for an empty slot for our drive info */
1990         for (i = 0; i < CISS_MAX_LUN; i++) {
1991
1992                 /* if not cxd0 case, and it's occupied, skip it. */
1993                 if (h->drv[i] && i != 0)
1994                         continue;
1995                 /*
1996                  * If it's cxd0 case, and drv is alloc'ed already, and a
1997                  * disk is configured there, skip it.
1998                  */
1999                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2000                         continue;
2001
2002                 /*
2003                  * We've found an empty slot.  Update highest_lun
2004                  * provided this isn't just the fake cxd0 controller node.
2005                  */
2006                 if (i > h->highest_lun && !controller_node)
2007                         h->highest_lun = i;
2008
2009                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2010                 if (i == 0 && h->drv[i] != NULL)
2011                         return i;
2012
2013                 /*
2014                  * Found an empty slot, not already alloc'ed.  Allocate it.
2015                  * Mark it with raid_level == -1, so we know it's new later on.
2016                  */
2017                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2018                 if (!drv)
2019                         return -1;
2020                 drv->raid_level = -1; /* so we know it's new */
2021                 h->drv[i] = drv;
2022                 return i;
2023         }
2024         return -1;
2025 }
2026
2027 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2028 {
2029         kfree(h->drv[drv_index]);
2030         h->drv[drv_index] = NULL;
2031 }
2032
2033 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2034 {
2035         put_disk(h->gendisk[drv_index]);
2036         h->gendisk[drv_index] = NULL;
2037 }
2038
2039 /* cciss_add_gendisk finds a free hba[]->drv structure
2040  * and allocates a gendisk if needed, and sets the lunid
2041  * in the drvinfo structure.   It returns the index into
2042  * the ->drv[] array, or -1 if none are free.
2043  * is_controller_node indicates whether highest_lun should
2044  * count this disk, or if it's only being added to provide
2045  * a means to talk to the controller in case no logical
2046  * drives have yet been configured.
2047  */
2048 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2049         int controller_node)
2050 {
2051         int drv_index;
2052
2053         drv_index = cciss_alloc_drive_info(h, controller_node);
2054         if (drv_index == -1)
2055                 return -1;
2056
2057         /*Check if the gendisk needs to be allocated */
2058         if (!h->gendisk[drv_index]) {
2059                 h->gendisk[drv_index] =
2060                         alloc_disk(1 << NWD_SHIFT);
2061                 if (!h->gendisk[drv_index]) {
2062                         printk(KERN_ERR "cciss%d: could not "
2063                                 "allocate a new disk %d\n",
2064                                 h->ctlr, drv_index);
2065                         goto err_free_drive_info;
2066                 }
2067         }
2068         memcpy(h->drv[drv_index]->LunID, lunid,
2069                 sizeof(h->drv[drv_index]->LunID));
2070         if (cciss_create_ld_sysfs_entry(h, drv_index))
2071                 goto err_free_disk;
2072         /* Don't need to mark this busy because nobody */
2073         /* else knows about this disk yet to contend */
2074         /* for access to it. */
2075         h->drv[drv_index]->busy_configuring = 0;
2076         wmb();
2077         return drv_index;
2078
2079 err_free_disk:
2080         cciss_free_gendisk(h, drv_index);
2081 err_free_drive_info:
2082         cciss_free_drive_info(h, drv_index);
2083         return -1;
2084 }
2085
2086 /* This is for the special case of a controller which
2087  * has no logical drives.  In this case, we still need
2088  * to register a disk so the controller can be accessed
2089  * by the Array Config Utility.
2090  */
2091 static void cciss_add_controller_node(ctlr_info_t *h)
2092 {
2093         struct gendisk *disk;
2094         int drv_index;
2095
2096         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2097                 return;
2098
2099         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2100         if (drv_index == -1)
2101                 goto error;
2102         h->drv[drv_index]->block_size = 512;
2103         h->drv[drv_index]->nr_blocks = 0;
2104         h->drv[drv_index]->heads = 0;
2105         h->drv[drv_index]->sectors = 0;
2106         h->drv[drv_index]->cylinders = 0;
2107         h->drv[drv_index]->raid_level = -1;
2108         memset(h->drv[drv_index]->serial_no, 0, 16);
2109         disk = h->gendisk[drv_index];
2110         if (cciss_add_disk(h, disk, drv_index) == 0)
2111                 return;
2112         cciss_free_gendisk(h, drv_index);
2113         cciss_free_drive_info(h, drv_index);
2114 error:
2115         printk(KERN_WARNING "cciss%d: could not "
2116                 "add disk 0.\n", h->ctlr);
2117         return;
2118 }
2119
2120 /* This function will add and remove logical drives from the Logical
2121  * drive array of the controller and maintain persistency of ordering
2122  * so that mount points are preserved until the next reboot.  This allows
2123  * for the removal of logical drives in the middle of the drive array
2124  * without a re-ordering of those drives.
2125  * INPUT
2126  * h            = The controller to perform the operations on
2127  */
2128 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2129         int via_ioctl)
2130 {
2131         int ctlr = h->ctlr;
2132         int num_luns;
2133         ReportLunData_struct *ld_buff = NULL;
2134         int return_code;
2135         int listlength = 0;
2136         int i;
2137         int drv_found;
2138         int drv_index = 0;
2139         unsigned char lunid[8] = CTLR_LUNID;
2140         unsigned long flags;
2141
2142         if (!capable(CAP_SYS_RAWIO))
2143                 return -EPERM;
2144
2145         /* Set busy_configuring flag for this operation */
2146         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2147         if (h->busy_configuring) {
2148                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2149                 return -EBUSY;
2150         }
2151         h->busy_configuring = 1;
2152         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2153
2154         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2155         if (ld_buff == NULL)
2156                 goto mem_msg;
2157
2158         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2159                                       sizeof(ReportLunData_struct),
2160                                       0, CTLR_LUNID, TYPE_CMD);
2161
2162         if (return_code == IO_OK)
2163                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2164         else {  /* reading number of logical volumes failed */
2165                 printk(KERN_WARNING "cciss: report logical volume"
2166                        " command failed\n");
2167                 listlength = 0;
2168                 goto freeret;
2169         }
2170
2171         num_luns = listlength / 8;      /* 8 bytes per entry */
2172         if (num_luns > CISS_MAX_LUN) {
2173                 num_luns = CISS_MAX_LUN;
2174                 printk(KERN_WARNING "cciss: more luns configured"
2175                        " on controller than can be handled by"
2176                        " this driver.\n");
2177         }
2178
2179         if (num_luns == 0)
2180                 cciss_add_controller_node(h);
2181
2182         /* Compare controller drive array to driver's drive array
2183          * to see if any drives are missing on the controller due
2184          * to action of Array Config Utility (user deletes drive)
2185          * and deregister logical drives which have disappeared.
2186          */
2187         for (i = 0; i <= h->highest_lun; i++) {
2188                 int j;
2189                 drv_found = 0;
2190
2191                 /* skip holes in the array from already deleted drives */
2192                 if (h->drv[i] == NULL)
2193                         continue;
2194
2195                 for (j = 0; j < num_luns; j++) {
2196                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2197                         if (memcmp(h->drv[i]->LunID, lunid,
2198                                 sizeof(lunid)) == 0) {
2199                                 drv_found = 1;
2200                                 break;
2201                         }
2202                 }
2203                 if (!drv_found) {
2204                         /* Deregister it from the OS, it's gone. */
2205                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2206                         h->drv[i]->busy_configuring = 1;
2207                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2208                         return_code = deregister_disk(h, i, 1, via_ioctl);
2209                         if (h->drv[i] != NULL)
2210                                 h->drv[i]->busy_configuring = 0;
2211                 }
2212         }
2213
2214         /* Compare controller drive array to driver's drive array.
2215          * Check for updates in the drive information and any new drives
2216          * on the controller due to ACU adding logical drives, or changing
2217          * a logical drive's size, etc.  Reregister any new/changed drives
2218          */
2219         for (i = 0; i < num_luns; i++) {
2220                 int j;
2221
2222                 drv_found = 0;
2223
2224                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2225                 /* Find if the LUN is already in the drive array
2226                  * of the driver.  If so then update its info
2227                  * if not in use.  If it does not exist then find
2228                  * the first free index and add it.
2229                  */
2230                 for (j = 0; j <= h->highest_lun; j++) {
2231                         if (h->drv[j] != NULL &&
2232                                 memcmp(h->drv[j]->LunID, lunid,
2233                                         sizeof(h->drv[j]->LunID)) == 0) {
2234                                 drv_index = j;
2235                                 drv_found = 1;
2236                                 break;
2237                         }
2238                 }
2239
2240                 /* check if the drive was found already in the array */
2241                 if (!drv_found) {
2242                         drv_index = cciss_add_gendisk(h, lunid, 0);
2243                         if (drv_index == -1)
2244                                 goto freeret;
2245                 }
2246                 cciss_update_drive_info(ctlr, drv_index, first_time,
2247                         via_ioctl);
2248         }               /* end for */
2249
2250 freeret:
2251         kfree(ld_buff);
2252         h->busy_configuring = 0;
2253         /* We return -1 here to tell the ACU that we have registered/updated
2254          * all of the drives that we can and to keep it from calling us
2255          * additional times.
2256          */
2257         return -1;
2258 mem_msg:
2259         printk(KERN_ERR "cciss: out of memory\n");
2260         h->busy_configuring = 0;
2261         goto freeret;
2262 }
2263
2264 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2265 {
2266         /* zero out the disk size info */
2267         drive_info->nr_blocks = 0;
2268         drive_info->block_size = 0;
2269         drive_info->heads = 0;
2270         drive_info->sectors = 0;
2271         drive_info->cylinders = 0;
2272         drive_info->raid_level = -1;
2273         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2274         memset(drive_info->model, 0, sizeof(drive_info->model));
2275         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2276         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2277         /*
2278          * don't clear the LUNID though, we need to remember which
2279          * one this one is.
2280          */
2281 }
2282
2283 /* This function will deregister the disk and it's queue from the
2284  * kernel.  It must be called with the controller lock held and the
2285  * drv structures busy_configuring flag set.  It's parameters are:
2286  *
2287  * disk = This is the disk to be deregistered
2288  * drv  = This is the drive_info_struct associated with the disk to be
2289  *        deregistered.  It contains information about the disk used
2290  *        by the driver.
2291  * clear_all = This flag determines whether or not the disk information
2292  *             is going to be completely cleared out and the highest_lun
2293  *             reset.  Sometimes we want to clear out information about
2294  *             the disk in preparation for re-adding it.  In this case
2295  *             the highest_lun should be left unchanged and the LunID
2296  *             should not be cleared.
2297  * via_ioctl
2298  *    This indicates whether we've reached this path via ioctl.
2299  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2300  *    If this path is reached via ioctl(), then the max_usage_count will
2301  *    be 1, as the process calling ioctl() has got to have the device open.
2302  *    If we get here via sysfs, then the max usage count will be zero.
2303 */
2304 static int deregister_disk(ctlr_info_t *h, int drv_index,
2305                            int clear_all, int via_ioctl)
2306 {
2307         int i;
2308         struct gendisk *disk;
2309         drive_info_struct *drv;
2310         int recalculate_highest_lun;
2311
2312         if (!capable(CAP_SYS_RAWIO))
2313                 return -EPERM;
2314
2315         drv = h->drv[drv_index];
2316         disk = h->gendisk[drv_index];
2317
2318         /* make sure logical volume is NOT is use */
2319         if (clear_all || (h->gendisk[0] == disk)) {
2320                 if (drv->usage_count > via_ioctl)
2321                         return -EBUSY;
2322         } else if (drv->usage_count > 0)
2323                 return -EBUSY;
2324
2325         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2326
2327         /* invalidate the devices and deregister the disk.  If it is disk
2328          * zero do not deregister it but just zero out it's values.  This
2329          * allows us to delete disk zero but keep the controller registered.
2330          */
2331         if (h->gendisk[0] != disk) {
2332                 struct request_queue *q = disk->queue;
2333                 if (disk->flags & GENHD_FL_UP) {
2334                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2335                         del_gendisk(disk);
2336                 }
2337                 if (q)
2338                         blk_cleanup_queue(q);
2339                 /* If clear_all is set then we are deleting the logical
2340                  * drive, not just refreshing its info.  For drives
2341                  * other than disk 0 we will call put_disk.  We do not
2342                  * do this for disk 0 as we need it to be able to
2343                  * configure the controller.
2344                  */
2345                 if (clear_all){
2346                         /* This isn't pretty, but we need to find the
2347                          * disk in our array and NULL our the pointer.
2348                          * This is so that we will call alloc_disk if
2349                          * this index is used again later.
2350                          */
2351                         for (i=0; i < CISS_MAX_LUN; i++){
2352                                 if (h->gendisk[i] == disk) {
2353                                         h->gendisk[i] = NULL;
2354                                         break;
2355                                 }
2356                         }
2357                         put_disk(disk);
2358                 }
2359         } else {
2360                 set_capacity(disk, 0);
2361                 cciss_clear_drive_info(drv);
2362         }
2363
2364         --h->num_luns;
2365
2366         /* if it was the last disk, find the new hightest lun */
2367         if (clear_all && recalculate_highest_lun) {
2368                 int i, newhighest = -1;
2369                 for (i = 0; i <= h->highest_lun; i++) {
2370                         /* if the disk has size > 0, it is available */
2371                         if (h->drv[i] && h->drv[i]->heads)
2372                                 newhighest = i;
2373                 }
2374                 h->highest_lun = newhighest;
2375         }
2376         return 0;
2377 }
2378
2379 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2380                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2381                 int cmd_type)
2382 {
2383         ctlr_info_t *h = hba[ctlr];
2384         u64bit buff_dma_handle;
2385         int status = IO_OK;
2386
2387         c->cmd_type = CMD_IOCTL_PEND;
2388         c->Header.ReplyQueue = 0;
2389         if (buff != NULL) {
2390                 c->Header.SGList = 1;
2391                 c->Header.SGTotal = 1;
2392         } else {
2393                 c->Header.SGList = 0;
2394                 c->Header.SGTotal = 0;
2395         }
2396         c->Header.Tag.lower = c->busaddr;
2397         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2398
2399         c->Request.Type.Type = cmd_type;
2400         if (cmd_type == TYPE_CMD) {
2401                 switch (cmd) {
2402                 case CISS_INQUIRY:
2403                         /* are we trying to read a vital product page */
2404                         if (page_code != 0) {
2405                                 c->Request.CDB[1] = 0x01;
2406                                 c->Request.CDB[2] = page_code;
2407                         }
2408                         c->Request.CDBLen = 6;
2409                         c->Request.Type.Attribute = ATTR_SIMPLE;
2410                         c->Request.Type.Direction = XFER_READ;
2411                         c->Request.Timeout = 0;
2412                         c->Request.CDB[0] = CISS_INQUIRY;
2413                         c->Request.CDB[4] = size & 0xFF;
2414                         break;
2415                 case CISS_REPORT_LOG:
2416                 case CISS_REPORT_PHYS:
2417                         /* Talking to controller so It's a physical command
2418                            mode = 00 target = 0.  Nothing to write.
2419                          */
2420                         c->Request.CDBLen = 12;
2421                         c->Request.Type.Attribute = ATTR_SIMPLE;
2422                         c->Request.Type.Direction = XFER_READ;
2423                         c->Request.Timeout = 0;
2424                         c->Request.CDB[0] = cmd;
2425                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
2426                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2427                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2428                         c->Request.CDB[9] = size & 0xFF;
2429                         break;
2430
2431                 case CCISS_READ_CAPACITY:
2432                         c->Request.CDBLen = 10;
2433                         c->Request.Type.Attribute = ATTR_SIMPLE;
2434                         c->Request.Type.Direction = XFER_READ;
2435                         c->Request.Timeout = 0;
2436                         c->Request.CDB[0] = cmd;
2437                         break;
2438                 case CCISS_READ_CAPACITY_16:
2439                         c->Request.CDBLen = 16;
2440                         c->Request.Type.Attribute = ATTR_SIMPLE;
2441                         c->Request.Type.Direction = XFER_READ;
2442                         c->Request.Timeout = 0;
2443                         c->Request.CDB[0] = cmd;
2444                         c->Request.CDB[1] = 0x10;
2445                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2446                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2447                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2448                         c->Request.CDB[13] = size & 0xFF;
2449                         c->Request.Timeout = 0;
2450                         c->Request.CDB[0] = cmd;
2451                         break;
2452                 case CCISS_CACHE_FLUSH:
2453                         c->Request.CDBLen = 12;
2454                         c->Request.Type.Attribute = ATTR_SIMPLE;
2455                         c->Request.Type.Direction = XFER_WRITE;
2456                         c->Request.Timeout = 0;
2457                         c->Request.CDB[0] = BMIC_WRITE;
2458                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2459                         break;
2460                 case TEST_UNIT_READY:
2461                         c->Request.CDBLen = 6;
2462                         c->Request.Type.Attribute = ATTR_SIMPLE;
2463                         c->Request.Type.Direction = XFER_NONE;
2464                         c->Request.Timeout = 0;
2465                         break;
2466                 default:
2467                         printk(KERN_WARNING
2468                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2469                         return IO_ERROR;
2470                 }
2471         } else if (cmd_type == TYPE_MSG) {
2472                 switch (cmd) {
2473                 case 0: /* ABORT message */
2474                         c->Request.CDBLen = 12;
2475                         c->Request.Type.Attribute = ATTR_SIMPLE;
2476                         c->Request.Type.Direction = XFER_WRITE;
2477                         c->Request.Timeout = 0;
2478                         c->Request.CDB[0] = cmd;        /* abort */
2479                         c->Request.CDB[1] = 0;  /* abort a command */
2480                         /* buff contains the tag of the command to abort */
2481                         memcpy(&c->Request.CDB[4], buff, 8);
2482                         break;
2483                 case 1: /* RESET message */
2484                         c->Request.CDBLen = 16;
2485                         c->Request.Type.Attribute = ATTR_SIMPLE;
2486                         c->Request.Type.Direction = XFER_NONE;
2487                         c->Request.Timeout = 0;
2488                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2489                         c->Request.CDB[0] = cmd;        /* reset */
2490                         c->Request.CDB[1] = 0x03;       /* reset a target */
2491                         break;
2492                 case 3: /* No-Op message */
2493                         c->Request.CDBLen = 1;
2494                         c->Request.Type.Attribute = ATTR_SIMPLE;
2495                         c->Request.Type.Direction = XFER_WRITE;
2496                         c->Request.Timeout = 0;
2497                         c->Request.CDB[0] = cmd;
2498                         break;
2499                 default:
2500                         printk(KERN_WARNING
2501                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2502                         return IO_ERROR;
2503                 }
2504         } else {
2505                 printk(KERN_WARNING
2506                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2507                 return IO_ERROR;
2508         }
2509         /* Fill in the scatter gather information */
2510         if (size > 0) {
2511                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2512                                                              buff, size,
2513                                                              PCI_DMA_BIDIRECTIONAL);
2514                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2515                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2516                 c->SG[0].Len = size;
2517                 c->SG[0].Ext = 0;       /* we are not chaining */
2518         }
2519         return status;
2520 }
2521
2522 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2523 {
2524         switch (c->err_info->ScsiStatus) {
2525         case SAM_STAT_GOOD:
2526                 return IO_OK;
2527         case SAM_STAT_CHECK_CONDITION:
2528                 switch (0xf & c->err_info->SenseInfo[2]) {
2529                 case 0: return IO_OK; /* no sense */
2530                 case 1: return IO_OK; /* recovered error */
2531                 default:
2532                         if (check_for_unit_attention(h, c))
2533                                 return IO_NEEDS_RETRY;
2534                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2535                                 "check condition, sense key = 0x%02x\n",
2536                                 h->ctlr, c->Request.CDB[0],
2537                                 c->err_info->SenseInfo[2]);
2538                 }
2539                 break;
2540         default:
2541                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2542                         "scsi status = 0x%02x\n", h->ctlr,
2543                         c->Request.CDB[0], c->err_info->ScsiStatus);
2544                 break;
2545         }
2546         return IO_ERROR;
2547 }
2548
2549 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2550 {
2551         int return_status = IO_OK;
2552
2553         if (c->err_info->CommandStatus == CMD_SUCCESS)
2554                 return IO_OK;
2555
2556         switch (c->err_info->CommandStatus) {
2557         case CMD_TARGET_STATUS:
2558                 return_status = check_target_status(h, c);
2559                 break;
2560         case CMD_DATA_UNDERRUN:
2561         case CMD_DATA_OVERRUN:
2562                 /* expected for inquiry and report lun commands */
2563                 break;
2564         case CMD_INVALID:
2565                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2566                        "reported invalid\n", c->Request.CDB[0]);
2567                 return_status = IO_ERROR;
2568                 break;
2569         case CMD_PROTOCOL_ERR:
2570                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2571                        "protocol error \n", c->Request.CDB[0]);
2572                 return_status = IO_ERROR;
2573                 break;
2574         case CMD_HARDWARE_ERR:
2575                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2576                        " hardware error\n", c->Request.CDB[0]);
2577                 return_status = IO_ERROR;
2578                 break;
2579         case CMD_CONNECTION_LOST:
2580                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2581                        "connection lost\n", c->Request.CDB[0]);
2582                 return_status = IO_ERROR;
2583                 break;
2584         case CMD_ABORTED:
2585                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2586                        "aborted\n", c->Request.CDB[0]);
2587                 return_status = IO_ERROR;
2588                 break;
2589         case CMD_ABORT_FAILED:
2590                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2591                        "abort failed\n", c->Request.CDB[0]);
2592                 return_status = IO_ERROR;
2593                 break;
2594         case CMD_UNSOLICITED_ABORT:
2595                 printk(KERN_WARNING
2596                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2597                         c->Request.CDB[0]);
2598                 return_status = IO_NEEDS_RETRY;
2599                 break;
2600         default:
2601                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2602                        "unknown status %x\n", c->Request.CDB[0],
2603                        c->err_info->CommandStatus);
2604                 return_status = IO_ERROR;
2605         }
2606         return return_status;
2607 }
2608
2609 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2610         int attempt_retry)
2611 {
2612         DECLARE_COMPLETION_ONSTACK(wait);
2613         u64bit buff_dma_handle;
2614         unsigned long flags;
2615         int return_status = IO_OK;
2616
2617 resend_cmd2:
2618         c->waiting = &wait;
2619         /* Put the request on the tail of the queue and send it */
2620         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2621         addQ(&h->reqQ, c);
2622         h->Qdepth++;
2623         start_io(h);
2624         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2625
2626         wait_for_completion(&wait);
2627
2628         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2629                 goto command_done;
2630
2631         return_status = process_sendcmd_error(h, c);
2632
2633         if (return_status == IO_NEEDS_RETRY &&
2634                 c->retry_count < MAX_CMD_RETRIES) {
2635                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2636                         c->Request.CDB[0]);
2637                 c->retry_count++;
2638                 /* erase the old error information */
2639                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2640                 return_status = IO_OK;
2641                 INIT_COMPLETION(wait);
2642                 goto resend_cmd2;
2643         }
2644
2645 command_done:
2646         /* unlock the buffers from DMA */
2647         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2648         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2649         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2650                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2651         return return_status;
2652 }
2653
2654 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2655                            __u8 page_code, unsigned char scsi3addr[],
2656                         int cmd_type)
2657 {
2658         ctlr_info_t *h = hba[ctlr];
2659         CommandList_struct *c;
2660         int return_status;
2661
2662         c = cmd_alloc(h, 0);
2663         if (!c)
2664                 return -ENOMEM;
2665         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2666                 scsi3addr, cmd_type);
2667         if (return_status == IO_OK)
2668                 return_status = sendcmd_withirq_core(h, c, 1);
2669
2670         cmd_free(h, c, 0);
2671         return return_status;
2672 }
2673
2674 static void cciss_geometry_inquiry(int ctlr, int logvol,
2675                                    sector_t total_size,
2676                                    unsigned int block_size,
2677                                    InquiryData_struct *inq_buff,
2678                                    drive_info_struct *drv)
2679 {
2680         int return_code;
2681         unsigned long t;
2682         unsigned char scsi3addr[8];
2683
2684         memset(inq_buff, 0, sizeof(InquiryData_struct));
2685         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2686         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2687                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2688         if (return_code == IO_OK) {
2689                 if (inq_buff->data_byte[8] == 0xFF) {
2690                         printk(KERN_WARNING
2691                                "cciss: reading geometry failed, volume "
2692                                "does not support reading geometry\n");
2693                         drv->heads = 255;
2694                         drv->sectors = 32;      // Sectors per track
2695                         drv->cylinders = total_size + 1;
2696                         drv->raid_level = RAID_UNKNOWN;
2697                 } else {
2698                         drv->heads = inq_buff->data_byte[6];
2699                         drv->sectors = inq_buff->data_byte[7];
2700                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2701                         drv->cylinders += inq_buff->data_byte[5];
2702                         drv->raid_level = inq_buff->data_byte[8];
2703                 }
2704                 drv->block_size = block_size;
2705                 drv->nr_blocks = total_size + 1;
2706                 t = drv->heads * drv->sectors;
2707                 if (t > 1) {
2708                         sector_t real_size = total_size + 1;
2709                         unsigned long rem = sector_div(real_size, t);
2710                         if (rem)
2711                                 real_size++;
2712                         drv->cylinders = real_size;
2713                 }
2714         } else {                /* Get geometry failed */
2715                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2716         }
2717 }
2718
2719 static void
2720 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2721                     unsigned int *block_size)
2722 {
2723         ReadCapdata_struct *buf;
2724         int return_code;
2725         unsigned char scsi3addr[8];
2726
2727         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2728         if (!buf) {
2729                 printk(KERN_WARNING "cciss: out of memory\n");
2730                 return;
2731         }
2732
2733         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2734         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2735                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2736         if (return_code == IO_OK) {
2737                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2738                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2739         } else {                /* read capacity command failed */
2740                 printk(KERN_WARNING "cciss: read capacity failed\n");
2741                 *total_size = 0;
2742                 *block_size = BLOCK_SIZE;
2743         }
2744         kfree(buf);
2745 }
2746
2747 static void cciss_read_capacity_16(int ctlr, int logvol,
2748         sector_t *total_size, unsigned int *block_size)
2749 {
2750         ReadCapdata_struct_16 *buf;
2751         int return_code;
2752         unsigned char scsi3addr[8];
2753
2754         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2755         if (!buf) {
2756                 printk(KERN_WARNING "cciss: out of memory\n");
2757                 return;
2758         }
2759
2760         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2761         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2762                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2763                         0, scsi3addr, TYPE_CMD);
2764         if (return_code == IO_OK) {
2765                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2766                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2767         } else {                /* read capacity command failed */
2768                 printk(KERN_WARNING "cciss: read capacity failed\n");
2769                 *total_size = 0;
2770                 *block_size = BLOCK_SIZE;
2771         }
2772         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2773                (unsigned long long)*total_size+1, *block_size);
2774         kfree(buf);
2775 }
2776
2777 static int cciss_revalidate(struct gendisk *disk)
2778 {
2779         ctlr_info_t *h = get_host(disk);
2780         drive_info_struct *drv = get_drv(disk);
2781         int logvol;
2782         int FOUND = 0;
2783         unsigned int block_size;
2784         sector_t total_size;
2785         InquiryData_struct *inq_buff = NULL;
2786
2787         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2788                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2789                         sizeof(drv->LunID)) == 0) {
2790                         FOUND = 1;
2791                         break;
2792                 }
2793         }
2794
2795         if (!FOUND)
2796                 return 1;
2797
2798         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2799         if (inq_buff == NULL) {
2800                 printk(KERN_WARNING "cciss: out of memory\n");
2801                 return 1;
2802         }
2803         if (h->cciss_read == CCISS_READ_10) {
2804                 cciss_read_capacity(h->ctlr, logvol,
2805                                         &total_size, &block_size);
2806         } else {
2807                 cciss_read_capacity_16(h->ctlr, logvol,
2808                                         &total_size, &block_size);
2809         }
2810         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2811                                inq_buff, drv);
2812
2813         blk_queue_logical_block_size(drv->queue, drv->block_size);
2814         set_capacity(disk, drv->nr_blocks);
2815
2816         kfree(inq_buff);
2817         return 0;
2818 }
2819
2820 /*
2821  * Map (physical) PCI mem into (virtual) kernel space
2822  */
2823 static void __iomem *remap_pci_mem(ulong base, ulong size)
2824 {
2825         ulong page_base = ((ulong) base) & PAGE_MASK;
2826         ulong page_offs = ((ulong) base) - page_base;
2827         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2828
2829         return page_remapped ? (page_remapped + page_offs) : NULL;
2830 }
2831
2832 /*
2833  * Takes jobs of the Q and sends them to the hardware, then puts it on
2834  * the Q to wait for completion.
2835  */
2836 static void start_io(ctlr_info_t *h)
2837 {
2838         CommandList_struct *c;
2839
2840         while (!hlist_empty(&h->reqQ)) {
2841                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2842                 /* can't do anything if fifo is full */
2843                 if ((h->access.fifo_full(h))) {
2844                         printk(KERN_WARNING "cciss: fifo full\n");
2845                         break;
2846                 }
2847
2848                 /* Get the first entry from the Request Q */
2849                 removeQ(c);
2850                 h->Qdepth--;
2851
2852                 /* Tell the controller execute command */
2853                 h->access.submit_command(h, c);
2854
2855                 /* Put job onto the completed Q */
2856                 addQ(&h->cmpQ, c);
2857         }
2858 }
2859
2860 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2861 /* Zeros out the error record and then resends the command back */
2862 /* to the controller */
2863 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2864 {
2865         /* erase the old error information */
2866         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2867
2868         /* add it to software queue and then send it to the controller */
2869         addQ(&h->reqQ, c);
2870         h->Qdepth++;
2871         if (h->Qdepth > h->maxQsinceinit)
2872                 h->maxQsinceinit = h->Qdepth;
2873
2874         start_io(h);
2875 }
2876
2877 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2878         unsigned int msg_byte, unsigned int host_byte,
2879         unsigned int driver_byte)
2880 {
2881         /* inverse of macros in scsi.h */
2882         return (scsi_status_byte & 0xff) |
2883                 ((msg_byte & 0xff) << 8) |
2884                 ((host_byte & 0xff) << 16) |
2885                 ((driver_byte & 0xff) << 24);
2886 }
2887
2888 static inline int evaluate_target_status(ctlr_info_t *h,
2889                         CommandList_struct *cmd, int *retry_cmd)
2890 {
2891         unsigned char sense_key;
2892         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2893         int error_value;
2894
2895         *retry_cmd = 0;
2896         /* If we get in here, it means we got "target status", that is, scsi status */
2897         status_byte = cmd->err_info->ScsiStatus;
2898         driver_byte = DRIVER_OK;
2899         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2900
2901         if (blk_pc_request(cmd->rq))
2902                 host_byte = DID_PASSTHROUGH;
2903         else
2904                 host_byte = DID_OK;
2905
2906         error_value = make_status_bytes(status_byte, msg_byte,
2907                 host_byte, driver_byte);
2908
2909         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2910                 if (!blk_pc_request(cmd->rq))
2911                         printk(KERN_WARNING "cciss: cmd %p "
2912                                "has SCSI Status 0x%x\n",
2913                                cmd, cmd->err_info->ScsiStatus);
2914                 return error_value;
2915         }
2916
2917         /* check the sense key */
2918         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2919         /* no status or recovered error */
2920         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
2921                 error_value = 0;
2922
2923         if (check_for_unit_attention(h, cmd)) {
2924                 *retry_cmd = !blk_pc_request(cmd->rq);
2925                 return 0;
2926         }
2927
2928         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
2929                 if (error_value != 0)
2930                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
2931                                " sense key = 0x%x\n", cmd, sense_key);
2932                 return error_value;
2933         }
2934
2935         /* SG_IO or similar, copy sense data back */
2936         if (cmd->rq->sense) {
2937                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2938                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2939                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2940                         cmd->rq->sense_len);
2941         } else
2942                 cmd->rq->sense_len = 0;
2943
2944         return error_value;
2945 }
2946
2947 /* checks the status of the job and calls complete buffers to mark all
2948  * buffers for the completed job. Note that this function does not need
2949  * to hold the hba/queue lock.
2950  */
2951 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
2952                                     int timeout)
2953 {
2954         int retry_cmd = 0;
2955         struct request *rq = cmd->rq;
2956
2957         rq->errors = 0;
2958
2959         if (timeout)
2960                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
2961
2962         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
2963                 goto after_error_processing;
2964
2965         switch (cmd->err_info->CommandStatus) {
2966         case CMD_TARGET_STATUS:
2967                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
2968                 break;
2969         case CMD_DATA_UNDERRUN:
2970                 if (blk_fs_request(cmd->rq)) {
2971                         printk(KERN_WARNING "cciss: cmd %p has"
2972                                " completed with data underrun "
2973                                "reported\n", cmd);
2974                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
2975                 }
2976                 break;
2977         case CMD_DATA_OVERRUN:
2978                 if (blk_fs_request(cmd->rq))
2979                         printk(KERN_WARNING "cciss: cmd %p has"
2980                                " completed with data overrun "
2981                                "reported\n", cmd);
2982                 break;
2983         case CMD_INVALID:
2984                 printk(KERN_WARNING "cciss: cmd %p is "
2985                        "reported invalid\n", cmd);
2986                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2987                         cmd->err_info->CommandStatus, DRIVER_OK,
2988                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2989                 break;
2990         case CMD_PROTOCOL_ERR:
2991                 printk(KERN_WARNING "cciss: cmd %p has "
2992                        "protocol error \n", cmd);
2993                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
2994                         cmd->err_info->CommandStatus, DRIVER_OK,
2995                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
2996                 break;
2997         case CMD_HARDWARE_ERR:
2998                 printk(KERN_WARNING "cciss: cmd %p had "
2999                        " hardware error\n", cmd);
3000                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3001                         cmd->err_info->CommandStatus, DRIVER_OK,
3002                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3003                 break;
3004         case CMD_CONNECTION_LOST:
3005                 printk(KERN_WARNING "cciss: cmd %p had "
3006                        "connection lost\n", cmd);
3007                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3008                         cmd->err_info->CommandStatus, DRIVER_OK,
3009                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3010                 break;
3011         case CMD_ABORTED:
3012                 printk(KERN_WARNING "cciss: cmd %p was "
3013                        "aborted\n", cmd);
3014                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3015                         cmd->err_info->CommandStatus, DRIVER_OK,
3016                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3017                 break;
3018         case CMD_ABORT_FAILED:
3019                 printk(KERN_WARNING "cciss: cmd %p reports "
3020                        "abort failed\n", cmd);
3021                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3022                         cmd->err_info->CommandStatus, DRIVER_OK,
3023                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3024                 break;
3025         case CMD_UNSOLICITED_ABORT:
3026                 printk(KERN_WARNING "cciss%d: unsolicited "
3027                        "abort %p\n", h->ctlr, cmd);
3028                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3029                         retry_cmd = 1;
3030                         printk(KERN_WARNING
3031                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3032                         cmd->retry_count++;
3033                 } else
3034                         printk(KERN_WARNING
3035                                "cciss%d: %p retried too "
3036                                "many times\n", h->ctlr, cmd);
3037                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3038                         cmd->err_info->CommandStatus, DRIVER_OK,
3039                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3040                 break;
3041         case CMD_TIMEOUT:
3042                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3043                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3044                         cmd->err_info->CommandStatus, DRIVER_OK,
3045                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3046                 break;
3047         default:
3048                 printk(KERN_WARNING "cciss: cmd %p returned "
3049                        "unknown status %x\n", cmd,
3050                        cmd->err_info->CommandStatus);
3051                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3052                         cmd->err_info->CommandStatus, DRIVER_OK,
3053                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3054         }
3055
3056 after_error_processing:
3057
3058         /* We need to return this command */
3059         if (retry_cmd) {
3060                 resend_cciss_cmd(h, cmd);
3061                 return;
3062         }
3063         cmd->rq->completion_data = cmd;
3064         blk_complete_request(cmd->rq);
3065 }
3066
3067 /*
3068  * Get a request and submit it to the controller.
3069  */
3070 static void do_cciss_request(struct request_queue *q)
3071 {
3072         ctlr_info_t *h = q->queuedata;
3073         CommandList_struct *c;
3074         sector_t start_blk;
3075         int seg;
3076         struct request *creq;
3077         u64bit temp64;
3078         struct scatterlist *tmp_sg;
3079         SGDescriptor_struct *curr_sg;
3080         drive_info_struct *drv;
3081         int i, dir;
3082         int nseg = 0;
3083         int sg_index = 0;
3084         int chained = 0;
3085
3086         /* We call start_io here in case there is a command waiting on the
3087          * queue that has not been sent.
3088          */
3089         if (blk_queue_plugged(q))
3090                 goto startio;
3091
3092       queue:
3093         creq = blk_peek_request(q);
3094         if (!creq)
3095                 goto startio;
3096
3097         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3098
3099         if ((c = cmd_alloc(h, 1)) == NULL)
3100                 goto full;
3101
3102         blk_start_request(creq);
3103
3104         tmp_sg = h->scatter_list[c->cmdindex];
3105         spin_unlock_irq(q->queue_lock);
3106
3107         c->cmd_type = CMD_RWREQ;
3108         c->rq = creq;
3109
3110         /* fill in the request */
3111         drv = creq->rq_disk->private_data;
3112         c->Header.ReplyQueue = 0;       // unused in simple mode
3113         /* got command from pool, so use the command block index instead */
3114         /* for direct lookups. */
3115         /* The first 2 bits are reserved for controller error reporting. */
3116         c->Header.Tag.lower = (c->cmdindex << 3);
3117         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
3118         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3119         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
3120         c->Request.Type.Type = TYPE_CMD;        // It is a command.
3121         c->Request.Type.Attribute = ATTR_SIMPLE;
3122         c->Request.Type.Direction =
3123             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3124         c->Request.Timeout = 0; // Don't time out
3125         c->Request.CDB[0] =
3126             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3127         start_blk = blk_rq_pos(creq);
3128 #ifdef CCISS_DEBUG
3129         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3130                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3131 #endif                          /* CCISS_DEBUG */
3132
3133         sg_init_table(tmp_sg, h->maxsgentries);
3134         seg = blk_rq_map_sg(q, creq, tmp_sg);
3135
3136         /* get the DMA records for the setup */
3137         if (c->Request.Type.Direction == XFER_READ)
3138                 dir = PCI_DMA_FROMDEVICE;
3139         else
3140                 dir = PCI_DMA_TODEVICE;
3141
3142         curr_sg = c->SG;
3143         sg_index = 0;
3144         chained = 0;
3145
3146         for (i = 0; i < seg; i++) {
3147                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3148                         !chained && ((seg - i) > 1)) {
3149                         nseg = seg - i;
3150                         curr_sg[sg_index].Len = (nseg) *
3151                                         sizeof(SGDescriptor_struct);
3152                         curr_sg[sg_index].Ext = CCISS_SG_CHAIN;
3153
3154                         /* Point to next chain block. */
3155                         curr_sg = h->cmd_sg_list[c->cmdindex]->sgchain;
3156                         sg_index = 0;
3157                         chained = 1;
3158                 }
3159                 curr_sg[sg_index].Len = tmp_sg[i].length;
3160                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3161                                                 tmp_sg[i].offset,
3162                                                 tmp_sg[i].length, dir);
3163                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3164                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3165                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3166
3167                 ++sg_index;
3168         }
3169
3170         if (chained) {
3171                 int len;
3172                 curr_sg = c->SG;
3173                 sg_index = h->max_cmd_sgentries - 1;
3174                 len = curr_sg[sg_index].Len;
3175                 /* Setup pointer to next chain block.
3176                  * Fill out last element in current chain
3177                  * block with address of next chain block.
3178                  */
3179                 temp64.val = pci_map_single(h->pdev,
3180                                         h->cmd_sg_list[c->cmdindex]->sgchain,
3181                                         len, dir);
3182
3183                 h->cmd_sg_list[c->cmdindex]->sg_chain_dma = temp64.val;
3184                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3185                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3186
3187                 pci_dma_sync_single_for_device(h->pdev,
3188                                 h->cmd_sg_list[c->cmdindex]->sg_chain_dma,
3189                                 len, dir);
3190         }
3191
3192         /* track how many SG entries we are using */
3193         if (seg > h->maxSG)
3194                 h->maxSG = seg;
3195
3196 #ifdef CCISS_DEBUG
3197         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3198                         "chained[%d]\n",
3199                         blk_rq_sectors(creq), seg, chained);
3200 #endif                          /* CCISS_DEBUG */
3201
3202         c->Header.SGList = c->Header.SGTotal = seg + chained;
3203         if (seg > h->max_cmd_sgentries)
3204                 c->Header.SGList = h->max_cmd_sgentries;
3205
3206         if (likely(blk_fs_request(creq))) {
3207                 if(h->cciss_read == CCISS_READ_10) {
3208                         c->Request.CDB[1] = 0;
3209                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
3210                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3211                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3212                         c->Request.CDB[5] = start_blk & 0xff;
3213                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
3214                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3215                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3216                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3217                 } else {
3218                         u32 upper32 = upper_32_bits(start_blk);
3219
3220                         c->Request.CDBLen = 16;
3221                         c->Request.CDB[1]= 0;
3222                         c->Request.CDB[2]= (upper32 >> 24) & 0xff;      //MSB
3223                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3224                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3225                         c->Request.CDB[5]= upper32 & 0xff;
3226                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3227                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3228                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3229                         c->Request.CDB[9]= start_blk & 0xff;
3230                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3231                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3232                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3233                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3234                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3235                 }
3236         } else if (blk_pc_request(creq)) {
3237                 c->Request.CDBLen = creq->cmd_len;
3238                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3239         } else {
3240                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3241                 BUG();
3242         }
3243
3244         spin_lock_irq(q->queue_lock);
3245
3246         addQ(&h->reqQ, c);
3247         h->Qdepth++;
3248         if (h->Qdepth > h->maxQsinceinit)
3249                 h->maxQsinceinit = h->Qdepth;
3250
3251         goto queue;
3252 full:
3253         blk_stop_queue(q);
3254 startio:
3255         /* We will already have the driver lock here so not need
3256          * to lock it.
3257          */
3258         start_io(h);
3259 }
3260
3261 static inline unsigned long get_next_completion(ctlr_info_t *h)
3262 {
3263         return h->access.command_completed(h);
3264 }
3265
3266 static inline int interrupt_pending(ctlr_info_t *h)
3267 {
3268         return h->access.intr_pending(h);
3269 }
3270
3271 static inline long interrupt_not_for_us(ctlr_info_t *h)
3272 {
3273         return (((h->access.intr_pending(h) == 0) ||
3274                  (h->interrupts_enabled == 0)));
3275 }
3276
3277 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3278 {
3279         ctlr_info_t *h = dev_id;
3280         CommandList_struct *c;
3281         unsigned long flags;
3282         __u32 a, a1, a2;
3283
3284         if (interrupt_not_for_us(h))
3285                 return IRQ_NONE;
3286         /*
3287          * If there are completed commands in the completion queue,
3288          * we had better do something about it.
3289          */
3290         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3291         while (interrupt_pending(h)) {
3292                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3293                         a1 = a;
3294                         if ((a & 0x04)) {
3295                                 a2 = (a >> 3);
3296                                 if (a2 >= h->nr_cmds) {
3297                                         printk(KERN_WARNING
3298                                                "cciss: controller cciss%d failed, stopping.\n",
3299                                                h->ctlr);
3300                                         fail_all_cmds(h->ctlr);
3301                                         return IRQ_HANDLED;
3302                                 }
3303
3304                                 c = h->cmd_pool + a2;
3305                                 a = c->busaddr;
3306
3307                         } else {
3308                                 struct hlist_node *tmp;
3309
3310                                 a &= ~3;
3311                                 c = NULL;
3312                                 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3313                                         if (c->busaddr == a)
3314                                                 break;
3315                                 }
3316                         }
3317                         /*
3318                          * If we've found the command, take it off the
3319                          * completion Q and free it
3320                          */
3321                         if (c && c->busaddr == a) {
3322                                 removeQ(c);
3323                                 if (c->cmd_type == CMD_RWREQ) {
3324                                         complete_command(h, c, 0);
3325                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3326                                         complete(c->waiting);
3327                                 }
3328 #                               ifdef CONFIG_CISS_SCSI_TAPE
3329                                 else if (c->cmd_type == CMD_SCSI)
3330                                         complete_scsi_command(c, 0, a1);
3331 #                               endif
3332                                 continue;
3333                         }
3334                 }
3335         }
3336
3337         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3338         return IRQ_HANDLED;
3339 }
3340
3341 /**
3342  * add_to_scan_list() - add controller to rescan queue
3343  * @h:                Pointer to the controller.
3344  *
3345  * Adds the controller to the rescan queue if not already on the queue.
3346  *
3347  * returns 1 if added to the queue, 0 if skipped (could be on the
3348  * queue already, or the controller could be initializing or shutting
3349  * down).
3350  **/
3351 static int add_to_scan_list(struct ctlr_info *h)
3352 {
3353         struct ctlr_info *test_h;
3354         int found = 0;
3355         int ret = 0;
3356
3357         if (h->busy_initializing)
3358                 return 0;
3359
3360         if (!mutex_trylock(&h->busy_shutting_down))
3361                 return 0;
3362
3363         mutex_lock(&scan_mutex);
3364         list_for_each_entry(test_h, &scan_q, scan_list) {
3365                 if (test_h == h) {
3366                         found = 1;
3367                         break;
3368                 }
3369         }
3370         if (!found && !h->busy_scanning) {
3371                 INIT_COMPLETION(h->scan_wait);
3372                 list_add_tail(&h->scan_list, &scan_q);
3373                 ret = 1;
3374         }
3375         mutex_unlock(&scan_mutex);
3376         mutex_unlock(&h->busy_shutting_down);
3377
3378         return ret;
3379 }
3380
3381 /**
3382  * remove_from_scan_list() - remove controller from rescan queue
3383  * @h:                     Pointer to the controller.
3384  *
3385  * Removes the controller from the rescan queue if present. Blocks if
3386  * the controller is currently conducting a rescan.  The controller
3387  * can be in one of three states:
3388  * 1. Doesn't need a scan
3389  * 2. On the scan list, but not scanning yet (we remove it)
3390  * 3. Busy scanning (and not on the list). In this case we want to wait for
3391  *    the scan to complete to make sure the scanning thread for this
3392  *    controller is completely idle.
3393  **/
3394 static void remove_from_scan_list(struct ctlr_info *h)
3395 {
3396         struct ctlr_info *test_h, *tmp_h;
3397
3398         mutex_lock(&scan_mutex);
3399         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3400                 if (test_h == h) { /* state 2. */
3401                         list_del(&h->scan_list);
3402                         complete_all(&h->scan_wait);
3403                         mutex_unlock(&scan_mutex);
3404                         return;
3405                 }
3406         }
3407         if (h->busy_scanning) { /* state 3. */
3408                 mutex_unlock(&scan_mutex);
3409                 wait_for_completion(&h->scan_wait);
3410         } else { /* state 1, nothing to do. */
3411                 mutex_unlock(&scan_mutex);
3412         }
3413 }
3414
3415 /**
3416  * scan_thread() - kernel thread used to rescan controllers
3417  * @data:        Ignored.
3418  *
3419  * A kernel thread used scan for drive topology changes on
3420  * controllers. The thread processes only one controller at a time
3421  * using a queue.  Controllers are added to the queue using
3422  * add_to_scan_list() and removed from the queue either after done
3423  * processing or using remove_from_scan_list().
3424  *
3425  * returns 0.
3426  **/
3427 static int scan_thread(void *data)
3428 {
3429         struct ctlr_info *h;
3430
3431         while (1) {
3432                 set_current_state(TASK_INTERRUPTIBLE);
3433                 schedule();
3434                 if (kthread_should_stop())
3435                         break;
3436
3437                 while (1) {
3438                         mutex_lock(&scan_mutex);
3439                         if (list_empty(&scan_q)) {
3440                                 mutex_unlock(&scan_mutex);
3441                                 break;
3442                         }
3443
3444                         h = list_entry(scan_q.next,
3445                                        struct ctlr_info,
3446                                        scan_list);
3447                         list_del(&h->scan_list);
3448                         h->busy_scanning = 1;
3449                         mutex_unlock(&scan_mutex);
3450
3451                         rebuild_lun_table(h, 0, 0);
3452                         complete_all(&h->scan_wait);
3453                         mutex_lock(&scan_mutex);
3454                         h->busy_scanning = 0;
3455                         mutex_unlock(&scan_mutex);
3456                 }
3457         }
3458
3459         return 0;
3460 }
3461
3462 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3463 {
3464         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3465                 return 0;
3466
3467         switch (c->err_info->SenseInfo[12]) {
3468         case STATE_CHANGED:
3469                 printk(KERN_WARNING "cciss%d: a state change "
3470                         "detected, command retried\n", h->ctlr);
3471                 return 1;
3472         break;
3473         case LUN_FAILED:
3474                 printk(KERN_WARNING "cciss%d: LUN failure "
3475                         "detected, action required\n", h->ctlr);
3476                 return 1;
3477         break;
3478         case REPORT_LUNS_CHANGED:
3479                 printk(KERN_WARNING "cciss%d: report LUN data "
3480                         "changed\n", h->ctlr);
3481         /*
3482          * Here, we could call add_to_scan_list and wake up the scan thread,
3483          * except that it's quite likely that we will get more than one
3484          * REPORT_LUNS_CHANGED condition in quick succession, which means
3485          * that those which occur after the first one will likely happen
3486          * *during* the scan_thread's rescan.  And the rescan code is not
3487          * robust enough to restart in the middle, undoing what it has already
3488          * done, and it's not clear that it's even possible to do this, since
3489          * part of what it does is notify the block layer, which starts
3490          * doing it's own i/o to read partition tables and so on, and the
3491          * driver doesn't have visibility to know what might need undoing.
3492          * In any event, if possible, it is horribly complicated to get right
3493          * so we just don't do it for now.
3494          *
3495          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3496          */
3497                 return 1;
3498         break;
3499         case POWER_OR_RESET:
3500                 printk(KERN_WARNING "cciss%d: a power on "
3501                         "or device reset detected\n", h->ctlr);
3502                 return 1;
3503         break;
3504         case UNIT_ATTENTION_CLEARED:
3505                 printk(KERN_WARNING "cciss%d: unit attention "
3506                     "cleared by another initiator\n", h->ctlr);
3507                 return 1;
3508         break;
3509         default:
3510                 printk(KERN_WARNING "cciss%d: unknown "
3511                         "unit attention detected\n", h->ctlr);
3512                                 return 1;
3513         }
3514 }
3515
3516 /*
3517  *  We cannot read the structure directly, for portability we must use
3518  *   the io functions.
3519  *   This is for debug only.
3520  */
3521 #ifdef CCISS_DEBUG
3522 static void print_cfg_table(CfgTable_struct *tb)
3523 {
3524         int i;
3525         char temp_name[17];
3526
3527         printk("Controller Configuration information\n");
3528         printk("------------------------------------\n");
3529         for (i = 0; i < 4; i++)
3530                 temp_name[i] = readb(&(tb->Signature[i]));
3531         temp_name[4] = '\0';
3532         printk("   Signature = %s\n", temp_name);
3533         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3534         printk("   Transport methods supported = 0x%x\n",
3535                readl(&(tb->TransportSupport)));
3536         printk("   Transport methods active = 0x%x\n",
3537                readl(&(tb->TransportActive)));
3538         printk("   Requested transport Method = 0x%x\n",
3539                readl(&(tb->HostWrite.TransportRequest)));
3540         printk("   Coalesce Interrupt Delay = 0x%x\n",
3541                readl(&(tb->HostWrite.CoalIntDelay)));
3542         printk("   Coalesce Interrupt Count = 0x%x\n",
3543                readl(&(tb->HostWrite.CoalIntCount)));
3544         printk("   Max outstanding commands = 0x%d\n",
3545                readl(&(tb->CmdsOutMax)));
3546         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3547         for (i = 0; i < 16; i++)
3548                 temp_name[i] = readb(&(tb->ServerName[i]));
3549         temp_name[16] = '\0';
3550         printk("   Server Name = %s\n", temp_name);
3551         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3552 }
3553 #endif                          /* CCISS_DEBUG */
3554
3555 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3556 {
3557         int i, offset, mem_type, bar_type;
3558         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3559                 return 0;
3560         offset = 0;
3561         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3562                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3563                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3564                         offset += 4;
3565                 else {
3566                         mem_type = pci_resource_flags(pdev, i) &
3567                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3568                         switch (mem_type) {
3569                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3570                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3571                                 offset += 4;    /* 32 bit */
3572                                 break;
3573                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3574                                 offset += 8;
3575                                 break;
3576                         default:        /* reserved in PCI 2.2 */
3577                                 printk(KERN_WARNING
3578                                        "Base address is invalid\n");
3579                                 return -1;
3580                                 break;
3581                         }
3582                 }
3583                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3584                         return i + 1;
3585         }
3586         return -1;
3587 }
3588
3589 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3590  * controllers that are capable. If not, we use IO-APIC mode.
3591  */
3592
3593 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3594                                            struct pci_dev *pdev, __u32 board_id)
3595 {
3596 #ifdef CONFIG_PCI_MSI
3597         int err;
3598         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3599         {0, 2}, {0, 3}
3600         };
3601
3602         /* Some boards advertise MSI but don't really support it */
3603         if ((board_id == 0x40700E11) ||
3604             (board_id == 0x40800E11) ||
3605             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3606                 goto default_int_mode;
3607
3608         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3609                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3610                 if (!err) {
3611                         c->intr[0] = cciss_msix_entries[0].vector;
3612                         c->intr[1] = cciss_msix_entries[1].vector;
3613                         c->intr[2] = cciss_msix_entries[2].vector;
3614                         c->intr[3] = cciss_msix_entries[3].vector;
3615                         c->msix_vector = 1;
3616                         return;
3617                 }
3618                 if (err > 0) {
3619                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3620                                "available\n", err);
3621                         goto default_int_mode;
3622                 } else {
3623                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3624                                err);
3625                         goto default_int_mode;
3626                 }
3627         }
3628         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3629                 if (!pci_enable_msi(pdev)) {
3630                         c->msi_vector = 1;
3631                 } else {
3632                         printk(KERN_WARNING "cciss: MSI init failed\n");
3633                 }
3634         }
3635 default_int_mode:
3636 #endif                          /* CONFIG_PCI_MSI */
3637         /* if we get here we're going to use the default interrupt mode */
3638         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3639         return;
3640 }
3641
3642 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3643 {
3644         ushort subsystem_vendor_id, subsystem_device_id, command;
3645         __u32 board_id, scratchpad = 0;
3646         __u64 cfg_offset;
3647         __u32 cfg_base_addr;
3648         __u64 cfg_base_addr_index;
3649         int i, prod_index, err;
3650
3651         subsystem_vendor_id = pdev->subsystem_vendor;
3652         subsystem_device_id = pdev->subsystem_device;
3653         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3654                     subsystem_vendor_id);
3655
3656         for (i = 0; i < ARRAY_SIZE(products); i++) {
3657                 /* Stand aside for hpsa driver on request */
3658                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3659                         return -ENODEV;
3660                 if (board_id == products[i].board_id)
3661                         break;
3662         }
3663         prod_index = i;
3664         if (prod_index == ARRAY_SIZE(products)) {
3665                 dev_warn(&pdev->dev,
3666                         "unrecognized board ID: 0x%08lx, ignoring.\n",
3667                         (unsigned long) board_id);
3668                 return -ENODEV;
3669         }
3670
3671         /* check to see if controller has been disabled */
3672         /* BEFORE trying to enable it */
3673         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3674         if (!(command & 0x02)) {
3675                 printk(KERN_WARNING
3676                        "cciss: controller appears to be disabled\n");
3677                 return -ENODEV;
3678         }
3679
3680         err = pci_enable_device(pdev);
3681         if (err) {
3682                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3683                 return err;
3684         }
3685
3686         err = pci_request_regions(pdev, "cciss");
3687         if (err) {
3688                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3689                        "aborting\n");
3690                 return err;
3691         }
3692
3693 #ifdef CCISS_DEBUG
3694         printk("command = %x\n", command);
3695         printk("irq = %x\n", pdev->irq);
3696         printk("board_id = %x\n", board_id);
3697 #endif                          /* CCISS_DEBUG */
3698
3699 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3700  * else we use the IO-APIC interrupt assigned to us by system ROM.
3701  */
3702         cciss_interrupt_mode(c, pdev, board_id);
3703
3704         /* find the memory BAR */
3705         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3706                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3707                         break;
3708         }
3709         if (i == DEVICE_COUNT_RESOURCE) {
3710                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3711                 err = -ENODEV;
3712                 goto err_out_free_res;
3713         }
3714
3715         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3716                                                  * already removed
3717                                                  */
3718
3719 #ifdef CCISS_DEBUG
3720         printk("address 0 = %lx\n", c->paddr);
3721 #endif                          /* CCISS_DEBUG */
3722         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3723
3724         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3725          * We poll for up to 120 secs, once per 100ms. */
3726         for (i = 0; i < 1200; i++) {
3727                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3728                 if (scratchpad == CCISS_FIRMWARE_READY)
3729                         break;
3730                 set_current_state(TASK_INTERRUPTIBLE);
3731                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
3732         }
3733         if (scratchpad != CCISS_FIRMWARE_READY) {
3734                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3735                 err = -ENODEV;
3736                 goto err_out_free_res;
3737         }
3738
3739         /* get the address index number */
3740         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3741         cfg_base_addr &= (__u32) 0x0000ffff;
3742 #ifdef CCISS_DEBUG
3743         printk("cfg base address = %x\n", cfg_base_addr);
3744 #endif                          /* CCISS_DEBUG */
3745         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3746 #ifdef CCISS_DEBUG
3747         printk("cfg base address index = %llx\n",
3748                 (unsigned long long)cfg_base_addr_index);
3749 #endif                          /* CCISS_DEBUG */
3750         if (cfg_base_addr_index == -1) {
3751                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3752                 err = -ENODEV;
3753                 goto err_out_free_res;
3754         }
3755
3756         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3757 #ifdef CCISS_DEBUG
3758         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3759 #endif                          /* CCISS_DEBUG */
3760         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3761                                                        cfg_base_addr_index) +
3762                                     cfg_offset, sizeof(CfgTable_struct));
3763         c->board_id = board_id;
3764
3765 #ifdef CCISS_DEBUG
3766         print_cfg_table(c->cfgtable);
3767 #endif                          /* CCISS_DEBUG */
3768
3769         /* Some controllers support Zero Memory Raid (ZMR).
3770          * When configured in ZMR mode the number of supported
3771          * commands drops to 64. So instead of just setting an
3772          * arbitrary value we make the driver a little smarter.
3773          * We read the config table to tell us how many commands
3774          * are supported on the controller then subtract 4 to
3775          * leave a little room for ioctl calls.
3776          */
3777         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3778         c->maxsgentries = readl(&(c->cfgtable->MaxSGElements));
3779
3780         /*
3781          * Limit native command to 32 s/g elements to save dma'able memory.
3782          * Howvever spec says if 0, use 31
3783          */
3784
3785         c->max_cmd_sgentries = 31;
3786         if (c->maxsgentries > 512) {
3787                 c->max_cmd_sgentries = 32;
3788                 c->chainsize = c->maxsgentries - c->max_cmd_sgentries + 1;
3789                 c->maxsgentries -= 1;   /* account for chain pointer */
3790         } else {
3791                 c->maxsgentries = 31;   /* Default to traditional value */
3792                 c->chainsize = 0;       /* traditional */
3793         }
3794
3795         c->product_name = products[prod_index].product_name;
3796         c->access = *(products[prod_index].access);
3797         c->nr_cmds = c->max_commands - 4;
3798         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3799             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3800             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3801             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3802                 printk("Does not appear to be a valid CISS config table\n");
3803                 err = -ENODEV;
3804                 goto err_out_free_res;
3805         }
3806 #ifdef CONFIG_X86
3807         {
3808                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3809                 __u32 prefetch;
3810                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3811                 prefetch |= 0x100;
3812                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3813         }
3814 #endif
3815
3816         /* Disabling DMA prefetch and refetch for the P600.
3817          * An ASIC bug may result in accesses to invalid memory addresses.
3818          * We've disabled prefetch for some time now. Testing with XEN
3819          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3820          */
3821         if(board_id == 0x3225103C) {
3822                 __u32 dma_prefetch;
3823                 __u32 dma_refetch;
3824                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3825                 dma_prefetch |= 0x8000;
3826                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3827                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3828                 dma_refetch |= 0x1;
3829                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3830         }
3831
3832 #ifdef CCISS_DEBUG
3833         printk("Trying to put board into Simple mode\n");
3834 #endif                          /* CCISS_DEBUG */
3835         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3836         /* Update the field, and then ring the doorbell */
3837         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3838         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3839
3840         /* under certain very rare conditions, this can take awhile.
3841          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3842          * as we enter this code.) */
3843         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3844                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3845                         break;
3846                 /* delay and try again */
3847                 set_current_state(TASK_INTERRUPTIBLE);
3848                 schedule_timeout(msecs_to_jiffies(1));
3849         }
3850
3851 #ifdef CCISS_DEBUG
3852         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3853                readl(c->vaddr + SA5_DOORBELL));
3854 #endif                          /* CCISS_DEBUG */
3855 #ifdef CCISS_DEBUG
3856         print_cfg_table(c->cfgtable);
3857 #endif                          /* CCISS_DEBUG */
3858
3859         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3860                 printk(KERN_WARNING "cciss: unable to get board into"
3861                        " simple mode\n");
3862                 err = -ENODEV;
3863                 goto err_out_free_res;
3864         }
3865         return 0;
3866
3867 err_out_free_res:
3868         /*
3869          * Deliberately omit pci_disable_device(): it does something nasty to
3870          * Smart Array controllers that pci_enable_device does not undo
3871          */
3872         pci_release_regions(pdev);
3873         return err;
3874 }
3875
3876 /* Function to find the first free pointer into our hba[] array
3877  * Returns -1 if no free entries are left.
3878  */
3879 static int alloc_cciss_hba(void)
3880 {
3881         int i;
3882
3883         for (i = 0; i < MAX_CTLR; i++) {
3884                 if (!hba[i]) {
3885                         ctlr_info_t *p;
3886
3887                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3888                         if (!p)
3889                                 goto Enomem;
3890                         hba[i] = p;
3891                         return i;
3892                 }
3893         }
3894         printk(KERN_WARNING "cciss: This driver supports a maximum"
3895                " of %d controllers.\n", MAX_CTLR);
3896         return -1;
3897 Enomem:
3898         printk(KERN_ERR "cciss: out of memory.\n");
3899         return -1;
3900 }
3901
3902 static void free_hba(int n)
3903 {
3904         ctlr_info_t *h = hba[n];
3905         int i;
3906
3907         hba[n] = NULL;
3908         for (i = 0; i < h->highest_lun + 1; i++)
3909                 if (h->gendisk[i] != NULL)
3910                         put_disk(h->gendisk[i]);
3911         kfree(h);
3912 }
3913
3914 /* Send a message CDB to the firmware. */
3915 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3916 {
3917         typedef struct {
3918                 CommandListHeader_struct CommandHeader;
3919                 RequestBlock_struct Request;
3920                 ErrDescriptor_struct ErrorDescriptor;
3921         } Command;
3922         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3923         Command *cmd;
3924         dma_addr_t paddr64;
3925         uint32_t paddr32, tag;
3926         void __iomem *vaddr;
3927         int i, err;
3928
3929         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3930         if (vaddr == NULL)
3931                 return -ENOMEM;
3932
3933         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3934            CCISS commands, so they must be allocated from the lower 4GiB of
3935            memory. */
3936         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3937         if (err) {
3938                 iounmap(vaddr);
3939                 return -ENOMEM;
3940         }
3941
3942         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3943         if (cmd == NULL) {
3944                 iounmap(vaddr);
3945                 return -ENOMEM;
3946         }
3947
3948         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
3949            although there's no guarantee, we assume that the address is at
3950            least 4-byte aligned (most likely, it's page-aligned). */
3951         paddr32 = paddr64;
3952
3953         cmd->CommandHeader.ReplyQueue = 0;
3954         cmd->CommandHeader.SGList = 0;
3955         cmd->CommandHeader.SGTotal = 0;
3956         cmd->CommandHeader.Tag.lower = paddr32;
3957         cmd->CommandHeader.Tag.upper = 0;
3958         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
3959
3960         cmd->Request.CDBLen = 16;
3961         cmd->Request.Type.Type = TYPE_MSG;
3962         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
3963         cmd->Request.Type.Direction = XFER_NONE;
3964         cmd->Request.Timeout = 0; /* Don't time out */
3965         cmd->Request.CDB[0] = opcode;
3966         cmd->Request.CDB[1] = type;
3967         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
3968
3969         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
3970         cmd->ErrorDescriptor.Addr.upper = 0;
3971         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
3972
3973         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
3974
3975         for (i = 0; i < 10; i++) {
3976                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
3977                 if ((tag & ~3) == paddr32)
3978                         break;
3979                 schedule_timeout_uninterruptible(HZ);
3980         }
3981
3982         iounmap(vaddr);
3983
3984         /* we leak the DMA buffer here ... no choice since the controller could
3985            still complete the command. */
3986         if (i == 10) {
3987                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
3988                         opcode, type);
3989                 return -ETIMEDOUT;
3990         }
3991
3992         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
3993
3994         if (tag & 2) {
3995                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
3996                         opcode, type);
3997                 return -EIO;
3998         }
3999
4000         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4001                 opcode, type);
4002         return 0;
4003 }
4004
4005 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4006 #define cciss_noop(p) cciss_message(p, 3, 0)
4007
4008 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4009 {
4010 /* the #defines are stolen from drivers/pci/msi.h. */
4011 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4012 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4013
4014         int pos;
4015         u16 control = 0;
4016
4017         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4018         if (pos) {
4019                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4020                 if (control & PCI_MSI_FLAGS_ENABLE) {
4021                         printk(KERN_INFO "cciss: resetting MSI\n");
4022                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4023                 }
4024         }
4025
4026         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4027         if (pos) {
4028                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4029                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4030                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4031                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4032                 }
4033         }
4034
4035         return 0;
4036 }
4037
4038 /* This does a hard reset of the controller using PCI power management
4039  * states. */
4040 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4041 {
4042         u16 pmcsr, saved_config_space[32];
4043         int i, pos;
4044
4045         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4046
4047         /* This is very nearly the same thing as
4048
4049            pci_save_state(pci_dev);
4050            pci_set_power_state(pci_dev, PCI_D3hot);
4051            pci_set_power_state(pci_dev, PCI_D0);
4052            pci_restore_state(pci_dev);
4053
4054            but we can't use these nice canned kernel routines on
4055            kexec, because they also check the MSI/MSI-X state in PCI
4056            configuration space and do the wrong thing when it is
4057            set/cleared.  Also, the pci_save/restore_state functions
4058            violate the ordering requirements for restoring the
4059            configuration space from the CCISS document (see the
4060            comment below).  So we roll our own .... */
4061
4062         for (i = 0; i < 32; i++)
4063                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4064
4065         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4066         if (pos == 0) {
4067                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4068                 return -ENODEV;
4069         }
4070
4071         /* Quoting from the Open CISS Specification: "The Power
4072          * Management Control/Status Register (CSR) controls the power
4073          * state of the device.  The normal operating state is D0,
4074          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4075          * the controller, place the interface device in D3 then to
4076          * D0, this causes a secondary PCI reset which will reset the
4077          * controller." */
4078
4079         /* enter the D3hot power management state */
4080         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4081         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4082         pmcsr |= PCI_D3hot;
4083         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4084
4085         schedule_timeout_uninterruptible(HZ >> 1);
4086
4087         /* enter the D0 power management state */
4088         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4089         pmcsr |= PCI_D0;
4090         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4091
4092         schedule_timeout_uninterruptible(HZ >> 1);
4093
4094         /* Restore the PCI configuration space.  The Open CISS
4095          * Specification says, "Restore the PCI Configuration
4096          * Registers, offsets 00h through 60h. It is important to
4097          * restore the command register, 16-bits at offset 04h,
4098          * last. Do not restore the configuration status register,
4099          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4100         for (i = 0; i < 32; i++) {
4101                 if (i == 2 || i == 3)
4102                         continue;
4103                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4104         }
4105         wmb();
4106         pci_write_config_word(pdev, 4, saved_config_space[2]);
4107
4108         return 0;
4109 }
4110
4111 /*
4112  *  This is it.  Find all the controllers and register them.  I really hate
4113  *  stealing all these major device numbers.
4114  *  returns the number of block devices registered.
4115  */
4116 static int __devinit cciss_init_one(struct pci_dev *pdev,
4117                                     const struct pci_device_id *ent)
4118 {
4119         int i;
4120         int j = 0;
4121         int k = 0;
4122         int rc;
4123         int dac, return_code;
4124         InquiryData_struct *inq_buff;
4125
4126         if (reset_devices) {
4127                 /* Reset the controller with a PCI power-cycle */
4128                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4129                         return -ENODEV;
4130
4131                 /* Now try to get the controller to respond to a no-op. Some
4132                    devices (notably the HP Smart Array 5i Controller) need
4133                    up to 30 seconds to respond. */
4134                 for (i=0; i<30; i++) {
4135                         if (cciss_noop(pdev) == 0)
4136                                 break;
4137
4138                         schedule_timeout_uninterruptible(HZ);
4139                 }
4140                 if (i == 30) {
4141                         printk(KERN_ERR "cciss: controller seems dead\n");
4142                         return -EBUSY;
4143                 }
4144         }
4145
4146         i = alloc_cciss_hba();
4147         if (i < 0)
4148                 return -1;
4149
4150         hba[i]->busy_initializing = 1;
4151         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4152         INIT_HLIST_HEAD(&hba[i]->reqQ);
4153         mutex_init(&hba[i]->busy_shutting_down);
4154
4155         if (cciss_pci_init(hba[i], pdev) != 0)
4156                 goto clean_no_release_regions;
4157
4158         sprintf(hba[i]->devname, "cciss%d", i);
4159         hba[i]->ctlr = i;
4160         hba[i]->pdev = pdev;
4161
4162         init_completion(&hba[i]->scan_wait);
4163
4164         if (cciss_create_hba_sysfs_entry(hba[i]))
4165                 goto clean0;
4166
4167         /* configure PCI DMA stuff */
4168         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4169                 dac = 1;
4170         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4171                 dac = 0;
4172         else {
4173                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4174                 goto clean1;
4175         }
4176
4177         /*
4178          * register with the major number, or get a dynamic major number
4179          * by passing 0 as argument.  This is done for greater than
4180          * 8 controller support.
4181          */
4182         if (i < MAX_CTLR_ORIG)
4183                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4184         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4185         if (rc == -EBUSY || rc == -EINVAL) {
4186                 printk(KERN_ERR
4187                        "cciss:  Unable to get major number %d for %s "
4188                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4189                 goto clean1;
4190         } else {
4191                 if (i >= MAX_CTLR_ORIG)
4192                         hba[i]->major = rc;
4193         }
4194
4195         /* make sure the board interrupts are off */
4196         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4197         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
4198                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
4199                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4200                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4201                 goto clean2;
4202         }
4203
4204         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4205                hba[i]->devname, pdev->device, pci_name(pdev),
4206                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4207
4208         hba[i]->cmd_pool_bits =
4209             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4210                         * sizeof(unsigned long), GFP_KERNEL);
4211         hba[i]->cmd_pool = (CommandList_struct *)
4212             pci_alloc_consistent(hba[i]->pdev,
4213                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4214                     &(hba[i]->cmd_pool_dhandle));
4215         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4216             pci_alloc_consistent(hba[i]->pdev,
4217                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4218                     &(hba[i]->errinfo_pool_dhandle));
4219         if ((hba[i]->cmd_pool_bits == NULL)
4220             || (hba[i]->cmd_pool == NULL)
4221             || (hba[i]->errinfo_pool == NULL)) {
4222                 printk(KERN_ERR "cciss: out of memory");
4223                 goto clean4;
4224         }
4225
4226         /* Need space for temp scatter list */
4227         hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4228                                                 sizeof(struct scatterlist *),
4229                                                 GFP_KERNEL);
4230         for (k = 0; k < hba[i]->nr_cmds; k++) {
4231                 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4232                                                         hba[i]->maxsgentries,
4233                                                         GFP_KERNEL);
4234                 if (hba[i]->scatter_list[k] == NULL) {
4235                         printk(KERN_ERR "cciss%d: could not allocate "
4236                                 "s/g lists\n", i);
4237                         goto clean4;
4238                 }
4239         }
4240         hba[i]->cmd_sg_list = kmalloc(sizeof(struct Cmd_sg_list *) *
4241                                                 hba[i]->nr_cmds,
4242                                                 GFP_KERNEL);
4243         if (!hba[i]->cmd_sg_list) {
4244                 printk(KERN_ERR "cciss%d: Cannot get memory for "
4245                         "s/g chaining.\n", i);
4246                 goto clean4;
4247         }
4248         /* Build up chain blocks for each command */
4249         if (hba[i]->chainsize > 0) {
4250                 for (j = 0; j < hba[i]->nr_cmds; j++) {
4251                         hba[i]->cmd_sg_list[j] =
4252                                         kmalloc(sizeof(struct Cmd_sg_list),
4253                                                         GFP_KERNEL);
4254                         if (!hba[i]->cmd_sg_list[j]) {
4255                                 printk(KERN_ERR "cciss%d: Cannot get memory "
4256                                         "for chain block.\n", i);
4257                                 goto clean4;
4258                         }
4259                         /* Need a block of chainsized s/g elements. */
4260                         hba[i]->cmd_sg_list[j]->sgchain =
4261                                         kmalloc((hba[i]->chainsize *
4262                                                 sizeof(SGDescriptor_struct)),
4263                                                 GFP_KERNEL);
4264                         if (!hba[i]->cmd_sg_list[j]->sgchain) {
4265                                 printk(KERN_ERR "cciss%d: Cannot get memory "
4266                                         "for s/g chains\n", i);
4267                                 goto clean4;
4268                         }
4269                 }
4270         }
4271
4272         spin_lock_init(&hba[i]->lock);
4273
4274         /* Initialize the pdev driver private data.
4275            have it point to hba[i].  */
4276         pci_set_drvdata(pdev, hba[i]);
4277         /* command and error info recs zeroed out before
4278            they are used */
4279         memset(hba[i]->cmd_pool_bits, 0,
4280                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4281                         * sizeof(unsigned long));
4282
4283         hba[i]->num_luns = 0;
4284         hba[i]->highest_lun = -1;
4285         for (j = 0; j < CISS_MAX_LUN; j++) {
4286                 hba[i]->drv[j] = NULL;
4287                 hba[i]->gendisk[j] = NULL;
4288         }
4289
4290         cciss_scsi_setup(i);
4291
4292         /* Turn the interrupts on so we can service requests */
4293         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4294
4295         /* Get the firmware version */
4296         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4297         if (inq_buff == NULL) {
4298                 printk(KERN_ERR "cciss: out of memory\n");
4299                 goto clean4;
4300         }
4301
4302         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4303                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4304         if (return_code == IO_OK) {
4305                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4306                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4307                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4308                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4309         } else {         /* send command failed */
4310                 printk(KERN_WARNING "cciss: unable to determine firmware"
4311                         " version of controller\n");
4312         }
4313         kfree(inq_buff);
4314
4315         cciss_procinit(i);
4316
4317         hba[i]->cciss_max_sectors = 8192;
4318
4319         rebuild_lun_table(hba[i], 1, 0);
4320         hba[i]->busy_initializing = 0;
4321         return 1;
4322
4323 clean4:
4324         kfree(hba[i]->cmd_pool_bits);
4325         /* Free up sg elements */
4326         for (k = 0; k < hba[i]->nr_cmds; k++)
4327                 kfree(hba[i]->scatter_list[k]);
4328         kfree(hba[i]->scatter_list);
4329         /* Only free up extra s/g lists if controller supports them */
4330         if (hba[i]->chainsize > 0) {
4331                 for (j = 0; j < hba[i]->nr_cmds; j++) {
4332                         if (hba[i]->cmd_sg_list[j]) {
4333                                 kfree(hba[i]->cmd_sg_list[j]->sgchain);
4334                                 kfree(hba[i]->cmd_sg_list[j]);
4335                         }
4336                 }
4337                 kfree(hba[i]->cmd_sg_list);
4338         }
4339         if (hba[i]->cmd_pool)
4340                 pci_free_consistent(hba[i]->pdev,
4341                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4342                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4343         if (hba[i]->errinfo_pool)
4344                 pci_free_consistent(hba[i]->pdev,
4345                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4346                                     hba[i]->errinfo_pool,
4347                                     hba[i]->errinfo_pool_dhandle);
4348         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4349 clean2:
4350         unregister_blkdev(hba[i]->major, hba[i]->devname);
4351 clean1:
4352         cciss_destroy_hba_sysfs_entry(hba[i]);
4353 clean0:
4354         pci_release_regions(pdev);
4355 clean_no_release_regions:
4356         hba[i]->busy_initializing = 0;
4357
4358         /*
4359          * Deliberately omit pci_disable_device(): it does something nasty to
4360          * Smart Array controllers that pci_enable_device does not undo
4361          */
4362         pci_set_drvdata(pdev, NULL);
4363         free_hba(i);
4364         return -1;
4365 }
4366
4367 static void cciss_shutdown(struct pci_dev *pdev)
4368 {
4369         ctlr_info_t *h;
4370         char *flush_buf;
4371         int return_code;
4372
4373         h = pci_get_drvdata(pdev);
4374         flush_buf = kzalloc(4, GFP_KERNEL);
4375         if (!flush_buf) {
4376                 printk(KERN_WARNING
4377                         "cciss:%d cache not flushed, out of memory.\n",
4378                         h->ctlr);
4379                 return;
4380         }
4381         /* write all data in the battery backed cache to disk */
4382         memset(flush_buf, 0, 4);
4383         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4384                 4, 0, CTLR_LUNID, TYPE_CMD);
4385         kfree(flush_buf);
4386         if (return_code != IO_OK)
4387                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4388                         h->ctlr);
4389         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4390         free_irq(h->intr[2], h);
4391 }
4392
4393 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4394 {
4395         ctlr_info_t *tmp_ptr;
4396         int i, j;
4397
4398         if (pci_get_drvdata(pdev) == NULL) {
4399                 printk(KERN_ERR "cciss: Unable to remove device \n");
4400                 return;
4401         }
4402
4403         tmp_ptr = pci_get_drvdata(pdev);
4404         i = tmp_ptr->ctlr;
4405         if (hba[i] == NULL) {
4406                 printk(KERN_ERR "cciss: device appears to "
4407                        "already be removed \n");
4408                 return;
4409         }
4410
4411         mutex_lock(&hba[i]->busy_shutting_down);
4412
4413         remove_from_scan_list(hba[i]);
4414         remove_proc_entry(hba[i]->devname, proc_cciss);
4415         unregister_blkdev(hba[i]->major, hba[i]->devname);
4416
4417         /* remove it from the disk list */
4418         for (j = 0; j < CISS_MAX_LUN; j++) {
4419                 struct gendisk *disk = hba[i]->gendisk[j];
4420                 if (disk) {
4421                         struct request_queue *q = disk->queue;
4422
4423                         if (disk->flags & GENHD_FL_UP) {
4424                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4425                                 del_gendisk(disk);
4426                         }
4427                         if (q)
4428                                 blk_cleanup_queue(q);
4429                 }
4430         }
4431
4432 #ifdef CONFIG_CISS_SCSI_TAPE
4433         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4434 #endif
4435
4436         cciss_shutdown(pdev);
4437
4438 #ifdef CONFIG_PCI_MSI
4439         if (hba[i]->msix_vector)
4440                 pci_disable_msix(hba[i]->pdev);
4441         else if (hba[i]->msi_vector)
4442                 pci_disable_msi(hba[i]->pdev);
4443 #endif                          /* CONFIG_PCI_MSI */
4444
4445         iounmap(hba[i]->vaddr);
4446
4447         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4448                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4449         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4450                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4451         kfree(hba[i]->cmd_pool_bits);
4452         /* Free up sg elements */
4453         for (j = 0; j < hba[i]->nr_cmds; j++)
4454                 kfree(hba[i]->scatter_list[j]);
4455         kfree(hba[i]->scatter_list);
4456         /* Only free up extra s/g lists if controller supports them */
4457         if (hba[i]->chainsize > 0) {
4458                 for (j = 0; j < hba[i]->nr_cmds; j++) {
4459                         if (hba[i]->cmd_sg_list[j]) {
4460                                 kfree(hba[i]->cmd_sg_list[j]->sgchain);
4461                                 kfree(hba[i]->cmd_sg_list[j]);
4462                         }
4463                 }
4464                 kfree(hba[i]->cmd_sg_list);
4465         }
4466         /*
4467          * Deliberately omit pci_disable_device(): it does something nasty to
4468          * Smart Array controllers that pci_enable_device does not undo
4469          */
4470         pci_release_regions(pdev);
4471         pci_set_drvdata(pdev, NULL);
4472         cciss_destroy_hba_sysfs_entry(hba[i]);
4473         mutex_unlock(&hba[i]->busy_shutting_down);
4474         free_hba(i);
4475 }
4476
4477 static struct pci_driver cciss_pci_driver = {
4478         .name = "cciss",
4479         .probe = cciss_init_one,
4480         .remove = __devexit_p(cciss_remove_one),
4481         .id_table = cciss_pci_device_id,        /* id_table */
4482         .shutdown = cciss_shutdown,
4483 };
4484
4485 /*
4486  *  This is it.  Register the PCI driver information for the cards we control
4487  *  the OS will call our registered routines when it finds one of our cards.
4488  */
4489 static int __init cciss_init(void)
4490 {
4491         int err;
4492
4493         /*
4494          * The hardware requires that commands are aligned on a 64-bit
4495          * boundary. Given that we use pci_alloc_consistent() to allocate an
4496          * array of them, the size must be a multiple of 8 bytes.
4497          */
4498         BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4499
4500         printk(KERN_INFO DRIVER_NAME "\n");
4501
4502         err = bus_register(&cciss_bus_type);
4503         if (err)
4504                 return err;
4505
4506         /* Start the scan thread */
4507         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4508         if (IS_ERR(cciss_scan_thread)) {
4509                 err = PTR_ERR(cciss_scan_thread);
4510                 goto err_bus_unregister;
4511         }
4512
4513         /* Register for our PCI devices */
4514         err = pci_register_driver(&cciss_pci_driver);
4515         if (err)
4516                 goto err_thread_stop;
4517
4518         return err;
4519
4520 err_thread_stop:
4521         kthread_stop(cciss_scan_thread);
4522 err_bus_unregister:
4523         bus_unregister(&cciss_bus_type);
4524
4525         return err;
4526 }
4527
4528 static void __exit cciss_cleanup(void)
4529 {
4530         int i;
4531
4532         pci_unregister_driver(&cciss_pci_driver);
4533         /* double check that all controller entrys have been removed */
4534         for (i = 0; i < MAX_CTLR; i++) {
4535                 if (hba[i] != NULL) {
4536                         printk(KERN_WARNING "cciss: had to remove"
4537                                " controller %d\n", i);
4538                         cciss_remove_one(hba[i]->pdev);
4539                 }
4540         }
4541         kthread_stop(cciss_scan_thread);
4542         remove_proc_entry("driver/cciss", NULL);
4543         bus_unregister(&cciss_bus_type);
4544 }
4545
4546 static void fail_all_cmds(unsigned long ctlr)
4547 {
4548         /* If we get here, the board is apparently dead. */
4549         ctlr_info_t *h = hba[ctlr];
4550         CommandList_struct *c;
4551         unsigned long flags;
4552
4553         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4554         h->alive = 0;           /* the controller apparently died... */
4555
4556         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4557
4558         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
4559
4560         /* move everything off the request queue onto the completed queue */
4561         while (!hlist_empty(&h->reqQ)) {
4562                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4563                 removeQ(c);
4564                 h->Qdepth--;
4565                 addQ(&h->cmpQ, c);
4566         }
4567
4568         /* Now, fail everything on the completed queue with a HW error */
4569         while (!hlist_empty(&h->cmpQ)) {
4570                 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4571                 removeQ(c);
4572                 if (c->cmd_type != CMD_MSG_STALE)
4573                         c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4574                 if (c->cmd_type == CMD_RWREQ) {
4575                         complete_command(h, c, 0);
4576                 } else if (c->cmd_type == CMD_IOCTL_PEND)
4577                         complete(c->waiting);
4578 #ifdef CONFIG_CISS_SCSI_TAPE
4579                 else if (c->cmd_type == CMD_SCSI)
4580                         complete_scsi_command(c, 0, 0);
4581 #endif
4582         }
4583         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4584         return;
4585 }
4586
4587 module_init(cciss_init);
4588 module_exit(cciss_cleanup);