drivers/block/cciss.c

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