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/delay.h>
  30 #include <linux/major.h>
  31 #include <linux/fs.h>
  32 #include <linux/bio.h>
  33 #include <linux/blkpg.h>
  34 #include <linux/timer.h>
  35 #include <linux/proc_fs.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/init.h>
  38 #include <linux/jiffies.h>
  39 #include <linux/hdreg.h>
  40 #include <linux/spinlock.h>
  41 #include <linux/compat.h>
  42 #include <linux/mutex.h>
  43 #include <asm/uaccess.h>
  44 #include <asm/io.h>
  45
  46 #include <linux/dma-mapping.h>
  47 #include <linux/blkdev.h>
  48 #include <linux/genhd.h>
  49 #include <linux/completion.h>
  50 #include <scsi/scsi.h>
  51 #include <scsi/sg.h>
  52 #include <scsi/scsi_ioctl.h>
  53 #include <linux/cdrom.h>
  54 #include <linux/scatterlist.h>
  55 #include <linux/kthread.h>
  56
  57 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  58 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  59 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  60
  61 /* Embedded module documentation macros - see modules.h */
  62 MODULE_AUTHOR("Hewlett-Packard Company");
  63 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  64 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  65 MODULE_VERSION("3.6.26");
  66 MODULE_LICENSE("GPL");
  67
  68 static DEFINE_MUTEX(cciss_mutex);
  69 static struct proc_dir_entry *proc_cciss;
  70
  71 #include "cciss_cmd.h"
  72 #include "cciss.h"
  73 #include <linux/cciss_ioctl.h>
  74
  75 /* define the PCI info for the cards we can control */
  76 static const struct pci_device_id cciss_pci_device_id[] = {
  77         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  78         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  79         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  80         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  86         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  87         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  88         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  89         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
  97         {0,}
  98 };
  99
 100 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 101
 102 /*  board_id = Subsystem Device ID & Vendor ID
 103  *  product = Marketing Name for the board
 104  *  access = Address of the struct of function pointers
 105  */
 106 static struct board_type products[] = {
 107         {0x40700E11, "Smart Array 5300", &SA5_access},
 108         {0x40800E11, "Smart Array 5i", &SA5B_access},
 109         {0x40820E11, "Smart Array 532", &SA5B_access},
 110         {0x40830E11, "Smart Array 5312", &SA5B_access},
 111         {0x409A0E11, "Smart Array 641", &SA5_access},
 112         {0x409B0E11, "Smart Array 642", &SA5_access},
 113         {0x409C0E11, "Smart Array 6400", &SA5_access},
 114         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 115         {0x40910E11, "Smart Array 6i", &SA5_access},
 116         {0x3225103C, "Smart Array P600", &SA5_access},
 117         {0x3223103C, "Smart Array P800", &SA5_access},
 118         {0x3234103C, "Smart Array P400", &SA5_access},
 119         {0x3235103C, "Smart Array P400i", &SA5_access},
 120         {0x3211103C, "Smart Array E200i", &SA5_access},
 121         {0x3212103C, "Smart Array E200", &SA5_access},
 122         {0x3213103C, "Smart Array E200i", &SA5_access},
 123         {0x3214103C, "Smart Array E200i", &SA5_access},
 124         {0x3215103C, "Smart Array E200i", &SA5_access},
 125         {0x3237103C, "Smart Array E500", &SA5_access},
 126         {0x3223103C, "Smart Array P800", &SA5_access},
 127         {0x3234103C, "Smart Array P400", &SA5_access},
 128         {0x323D103C, "Smart Array P700m", &SA5_access},
 129 };
 130
 131 /* How long to wait (in milliseconds) for board to go into simple mode */
 132 #define MAX_CONFIG_WAIT 30000
 133 #define MAX_IOCTL_CONFIG_WAIT 1000
 134
 135 /*define how many times we will try a command because of bus resets */
 136 #define MAX_CMD_RETRIES 3
 137
 138 #define MAX_CTLR        32
 139
 140 /* Originally cciss driver only supports 8 major numbers */
 141 #define MAX_CTLR_ORIG   8
 142
 143 static ctlr_info_t *hba[MAX_CTLR];
 144
 145 static struct task_struct *cciss_scan_thread;
 146 static DEFINE_MUTEX(scan_mutex);
 147 static LIST_HEAD(scan_q);
 148
 149 static void do_cciss_request(struct request_queue *q);
 150 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 151 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 152 static int cciss_open(struct block_device *bdev, fmode_t mode);
 153 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 154 static int cciss_release(struct gendisk *disk, fmode_t mode);
 155 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 156                     unsigned int cmd, unsigned long arg);
 157 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 158                        unsigned int cmd, unsigned long arg);
 159 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 160
 161 static int cciss_revalidate(struct gendisk *disk);
 162 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 163 static int deregister_disk(ctlr_info_t *h, int drv_index,
 164                            int clear_all, int via_ioctl);
 165
 166 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 167                         sector_t *total_size, unsigned int *block_size);
 168 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 169                         sector_t *total_size, unsigned int *block_size);
 170 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 171                         sector_t total_size,
 172                         unsigned int block_size, InquiryData_struct *inq_buff,
 173                                    drive_info_struct *drv);
 174 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
 175 static void start_io(ctlr_info_t *h);
 176 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 177                         __u8 page_code, unsigned char scsi3addr[],
 178                         int cmd_type);
 179 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 180         int attempt_retry);
 181 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 182
 183 static int add_to_scan_list(struct ctlr_info *h);
 184 static int scan_thread(void *data);
 185 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 186 static void cciss_hba_release(struct device *dev);
 187 static void cciss_device_release(struct device *dev);
 188 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 189 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 190 static inline u32 next_command(ctlr_info_t *h);
 191 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
 192         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 193         u64 *cfg_offset);
 194 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 195         unsigned long *memory_bar);
 196
 197
 198 /* performant mode helper functions */
 199 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 200                                 int *bucket_map);
 201 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 202
 203 #ifdef CONFIG_PROC_FS
 204 static void cciss_procinit(ctlr_info_t *h);
 205 #else
 206 static void cciss_procinit(ctlr_info_t *h)
 207 {
 208 }
 209 #endif                          /* CONFIG_PROC_FS */
 210
 211 #ifdef CONFIG_COMPAT
 212 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 213                               unsigned, unsigned long);
 214 #endif
 215
 216 static const struct block_device_operations cciss_fops = {
 217         .owner = THIS_MODULE,
 218         .open = cciss_unlocked_open,
 219         .release = cciss_release,
 220         .ioctl = do_ioctl,
 221         .getgeo = cciss_getgeo,
 222 #ifdef CONFIG_COMPAT
 223         .compat_ioctl = cciss_compat_ioctl,
 224 #endif
 225         .revalidate_disk = cciss_revalidate,
 226 };
 227
 228 /* set_performant_mode: Modify the tag for cciss performant
 229  * set bit 0 for pull model, bits 3-1 for block fetch
 230  * register number
 231  */
 232 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 233 {
 234         if (likely(h->transMethod == CFGTBL_Trans_Performant))
 235                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 236 }
 237
 238 /*
 239  * Enqueuing and dequeuing functions for cmdlists.
 240  */
 241 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 242 {
 243         hlist_add_head(&c->list, list);
 244 }
 245
 246 static inline void removeQ(CommandList_struct *c)
 247 {
 248         /*
 249          * After kexec/dump some commands might still
 250          * be in flight, which the firmware will try
 251          * to complete. Resetting the firmware doesn't work
 252          * with old fw revisions, so we have to mark
 253          * them off as 'stale' to prevent the driver from
 254          * falling over.
 255          */
 256         if (WARN_ON(hlist_unhashed(&c->list))) {
 257                 c->cmd_type = CMD_MSG_STALE;
 258                 return;
 259         }
 260
 261         hlist_del_init(&c->list);
 262 }
 263
 264 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 265         CommandList_struct *c)
 266 {
 267         unsigned long flags;
 268         set_performant_mode(h, c);
 269         spin_lock_irqsave(&h->lock, flags);
 270         addQ(&h->reqQ, c);
 271         h->Qdepth++;
 272         if (h->Qdepth > h->maxQsinceinit)
 273                 h->maxQsinceinit = h->Qdepth;
 274         start_io(h);
 275         spin_unlock_irqrestore(&h->lock, flags);
 276 }
 277
 278 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 279         int nr_cmds)
 280 {
 281         int i;
 282
 283         if (!cmd_sg_list)
 284                 return;
 285         for (i = 0; i < nr_cmds; i++) {
 286                 kfree(cmd_sg_list[i]);
 287                 cmd_sg_list[i] = NULL;
 288         }
 289         kfree(cmd_sg_list);
 290 }
 291
 292 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 293         ctlr_info_t *h, int chainsize, int nr_cmds)
 294 {
 295         int j;
 296         SGDescriptor_struct **cmd_sg_list;
 297
 298         if (chainsize <= 0)
 299                 return NULL;
 300
 301         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 302         if (!cmd_sg_list)
 303                 return NULL;
 304
 305         /* Build up chain blocks for each command */
 306         for (j = 0; j < nr_cmds; j++) {
 307                 /* Need a block of chainsized s/g elements. */
 308                 cmd_sg_list[j] = kmalloc((chainsize *
 309                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 310                 if (!cmd_sg_list[j]) {
 311                         dev_err(&h->pdev->dev, "Cannot get memory "
 312                                 "for s/g chains.\n");
 313                         goto clean;
 314                 }
 315         }
 316         return cmd_sg_list;
 317 clean:
 318         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 319         return NULL;
 320 }
 321
 322 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 323 {
 324         SGDescriptor_struct *chain_sg;
 325         u64bit temp64;
 326
 327         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 328                 return;
 329
 330         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 331         temp64.val32.lower = chain_sg->Addr.lower;
 332         temp64.val32.upper = chain_sg->Addr.upper;
 333         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 334 }
 335
 336 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 337         SGDescriptor_struct *chain_block, int len)
 338 {
 339         SGDescriptor_struct *chain_sg;
 340         u64bit temp64;
 341
 342         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 343         chain_sg->Ext = CCISS_SG_CHAIN;
 344         chain_sg->Len = len;
 345         temp64.val = pci_map_single(h->pdev, chain_block, len,
 346                                 PCI_DMA_TODEVICE);
 347         chain_sg->Addr.lower = temp64.val32.lower;
 348         chain_sg->Addr.upper = temp64.val32.upper;
 349 }
 350
 351 #include "cciss_scsi.c"         /* For SCSI tape support */
 352
 353 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 354         "UNKNOWN"
 355 };
 356 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 357
 358 #ifdef CONFIG_PROC_FS
 359
 360 /*
 361  * Report information about this controller.
 362  */
 363 #define ENG_GIG 1000000000
 364 #define ENG_GIG_FACTOR (ENG_GIG/512)
 365 #define ENGAGE_SCSI     "engage scsi"
 366
 367 static void cciss_seq_show_header(struct seq_file *seq)
 368 {
 369         ctlr_info_t *h = seq->private;
 370
 371         seq_printf(seq, "%s: HP %s Controller\n"
 372                 "Board ID: 0x%08lx\n"
 373                 "Firmware Version: %c%c%c%c\n"
 374                 "IRQ: %d\n"
 375                 "Logical drives: %d\n"
 376                 "Current Q depth: %d\n"
 377                 "Current # commands on controller: %d\n"
 378                 "Max Q depth since init: %d\n"
 379                 "Max # commands on controller since init: %d\n"
 380                 "Max SG entries since init: %d\n",
 381                 h->devname,
 382                 h->product_name,
 383                 (unsigned long)h->board_id,
 384                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 385                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 386                 h->num_luns,
 387                 h->Qdepth, h->commands_outstanding,
 388                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 389
 390 #ifdef CONFIG_CISS_SCSI_TAPE
 391         cciss_seq_tape_report(seq, h);
 392 #endif /* CONFIG_CISS_SCSI_TAPE */
 393 }
 394
 395 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 396 {
 397         ctlr_info_t *h = seq->private;
 398         unsigned long flags;
 399
 400         /* prevent displaying bogus info during configuration
 401          * or deconfiguration of a logical volume
 402          */
 403         spin_lock_irqsave(&h->lock, flags);
 404         if (h->busy_configuring) {
 405                 spin_unlock_irqrestore(&h->lock, flags);
 406                 return ERR_PTR(-EBUSY);
 407         }
 408         h->busy_configuring = 1;
 409         spin_unlock_irqrestore(&h->lock, flags);
 410
 411         if (*pos == 0)
 412                 cciss_seq_show_header(seq);
 413
 414         return pos;
 415 }
 416
 417 static int cciss_seq_show(struct seq_file *seq, void *v)
 418 {
 419         sector_t vol_sz, vol_sz_frac;
 420         ctlr_info_t *h = seq->private;
 421         unsigned ctlr = h->ctlr;
 422         loff_t *pos = v;
 423         drive_info_struct *drv = h->drv[*pos];
 424
 425         if (*pos > h->highest_lun)
 426                 return 0;
 427
 428         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 429                 return 0;
 430
 431         if (drv->heads == 0)
 432                 return 0;
 433
 434         vol_sz = drv->nr_blocks;
 435         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 436         vol_sz_frac *= 100;
 437         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 438
 439         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 440                 drv->raid_level = RAID_UNKNOWN;
 441         seq_printf(seq, "cciss/c%dd%d:"
 442                         "\t%4u.%02uGB\tRAID %s\n",
 443                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 444                         raid_label[drv->raid_level]);
 445         return 0;
 446 }
 447
 448 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 449 {
 450         ctlr_info_t *h = seq->private;
 451
 452         if (*pos > h->highest_lun)
 453                 return NULL;
 454         *pos += 1;
 455
 456         return pos;
 457 }
 458
 459 static void cciss_seq_stop(struct seq_file *seq, void *v)
 460 {
 461         ctlr_info_t *h = seq->private;
 462
 463         /* Only reset h->busy_configuring if we succeeded in setting
 464          * it during cciss_seq_start. */
 465         if (v == ERR_PTR(-EBUSY))
 466                 return;
 467
 468         h->busy_configuring = 0;
 469 }
 470
 471 static const struct seq_operations cciss_seq_ops = {
 472         .start = cciss_seq_start,
 473         .show  = cciss_seq_show,
 474         .next  = cciss_seq_next,
 475         .stop  = cciss_seq_stop,
 476 };
 477
 478 static int cciss_seq_open(struct inode *inode, struct file *file)
 479 {
 480         int ret = seq_open(file, &cciss_seq_ops);
 481         struct seq_file *seq = file->private_data;
 482
 483         if (!ret)
 484                 seq->private = PDE(inode)->data;
 485
 486         return ret;
 487 }
 488
 489 static ssize_t
 490 cciss_proc_write(struct file *file, const char __user *buf,
 491                  size_t length, loff_t *ppos)
 492 {
 493         int err;
 494         char *buffer;
 495
 496 #ifndef CONFIG_CISS_SCSI_TAPE
 497         return -EINVAL;
 498 #endif
 499
 500         if (!buf || length > PAGE_SIZE - 1)
 501                 return -EINVAL;
 502
 503         buffer = (char *)__get_free_page(GFP_KERNEL);
 504         if (!buffer)
 505                 return -ENOMEM;
 506
 507         err = -EFAULT;
 508         if (copy_from_user(buffer, buf, length))
 509                 goto out;
 510         buffer[length] = '\0';
 511
 512 #ifdef CONFIG_CISS_SCSI_TAPE
 513         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 514                 struct seq_file *seq = file->private_data;
 515                 ctlr_info_t *h = seq->private;
 516
 517                 err = cciss_engage_scsi(h);
 518                 if (err == 0)
 519                         err = length;
 520         } else
 521 #endif /* CONFIG_CISS_SCSI_TAPE */
 522                 err = -EINVAL;
 523         /* might be nice to have "disengage" too, but it's not
 524            safely possible. (only 1 module use count, lock issues.) */
 525
 526 out:
 527         free_page((unsigned long)buffer);
 528         return err;
 529 }
 530
 531 static const struct file_operations cciss_proc_fops = {
 532         .owner   = THIS_MODULE,
 533         .open    = cciss_seq_open,
 534         .read    = seq_read,
 535         .llseek  = seq_lseek,
 536         .release = seq_release,
 537         .write   = cciss_proc_write,
 538 };
 539
 540 static void __devinit cciss_procinit(ctlr_info_t *h)
 541 {
 542         struct proc_dir_entry *pde;
 543
 544         if (proc_cciss == NULL)
 545                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 546         if (!proc_cciss)
 547                 return;
 548         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 549                                         S_IROTH, proc_cciss,
 550                                         &cciss_proc_fops, h);
 551 }
 552 #endif                          /* CONFIG_PROC_FS */
 553
 554 #define MAX_PRODUCT_NAME_LEN 19
 555
 556 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 557 #define to_drv(n) container_of(n, drive_info_struct, dev)
 558
 559 static ssize_t host_store_rescan(struct device *dev,
 560                                  struct device_attribute *attr,
 561                                  const char *buf, size_t count)
 562 {
 563         struct ctlr_info *h = to_hba(dev);
 564
 565         add_to_scan_list(h);
 566         wake_up_process(cciss_scan_thread);
 567         wait_for_completion_interruptible(&h->scan_wait);
 568
 569         return count;
 570 }
 571 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 572
 573 static ssize_t dev_show_unique_id(struct device *dev,
 574                                  struct device_attribute *attr,
 575                                  char *buf)
 576 {
 577         drive_info_struct *drv = to_drv(dev);
 578         struct ctlr_info *h = to_hba(drv->dev.parent);
 579         __u8 sn[16];
 580         unsigned long flags;
 581         int ret = 0;
 582
 583         spin_lock_irqsave(&h->lock, flags);
 584         if (h->busy_configuring)
 585                 ret = -EBUSY;
 586         else
 587                 memcpy(sn, drv->serial_no, sizeof(sn));
 588         spin_unlock_irqrestore(&h->lock, flags);
 589
 590         if (ret)
 591                 return ret;
 592         else
 593                 return snprintf(buf, 16 * 2 + 2,
 594                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 595                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 596                                 sn[0], sn[1], sn[2], sn[3],
 597                                 sn[4], sn[5], sn[6], sn[7],
 598                                 sn[8], sn[9], sn[10], sn[11],
 599                                 sn[12], sn[13], sn[14], sn[15]);
 600 }
 601 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 602
 603 static ssize_t dev_show_vendor(struct device *dev,
 604                                struct device_attribute *attr,
 605                                char *buf)
 606 {
 607         drive_info_struct *drv = to_drv(dev);
 608         struct ctlr_info *h = to_hba(drv->dev.parent);
 609         char vendor[VENDOR_LEN + 1];
 610         unsigned long flags;
 611         int ret = 0;
 612
 613         spin_lock_irqsave(&h->lock, flags);
 614         if (h->busy_configuring)
 615                 ret = -EBUSY;
 616         else
 617                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 618         spin_unlock_irqrestore(&h->lock, flags);
 619
 620         if (ret)
 621                 return ret;
 622         else
 623                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 624 }
 625 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 626
 627 static ssize_t dev_show_model(struct device *dev,
 628                               struct device_attribute *attr,
 629                               char *buf)
 630 {
 631         drive_info_struct *drv = to_drv(dev);
 632         struct ctlr_info *h = to_hba(drv->dev.parent);
 633         char model[MODEL_LEN + 1];
 634         unsigned long flags;
 635         int ret = 0;
 636
 637         spin_lock_irqsave(&h->lock, flags);
 638         if (h->busy_configuring)
 639                 ret = -EBUSY;
 640         else
 641                 memcpy(model, drv->model, MODEL_LEN + 1);
 642         spin_unlock_irqrestore(&h->lock, flags);
 643
 644         if (ret)
 645                 return ret;
 646         else
 647                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 648 }
 649 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 650
 651 static ssize_t dev_show_rev(struct device *dev,
 652                             struct device_attribute *attr,
 653                             char *buf)
 654 {
 655         drive_info_struct *drv = to_drv(dev);
 656         struct ctlr_info *h = to_hba(drv->dev.parent);
 657         char rev[REV_LEN + 1];
 658         unsigned long flags;
 659         int ret = 0;
 660
 661         spin_lock_irqsave(&h->lock, flags);
 662         if (h->busy_configuring)
 663                 ret = -EBUSY;
 664         else
 665                 memcpy(rev, drv->rev, REV_LEN + 1);
 666         spin_unlock_irqrestore(&h->lock, flags);
 667
 668         if (ret)
 669                 return ret;
 670         else
 671                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 672 }
 673 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 674
 675 static ssize_t cciss_show_lunid(struct device *dev,
 676                                 struct device_attribute *attr, char *buf)
 677 {
 678         drive_info_struct *drv = to_drv(dev);
 679         struct ctlr_info *h = to_hba(drv->dev.parent);
 680         unsigned long flags;
 681         unsigned char lunid[8];
 682
 683         spin_lock_irqsave(&h->lock, flags);
 684         if (h->busy_configuring) {
 685                 spin_unlock_irqrestore(&h->lock, flags);
 686                 return -EBUSY;
 687         }
 688         if (!drv->heads) {
 689                 spin_unlock_irqrestore(&h->lock, flags);
 690                 return -ENOTTY;
 691         }
 692         memcpy(lunid, drv->LunID, sizeof(lunid));
 693         spin_unlock_irqrestore(&h->lock, flags);
 694         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 695                 lunid[0], lunid[1], lunid[2], lunid[3],
 696                 lunid[4], lunid[5], lunid[6], lunid[7]);
 697 }
 698 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 699
 700 static ssize_t cciss_show_raid_level(struct device *dev,
 701                                      struct device_attribute *attr, char *buf)
 702 {
 703         drive_info_struct *drv = to_drv(dev);
 704         struct ctlr_info *h = to_hba(drv->dev.parent);
 705         int raid;
 706         unsigned long flags;
 707
 708         spin_lock_irqsave(&h->lock, flags);
 709         if (h->busy_configuring) {
 710                 spin_unlock_irqrestore(&h->lock, flags);
 711                 return -EBUSY;
 712         }
 713         raid = drv->raid_level;
 714         spin_unlock_irqrestore(&h->lock, flags);
 715         if (raid < 0 || raid > RAID_UNKNOWN)
 716                 raid = RAID_UNKNOWN;
 717
 718         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 719                         raid_label[raid]);
 720 }
 721 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 722
 723 static ssize_t cciss_show_usage_count(struct device *dev,
 724                                       struct device_attribute *attr, char *buf)
 725 {
 726         drive_info_struct *drv = to_drv(dev);
 727         struct ctlr_info *h = to_hba(drv->dev.parent);
 728         unsigned long flags;
 729         int count;
 730
 731         spin_lock_irqsave(&h->lock, flags);
 732         if (h->busy_configuring) {
 733                 spin_unlock_irqrestore(&h->lock, flags);
 734                 return -EBUSY;
 735         }
 736         count = drv->usage_count;
 737         spin_unlock_irqrestore(&h->lock, flags);
 738         return snprintf(buf, 20, "%d\n", count);
 739 }
 740 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 741
 742 static struct attribute *cciss_host_attrs[] = {
 743         &dev_attr_rescan.attr,
 744         NULL
 745 };
 746
 747 static struct attribute_group cciss_host_attr_group = {
 748         .attrs = cciss_host_attrs,
 749 };
 750
 751 static const struct attribute_group *cciss_host_attr_groups[] = {
 752         &cciss_host_attr_group,
 753         NULL
 754 };
 755
 756 static struct device_type cciss_host_type = {
 757         .name           = "cciss_host",
 758         .groups         = cciss_host_attr_groups,
 759         .release        = cciss_hba_release,
 760 };
 761
 762 static struct attribute *cciss_dev_attrs[] = {
 763         &dev_attr_unique_id.attr,
 764         &dev_attr_model.attr,
 765         &dev_attr_vendor.attr,
 766         &dev_attr_rev.attr,
 767         &dev_attr_lunid.attr,
 768         &dev_attr_raid_level.attr,
 769         &dev_attr_usage_count.attr,
 770         NULL
 771 };
 772
 773 static struct attribute_group cciss_dev_attr_group = {
 774         .attrs = cciss_dev_attrs,
 775 };
 776
 777 static const struct attribute_group *cciss_dev_attr_groups[] = {
 778         &cciss_dev_attr_group,
 779         NULL
 780 };
 781
 782 static struct device_type cciss_dev_type = {
 783         .name           = "cciss_device",
 784         .groups         = cciss_dev_attr_groups,
 785         .release        = cciss_device_release,
 786 };
 787
 788 static struct bus_type cciss_bus_type = {
 789         .name           = "cciss",
 790 };
 791
 792 /*
 793  * cciss_hba_release is called when the reference count
 794  * of h->dev goes to zero.
 795  */
 796 static void cciss_hba_release(struct device *dev)
 797 {
 798         /*
 799          * nothing to do, but need this to avoid a warning
 800          * about not having a release handler from lib/kref.c.
 801          */
 802 }
 803
 804 /*
 805  * Initialize sysfs entry for each controller.  This sets up and registers
 806  * the 'cciss#' directory for each individual controller under
 807  * /sys/bus/pci/devices/<dev>/.
 808  */
 809 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 810 {
 811         device_initialize(&h->dev);
 812         h->dev.type = &cciss_host_type;
 813         h->dev.bus = &cciss_bus_type;
 814         dev_set_name(&h->dev, "%s", h->devname);
 815         h->dev.parent = &h->pdev->dev;
 816
 817         return device_add(&h->dev);
 818 }
 819
 820 /*
 821  * Remove sysfs entries for an hba.
 822  */
 823 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 824 {
 825         device_del(&h->dev);
 826         put_device(&h->dev); /* final put. */
 827 }
 828
 829 /* cciss_device_release is called when the reference count
 830  * of h->drv[x]dev goes to zero.
 831  */
 832 static void cciss_device_release(struct device *dev)
 833 {
 834         drive_info_struct *drv = to_drv(dev);
 835         kfree(drv);
 836 }
 837
 838 /*
 839  * Initialize sysfs for each logical drive.  This sets up and registers
 840  * the 'c#d#' directory for each individual logical drive under
 841  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 842  * /sys/block/cciss!c#d# to this entry.
 843  */
 844 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 845                                        int drv_index)
 846 {
 847         struct device *dev;
 848
 849         if (h->drv[drv_index]->device_initialized)
 850                 return 0;
 851
 852         dev = &h->drv[drv_index]->dev;
 853         device_initialize(dev);
 854         dev->type = &cciss_dev_type;
 855         dev->bus = &cciss_bus_type;
 856         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 857         dev->parent = &h->dev;
 858         h->drv[drv_index]->device_initialized = 1;
 859         return device_add(dev);
 860 }
 861
 862 /*
 863  * Remove sysfs entries for a logical drive.
 864  */
 865 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 866         int ctlr_exiting)
 867 {
 868         struct device *dev = &h->drv[drv_index]->dev;
 869
 870         /* special case for c*d0, we only destroy it on controller exit */
 871         if (drv_index == 0 && !ctlr_exiting)
 872                 return;
 873
 874         device_del(dev);
 875         put_device(dev); /* the "final" put. */
 876         h->drv[drv_index] = NULL;
 877 }
 878
 879 /*
 880  * For operations that cannot sleep, a command block is allocated at init,
 881  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 882  * which ones are free or in use.
 883  */
 884 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 885 {
 886         CommandList_struct *c;
 887         int i;
 888         u64bit temp64;
 889         dma_addr_t cmd_dma_handle, err_dma_handle;
 890
 891         do {
 892                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 893                 if (i == h->nr_cmds)
 894                         return NULL;
 895         } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
 896                   h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 897         c = h->cmd_pool + i;
 898         memset(c, 0, sizeof(CommandList_struct));
 899         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
 900         c->err_info = h->errinfo_pool + i;
 901         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 902         err_dma_handle = h->errinfo_pool_dhandle
 903             + i * sizeof(ErrorInfo_struct);
 904         h->nr_allocs++;
 905
 906         c->cmdindex = i;
 907
 908         INIT_HLIST_NODE(&c->list);
 909         c->busaddr = (__u32) cmd_dma_handle;
 910         temp64.val = (__u64) err_dma_handle;
 911         c->ErrDesc.Addr.lower = temp64.val32.lower;
 912         c->ErrDesc.Addr.upper = temp64.val32.upper;
 913         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 914
 915         c->ctlr = h->ctlr;
 916         return c;
 917 }
 918
 919 /* allocate a command using pci_alloc_consistent, used for ioctls,
 920  * etc., not for the main i/o path.
 921  */
 922 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
 923 {
 924         CommandList_struct *c;
 925         u64bit temp64;
 926         dma_addr_t cmd_dma_handle, err_dma_handle;
 927
 928         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 929                 sizeof(CommandList_struct), &cmd_dma_handle);
 930         if (c == NULL)
 931                 return NULL;
 932         memset(c, 0, sizeof(CommandList_struct));
 933
 934         c->cmdindex = -1;
 935
 936         c->err_info = (ErrorInfo_struct *)
 937             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 938                     &err_dma_handle);
 939
 940         if (c->err_info == NULL) {
 941                 pci_free_consistent(h->pdev,
 942                         sizeof(CommandList_struct), c, cmd_dma_handle);
 943                 return NULL;
 944         }
 945         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 946
 947         INIT_HLIST_NODE(&c->list);
 948         c->busaddr = (__u32) cmd_dma_handle;
 949         temp64.val = (__u64) err_dma_handle;
 950         c->ErrDesc.Addr.lower = temp64.val32.lower;
 951         c->ErrDesc.Addr.upper = temp64.val32.upper;
 952         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 953
 954         c->ctlr = h->ctlr;
 955         return c;
 956 }
 957
 958 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
 959 {
 960         int i;
 961
 962         i = c - h->cmd_pool;
 963         clear_bit(i & (BITS_PER_LONG - 1),
 964                   h->cmd_pool_bits + (i / BITS_PER_LONG));
 965         h->nr_frees++;
 966 }
 967
 968 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
 969 {
 970         u64bit temp64;
 971
 972         temp64.val32.lower = c->ErrDesc.Addr.lower;
 973         temp64.val32.upper = c->ErrDesc.Addr.upper;
 974         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 975                             c->err_info, (dma_addr_t) temp64.val);
 976         pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 977                             c, (dma_addr_t) c->busaddr);
 978 }
 979
 980 static inline ctlr_info_t *get_host(struct gendisk *disk)
 981 {
 982         return disk->queue->queuedata;
 983 }
 984
 985 static inline drive_info_struct *get_drv(struct gendisk *disk)
 986 {
 987         return disk->private_data;
 988 }
 989
 990 /*
 991  * Open.  Make sure the device is really there.
 992  */
 993 static int cciss_open(struct block_device *bdev, fmode_t mode)
 994 {
 995         ctlr_info_t *h = get_host(bdev->bd_disk);
 996         drive_info_struct *drv = get_drv(bdev->bd_disk);
 997
 998         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
 999         if (drv->busy_configuring)
1000                 return -EBUSY;
1001         /*
1002          * Root is allowed to open raw volume zero even if it's not configured
1003          * so array config can still work. Root is also allowed to open any
1004          * volume that has a LUN ID, so it can issue IOCTL to reread the
1005          * disk information.  I don't think I really like this
1006          * but I'm already using way to many device nodes to claim another one
1007          * for "raw controller".
1008          */
1009         if (drv->heads == 0) {
1010                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1011                         /* if not node 0 make sure it is a partition = 0 */
1012                         if (MINOR(bdev->bd_dev) & 0x0f) {
1013                                 return -ENXIO;
1014                                 /* if it is, make sure we have a LUN ID */
1015                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1016                                 sizeof(drv->LunID))) {
1017                                 return -ENXIO;
1018                         }
1019                 }
1020                 if (!capable(CAP_SYS_ADMIN))
1021                         return -EPERM;
1022         }
1023         drv->usage_count++;
1024         h->usage_count++;
1025         return 0;
1026 }
1027
1028 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1029 {
1030         int ret;
1031
1032         mutex_lock(&cciss_mutex);
1033         ret = cciss_open(bdev, mode);
1034         mutex_unlock(&cciss_mutex);
1035
1036         return ret;
1037 }
1038
1039 /*
1040  * Close.  Sync first.
1041  */
1042 static int cciss_release(struct gendisk *disk, fmode_t mode)
1043 {
1044         ctlr_info_t *h;
1045         drive_info_struct *drv;
1046
1047         mutex_lock(&cciss_mutex);
1048         h = get_host(disk);
1049         drv = get_drv(disk);
1050         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1051         drv->usage_count--;
1052         h->usage_count--;
1053         mutex_unlock(&cciss_mutex);
1054         return 0;
1055 }
1056
1057 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1058                     unsigned cmd, unsigned long arg)
1059 {
1060         int ret;
1061         mutex_lock(&cciss_mutex);
1062         ret = cciss_ioctl(bdev, mode, cmd, arg);
1063         mutex_unlock(&cciss_mutex);
1064         return ret;
1065 }
1066
1067 #ifdef CONFIG_COMPAT
1068
1069 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1070                                   unsigned cmd, unsigned long arg);
1071 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1072                                       unsigned cmd, unsigned long arg);
1073
1074 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1075                               unsigned cmd, unsigned long arg)
1076 {
1077         switch (cmd) {
1078         case CCISS_GETPCIINFO:
1079         case CCISS_GETINTINFO:
1080         case CCISS_SETINTINFO:
1081         case CCISS_GETNODENAME:
1082         case CCISS_SETNODENAME:
1083         case CCISS_GETHEARTBEAT:
1084         case CCISS_GETBUSTYPES:
1085         case CCISS_GETFIRMVER:
1086         case CCISS_GETDRIVVER:
1087         case CCISS_REVALIDVOLS:
1088         case CCISS_DEREGDISK:
1089         case CCISS_REGNEWDISK:
1090         case CCISS_REGNEWD:
1091         case CCISS_RESCANDISK:
1092         case CCISS_GETLUNINFO:
1093                 return do_ioctl(bdev, mode, cmd, arg);
1094
1095         case CCISS_PASSTHRU32:
1096                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1097         case CCISS_BIG_PASSTHRU32:
1098                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1099
1100         default:
1101                 return -ENOIOCTLCMD;
1102         }
1103 }
1104
1105 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1106                                   unsigned cmd, unsigned long arg)
1107 {
1108         IOCTL32_Command_struct __user *arg32 =
1109             (IOCTL32_Command_struct __user *) arg;
1110         IOCTL_Command_struct arg64;
1111         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1112         int err;
1113         u32 cp;
1114
1115         err = 0;
1116         err |=
1117             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1118                            sizeof(arg64.LUN_info));
1119         err |=
1120             copy_from_user(&arg64.Request, &arg32->Request,
1121                            sizeof(arg64.Request));
1122         err |=
1123             copy_from_user(&arg64.error_info, &arg32->error_info,
1124                            sizeof(arg64.error_info));
1125         err |= get_user(arg64.buf_size, &arg32->buf_size);
1126         err |= get_user(cp, &arg32->buf);
1127         arg64.buf = compat_ptr(cp);
1128         err |= copy_to_user(p, &arg64, sizeof(arg64));
1129
1130         if (err)
1131                 return -EFAULT;
1132
1133         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1134         if (err)
1135                 return err;
1136         err |=
1137             copy_in_user(&arg32->error_info, &p->error_info,
1138                          sizeof(arg32->error_info));
1139         if (err)
1140                 return -EFAULT;
1141         return err;
1142 }
1143
1144 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1145                                       unsigned cmd, unsigned long arg)
1146 {
1147         BIG_IOCTL32_Command_struct __user *arg32 =
1148             (BIG_IOCTL32_Command_struct __user *) arg;
1149         BIG_IOCTL_Command_struct arg64;
1150         BIG_IOCTL_Command_struct __user *p =
1151             compat_alloc_user_space(sizeof(arg64));
1152         int err;
1153         u32 cp;
1154
1155         memset(&arg64, 0, sizeof(arg64));
1156         err = 0;
1157         err |=
1158             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1159                            sizeof(arg64.LUN_info));
1160         err |=
1161             copy_from_user(&arg64.Request, &arg32->Request,
1162                            sizeof(arg64.Request));
1163         err |=
1164             copy_from_user(&arg64.error_info, &arg32->error_info,
1165                            sizeof(arg64.error_info));
1166         err |= get_user(arg64.buf_size, &arg32->buf_size);
1167         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1168         err |= get_user(cp, &arg32->buf);
1169         arg64.buf = compat_ptr(cp);
1170         err |= copy_to_user(p, &arg64, sizeof(arg64));
1171
1172         if (err)
1173                 return -EFAULT;
1174
1175         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1176         if (err)
1177                 return err;
1178         err |=
1179             copy_in_user(&arg32->error_info, &p->error_info,
1180                          sizeof(arg32->error_info));
1181         if (err)
1182                 return -EFAULT;
1183         return err;
1184 }
1185 #endif
1186
1187 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1188 {
1189         drive_info_struct *drv = get_drv(bdev->bd_disk);
1190
1191         if (!drv->cylinders)
1192                 return -ENXIO;
1193
1194         geo->heads = drv->heads;
1195         geo->sectors = drv->sectors;
1196         geo->cylinders = drv->cylinders;
1197         return 0;
1198 }
1199
1200 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1201 {
1202         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1203                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1204                 (void)check_for_unit_attention(h, c);
1205 }
1206
1207 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1208 {
1209         cciss_pci_info_struct pciinfo;
1210
1211         if (!argp)
1212                 return -EINVAL;
1213         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1214         pciinfo.bus = h->pdev->bus->number;
1215         pciinfo.dev_fn = h->pdev->devfn;
1216         pciinfo.board_id = h->board_id;
1217         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1218                 return -EFAULT;
1219         return 0;
1220 }
1221
1222 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1223 {
1224         cciss_coalint_struct intinfo;
1225
1226         if (!argp)
1227                 return -EINVAL;
1228         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1229         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1230         if (copy_to_user
1231             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1232                 return -EFAULT;
1233         return 0;
1234 }
1235
1236 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1237 {
1238         cciss_coalint_struct intinfo;
1239         unsigned long flags;
1240         int i;
1241
1242         if (!argp)
1243                 return -EINVAL;
1244         if (!capable(CAP_SYS_ADMIN))
1245                 return -EPERM;
1246         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1247                 return -EFAULT;
1248         if ((intinfo.delay == 0) && (intinfo.count == 0))
1249                 return -EINVAL;
1250         spin_lock_irqsave(&h->lock, flags);
1251         /* Update the field, and then ring the doorbell */
1252         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1253         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1254         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1255
1256         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1257                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1258                         break;
1259                 udelay(1000); /* delay and try again */
1260         }
1261         spin_unlock_irqrestore(&h->lock, flags);
1262         if (i >= MAX_IOCTL_CONFIG_WAIT)
1263                 return -EAGAIN;
1264         return 0;
1265 }
1266
1267 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1268 {
1269         NodeName_type NodeName;
1270         int i;
1271
1272         if (!argp)
1273                 return -EINVAL;
1274         for (i = 0; i < 16; i++)
1275                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1276         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1277                 return -EFAULT;
1278         return 0;
1279 }
1280
1281 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1282 {
1283         NodeName_type NodeName;
1284         unsigned long flags;
1285         int i;
1286
1287         if (!argp)
1288                 return -EINVAL;
1289         if (!capable(CAP_SYS_ADMIN))
1290                 return -EPERM;
1291         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1292                 return -EFAULT;
1293         spin_lock_irqsave(&h->lock, flags);
1294         /* Update the field, and then ring the doorbell */
1295         for (i = 0; i < 16; i++)
1296                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1297         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1298         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1299                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1300                         break;
1301                 udelay(1000); /* delay and try again */
1302         }
1303         spin_unlock_irqrestore(&h->lock, flags);
1304         if (i >= MAX_IOCTL_CONFIG_WAIT)
1305                 return -EAGAIN;
1306         return 0;
1307 }
1308
1309 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1310 {
1311         Heartbeat_type heartbeat;
1312
1313         if (!argp)
1314                 return -EINVAL;
1315         heartbeat = readl(&h->cfgtable->HeartBeat);
1316         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1317                 return -EFAULT;
1318         return 0;
1319 }
1320
1321 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1322 {
1323         BusTypes_type BusTypes;
1324
1325         if (!argp)
1326                 return -EINVAL;
1327         BusTypes = readl(&h->cfgtable->BusTypes);
1328         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1329                 return -EFAULT;
1330         return 0;
1331 }
1332
1333 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1334 {
1335         FirmwareVer_type firmware;
1336
1337         if (!argp)
1338                 return -EINVAL;
1339         memcpy(firmware, h->firm_ver, 4);
1340
1341         if (copy_to_user
1342             (argp, firmware, sizeof(FirmwareVer_type)))
1343                 return -EFAULT;
1344         return 0;
1345 }
1346
1347 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1348 {
1349         DriverVer_type DriverVer = DRIVER_VERSION;
1350
1351         if (!argp)
1352                 return -EINVAL;
1353         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1354                 return -EFAULT;
1355         return 0;
1356 }
1357
1358 static int cciss_getluninfo(ctlr_info_t *h,
1359         struct gendisk *disk, void __user *argp)
1360 {
1361         LogvolInfo_struct luninfo;
1362         drive_info_struct *drv = get_drv(disk);
1363
1364         if (!argp)
1365                 return -EINVAL;
1366         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1367         luninfo.num_opens = drv->usage_count;
1368         luninfo.num_parts = 0;
1369         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1370                 return -EFAULT;
1371         return 0;
1372 }
1373
1374 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1375 {
1376         IOCTL_Command_struct iocommand;
1377         CommandList_struct *c;
1378         char *buff = NULL;
1379         u64bit temp64;
1380         DECLARE_COMPLETION_ONSTACK(wait);
1381
1382         if (!argp)
1383                 return -EINVAL;
1384
1385         if (!capable(CAP_SYS_RAWIO))
1386                 return -EPERM;
1387
1388         if (copy_from_user
1389             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1390                 return -EFAULT;
1391         if ((iocommand.buf_size < 1) &&
1392             (iocommand.Request.Type.Direction != XFER_NONE)) {
1393                 return -EINVAL;
1394         }
1395         if (iocommand.buf_size > 0) {
1396                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1397                 if (buff == NULL)
1398                         return -EFAULT;
1399         }
1400         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1401                 /* Copy the data into the buffer we created */
1402                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1403                         kfree(buff);
1404                         return -EFAULT;
1405                 }
1406         } else {
1407                 memset(buff, 0, iocommand.buf_size);
1408         }
1409         c = cmd_special_alloc(h);
1410         if (!c) {
1411                 kfree(buff);
1412                 return -ENOMEM;
1413         }
1414         /* Fill in the command type */
1415         c->cmd_type = CMD_IOCTL_PEND;
1416         /* Fill in Command Header */
1417         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1418         if (iocommand.buf_size > 0) { /* buffer to fill */
1419                 c->Header.SGList = 1;
1420                 c->Header.SGTotal = 1;
1421         } else { /* no buffers to fill */
1422                 c->Header.SGList = 0;
1423                 c->Header.SGTotal = 0;
1424         }
1425         c->Header.LUN = iocommand.LUN_info;
1426         /* use the kernel address the cmd block for tag */
1427         c->Header.Tag.lower = c->busaddr;
1428
1429         /* Fill in Request block */
1430         c->Request = iocommand.Request;
1431
1432         /* Fill in the scatter gather information */
1433         if (iocommand.buf_size > 0) {
1434                 temp64.val = pci_map_single(h->pdev, buff,
1435                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1436                 c->SG[0].Addr.lower = temp64.val32.lower;
1437                 c->SG[0].Addr.upper = temp64.val32.upper;
1438                 c->SG[0].Len = iocommand.buf_size;
1439                 c->SG[0].Ext = 0;  /* we are not chaining */
1440         }
1441         c->waiting = &wait;
1442
1443         enqueue_cmd_and_start_io(h, c);
1444         wait_for_completion(&wait);
1445
1446         /* unlock the buffers from DMA */
1447         temp64.val32.lower = c->SG[0].Addr.lower;
1448         temp64.val32.upper = c->SG[0].Addr.upper;
1449         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1450                          PCI_DMA_BIDIRECTIONAL);
1451         check_ioctl_unit_attention(h, c);
1452
1453         /* Copy the error information out */
1454         iocommand.error_info = *(c->err_info);
1455         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1456                 kfree(buff);
1457                 cmd_special_free(h, c);
1458                 return -EFAULT;
1459         }
1460
1461         if (iocommand.Request.Type.Direction == XFER_READ) {
1462                 /* Copy the data out of the buffer we created */
1463                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1464                         kfree(buff);
1465                         cmd_special_free(h, c);
1466                         return -EFAULT;
1467                 }
1468         }
1469         kfree(buff);
1470         cmd_special_free(h, c);
1471         return 0;
1472 }
1473
1474 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1475 {
1476         BIG_IOCTL_Command_struct *ioc;
1477         CommandList_struct *c;
1478         unsigned char **buff = NULL;
1479         int *buff_size = NULL;
1480         u64bit temp64;
1481         BYTE sg_used = 0;
1482         int status = 0;
1483         int i;
1484         DECLARE_COMPLETION_ONSTACK(wait);
1485         __u32 left;
1486         __u32 sz;
1487         BYTE __user *data_ptr;
1488
1489         if (!argp)
1490                 return -EINVAL;
1491         if (!capable(CAP_SYS_RAWIO))
1492                 return -EPERM;
1493         ioc = (BIG_IOCTL_Command_struct *)
1494             kmalloc(sizeof(*ioc), GFP_KERNEL);
1495         if (!ioc) {
1496                 status = -ENOMEM;
1497                 goto cleanup1;
1498         }
1499         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1500                 status = -EFAULT;
1501                 goto cleanup1;
1502         }
1503         if ((ioc->buf_size < 1) &&
1504             (ioc->Request.Type.Direction != XFER_NONE)) {
1505                 status = -EINVAL;
1506                 goto cleanup1;
1507         }
1508         /* Check kmalloc limits  using all SGs */
1509         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1510                 status = -EINVAL;
1511                 goto cleanup1;
1512         }
1513         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1514                 status = -EINVAL;
1515                 goto cleanup1;
1516         }
1517         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1518         if (!buff) {
1519                 status = -ENOMEM;
1520                 goto cleanup1;
1521         }
1522         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1523         if (!buff_size) {
1524                 status = -ENOMEM;
1525                 goto cleanup1;
1526         }
1527         left = ioc->buf_size;
1528         data_ptr = ioc->buf;
1529         while (left) {
1530                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1531                 buff_size[sg_used] = sz;
1532                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1533                 if (buff[sg_used] == NULL) {
1534                         status = -ENOMEM;
1535                         goto cleanup1;
1536                 }
1537                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1538                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1539                                 status = -EFAULT;
1540                                 goto cleanup1;
1541                         }
1542                 } else {
1543                         memset(buff[sg_used], 0, sz);
1544                 }
1545                 left -= sz;
1546                 data_ptr += sz;
1547                 sg_used++;
1548         }
1549         c = cmd_special_alloc(h);
1550         if (!c) {
1551                 status = -ENOMEM;
1552                 goto cleanup1;
1553         }
1554         c->cmd_type = CMD_IOCTL_PEND;
1555         c->Header.ReplyQueue = 0;
1556         c->Header.SGList = sg_used;
1557         c->Header.SGTotal = sg_used;
1558         c->Header.LUN = ioc->LUN_info;
1559         c->Header.Tag.lower = c->busaddr;
1560
1561         c->Request = ioc->Request;
1562         for (i = 0; i < sg_used; i++) {
1563                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1564                                     PCI_DMA_BIDIRECTIONAL);
1565                 c->SG[i].Addr.lower = temp64.val32.lower;
1566                 c->SG[i].Addr.upper = temp64.val32.upper;
1567                 c->SG[i].Len = buff_size[i];
1568                 c->SG[i].Ext = 0;       /* we are not chaining */
1569         }
1570         c->waiting = &wait;
1571         enqueue_cmd_and_start_io(h, c);
1572         wait_for_completion(&wait);
1573         /* unlock the buffers from DMA */
1574         for (i = 0; i < sg_used; i++) {
1575                 temp64.val32.lower = c->SG[i].Addr.lower;
1576                 temp64.val32.upper = c->SG[i].Addr.upper;
1577                 pci_unmap_single(h->pdev,
1578                         (dma_addr_t) temp64.val, buff_size[i],
1579                         PCI_DMA_BIDIRECTIONAL);
1580         }
1581         check_ioctl_unit_attention(h, c);
1582         /* Copy the error information out */
1583         ioc->error_info = *(c->err_info);
1584         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1585                 cmd_special_free(h, c);
1586                 status = -EFAULT;
1587                 goto cleanup1;
1588         }
1589         if (ioc->Request.Type.Direction == XFER_READ) {
1590                 /* Copy the data out of the buffer we created */
1591                 BYTE __user *ptr = ioc->buf;
1592                 for (i = 0; i < sg_used; i++) {
1593                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1594                                 cmd_special_free(h, c);
1595                                 status = -EFAULT;
1596                                 goto cleanup1;
1597                         }
1598                         ptr += buff_size[i];
1599                 }
1600         }
1601         cmd_special_free(h, c);
1602         status = 0;
1603 cleanup1:
1604         if (buff) {
1605                 for (i = 0; i < sg_used; i++)
1606                         kfree(buff[i]);
1607                 kfree(buff);
1608         }
1609         kfree(buff_size);
1610         kfree(ioc);
1611         return status;
1612 }
1613
1614 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1615         unsigned int cmd, unsigned long arg)
1616 {
1617         struct gendisk *disk = bdev->bd_disk;
1618         ctlr_info_t *h = get_host(disk);
1619         void __user *argp = (void __user *)arg;
1620
1621         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1622                 cmd, arg);
1623         switch (cmd) {
1624         case CCISS_GETPCIINFO:
1625                 return cciss_getpciinfo(h, argp);
1626         case CCISS_GETINTINFO:
1627                 return cciss_getintinfo(h, argp);
1628         case CCISS_SETINTINFO:
1629                 return cciss_setintinfo(h, argp);
1630         case CCISS_GETNODENAME:
1631                 return cciss_getnodename(h, argp);
1632         case CCISS_SETNODENAME:
1633                 return cciss_setnodename(h, argp);
1634         case CCISS_GETHEARTBEAT:
1635                 return cciss_getheartbeat(h, argp);
1636         case CCISS_GETBUSTYPES:
1637                 return cciss_getbustypes(h, argp);
1638         case CCISS_GETFIRMVER:
1639                 return cciss_getfirmver(h, argp);
1640         case CCISS_GETDRIVVER:
1641                 return cciss_getdrivver(h, argp);
1642         case CCISS_DEREGDISK:
1643         case CCISS_REGNEWD:
1644         case CCISS_REVALIDVOLS:
1645                 return rebuild_lun_table(h, 0, 1);
1646         case CCISS_GETLUNINFO:
1647                 return cciss_getluninfo(h, disk, argp);
1648         case CCISS_PASSTHRU:
1649                 return cciss_passthru(h, argp);
1650         case CCISS_BIG_PASSTHRU:
1651                 return cciss_bigpassthru(h, argp);
1652
1653         /* scsi_cmd_ioctl handles these, below, though some are not */
1654         /* very meaningful for cciss.  SG_IO is the main one people want. */
1655
1656         case SG_GET_VERSION_NUM:
1657         case SG_SET_TIMEOUT:
1658         case SG_GET_TIMEOUT:
1659         case SG_GET_RESERVED_SIZE:
1660         case SG_SET_RESERVED_SIZE:
1661         case SG_EMULATED_HOST:
1662         case SG_IO:
1663         case SCSI_IOCTL_SEND_COMMAND:
1664                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1665
1666         /* scsi_cmd_ioctl would normally handle these, below, but */
1667         /* they aren't a good fit for cciss, as CD-ROMs are */
1668         /* not supported, and we don't have any bus/target/lun */
1669         /* which we present to the kernel. */
1670
1671         case CDROM_SEND_PACKET:
1672         case CDROMCLOSETRAY:
1673         case CDROMEJECT:
1674         case SCSI_IOCTL_GET_IDLUN:
1675         case SCSI_IOCTL_GET_BUS_NUMBER:
1676         default:
1677                 return -ENOTTY;
1678         }
1679 }
1680
1681 static void cciss_check_queues(ctlr_info_t *h)
1682 {
1683         int start_queue = h->next_to_run;
1684         int i;
1685
1686         /* check to see if we have maxed out the number of commands that can
1687          * be placed on the queue.  If so then exit.  We do this check here
1688          * in case the interrupt we serviced was from an ioctl and did not
1689          * free any new commands.
1690          */
1691         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1692                 return;
1693
1694         /* We have room on the queue for more commands.  Now we need to queue
1695          * them up.  We will also keep track of the next queue to run so
1696          * that every queue gets a chance to be started first.
1697          */
1698         for (i = 0; i < h->highest_lun + 1; i++) {
1699                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1700                 /* make sure the disk has been added and the drive is real
1701                  * because this can be called from the middle of init_one.
1702                  */
1703                 if (!h->drv[curr_queue])
1704                         continue;
1705                 if (!(h->drv[curr_queue]->queue) ||
1706                         !(h->drv[curr_queue]->heads))
1707                         continue;
1708                 blk_start_queue(h->gendisk[curr_queue]->queue);
1709
1710                 /* check to see if we have maxed out the number of commands
1711                  * that can be placed on the queue.
1712                  */
1713                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1714                         if (curr_queue == start_queue) {
1715                                 h->next_to_run =
1716                                     (start_queue + 1) % (h->highest_lun + 1);
1717                                 break;
1718                         } else {
1719                                 h->next_to_run = curr_queue;
1720                                 break;
1721                         }
1722                 }
1723         }
1724 }
1725
1726 static void cciss_softirq_done(struct request *rq)
1727 {
1728         CommandList_struct *c = rq->completion_data;
1729         ctlr_info_t *h = hba[c->ctlr];
1730         SGDescriptor_struct *curr_sg = c->SG;
1731         u64bit temp64;
1732         unsigned long flags;
1733         int i, ddir;
1734         int sg_index = 0;
1735
1736         if (c->Request.Type.Direction == XFER_READ)
1737                 ddir = PCI_DMA_FROMDEVICE;
1738         else
1739                 ddir = PCI_DMA_TODEVICE;
1740
1741         /* command did not need to be retried */
1742         /* unmap the DMA mapping for all the scatter gather elements */
1743         for (i = 0; i < c->Header.SGList; i++) {
1744                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1745                         cciss_unmap_sg_chain_block(h, c);
1746                         /* Point to the next block */
1747                         curr_sg = h->cmd_sg_list[c->cmdindex];
1748                         sg_index = 0;
1749                 }
1750                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1751                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1752                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1753                                 ddir);
1754                 ++sg_index;
1755         }
1756
1757         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1758
1759         /* set the residual count for pc requests */
1760         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1761                 rq->resid_len = c->err_info->ResidualCnt;
1762
1763         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1764
1765         spin_lock_irqsave(&h->lock, flags);
1766         cmd_free(h, c);
1767         cciss_check_queues(h);
1768         spin_unlock_irqrestore(&h->lock, flags);
1769 }
1770
1771 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1772         unsigned char scsi3addr[], uint32_t log_unit)
1773 {
1774         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1775                 sizeof(h->drv[log_unit]->LunID));
1776 }
1777
1778 /* This function gets the SCSI vendor, model, and revision of a logical drive
1779  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1780  * they cannot be read.
1781  */
1782 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1783                                    char *vendor, char *model, char *rev)
1784 {
1785         int rc;
1786         InquiryData_struct *inq_buf;
1787         unsigned char scsi3addr[8];
1788
1789         *vendor = '\0';
1790         *model = '\0';
1791         *rev = '\0';
1792
1793         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1794         if (!inq_buf)
1795                 return;
1796
1797         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1798         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1799                         scsi3addr, TYPE_CMD);
1800         if (rc == IO_OK) {
1801                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1802                 vendor[VENDOR_LEN] = '\0';
1803                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1804                 model[MODEL_LEN] = '\0';
1805                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1806                 rev[REV_LEN] = '\0';
1807         }
1808
1809         kfree(inq_buf);
1810         return;
1811 }
1812
1813 /* This function gets the serial number of a logical drive via
1814  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1815  * number cannot be had, for whatever reason, 16 bytes of 0xff
1816  * are returned instead.
1817  */
1818 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1819                                 unsigned char *serial_no, int buflen)
1820 {
1821 #define PAGE_83_INQ_BYTES 64
1822         int rc;
1823         unsigned char *buf;
1824         unsigned char scsi3addr[8];
1825
1826         if (buflen > 16)
1827                 buflen = 16;
1828         memset(serial_no, 0xff, buflen);
1829         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1830         if (!buf)
1831                 return;
1832         memset(serial_no, 0, buflen);
1833         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1834         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1835                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1836         if (rc == IO_OK)
1837                 memcpy(serial_no, &buf[8], buflen);
1838         kfree(buf);
1839         return;
1840 }
1841
1842 /*
1843  * cciss_add_disk sets up the block device queue for a logical drive
1844  */
1845 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1846                                 int drv_index)
1847 {
1848         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1849         if (!disk->queue)
1850                 goto init_queue_failure;
1851         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1852         disk->major = h->major;
1853         disk->first_minor = drv_index << NWD_SHIFT;
1854         disk->fops = &cciss_fops;
1855         if (cciss_create_ld_sysfs_entry(h, drv_index))
1856                 goto cleanup_queue;
1857         disk->private_data = h->drv[drv_index];
1858         disk->driverfs_dev = &h->drv[drv_index]->dev;
1859
1860         /* Set up queue information */
1861         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1862
1863         /* This is a hardware imposed limit. */
1864         blk_queue_max_segments(disk->queue, h->maxsgentries);
1865
1866         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1867
1868         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1869
1870         disk->queue->queuedata = h;
1871
1872         blk_queue_logical_block_size(disk->queue,
1873                                      h->drv[drv_index]->block_size);
1874
1875         /* Make sure all queue data is written out before */
1876         /* setting h->drv[drv_index]->queue, as setting this */
1877         /* allows the interrupt handler to start the queue */
1878         wmb();
1879         h->drv[drv_index]->queue = disk->queue;
1880         add_disk(disk);
1881         return 0;
1882
1883 cleanup_queue:
1884         blk_cleanup_queue(disk->queue);
1885         disk->queue = NULL;
1886 init_queue_failure:
1887         return -1;
1888 }
1889
1890 /* This function will check the usage_count of the drive to be updated/added.
1891  * If the usage_count is zero and it is a heretofore unknown drive, or,
1892  * the drive's capacity, geometry, or serial number has changed,
1893  * then the drive information will be updated and the disk will be
1894  * re-registered with the kernel.  If these conditions don't hold,
1895  * then it will be left alone for the next reboot.  The exception to this
1896  * is disk 0 which will always be left registered with the kernel since it
1897  * is also the controller node.  Any changes to disk 0 will show up on
1898  * the next reboot.
1899  */
1900 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1901         int first_time, int via_ioctl)
1902 {
1903         struct gendisk *disk;
1904         InquiryData_struct *inq_buff = NULL;
1905         unsigned int block_size;
1906         sector_t total_size;
1907         unsigned long flags = 0;
1908         int ret = 0;
1909         drive_info_struct *drvinfo;
1910
1911         /* Get information about the disk and modify the driver structure */
1912         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1913         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1914         if (inq_buff == NULL || drvinfo == NULL)
1915                 goto mem_msg;
1916
1917         /* testing to see if 16-byte CDBs are already being used */
1918         if (h->cciss_read == CCISS_READ_16) {
1919                 cciss_read_capacity_16(h, drv_index,
1920                         &total_size, &block_size);
1921
1922         } else {
1923                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1924                 /* if read_capacity returns all F's this volume is >2TB */
1925                 /* in size so we switch to 16-byte CDB's for all */
1926                 /* read/write ops */
1927                 if (total_size == 0xFFFFFFFFULL) {
1928                         cciss_read_capacity_16(h, drv_index,
1929                         &total_size, &block_size);
1930                         h->cciss_read = CCISS_READ_16;
1931                         h->cciss_write = CCISS_WRITE_16;
1932                 } else {
1933                         h->cciss_read = CCISS_READ_10;
1934                         h->cciss_write = CCISS_WRITE_10;
1935                 }
1936         }
1937
1938         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1939                                inq_buff, drvinfo);
1940         drvinfo->block_size = block_size;
1941         drvinfo->nr_blocks = total_size + 1;
1942
1943         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1944                                 drvinfo->model, drvinfo->rev);
1945         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1946                         sizeof(drvinfo->serial_no));
1947         /* Save the lunid in case we deregister the disk, below. */
1948         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1949                 sizeof(drvinfo->LunID));
1950
1951         /* Is it the same disk we already know, and nothing's changed? */
1952         if (h->drv[drv_index]->raid_level != -1 &&
1953                 ((memcmp(drvinfo->serial_no,
1954                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1955                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1956                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1957                 drvinfo->heads == h->drv[drv_index]->heads &&
1958                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1959                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1960                         /* The disk is unchanged, nothing to update */
1961                         goto freeret;
1962
1963         /* If we get here it's not the same disk, or something's changed,
1964          * so we need to * deregister it, and re-register it, if it's not
1965          * in use.
1966          * If the disk already exists then deregister it before proceeding
1967          * (unless it's the first disk (for the controller node).
1968          */
1969         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1970                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
1971                 spin_lock_irqsave(&h->lock, flags);
1972                 h->drv[drv_index]->busy_configuring = 1;
1973                 spin_unlock_irqrestore(&h->lock, flags);
1974
1975                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1976                  * which keeps the interrupt handler from starting
1977                  * the queue.
1978                  */
1979                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1980         }
1981
1982         /* If the disk is in use return */
1983         if (ret)
1984                 goto freeret;
1985
1986         /* Save the new information from cciss_geometry_inquiry
1987          * and serial number inquiry.  If the disk was deregistered
1988          * above, then h->drv[drv_index] will be NULL.
1989          */
1990         if (h->drv[drv_index] == NULL) {
1991                 drvinfo->device_initialized = 0;
1992                 h->drv[drv_index] = drvinfo;
1993                 drvinfo = NULL; /* so it won't be freed below. */
1994         } else {
1995                 /* special case for cxd0 */
1996                 h->drv[drv_index]->block_size = drvinfo->block_size;
1997                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
1998                 h->drv[drv_index]->heads = drvinfo->heads;
1999                 h->drv[drv_index]->sectors = drvinfo->sectors;
2000                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2001                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2002                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2003                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2004                         VENDOR_LEN + 1);
2005                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2006                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2007         }
2008
2009         ++h->num_luns;
2010         disk = h->gendisk[drv_index];
2011         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2012
2013         /* If it's not disk 0 (drv_index != 0)
2014          * or if it was disk 0, but there was previously
2015          * no actual corresponding configured logical drive
2016          * (raid_leve == -1) then we want to update the
2017          * logical drive's information.
2018          */
2019         if (drv_index || first_time) {
2020                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2021                         cciss_free_gendisk(h, drv_index);
2022                         cciss_free_drive_info(h, drv_index);
2023                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2024                                 drv_index);
2025                         --h->num_luns;
2026                 }
2027         }
2028
2029 freeret:
2030         kfree(inq_buff);
2031         kfree(drvinfo);
2032         return;
2033 mem_msg:
2034         dev_err(&h->pdev->dev, "out of memory\n");
2035         goto freeret;
2036 }
2037
2038 /* This function will find the first index of the controllers drive array
2039  * that has a null drv pointer and allocate the drive info struct and
2040  * will return that index   This is where new drives will be added.
2041  * If the index to be returned is greater than the highest_lun index for
2042  * the controller then highest_lun is set * to this new index.
2043  * If there are no available indexes or if tha allocation fails, then -1
2044  * is returned.  * "controller_node" is used to know if this is a real
2045  * logical drive, or just the controller node, which determines if this
2046  * counts towards highest_lun.
2047  */
2048 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2049 {
2050         int i;
2051         drive_info_struct *drv;
2052
2053         /* Search for an empty slot for our drive info */
2054         for (i = 0; i < CISS_MAX_LUN; i++) {
2055
2056                 /* if not cxd0 case, and it's occupied, skip it. */
2057                 if (h->drv[i] && i != 0)
2058                         continue;
2059                 /*
2060                  * If it's cxd0 case, and drv is alloc'ed already, and a
2061                  * disk is configured there, skip it.
2062                  */
2063                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2064                         continue;
2065
2066                 /*
2067                  * We've found an empty slot.  Update highest_lun
2068                  * provided this isn't just the fake cxd0 controller node.
2069                  */
2070                 if (i > h->highest_lun && !controller_node)
2071                         h->highest_lun = i;
2072
2073                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2074                 if (i == 0 && h->drv[i] != NULL)
2075                         return i;
2076
2077                 /*
2078                  * Found an empty slot, not already alloc'ed.  Allocate it.
2079                  * Mark it with raid_level == -1, so we know it's new later on.
2080                  */
2081                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2082                 if (!drv)
2083                         return -1;
2084                 drv->raid_level = -1; /* so we know it's new */
2085                 h->drv[i] = drv;
2086                 return i;
2087         }
2088         return -1;
2089 }
2090
2091 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2092 {
2093         kfree(h->drv[drv_index]);
2094         h->drv[drv_index] = NULL;
2095 }
2096
2097 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2098 {
2099         put_disk(h->gendisk[drv_index]);
2100         h->gendisk[drv_index] = NULL;
2101 }
2102
2103 /* cciss_add_gendisk finds a free hba[]->drv structure
2104  * and allocates a gendisk if needed, and sets the lunid
2105  * in the drvinfo structure.   It returns the index into
2106  * the ->drv[] array, or -1 if none are free.
2107  * is_controller_node indicates whether highest_lun should
2108  * count this disk, or if it's only being added to provide
2109  * a means to talk to the controller in case no logical
2110  * drives have yet been configured.
2111  */
2112 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2113         int controller_node)
2114 {
2115         int drv_index;
2116
2117         drv_index = cciss_alloc_drive_info(h, controller_node);
2118         if (drv_index == -1)
2119                 return -1;
2120
2121         /*Check if the gendisk needs to be allocated */
2122         if (!h->gendisk[drv_index]) {
2123                 h->gendisk[drv_index] =
2124                         alloc_disk(1 << NWD_SHIFT);
2125                 if (!h->gendisk[drv_index]) {
2126                         dev_err(&h->pdev->dev,
2127                                 "could not allocate a new disk %d\n",
2128                                 drv_index);
2129                         goto err_free_drive_info;
2130                 }
2131         }
2132         memcpy(h->drv[drv_index]->LunID, lunid,
2133                 sizeof(h->drv[drv_index]->LunID));
2134         if (cciss_create_ld_sysfs_entry(h, drv_index))
2135                 goto err_free_disk;
2136         /* Don't need to mark this busy because nobody */
2137         /* else knows about this disk yet to contend */
2138         /* for access to it. */
2139         h->drv[drv_index]->busy_configuring = 0;
2140         wmb();
2141         return drv_index;
2142
2143 err_free_disk:
2144         cciss_free_gendisk(h, drv_index);
2145 err_free_drive_info:
2146         cciss_free_drive_info(h, drv_index);
2147         return -1;
2148 }
2149
2150 /* This is for the special case of a controller which
2151  * has no logical drives.  In this case, we still need
2152  * to register a disk so the controller can be accessed
2153  * by the Array Config Utility.
2154  */
2155 static void cciss_add_controller_node(ctlr_info_t *h)
2156 {
2157         struct gendisk *disk;
2158         int drv_index;
2159
2160         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2161                 return;
2162
2163         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2164         if (drv_index == -1)
2165                 goto error;
2166         h->drv[drv_index]->block_size = 512;
2167         h->drv[drv_index]->nr_blocks = 0;
2168         h->drv[drv_index]->heads = 0;
2169         h->drv[drv_index]->sectors = 0;
2170         h->drv[drv_index]->cylinders = 0;
2171         h->drv[drv_index]->raid_level = -1;
2172         memset(h->drv[drv_index]->serial_no, 0, 16);
2173         disk = h->gendisk[drv_index];
2174         if (cciss_add_disk(h, disk, drv_index) == 0)
2175                 return;
2176         cciss_free_gendisk(h, drv_index);
2177         cciss_free_drive_info(h, drv_index);
2178 error:
2179         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2180         return;
2181 }
2182
2183 /* This function will add and remove logical drives from the Logical
2184  * drive array of the controller and maintain persistency of ordering
2185  * so that mount points are preserved until the next reboot.  This allows
2186  * for the removal of logical drives in the middle of the drive array
2187  * without a re-ordering of those drives.
2188  * INPUT
2189  * h            = The controller to perform the operations on
2190  */
2191 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2192         int via_ioctl)
2193 {
2194         int num_luns;
2195         ReportLunData_struct *ld_buff = NULL;
2196         int return_code;
2197         int listlength = 0;
2198         int i;
2199         int drv_found;
2200         int drv_index = 0;
2201         unsigned char lunid[8] = CTLR_LUNID;
2202         unsigned long flags;
2203
2204         if (!capable(CAP_SYS_RAWIO))
2205                 return -EPERM;
2206
2207         /* Set busy_configuring flag for this operation */
2208         spin_lock_irqsave(&h->lock, flags);
2209         if (h->busy_configuring) {
2210                 spin_unlock_irqrestore(&h->lock, flags);
2211                 return -EBUSY;
2212         }
2213         h->busy_configuring = 1;
2214         spin_unlock_irqrestore(&h->lock, flags);
2215
2216         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2217         if (ld_buff == NULL)
2218                 goto mem_msg;
2219
2220         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2221                                       sizeof(ReportLunData_struct),
2222                                       0, CTLR_LUNID, TYPE_CMD);
2223
2224         if (return_code == IO_OK)
2225                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2226         else {  /* reading number of logical volumes failed */
2227                 dev_warn(&h->pdev->dev,
2228                         "report logical volume command failed\n");
2229                 listlength = 0;
2230                 goto freeret;
2231         }
2232
2233         num_luns = listlength / 8;      /* 8 bytes per entry */
2234         if (num_luns > CISS_MAX_LUN) {
2235                 num_luns = CISS_MAX_LUN;
2236                 dev_warn(&h->pdev->dev, "more luns configured"
2237                        " on controller than can be handled by"
2238                        " this driver.\n");
2239         }
2240
2241         if (num_luns == 0)
2242                 cciss_add_controller_node(h);
2243
2244         /* Compare controller drive array to driver's drive array
2245          * to see if any drives are missing on the controller due
2246          * to action of Array Config Utility (user deletes drive)
2247          * and deregister logical drives which have disappeared.
2248          */
2249         for (i = 0; i <= h->highest_lun; i++) {
2250                 int j;
2251                 drv_found = 0;
2252
2253                 /* skip holes in the array from already deleted drives */
2254                 if (h->drv[i] == NULL)
2255                         continue;
2256
2257                 for (j = 0; j < num_luns; j++) {
2258                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2259                         if (memcmp(h->drv[i]->LunID, lunid,
2260                                 sizeof(lunid)) == 0) {
2261                                 drv_found = 1;
2262                                 break;
2263                         }
2264                 }
2265                 if (!drv_found) {
2266                         /* Deregister it from the OS, it's gone. */
2267                         spin_lock_irqsave(&h->lock, flags);
2268                         h->drv[i]->busy_configuring = 1;
2269                         spin_unlock_irqrestore(&h->lock, flags);
2270                         return_code = deregister_disk(h, i, 1, via_ioctl);
2271                         if (h->drv[i] != NULL)
2272                                 h->drv[i]->busy_configuring = 0;
2273                 }
2274         }
2275
2276         /* Compare controller drive array to driver's drive array.
2277          * Check for updates in the drive information and any new drives
2278          * on the controller due to ACU adding logical drives, or changing
2279          * a logical drive's size, etc.  Reregister any new/changed drives
2280          */
2281         for (i = 0; i < num_luns; i++) {
2282                 int j;
2283
2284                 drv_found = 0;
2285
2286                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2287                 /* Find if the LUN is already in the drive array
2288                  * of the driver.  If so then update its info
2289                  * if not in use.  If it does not exist then find
2290                  * the first free index and add it.
2291                  */
2292                 for (j = 0; j <= h->highest_lun; j++) {
2293                         if (h->drv[j] != NULL &&
2294                                 memcmp(h->drv[j]->LunID, lunid,
2295                                         sizeof(h->drv[j]->LunID)) == 0) {
2296                                 drv_index = j;
2297                                 drv_found = 1;
2298                                 break;
2299                         }
2300                 }
2301
2302                 /* check if the drive was found already in the array */
2303                 if (!drv_found) {
2304                         drv_index = cciss_add_gendisk(h, lunid, 0);
2305                         if (drv_index == -1)
2306                                 goto freeret;
2307                 }
2308                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2309         }               /* end for */
2310
2311 freeret:
2312         kfree(ld_buff);
2313         h->busy_configuring = 0;
2314         /* We return -1 here to tell the ACU that we have registered/updated
2315          * all of the drives that we can and to keep it from calling us
2316          * additional times.
2317          */
2318         return -1;
2319 mem_msg:
2320         dev_err(&h->pdev->dev, "out of memory\n");
2321         h->busy_configuring = 0;
2322         goto freeret;
2323 }
2324
2325 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2326 {
2327         /* zero out the disk size info */
2328         drive_info->nr_blocks = 0;
2329         drive_info->block_size = 0;
2330         drive_info->heads = 0;
2331         drive_info->sectors = 0;
2332         drive_info->cylinders = 0;
2333         drive_info->raid_level = -1;
2334         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2335         memset(drive_info->model, 0, sizeof(drive_info->model));
2336         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2337         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2338         /*
2339          * don't clear the LUNID though, we need to remember which
2340          * one this one is.
2341          */
2342 }
2343
2344 /* This function will deregister the disk and it's queue from the
2345  * kernel.  It must be called with the controller lock held and the
2346  * drv structures busy_configuring flag set.  It's parameters are:
2347  *
2348  * disk = This is the disk to be deregistered
2349  * drv  = This is the drive_info_struct associated with the disk to be
2350  *        deregistered.  It contains information about the disk used
2351  *        by the driver.
2352  * clear_all = This flag determines whether or not the disk information
2353  *             is going to be completely cleared out and the highest_lun
2354  *             reset.  Sometimes we want to clear out information about
2355  *             the disk in preparation for re-adding it.  In this case
2356  *             the highest_lun should be left unchanged and the LunID
2357  *             should not be cleared.
2358  * via_ioctl
2359  *    This indicates whether we've reached this path via ioctl.
2360  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2361  *    If this path is reached via ioctl(), then the max_usage_count will
2362  *    be 1, as the process calling ioctl() has got to have the device open.
2363  *    If we get here via sysfs, then the max usage count will be zero.
2364 */
2365 static int deregister_disk(ctlr_info_t *h, int drv_index,
2366                            int clear_all, int via_ioctl)
2367 {
2368         int i;
2369         struct gendisk *disk;
2370         drive_info_struct *drv;
2371         int recalculate_highest_lun;
2372
2373         if (!capable(CAP_SYS_RAWIO))
2374                 return -EPERM;
2375
2376         drv = h->drv[drv_index];
2377         disk = h->gendisk[drv_index];
2378
2379         /* make sure logical volume is NOT is use */
2380         if (clear_all || (h->gendisk[0] == disk)) {
2381                 if (drv->usage_count > via_ioctl)
2382                         return -EBUSY;
2383         } else if (drv->usage_count > 0)
2384                 return -EBUSY;
2385
2386         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2387
2388         /* invalidate the devices and deregister the disk.  If it is disk
2389          * zero do not deregister it but just zero out it's values.  This
2390          * allows us to delete disk zero but keep the controller registered.
2391          */
2392         if (h->gendisk[0] != disk) {
2393                 struct request_queue *q = disk->queue;
2394                 if (disk->flags & GENHD_FL_UP) {
2395                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2396                         del_gendisk(disk);
2397                 }
2398                 if (q)
2399                         blk_cleanup_queue(q);
2400                 /* If clear_all is set then we are deleting the logical
2401                  * drive, not just refreshing its info.  For drives
2402                  * other than disk 0 we will call put_disk.  We do not
2403                  * do this for disk 0 as we need it to be able to
2404                  * configure the controller.
2405                  */
2406                 if (clear_all){
2407                         /* This isn't pretty, but we need to find the
2408                          * disk in our array and NULL our the pointer.
2409                          * This is so that we will call alloc_disk if
2410                          * this index is used again later.
2411                          */
2412                         for (i=0; i < CISS_MAX_LUN; i++){
2413                                 if (h->gendisk[i] == disk) {
2414                                         h->gendisk[i] = NULL;
2415                                         break;
2416                                 }
2417                         }
2418                         put_disk(disk);
2419                 }
2420         } else {
2421                 set_capacity(disk, 0);
2422                 cciss_clear_drive_info(drv);
2423         }
2424
2425         --h->num_luns;
2426
2427         /* if it was the last disk, find the new hightest lun */
2428         if (clear_all && recalculate_highest_lun) {
2429                 int newhighest = -1;
2430                 for (i = 0; i <= h->highest_lun; i++) {
2431                         /* if the disk has size > 0, it is available */
2432                         if (h->drv[i] && h->drv[i]->heads)
2433                                 newhighest = i;
2434                 }
2435                 h->highest_lun = newhighest;
2436         }
2437         return 0;
2438 }
2439
2440 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2441                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2442                 int cmd_type)
2443 {
2444         u64bit buff_dma_handle;
2445         int status = IO_OK;
2446
2447         c->cmd_type = CMD_IOCTL_PEND;
2448         c->Header.ReplyQueue = 0;
2449         if (buff != NULL) {
2450                 c->Header.SGList = 1;
2451                 c->Header.SGTotal = 1;
2452         } else {
2453                 c->Header.SGList = 0;
2454                 c->Header.SGTotal = 0;
2455         }
2456         c->Header.Tag.lower = c->busaddr;
2457         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2458
2459         c->Request.Type.Type = cmd_type;
2460         if (cmd_type == TYPE_CMD) {
2461                 switch (cmd) {
2462                 case CISS_INQUIRY:
2463                         /* are we trying to read a vital product page */
2464                         if (page_code != 0) {
2465                                 c->Request.CDB[1] = 0x01;
2466                                 c->Request.CDB[2] = page_code;
2467                         }
2468                         c->Request.CDBLen = 6;
2469                         c->Request.Type.Attribute = ATTR_SIMPLE;
2470                         c->Request.Type.Direction = XFER_READ;
2471                         c->Request.Timeout = 0;
2472                         c->Request.CDB[0] = CISS_INQUIRY;
2473                         c->Request.CDB[4] = size & 0xFF;
2474                         break;
2475                 case CISS_REPORT_LOG:
2476                 case CISS_REPORT_PHYS:
2477                         /* Talking to controller so It's a physical command
2478                            mode = 00 target = 0.  Nothing to write.
2479                          */
2480                         c->Request.CDBLen = 12;
2481                         c->Request.Type.Attribute = ATTR_SIMPLE;
2482                         c->Request.Type.Direction = XFER_READ;
2483                         c->Request.Timeout = 0;
2484                         c->Request.CDB[0] = cmd;
2485                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2486                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2487                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2488                         c->Request.CDB[9] = size & 0xFF;
2489                         break;
2490
2491                 case CCISS_READ_CAPACITY:
2492                         c->Request.CDBLen = 10;
2493                         c->Request.Type.Attribute = ATTR_SIMPLE;
2494                         c->Request.Type.Direction = XFER_READ;
2495                         c->Request.Timeout = 0;
2496                         c->Request.CDB[0] = cmd;
2497                         break;
2498                 case CCISS_READ_CAPACITY_16:
2499                         c->Request.CDBLen = 16;
2500                         c->Request.Type.Attribute = ATTR_SIMPLE;
2501                         c->Request.Type.Direction = XFER_READ;
2502                         c->Request.Timeout = 0;
2503                         c->Request.CDB[0] = cmd;
2504                         c->Request.CDB[1] = 0x10;
2505                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2506                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2507                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2508                         c->Request.CDB[13] = size & 0xFF;
2509                         c->Request.Timeout = 0;
2510                         c->Request.CDB[0] = cmd;
2511                         break;
2512                 case CCISS_CACHE_FLUSH:
2513                         c->Request.CDBLen = 12;
2514                         c->Request.Type.Attribute = ATTR_SIMPLE;
2515                         c->Request.Type.Direction = XFER_WRITE;
2516                         c->Request.Timeout = 0;
2517                         c->Request.CDB[0] = BMIC_WRITE;
2518                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2519                         break;
2520                 case TEST_UNIT_READY:
2521                         c->Request.CDBLen = 6;
2522                         c->Request.Type.Attribute = ATTR_SIMPLE;
2523                         c->Request.Type.Direction = XFER_NONE;
2524                         c->Request.Timeout = 0;
2525                         break;
2526                 default:
2527                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2528                         return IO_ERROR;
2529                 }
2530         } else if (cmd_type == TYPE_MSG) {
2531                 switch (cmd) {
2532                 case 0: /* ABORT message */
2533                         c->Request.CDBLen = 12;
2534                         c->Request.Type.Attribute = ATTR_SIMPLE;
2535                         c->Request.Type.Direction = XFER_WRITE;
2536                         c->Request.Timeout = 0;
2537                         c->Request.CDB[0] = cmd;        /* abort */
2538                         c->Request.CDB[1] = 0;  /* abort a command */
2539                         /* buff contains the tag of the command to abort */
2540                         memcpy(&c->Request.CDB[4], buff, 8);
2541                         break;
2542                 case 1: /* RESET message */
2543                         c->Request.CDBLen = 16;
2544                         c->Request.Type.Attribute = ATTR_SIMPLE;
2545                         c->Request.Type.Direction = XFER_NONE;
2546                         c->Request.Timeout = 0;
2547                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2548                         c->Request.CDB[0] = cmd;        /* reset */
2549                         c->Request.CDB[1] = 0x03;       /* reset a target */
2550                         break;
2551                 case 3: /* No-Op message */
2552                         c->Request.CDBLen = 1;
2553                         c->Request.Type.Attribute = ATTR_SIMPLE;
2554                         c->Request.Type.Direction = XFER_WRITE;
2555                         c->Request.Timeout = 0;
2556                         c->Request.CDB[0] = cmd;
2557                         break;
2558                 default:
2559                         dev_warn(&h->pdev->dev,
2560                                 "unknown message type %d\n", cmd);
2561                         return IO_ERROR;
2562                 }
2563         } else {
2564                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2565                 return IO_ERROR;
2566         }
2567         /* Fill in the scatter gather information */
2568         if (size > 0) {
2569                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2570                                                              buff, size,
2571                                                              PCI_DMA_BIDIRECTIONAL);
2572                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2573                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2574                 c->SG[0].Len = size;
2575                 c->SG[0].Ext = 0;       /* we are not chaining */
2576         }
2577         return status;
2578 }
2579
2580 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2581 {
2582         switch (c->err_info->ScsiStatus) {
2583         case SAM_STAT_GOOD:
2584                 return IO_OK;
2585         case SAM_STAT_CHECK_CONDITION:
2586                 switch (0xf & c->err_info->SenseInfo[2]) {
2587                 case 0: return IO_OK; /* no sense */
2588                 case 1: return IO_OK; /* recovered error */
2589                 default:
2590                         if (check_for_unit_attention(h, c))
2591                                 return IO_NEEDS_RETRY;
2592                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2593                                 "check condition, sense key = 0x%02x\n",
2594                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2595                 }
2596                 break;
2597         default:
2598                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2599                         "scsi status = 0x%02x\n",
2600                         c->Request.CDB[0], c->err_info->ScsiStatus);
2601                 break;
2602         }
2603         return IO_ERROR;
2604 }
2605
2606 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2607 {
2608         int return_status = IO_OK;
2609
2610         if (c->err_info->CommandStatus == CMD_SUCCESS)
2611                 return IO_OK;
2612
2613         switch (c->err_info->CommandStatus) {
2614         case CMD_TARGET_STATUS:
2615                 return_status = check_target_status(h, c);
2616                 break;
2617         case CMD_DATA_UNDERRUN:
2618         case CMD_DATA_OVERRUN:
2619                 /* expected for inquiry and report lun commands */
2620                 break;
2621         case CMD_INVALID:
2622                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2623                        "reported invalid\n", c->Request.CDB[0]);
2624                 return_status = IO_ERROR;
2625                 break;
2626         case CMD_PROTOCOL_ERR:
2627                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2628                        "protocol error\n", c->Request.CDB[0]);
2629                 return_status = IO_ERROR;
2630                 break;
2631         case CMD_HARDWARE_ERR:
2632                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2633                        " hardware error\n", c->Request.CDB[0]);
2634                 return_status = IO_ERROR;
2635                 break;
2636         case CMD_CONNECTION_LOST:
2637                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2638                        "connection lost\n", c->Request.CDB[0]);
2639                 return_status = IO_ERROR;
2640                 break;
2641         case CMD_ABORTED:
2642                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2643                        "aborted\n", c->Request.CDB[0]);
2644                 return_status = IO_ERROR;
2645                 break;
2646         case CMD_ABORT_FAILED:
2647                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2648                        "abort failed\n", c->Request.CDB[0]);
2649                 return_status = IO_ERROR;
2650                 break;
2651         case CMD_UNSOLICITED_ABORT:
2652                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2653                         c->Request.CDB[0]);
2654                 return_status = IO_NEEDS_RETRY;
2655                 break;
2656         default:
2657                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2658                        "unknown status %x\n", c->Request.CDB[0],
2659                        c->err_info->CommandStatus);
2660                 return_status = IO_ERROR;
2661         }
2662         return return_status;
2663 }
2664
2665 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2666         int attempt_retry)
2667 {
2668         DECLARE_COMPLETION_ONSTACK(wait);
2669         u64bit buff_dma_handle;
2670         int return_status = IO_OK;
2671
2672 resend_cmd2:
2673         c->waiting = &wait;
2674         enqueue_cmd_and_start_io(h, c);
2675
2676         wait_for_completion(&wait);
2677
2678         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2679                 goto command_done;
2680
2681         return_status = process_sendcmd_error(h, c);
2682
2683         if (return_status == IO_NEEDS_RETRY &&
2684                 c->retry_count < MAX_CMD_RETRIES) {
2685                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2686                         c->Request.CDB[0]);
2687                 c->retry_count++;
2688                 /* erase the old error information */
2689                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2690                 return_status = IO_OK;
2691                 INIT_COMPLETION(wait);
2692                 goto resend_cmd2;
2693         }
2694
2695 command_done:
2696         /* unlock the buffers from DMA */
2697         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2698         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2699         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2700                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2701         return return_status;
2702 }
2703
2704 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2705                            __u8 page_code, unsigned char scsi3addr[],
2706                         int cmd_type)
2707 {
2708         CommandList_struct *c;
2709         int return_status;
2710
2711         c = cmd_special_alloc(h);
2712         if (!c)
2713                 return -ENOMEM;
2714         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2715                 scsi3addr, cmd_type);
2716         if (return_status == IO_OK)
2717                 return_status = sendcmd_withirq_core(h, c, 1);
2718
2719         cmd_special_free(h, c);
2720         return return_status;
2721 }
2722
2723 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2724                                    sector_t total_size,
2725                                    unsigned int block_size,
2726                                    InquiryData_struct *inq_buff,
2727                                    drive_info_struct *drv)
2728 {
2729         int return_code;
2730         unsigned long t;
2731         unsigned char scsi3addr[8];
2732
2733         memset(inq_buff, 0, sizeof(InquiryData_struct));
2734         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2735         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2736                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2737         if (return_code == IO_OK) {
2738                 if (inq_buff->data_byte[8] == 0xFF) {
2739                         dev_warn(&h->pdev->dev,
2740                                "reading geometry failed, volume "
2741                                "does not support reading geometry\n");
2742                         drv->heads = 255;
2743                         drv->sectors = 32;      /* Sectors per track */
2744                         drv->cylinders = total_size + 1;
2745                         drv->raid_level = RAID_UNKNOWN;
2746                 } else {
2747                         drv->heads = inq_buff->data_byte[6];
2748                         drv->sectors = inq_buff->data_byte[7];
2749                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2750                         drv->cylinders += inq_buff->data_byte[5];
2751                         drv->raid_level = inq_buff->data_byte[8];
2752                 }
2753                 drv->block_size = block_size;
2754                 drv->nr_blocks = total_size + 1;
2755                 t = drv->heads * drv->sectors;
2756                 if (t > 1) {
2757                         sector_t real_size = total_size + 1;
2758                         unsigned long rem = sector_div(real_size, t);
2759                         if (rem)
2760                                 real_size++;
2761                         drv->cylinders = real_size;
2762                 }
2763         } else {                /* Get geometry failed */
2764                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2765         }
2766 }
2767
2768 static void
2769 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2770                     unsigned int *block_size)
2771 {
2772         ReadCapdata_struct *buf;
2773         int return_code;
2774         unsigned char scsi3addr[8];
2775
2776         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2777         if (!buf) {
2778                 dev_warn(&h->pdev->dev, "out of memory\n");
2779                 return;
2780         }
2781
2782         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2783         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2784                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2785         if (return_code == IO_OK) {
2786                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2787                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2788         } else {                /* read capacity command failed */
2789                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2790                 *total_size = 0;
2791                 *block_size = BLOCK_SIZE;
2792         }
2793         kfree(buf);
2794 }
2795
2796 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2797         sector_t *total_size, unsigned int *block_size)
2798 {
2799         ReadCapdata_struct_16 *buf;
2800         int return_code;
2801         unsigned char scsi3addr[8];
2802
2803         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2804         if (!buf) {
2805                 dev_warn(&h->pdev->dev, "out of memory\n");
2806                 return;
2807         }
2808
2809         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2810         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2811                 buf, sizeof(ReadCapdata_struct_16),
2812                         0, scsi3addr, TYPE_CMD);
2813         if (return_code == IO_OK) {
2814                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2815                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2816         } else {                /* read capacity command failed */
2817                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2818                 *total_size = 0;
2819                 *block_size = BLOCK_SIZE;
2820         }
2821         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2822                (unsigned long long)*total_size+1, *block_size);
2823         kfree(buf);
2824 }
2825
2826 static int cciss_revalidate(struct gendisk *disk)
2827 {
2828         ctlr_info_t *h = get_host(disk);
2829         drive_info_struct *drv = get_drv(disk);
2830         int logvol;
2831         int FOUND = 0;
2832         unsigned int block_size;
2833         sector_t total_size;
2834         InquiryData_struct *inq_buff = NULL;
2835
2836         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2837                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2838                         sizeof(drv->LunID)) == 0) {
2839                         FOUND = 1;
2840                         break;
2841                 }
2842         }
2843
2844         if (!FOUND)
2845                 return 1;
2846
2847         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2848         if (inq_buff == NULL) {
2849                 dev_warn(&h->pdev->dev, "out of memory\n");
2850                 return 1;
2851         }
2852         if (h->cciss_read == CCISS_READ_10) {
2853                 cciss_read_capacity(h, logvol,
2854                                         &total_size, &block_size);
2855         } else {
2856                 cciss_read_capacity_16(h, logvol,
2857                                         &total_size, &block_size);
2858         }
2859         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2860                                inq_buff, drv);
2861
2862         blk_queue_logical_block_size(drv->queue, drv->block_size);
2863         set_capacity(disk, drv->nr_blocks);
2864
2865         kfree(inq_buff);
2866         return 0;
2867 }
2868
2869 /*
2870  * Map (physical) PCI mem into (virtual) kernel space
2871  */
2872 static void __iomem *remap_pci_mem(ulong base, ulong size)
2873 {
2874         ulong page_base = ((ulong) base) & PAGE_MASK;
2875         ulong page_offs = ((ulong) base) - page_base;
2876         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2877
2878         return page_remapped ? (page_remapped + page_offs) : NULL;
2879 }
2880
2881 /*
2882  * Takes jobs of the Q and sends them to the hardware, then puts it on
2883  * the Q to wait for completion.
2884  */
2885 static void start_io(ctlr_info_t *h)
2886 {
2887         CommandList_struct *c;
2888
2889         while (!hlist_empty(&h->reqQ)) {
2890                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2891                 /* can't do anything if fifo is full */
2892                 if ((h->access.fifo_full(h))) {
2893                         dev_warn(&h->pdev->dev, "fifo full\n");
2894                         break;
2895                 }
2896
2897                 /* Get the first entry from the Request Q */
2898                 removeQ(c);
2899                 h->Qdepth--;
2900
2901                 /* Tell the controller execute command */
2902                 h->access.submit_command(h, c);
2903
2904                 /* Put job onto the completed Q */
2905                 addQ(&h->cmpQ, c);
2906         }
2907 }
2908
2909 /* Assumes that h->lock is held. */
2910 /* Zeros out the error record and then resends the command back */
2911 /* to the controller */
2912 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2913 {
2914         /* erase the old error information */
2915         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2916
2917         /* add it to software queue and then send it to the controller */
2918         addQ(&h->reqQ, c);
2919         h->Qdepth++;
2920         if (h->Qdepth > h->maxQsinceinit)
2921                 h->maxQsinceinit = h->Qdepth;
2922
2923         start_io(h);
2924 }
2925
2926 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2927         unsigned int msg_byte, unsigned int host_byte,
2928         unsigned int driver_byte)
2929 {
2930         /* inverse of macros in scsi.h */
2931         return (scsi_status_byte & 0xff) |
2932                 ((msg_byte & 0xff) << 8) |
2933                 ((host_byte & 0xff) << 16) |
2934                 ((driver_byte & 0xff) << 24);
2935 }
2936
2937 static inline int evaluate_target_status(ctlr_info_t *h,
2938                         CommandList_struct *cmd, int *retry_cmd)
2939 {
2940         unsigned char sense_key;
2941         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2942         int error_value;
2943
2944         *retry_cmd = 0;
2945         /* If we get in here, it means we got "target status", that is, scsi status */
2946         status_byte = cmd->err_info->ScsiStatus;
2947         driver_byte = DRIVER_OK;
2948         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2949
2950         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2951                 host_byte = DID_PASSTHROUGH;
2952         else
2953                 host_byte = DID_OK;
2954
2955         error_value = make_status_bytes(status_byte, msg_byte,
2956                 host_byte, driver_byte);
2957
2958         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2959                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
2960                         dev_warn(&h->pdev->dev, "cmd %p "
2961                                "has SCSI Status 0x%x\n",
2962                                cmd, cmd->err_info->ScsiStatus);
2963                 return error_value;
2964         }
2965
2966         /* check the sense key */
2967         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2968         /* no status or recovered error */
2969         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
2970             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
2971                 error_value = 0;
2972
2973         if (check_for_unit_attention(h, cmd)) {
2974                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
2975                 return 0;
2976         }
2977
2978         /* Not SG_IO or similar? */
2979         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
2980                 if (error_value != 0)
2981                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
2982                                " sense key = 0x%x\n", cmd, sense_key);
2983                 return error_value;
2984         }
2985
2986         /* SG_IO or similar, copy sense data back */
2987         if (cmd->rq->sense) {
2988                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2989                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2990                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2991                         cmd->rq->sense_len);
2992         } else
2993                 cmd->rq->sense_len = 0;
2994
2995         return error_value;
2996 }
2997
2998 /* checks the status of the job and calls complete buffers to mark all
2999  * buffers for the completed job. Note that this function does not need
3000  * to hold the hba/queue lock.
3001  */
3002 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3003                                     int timeout)
3004 {
3005         int retry_cmd = 0;
3006         struct request *rq = cmd->rq;
3007
3008         rq->errors = 0;
3009
3010         if (timeout)
3011                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3012
3013         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3014                 goto after_error_processing;
3015
3016         switch (cmd->err_info->CommandStatus) {
3017         case CMD_TARGET_STATUS:
3018                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3019                 break;
3020         case CMD_DATA_UNDERRUN:
3021                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3022                         dev_warn(&h->pdev->dev, "cmd %p has"
3023                                " completed with data underrun "
3024                                "reported\n", cmd);
3025                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3026                 }
3027                 break;
3028         case CMD_DATA_OVERRUN:
3029                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3030                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3031                                " completed with data overrun "
3032                                "reported\n", cmd);
3033                 break;
3034         case CMD_INVALID:
3035                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3036                        "reported invalid\n", cmd);
3037                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3038                         cmd->err_info->CommandStatus, DRIVER_OK,
3039                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3040                                 DID_PASSTHROUGH : DID_ERROR);
3041                 break;
3042         case CMD_PROTOCOL_ERR:
3043                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3044                        "protocol error\n", cmd);
3045                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3046                         cmd->err_info->CommandStatus, DRIVER_OK,
3047                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3048                                 DID_PASSTHROUGH : DID_ERROR);
3049                 break;
3050         case CMD_HARDWARE_ERR:
3051                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3052                        " hardware error\n", cmd);
3053                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3054                         cmd->err_info->CommandStatus, DRIVER_OK,
3055                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3056                                 DID_PASSTHROUGH : DID_ERROR);
3057                 break;
3058         case CMD_CONNECTION_LOST:
3059                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3060                        "connection lost\n", cmd);
3061                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3062                         cmd->err_info->CommandStatus, DRIVER_OK,
3063                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3064                                 DID_PASSTHROUGH : DID_ERROR);
3065                 break;
3066         case CMD_ABORTED:
3067                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3068                        "aborted\n", cmd);
3069                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3070                         cmd->err_info->CommandStatus, DRIVER_OK,
3071                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3072                                 DID_PASSTHROUGH : DID_ABORT);
3073                 break;
3074         case CMD_ABORT_FAILED:
3075                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3076                        "abort failed\n", cmd);
3077                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3078                         cmd->err_info->CommandStatus, DRIVER_OK,
3079                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3080                                 DID_PASSTHROUGH : DID_ERROR);
3081                 break;
3082         case CMD_UNSOLICITED_ABORT:
3083                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3084                        "abort %p\n", h->ctlr, cmd);
3085                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3086                         retry_cmd = 1;
3087                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3088                         cmd->retry_count++;
3089                 } else
3090                         dev_warn(&h->pdev->dev,
3091                                 "%p retried too many times\n", cmd);
3092                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3093                         cmd->err_info->CommandStatus, DRIVER_OK,
3094                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3095                                 DID_PASSTHROUGH : DID_ABORT);
3096                 break;
3097         case CMD_TIMEOUT:
3098                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3099                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3100                         cmd->err_info->CommandStatus, DRIVER_OK,
3101                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3102                                 DID_PASSTHROUGH : DID_ERROR);
3103                 break;
3104         default:
3105                 dev_warn(&h->pdev->dev, "cmd %p returned "
3106                        "unknown status %x\n", cmd,
3107                        cmd->err_info->CommandStatus);
3108                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3109                         cmd->err_info->CommandStatus, DRIVER_OK,
3110                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3111                                 DID_PASSTHROUGH : DID_ERROR);
3112         }
3113
3114 after_error_processing:
3115
3116         /* We need to return this command */
3117         if (retry_cmd) {
3118                 resend_cciss_cmd(h, cmd);
3119                 return;
3120         }
3121         cmd->rq->completion_data = cmd;
3122         blk_complete_request(cmd->rq);
3123 }
3124
3125 static inline u32 cciss_tag_contains_index(u32 tag)
3126 {
3127 #define DIRECT_LOOKUP_BIT 0x10
3128         return tag & DIRECT_LOOKUP_BIT;
3129 }
3130
3131 static inline u32 cciss_tag_to_index(u32 tag)
3132 {
3133 #define DIRECT_LOOKUP_SHIFT 5
3134         return tag >> DIRECT_LOOKUP_SHIFT;
3135 }
3136
3137 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3138 {
3139 #define CCISS_ERROR_BITS 0x03
3140         return tag & ~CCISS_ERROR_BITS;
3141 }
3142
3143 static inline void cciss_mark_tag_indexed(u32 *tag)
3144 {
3145         *tag |= DIRECT_LOOKUP_BIT;
3146 }
3147
3148 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3149 {
3150         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3151 }
3152
3153 /*
3154  * Get a request and submit it to the controller.
3155  */
3156 static void do_cciss_request(struct request_queue *q)
3157 {
3158         ctlr_info_t *h = q->queuedata;
3159         CommandList_struct *c;
3160         sector_t start_blk;
3161         int seg;
3162         struct request *creq;
3163         u64bit temp64;
3164         struct scatterlist *tmp_sg;
3165         SGDescriptor_struct *curr_sg;
3166         drive_info_struct *drv;
3167         int i, dir;
3168         int sg_index = 0;
3169         int chained = 0;
3170
3171         /* We call start_io here in case there is a command waiting on the
3172          * queue that has not been sent.
3173          */
3174         if (blk_queue_plugged(q))
3175                 goto startio;
3176
3177       queue:
3178         creq = blk_peek_request(q);
3179         if (!creq)
3180                 goto startio;
3181
3182         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3183
3184         c = cmd_alloc(h);
3185         if (!c)
3186                 goto full;
3187
3188         blk_start_request(creq);
3189
3190         tmp_sg = h->scatter_list[c->cmdindex];
3191         spin_unlock_irq(q->queue_lock);
3192
3193         c->cmd_type = CMD_RWREQ;
3194         c->rq = creq;
3195
3196         /* fill in the request */
3197         drv = creq->rq_disk->private_data;
3198         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3199         /* got command from pool, so use the command block index instead */
3200         /* for direct lookups. */
3201         /* The first 2 bits are reserved for controller error reporting. */
3202         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3203         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3204         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3205         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3206         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3207         c->Request.Type.Attribute = ATTR_SIMPLE;
3208         c->Request.Type.Direction =
3209             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3210         c->Request.Timeout = 0; /* Don't time out */
3211         c->Request.CDB[0] =
3212             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3213         start_blk = blk_rq_pos(creq);
3214         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3215                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3216         sg_init_table(tmp_sg, h->maxsgentries);
3217         seg = blk_rq_map_sg(q, creq, tmp_sg);
3218
3219         /* get the DMA records for the setup */
3220         if (c->Request.Type.Direction == XFER_READ)
3221                 dir = PCI_DMA_FROMDEVICE;
3222         else
3223                 dir = PCI_DMA_TODEVICE;
3224
3225         curr_sg = c->SG;
3226         sg_index = 0;
3227         chained = 0;
3228
3229         for (i = 0; i < seg; i++) {
3230                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3231                         !chained && ((seg - i) > 1)) {
3232                         /* Point to next chain block. */
3233                         curr_sg = h->cmd_sg_list[c->cmdindex];
3234                         sg_index = 0;
3235                         chained = 1;
3236                 }
3237                 curr_sg[sg_index].Len = tmp_sg[i].length;
3238                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3239                                                 tmp_sg[i].offset,
3240                                                 tmp_sg[i].length, dir);
3241                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3242                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3243                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3244                 ++sg_index;
3245         }
3246         if (chained)
3247                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3248                         (seg - (h->max_cmd_sgentries - 1)) *
3249                                 sizeof(SGDescriptor_struct));
3250
3251         /* track how many SG entries we are using */
3252         if (seg > h->maxSG)
3253                 h->maxSG = seg;
3254
3255         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3256                         "chained[%d]\n",
3257                         blk_rq_sectors(creq), seg, chained);
3258
3259         c->Header.SGTotal = seg + chained;
3260         if (seg <= h->max_cmd_sgentries)
3261                 c->Header.SGList = c->Header.SGTotal;
3262         else
3263                 c->Header.SGList = h->max_cmd_sgentries;
3264         set_performant_mode(h, c);
3265
3266         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3267                 if(h->cciss_read == CCISS_READ_10) {
3268                         c->Request.CDB[1] = 0;
3269                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3270                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3271                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3272                         c->Request.CDB[5] = start_blk & 0xff;
3273                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3274                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3275                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3276                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3277                 } else {
3278                         u32 upper32 = upper_32_bits(start_blk);
3279
3280                         c->Request.CDBLen = 16;
3281                         c->Request.CDB[1]= 0;
3282                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3283                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3284                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3285                         c->Request.CDB[5]= upper32 & 0xff;
3286                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3287                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3288                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3289                         c->Request.CDB[9]= start_blk & 0xff;
3290                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3291                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3292                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3293                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3294                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3295                 }
3296         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3297                 c->Request.CDBLen = creq->cmd_len;
3298                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3299         } else {
3300                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3301                         creq->cmd_type);
3302                 BUG();
3303         }
3304
3305         spin_lock_irq(q->queue_lock);
3306
3307         addQ(&h->reqQ, c);
3308         h->Qdepth++;
3309         if (h->Qdepth > h->maxQsinceinit)
3310                 h->maxQsinceinit = h->Qdepth;
3311
3312         goto queue;
3313 full:
3314         blk_stop_queue(q);
3315 startio:
3316         /* We will already have the driver lock here so not need
3317          * to lock it.
3318          */
3319         start_io(h);
3320 }
3321
3322 static inline unsigned long get_next_completion(ctlr_info_t *h)
3323 {
3324         return h->access.command_completed(h);
3325 }
3326
3327 static inline int interrupt_pending(ctlr_info_t *h)
3328 {
3329         return h->access.intr_pending(h);
3330 }
3331
3332 static inline long interrupt_not_for_us(ctlr_info_t *h)
3333 {
3334         return ((h->access.intr_pending(h) == 0) ||
3335                 (h->interrupts_enabled == 0));
3336 }
3337
3338 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3339                         u32 raw_tag)
3340 {
3341         if (unlikely(tag_index >= h->nr_cmds)) {
3342                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3343                 return 1;
3344         }
3345         return 0;
3346 }
3347
3348 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3349                                 u32 raw_tag)
3350 {
3351         removeQ(c);
3352         if (likely(c->cmd_type == CMD_RWREQ))
3353                 complete_command(h, c, 0);
3354         else if (c->cmd_type == CMD_IOCTL_PEND)
3355                 complete(c->waiting);
3356 #ifdef CONFIG_CISS_SCSI_TAPE
3357         else if (c->cmd_type == CMD_SCSI)
3358                 complete_scsi_command(c, 0, raw_tag);
3359 #endif
3360 }
3361
3362 static inline u32 next_command(ctlr_info_t *h)
3363 {
3364         u32 a;
3365
3366         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3367                 return h->access.command_completed(h);
3368
3369         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3370                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3371                 (h->reply_pool_head)++;
3372                 h->commands_outstanding--;
3373         } else {
3374                 a = FIFO_EMPTY;
3375         }
3376         /* Check for wraparound */
3377         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3378                 h->reply_pool_head = h->reply_pool;
3379                 h->reply_pool_wraparound ^= 1;
3380         }
3381         return a;
3382 }
3383
3384 /* process completion of an indexed ("direct lookup") command */
3385 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3386 {
3387         u32 tag_index;
3388         CommandList_struct *c;
3389
3390         tag_index = cciss_tag_to_index(raw_tag);
3391         if (bad_tag(h, tag_index, raw_tag))
3392                 return next_command(h);
3393         c = h->cmd_pool + tag_index;
3394         finish_cmd(h, c, raw_tag);
3395         return next_command(h);
3396 }
3397
3398 /* process completion of a non-indexed command */
3399 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3400 {
3401         u32 tag;
3402         CommandList_struct *c = NULL;
3403         struct hlist_node *tmp;
3404         __u32 busaddr_masked, tag_masked;
3405
3406         tag = cciss_tag_discard_error_bits(raw_tag);
3407         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3408                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3409                 tag_masked = cciss_tag_discard_error_bits(tag);
3410                 if (busaddr_masked == tag_masked) {
3411                         finish_cmd(h, c, raw_tag);
3412                         return next_command(h);
3413                 }
3414         }
3415         bad_tag(h, h->nr_cmds + 1, raw_tag);
3416         return next_command(h);
3417 }
3418
3419 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3420 {
3421         ctlr_info_t *h = dev_id;
3422         unsigned long flags;
3423         u32 raw_tag;
3424
3425         if (interrupt_not_for_us(h))
3426                 return IRQ_NONE;
3427         spin_lock_irqsave(&h->lock, flags);
3428         while (interrupt_pending(h)) {
3429                 raw_tag = get_next_completion(h);
3430                 while (raw_tag != FIFO_EMPTY) {
3431                         if (cciss_tag_contains_index(raw_tag))
3432                                 raw_tag = process_indexed_cmd(h, raw_tag);
3433                         else
3434                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3435                 }
3436         }
3437         spin_unlock_irqrestore(&h->lock, flags);
3438         return IRQ_HANDLED;
3439 }
3440
3441 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3442  * check the interrupt pending register because it is not set.
3443  */
3444 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3445 {
3446         ctlr_info_t *h = dev_id;
3447         unsigned long flags;
3448         u32 raw_tag;
3449
3450         spin_lock_irqsave(&h->lock, flags);
3451         raw_tag = get_next_completion(h);
3452         while (raw_tag != FIFO_EMPTY) {
3453                 if (cciss_tag_contains_index(raw_tag))
3454                         raw_tag = process_indexed_cmd(h, raw_tag);
3455                 else
3456                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3457         }
3458         spin_unlock_irqrestore(&h->lock, flags);
3459         return IRQ_HANDLED;
3460 }
3461
3462 /**
3463  * add_to_scan_list() - add controller to rescan queue
3464  * @h:                Pointer to the controller.
3465  *
3466  * Adds the controller to the rescan queue if not already on the queue.
3467  *
3468  * returns 1 if added to the queue, 0 if skipped (could be on the
3469  * queue already, or the controller could be initializing or shutting
3470  * down).
3471  **/
3472 static int add_to_scan_list(struct ctlr_info *h)
3473 {
3474         struct ctlr_info *test_h;
3475         int found = 0;
3476         int ret = 0;
3477
3478         if (h->busy_initializing)
3479                 return 0;
3480
3481         if (!mutex_trylock(&h->busy_shutting_down))
3482                 return 0;
3483
3484         mutex_lock(&scan_mutex);
3485         list_for_each_entry(test_h, &scan_q, scan_list) {
3486                 if (test_h == h) {
3487                         found = 1;
3488                         break;
3489                 }
3490         }
3491         if (!found && !h->busy_scanning) {
3492                 INIT_COMPLETION(h->scan_wait);
3493                 list_add_tail(&h->scan_list, &scan_q);
3494                 ret = 1;
3495         }
3496         mutex_unlock(&scan_mutex);
3497         mutex_unlock(&h->busy_shutting_down);
3498
3499         return ret;
3500 }
3501
3502 /**
3503  * remove_from_scan_list() - remove controller from rescan queue
3504  * @h:                     Pointer to the controller.
3505  *
3506  * Removes the controller from the rescan queue if present. Blocks if
3507  * the controller is currently conducting a rescan.  The controller
3508  * can be in one of three states:
3509  * 1. Doesn't need a scan
3510  * 2. On the scan list, but not scanning yet (we remove it)
3511  * 3. Busy scanning (and not on the list). In this case we want to wait for
3512  *    the scan to complete to make sure the scanning thread for this
3513  *    controller is completely idle.
3514  **/
3515 static void remove_from_scan_list(struct ctlr_info *h)
3516 {
3517         struct ctlr_info *test_h, *tmp_h;
3518
3519         mutex_lock(&scan_mutex);
3520         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3521                 if (test_h == h) { /* state 2. */
3522                         list_del(&h->scan_list);
3523                         complete_all(&h->scan_wait);
3524                         mutex_unlock(&scan_mutex);
3525                         return;
3526                 }
3527         }
3528         if (h->busy_scanning) { /* state 3. */
3529                 mutex_unlock(&scan_mutex);
3530                 wait_for_completion(&h->scan_wait);
3531         } else { /* state 1, nothing to do. */
3532                 mutex_unlock(&scan_mutex);
3533         }
3534 }
3535
3536 /**
3537  * scan_thread() - kernel thread used to rescan controllers
3538  * @data:        Ignored.
3539  *
3540  * A kernel thread used scan for drive topology changes on
3541  * controllers. The thread processes only one controller at a time
3542  * using a queue.  Controllers are added to the queue using
3543  * add_to_scan_list() and removed from the queue either after done
3544  * processing or using remove_from_scan_list().
3545  *
3546  * returns 0.
3547  **/
3548 static int scan_thread(void *data)
3549 {
3550         struct ctlr_info *h;
3551
3552         while (1) {
3553                 set_current_state(TASK_INTERRUPTIBLE);
3554                 schedule();
3555                 if (kthread_should_stop())
3556                         break;
3557
3558                 while (1) {
3559                         mutex_lock(&scan_mutex);
3560                         if (list_empty(&scan_q)) {
3561                                 mutex_unlock(&scan_mutex);
3562                                 break;
3563                         }
3564
3565                         h = list_entry(scan_q.next,
3566                                        struct ctlr_info,
3567                                        scan_list);
3568                         list_del(&h->scan_list);
3569                         h->busy_scanning = 1;
3570                         mutex_unlock(&scan_mutex);
3571
3572                         rebuild_lun_table(h, 0, 0);
3573                         complete_all(&h->scan_wait);
3574                         mutex_lock(&scan_mutex);
3575                         h->busy_scanning = 0;
3576                         mutex_unlock(&scan_mutex);
3577                 }
3578         }
3579
3580         return 0;
3581 }
3582
3583 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3584 {
3585         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3586                 return 0;
3587
3588         switch (c->err_info->SenseInfo[12]) {
3589         case STATE_CHANGED:
3590                 dev_warn(&h->pdev->dev, "a state change "
3591                         "detected, command retried\n");
3592                 return 1;
3593         break;
3594         case LUN_FAILED:
3595                 dev_warn(&h->pdev->dev, "LUN failure "
3596                         "detected, action required\n");
3597                 return 1;
3598         break;
3599         case REPORT_LUNS_CHANGED:
3600                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3601         /*
3602          * Here, we could call add_to_scan_list and wake up the scan thread,
3603          * except that it's quite likely that we will get more than one
3604          * REPORT_LUNS_CHANGED condition in quick succession, which means
3605          * that those which occur after the first one will likely happen
3606          * *during* the scan_thread's rescan.  And the rescan code is not
3607          * robust enough to restart in the middle, undoing what it has already
3608          * done, and it's not clear that it's even possible to do this, since
3609          * part of what it does is notify the block layer, which starts
3610          * doing it's own i/o to read partition tables and so on, and the
3611          * driver doesn't have visibility to know what might need undoing.
3612          * In any event, if possible, it is horribly complicated to get right
3613          * so we just don't do it for now.
3614          *
3615          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3616          */
3617                 return 1;
3618         break;
3619         case POWER_OR_RESET:
3620                 dev_warn(&h->pdev->dev,
3621                         "a power on or device reset detected\n");
3622                 return 1;
3623         break;
3624         case UNIT_ATTENTION_CLEARED:
3625                 dev_warn(&h->pdev->dev,
3626                         "unit attention cleared by another initiator\n");
3627                 return 1;
3628         break;
3629         default:
3630                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3631                 return 1;
3632         }
3633 }
3634
3635 /*
3636  *  We cannot read the structure directly, for portability we must use
3637  *   the io functions.
3638  *   This is for debug only.
3639  */
3640 static void print_cfg_table(ctlr_info_t *h)
3641 {
3642         int i;
3643         char temp_name[17];
3644         CfgTable_struct *tb = h->cfgtable;
3645
3646         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3647         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3648         for (i = 0; i < 4; i++)
3649                 temp_name[i] = readb(&(tb->Signature[i]));
3650         temp_name[4] = '\0';
3651         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3652         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3653                 readl(&(tb->SpecValence)));
3654         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3655                readl(&(tb->TransportSupport)));
3656         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3657                readl(&(tb->TransportActive)));
3658         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3659                readl(&(tb->HostWrite.TransportRequest)));
3660         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3661                readl(&(tb->HostWrite.CoalIntDelay)));
3662         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3663                readl(&(tb->HostWrite.CoalIntCount)));
3664         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3665                readl(&(tb->CmdsOutMax)));
3666         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3667                 readl(&(tb->BusTypes)));
3668         for (i = 0; i < 16; i++)
3669                 temp_name[i] = readb(&(tb->ServerName[i]));
3670         temp_name[16] = '\0';
3671         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3672         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3673                 readl(&(tb->HeartBeat)));
3674 }
3675
3676 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3677 {
3678         int i, offset, mem_type, bar_type;
3679         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3680                 return 0;
3681         offset = 0;
3682         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3683                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3684                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3685                         offset += 4;
3686                 else {
3687                         mem_type = pci_resource_flags(pdev, i) &
3688                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3689                         switch (mem_type) {
3690                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3691                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3692                                 offset += 4;    /* 32 bit */
3693                                 break;
3694                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3695                                 offset += 8;
3696                                 break;
3697                         default:        /* reserved in PCI 2.2 */
3698                                 dev_warn(&pdev->dev,
3699                                        "Base address is invalid\n");
3700                                 return -1;
3701                                 break;
3702                         }
3703                 }
3704                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3705                         return i + 1;
3706         }
3707         return -1;
3708 }
3709
3710 /* Fill in bucket_map[], given nsgs (the max number of
3711  * scatter gather elements supported) and bucket[],
3712  * which is an array of 8 integers.  The bucket[] array
3713  * contains 8 different DMA transfer sizes (in 16
3714  * byte increments) which the controller uses to fetch
3715  * commands.  This function fills in bucket_map[], which
3716  * maps a given number of scatter gather elements to one of
3717  * the 8 DMA transfer sizes.  The point of it is to allow the
3718  * controller to only do as much DMA as needed to fetch the
3719  * command, with the DMA transfer size encoded in the lower
3720  * bits of the command address.
3721  */
3722 static void  calc_bucket_map(int bucket[], int num_buckets,
3723         int nsgs, int *bucket_map)
3724 {
3725         int i, j, b, size;
3726
3727         /* even a command with 0 SGs requires 4 blocks */
3728 #define MINIMUM_TRANSFER_BLOCKS 4
3729 #define NUM_BUCKETS 8
3730         /* Note, bucket_map must have nsgs+1 entries. */
3731         for (i = 0; i <= nsgs; i++) {
3732                 /* Compute size of a command with i SG entries */
3733                 size = i + MINIMUM_TRANSFER_BLOCKS;
3734                 b = num_buckets; /* Assume the biggest bucket */
3735                 /* Find the bucket that is just big enough */
3736                 for (j = 0; j < 8; j++) {
3737                         if (bucket[j] >= size) {
3738                                 b = j;
3739                                 break;
3740                         }
3741                 }
3742                 /* for a command with i SG entries, use bucket b. */
3743                 bucket_map[i] = b;
3744         }
3745 }
3746
3747 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3748 {
3749         int i;
3750
3751         /* under certain very rare conditions, this can take awhile.
3752          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3753          * as we enter this code.) */
3754         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3755                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3756                         break;
3757                 usleep_range(10000, 20000);
3758         }
3759 }
3760
3761 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3762 {
3763         /* This is a bit complicated.  There are 8 registers on
3764          * the controller which we write to to tell it 8 different
3765          * sizes of commands which there may be.  It's a way of
3766          * reducing the DMA done to fetch each command.  Encoded into
3767          * each command's tag are 3 bits which communicate to the controller
3768          * which of the eight sizes that command fits within.  The size of
3769          * each command depends on how many scatter gather entries there are.
3770          * Each SG entry requires 16 bytes.  The eight registers are programmed
3771          * with the number of 16-byte blocks a command of that size requires.
3772          * The smallest command possible requires 5 such 16 byte blocks.
3773          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3774          * blocks.  Note, this only extends to the SG entries contained
3775          * within the command block, and does not extend to chained blocks
3776          * of SG elements.   bft[] contains the eight values we write to
3777          * the registers.  They are not evenly distributed, but have more
3778          * sizes for small commands, and fewer sizes for larger commands.
3779          */
3780         __u32 trans_offset;
3781         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3782                         /*
3783                          *  5 = 1 s/g entry or 4k
3784                          *  6 = 2 s/g entry or 8k
3785                          *  8 = 4 s/g entry or 16k
3786                          * 10 = 6 s/g entry or 24k
3787                          */
3788         unsigned long register_value;
3789         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3790
3791         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3792
3793         /* Controller spec: zero out this buffer. */
3794         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3795         h->reply_pool_head = h->reply_pool;
3796
3797         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3798         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3799                                 h->blockFetchTable);
3800         writel(bft[0], &h->transtable->BlockFetch0);
3801         writel(bft[1], &h->transtable->BlockFetch1);
3802         writel(bft[2], &h->transtable->BlockFetch2);
3803         writel(bft[3], &h->transtable->BlockFetch3);
3804         writel(bft[4], &h->transtable->BlockFetch4);
3805         writel(bft[5], &h->transtable->BlockFetch5);
3806         writel(bft[6], &h->transtable->BlockFetch6);
3807         writel(bft[7], &h->transtable->BlockFetch7);
3808
3809         /* size of controller ring buffer */
3810         writel(h->max_commands, &h->transtable->RepQSize);
3811         writel(1, &h->transtable->RepQCount);
3812         writel(0, &h->transtable->RepQCtrAddrLow32);
3813         writel(0, &h->transtable->RepQCtrAddrHigh32);
3814         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3815         writel(0, &h->transtable->RepQAddr0High32);
3816         writel(CFGTBL_Trans_Performant,
3817                         &(h->cfgtable->HostWrite.TransportRequest));
3818
3819         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3820         cciss_wait_for_mode_change_ack(h);
3821         register_value = readl(&(h->cfgtable->TransportActive));
3822         if (!(register_value & CFGTBL_Trans_Performant))
3823                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3824                                         " performant mode\n");
3825 }
3826
3827 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3828 {
3829         __u32 trans_support;
3830
3831         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3832         /* Attempt to put controller into performant mode if supported */
3833         /* Does board support performant mode? */
3834         trans_support = readl(&(h->cfgtable->TransportSupport));
3835         if (!(trans_support & PERFORMANT_MODE))
3836                 return;
3837
3838         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3839         /* Performant mode demands commands on a 32 byte boundary
3840          * pci_alloc_consistent aligns on page boundarys already.
3841          * Just need to check if divisible by 32
3842          */
3843         if ((sizeof(CommandList_struct) % 32) != 0) {
3844                 dev_warn(&h->pdev->dev, "%s %d %s\n",
3845                         "cciss info: command size[",
3846                         (int)sizeof(CommandList_struct),
3847                         "] not divisible by 32, no performant mode..\n");
3848                 return;
3849         }
3850
3851         /* Performant mode ring buffer and supporting data structures */
3852         h->reply_pool = (__u64 *)pci_alloc_consistent(
3853                 h->pdev, h->max_commands * sizeof(__u64),
3854                 &(h->reply_pool_dhandle));
3855
3856         /* Need a block fetch table for performant mode */
3857         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3858                 sizeof(__u32)), GFP_KERNEL);
3859
3860         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3861                 goto clean_up;
3862
3863         cciss_enter_performant_mode(h);
3864
3865         /* Change the access methods to the performant access methods */
3866         h->access = SA5_performant_access;
3867         h->transMethod = CFGTBL_Trans_Performant;
3868
3869         return;
3870 clean_up:
3871         kfree(h->blockFetchTable);
3872         if (h->reply_pool)
3873                 pci_free_consistent(h->pdev,
3874                                 h->max_commands * sizeof(__u64),
3875                                 h->reply_pool,
3876                                 h->reply_pool_dhandle);
3877         return;
3878
3879 } /* cciss_put_controller_into_performant_mode */
3880
3881 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3882  * controllers that are capable. If not, we use IO-APIC mode.
3883  */
3884
3885 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3886 {
3887 #ifdef CONFIG_PCI_MSI
3888         int err;
3889         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3890         {0, 2}, {0, 3}
3891         };
3892
3893         /* Some boards advertise MSI but don't really support it */
3894         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3895             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3896                 goto default_int_mode;
3897
3898         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3899                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3900                 if (!err) {
3901                         h->intr[0] = cciss_msix_entries[0].vector;
3902                         h->intr[1] = cciss_msix_entries[1].vector;
3903                         h->intr[2] = cciss_msix_entries[2].vector;
3904                         h->intr[3] = cciss_msix_entries[3].vector;
3905                         h->msix_vector = 1;
3906                         return;
3907                 }
3908                 if (err > 0) {
3909                         dev_warn(&h->pdev->dev,
3910                                 "only %d MSI-X vectors available\n", err);
3911                         goto default_int_mode;
3912                 } else {
3913                         dev_warn(&h->pdev->dev,
3914                                 "MSI-X init failed %d\n", err);
3915                         goto default_int_mode;
3916                 }
3917         }
3918         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3919                 if (!pci_enable_msi(h->pdev))
3920                         h->msi_vector = 1;
3921                 else
3922                         dev_warn(&h->pdev->dev, "MSI init failed\n");
3923         }
3924 default_int_mode:
3925 #endif                          /* CONFIG_PCI_MSI */
3926         /* if we get here we're going to use the default interrupt mode */
3927         h->intr[PERF_MODE_INT] = h->pdev->irq;
3928         return;
3929 }
3930
3931 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3932 {
3933         int i;
3934         u32 subsystem_vendor_id, subsystem_device_id;
3935
3936         subsystem_vendor_id = pdev->subsystem_vendor;
3937         subsystem_device_id = pdev->subsystem_device;
3938         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3939                         subsystem_vendor_id;
3940
3941         for (i = 0; i < ARRAY_SIZE(products); i++)
3942                 if (*board_id == products[i].board_id)
3943                         return i;
3944         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
3945                 *board_id);
3946         return -ENODEV;
3947 }
3948
3949 static inline bool cciss_board_disabled(ctlr_info_t *h)
3950 {
3951         u16 command;
3952
3953         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
3954         return ((command & PCI_COMMAND_MEMORY) == 0);
3955 }
3956
3957 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
3958         unsigned long *memory_bar)
3959 {
3960         int i;
3961
3962         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
3963                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
3964                         /* addressing mode bits already removed */
3965                         *memory_bar = pci_resource_start(pdev, i);
3966                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
3967                                 *memory_bar);
3968                         return 0;
3969                 }
3970         dev_warn(&pdev->dev, "no memory BAR found\n");
3971         return -ENODEV;
3972 }
3973
3974 static int __devinit cciss_wait_for_board_state(struct pci_dev *pdev,
3975         void __iomem *vaddr, int wait_for_ready)
3976 #define BOARD_READY 1
3977 #define BOARD_NOT_READY 0
3978 {
3979         int i, iterations;
3980         u32 scratchpad;
3981
3982         if (wait_for_ready)
3983                 iterations = CCISS_BOARD_READY_ITERATIONS;
3984         else
3985                 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
3986
3987         for (i = 0; i < iterations; i++) {
3988                 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
3989                 if (wait_for_ready) {
3990                         if (scratchpad == CCISS_FIRMWARE_READY)
3991                                 return 0;
3992                 } else {
3993                         if (scratchpad != CCISS_FIRMWARE_READY)
3994                                 return 0;
3995                 }
3996                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
3997         }
3998         dev_warn(&pdev->dev, "board not ready, timed out.\n");
3999         return -ENODEV;
4000 }
4001
4002 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4003         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4004         u64 *cfg_offset)
4005 {
4006         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4007         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4008         *cfg_base_addr &= (u32) 0x0000ffff;
4009         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4010         if (*cfg_base_addr_index == -1) {
4011                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4012                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4013                 return -ENODEV;
4014         }
4015         return 0;
4016 }
4017
4018 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4019 {
4020         u64 cfg_offset;
4021         u32 cfg_base_addr;
4022         u64 cfg_base_addr_index;
4023         u32 trans_offset;
4024         int rc;
4025
4026         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4027                 &cfg_base_addr_index, &cfg_offset);
4028         if (rc)
4029                 return rc;
4030         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4031                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4032         if (!h->cfgtable)
4033                 return -ENOMEM;
4034         /* Find performant mode table. */
4035         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4036         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4037                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4038                                 sizeof(*h->transtable));
4039         if (!h->transtable)
4040                 return -ENOMEM;
4041         return 0;
4042 }
4043
4044 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4045 {
4046         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4047
4048         /* Limit commands in memory limited kdump scenario. */
4049         if (reset_devices && h->max_commands > 32)
4050                 h->max_commands = 32;
4051
4052         if (h->max_commands < 16) {
4053                 dev_warn(&h->pdev->dev, "Controller reports "
4054                         "max supported commands of %d, an obvious lie. "
4055                         "Using 16.  Ensure that firmware is up to date.\n",
4056                         h->max_commands);
4057                 h->max_commands = 16;
4058         }
4059 }
4060
4061 /* Interrogate the hardware for some limits:
4062  * max commands, max SG elements without chaining, and with chaining,
4063  * SG chain block size, etc.
4064  */
4065 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4066 {
4067         cciss_get_max_perf_mode_cmds(h);
4068         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4069         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4070         /*
4071          * Limit in-command s/g elements to 32 save dma'able memory.
4072          * Howvever spec says if 0, use 31
4073          */
4074         h->max_cmd_sgentries = 31;
4075         if (h->maxsgentries > 512) {
4076                 h->max_cmd_sgentries = 32;
4077                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4078                 h->maxsgentries--; /* save one for chain pointer */
4079         } else {
4080                 h->maxsgentries = 31; /* default to traditional values */
4081                 h->chainsize = 0;
4082         }
4083 }
4084
4085 static inline bool CISS_signature_present(ctlr_info_t *h)
4086 {
4087         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4088             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4089             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4090             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4091                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4092                 return false;
4093         }
4094         return true;
4095 }
4096
4097 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4098 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4099 {
4100 #ifdef CONFIG_X86
4101         u32 prefetch;
4102
4103         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4104         prefetch |= 0x100;
4105         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4106 #endif
4107 }
4108
4109 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4110  * in a prefetch beyond physical memory.
4111  */
4112 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4113 {
4114         u32 dma_prefetch;
4115         __u32 dma_refetch;
4116
4117         if (h->board_id != 0x3225103C)
4118                 return;
4119         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4120         dma_prefetch |= 0x8000;
4121         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4122         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4123         dma_refetch |= 0x1;
4124         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4125 }
4126
4127 static int __devinit cciss_pci_init(ctlr_info_t *h)
4128 {
4129         int prod_index, err;
4130
4131         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4132         if (prod_index < 0)
4133                 return -ENODEV;
4134         h->product_name = products[prod_index].product_name;
4135         h->access = *(products[prod_index].access);
4136
4137         if (cciss_board_disabled(h)) {
4138                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4139                 return -ENODEV;
4140         }
4141         err = pci_enable_device(h->pdev);
4142         if (err) {
4143                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4144                 return err;
4145         }
4146
4147         err = pci_request_regions(h->pdev, "cciss");
4148         if (err) {
4149                 dev_warn(&h->pdev->dev,
4150                         "Cannot obtain PCI resources, aborting\n");
4151                 return err;
4152         }
4153
4154         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4155         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4156
4157 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4158  * else we use the IO-APIC interrupt assigned to us by system ROM.
4159  */
4160         cciss_interrupt_mode(h);
4161         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4162         if (err)
4163                 goto err_out_free_res;
4164         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4165         if (!h->vaddr) {
4166                 err = -ENOMEM;
4167                 goto err_out_free_res;
4168         }
4169         err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4170         if (err)
4171                 goto err_out_free_res;
4172         err = cciss_find_cfgtables(h);
4173         if (err)
4174                 goto err_out_free_res;
4175         print_cfg_table(h);
4176         cciss_find_board_params(h);
4177
4178         if (!CISS_signature_present(h)) {
4179                 err = -ENODEV;
4180                 goto err_out_free_res;
4181         }
4182         cciss_enable_scsi_prefetch(h);
4183         cciss_p600_dma_prefetch_quirk(h);
4184         cciss_put_controller_into_performant_mode(h);
4185         return 0;
4186
4187 err_out_free_res:
4188         /*
4189          * Deliberately omit pci_disable_device(): it does something nasty to
4190          * Smart Array controllers that pci_enable_device does not undo
4191          */
4192         if (h->transtable)
4193                 iounmap(h->transtable);
4194         if (h->cfgtable)
4195                 iounmap(h->cfgtable);
4196         if (h->vaddr)
4197                 iounmap(h->vaddr);
4198         pci_release_regions(h->pdev);
4199         return err;
4200 }
4201
4202 /* Function to find the first free pointer into our hba[] array
4203  * Returns -1 if no free entries are left.
4204  */
4205 static int alloc_cciss_hba(struct pci_dev *pdev)
4206 {
4207         int i;
4208
4209         for (i = 0; i < MAX_CTLR; i++) {
4210                 if (!hba[i]) {
4211                         ctlr_info_t *h;
4212
4213                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4214                         if (!h)
4215                                 goto Enomem;
4216                         hba[i] = h;
4217                         return i;
4218                 }
4219         }
4220         dev_warn(&pdev->dev, "This driver supports a maximum"
4221                " of %d controllers.\n", MAX_CTLR);
4222         return -1;
4223 Enomem:
4224         dev_warn(&pdev->dev, "out of memory.\n");
4225         return -1;
4226 }
4227
4228 static void free_hba(ctlr_info_t *h)
4229 {
4230         int i;
4231
4232         hba[h->ctlr] = NULL;
4233         for (i = 0; i < h->highest_lun + 1; i++)
4234                 if (h->gendisk[i] != NULL)
4235                         put_disk(h->gendisk[i]);
4236         kfree(h);
4237 }
4238
4239 /* Send a message CDB to the firmware. */
4240 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4241 {
4242         typedef struct {
4243                 CommandListHeader_struct CommandHeader;
4244                 RequestBlock_struct Request;
4245                 ErrDescriptor_struct ErrorDescriptor;
4246         } Command;
4247         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4248         Command *cmd;
4249         dma_addr_t paddr64;
4250         uint32_t paddr32, tag;
4251         void __iomem *vaddr;
4252         int i, err;
4253
4254         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4255         if (vaddr == NULL)
4256                 return -ENOMEM;
4257
4258         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4259            CCISS commands, so they must be allocated from the lower 4GiB of
4260            memory. */
4261         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4262         if (err) {
4263                 iounmap(vaddr);
4264                 return -ENOMEM;
4265         }
4266
4267         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4268         if (cmd == NULL) {
4269                 iounmap(vaddr);
4270                 return -ENOMEM;
4271         }
4272
4273         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4274            although there's no guarantee, we assume that the address is at
4275            least 4-byte aligned (most likely, it's page-aligned). */
4276         paddr32 = paddr64;
4277
4278         cmd->CommandHeader.ReplyQueue = 0;
4279         cmd->CommandHeader.SGList = 0;
4280         cmd->CommandHeader.SGTotal = 0;
4281         cmd->CommandHeader.Tag.lower = paddr32;
4282         cmd->CommandHeader.Tag.upper = 0;
4283         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4284
4285         cmd->Request.CDBLen = 16;
4286         cmd->Request.Type.Type = TYPE_MSG;
4287         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4288         cmd->Request.Type.Direction = XFER_NONE;
4289         cmd->Request.Timeout = 0; /* Don't time out */
4290         cmd->Request.CDB[0] = opcode;
4291         cmd->Request.CDB[1] = type;
4292         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4293
4294         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4295         cmd->ErrorDescriptor.Addr.upper = 0;
4296         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4297
4298         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4299
4300         for (i = 0; i < 10; i++) {
4301                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4302                 if ((tag & ~3) == paddr32)
4303                         break;
4304                 schedule_timeout_uninterruptible(HZ);
4305         }
4306
4307         iounmap(vaddr);
4308
4309         /* we leak the DMA buffer here ... no choice since the controller could
4310            still complete the command. */
4311         if (i == 10) {
4312                 dev_err(&pdev->dev,
4313                         "controller message %02x:%02x timed out\n",
4314                         opcode, type);
4315                 return -ETIMEDOUT;
4316         }
4317
4318         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4319
4320         if (tag & 2) {
4321                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4322                         opcode, type);
4323                 return -EIO;
4324         }
4325
4326         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4327                 opcode, type);
4328         return 0;
4329 }
4330
4331 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4332 #define cciss_noop(p) cciss_message(p, 3, 0)
4333
4334 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4335         void * __iomem vaddr, bool use_doorbell)
4336 {
4337         u16 pmcsr;
4338         int pos;
4339
4340         if (use_doorbell) {
4341                 /* For everything after the P600, the PCI power state method
4342                  * of resetting the controller doesn't work, so we have this
4343                  * other way using the doorbell register.
4344                  */
4345                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4346                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4347                 msleep(1000);
4348         } else { /* Try to do it the PCI power state way */
4349
4350                 /* Quoting from the Open CISS Specification: "The Power
4351                  * Management Control/Status Register (CSR) controls the power
4352                  * state of the device.  The normal operating state is D0,
4353                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4354                  * the controller, place the interface device in D3 then to D0,
4355                  * this causes a secondary PCI reset which will reset the
4356                  * controller." */
4357
4358                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4359                 if (pos == 0) {
4360                         dev_err(&pdev->dev,
4361                                 "cciss_controller_hard_reset: "
4362                                 "PCI PM not supported\n");
4363                         return -ENODEV;
4364                 }
4365                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4366                 /* enter the D3hot power management state */
4367                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4368                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4369                 pmcsr |= PCI_D3hot;
4370                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4371
4372                 msleep(500);
4373
4374                 /* enter the D0 power management state */
4375                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4376                 pmcsr |= PCI_D0;
4377                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4378
4379                 msleep(500);
4380         }
4381         return 0;
4382 }
4383
4384 /* This does a hard reset of the controller using PCI power management
4385  * states or using the doorbell register. */
4386 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4387 {
4388         u64 cfg_offset;
4389         u32 cfg_base_addr;
4390         u64 cfg_base_addr_index;
4391         void __iomem *vaddr;
4392         unsigned long paddr;
4393         u32 misc_fw_support, active_transport;
4394         int rc;
4395         CfgTable_struct __iomem *cfgtable;
4396         bool use_doorbell;
4397         u32 board_id;
4398         u16 command_register;
4399
4400         /* For controllers as old a the p600, this is very nearly
4401          * the same thing as
4402          *
4403          * pci_save_state(pci_dev);
4404          * pci_set_power_state(pci_dev, PCI_D3hot);
4405          * pci_set_power_state(pci_dev, PCI_D0);
4406          * pci_restore_state(pci_dev);
4407          *
4408          * For controllers newer than the P600, the pci power state
4409          * method of resetting doesn't work so we have another way
4410          * using the doorbell register.
4411          */
4412
4413         /* Exclude 640x boards.  These are two pci devices in one slot
4414          * which share a battery backed cache module.  One controls the
4415          * cache, the other accesses the cache through the one that controls
4416          * it.  If we reset the one controlling the cache, the other will
4417          * likely not be happy.  Just forbid resetting this conjoined mess.
4418          */
4419         cciss_lookup_board_id(pdev, &board_id);
4420         if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4421                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4422                                 "due to shared cache module.");
4423                 return -ENODEV;
4424         }
4425
4426         /* Save the PCI command register */
4427         pci_read_config_word(pdev, 4, &command_register);
4428         /* Turn the board off.  This is so that later pci_restore_state()
4429          * won't turn the board on before the rest of config space is ready.
4430          */
4431         pci_disable_device(pdev);
4432         pci_save_state(pdev);
4433
4434         /* find the first memory BAR, so we can find the cfg table */
4435         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4436         if (rc)
4437                 return rc;
4438         vaddr = remap_pci_mem(paddr, 0x250);
4439         if (!vaddr)
4440                 return -ENOMEM;
4441
4442         /* find cfgtable in order to check if reset via doorbell is supported */
4443         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4444                                         &cfg_base_addr_index, &cfg_offset);
4445         if (rc)
4446                 goto unmap_vaddr;
4447         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4448                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4449         if (!cfgtable) {
4450                 rc = -ENOMEM;
4451                 goto unmap_vaddr;
4452         }
4453
4454         /* If reset via doorbell register is supported, use that. */
4455         misc_fw_support = readl(&cfgtable->misc_fw_support);
4456         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4457
4458         /* The doorbell reset seems to cause lockups on some Smart
4459          * Arrays (e.g. P410, P410i, maybe others).  Until this is
4460          * fixed or at least isolated, avoid the doorbell reset.
4461          */
4462         use_doorbell = 0;
4463
4464         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4465         if (rc)
4466                 goto unmap_cfgtable;
4467         pci_restore_state(pdev);
4468         rc = pci_enable_device(pdev);
4469         if (rc) {
4470                 dev_warn(&pdev->dev, "failed to enable device.\n");
4471                 goto unmap_cfgtable;
4472         }
4473         pci_write_config_word(pdev, 4, command_register);
4474
4475         /* Some devices (notably the HP Smart Array 5i Controller)
4476            need a little pause here */
4477         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4478
4479         /* Wait for board to become not ready, then ready. */
4480         dev_info(&pdev->dev, "Waiting for board to become ready.\n");
4481         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4482         if (rc) /* Don't bail, might be E500, etc. which can't be reset */
4483                 dev_warn(&pdev->dev,
4484                         "failed waiting for board to become not ready\n");
4485         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4486         if (rc) {
4487                 dev_warn(&pdev->dev,
4488                         "failed waiting for board to become ready\n");
4489                 goto unmap_cfgtable;
4490         }
4491         dev_info(&pdev->dev, "board ready.\n");
4492
4493         /* Controller should be in simple mode at this point.  If it's not,
4494          * It means we're on one of those controllers which doesn't support
4495          * the doorbell reset method and on which the PCI power management reset
4496          * method doesn't work (P800, for example.)
4497          * In those cases, don't try to proceed, as it generally doesn't work.
4498          */
4499         active_transport = readl(&cfgtable->TransportActive);
4500         if (active_transport & PERFORMANT_MODE) {
4501                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4502                         " Ignoring controller.\n");
4503                 rc = -ENODEV;
4504         }
4505
4506 unmap_cfgtable:
4507         iounmap(cfgtable);
4508
4509 unmap_vaddr:
4510         iounmap(vaddr);
4511         return rc;
4512 }
4513
4514 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4515 {
4516         int rc, i;
4517
4518         if (!reset_devices)
4519                 return 0;
4520
4521         /* Reset the controller with a PCI power-cycle or via doorbell */
4522         rc = cciss_kdump_hard_reset_controller(pdev);
4523
4524         /* -ENOTSUPP here means we cannot reset the controller
4525          * but it's already (and still) up and running in
4526          * "performant mode".  Or, it might be 640x, which can't reset
4527          * due to concerns about shared bbwc between 6402/6404 pair.
4528          */
4529         if (rc == -ENOTSUPP)
4530                 return 0; /* just try to do the kdump anyhow. */
4531         if (rc)
4532                 return -ENODEV;
4533
4534         /* Now try to get the controller to respond to a no-op */
4535         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4536                 if (cciss_noop(pdev) == 0)
4537                         break;
4538                 else
4539                         dev_warn(&pdev->dev, "no-op failed%s\n",
4540                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4541                                         "; re-trying" : ""));
4542                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4543         }
4544         return 0;
4545 }
4546
4547 /*
4548  *  This is it.  Find all the controllers and register them.  I really hate
4549  *  stealing all these major device numbers.
4550  *  returns the number of block devices registered.
4551  */
4552 static int __devinit cciss_init_one(struct pci_dev *pdev,
4553                                     const struct pci_device_id *ent)
4554 {
4555         int i;
4556         int j = 0;
4557         int k = 0;
4558         int rc;
4559         int dac, return_code;
4560         InquiryData_struct *inq_buff;
4561         ctlr_info_t *h;
4562
4563         rc = cciss_init_reset_devices(pdev);
4564         if (rc)
4565                 return rc;
4566         i = alloc_cciss_hba(pdev);
4567         if (i < 0)
4568                 return -1;
4569
4570         h = hba[i];
4571         h->pdev = pdev;
4572         h->busy_initializing = 1;
4573         INIT_HLIST_HEAD(&h->cmpQ);
4574         INIT_HLIST_HEAD(&h->reqQ);
4575         mutex_init(&h->busy_shutting_down);
4576
4577         if (cciss_pci_init(h) != 0)
4578                 goto clean_no_release_regions;
4579
4580         sprintf(h->devname, "cciss%d", i);
4581         h->ctlr = i;
4582
4583         init_completion(&h->scan_wait);
4584
4585         if (cciss_create_hba_sysfs_entry(h))
4586                 goto clean0;
4587
4588         /* configure PCI DMA stuff */
4589         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4590                 dac = 1;
4591         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4592                 dac = 0;
4593         else {
4594                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4595                 goto clean1;
4596         }
4597
4598         /*
4599          * register with the major number, or get a dynamic major number
4600          * by passing 0 as argument.  This is done for greater than
4601          * 8 controller support.
4602          */
4603         if (i < MAX_CTLR_ORIG)
4604                 h->major = COMPAQ_CISS_MAJOR + i;
4605         rc = register_blkdev(h->major, h->devname);
4606         if (rc == -EBUSY || rc == -EINVAL) {
4607                 dev_err(&h->pdev->dev,
4608                        "Unable to get major number %d for %s "
4609                        "on hba %d\n", h->major, h->devname, i);
4610                 goto clean1;
4611         } else {
4612                 if (i >= MAX_CTLR_ORIG)
4613                         h->major = rc;
4614         }
4615
4616         /* make sure the board interrupts are off */
4617         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4618         if (h->msi_vector || h->msix_vector) {
4619                 if (request_irq(h->intr[PERF_MODE_INT],
4620                                 do_cciss_msix_intr,
4621                                 IRQF_DISABLED, h->devname, h)) {
4622                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4623                                h->intr[PERF_MODE_INT], h->devname);
4624                         goto clean2;
4625                 }
4626         } else {
4627                 if (request_irq(h->intr[PERF_MODE_INT], do_cciss_intx,
4628                                 IRQF_DISABLED, h->devname, h)) {
4629                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4630                                h->intr[PERF_MODE_INT], h->devname);
4631                         goto clean2;
4632                 }
4633         }
4634
4635         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4636                h->devname, pdev->device, pci_name(pdev),
4637                h->intr[PERF_MODE_INT], dac ? "" : " not");
4638
4639         h->cmd_pool_bits =
4640             kmalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4641                         * sizeof(unsigned long), GFP_KERNEL);
4642         h->cmd_pool = (CommandList_struct *)
4643             pci_alloc_consistent(h->pdev,
4644                     h->nr_cmds * sizeof(CommandList_struct),
4645                     &(h->cmd_pool_dhandle));
4646         h->errinfo_pool = (ErrorInfo_struct *)
4647             pci_alloc_consistent(h->pdev,
4648                     h->nr_cmds * sizeof(ErrorInfo_struct),
4649                     &(h->errinfo_pool_dhandle));
4650         if ((h->cmd_pool_bits == NULL)
4651             || (h->cmd_pool == NULL)
4652             || (h->errinfo_pool == NULL)) {
4653                 dev_err(&h->pdev->dev, "out of memory");
4654                 goto clean4;
4655         }
4656
4657         /* Need space for temp scatter list */
4658         h->scatter_list = kmalloc(h->max_commands *
4659                                                 sizeof(struct scatterlist *),
4660                                                 GFP_KERNEL);
4661         if (!h->scatter_list)
4662                 goto clean4;
4663
4664         for (k = 0; k < h->nr_cmds; k++) {
4665                 h->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4666                                                         h->maxsgentries,
4667                                                         GFP_KERNEL);
4668                 if (h->scatter_list[k] == NULL) {
4669                         dev_err(&h->pdev->dev,
4670                                 "could not allocate s/g lists\n");
4671                         goto clean4;
4672                 }
4673         }
4674         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4675                 h->chainsize, h->nr_cmds);
4676         if (!h->cmd_sg_list && h->chainsize > 0)
4677                 goto clean4;
4678
4679         spin_lock_init(&h->lock);
4680
4681         /* Initialize the pdev driver private data.
4682            have it point to h.  */
4683         pci_set_drvdata(pdev, h);
4684         /* command and error info recs zeroed out before
4685            they are used */
4686         memset(h->cmd_pool_bits, 0,
4687                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4688                         * sizeof(unsigned long));
4689
4690         h->num_luns = 0;
4691         h->highest_lun = -1;
4692         for (j = 0; j < CISS_MAX_LUN; j++) {
4693                 h->drv[j] = NULL;
4694                 h->gendisk[j] = NULL;
4695         }
4696
4697         cciss_scsi_setup(h);
4698
4699         /* Turn the interrupts on so we can service requests */
4700         h->access.set_intr_mask(h, CCISS_INTR_ON);
4701
4702         /* Get the firmware version */
4703         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4704         if (inq_buff == NULL) {
4705                 dev_err(&h->pdev->dev, "out of memory\n");
4706                 goto clean4;
4707         }
4708
4709         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4710                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4711         if (return_code == IO_OK) {
4712                 h->firm_ver[0] = inq_buff->data_byte[32];
4713                 h->firm_ver[1] = inq_buff->data_byte[33];
4714                 h->firm_ver[2] = inq_buff->data_byte[34];
4715                 h->firm_ver[3] = inq_buff->data_byte[35];
4716         } else {         /* send command failed */
4717                 dev_warn(&h->pdev->dev, "unable to determine firmware"
4718                         " version of controller\n");
4719         }
4720         kfree(inq_buff);
4721
4722         cciss_procinit(h);
4723
4724         h->cciss_max_sectors = 8192;
4725
4726         rebuild_lun_table(h, 1, 0);
4727         h->busy_initializing = 0;
4728         return 1;
4729
4730 clean4:
4731         kfree(h->cmd_pool_bits);
4732         /* Free up sg elements */
4733         for (k-- ; k >= 0; k--)
4734                 kfree(h->scatter_list[k]);
4735         kfree(h->scatter_list);
4736         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4737         if (h->cmd_pool)
4738                 pci_free_consistent(h->pdev,
4739                                     h->nr_cmds * sizeof(CommandList_struct),
4740                                     h->cmd_pool, h->cmd_pool_dhandle);
4741         if (h->errinfo_pool)
4742                 pci_free_consistent(h->pdev,
4743                                     h->nr_cmds * sizeof(ErrorInfo_struct),
4744                                     h->errinfo_pool,
4745                                     h->errinfo_pool_dhandle);
4746         free_irq(h->intr[PERF_MODE_INT], h);
4747 clean2:
4748         unregister_blkdev(h->major, h->devname);
4749 clean1:
4750         cciss_destroy_hba_sysfs_entry(h);
4751 clean0:
4752         pci_release_regions(pdev);
4753 clean_no_release_regions:
4754         h->busy_initializing = 0;
4755
4756         /*
4757          * Deliberately omit pci_disable_device(): it does something nasty to
4758          * Smart Array controllers that pci_enable_device does not undo
4759          */
4760         pci_set_drvdata(pdev, NULL);
4761         free_hba(h);
4762         return -1;
4763 }
4764
4765 static void cciss_shutdown(struct pci_dev *pdev)
4766 {
4767         ctlr_info_t *h;
4768         char *flush_buf;
4769         int return_code;
4770
4771         h = pci_get_drvdata(pdev);
4772         flush_buf = kzalloc(4, GFP_KERNEL);
4773         if (!flush_buf) {
4774                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
4775                 return;
4776         }
4777         /* write all data in the battery backed cache to disk */
4778         memset(flush_buf, 0, 4);
4779         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4780                 4, 0, CTLR_LUNID, TYPE_CMD);
4781         kfree(flush_buf);
4782         if (return_code != IO_OK)
4783                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
4784         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4785         free_irq(h->intr[PERF_MODE_INT], h);
4786 }
4787
4788 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4789 {
4790         ctlr_info_t *h;
4791         int i, j;
4792
4793         if (pci_get_drvdata(pdev) == NULL) {
4794                 dev_err(&pdev->dev, "Unable to remove device\n");
4795                 return;
4796         }
4797
4798         h = pci_get_drvdata(pdev);
4799         i = h->ctlr;
4800         if (hba[i] == NULL) {
4801                 dev_err(&pdev->dev, "device appears to already be removed\n");
4802                 return;
4803         }
4804
4805         mutex_lock(&h->busy_shutting_down);
4806
4807         remove_from_scan_list(h);
4808         remove_proc_entry(h->devname, proc_cciss);
4809         unregister_blkdev(h->major, h->devname);
4810
4811         /* remove it from the disk list */
4812         for (j = 0; j < CISS_MAX_LUN; j++) {
4813                 struct gendisk *disk = h->gendisk[j];
4814                 if (disk) {
4815                         struct request_queue *q = disk->queue;
4816
4817                         if (disk->flags & GENHD_FL_UP) {
4818                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
4819                                 del_gendisk(disk);
4820                         }
4821                         if (q)
4822                                 blk_cleanup_queue(q);
4823                 }
4824         }
4825
4826 #ifdef CONFIG_CISS_SCSI_TAPE
4827         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
4828 #endif
4829
4830         cciss_shutdown(pdev);
4831
4832 #ifdef CONFIG_PCI_MSI
4833         if (h->msix_vector)
4834                 pci_disable_msix(h->pdev);
4835         else if (h->msi_vector)
4836                 pci_disable_msi(h->pdev);
4837 #endif                          /* CONFIG_PCI_MSI */
4838
4839         iounmap(h->transtable);
4840         iounmap(h->cfgtable);
4841         iounmap(h->vaddr);
4842
4843         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(CommandList_struct),
4844                             h->cmd_pool, h->cmd_pool_dhandle);
4845         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(ErrorInfo_struct),
4846                             h->errinfo_pool, h->errinfo_pool_dhandle);
4847         kfree(h->cmd_pool_bits);
4848         /* Free up sg elements */
4849         for (j = 0; j < h->nr_cmds; j++)
4850                 kfree(h->scatter_list[j]);
4851         kfree(h->scatter_list);
4852         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4853         /*
4854          * Deliberately omit pci_disable_device(): it does something nasty to
4855          * Smart Array controllers that pci_enable_device does not undo
4856          */
4857         pci_release_regions(pdev);
4858         pci_set_drvdata(pdev, NULL);
4859         cciss_destroy_hba_sysfs_entry(h);
4860         mutex_unlock(&h->busy_shutting_down);
4861         free_hba(h);
4862 }
4863
4864 static struct pci_driver cciss_pci_driver = {
4865         .name = "cciss",
4866         .probe = cciss_init_one,
4867         .remove = __devexit_p(cciss_remove_one),
4868         .id_table = cciss_pci_device_id,        /* id_table */
4869         .shutdown = cciss_shutdown,
4870 };
4871
4872 /*
4873  *  This is it.  Register the PCI driver information for the cards we control
4874  *  the OS will call our registered routines when it finds one of our cards.
4875  */
4876 static int __init cciss_init(void)
4877 {
4878         int err;
4879
4880         /*
4881          * The hardware requires that commands are aligned on a 64-bit
4882          * boundary. Given that we use pci_alloc_consistent() to allocate an
4883          * array of them, the size must be a multiple of 8 bytes.
4884          */
4885         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4886         printk(KERN_INFO DRIVER_NAME "\n");
4887
4888         err = bus_register(&cciss_bus_type);
4889         if (err)
4890                 return err;
4891
4892         /* Start the scan thread */
4893         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4894         if (IS_ERR(cciss_scan_thread)) {
4895                 err = PTR_ERR(cciss_scan_thread);
4896                 goto err_bus_unregister;
4897         }
4898
4899         /* Register for our PCI devices */
4900         err = pci_register_driver(&cciss_pci_driver);
4901         if (err)
4902                 goto err_thread_stop;
4903
4904         return err;
4905
4906 err_thread_stop:
4907         kthread_stop(cciss_scan_thread);
4908 err_bus_unregister:
4909         bus_unregister(&cciss_bus_type);
4910
4911         return err;
4912 }
4913
4914 static void __exit cciss_cleanup(void)
4915 {
4916         int i;
4917
4918         pci_unregister_driver(&cciss_pci_driver);
4919         /* double check that all controller entrys have been removed */
4920         for (i = 0; i < MAX_CTLR; i++) {
4921                 if (hba[i] != NULL) {
4922                         dev_warn(&hba[i]->pdev->dev,
4923                                 "had to remove controller\n");
4924                         cciss_remove_one(hba[i]->pdev);
4925                 }
4926         }
4927         kthread_stop(cciss_scan_thread);
4928         if (proc_cciss)
4929                 remove_proc_entry("driver/cciss", NULL);
4930         bus_unregister(&cciss_bus_type);
4931 }
4932
4933 module_init(cciss_init);
4934 module_exit(cciss_cleanup);