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/pci-aspm.h>
  28 #include <linux/kernel.h>
  29 #include <linux/slab.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/compat.h>
  43 #include <linux/mutex.h>
  44 #include <linux/bitmap.h>
  45 #include <linux/io.h>
  46 #include <linux/uaccess.h>
  47
  48 #include <linux/dma-mapping.h>
  49 #include <linux/blkdev.h>
  50 #include <linux/genhd.h>
  51 #include <linux/completion.h>
  52 #include <scsi/scsi.h>
  53 #include <scsi/sg.h>
  54 #include <scsi/scsi_ioctl.h>
  55 #include <scsi/scsi_request.h>
  56 #include <linux/cdrom.h>
  57 #include <linux/scatterlist.h>
  58 #include <linux/kthread.h>
  59
  60 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  61 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  62 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  63
  64 /* Embedded module documentation macros - see modules.h */
  65 MODULE_AUTHOR("Hewlett-Packard Company");
  66 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  67 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  68 MODULE_VERSION("3.6.26");
  69 MODULE_LICENSE("GPL");
  70 static int cciss_tape_cmds = 6;
  71 module_param(cciss_tape_cmds, int, 0644);
  72 MODULE_PARM_DESC(cciss_tape_cmds,
  73         "number of commands to allocate for tape devices (default: 6)");
  74 static int cciss_simple_mode;
  75 module_param(cciss_simple_mode, int, S_IRUGO|S_IWUSR);
  76 MODULE_PARM_DESC(cciss_simple_mode,
  77         "Use 'simple mode' rather than 'performant mode'");
  78
  79 static int cciss_allow_hpsa;
  80 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  81 MODULE_PARM_DESC(cciss_allow_hpsa,
  82         "Prevent cciss driver from accessing hardware known to be "
  83         " supported by the hpsa driver");
  84
  85 static DEFINE_MUTEX(cciss_mutex);
  86 static struct proc_dir_entry *proc_cciss;
  87
  88 #include "cciss_cmd.h"
  89 #include "cciss.h"
  90 #include <linux/cciss_ioctl.h>
  91
  92 /* define the PCI info for the cards we can control */
  93 static const struct pci_device_id cciss_pci_device_id[] = {
  94         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  95         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  96         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  97         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  98         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  99         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
 100         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
 101         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
 102         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
 110         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
 111         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
 112         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 113         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 114         {0,}
 115 };
 116
 117 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 118
 119 /*  board_id = Subsystem Device ID & Vendor ID
 120  *  product = Marketing Name for the board
 121  *  access = Address of the struct of function pointers
 122  */
 123 static struct board_type products[] = {
 124         {0x40700E11, "Smart Array 5300", &SA5_access},
 125         {0x40800E11, "Smart Array 5i", &SA5B_access},
 126         {0x40820E11, "Smart Array 532", &SA5B_access},
 127         {0x40830E11, "Smart Array 5312", &SA5B_access},
 128         {0x409A0E11, "Smart Array 641", &SA5_access},
 129         {0x409B0E11, "Smart Array 642", &SA5_access},
 130         {0x409C0E11, "Smart Array 6400", &SA5_access},
 131         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 132         {0x40910E11, "Smart Array 6i", &SA5_access},
 133         {0x3225103C, "Smart Array P600", &SA5_access},
 134         {0x3223103C, "Smart Array P800", &SA5_access},
 135         {0x3234103C, "Smart Array P400", &SA5_access},
 136         {0x3235103C, "Smart Array P400i", &SA5_access},
 137         {0x3211103C, "Smart Array E200i", &SA5_access},
 138         {0x3212103C, "Smart Array E200", &SA5_access},
 139         {0x3213103C, "Smart Array E200i", &SA5_access},
 140         {0x3214103C, "Smart Array E200i", &SA5_access},
 141         {0x3215103C, "Smart Array E200i", &SA5_access},
 142         {0x3237103C, "Smart Array E500", &SA5_access},
 143         {0x323D103C, "Smart Array P700m", &SA5_access},
 144 };
 145
 146 /* How long to wait (in milliseconds) for board to go into simple mode */
 147 #define MAX_CONFIG_WAIT 30000
 148 #define MAX_IOCTL_CONFIG_WAIT 1000
 149
 150 /*define how many times we will try a command because of bus resets */
 151 #define MAX_CMD_RETRIES 3
 152
 153 #define MAX_CTLR        32
 154
 155 /* Originally cciss driver only supports 8 major numbers */
 156 #define MAX_CTLR_ORIG   8
 157
 158 static ctlr_info_t *hba[MAX_CTLR];
 159
 160 static struct task_struct *cciss_scan_thread;
 161 static DEFINE_MUTEX(scan_mutex);
 162 static LIST_HEAD(scan_q);
 163
 164 static void do_cciss_request(struct request_queue *q);
 165 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 166 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 167 static int cciss_open(struct block_device *bdev, fmode_t mode);
 168 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 169 static void cciss_release(struct gendisk *disk, fmode_t mode);
 170 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 171                        unsigned int cmd, unsigned long arg);
 172 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 173
 174 static int cciss_revalidate(struct gendisk *disk);
 175 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 176 static int deregister_disk(ctlr_info_t *h, int drv_index,
 177                            int clear_all, int via_ioctl);
 178
 179 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 180                         sector_t *total_size, unsigned int *block_size);
 181 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 182                         sector_t *total_size, unsigned int *block_size);
 183 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 184                         sector_t total_size,
 185                         unsigned int block_size, InquiryData_struct *inq_buff,
 186                                    drive_info_struct *drv);
 187 static void cciss_interrupt_mode(ctlr_info_t *);
 188 static int cciss_enter_simple_mode(struct ctlr_info *h);
 189 static void start_io(ctlr_info_t *h);
 190 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 191                         __u8 page_code, unsigned char scsi3addr[],
 192                         int cmd_type);
 193 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 194         int attempt_retry);
 195 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 196
 197 static int add_to_scan_list(struct ctlr_info *h);
 198 static int scan_thread(void *data);
 199 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 200 static void cciss_hba_release(struct device *dev);
 201 static void cciss_device_release(struct device *dev);
 202 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 203 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 204 static inline u32 next_command(ctlr_info_t *h);
 205 static int cciss_find_cfg_addrs(struct pci_dev *pdev, void __iomem *vaddr,
 206                                 u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 207                                 u64 *cfg_offset);
 208 static int cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 209                                      unsigned long *memory_bar);
 210 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag);
 211 static int write_driver_ver_to_cfgtable(CfgTable_struct __iomem *cfgtable);
 212
 213 /* performant mode helper functions */
 214 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 215                                 int *bucket_map);
 216 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 217
 218 #ifdef CONFIG_PROC_FS
 219 static void cciss_procinit(ctlr_info_t *h);
 220 #else
 221 static void cciss_procinit(ctlr_info_t *h)
 222 {
 223 }
 224 #endif                          /* CONFIG_PROC_FS */
 225
 226 #ifdef CONFIG_COMPAT
 227 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 228                               unsigned, unsigned long);
 229 #endif
 230
 231 static const struct block_device_operations cciss_fops = {
 232         .owner = THIS_MODULE,
 233         .open = cciss_unlocked_open,
 234         .release = cciss_release,
 235         .ioctl = cciss_ioctl,
 236         .getgeo = cciss_getgeo,
 237 #ifdef CONFIG_COMPAT
 238         .compat_ioctl = cciss_compat_ioctl,
 239 #endif
 240         .revalidate_disk = cciss_revalidate,
 241 };
 242
 243 /* set_performant_mode: Modify the tag for cciss performant
 244  * set bit 0 for pull model, bits 3-1 for block fetch
 245  * register number
 246  */
 247 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 248 {
 249         if (likely(h->transMethod & CFGTBL_Trans_Performant))
 250                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 251 }
 252
 253 /*
 254  * Enqueuing and dequeuing functions for cmdlists.
 255  */
 256 static inline void addQ(struct list_head *list, CommandList_struct *c)
 257 {
 258         list_add_tail(&c->list, list);
 259 }
 260
 261 static inline void removeQ(CommandList_struct *c)
 262 {
 263         /*
 264          * After kexec/dump some commands might still
 265          * be in flight, which the firmware will try
 266          * to complete. Resetting the firmware doesn't work
 267          * with old fw revisions, so we have to mark
 268          * them off as 'stale' to prevent the driver from
 269          * falling over.
 270          */
 271         if (WARN_ON(list_empty(&c->list))) {
 272                 c->cmd_type = CMD_MSG_STALE;
 273                 return;
 274         }
 275
 276         list_del_init(&c->list);
 277 }
 278
 279 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 280         CommandList_struct *c)
 281 {
 282         unsigned long flags;
 283         set_performant_mode(h, c);
 284         spin_lock_irqsave(&h->lock, flags);
 285         addQ(&h->reqQ, c);
 286         h->Qdepth++;
 287         if (h->Qdepth > h->maxQsinceinit)
 288                 h->maxQsinceinit = h->Qdepth;
 289         start_io(h);
 290         spin_unlock_irqrestore(&h->lock, flags);
 291 }
 292
 293 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 294         int nr_cmds)
 295 {
 296         int i;
 297
 298         if (!cmd_sg_list)
 299                 return;
 300         for (i = 0; i < nr_cmds; i++) {
 301                 kfree(cmd_sg_list[i]);
 302                 cmd_sg_list[i] = NULL;
 303         }
 304         kfree(cmd_sg_list);
 305 }
 306
 307 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 308         ctlr_info_t *h, int chainsize, int nr_cmds)
 309 {
 310         int j;
 311         SGDescriptor_struct **cmd_sg_list;
 312
 313         if (chainsize <= 0)
 314                 return NULL;
 315
 316         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 317         if (!cmd_sg_list)
 318                 return NULL;
 319
 320         /* Build up chain blocks for each command */
 321         for (j = 0; j < nr_cmds; j++) {
 322                 /* Need a block of chainsized s/g elements. */
 323                 cmd_sg_list[j] = kmalloc((chainsize *
 324                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 325                 if (!cmd_sg_list[j]) {
 326                         dev_err(&h->pdev->dev, "Cannot get memory "
 327                                 "for s/g chains.\n");
 328                         goto clean;
 329                 }
 330         }
 331         return cmd_sg_list;
 332 clean:
 333         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 334         return NULL;
 335 }
 336
 337 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 338 {
 339         SGDescriptor_struct *chain_sg;
 340         u64bit temp64;
 341
 342         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 343                 return;
 344
 345         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 346         temp64.val32.lower = chain_sg->Addr.lower;
 347         temp64.val32.upper = chain_sg->Addr.upper;
 348         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 349 }
 350
 351 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 352         SGDescriptor_struct *chain_block, int len)
 353 {
 354         SGDescriptor_struct *chain_sg;
 355         u64bit temp64;
 356
 357         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 358         chain_sg->Ext = CCISS_SG_CHAIN;
 359         chain_sg->Len = len;
 360         temp64.val = pci_map_single(h->pdev, chain_block, len,
 361                                 PCI_DMA_TODEVICE);
 362         chain_sg->Addr.lower = temp64.val32.lower;
 363         chain_sg->Addr.upper = temp64.val32.upper;
 364 }
 365
 366 #include "cciss_scsi.c"         /* For SCSI tape support */
 367
 368 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 369         "UNKNOWN"
 370 };
 371 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 372
 373 #ifdef CONFIG_PROC_FS
 374
 375 /*
 376  * Report information about this controller.
 377  */
 378 #define ENG_GIG 1000000000
 379 #define ENG_GIG_FACTOR (ENG_GIG/512)
 380 #define ENGAGE_SCSI     "engage scsi"
 381
 382 static void cciss_seq_show_header(struct seq_file *seq)
 383 {
 384         ctlr_info_t *h = seq->private;
 385
 386         seq_printf(seq, "%s: HP %s Controller\n"
 387                 "Board ID: 0x%08lx\n"
 388                 "Firmware Version: %c%c%c%c\n"
 389                 "IRQ: %d\n"
 390                 "Logical drives: %d\n"
 391                 "Current Q depth: %d\n"
 392                 "Current # commands on controller: %d\n"
 393                 "Max Q depth since init: %d\n"
 394                 "Max # commands on controller since init: %d\n"
 395                 "Max SG entries since init: %d\n",
 396                 h->devname,
 397                 h->product_name,
 398                 (unsigned long)h->board_id,
 399                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 400                 h->firm_ver[3], (unsigned int)h->intr[h->intr_mode],
 401                 h->num_luns,
 402                 h->Qdepth, h->commands_outstanding,
 403                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 404
 405 #ifdef CONFIG_CISS_SCSI_TAPE
 406         cciss_seq_tape_report(seq, h);
 407 #endif /* CONFIG_CISS_SCSI_TAPE */
 408 }
 409
 410 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 411 {
 412         ctlr_info_t *h = seq->private;
 413         unsigned long flags;
 414
 415         /* prevent displaying bogus info during configuration
 416          * or deconfiguration of a logical volume
 417          */
 418         spin_lock_irqsave(&h->lock, flags);
 419         if (h->busy_configuring) {
 420                 spin_unlock_irqrestore(&h->lock, flags);
 421                 return ERR_PTR(-EBUSY);
 422         }
 423         h->busy_configuring = 1;
 424         spin_unlock_irqrestore(&h->lock, flags);
 425
 426         if (*pos == 0)
 427                 cciss_seq_show_header(seq);
 428
 429         return pos;
 430 }
 431
 432 static int cciss_seq_show(struct seq_file *seq, void *v)
 433 {
 434         sector_t vol_sz, vol_sz_frac;
 435         ctlr_info_t *h = seq->private;
 436         unsigned ctlr = h->ctlr;
 437         loff_t *pos = v;
 438         drive_info_struct *drv = h->drv[*pos];
 439
 440         if (*pos > h->highest_lun)
 441                 return 0;
 442
 443         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 444                 return 0;
 445
 446         if (drv->heads == 0)
 447                 return 0;
 448
 449         vol_sz = drv->nr_blocks;
 450         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 451         vol_sz_frac *= 100;
 452         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 453
 454         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 455                 drv->raid_level = RAID_UNKNOWN;
 456         seq_printf(seq, "cciss/c%dd%d:"
 457                         "\t%4u.%02uGB\tRAID %s\n",
 458                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 459                         raid_label[drv->raid_level]);
 460         return 0;
 461 }
 462
 463 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 464 {
 465         ctlr_info_t *h = seq->private;
 466
 467         if (*pos > h->highest_lun)
 468                 return NULL;
 469         *pos += 1;
 470
 471         return pos;
 472 }
 473
 474 static void cciss_seq_stop(struct seq_file *seq, void *v)
 475 {
 476         ctlr_info_t *h = seq->private;
 477
 478         /* Only reset h->busy_configuring if we succeeded in setting
 479          * it during cciss_seq_start. */
 480         if (v == ERR_PTR(-EBUSY))
 481                 return;
 482
 483         h->busy_configuring = 0;
 484 }
 485
 486 static const struct seq_operations cciss_seq_ops = {
 487         .start = cciss_seq_start,
 488         .show  = cciss_seq_show,
 489         .next  = cciss_seq_next,
 490         .stop  = cciss_seq_stop,
 491 };
 492
 493 static int cciss_seq_open(struct inode *inode, struct file *file)
 494 {
 495         int ret = seq_open(file, &cciss_seq_ops);
 496         struct seq_file *seq = file->private_data;
 497
 498         if (!ret)
 499                 seq->private = PDE_DATA(inode);
 500
 501         return ret;
 502 }
 503
 504 static ssize_t
 505 cciss_proc_write(struct file *file, const char __user *buf,
 506                  size_t length, loff_t *ppos)
 507 {
 508         int err;
 509         char *buffer;
 510
 511 #ifndef CONFIG_CISS_SCSI_TAPE
 512         return -EINVAL;
 513 #endif
 514
 515         if (!buf || length > PAGE_SIZE - 1)
 516                 return -EINVAL;
 517
 518         buffer = memdup_user_nul(buf, length);
 519         if (IS_ERR(buffer))
 520                 return PTR_ERR(buffer);
 521
 522 #ifdef CONFIG_CISS_SCSI_TAPE
 523         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 524                 struct seq_file *seq = file->private_data;
 525                 ctlr_info_t *h = seq->private;
 526
 527                 err = cciss_engage_scsi(h);
 528                 if (err == 0)
 529                         err = length;
 530         } else
 531 #endif /* CONFIG_CISS_SCSI_TAPE */
 532                 err = -EINVAL;
 533         /* might be nice to have "disengage" too, but it's not
 534            safely possible. (only 1 module use count, lock issues.) */
 535
 536         kfree(buffer);
 537         return err;
 538 }
 539
 540 static const struct file_operations cciss_proc_fops = {
 541         .owner   = THIS_MODULE,
 542         .open    = cciss_seq_open,
 543         .read    = seq_read,
 544         .llseek  = seq_lseek,
 545         .release = seq_release,
 546         .write   = cciss_proc_write,
 547 };
 548
 549 static void cciss_procinit(ctlr_info_t *h)
 550 {
 551         struct proc_dir_entry *pde;
 552
 553         if (proc_cciss == NULL)
 554                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 555         if (!proc_cciss)
 556                 return;
 557         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 558                                         S_IROTH, proc_cciss,
 559                                         &cciss_proc_fops, h);
 560 }
 561 #endif                          /* CONFIG_PROC_FS */
 562
 563 #define MAX_PRODUCT_NAME_LEN 19
 564
 565 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 566 #define to_drv(n) container_of(n, drive_info_struct, dev)
 567
 568 /* List of controllers which cannot be hard reset on kexec with reset_devices */
 569 static u32 unresettable_controller[] = {
 570         0x3223103C, /* Smart Array P800 */
 571         0x3234103C, /* Smart Array P400 */
 572         0x3235103C, /* Smart Array P400i */
 573         0x3211103C, /* Smart Array E200i */
 574         0x3212103C, /* Smart Array E200 */
 575         0x3213103C, /* Smart Array E200i */
 576         0x3214103C, /* Smart Array E200i */
 577         0x3215103C, /* Smart Array E200i */
 578         0x3237103C, /* Smart Array E500 */
 579         0x323D103C, /* Smart Array P700m */
 580         0x40800E11, /* Smart Array 5i */
 581         0x409C0E11, /* Smart Array 6400 */
 582         0x409D0E11, /* Smart Array 6400 EM */
 583         0x40700E11, /* Smart Array 5300 */
 584         0x40820E11, /* Smart Array 532 */
 585         0x40830E11, /* Smart Array 5312 */
 586         0x409A0E11, /* Smart Array 641 */
 587         0x409B0E11, /* Smart Array 642 */
 588         0x40910E11, /* Smart Array 6i */
 589 };
 590
 591 /* List of controllers which cannot even be soft reset */
 592 static u32 soft_unresettable_controller[] = {
 593         0x40800E11, /* Smart Array 5i */
 594         0x40700E11, /* Smart Array 5300 */
 595         0x40820E11, /* Smart Array 532 */
 596         0x40830E11, /* Smart Array 5312 */
 597         0x409A0E11, /* Smart Array 641 */
 598         0x409B0E11, /* Smart Array 642 */
 599         0x40910E11, /* Smart Array 6i */
 600         /* Exclude 640x boards.  These are two pci devices in one slot
 601          * which share a battery backed cache module.  One controls the
 602          * cache, the other accesses the cache through the one that controls
 603          * it.  If we reset the one controlling the cache, the other will
 604          * likely not be happy.  Just forbid resetting this conjoined mess.
 605          */
 606         0x409C0E11, /* Smart Array 6400 */
 607         0x409D0E11, /* Smart Array 6400 EM */
 608 };
 609
 610 static int ctlr_is_hard_resettable(u32 board_id)
 611 {
 612         int i;
 613
 614         for (i = 0; i < ARRAY_SIZE(unresettable_controller); i++)
 615                 if (unresettable_controller[i] == board_id)
 616                         return 0;
 617         return 1;
 618 }
 619
 620 static int ctlr_is_soft_resettable(u32 board_id)
 621 {
 622         int i;
 623
 624         for (i = 0; i < ARRAY_SIZE(soft_unresettable_controller); i++)
 625                 if (soft_unresettable_controller[i] == board_id)
 626                         return 0;
 627         return 1;
 628 }
 629
 630 static int ctlr_is_resettable(u32 board_id)
 631 {
 632         return ctlr_is_hard_resettable(board_id) ||
 633                 ctlr_is_soft_resettable(board_id);
 634 }
 635
 636 static ssize_t host_show_resettable(struct device *dev,
 637                                     struct device_attribute *attr,
 638                                     char *buf)
 639 {
 640         struct ctlr_info *h = to_hba(dev);
 641
 642         return snprintf(buf, 20, "%d\n", ctlr_is_resettable(h->board_id));
 643 }
 644 static DEVICE_ATTR(resettable, S_IRUGO, host_show_resettable, NULL);
 645
 646 static ssize_t host_store_rescan(struct device *dev,
 647                                  struct device_attribute *attr,
 648                                  const char *buf, size_t count)
 649 {
 650         struct ctlr_info *h = to_hba(dev);
 651
 652         add_to_scan_list(h);
 653         wake_up_process(cciss_scan_thread);
 654         wait_for_completion_interruptible(&h->scan_wait);
 655
 656         return count;
 657 }
 658 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 659
 660 static ssize_t host_show_transport_mode(struct device *dev,
 661                                  struct device_attribute *attr,
 662                                  char *buf)
 663 {
 664         struct ctlr_info *h = to_hba(dev);
 665
 666         return snprintf(buf, 20, "%s\n",
 667                 h->transMethod & CFGTBL_Trans_Performant ?
 668                         "performant" : "simple");
 669 }
 670 static DEVICE_ATTR(transport_mode, S_IRUGO, host_show_transport_mode, NULL);
 671
 672 static ssize_t dev_show_unique_id(struct device *dev,
 673                                  struct device_attribute *attr,
 674                                  char *buf)
 675 {
 676         drive_info_struct *drv = to_drv(dev);
 677         struct ctlr_info *h = to_hba(drv->dev.parent);
 678         __u8 sn[16];
 679         unsigned long flags;
 680         int ret = 0;
 681
 682         spin_lock_irqsave(&h->lock, flags);
 683         if (h->busy_configuring)
 684                 ret = -EBUSY;
 685         else
 686                 memcpy(sn, drv->serial_no, sizeof(sn));
 687         spin_unlock_irqrestore(&h->lock, flags);
 688
 689         if (ret)
 690                 return ret;
 691         else
 692                 return snprintf(buf, 16 * 2 + 2,
 693                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 694                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 695                                 sn[0], sn[1], sn[2], sn[3],
 696                                 sn[4], sn[5], sn[6], sn[7],
 697                                 sn[8], sn[9], sn[10], sn[11],
 698                                 sn[12], sn[13], sn[14], sn[15]);
 699 }
 700 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 701
 702 static ssize_t dev_show_vendor(struct device *dev,
 703                                struct device_attribute *attr,
 704                                char *buf)
 705 {
 706         drive_info_struct *drv = to_drv(dev);
 707         struct ctlr_info *h = to_hba(drv->dev.parent);
 708         char vendor[VENDOR_LEN + 1];
 709         unsigned long flags;
 710         int ret = 0;
 711
 712         spin_lock_irqsave(&h->lock, flags);
 713         if (h->busy_configuring)
 714                 ret = -EBUSY;
 715         else
 716                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 717         spin_unlock_irqrestore(&h->lock, flags);
 718
 719         if (ret)
 720                 return ret;
 721         else
 722                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 723 }
 724 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 725
 726 static ssize_t dev_show_model(struct device *dev,
 727                               struct device_attribute *attr,
 728                               char *buf)
 729 {
 730         drive_info_struct *drv = to_drv(dev);
 731         struct ctlr_info *h = to_hba(drv->dev.parent);
 732         char model[MODEL_LEN + 1];
 733         unsigned long flags;
 734         int ret = 0;
 735
 736         spin_lock_irqsave(&h->lock, flags);
 737         if (h->busy_configuring)
 738                 ret = -EBUSY;
 739         else
 740                 memcpy(model, drv->model, MODEL_LEN + 1);
 741         spin_unlock_irqrestore(&h->lock, flags);
 742
 743         if (ret)
 744                 return ret;
 745         else
 746                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 747 }
 748 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 749
 750 static ssize_t dev_show_rev(struct device *dev,
 751                             struct device_attribute *attr,
 752                             char *buf)
 753 {
 754         drive_info_struct *drv = to_drv(dev);
 755         struct ctlr_info *h = to_hba(drv->dev.parent);
 756         char rev[REV_LEN + 1];
 757         unsigned long flags;
 758         int ret = 0;
 759
 760         spin_lock_irqsave(&h->lock, flags);
 761         if (h->busy_configuring)
 762                 ret = -EBUSY;
 763         else
 764                 memcpy(rev, drv->rev, REV_LEN + 1);
 765         spin_unlock_irqrestore(&h->lock, flags);
 766
 767         if (ret)
 768                 return ret;
 769         else
 770                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 771 }
 772 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 773
 774 static ssize_t cciss_show_lunid(struct device *dev,
 775                                 struct device_attribute *attr, char *buf)
 776 {
 777         drive_info_struct *drv = to_drv(dev);
 778         struct ctlr_info *h = to_hba(drv->dev.parent);
 779         unsigned long flags;
 780         unsigned char lunid[8];
 781
 782         spin_lock_irqsave(&h->lock, flags);
 783         if (h->busy_configuring) {
 784                 spin_unlock_irqrestore(&h->lock, flags);
 785                 return -EBUSY;
 786         }
 787         if (!drv->heads) {
 788                 spin_unlock_irqrestore(&h->lock, flags);
 789                 return -ENOTTY;
 790         }
 791         memcpy(lunid, drv->LunID, sizeof(lunid));
 792         spin_unlock_irqrestore(&h->lock, flags);
 793         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 794                 lunid[0], lunid[1], lunid[2], lunid[3],
 795                 lunid[4], lunid[5], lunid[6], lunid[7]);
 796 }
 797 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 798
 799 static ssize_t cciss_show_raid_level(struct device *dev,
 800                                      struct device_attribute *attr, char *buf)
 801 {
 802         drive_info_struct *drv = to_drv(dev);
 803         struct ctlr_info *h = to_hba(drv->dev.parent);
 804         int raid;
 805         unsigned long flags;
 806
 807         spin_lock_irqsave(&h->lock, flags);
 808         if (h->busy_configuring) {
 809                 spin_unlock_irqrestore(&h->lock, flags);
 810                 return -EBUSY;
 811         }
 812         raid = drv->raid_level;
 813         spin_unlock_irqrestore(&h->lock, flags);
 814         if (raid < 0 || raid > RAID_UNKNOWN)
 815                 raid = RAID_UNKNOWN;
 816
 817         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 818                         raid_label[raid]);
 819 }
 820 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 821
 822 static ssize_t cciss_show_usage_count(struct device *dev,
 823                                       struct device_attribute *attr, char *buf)
 824 {
 825         drive_info_struct *drv = to_drv(dev);
 826         struct ctlr_info *h = to_hba(drv->dev.parent);
 827         unsigned long flags;
 828         int count;
 829
 830         spin_lock_irqsave(&h->lock, flags);
 831         if (h->busy_configuring) {
 832                 spin_unlock_irqrestore(&h->lock, flags);
 833                 return -EBUSY;
 834         }
 835         count = drv->usage_count;
 836         spin_unlock_irqrestore(&h->lock, flags);
 837         return snprintf(buf, 20, "%d\n", count);
 838 }
 839 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 840
 841 static struct attribute *cciss_host_attrs[] = {
 842         &dev_attr_rescan.attr,
 843         &dev_attr_resettable.attr,
 844         &dev_attr_transport_mode.attr,
 845         NULL
 846 };
 847
 848 static struct attribute_group cciss_host_attr_group = {
 849         .attrs = cciss_host_attrs,
 850 };
 851
 852 static const struct attribute_group *cciss_host_attr_groups[] = {
 853         &cciss_host_attr_group,
 854         NULL
 855 };
 856
 857 static struct device_type cciss_host_type = {
 858         .name           = "cciss_host",
 859         .groups         = cciss_host_attr_groups,
 860         .release        = cciss_hba_release,
 861 };
 862
 863 static struct attribute *cciss_dev_attrs[] = {
 864         &dev_attr_unique_id.attr,
 865         &dev_attr_model.attr,
 866         &dev_attr_vendor.attr,
 867         &dev_attr_rev.attr,
 868         &dev_attr_lunid.attr,
 869         &dev_attr_raid_level.attr,
 870         &dev_attr_usage_count.attr,
 871         NULL
 872 };
 873
 874 static struct attribute_group cciss_dev_attr_group = {
 875         .attrs = cciss_dev_attrs,
 876 };
 877
 878 static const struct attribute_group *cciss_dev_attr_groups[] = {
 879         &cciss_dev_attr_group,
 880         NULL
 881 };
 882
 883 static struct device_type cciss_dev_type = {
 884         .name           = "cciss_device",
 885         .groups         = cciss_dev_attr_groups,
 886         .release        = cciss_device_release,
 887 };
 888
 889 static struct bus_type cciss_bus_type = {
 890         .name           = "cciss",
 891 };
 892
 893 /*
 894  * cciss_hba_release is called when the reference count
 895  * of h->dev goes to zero.
 896  */
 897 static void cciss_hba_release(struct device *dev)
 898 {
 899         /*
 900          * nothing to do, but need this to avoid a warning
 901          * about not having a release handler from lib/kref.c.
 902          */
 903 }
 904
 905 /*
 906  * Initialize sysfs entry for each controller.  This sets up and registers
 907  * the 'cciss#' directory for each individual controller under
 908  * /sys/bus/pci/devices/<dev>/.
 909  */
 910 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 911 {
 912         device_initialize(&h->dev);
 913         h->dev.type = &cciss_host_type;
 914         h->dev.bus = &cciss_bus_type;
 915         dev_set_name(&h->dev, "%s", h->devname);
 916         h->dev.parent = &h->pdev->dev;
 917
 918         return device_add(&h->dev);
 919 }
 920
 921 /*
 922  * Remove sysfs entries for an hba.
 923  */
 924 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 925 {
 926         device_del(&h->dev);
 927         put_device(&h->dev); /* final put. */
 928 }
 929
 930 /* cciss_device_release is called when the reference count
 931  * of h->drv[x]dev goes to zero.
 932  */
 933 static void cciss_device_release(struct device *dev)
 934 {
 935         drive_info_struct *drv = to_drv(dev);
 936         kfree(drv);
 937 }
 938
 939 /*
 940  * Initialize sysfs for each logical drive.  This sets up and registers
 941  * the 'c#d#' directory for each individual logical drive under
 942  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 943  * /sys/block/cciss!c#d# to this entry.
 944  */
 945 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 946                                        int drv_index)
 947 {
 948         struct device *dev;
 949
 950         if (h->drv[drv_index]->device_initialized)
 951                 return 0;
 952
 953         dev = &h->drv[drv_index]->dev;
 954         device_initialize(dev);
 955         dev->type = &cciss_dev_type;
 956         dev->bus = &cciss_bus_type;
 957         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 958         dev->parent = &h->dev;
 959         h->drv[drv_index]->device_initialized = 1;
 960         return device_add(dev);
 961 }
 962
 963 /*
 964  * Remove sysfs entries for a logical drive.
 965  */
 966 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 967         int ctlr_exiting)
 968 {
 969         struct device *dev = &h->drv[drv_index]->dev;
 970
 971         /* special case for c*d0, we only destroy it on controller exit */
 972         if (drv_index == 0 && !ctlr_exiting)
 973                 return;
 974
 975         device_del(dev);
 976         put_device(dev); /* the "final" put. */
 977         h->drv[drv_index] = NULL;
 978 }
 979
 980 /*
 981  * For operations that cannot sleep, a command block is allocated at init,
 982  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 983  * which ones are free or in use.
 984  */
 985 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 986 {
 987         CommandList_struct *c;
 988         int i;
 989         u64bit temp64;
 990         dma_addr_t cmd_dma_handle, err_dma_handle;
 991
 992         do {
 993                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 994                 if (i == h->nr_cmds)
 995                         return NULL;
 996         } while (test_and_set_bit(i, h->cmd_pool_bits) != 0);
 997         c = h->cmd_pool + i;
 998         memset(c, 0, sizeof(CommandList_struct));
 999         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
1000         c->err_info = h->errinfo_pool + i;
1001         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
1002         err_dma_handle = h->errinfo_pool_dhandle
1003             + i * sizeof(ErrorInfo_struct);
1004         h->nr_allocs++;
1005
1006         c->cmdindex = i;
1007
1008         INIT_LIST_HEAD(&c->list);
1009         c->busaddr = (__u32) cmd_dma_handle;
1010         temp64.val = (__u64) err_dma_handle;
1011         c->ErrDesc.Addr.lower = temp64.val32.lower;
1012         c->ErrDesc.Addr.upper = temp64.val32.upper;
1013         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
1014
1015         c->ctlr = h->ctlr;
1016         return c;
1017 }
1018
1019 /* allocate a command using pci_alloc_consistent, used for ioctls,
1020  * etc., not for the main i/o path.
1021  */
1022 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
1023 {
1024         CommandList_struct *c;
1025         u64bit temp64;
1026         dma_addr_t cmd_dma_handle, err_dma_handle;
1027
1028         c = pci_zalloc_consistent(h->pdev, sizeof(CommandList_struct),
1029                                   &cmd_dma_handle);
1030         if (c == NULL)
1031                 return NULL;
1032
1033         c->cmdindex = -1;
1034
1035         c->err_info = pci_zalloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
1036                                             &err_dma_handle);
1037
1038         if (c->err_info == NULL) {
1039                 pci_free_consistent(h->pdev,
1040                         sizeof(CommandList_struct), c, cmd_dma_handle);
1041                 return NULL;
1042         }
1043
1044         INIT_LIST_HEAD(&c->list);
1045         c->busaddr = (__u32) cmd_dma_handle;
1046         temp64.val = (__u64) err_dma_handle;
1047         c->ErrDesc.Addr.lower = temp64.val32.lower;
1048         c->ErrDesc.Addr.upper = temp64.val32.upper;
1049         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
1050
1051         c->ctlr = h->ctlr;
1052         return c;
1053 }
1054
1055 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
1056 {
1057         int i;
1058
1059         i = c - h->cmd_pool;
1060         clear_bit(i, h->cmd_pool_bits);
1061         h->nr_frees++;
1062 }
1063
1064 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
1065 {
1066         u64bit temp64;
1067
1068         temp64.val32.lower = c->ErrDesc.Addr.lower;
1069         temp64.val32.upper = c->ErrDesc.Addr.upper;
1070         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1071                             c->err_info, (dma_addr_t) temp64.val);
1072         pci_free_consistent(h->pdev, sizeof(CommandList_struct), c,
1073                 (dma_addr_t) cciss_tag_discard_error_bits(h, (u32) c->busaddr));
1074 }
1075
1076 static inline ctlr_info_t *get_host(struct gendisk *disk)
1077 {
1078         return disk->queue->queuedata;
1079 }
1080
1081 static inline drive_info_struct *get_drv(struct gendisk *disk)
1082 {
1083         return disk->private_data;
1084 }
1085
1086 /*
1087  * Open.  Make sure the device is really there.
1088  */
1089 static int cciss_open(struct block_device *bdev, fmode_t mode)
1090 {
1091         ctlr_info_t *h = get_host(bdev->bd_disk);
1092         drive_info_struct *drv = get_drv(bdev->bd_disk);
1093
1094         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1095         if (drv->busy_configuring)
1096                 return -EBUSY;
1097         /*
1098          * Root is allowed to open raw volume zero even if it's not configured
1099          * so array config can still work. Root is also allowed to open any
1100          * volume that has a LUN ID, so it can issue IOCTL to reread the
1101          * disk information.  I don't think I really like this
1102          * but I'm already using way to many device nodes to claim another one
1103          * for "raw controller".
1104          */
1105         if (drv->heads == 0) {
1106                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1107                         /* if not node 0 make sure it is a partition = 0 */
1108                         if (MINOR(bdev->bd_dev) & 0x0f) {
1109                                 return -ENXIO;
1110                                 /* if it is, make sure we have a LUN ID */
1111                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1112                                 sizeof(drv->LunID))) {
1113                                 return -ENXIO;
1114                         }
1115                 }
1116                 if (!capable(CAP_SYS_ADMIN))
1117                         return -EPERM;
1118         }
1119         drv->usage_count++;
1120         h->usage_count++;
1121         return 0;
1122 }
1123
1124 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1125 {
1126         int ret;
1127
1128         mutex_lock(&cciss_mutex);
1129         ret = cciss_open(bdev, mode);
1130         mutex_unlock(&cciss_mutex);
1131
1132         return ret;
1133 }
1134
1135 /*
1136  * Close.  Sync first.
1137  */
1138 static void cciss_release(struct gendisk *disk, fmode_t mode)
1139 {
1140         ctlr_info_t *h;
1141         drive_info_struct *drv;
1142
1143         mutex_lock(&cciss_mutex);
1144         h = get_host(disk);
1145         drv = get_drv(disk);
1146         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1147         drv->usage_count--;
1148         h->usage_count--;
1149         mutex_unlock(&cciss_mutex);
1150 }
1151
1152 #ifdef CONFIG_COMPAT
1153
1154 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1155                                   unsigned cmd, unsigned long arg);
1156 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1157                                       unsigned cmd, unsigned long arg);
1158
1159 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1160                               unsigned cmd, unsigned long arg)
1161 {
1162         switch (cmd) {
1163         case CCISS_GETPCIINFO:
1164         case CCISS_GETINTINFO:
1165         case CCISS_SETINTINFO:
1166         case CCISS_GETNODENAME:
1167         case CCISS_SETNODENAME:
1168         case CCISS_GETHEARTBEAT:
1169         case CCISS_GETBUSTYPES:
1170         case CCISS_GETFIRMVER:
1171         case CCISS_GETDRIVVER:
1172         case CCISS_REVALIDVOLS:
1173         case CCISS_DEREGDISK:
1174         case CCISS_REGNEWDISK:
1175         case CCISS_REGNEWD:
1176         case CCISS_RESCANDISK:
1177         case CCISS_GETLUNINFO:
1178                 return cciss_ioctl(bdev, mode, cmd, arg);
1179
1180         case CCISS_PASSTHRU32:
1181                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1182         case CCISS_BIG_PASSTHRU32:
1183                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1184
1185         default:
1186                 return -ENOIOCTLCMD;
1187         }
1188 }
1189
1190 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1191                                   unsigned cmd, unsigned long arg)
1192 {
1193         IOCTL32_Command_struct __user *arg32 =
1194             (IOCTL32_Command_struct __user *) arg;
1195         IOCTL_Command_struct arg64;
1196         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1197         int err;
1198         u32 cp;
1199
1200         memset(&arg64, 0, sizeof(arg64));
1201         err = 0;
1202         err |=
1203             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1204                            sizeof(arg64.LUN_info));
1205         err |=
1206             copy_from_user(&arg64.Request, &arg32->Request,
1207                            sizeof(arg64.Request));
1208         err |=
1209             copy_from_user(&arg64.error_info, &arg32->error_info,
1210                            sizeof(arg64.error_info));
1211         err |= get_user(arg64.buf_size, &arg32->buf_size);
1212         err |= get_user(cp, &arg32->buf);
1213         arg64.buf = compat_ptr(cp);
1214         err |= copy_to_user(p, &arg64, sizeof(arg64));
1215
1216         if (err)
1217                 return -EFAULT;
1218
1219         err = cciss_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1220         if (err)
1221                 return err;
1222         err |=
1223             copy_in_user(&arg32->error_info, &p->error_info,
1224                          sizeof(arg32->error_info));
1225         if (err)
1226                 return -EFAULT;
1227         return err;
1228 }
1229
1230 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1231                                       unsigned cmd, unsigned long arg)
1232 {
1233         BIG_IOCTL32_Command_struct __user *arg32 =
1234             (BIG_IOCTL32_Command_struct __user *) arg;
1235         BIG_IOCTL_Command_struct arg64;
1236         BIG_IOCTL_Command_struct __user *p =
1237             compat_alloc_user_space(sizeof(arg64));
1238         int err;
1239         u32 cp;
1240
1241         memset(&arg64, 0, sizeof(arg64));
1242         err = 0;
1243         err |=
1244             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1245                            sizeof(arg64.LUN_info));
1246         err |=
1247             copy_from_user(&arg64.Request, &arg32->Request,
1248                            sizeof(arg64.Request));
1249         err |=
1250             copy_from_user(&arg64.error_info, &arg32->error_info,
1251                            sizeof(arg64.error_info));
1252         err |= get_user(arg64.buf_size, &arg32->buf_size);
1253         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1254         err |= get_user(cp, &arg32->buf);
1255         arg64.buf = compat_ptr(cp);
1256         err |= copy_to_user(p, &arg64, sizeof(arg64));
1257
1258         if (err)
1259                 return -EFAULT;
1260
1261         err = cciss_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1262         if (err)
1263                 return err;
1264         err |=
1265             copy_in_user(&arg32->error_info, &p->error_info,
1266                          sizeof(arg32->error_info));
1267         if (err)
1268                 return -EFAULT;
1269         return err;
1270 }
1271 #endif
1272
1273 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1274 {
1275         drive_info_struct *drv = get_drv(bdev->bd_disk);
1276
1277         if (!drv->cylinders)
1278                 return -ENXIO;
1279
1280         geo->heads = drv->heads;
1281         geo->sectors = drv->sectors;
1282         geo->cylinders = drv->cylinders;
1283         return 0;
1284 }
1285
1286 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1287 {
1288         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1289                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1290                 (void)check_for_unit_attention(h, c);
1291 }
1292
1293 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1294 {
1295         cciss_pci_info_struct pciinfo;
1296
1297         if (!argp)
1298                 return -EINVAL;
1299         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1300         pciinfo.bus = h->pdev->bus->number;
1301         pciinfo.dev_fn = h->pdev->devfn;
1302         pciinfo.board_id = h->board_id;
1303         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1304                 return -EFAULT;
1305         return 0;
1306 }
1307
1308 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1309 {
1310         cciss_coalint_struct intinfo;
1311         unsigned long flags;
1312
1313         if (!argp)
1314                 return -EINVAL;
1315         spin_lock_irqsave(&h->lock, flags);
1316         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1317         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1318         spin_unlock_irqrestore(&h->lock, flags);
1319         if (copy_to_user
1320             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1321                 return -EFAULT;
1322         return 0;
1323 }
1324
1325 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1326 {
1327         cciss_coalint_struct intinfo;
1328         unsigned long flags;
1329         int i;
1330
1331         if (!argp)
1332                 return -EINVAL;
1333         if (!capable(CAP_SYS_ADMIN))
1334                 return -EPERM;
1335         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1336                 return -EFAULT;
1337         if ((intinfo.delay == 0) && (intinfo.count == 0))
1338                 return -EINVAL;
1339         spin_lock_irqsave(&h->lock, flags);
1340         /* Update the field, and then ring the doorbell */
1341         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1342         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1343         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1344
1345         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1346                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1347                         break;
1348                 udelay(1000); /* delay and try again */
1349         }
1350         spin_unlock_irqrestore(&h->lock, flags);
1351         if (i >= MAX_IOCTL_CONFIG_WAIT)
1352                 return -EAGAIN;
1353         return 0;
1354 }
1355
1356 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1357 {
1358         NodeName_type NodeName;
1359         unsigned long flags;
1360         int i;
1361
1362         if (!argp)
1363                 return -EINVAL;
1364         spin_lock_irqsave(&h->lock, flags);
1365         for (i = 0; i < 16; i++)
1366                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1367         spin_unlock_irqrestore(&h->lock, flags);
1368         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1369                 return -EFAULT;
1370         return 0;
1371 }
1372
1373 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1374 {
1375         NodeName_type NodeName;
1376         unsigned long flags;
1377         int i;
1378
1379         if (!argp)
1380                 return -EINVAL;
1381         if (!capable(CAP_SYS_ADMIN))
1382                 return -EPERM;
1383         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1384                 return -EFAULT;
1385         spin_lock_irqsave(&h->lock, flags);
1386         /* Update the field, and then ring the doorbell */
1387         for (i = 0; i < 16; i++)
1388                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1389         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1390         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1391                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1392                         break;
1393                 udelay(1000); /* delay and try again */
1394         }
1395         spin_unlock_irqrestore(&h->lock, flags);
1396         if (i >= MAX_IOCTL_CONFIG_WAIT)
1397                 return -EAGAIN;
1398         return 0;
1399 }
1400
1401 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1402 {
1403         Heartbeat_type heartbeat;
1404         unsigned long flags;
1405
1406         if (!argp)
1407                 return -EINVAL;
1408         spin_lock_irqsave(&h->lock, flags);
1409         heartbeat = readl(&h->cfgtable->HeartBeat);
1410         spin_unlock_irqrestore(&h->lock, flags);
1411         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1412                 return -EFAULT;
1413         return 0;
1414 }
1415
1416 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1417 {
1418         BusTypes_type BusTypes;
1419         unsigned long flags;
1420
1421         if (!argp)
1422                 return -EINVAL;
1423         spin_lock_irqsave(&h->lock, flags);
1424         BusTypes = readl(&h->cfgtable->BusTypes);
1425         spin_unlock_irqrestore(&h->lock, flags);
1426         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1427                 return -EFAULT;
1428         return 0;
1429 }
1430
1431 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1432 {
1433         FirmwareVer_type firmware;
1434
1435         if (!argp)
1436                 return -EINVAL;
1437         memcpy(firmware, h->firm_ver, 4);
1438
1439         if (copy_to_user
1440             (argp, firmware, sizeof(FirmwareVer_type)))
1441                 return -EFAULT;
1442         return 0;
1443 }
1444
1445 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1446 {
1447         DriverVer_type DriverVer = DRIVER_VERSION;
1448
1449         if (!argp)
1450                 return -EINVAL;
1451         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1452                 return -EFAULT;
1453         return 0;
1454 }
1455
1456 static int cciss_getluninfo(ctlr_info_t *h,
1457         struct gendisk *disk, void __user *argp)
1458 {
1459         LogvolInfo_struct luninfo;
1460         drive_info_struct *drv = get_drv(disk);
1461
1462         if (!argp)
1463                 return -EINVAL;
1464         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1465         luninfo.num_opens = drv->usage_count;
1466         luninfo.num_parts = 0;
1467         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1468                 return -EFAULT;
1469         return 0;
1470 }
1471
1472 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1473 {
1474         IOCTL_Command_struct iocommand;
1475         CommandList_struct *c;
1476         char *buff = NULL;
1477         u64bit temp64;
1478         DECLARE_COMPLETION_ONSTACK(wait);
1479
1480         if (!argp)
1481                 return -EINVAL;
1482
1483         if (!capable(CAP_SYS_RAWIO))
1484                 return -EPERM;
1485
1486         if (copy_from_user
1487             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1488                 return -EFAULT;
1489         if ((iocommand.buf_size < 1) &&
1490             (iocommand.Request.Type.Direction != XFER_NONE)) {
1491                 return -EINVAL;
1492         }
1493         if (iocommand.buf_size > 0) {
1494                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1495                 if (buff == NULL)
1496                         return -EFAULT;
1497         }
1498         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1499                 /* Copy the data into the buffer we created */
1500                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1501                         kfree(buff);
1502                         return -EFAULT;
1503                 }
1504         } else {
1505                 memset(buff, 0, iocommand.buf_size);
1506         }
1507         c = cmd_special_alloc(h);
1508         if (!c) {
1509                 kfree(buff);
1510                 return -ENOMEM;
1511         }
1512         /* Fill in the command type */
1513         c->cmd_type = CMD_IOCTL_PEND;
1514         /* Fill in Command Header */
1515         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1516         if (iocommand.buf_size > 0) { /* buffer to fill */
1517                 c->Header.SGList = 1;
1518                 c->Header.SGTotal = 1;
1519         } else { /* no buffers to fill */
1520                 c->Header.SGList = 0;
1521                 c->Header.SGTotal = 0;
1522         }
1523         c->Header.LUN = iocommand.LUN_info;
1524         /* use the kernel address the cmd block for tag */
1525         c->Header.Tag.lower = c->busaddr;
1526
1527         /* Fill in Request block */
1528         c->Request = iocommand.Request;
1529
1530         /* Fill in the scatter gather information */
1531         if (iocommand.buf_size > 0) {
1532                 temp64.val = pci_map_single(h->pdev, buff,
1533                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1534                 c->SG[0].Addr.lower = temp64.val32.lower;
1535                 c->SG[0].Addr.upper = temp64.val32.upper;
1536                 c->SG[0].Len = iocommand.buf_size;
1537                 c->SG[0].Ext = 0;  /* we are not chaining */
1538         }
1539         c->waiting = &wait;
1540
1541         enqueue_cmd_and_start_io(h, c);
1542         wait_for_completion(&wait);
1543
1544         /* unlock the buffers from DMA */
1545         temp64.val32.lower = c->SG[0].Addr.lower;
1546         temp64.val32.upper = c->SG[0].Addr.upper;
1547         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1548                          PCI_DMA_BIDIRECTIONAL);
1549         check_ioctl_unit_attention(h, c);
1550
1551         /* Copy the error information out */
1552         iocommand.error_info = *(c->err_info);
1553         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1554                 kfree(buff);
1555                 cmd_special_free(h, c);
1556                 return -EFAULT;
1557         }
1558
1559         if (iocommand.Request.Type.Direction == XFER_READ) {
1560                 /* Copy the data out of the buffer we created */
1561                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1562                         kfree(buff);
1563                         cmd_special_free(h, c);
1564                         return -EFAULT;
1565                 }
1566         }
1567         kfree(buff);
1568         cmd_special_free(h, c);
1569         return 0;
1570 }
1571
1572 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1573 {
1574         BIG_IOCTL_Command_struct *ioc;
1575         CommandList_struct *c;
1576         unsigned char **buff = NULL;
1577         int *buff_size = NULL;
1578         u64bit temp64;
1579         BYTE sg_used = 0;
1580         int status = 0;
1581         int i;
1582         DECLARE_COMPLETION_ONSTACK(wait);
1583         __u32 left;
1584         __u32 sz;
1585         BYTE __user *data_ptr;
1586
1587         if (!argp)
1588                 return -EINVAL;
1589         if (!capable(CAP_SYS_RAWIO))
1590                 return -EPERM;
1591         ioc = kmalloc(sizeof(*ioc), GFP_KERNEL);
1592         if (!ioc) {
1593                 status = -ENOMEM;
1594                 goto cleanup1;
1595         }
1596         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1597                 status = -EFAULT;
1598                 goto cleanup1;
1599         }
1600         if ((ioc->buf_size < 1) &&
1601             (ioc->Request.Type.Direction != XFER_NONE)) {
1602                 status = -EINVAL;
1603                 goto cleanup1;
1604         }
1605         /* Check kmalloc limits  using all SGs */
1606         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1607                 status = -EINVAL;
1608                 goto cleanup1;
1609         }
1610         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1611                 status = -EINVAL;
1612                 goto cleanup1;
1613         }
1614         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1615         if (!buff) {
1616                 status = -ENOMEM;
1617                 goto cleanup1;
1618         }
1619         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1620         if (!buff_size) {
1621                 status = -ENOMEM;
1622                 goto cleanup1;
1623         }
1624         left = ioc->buf_size;
1625         data_ptr = ioc->buf;
1626         while (left) {
1627                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1628                 buff_size[sg_used] = sz;
1629                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1630                 if (buff[sg_used] == NULL) {
1631                         status = -ENOMEM;
1632                         goto cleanup1;
1633                 }
1634                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1635                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1636                                 status = -EFAULT;
1637                                 goto cleanup1;
1638                         }
1639                 } else {
1640                         memset(buff[sg_used], 0, sz);
1641                 }
1642                 left -= sz;
1643                 data_ptr += sz;
1644                 sg_used++;
1645         }
1646         c = cmd_special_alloc(h);
1647         if (!c) {
1648                 status = -ENOMEM;
1649                 goto cleanup1;
1650         }
1651         c->cmd_type = CMD_IOCTL_PEND;
1652         c->Header.ReplyQueue = 0;
1653         c->Header.SGList = sg_used;
1654         c->Header.SGTotal = sg_used;
1655         c->Header.LUN = ioc->LUN_info;
1656         c->Header.Tag.lower = c->busaddr;
1657
1658         c->Request = ioc->Request;
1659         for (i = 0; i < sg_used; i++) {
1660                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1661                                     PCI_DMA_BIDIRECTIONAL);
1662                 c->SG[i].Addr.lower = temp64.val32.lower;
1663                 c->SG[i].Addr.upper = temp64.val32.upper;
1664                 c->SG[i].Len = buff_size[i];
1665                 c->SG[i].Ext = 0;       /* we are not chaining */
1666         }
1667         c->waiting = &wait;
1668         enqueue_cmd_and_start_io(h, c);
1669         wait_for_completion(&wait);
1670         /* unlock the buffers from DMA */
1671         for (i = 0; i < sg_used; i++) {
1672                 temp64.val32.lower = c->SG[i].Addr.lower;
1673                 temp64.val32.upper = c->SG[i].Addr.upper;
1674                 pci_unmap_single(h->pdev,
1675                         (dma_addr_t) temp64.val, buff_size[i],
1676                         PCI_DMA_BIDIRECTIONAL);
1677         }
1678         check_ioctl_unit_attention(h, c);
1679         /* Copy the error information out */
1680         ioc->error_info = *(c->err_info);
1681         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1682                 cmd_special_free(h, c);
1683                 status = -EFAULT;
1684                 goto cleanup1;
1685         }
1686         if (ioc->Request.Type.Direction == XFER_READ) {
1687                 /* Copy the data out of the buffer we created */
1688                 BYTE __user *ptr = ioc->buf;
1689                 for (i = 0; i < sg_used; i++) {
1690                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1691                                 cmd_special_free(h, c);
1692                                 status = -EFAULT;
1693                                 goto cleanup1;
1694                         }
1695                         ptr += buff_size[i];
1696                 }
1697         }
1698         cmd_special_free(h, c);
1699         status = 0;
1700 cleanup1:
1701         if (buff) {
1702                 for (i = 0; i < sg_used; i++)
1703                         kfree(buff[i]);
1704                 kfree(buff);
1705         }
1706         kfree(buff_size);
1707         kfree(ioc);
1708         return status;
1709 }
1710
1711 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1712         unsigned int cmd, unsigned long arg)
1713 {
1714         struct gendisk *disk = bdev->bd_disk;
1715         ctlr_info_t *h = get_host(disk);
1716         void __user *argp = (void __user *)arg;
1717
1718         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1719                 cmd, arg);
1720         switch (cmd) {
1721         case CCISS_GETPCIINFO:
1722                 return cciss_getpciinfo(h, argp);
1723         case CCISS_GETINTINFO:
1724                 return cciss_getintinfo(h, argp);
1725         case CCISS_SETINTINFO:
1726                 return cciss_setintinfo(h, argp);
1727         case CCISS_GETNODENAME:
1728                 return cciss_getnodename(h, argp);
1729         case CCISS_SETNODENAME:
1730                 return cciss_setnodename(h, argp);
1731         case CCISS_GETHEARTBEAT:
1732                 return cciss_getheartbeat(h, argp);
1733         case CCISS_GETBUSTYPES:
1734                 return cciss_getbustypes(h, argp);
1735         case CCISS_GETFIRMVER:
1736                 return cciss_getfirmver(h, argp);
1737         case CCISS_GETDRIVVER:
1738                 return cciss_getdrivver(h, argp);
1739         case CCISS_DEREGDISK:
1740         case CCISS_REGNEWD:
1741         case CCISS_REVALIDVOLS:
1742                 return rebuild_lun_table(h, 0, 1);
1743         case CCISS_GETLUNINFO:
1744                 return cciss_getluninfo(h, disk, argp);
1745         case CCISS_PASSTHRU:
1746                 return cciss_passthru(h, argp);
1747         case CCISS_BIG_PASSTHRU:
1748                 return cciss_bigpassthru(h, argp);
1749
1750         /* scsi_cmd_blk_ioctl handles these, below, though some are not */
1751         /* very meaningful for cciss.  SG_IO is the main one people want. */
1752
1753         case SG_GET_VERSION_NUM:
1754         case SG_SET_TIMEOUT:
1755         case SG_GET_TIMEOUT:
1756         case SG_GET_RESERVED_SIZE:
1757         case SG_SET_RESERVED_SIZE:
1758         case SG_EMULATED_HOST:
1759         case SG_IO:
1760         case SCSI_IOCTL_SEND_COMMAND:
1761                 return scsi_cmd_blk_ioctl(bdev, mode, cmd, argp);
1762
1763         /* scsi_cmd_blk_ioctl would normally handle these, below, but */
1764         /* they aren't a good fit for cciss, as CD-ROMs are */
1765         /* not supported, and we don't have any bus/target/lun */
1766         /* which we present to the kernel. */
1767
1768         case CDROM_SEND_PACKET:
1769         case CDROMCLOSETRAY:
1770         case CDROMEJECT:
1771         case SCSI_IOCTL_GET_IDLUN:
1772         case SCSI_IOCTL_GET_BUS_NUMBER:
1773         default:
1774                 return -ENOTTY;
1775         }
1776 }
1777
1778 static void cciss_check_queues(ctlr_info_t *h)
1779 {
1780         int start_queue = h->next_to_run;
1781         int i;
1782
1783         /* check to see if we have maxed out the number of commands that can
1784          * be placed on the queue.  If so then exit.  We do this check here
1785          * in case the interrupt we serviced was from an ioctl and did not
1786          * free any new commands.
1787          */
1788         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1789                 return;
1790
1791         /* We have room on the queue for more commands.  Now we need to queue
1792          * them up.  We will also keep track of the next queue to run so
1793          * that every queue gets a chance to be started first.
1794          */
1795         for (i = 0; i < h->highest_lun + 1; i++) {
1796                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1797                 /* make sure the disk has been added and the drive is real
1798                  * because this can be called from the middle of init_one.
1799                  */
1800                 if (!h->drv[curr_queue])
1801                         continue;
1802                 if (!(h->drv[curr_queue]->queue) ||
1803                         !(h->drv[curr_queue]->heads))
1804                         continue;
1805                 blk_start_queue(h->gendisk[curr_queue]->queue);
1806
1807                 /* check to see if we have maxed out the number of commands
1808                  * that can be placed on the queue.
1809                  */
1810                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1811                         if (curr_queue == start_queue) {
1812                                 h->next_to_run =
1813                                     (start_queue + 1) % (h->highest_lun + 1);
1814                                 break;
1815                         } else {
1816                                 h->next_to_run = curr_queue;
1817                                 break;
1818                         }
1819                 }
1820         }
1821 }
1822
1823 static void cciss_softirq_done(struct request *rq)
1824 {
1825         CommandList_struct *c = rq->completion_data;
1826         ctlr_info_t *h = hba[c->ctlr];
1827         SGDescriptor_struct *curr_sg = c->SG;
1828         u64bit temp64;
1829         unsigned long flags;
1830         int i, ddir;
1831         int sg_index = 0;
1832
1833         if (c->Request.Type.Direction == XFER_READ)
1834                 ddir = PCI_DMA_FROMDEVICE;
1835         else
1836                 ddir = PCI_DMA_TODEVICE;
1837
1838         /* command did not need to be retried */
1839         /* unmap the DMA mapping for all the scatter gather elements */
1840         for (i = 0; i < c->Header.SGList; i++) {
1841                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1842                         cciss_unmap_sg_chain_block(h, c);
1843                         /* Point to the next block */
1844                         curr_sg = h->cmd_sg_list[c->cmdindex];
1845                         sg_index = 0;
1846                 }
1847                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1848                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1849                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1850                                 ddir);
1851                 ++sg_index;
1852         }
1853
1854         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1855
1856         /* set the residual count for pc requests */
1857         if (blk_rq_is_passthrough(rq))
1858                 scsi_req(rq)->resid_len = c->err_info->ResidualCnt;
1859
1860         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1861
1862         spin_lock_irqsave(&h->lock, flags);
1863         cmd_free(h, c);
1864         cciss_check_queues(h);
1865         spin_unlock_irqrestore(&h->lock, flags);
1866 }
1867
1868 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1869         unsigned char scsi3addr[], uint32_t log_unit)
1870 {
1871         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1872                 sizeof(h->drv[log_unit]->LunID));
1873 }
1874
1875 /* This function gets the SCSI vendor, model, and revision of a logical drive
1876  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1877  * they cannot be read.
1878  */
1879 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1880                                    char *vendor, char *model, char *rev)
1881 {
1882         int rc;
1883         InquiryData_struct *inq_buf;
1884         unsigned char scsi3addr[8];
1885
1886         *vendor = '\0';
1887         *model = '\0';
1888         *rev = '\0';
1889
1890         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1891         if (!inq_buf)
1892                 return;
1893
1894         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1895         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1896                         scsi3addr, TYPE_CMD);
1897         if (rc == IO_OK) {
1898                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1899                 vendor[VENDOR_LEN] = '\0';
1900                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1901                 model[MODEL_LEN] = '\0';
1902                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1903                 rev[REV_LEN] = '\0';
1904         }
1905
1906         kfree(inq_buf);
1907         return;
1908 }
1909
1910 /* This function gets the serial number of a logical drive via
1911  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1912  * number cannot be had, for whatever reason, 16 bytes of 0xff
1913  * are returned instead.
1914  */
1915 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1916                                 unsigned char *serial_no, int buflen)
1917 {
1918 #define PAGE_83_INQ_BYTES 64
1919         int rc;
1920         unsigned char *buf;
1921         unsigned char scsi3addr[8];
1922
1923         if (buflen > 16)
1924                 buflen = 16;
1925         memset(serial_no, 0xff, buflen);
1926         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1927         if (!buf)
1928                 return;
1929         memset(serial_no, 0, buflen);
1930         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1931         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1932                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1933         if (rc == IO_OK)
1934                 memcpy(serial_no, &buf[8], buflen);
1935         kfree(buf);
1936         return;
1937 }
1938
1939 /*
1940  * cciss_add_disk sets up the block device queue for a logical drive
1941  */
1942 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1943                                 int drv_index)
1944 {
1945         disk->queue = blk_alloc_queue(GFP_KERNEL);
1946         if (!disk->queue)
1947                 goto init_queue_failure;
1948
1949         disk->queue->cmd_size = sizeof(struct scsi_request);
1950         disk->queue->request_fn = do_cciss_request;
1951         disk->queue->queue_lock = &h->lock;
1952         if (blk_init_allocated_queue(disk->queue) < 0)
1953                 goto cleanup_queue;
1954
1955         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1956         disk->major = h->major;
1957         disk->first_minor = drv_index << NWD_SHIFT;
1958         disk->fops = &cciss_fops;
1959         if (cciss_create_ld_sysfs_entry(h, drv_index))
1960                 goto cleanup_queue;
1961         disk->private_data = h->drv[drv_index];
1962
1963         /* Set up queue information */
1964         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1965
1966         /* This is a hardware imposed limit. */
1967         blk_queue_max_segments(disk->queue, h->maxsgentries);
1968
1969         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1970
1971         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1972
1973         disk->queue->queuedata = h;
1974
1975         blk_queue_logical_block_size(disk->queue,
1976                                      h->drv[drv_index]->block_size);
1977
1978         /* Make sure all queue data is written out before */
1979         /* setting h->drv[drv_index]->queue, as setting this */
1980         /* allows the interrupt handler to start the queue */
1981         wmb();
1982         h->drv[drv_index]->queue = disk->queue;
1983         device_add_disk(&h->drv[drv_index]->dev, disk);
1984         return 0;
1985
1986 cleanup_queue:
1987         blk_cleanup_queue(disk->queue);
1988         disk->queue = NULL;
1989 init_queue_failure:
1990         return -1;
1991 }
1992
1993 /* This function will check the usage_count of the drive to be updated/added.
1994  * If the usage_count is zero and it is a heretofore unknown drive, or,
1995  * the drive's capacity, geometry, or serial number has changed,
1996  * then the drive information will be updated and the disk will be
1997  * re-registered with the kernel.  If these conditions don't hold,
1998  * then it will be left alone for the next reboot.  The exception to this
1999  * is disk 0 which will always be left registered with the kernel since it
2000  * is also the controller node.  Any changes to disk 0 will show up on
2001  * the next reboot.
2002  */
2003 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
2004         int first_time, int via_ioctl)
2005 {
2006         struct gendisk *disk;
2007         InquiryData_struct *inq_buff = NULL;
2008         unsigned int block_size;
2009         sector_t total_size;
2010         unsigned long flags = 0;
2011         int ret = 0;
2012         drive_info_struct *drvinfo;
2013
2014         /* Get information about the disk and modify the driver structure */
2015         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2016         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
2017         if (inq_buff == NULL || drvinfo == NULL)
2018                 goto mem_msg;
2019
2020         /* testing to see if 16-byte CDBs are already being used */
2021         if (h->cciss_read == CCISS_READ_16) {
2022                 cciss_read_capacity_16(h, drv_index,
2023                         &total_size, &block_size);
2024
2025         } else {
2026                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
2027                 /* if read_capacity returns all F's this volume is >2TB */
2028                 /* in size so we switch to 16-byte CDB's for all */
2029                 /* read/write ops */
2030                 if (total_size == 0xFFFFFFFFULL) {
2031                         cciss_read_capacity_16(h, drv_index,
2032                         &total_size, &block_size);
2033                         h->cciss_read = CCISS_READ_16;
2034                         h->cciss_write = CCISS_WRITE_16;
2035                 } else {
2036                         h->cciss_read = CCISS_READ_10;
2037                         h->cciss_write = CCISS_WRITE_10;
2038                 }
2039         }
2040
2041         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
2042                                inq_buff, drvinfo);
2043         drvinfo->block_size = block_size;
2044         drvinfo->nr_blocks = total_size + 1;
2045
2046         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
2047                                 drvinfo->model, drvinfo->rev);
2048         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
2049                         sizeof(drvinfo->serial_no));
2050         /* Save the lunid in case we deregister the disk, below. */
2051         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
2052                 sizeof(drvinfo->LunID));
2053
2054         /* Is it the same disk we already know, and nothing's changed? */
2055         if (h->drv[drv_index]->raid_level != -1 &&
2056                 ((memcmp(drvinfo->serial_no,
2057                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2058                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2059                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2060                 drvinfo->heads == h->drv[drv_index]->heads &&
2061                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2062                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2063                         /* The disk is unchanged, nothing to update */
2064                         goto freeret;
2065
2066         /* If we get here it's not the same disk, or something's changed,
2067          * so we need to * deregister it, and re-register it, if it's not
2068          * in use.
2069          * If the disk already exists then deregister it before proceeding
2070          * (unless it's the first disk (for the controller node).
2071          */
2072         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2073                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2074                 spin_lock_irqsave(&h->lock, flags);
2075                 h->drv[drv_index]->busy_configuring = 1;
2076                 spin_unlock_irqrestore(&h->lock, flags);
2077
2078                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2079                  * which keeps the interrupt handler from starting
2080                  * the queue.
2081                  */
2082                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2083         }
2084
2085         /* If the disk is in use return */
2086         if (ret)
2087                 goto freeret;
2088
2089         /* Save the new information from cciss_geometry_inquiry
2090          * and serial number inquiry.  If the disk was deregistered
2091          * above, then h->drv[drv_index] will be NULL.
2092          */
2093         if (h->drv[drv_index] == NULL) {
2094                 drvinfo->device_initialized = 0;
2095                 h->drv[drv_index] = drvinfo;
2096                 drvinfo = NULL; /* so it won't be freed below. */
2097         } else {
2098                 /* special case for cxd0 */
2099                 h->drv[drv_index]->block_size = drvinfo->block_size;
2100                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2101                 h->drv[drv_index]->heads = drvinfo->heads;
2102                 h->drv[drv_index]->sectors = drvinfo->sectors;
2103                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2104                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2105                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2106                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2107                         VENDOR_LEN + 1);
2108                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2109                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2110         }
2111
2112         ++h->num_luns;
2113         disk = h->gendisk[drv_index];
2114         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2115
2116         /* If it's not disk 0 (drv_index != 0)
2117          * or if it was disk 0, but there was previously
2118          * no actual corresponding configured logical drive
2119          * (raid_leve == -1) then we want to update the
2120          * logical drive's information.
2121          */
2122         if (drv_index || first_time) {
2123                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2124                         cciss_free_gendisk(h, drv_index);
2125                         cciss_free_drive_info(h, drv_index);
2126                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2127                                 drv_index);
2128                         --h->num_luns;
2129                 }
2130         }
2131
2132 freeret:
2133         kfree(inq_buff);
2134         kfree(drvinfo);
2135         return;
2136 mem_msg:
2137         dev_err(&h->pdev->dev, "out of memory\n");
2138         goto freeret;
2139 }
2140
2141 /* This function will find the first index of the controllers drive array
2142  * that has a null drv pointer and allocate the drive info struct and
2143  * will return that index   This is where new drives will be added.
2144  * If the index to be returned is greater than the highest_lun index for
2145  * the controller then highest_lun is set * to this new index.
2146  * If there are no available indexes or if tha allocation fails, then -1
2147  * is returned.  * "controller_node" is used to know if this is a real
2148  * logical drive, or just the controller node, which determines if this
2149  * counts towards highest_lun.
2150  */
2151 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2152 {
2153         int i;
2154         drive_info_struct *drv;
2155
2156         /* Search for an empty slot for our drive info */
2157         for (i = 0; i < CISS_MAX_LUN; i++) {
2158
2159                 /* if not cxd0 case, and it's occupied, skip it. */
2160                 if (h->drv[i] && i != 0)
2161                         continue;
2162                 /*
2163                  * If it's cxd0 case, and drv is alloc'ed already, and a
2164                  * disk is configured there, skip it.
2165                  */
2166                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2167                         continue;
2168
2169                 /*
2170                  * We've found an empty slot.  Update highest_lun
2171                  * provided this isn't just the fake cxd0 controller node.
2172                  */
2173                 if (i > h->highest_lun && !controller_node)
2174                         h->highest_lun = i;
2175
2176                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2177                 if (i == 0 && h->drv[i] != NULL)
2178                         return i;
2179
2180                 /*
2181                  * Found an empty slot, not already alloc'ed.  Allocate it.
2182                  * Mark it with raid_level == -1, so we know it's new later on.
2183                  */
2184                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2185                 if (!drv)
2186                         return -1;
2187                 drv->raid_level = -1; /* so we know it's new */
2188                 h->drv[i] = drv;
2189                 return i;
2190         }
2191         return -1;
2192 }
2193
2194 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2195 {
2196         kfree(h->drv[drv_index]);
2197         h->drv[drv_index] = NULL;
2198 }
2199
2200 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2201 {
2202         put_disk(h->gendisk[drv_index]);
2203         h->gendisk[drv_index] = NULL;
2204 }
2205
2206 /* cciss_add_gendisk finds a free hba[]->drv structure
2207  * and allocates a gendisk if needed, and sets the lunid
2208  * in the drvinfo structure.   It returns the index into
2209  * the ->drv[] array, or -1 if none are free.
2210  * is_controller_node indicates whether highest_lun should
2211  * count this disk, or if it's only being added to provide
2212  * a means to talk to the controller in case no logical
2213  * drives have yet been configured.
2214  */
2215 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2216         int controller_node)
2217 {
2218         int drv_index;
2219
2220         drv_index = cciss_alloc_drive_info(h, controller_node);
2221         if (drv_index == -1)
2222                 return -1;
2223
2224         /*Check if the gendisk needs to be allocated */
2225         if (!h->gendisk[drv_index]) {
2226                 h->gendisk[drv_index] =
2227                         alloc_disk(1 << NWD_SHIFT);
2228                 if (!h->gendisk[drv_index]) {
2229                         dev_err(&h->pdev->dev,
2230                                 "could not allocate a new disk %d\n",
2231                                 drv_index);
2232                         goto err_free_drive_info;
2233                 }
2234         }
2235         memcpy(h->drv[drv_index]->LunID, lunid,
2236                 sizeof(h->drv[drv_index]->LunID));
2237         if (cciss_create_ld_sysfs_entry(h, drv_index))
2238                 goto err_free_disk;
2239         /* Don't need to mark this busy because nobody */
2240         /* else knows about this disk yet to contend */
2241         /* for access to it. */
2242         h->drv[drv_index]->busy_configuring = 0;
2243         wmb();
2244         return drv_index;
2245
2246 err_free_disk:
2247         cciss_free_gendisk(h, drv_index);
2248 err_free_drive_info:
2249         cciss_free_drive_info(h, drv_index);
2250         return -1;
2251 }
2252
2253 /* This is for the special case of a controller which
2254  * has no logical drives.  In this case, we still need
2255  * to register a disk so the controller can be accessed
2256  * by the Array Config Utility.
2257  */
2258 static void cciss_add_controller_node(ctlr_info_t *h)
2259 {
2260         struct gendisk *disk;
2261         int drv_index;
2262
2263         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2264                 return;
2265
2266         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2267         if (drv_index == -1)
2268                 goto error;
2269         h->drv[drv_index]->block_size = 512;
2270         h->drv[drv_index]->nr_blocks = 0;
2271         h->drv[drv_index]->heads = 0;
2272         h->drv[drv_index]->sectors = 0;
2273         h->drv[drv_index]->cylinders = 0;
2274         h->drv[drv_index]->raid_level = -1;
2275         memset(h->drv[drv_index]->serial_no, 0, 16);
2276         disk = h->gendisk[drv_index];
2277         if (cciss_add_disk(h, disk, drv_index) == 0)
2278                 return;
2279         cciss_free_gendisk(h, drv_index);
2280         cciss_free_drive_info(h, drv_index);
2281 error:
2282         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2283         return;
2284 }
2285
2286 /* This function will add and remove logical drives from the Logical
2287  * drive array of the controller and maintain persistency of ordering
2288  * so that mount points are preserved until the next reboot.  This allows
2289  * for the removal of logical drives in the middle of the drive array
2290  * without a re-ordering of those drives.
2291  * INPUT
2292  * h            = The controller to perform the operations on
2293  */
2294 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2295         int via_ioctl)
2296 {
2297         int num_luns;
2298         ReportLunData_struct *ld_buff = NULL;
2299         int return_code;
2300         int listlength = 0;
2301         int i;
2302         int drv_found;
2303         int drv_index = 0;
2304         unsigned char lunid[8] = CTLR_LUNID;
2305         unsigned long flags;
2306
2307         if (!capable(CAP_SYS_RAWIO))
2308                 return -EPERM;
2309
2310         /* Set busy_configuring flag for this operation */
2311         spin_lock_irqsave(&h->lock, flags);
2312         if (h->busy_configuring) {
2313                 spin_unlock_irqrestore(&h->lock, flags);
2314                 return -EBUSY;
2315         }
2316         h->busy_configuring = 1;
2317         spin_unlock_irqrestore(&h->lock, flags);
2318
2319         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2320         if (ld_buff == NULL)
2321                 goto mem_msg;
2322
2323         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2324                                       sizeof(ReportLunData_struct),
2325                                       0, CTLR_LUNID, TYPE_CMD);
2326
2327         if (return_code == IO_OK)
2328                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2329         else {  /* reading number of logical volumes failed */
2330                 dev_warn(&h->pdev->dev,
2331                         "report logical volume command failed\n");
2332                 listlength = 0;
2333                 goto freeret;
2334         }
2335
2336         num_luns = listlength / 8;      /* 8 bytes per entry */
2337         if (num_luns > CISS_MAX_LUN) {
2338                 num_luns = CISS_MAX_LUN;
2339                 dev_warn(&h->pdev->dev, "more luns configured"
2340                        " on controller than can be handled by"
2341                        " this driver.\n");
2342         }
2343
2344         if (num_luns == 0)
2345                 cciss_add_controller_node(h);
2346
2347         /* Compare controller drive array to driver's drive array
2348          * to see if any drives are missing on the controller due
2349          * to action of Array Config Utility (user deletes drive)
2350          * and deregister logical drives which have disappeared.
2351          */
2352         for (i = 0; i <= h->highest_lun; i++) {
2353                 int j;
2354                 drv_found = 0;
2355
2356                 /* skip holes in the array from already deleted drives */
2357                 if (h->drv[i] == NULL)
2358                         continue;
2359
2360                 for (j = 0; j < num_luns; j++) {
2361                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2362                         if (memcmp(h->drv[i]->LunID, lunid,
2363                                 sizeof(lunid)) == 0) {
2364                                 drv_found = 1;
2365                                 break;
2366                         }
2367                 }
2368                 if (!drv_found) {
2369                         /* Deregister it from the OS, it's gone. */
2370                         spin_lock_irqsave(&h->lock, flags);
2371                         h->drv[i]->busy_configuring = 1;
2372                         spin_unlock_irqrestore(&h->lock, flags);
2373                         return_code = deregister_disk(h, i, 1, via_ioctl);
2374                         if (h->drv[i] != NULL)
2375                                 h->drv[i]->busy_configuring = 0;
2376                 }
2377         }
2378
2379         /* Compare controller drive array to driver's drive array.
2380          * Check for updates in the drive information and any new drives
2381          * on the controller due to ACU adding logical drives, or changing
2382          * a logical drive's size, etc.  Reregister any new/changed drives
2383          */
2384         for (i = 0; i < num_luns; i++) {
2385                 int j;
2386
2387                 drv_found = 0;
2388
2389                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2390                 /* Find if the LUN is already in the drive array
2391                  * of the driver.  If so then update its info
2392                  * if not in use.  If it does not exist then find
2393                  * the first free index and add it.
2394                  */
2395                 for (j = 0; j <= h->highest_lun; j++) {
2396                         if (h->drv[j] != NULL &&
2397                                 memcmp(h->drv[j]->LunID, lunid,
2398                                         sizeof(h->drv[j]->LunID)) == 0) {
2399                                 drv_index = j;
2400                                 drv_found = 1;
2401                                 break;
2402                         }
2403                 }
2404
2405                 /* check if the drive was found already in the array */
2406                 if (!drv_found) {
2407                         drv_index = cciss_add_gendisk(h, lunid, 0);
2408                         if (drv_index == -1)
2409                                 goto freeret;
2410                 }
2411                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2412         }               /* end for */
2413
2414 freeret:
2415         kfree(ld_buff);
2416         h->busy_configuring = 0;
2417         /* We return -1 here to tell the ACU that we have registered/updated
2418          * all of the drives that we can and to keep it from calling us
2419          * additional times.
2420          */
2421         return -1;
2422 mem_msg:
2423         dev_err(&h->pdev->dev, "out of memory\n");
2424         h->busy_configuring = 0;
2425         goto freeret;
2426 }
2427
2428 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2429 {
2430         /* zero out the disk size info */
2431         drive_info->nr_blocks = 0;
2432         drive_info->block_size = 0;
2433         drive_info->heads = 0;
2434         drive_info->sectors = 0;
2435         drive_info->cylinders = 0;
2436         drive_info->raid_level = -1;
2437         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2438         memset(drive_info->model, 0, sizeof(drive_info->model));
2439         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2440         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2441         /*
2442          * don't clear the LUNID though, we need to remember which
2443          * one this one is.
2444          */
2445 }
2446
2447 /* This function will deregister the disk and it's queue from the
2448  * kernel.  It must be called with the controller lock held and the
2449  * drv structures busy_configuring flag set.  It's parameters are:
2450  *
2451  * disk = This is the disk to be deregistered
2452  * drv  = This is the drive_info_struct associated with the disk to be
2453  *        deregistered.  It contains information about the disk used
2454  *        by the driver.
2455  * clear_all = This flag determines whether or not the disk information
2456  *             is going to be completely cleared out and the highest_lun
2457  *             reset.  Sometimes we want to clear out information about
2458  *             the disk in preparation for re-adding it.  In this case
2459  *             the highest_lun should be left unchanged and the LunID
2460  *             should not be cleared.
2461  * via_ioctl
2462  *    This indicates whether we've reached this path via ioctl.
2463  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2464  *    If this path is reached via ioctl(), then the max_usage_count will
2465  *    be 1, as the process calling ioctl() has got to have the device open.
2466  *    If we get here via sysfs, then the max usage count will be zero.
2467 */
2468 static int deregister_disk(ctlr_info_t *h, int drv_index,
2469                            int clear_all, int via_ioctl)
2470 {
2471         int i;
2472         struct gendisk *disk;
2473         drive_info_struct *drv;
2474         int recalculate_highest_lun;
2475
2476         if (!capable(CAP_SYS_RAWIO))
2477                 return -EPERM;
2478
2479         drv = h->drv[drv_index];
2480         disk = h->gendisk[drv_index];
2481
2482         /* make sure logical volume is NOT is use */
2483         if (clear_all || (h->gendisk[0] == disk)) {
2484                 if (drv->usage_count > via_ioctl)
2485                         return -EBUSY;
2486         } else if (drv->usage_count > 0)
2487                 return -EBUSY;
2488
2489         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2490
2491         /* invalidate the devices and deregister the disk.  If it is disk
2492          * zero do not deregister it but just zero out it's values.  This
2493          * allows us to delete disk zero but keep the controller registered.
2494          */
2495         if (h->gendisk[0] != disk) {
2496                 struct request_queue *q = disk->queue;
2497                 if (disk->flags & GENHD_FL_UP) {
2498                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2499                         del_gendisk(disk);
2500                 }
2501                 if (q)
2502                         blk_cleanup_queue(q);
2503                 /* If clear_all is set then we are deleting the logical
2504                  * drive, not just refreshing its info.  For drives
2505                  * other than disk 0 we will call put_disk.  We do not
2506                  * do this for disk 0 as we need it to be able to
2507                  * configure the controller.
2508                  */
2509                 if (clear_all){
2510                         /* This isn't pretty, but we need to find the
2511                          * disk in our array and NULL our the pointer.
2512                          * This is so that we will call alloc_disk if
2513                          * this index is used again later.
2514                          */
2515                         for (i=0; i < CISS_MAX_LUN; i++){
2516                                 if (h->gendisk[i] == disk) {
2517                                         h->gendisk[i] = NULL;
2518                                         break;
2519                                 }
2520                         }
2521                         put_disk(disk);
2522                 }
2523         } else {
2524                 set_capacity(disk, 0);
2525                 cciss_clear_drive_info(drv);
2526         }
2527
2528         --h->num_luns;
2529
2530         /* if it was the last disk, find the new hightest lun */
2531         if (clear_all && recalculate_highest_lun) {
2532                 int newhighest = -1;
2533                 for (i = 0; i <= h->highest_lun; i++) {
2534                         /* if the disk has size > 0, it is available */
2535                         if (h->drv[i] && h->drv[i]->heads)
2536                                 newhighest = i;
2537                 }
2538                 h->highest_lun = newhighest;
2539         }
2540         return 0;
2541 }
2542
2543 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2544                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2545                 int cmd_type)
2546 {
2547         u64bit buff_dma_handle;
2548         int status = IO_OK;
2549
2550         c->cmd_type = CMD_IOCTL_PEND;
2551         c->Header.ReplyQueue = 0;
2552         if (buff != NULL) {
2553                 c->Header.SGList = 1;
2554                 c->Header.SGTotal = 1;
2555         } else {
2556                 c->Header.SGList = 0;
2557                 c->Header.SGTotal = 0;
2558         }
2559         c->Header.Tag.lower = c->busaddr;
2560         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2561
2562         c->Request.Type.Type = cmd_type;
2563         if (cmd_type == TYPE_CMD) {
2564                 switch (cmd) {
2565                 case CISS_INQUIRY:
2566                         /* are we trying to read a vital product page */
2567                         if (page_code != 0) {
2568                                 c->Request.CDB[1] = 0x01;
2569                                 c->Request.CDB[2] = page_code;
2570                         }
2571                         c->Request.CDBLen = 6;
2572                         c->Request.Type.Attribute = ATTR_SIMPLE;
2573                         c->Request.Type.Direction = XFER_READ;
2574                         c->Request.Timeout = 0;
2575                         c->Request.CDB[0] = CISS_INQUIRY;
2576                         c->Request.CDB[4] = size & 0xFF;
2577                         break;
2578                 case CISS_REPORT_LOG:
2579                 case CISS_REPORT_PHYS:
2580                         /* Talking to controller so It's a physical command
2581                            mode = 00 target = 0.  Nothing to write.
2582                          */
2583                         c->Request.CDBLen = 12;
2584                         c->Request.Type.Attribute = ATTR_SIMPLE;
2585                         c->Request.Type.Direction = XFER_READ;
2586                         c->Request.Timeout = 0;
2587                         c->Request.CDB[0] = cmd;
2588                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2589                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2590                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2591                         c->Request.CDB[9] = size & 0xFF;
2592                         break;
2593
2594                 case CCISS_READ_CAPACITY:
2595                         c->Request.CDBLen = 10;
2596                         c->Request.Type.Attribute = ATTR_SIMPLE;
2597                         c->Request.Type.Direction = XFER_READ;
2598                         c->Request.Timeout = 0;
2599                         c->Request.CDB[0] = cmd;
2600                         break;
2601                 case CCISS_READ_CAPACITY_16:
2602                         c->Request.CDBLen = 16;
2603                         c->Request.Type.Attribute = ATTR_SIMPLE;
2604                         c->Request.Type.Direction = XFER_READ;
2605                         c->Request.Timeout = 0;
2606                         c->Request.CDB[0] = cmd;
2607                         c->Request.CDB[1] = 0x10;
2608                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2609                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2610                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2611                         c->Request.CDB[13] = size & 0xFF;
2612                         c->Request.Timeout = 0;
2613                         c->Request.CDB[0] = cmd;
2614                         break;
2615                 case CCISS_CACHE_FLUSH:
2616                         c->Request.CDBLen = 12;
2617                         c->Request.Type.Attribute = ATTR_SIMPLE;
2618                         c->Request.Type.Direction = XFER_WRITE;
2619                         c->Request.Timeout = 0;
2620                         c->Request.CDB[0] = BMIC_WRITE;
2621                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2622                         c->Request.CDB[7] = (size >> 8) & 0xFF;
2623                         c->Request.CDB[8] = size & 0xFF;
2624                         break;
2625                 case TEST_UNIT_READY:
2626                         c->Request.CDBLen = 6;
2627                         c->Request.Type.Attribute = ATTR_SIMPLE;
2628                         c->Request.Type.Direction = XFER_NONE;
2629                         c->Request.Timeout = 0;
2630                         break;
2631                 default:
2632                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2633                         return IO_ERROR;
2634                 }
2635         } else if (cmd_type == TYPE_MSG) {
2636                 switch (cmd) {
2637                 case CCISS_ABORT_MSG:
2638                         c->Request.CDBLen = 12;
2639                         c->Request.Type.Attribute = ATTR_SIMPLE;
2640                         c->Request.Type.Direction = XFER_WRITE;
2641                         c->Request.Timeout = 0;
2642                         c->Request.CDB[0] = cmd;        /* abort */
2643                         c->Request.CDB[1] = 0;  /* abort a command */
2644                         /* buff contains the tag of the command to abort */
2645                         memcpy(&c->Request.CDB[4], buff, 8);
2646                         break;
2647                 case CCISS_RESET_MSG:
2648                         c->Request.CDBLen = 16;
2649                         c->Request.Type.Attribute = ATTR_SIMPLE;
2650                         c->Request.Type.Direction = XFER_NONE;
2651                         c->Request.Timeout = 0;
2652                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2653                         c->Request.CDB[0] = cmd;        /* reset */
2654                         c->Request.CDB[1] = CCISS_RESET_TYPE_TARGET;
2655                         break;
2656                 case CCISS_NOOP_MSG:
2657                         c->Request.CDBLen = 1;
2658                         c->Request.Type.Attribute = ATTR_SIMPLE;
2659                         c->Request.Type.Direction = XFER_WRITE;
2660                         c->Request.Timeout = 0;
2661                         c->Request.CDB[0] = cmd;
2662                         break;
2663                 default:
2664                         dev_warn(&h->pdev->dev,
2665                                 "unknown message type %d\n", cmd);
2666                         return IO_ERROR;
2667                 }
2668         } else {
2669                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2670                 return IO_ERROR;
2671         }
2672         /* Fill in the scatter gather information */
2673         if (size > 0) {
2674                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2675                                                              buff, size,
2676                                                              PCI_DMA_BIDIRECTIONAL);
2677                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2678                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2679                 c->SG[0].Len = size;
2680                 c->SG[0].Ext = 0;       /* we are not chaining */
2681         }
2682         return status;
2683 }
2684
2685 static int cciss_send_reset(ctlr_info_t *h, unsigned char *scsi3addr,
2686                             u8 reset_type)
2687 {
2688         CommandList_struct *c;
2689         int return_status;
2690
2691         c = cmd_alloc(h);
2692         if (!c)
2693                 return -ENOMEM;
2694         return_status = fill_cmd(h, c, CCISS_RESET_MSG, NULL, 0, 0,
2695                 CTLR_LUNID, TYPE_MSG);
2696         c->Request.CDB[1] = reset_type; /* fill_cmd defaults to target reset */
2697         if (return_status != IO_OK) {
2698                 cmd_special_free(h, c);
2699                 return return_status;
2700         }
2701         c->waiting = NULL;
2702         enqueue_cmd_and_start_io(h, c);
2703         /* Don't wait for completion, the reset won't complete.  Don't free
2704          * the command either.  This is the last command we will send before
2705          * re-initializing everything, so it doesn't matter and won't leak.
2706          */
2707         return 0;
2708 }
2709
2710 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2711 {
2712         switch (c->err_info->ScsiStatus) {
2713         case SAM_STAT_GOOD:
2714                 return IO_OK;
2715         case SAM_STAT_CHECK_CONDITION:
2716                 switch (0xf & c->err_info->SenseInfo[2]) {
2717                 case 0: return IO_OK; /* no sense */
2718                 case 1: return IO_OK; /* recovered error */
2719                 default:
2720                         if (check_for_unit_attention(h, c))
2721                                 return IO_NEEDS_RETRY;
2722                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2723                                 "check condition, sense key = 0x%02x\n",
2724                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2725                 }
2726                 break;
2727         default:
2728                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2729                         "scsi status = 0x%02x\n",
2730                         c->Request.CDB[0], c->err_info->ScsiStatus);
2731                 break;
2732         }
2733         return IO_ERROR;
2734 }
2735
2736 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2737 {
2738         int return_status = IO_OK;
2739
2740         if (c->err_info->CommandStatus == CMD_SUCCESS)
2741                 return IO_OK;
2742
2743         switch (c->err_info->CommandStatus) {
2744         case CMD_TARGET_STATUS:
2745                 return_status = check_target_status(h, c);
2746                 break;
2747         case CMD_DATA_UNDERRUN:
2748         case CMD_DATA_OVERRUN:
2749                 /* expected for inquiry and report lun commands */
2750                 break;
2751         case CMD_INVALID:
2752                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2753                        "reported invalid\n", c->Request.CDB[0]);
2754                 return_status = IO_ERROR;
2755                 break;
2756         case CMD_PROTOCOL_ERR:
2757                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2758                        "protocol error\n", c->Request.CDB[0]);
2759                 return_status = IO_ERROR;
2760                 break;
2761         case CMD_HARDWARE_ERR:
2762                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2763                        " hardware error\n", c->Request.CDB[0]);
2764                 return_status = IO_ERROR;
2765                 break;
2766         case CMD_CONNECTION_LOST:
2767                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2768                        "connection lost\n", c->Request.CDB[0]);
2769                 return_status = IO_ERROR;
2770                 break;
2771         case CMD_ABORTED:
2772                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2773                        "aborted\n", c->Request.CDB[0]);
2774                 return_status = IO_ERROR;
2775                 break;
2776         case CMD_ABORT_FAILED:
2777                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2778                        "abort failed\n", c->Request.CDB[0]);
2779                 return_status = IO_ERROR;
2780                 break;
2781         case CMD_UNSOLICITED_ABORT:
2782                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2783                         c->Request.CDB[0]);
2784                 return_status = IO_NEEDS_RETRY;
2785                 break;
2786         case CMD_UNABORTABLE:
2787                 dev_warn(&h->pdev->dev, "cmd unabortable\n");
2788                 return_status = IO_ERROR;
2789                 break;
2790         default:
2791                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2792                        "unknown status %x\n", c->Request.CDB[0],
2793                        c->err_info->CommandStatus);
2794                 return_status = IO_ERROR;
2795         }
2796         return return_status;
2797 }
2798
2799 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2800         int attempt_retry)
2801 {
2802         DECLARE_COMPLETION_ONSTACK(wait);
2803         u64bit buff_dma_handle;
2804         int return_status = IO_OK;
2805
2806 resend_cmd2:
2807         c->waiting = &wait;
2808         enqueue_cmd_and_start_io(h, c);
2809
2810         wait_for_completion(&wait);
2811
2812         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2813                 goto command_done;
2814
2815         return_status = process_sendcmd_error(h, c);
2816
2817         if (return_status == IO_NEEDS_RETRY &&
2818                 c->retry_count < MAX_CMD_RETRIES) {
2819                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2820                         c->Request.CDB[0]);
2821                 c->retry_count++;
2822                 /* erase the old error information */
2823                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2824                 return_status = IO_OK;
2825                 reinit_completion(&wait);
2826                 goto resend_cmd2;
2827         }
2828
2829 command_done:
2830         /* unlock the buffers from DMA */
2831         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2832         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2833         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2834                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2835         return return_status;
2836 }
2837
2838 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2839                            __u8 page_code, unsigned char scsi3addr[],
2840                         int cmd_type)
2841 {
2842         CommandList_struct *c;
2843         int return_status;
2844
2845         c = cmd_special_alloc(h);
2846         if (!c)
2847                 return -ENOMEM;
2848         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2849                 scsi3addr, cmd_type);
2850         if (return_status == IO_OK)
2851                 return_status = sendcmd_withirq_core(h, c, 1);
2852
2853         cmd_special_free(h, c);
2854         return return_status;
2855 }
2856
2857 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2858                                    sector_t total_size,
2859                                    unsigned int block_size,
2860                                    InquiryData_struct *inq_buff,
2861                                    drive_info_struct *drv)
2862 {
2863         int return_code;
2864         unsigned long t;
2865         unsigned char scsi3addr[8];
2866
2867         memset(inq_buff, 0, sizeof(InquiryData_struct));
2868         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2869         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2870                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2871         if (return_code == IO_OK) {
2872                 if (inq_buff->data_byte[8] == 0xFF) {
2873                         dev_warn(&h->pdev->dev,
2874                                "reading geometry failed, volume "
2875                                "does not support reading geometry\n");
2876                         drv->heads = 255;
2877                         drv->sectors = 32;      /* Sectors per track */
2878                         drv->cylinders = total_size + 1;
2879                         drv->raid_level = RAID_UNKNOWN;
2880                 } else {
2881                         drv->heads = inq_buff->data_byte[6];
2882                         drv->sectors = inq_buff->data_byte[7];
2883                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2884                         drv->cylinders += inq_buff->data_byte[5];
2885                         drv->raid_level = inq_buff->data_byte[8];
2886                 }
2887                 drv->block_size = block_size;
2888                 drv->nr_blocks = total_size + 1;
2889                 t = drv->heads * drv->sectors;
2890                 if (t > 1) {
2891                         sector_t real_size = total_size + 1;
2892                         unsigned long rem = sector_div(real_size, t);
2893                         if (rem)
2894                                 real_size++;
2895                         drv->cylinders = real_size;
2896                 }
2897         } else {                /* Get geometry failed */
2898                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2899         }
2900 }
2901
2902 static void
2903 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2904                     unsigned int *block_size)
2905 {
2906         ReadCapdata_struct *buf;
2907         int return_code;
2908         unsigned char scsi3addr[8];
2909
2910         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2911         if (!buf) {
2912                 dev_warn(&h->pdev->dev, "out of memory\n");
2913                 return;
2914         }
2915
2916         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2917         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2918                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2919         if (return_code == IO_OK) {
2920                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2921                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2922         } else {                /* read capacity command failed */
2923                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2924                 *total_size = 0;
2925                 *block_size = BLOCK_SIZE;
2926         }
2927         kfree(buf);
2928 }
2929
2930 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2931         sector_t *total_size, unsigned int *block_size)
2932 {
2933         ReadCapdata_struct_16 *buf;
2934         int return_code;
2935         unsigned char scsi3addr[8];
2936
2937         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2938         if (!buf) {
2939                 dev_warn(&h->pdev->dev, "out of memory\n");
2940                 return;
2941         }
2942
2943         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2944         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2945                 buf, sizeof(ReadCapdata_struct_16),
2946                         0, scsi3addr, TYPE_CMD);
2947         if (return_code == IO_OK) {
2948                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2949                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2950         } else {                /* read capacity command failed */
2951                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2952                 *total_size = 0;
2953                 *block_size = BLOCK_SIZE;
2954         }
2955         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2956                (unsigned long long)*total_size+1, *block_size);
2957         kfree(buf);
2958 }
2959
2960 static int cciss_revalidate(struct gendisk *disk)
2961 {
2962         ctlr_info_t *h = get_host(disk);
2963         drive_info_struct *drv = get_drv(disk);
2964         int logvol;
2965         int FOUND = 0;
2966         unsigned int block_size;
2967         sector_t total_size;
2968         InquiryData_struct *inq_buff = NULL;
2969
2970         for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2971                 if (!h->drv[logvol])
2972                         continue;
2973                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2974                         sizeof(drv->LunID)) == 0) {
2975                         FOUND = 1;
2976                         break;
2977                 }
2978         }
2979
2980         if (!FOUND)
2981                 return 1;
2982
2983         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2984         if (inq_buff == NULL) {
2985                 dev_warn(&h->pdev->dev, "out of memory\n");
2986                 return 1;
2987         }
2988         if (h->cciss_read == CCISS_READ_10) {
2989                 cciss_read_capacity(h, logvol,
2990                                         &total_size, &block_size);
2991         } else {
2992                 cciss_read_capacity_16(h, logvol,
2993                                         &total_size, &block_size);
2994         }
2995         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2996                                inq_buff, drv);
2997
2998         blk_queue_logical_block_size(drv->queue, drv->block_size);
2999         set_capacity(disk, drv->nr_blocks);
3000
3001         kfree(inq_buff);
3002         return 0;
3003 }
3004
3005 /*
3006  * Map (physical) PCI mem into (virtual) kernel space
3007  */
3008 static void __iomem *remap_pci_mem(ulong base, ulong size)
3009 {
3010         ulong page_base = ((ulong) base) & PAGE_MASK;
3011         ulong page_offs = ((ulong) base) - page_base;
3012         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
3013
3014         return page_remapped ? (page_remapped + page_offs) : NULL;
3015 }
3016
3017 /*
3018  * Takes jobs of the Q and sends them to the hardware, then puts it on
3019  * the Q to wait for completion.
3020  */
3021 static void start_io(ctlr_info_t *h)
3022 {
3023         CommandList_struct *c;
3024
3025         while (!list_empty(&h->reqQ)) {
3026                 c = list_entry(h->reqQ.next, CommandList_struct, list);
3027                 /* can't do anything if fifo is full */
3028                 if ((h->access.fifo_full(h))) {
3029                         dev_warn(&h->pdev->dev, "fifo full\n");
3030                         break;
3031                 }
3032
3033                 /* Get the first entry from the Request Q */
3034                 removeQ(c);
3035                 h->Qdepth--;
3036
3037                 /* Tell the controller execute command */
3038                 h->access.submit_command(h, c);
3039
3040                 /* Put job onto the completed Q */
3041                 addQ(&h->cmpQ, c);
3042         }
3043 }
3044
3045 /* Assumes that h->lock is held. */
3046 /* Zeros out the error record and then resends the command back */
3047 /* to the controller */
3048 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
3049 {
3050         /* erase the old error information */
3051         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
3052
3053         /* add it to software queue and then send it to the controller */
3054         addQ(&h->reqQ, c);
3055         h->Qdepth++;
3056         if (h->Qdepth > h->maxQsinceinit)
3057                 h->maxQsinceinit = h->Qdepth;
3058
3059         start_io(h);
3060 }
3061
3062 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
3063         unsigned int msg_byte, unsigned int host_byte,
3064         unsigned int driver_byte)
3065 {
3066         /* inverse of macros in scsi.h */
3067         return (scsi_status_byte & 0xff) |
3068                 ((msg_byte & 0xff) << 8) |
3069                 ((host_byte & 0xff) << 16) |
3070                 ((driver_byte & 0xff) << 24);
3071 }
3072
3073 static inline int evaluate_target_status(ctlr_info_t *h,
3074                         CommandList_struct *cmd, int *retry_cmd)
3075 {
3076         unsigned char sense_key;
3077         unsigned char status_byte, msg_byte, host_byte, driver_byte;
3078         int error_value;
3079
3080         *retry_cmd = 0;
3081         /* If we get in here, it means we got "target status", that is, scsi status */
3082         status_byte = cmd->err_info->ScsiStatus;
3083         driver_byte = DRIVER_OK;
3084         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3085
3086         if (blk_rq_is_passthrough(cmd->rq))
3087                 host_byte = DID_PASSTHROUGH;
3088         else
3089                 host_byte = DID_OK;
3090
3091         error_value = make_status_bytes(status_byte, msg_byte,
3092                 host_byte, driver_byte);
3093
3094         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3095                 if (!blk_rq_is_passthrough(cmd->rq))
3096                         dev_warn(&h->pdev->dev, "cmd %p "
3097                                "has SCSI Status 0x%x\n",
3098                                cmd, cmd->err_info->ScsiStatus);
3099                 return error_value;
3100         }
3101
3102         /* check the sense key */
3103         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3104         /* no status or recovered error */
3105         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3106             !blk_rq_is_passthrough(cmd->rq))
3107                 error_value = 0;
3108
3109         if (check_for_unit_attention(h, cmd)) {
3110                 *retry_cmd = !blk_rq_is_passthrough(cmd->rq);
3111                 return 0;
3112         }
3113
3114         /* Not SG_IO or similar? */
3115         if (!blk_rq_is_passthrough(cmd->rq)) {
3116                 if (error_value != 0)
3117                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3118                                " sense key = 0x%x\n", cmd, sense_key);
3119                 return error_value;
3120         }
3121
3122         scsi_req(cmd->rq)->sense_len = cmd->err_info->SenseLen;
3123         return error_value;
3124 }
3125
3126 /* checks the status of the job and calls complete buffers to mark all
3127  * buffers for the completed job. Note that this function does not need
3128  * to hold the hba/queue lock.
3129  */
3130 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3131                                     int timeout)
3132 {
3133         int retry_cmd = 0;
3134         struct request *rq = cmd->rq;
3135
3136         rq->errors = 0;
3137
3138         if (timeout)
3139                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3140
3141         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3142                 goto after_error_processing;
3143
3144         switch (cmd->err_info->CommandStatus) {
3145         case CMD_TARGET_STATUS:
3146                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3147                 break;
3148         case CMD_DATA_UNDERRUN:
3149                 if (!blk_rq_is_passthrough(cmd->rq)) {
3150                         dev_warn(&h->pdev->dev, "cmd %p has"
3151                                " completed with data underrun "
3152                                "reported\n", cmd);
3153                 }
3154                 break;
3155         case CMD_DATA_OVERRUN:
3156                 if (!blk_rq_is_passthrough(cmd->rq))
3157                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3158                                " completed with data overrun "
3159                                "reported\n", cmd);
3160                 break;
3161         case CMD_INVALID:
3162                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3163                        "reported invalid\n", cmd);
3164                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3165                         cmd->err_info->CommandStatus, DRIVER_OK,
3166                         blk_rq_is_passthrough(cmd->rq) ?
3167                                 DID_PASSTHROUGH : DID_ERROR);
3168                 break;
3169         case CMD_PROTOCOL_ERR:
3170                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3171                        "protocol error\n", cmd);
3172                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3173                         cmd->err_info->CommandStatus, DRIVER_OK,
3174                         blk_rq_is_passthrough(cmd->rq) ?
3175                                 DID_PASSTHROUGH : DID_ERROR);
3176                 break;
3177         case CMD_HARDWARE_ERR:
3178                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3179                        " hardware error\n", cmd);
3180                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3181                         cmd->err_info->CommandStatus, DRIVER_OK,
3182                         blk_rq_is_passthrough(cmd->rq) ?
3183                                 DID_PASSTHROUGH : DID_ERROR);
3184                 break;
3185         case CMD_CONNECTION_LOST:
3186                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3187                        "connection lost\n", cmd);
3188                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3189                         cmd->err_info->CommandStatus, DRIVER_OK,
3190                         blk_rq_is_passthrough(cmd->rq) ?
3191                                 DID_PASSTHROUGH : DID_ERROR);
3192                 break;
3193         case CMD_ABORTED:
3194                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3195                        "aborted\n", cmd);
3196                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3197                         cmd->err_info->CommandStatus, DRIVER_OK,
3198                         blk_rq_is_passthrough(cmd->rq) ?
3199                                 DID_PASSTHROUGH : DID_ABORT);
3200                 break;
3201         case CMD_ABORT_FAILED:
3202                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3203                        "abort failed\n", cmd);
3204                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3205                         cmd->err_info->CommandStatus, DRIVER_OK,
3206                         blk_rq_is_passthrough(cmd->rq) ?
3207                                 DID_PASSTHROUGH : DID_ERROR);
3208                 break;
3209         case CMD_UNSOLICITED_ABORT:
3210                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3211                        "abort %p\n", h->ctlr, cmd);
3212                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3213                         retry_cmd = 1;
3214                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3215                         cmd->retry_count++;
3216                 } else
3217                         dev_warn(&h->pdev->dev,
3218                                 "%p retried too many times\n", cmd);
3219                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3220                         cmd->err_info->CommandStatus, DRIVER_OK,
3221                         blk_rq_is_passthrough(cmd->rq) ?
3222                                 DID_PASSTHROUGH : DID_ABORT);
3223                 break;
3224         case CMD_TIMEOUT:
3225                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3226                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3227                         cmd->err_info->CommandStatus, DRIVER_OK,
3228                         blk_rq_is_passthrough(cmd->rq) ?
3229                                 DID_PASSTHROUGH : DID_ERROR);
3230                 break;
3231         case CMD_UNABORTABLE:
3232                 dev_warn(&h->pdev->dev, "cmd %p unabortable\n", cmd);
3233                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3234                         cmd->err_info->CommandStatus, DRIVER_OK,
3235                         blk_rq_is_passthrough(cmd->rq) ?
3236                                 DID_PASSTHROUGH : DID_ERROR);
3237                 break;
3238         default:
3239                 dev_warn(&h->pdev->dev, "cmd %p returned "
3240                        "unknown status %x\n", cmd,
3241                        cmd->err_info->CommandStatus);
3242                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3243                         cmd->err_info->CommandStatus, DRIVER_OK,
3244                         blk_rq_is_passthrough(cmd->rq) ?
3245                                 DID_PASSTHROUGH : DID_ERROR);
3246         }
3247
3248 after_error_processing:
3249
3250         /* We need to return this command */
3251         if (retry_cmd) {
3252                 resend_cciss_cmd(h, cmd);
3253                 return;
3254         }
3255         cmd->rq->completion_data = cmd;
3256         blk_complete_request(cmd->rq);
3257 }
3258
3259 static inline u32 cciss_tag_contains_index(u32 tag)
3260 {
3261 #define DIRECT_LOOKUP_BIT 0x10
3262         return tag & DIRECT_LOOKUP_BIT;
3263 }
3264
3265 static inline u32 cciss_tag_to_index(u32 tag)
3266 {
3267 #define DIRECT_LOOKUP_SHIFT 5
3268         return tag >> DIRECT_LOOKUP_SHIFT;
3269 }
3270
3271 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag)
3272 {
3273 #define CCISS_PERF_ERROR_BITS ((1 << DIRECT_LOOKUP_SHIFT) - 1)
3274 #define CCISS_SIMPLE_ERROR_BITS 0x03
3275         if (likely(h->transMethod & CFGTBL_Trans_Performant))
3276                 return tag & ~CCISS_PERF_ERROR_BITS;
3277         return tag & ~CCISS_SIMPLE_ERROR_BITS;
3278 }
3279
3280 static inline void cciss_mark_tag_indexed(u32 *tag)
3281 {
3282         *tag |= DIRECT_LOOKUP_BIT;
3283 }
3284
3285 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3286 {
3287         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3288 }
3289
3290 /*
3291  * Get a request and submit it to the controller.
3292  */
3293 static void do_cciss_request(struct request_queue *q)
3294 {
3295         ctlr_info_t *h = q->queuedata;
3296         CommandList_struct *c;
3297         sector_t start_blk;
3298         int seg;
3299         struct request *creq;
3300         u64bit temp64;
3301         struct scatterlist *tmp_sg;
3302         SGDescriptor_struct *curr_sg;
3303         drive_info_struct *drv;
3304         int i, dir;
3305         int sg_index = 0;
3306         int chained = 0;
3307
3308       queue:
3309         creq = blk_peek_request(q);
3310         if (!creq)
3311                 goto startio;
3312
3313         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3314
3315         c = cmd_alloc(h);
3316         if (!c)
3317                 goto full;
3318
3319         blk_start_request(creq);
3320
3321         tmp_sg = h->scatter_list[c->cmdindex];
3322         spin_unlock_irq(q->queue_lock);
3323
3324         c->cmd_type = CMD_RWREQ;
3325         c->rq = creq;
3326
3327         /* fill in the request */
3328         drv = creq->rq_disk->private_data;
3329         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3330         /* got command from pool, so use the command block index instead */
3331         /* for direct lookups. */
3332         /* The first 2 bits are reserved for controller error reporting. */
3333         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3334         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3335         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3336         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3337         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3338         c->Request.Type.Attribute = ATTR_SIMPLE;
3339         c->Request.Type.Direction =
3340             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3341         c->Request.Timeout = 0; /* Don't time out */
3342         c->Request.CDB[0] =
3343             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3344         start_blk = blk_rq_pos(creq);
3345         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3346                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3347         sg_init_table(tmp_sg, h->maxsgentries);
3348         seg = blk_rq_map_sg(q, creq, tmp_sg);
3349
3350         /* get the DMA records for the setup */
3351         if (c->Request.Type.Direction == XFER_READ)
3352                 dir = PCI_DMA_FROMDEVICE;
3353         else
3354                 dir = PCI_DMA_TODEVICE;
3355
3356         curr_sg = c->SG;
3357         sg_index = 0;
3358         chained = 0;
3359
3360         for (i = 0; i < seg; i++) {
3361                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3362                         !chained && ((seg - i) > 1)) {
3363                         /* Point to next chain block. */
3364                         curr_sg = h->cmd_sg_list[c->cmdindex];
3365                         sg_index = 0;
3366                         chained = 1;
3367                 }
3368                 curr_sg[sg_index].Len = tmp_sg[i].length;
3369                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3370                                                 tmp_sg[i].offset,
3371                                                 tmp_sg[i].length, dir);
3372                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3373                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3374                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3375                 ++sg_index;
3376         }
3377         if (chained)
3378                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3379                         (seg - (h->max_cmd_sgentries - 1)) *
3380                                 sizeof(SGDescriptor_struct));
3381
3382         /* track how many SG entries we are using */
3383         if (seg > h->maxSG)
3384                 h->maxSG = seg;
3385
3386         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3387                         "chained[%d]\n",
3388                         blk_rq_sectors(creq), seg, chained);
3389
3390         c->Header.SGTotal = seg + chained;
3391         if (seg <= h->max_cmd_sgentries)
3392                 c->Header.SGList = c->Header.SGTotal;
3393         else
3394                 c->Header.SGList = h->max_cmd_sgentries;
3395         set_performant_mode(h, c);
3396
3397         switch (req_op(creq)) {
3398         case REQ_OP_READ:
3399         case REQ_OP_WRITE:
3400                 if(h->cciss_read == CCISS_READ_10) {
3401                         c->Request.CDB[1] = 0;
3402                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3403                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3404                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3405                         c->Request.CDB[5] = start_blk & 0xff;
3406                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3407                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3408                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3409                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3410                 } else {
3411                         u32 upper32 = upper_32_bits(start_blk);
3412
3413                         c->Request.CDBLen = 16;
3414                         c->Request.CDB[1]= 0;
3415                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3416                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3417                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3418                         c->Request.CDB[5]= upper32 & 0xff;
3419                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3420                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3421                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3422                         c->Request.CDB[9]= start_blk & 0xff;
3423                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3424                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3425                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3426                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3427                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3428                 }
3429                 break;
3430         case REQ_OP_SCSI_IN:
3431         case REQ_OP_SCSI_OUT:
3432                 c->Request.CDBLen = scsi_req(creq)->cmd_len;
3433                 memcpy(c->Request.CDB, scsi_req(creq)->cmd, BLK_MAX_CDB);
3434                 scsi_req(creq)->sense = c->err_info->SenseInfo;
3435                 break;
3436         default:
3437                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3438                         creq->cmd_flags);
3439                 BUG();
3440         }
3441
3442         spin_lock_irq(q->queue_lock);
3443
3444         addQ(&h->reqQ, c);
3445         h->Qdepth++;
3446         if (h->Qdepth > h->maxQsinceinit)
3447                 h->maxQsinceinit = h->Qdepth;
3448
3449         goto queue;
3450 full:
3451         blk_stop_queue(q);
3452 startio:
3453         /* We will already have the driver lock here so not need
3454          * to lock it.
3455          */
3456         start_io(h);
3457 }
3458
3459 static inline unsigned long get_next_completion(ctlr_info_t *h)
3460 {
3461         return h->access.command_completed(h);
3462 }
3463
3464 static inline int interrupt_pending(ctlr_info_t *h)
3465 {
3466         return h->access.intr_pending(h);
3467 }
3468
3469 static inline long interrupt_not_for_us(ctlr_info_t *h)
3470 {
3471         return ((h->access.intr_pending(h) == 0) ||
3472                 (h->interrupts_enabled == 0));
3473 }
3474
3475 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3476                         u32 raw_tag)
3477 {
3478         if (unlikely(tag_index >= h->nr_cmds)) {
3479                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3480                 return 1;
3481         }
3482         return 0;
3483 }
3484
3485 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3486                                 u32 raw_tag)
3487 {
3488         removeQ(c);
3489         if (likely(c->cmd_type == CMD_RWREQ))
3490                 complete_command(h, c, 0);
3491         else if (c->cmd_type == CMD_IOCTL_PEND)
3492                 complete(c->waiting);
3493 #ifdef CONFIG_CISS_SCSI_TAPE
3494         else if (c->cmd_type == CMD_SCSI)
3495                 complete_scsi_command(c, 0, raw_tag);
3496 #endif
3497 }
3498
3499 static inline u32 next_command(ctlr_info_t *h)
3500 {
3501         u32 a;
3502
3503         if (unlikely(!(h->transMethod & CFGTBL_Trans_Performant)))
3504                 return h->access.command_completed(h);
3505
3506         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3507                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3508                 (h->reply_pool_head)++;
3509                 h->commands_outstanding--;
3510         } else {
3511                 a = FIFO_EMPTY;
3512         }
3513         /* Check for wraparound */
3514         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3515                 h->reply_pool_head = h->reply_pool;
3516                 h->reply_pool_wraparound ^= 1;
3517         }
3518         return a;
3519 }
3520
3521 /* process completion of an indexed ("direct lookup") command */
3522 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3523 {
3524         u32 tag_index;
3525         CommandList_struct *c;
3526
3527         tag_index = cciss_tag_to_index(raw_tag);
3528         if (bad_tag(h, tag_index, raw_tag))
3529                 return next_command(h);
3530         c = h->cmd_pool + tag_index;
3531         finish_cmd(h, c, raw_tag);
3532         return next_command(h);
3533 }
3534
3535 /* process completion of a non-indexed command */
3536 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3537 {
3538         CommandList_struct *c = NULL;
3539         __u32 busaddr_masked, tag_masked;
3540
3541         tag_masked = cciss_tag_discard_error_bits(h, raw_tag);
3542         list_for_each_entry(c, &h->cmpQ, list) {
3543                 busaddr_masked = cciss_tag_discard_error_bits(h, c->busaddr);
3544                 if (busaddr_masked == tag_masked) {
3545                         finish_cmd(h, c, raw_tag);
3546                         return next_command(h);
3547                 }
3548         }
3549         bad_tag(h, h->nr_cmds + 1, raw_tag);
3550         return next_command(h);
3551 }
3552
3553 /* Some controllers, like p400, will give us one interrupt
3554  * after a soft reset, even if we turned interrupts off.
3555  * Only need to check for this in the cciss_xxx_discard_completions
3556  * functions.
3557  */
3558 static int ignore_bogus_interrupt(ctlr_info_t *h)
3559 {
3560         if (likely(!reset_devices))
3561                 return 0;
3562
3563         if (likely(h->interrupts_enabled))
3564                 return 0;
3565
3566         dev_info(&h->pdev->dev, "Received interrupt while interrupts disabled "
3567                 "(known firmware bug.)  Ignoring.\n");
3568
3569         return 1;
3570 }
3571
3572 static irqreturn_t cciss_intx_discard_completions(int irq, void *dev_id)
3573 {
3574         ctlr_info_t *h = dev_id;
3575         unsigned long flags;
3576         u32 raw_tag;
3577
3578         if (ignore_bogus_interrupt(h))
3579                 return IRQ_NONE;
3580
3581         if (interrupt_not_for_us(h))
3582                 return IRQ_NONE;
3583         spin_lock_irqsave(&h->lock, flags);
3584         while (interrupt_pending(h)) {
3585                 raw_tag = get_next_completion(h);
3586                 while (raw_tag != FIFO_EMPTY)
3587                         raw_tag = next_command(h);
3588         }
3589         spin_unlock_irqrestore(&h->lock, flags);
3590         return IRQ_HANDLED;
3591 }
3592
3593 static irqreturn_t cciss_msix_discard_completions(int irq, void *dev_id)
3594 {
3595         ctlr_info_t *h = dev_id;
3596         unsigned long flags;
3597         u32 raw_tag;
3598
3599         if (ignore_bogus_interrupt(h))
3600                 return IRQ_NONE;
3601
3602         spin_lock_irqsave(&h->lock, flags);
3603         raw_tag = get_next_completion(h);
3604         while (raw_tag != FIFO_EMPTY)
3605                 raw_tag = next_command(h);
3606         spin_unlock_irqrestore(&h->lock, flags);
3607         return IRQ_HANDLED;
3608 }
3609
3610 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3611 {
3612         ctlr_info_t *h = dev_id;
3613         unsigned long flags;
3614         u32 raw_tag;
3615
3616         if (interrupt_not_for_us(h))
3617                 return IRQ_NONE;
3618         spin_lock_irqsave(&h->lock, flags);
3619         while (interrupt_pending(h)) {
3620                 raw_tag = get_next_completion(h);
3621                 while (raw_tag != FIFO_EMPTY) {
3622                         if (cciss_tag_contains_index(raw_tag))
3623                                 raw_tag = process_indexed_cmd(h, raw_tag);
3624                         else
3625                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3626                 }
3627         }
3628         spin_unlock_irqrestore(&h->lock, flags);
3629         return IRQ_HANDLED;
3630 }
3631
3632 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3633  * check the interrupt pending register because it is not set.
3634  */
3635 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3636 {
3637         ctlr_info_t *h = dev_id;
3638         unsigned long flags;
3639         u32 raw_tag;
3640
3641         spin_lock_irqsave(&h->lock, flags);
3642         raw_tag = get_next_completion(h);
3643         while (raw_tag != FIFO_EMPTY) {
3644                 if (cciss_tag_contains_index(raw_tag))
3645                         raw_tag = process_indexed_cmd(h, raw_tag);
3646                 else
3647                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3648         }
3649         spin_unlock_irqrestore(&h->lock, flags);
3650         return IRQ_HANDLED;
3651 }
3652
3653 /**
3654  * add_to_scan_list() - add controller to rescan queue
3655  * @h:                Pointer to the controller.
3656  *
3657  * Adds the controller to the rescan queue if not already on the queue.
3658  *
3659  * returns 1 if added to the queue, 0 if skipped (could be on the
3660  * queue already, or the controller could be initializing or shutting
3661  * down).
3662  **/
3663 static int add_to_scan_list(struct ctlr_info *h)
3664 {
3665         struct ctlr_info *test_h;
3666         int found = 0;
3667         int ret = 0;
3668
3669         if (h->busy_initializing)
3670                 return 0;
3671
3672         if (!mutex_trylock(&h->busy_shutting_down))
3673                 return 0;
3674
3675         mutex_lock(&scan_mutex);
3676         list_for_each_entry(test_h, &scan_q, scan_list) {
3677                 if (test_h == h) {
3678                         found = 1;
3679                         break;
3680                 }
3681         }
3682         if (!found && !h->busy_scanning) {
3683                 reinit_completion(&h->scan_wait);
3684                 list_add_tail(&h->scan_list, &scan_q);
3685                 ret = 1;
3686         }
3687         mutex_unlock(&scan_mutex);
3688         mutex_unlock(&h->busy_shutting_down);
3689
3690         return ret;
3691 }
3692
3693 /**
3694  * remove_from_scan_list() - remove controller from rescan queue
3695  * @h:                     Pointer to the controller.
3696  *
3697  * Removes the controller from the rescan queue if present. Blocks if
3698  * the controller is currently conducting a rescan.  The controller
3699  * can be in one of three states:
3700  * 1. Doesn't need a scan
3701  * 2. On the scan list, but not scanning yet (we remove it)
3702  * 3. Busy scanning (and not on the list). In this case we want to wait for
3703  *    the scan to complete to make sure the scanning thread for this
3704  *    controller is completely idle.
3705  **/
3706 static void remove_from_scan_list(struct ctlr_info *h)
3707 {
3708         struct ctlr_info *test_h, *tmp_h;
3709
3710         mutex_lock(&scan_mutex);
3711         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3712                 if (test_h == h) { /* state 2. */
3713                         list_del(&h->scan_list);
3714                         complete_all(&h->scan_wait);
3715                         mutex_unlock(&scan_mutex);
3716                         return;
3717                 }
3718         }
3719         if (h->busy_scanning) { /* state 3. */
3720                 mutex_unlock(&scan_mutex);
3721                 wait_for_completion(&h->scan_wait);
3722         } else { /* state 1, nothing to do. */
3723                 mutex_unlock(&scan_mutex);
3724         }
3725 }
3726
3727 /**
3728  * scan_thread() - kernel thread used to rescan controllers
3729  * @data:        Ignored.
3730  *
3731  * A kernel thread used scan for drive topology changes on
3732  * controllers. The thread processes only one controller at a time
3733  * using a queue.  Controllers are added to the queue using
3734  * add_to_scan_list() and removed from the queue either after done
3735  * processing or using remove_from_scan_list().
3736  *
3737  * returns 0.
3738  **/
3739 static int scan_thread(void *data)
3740 {
3741         struct ctlr_info *h;
3742
3743         while (1) {
3744                 set_current_state(TASK_INTERRUPTIBLE);
3745                 schedule();
3746                 if (kthread_should_stop())
3747                         break;
3748
3749                 while (1) {
3750                         mutex_lock(&scan_mutex);
3751                         if (list_empty(&scan_q)) {
3752                                 mutex_unlock(&scan_mutex);
3753                                 break;
3754                         }
3755
3756                         h = list_entry(scan_q.next,
3757                                        struct ctlr_info,
3758                                        scan_list);
3759                         list_del(&h->scan_list);
3760                         h->busy_scanning = 1;
3761                         mutex_unlock(&scan_mutex);
3762
3763                         rebuild_lun_table(h, 0, 0);
3764                         complete_all(&h->scan_wait);
3765                         mutex_lock(&scan_mutex);
3766                         h->busy_scanning = 0;
3767                         mutex_unlock(&scan_mutex);
3768                 }
3769         }
3770
3771         return 0;
3772 }
3773
3774 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3775 {
3776         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3777                 return 0;
3778
3779         switch (c->err_info->SenseInfo[12]) {
3780         case STATE_CHANGED:
3781                 dev_warn(&h->pdev->dev, "a state change "
3782                         "detected, command retried\n");
3783                 return 1;
3784         break;
3785         case LUN_FAILED:
3786                 dev_warn(&h->pdev->dev, "LUN failure "
3787                         "detected, action required\n");
3788                 return 1;
3789         break;
3790         case REPORT_LUNS_CHANGED:
3791                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3792         /*
3793          * Here, we could call add_to_scan_list and wake up the scan thread,
3794          * except that it's quite likely that we will get more than one
3795          * REPORT_LUNS_CHANGED condition in quick succession, which means
3796          * that those which occur after the first one will likely happen
3797          * *during* the scan_thread's rescan.  And the rescan code is not
3798          * robust enough to restart in the middle, undoing what it has already
3799          * done, and it's not clear that it's even possible to do this, since
3800          * part of what it does is notify the block layer, which starts
3801          * doing it's own i/o to read partition tables and so on, and the
3802          * driver doesn't have visibility to know what might need undoing.
3803          * In any event, if possible, it is horribly complicated to get right
3804          * so we just don't do it for now.
3805          *
3806          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3807          */
3808                 return 1;
3809         break;
3810         case POWER_OR_RESET:
3811                 dev_warn(&h->pdev->dev,
3812                         "a power on or device reset detected\n");
3813                 return 1;
3814         break;
3815         case UNIT_ATTENTION_CLEARED:
3816                 dev_warn(&h->pdev->dev,
3817                         "unit attention cleared by another initiator\n");
3818                 return 1;
3819         break;
3820         default:
3821                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3822                 return 1;
3823         }
3824 }
3825
3826 /*
3827  *  We cannot read the structure directly, for portability we must use
3828  *   the io functions.
3829  *   This is for debug only.
3830  */
3831 static void print_cfg_table(ctlr_info_t *h)
3832 {
3833         int i;
3834         char temp_name[17];
3835         CfgTable_struct *tb = h->cfgtable;
3836
3837         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3838         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3839         for (i = 0; i < 4; i++)
3840                 temp_name[i] = readb(&(tb->Signature[i]));
3841         temp_name[4] = '\0';
3842         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3843         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3844                 readl(&(tb->SpecValence)));
3845         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3846                readl(&(tb->TransportSupport)));
3847         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3848                readl(&(tb->TransportActive)));
3849         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3850                readl(&(tb->HostWrite.TransportRequest)));
3851         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3852                readl(&(tb->HostWrite.CoalIntDelay)));
3853         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3854                readl(&(tb->HostWrite.CoalIntCount)));
3855         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%x\n",
3856                readl(&(tb->CmdsOutMax)));
3857         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3858                 readl(&(tb->BusTypes)));
3859         for (i = 0; i < 16; i++)
3860                 temp_name[i] = readb(&(tb->ServerName[i]));
3861         temp_name[16] = '\0';
3862         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3863         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3864                 readl(&(tb->HeartBeat)));
3865 }
3866
3867 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3868 {
3869         int i, offset, mem_type, bar_type;
3870         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3871                 return 0;
3872         offset = 0;
3873         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3874                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3875                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3876                         offset += 4;
3877                 else {
3878                         mem_type = pci_resource_flags(pdev, i) &
3879                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3880                         switch (mem_type) {
3881                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3882                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3883                                 offset += 4;    /* 32 bit */
3884                                 break;
3885                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3886                                 offset += 8;
3887                                 break;
3888                         default:        /* reserved in PCI 2.2 */
3889                                 dev_warn(&pdev->dev,
3890                                        "Base address is invalid\n");
3891                                 return -1;
3892                                 break;
3893                         }
3894                 }
3895                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3896                         return i + 1;
3897         }
3898         return -1;
3899 }
3900
3901 /* Fill in bucket_map[], given nsgs (the max number of
3902  * scatter gather elements supported) and bucket[],
3903  * which is an array of 8 integers.  The bucket[] array
3904  * contains 8 different DMA transfer sizes (in 16
3905  * byte increments) which the controller uses to fetch
3906  * commands.  This function fills in bucket_map[], which
3907  * maps a given number of scatter gather elements to one of
3908  * the 8 DMA transfer sizes.  The point of it is to allow the
3909  * controller to only do as much DMA as needed to fetch the
3910  * command, with the DMA transfer size encoded in the lower
3911  * bits of the command address.
3912  */
3913 static void  calc_bucket_map(int bucket[], int num_buckets,
3914         int nsgs, int *bucket_map)
3915 {
3916         int i, j, b, size;
3917
3918         /* even a command with 0 SGs requires 4 blocks */
3919 #define MINIMUM_TRANSFER_BLOCKS 4
3920 #define NUM_BUCKETS 8
3921         /* Note, bucket_map must have nsgs+1 entries. */
3922         for (i = 0; i <= nsgs; i++) {
3923                 /* Compute size of a command with i SG entries */
3924                 size = i + MINIMUM_TRANSFER_BLOCKS;
3925                 b = num_buckets; /* Assume the biggest bucket */
3926                 /* Find the bucket that is just big enough */
3927                 for (j = 0; j < 8; j++) {
3928                         if (bucket[j] >= size) {
3929                                 b = j;
3930                                 break;
3931                         }
3932                 }
3933                 /* for a command with i SG entries, use bucket b. */
3934                 bucket_map[i] = b;
3935         }
3936 }
3937
3938 static void cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3939 {
3940         int i;
3941
3942         /* under certain very rare conditions, this can take awhile.
3943          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3944          * as we enter this code.) */
3945         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3946                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3947                         break;
3948                 usleep_range(10000, 20000);
3949         }
3950 }
3951
3952 static void cciss_enter_performant_mode(ctlr_info_t *h, u32 use_short_tags)
3953 {
3954         /* This is a bit complicated.  There are 8 registers on
3955          * the controller which we write to to tell it 8 different
3956          * sizes of commands which there may be.  It's a way of
3957          * reducing the DMA done to fetch each command.  Encoded into
3958          * each command's tag are 3 bits which communicate to the controller
3959          * which of the eight sizes that command fits within.  The size of
3960          * each command depends on how many scatter gather entries there are.
3961          * Each SG entry requires 16 bytes.  The eight registers are programmed
3962          * with the number of 16-byte blocks a command of that size requires.
3963          * The smallest command possible requires 5 such 16 byte blocks.
3964          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3965          * blocks.  Note, this only extends to the SG entries contained
3966          * within the command block, and does not extend to chained blocks
3967          * of SG elements.   bft[] contains the eight values we write to
3968          * the registers.  They are not evenly distributed, but have more
3969          * sizes for small commands, and fewer sizes for larger commands.
3970          */
3971         __u32 trans_offset;
3972         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3973                         /*
3974                          *  5 = 1 s/g entry or 4k
3975                          *  6 = 2 s/g entry or 8k
3976                          *  8 = 4 s/g entry or 16k
3977                          * 10 = 6 s/g entry or 24k
3978                          */
3979         unsigned long register_value;
3980         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3981
3982         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3983
3984         /* Controller spec: zero out this buffer. */
3985         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3986         h->reply_pool_head = h->reply_pool;
3987
3988         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3989         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3990                                 h->blockFetchTable);
3991         writel(bft[0], &h->transtable->BlockFetch0);
3992         writel(bft[1], &h->transtable->BlockFetch1);
3993         writel(bft[2], &h->transtable->BlockFetch2);
3994         writel(bft[3], &h->transtable->BlockFetch3);
3995         writel(bft[4], &h->transtable->BlockFetch4);
3996         writel(bft[5], &h->transtable->BlockFetch5);
3997         writel(bft[6], &h->transtable->BlockFetch6);
3998         writel(bft[7], &h->transtable->BlockFetch7);
3999
4000         /* size of controller ring buffer */
4001         writel(h->max_commands, &h->transtable->RepQSize);
4002         writel(1, &h->transtable->RepQCount);
4003         writel(0, &h->transtable->RepQCtrAddrLow32);
4004         writel(0, &h->transtable->RepQCtrAddrHigh32);
4005         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
4006         writel(0, &h->transtable->RepQAddr0High32);
4007         writel(CFGTBL_Trans_Performant | use_short_tags,
4008                         &(h->cfgtable->HostWrite.TransportRequest));
4009
4010         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
4011         cciss_wait_for_mode_change_ack(h);
4012         register_value = readl(&(h->cfgtable->TransportActive));
4013         if (!(register_value & CFGTBL_Trans_Performant))
4014                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
4015                                         " performant mode\n");
4016 }
4017
4018 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h)
4019 {
4020         __u32 trans_support;
4021
4022         if (cciss_simple_mode)
4023                 return;
4024
4025         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
4026         /* Attempt to put controller into performant mode if supported */
4027         /* Does board support performant mode? */
4028         trans_support = readl(&(h->cfgtable->TransportSupport));
4029         if (!(trans_support & PERFORMANT_MODE))
4030                 return;
4031
4032         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
4033         /* Performant mode demands commands on a 32 byte boundary
4034          * pci_alloc_consistent aligns on page boundarys already.
4035          * Just need to check if divisible by 32
4036          */
4037         if ((sizeof(CommandList_struct) % 32) != 0) {
4038                 dev_warn(&h->pdev->dev, "%s %d %s\n",
4039                         "cciss info: command size[",
4040                         (int)sizeof(CommandList_struct),
4041                         "] not divisible by 32, no performant mode..\n");
4042                 return;
4043         }
4044
4045         /* Performant mode ring buffer and supporting data structures */
4046         h->reply_pool = (__u64 *)pci_alloc_consistent(
4047                 h->pdev, h->max_commands * sizeof(__u64),
4048                 &(h->reply_pool_dhandle));
4049
4050         /* Need a block fetch table for performant mode */
4051         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
4052                 sizeof(__u32)), GFP_KERNEL);
4053
4054         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
4055                 goto clean_up;
4056
4057         cciss_enter_performant_mode(h,
4058                 trans_support & CFGTBL_Trans_use_short_tags);
4059
4060         /* Change the access methods to the performant access methods */
4061         h->access = SA5_performant_access;
4062         h->transMethod = CFGTBL_Trans_Performant;
4063
4064         return;
4065 clean_up:
4066         kfree(h->blockFetchTable);
4067         if (h->reply_pool)
4068                 pci_free_consistent(h->pdev,
4069                                 h->max_commands * sizeof(__u64),
4070                                 h->reply_pool,
4071                                 h->reply_pool_dhandle);
4072         return;
4073
4074 } /* cciss_put_controller_into_performant_mode */
4075
4076 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
4077  * controllers that are capable. If not, we use IO-APIC mode.
4078  */
4079
4080 static void cciss_interrupt_mode(ctlr_info_t *h)
4081 {
4082         int ret;
4083
4084         /* Some boards advertise MSI but don't really support it */
4085         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
4086             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
4087                 goto default_int_mode;
4088
4089         ret = pci_alloc_irq_vectors(h->pdev, 4, 4, PCI_IRQ_MSIX);
4090         if (ret >= 0)   {
4091                 h->intr[0] = pci_irq_vector(h->pdev, 0);
4092                 h->intr[1] = pci_irq_vector(h->pdev, 1);
4093                 h->intr[2] = pci_irq_vector(h->pdev, 2);
4094                 h->intr[3] = pci_irq_vector(h->pdev, 3);
4095                 return;
4096         }
4097
4098         ret = pci_alloc_irq_vectors(h->pdev, 1, 1, PCI_IRQ_MSI);
4099
4100 default_int_mode:
4101         /* if we get here we're going to use the default interrupt mode */
4102         h->intr[h->intr_mode] = pci_irq_vector(h->pdev, 0);
4103         return;
4104 }
4105
4106 static int cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
4107 {
4108         int i;
4109         u32 subsystem_vendor_id, subsystem_device_id;
4110
4111         subsystem_vendor_id = pdev->subsystem_vendor;
4112         subsystem_device_id = pdev->subsystem_device;
4113         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
4114                         subsystem_vendor_id;
4115
4116         for (i = 0; i < ARRAY_SIZE(products); i++) {
4117                 /* Stand aside for hpsa driver on request */
4118                 if (cciss_allow_hpsa)
4119                         return -ENODEV;
4120                 if (*board_id == products[i].board_id)
4121                         return i;
4122         }
4123         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4124                 *board_id);
4125         return -ENODEV;
4126 }
4127
4128 static inline bool cciss_board_disabled(ctlr_info_t *h)
4129 {
4130         u16 command;
4131
4132         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4133         return ((command & PCI_COMMAND_MEMORY) == 0);
4134 }
4135
4136 static int cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4137                                      unsigned long *memory_bar)
4138 {
4139         int i;
4140
4141         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4142                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4143                         /* addressing mode bits already removed */
4144                         *memory_bar = pci_resource_start(pdev, i);
4145                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4146                                 *memory_bar);
4147                         return 0;
4148                 }
4149         dev_warn(&pdev->dev, "no memory BAR found\n");
4150         return -ENODEV;
4151 }
4152
4153 static int cciss_wait_for_board_state(struct pci_dev *pdev,
4154                                       void __iomem *vaddr, int wait_for_ready)
4155 #define BOARD_READY 1
4156 #define BOARD_NOT_READY 0
4157 {
4158         int i, iterations;
4159         u32 scratchpad;
4160
4161         if (wait_for_ready)
4162                 iterations = CCISS_BOARD_READY_ITERATIONS;
4163         else
4164                 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
4165
4166         for (i = 0; i < iterations; i++) {
4167                 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
4168                 if (wait_for_ready) {
4169                         if (scratchpad == CCISS_FIRMWARE_READY)
4170                                 return 0;
4171                 } else {
4172                         if (scratchpad != CCISS_FIRMWARE_READY)
4173                                 return 0;
4174                 }
4175                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4176         }
4177         dev_warn(&pdev->dev, "board not ready, timed out.\n");
4178         return -ENODEV;
4179 }
4180
4181 static int cciss_find_cfg_addrs(struct pci_dev *pdev, void __iomem *vaddr,
4182                                 u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4183                                 u64 *cfg_offset)
4184 {
4185         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4186         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4187         *cfg_base_addr &= (u32) 0x0000ffff;
4188         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4189         if (*cfg_base_addr_index == -1) {
4190                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4191                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4192                 return -ENODEV;
4193         }
4194         return 0;
4195 }
4196
4197 static int cciss_find_cfgtables(ctlr_info_t *h)
4198 {
4199         u64 cfg_offset;
4200         u32 cfg_base_addr;
4201         u64 cfg_base_addr_index;
4202         u32 trans_offset;
4203         int rc;
4204
4205         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4206                 &cfg_base_addr_index, &cfg_offset);
4207         if (rc)
4208                 return rc;
4209         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4210                 cfg_base_addr_index) + cfg_offset, sizeof(*h->cfgtable));
4211         if (!h->cfgtable)
4212                 return -ENOMEM;
4213         rc = write_driver_ver_to_cfgtable(h->cfgtable);
4214         if (rc)
4215                 return rc;
4216         /* Find performant mode table. */
4217         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4218         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4219                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4220                                 sizeof(*h->transtable));
4221         if (!h->transtable)
4222                 return -ENOMEM;
4223         return 0;
4224 }
4225
4226 static void cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4227 {
4228         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4229
4230         /* Limit commands in memory limited kdump scenario. */
4231         if (reset_devices && h->max_commands > 32)
4232                 h->max_commands = 32;
4233
4234         if (h->max_commands < 16) {
4235                 dev_warn(&h->pdev->dev, "Controller reports "
4236                         "max supported commands of %d, an obvious lie. "
4237                         "Using 16.  Ensure that firmware is up to date.\n",
4238                         h->max_commands);
4239                 h->max_commands = 16;
4240         }
4241 }
4242
4243 /* Interrogate the hardware for some limits:
4244  * max commands, max SG elements without chaining, and with chaining,
4245  * SG chain block size, etc.
4246  */
4247 static void cciss_find_board_params(ctlr_info_t *h)
4248 {
4249         cciss_get_max_perf_mode_cmds(h);
4250         h->nr_cmds = h->max_commands - 4 - cciss_tape_cmds;
4251         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4252         /*
4253          * The P600 may exhibit poor performnace under some workloads
4254          * if we use the value in the configuration table. Limit this
4255          * controller to MAXSGENTRIES (32) instead.
4256          */
4257         if (h->board_id == 0x3225103C)
4258                 h->maxsgentries = MAXSGENTRIES;
4259         /*
4260          * Limit in-command s/g elements to 32 save dma'able memory.
4261          * Howvever spec says if 0, use 31
4262          */
4263         h->max_cmd_sgentries = 31;
4264         if (h->maxsgentries > 512) {
4265                 h->max_cmd_sgentries = 32;
4266                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4267                 h->maxsgentries--; /* save one for chain pointer */
4268         } else {
4269                 h->maxsgentries = 31; /* default to traditional values */
4270                 h->chainsize = 0;
4271         }
4272 }
4273
4274 static inline bool CISS_signature_present(ctlr_info_t *h)
4275 {
4276         if (!check_signature(h->cfgtable->Signature, "CISS", 4)) {
4277                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4278                 return false;
4279         }
4280         return true;
4281 }
4282
4283 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4284 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4285 {
4286 #ifdef CONFIG_X86
4287         u32 prefetch;
4288
4289         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4290         prefetch |= 0x100;
4291         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4292 #endif
4293 }
4294
4295 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4296  * in a prefetch beyond physical memory.
4297  */
4298 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4299 {
4300         u32 dma_prefetch;
4301         __u32 dma_refetch;
4302
4303         if (h->board_id != 0x3225103C)
4304                 return;
4305         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4306         dma_prefetch |= 0x8000;
4307         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4308         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4309         dma_refetch |= 0x1;
4310         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4311 }
4312
4313 static int cciss_pci_init(ctlr_info_t *h)
4314 {
4315         int prod_index, err;
4316
4317         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4318         if (prod_index < 0)
4319                 return -ENODEV;
4320         h->product_name = products[prod_index].product_name;
4321         h->access = *(products[prod_index].access);
4322
4323         if (cciss_board_disabled(h)) {
4324                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4325                 return -ENODEV;
4326         }
4327
4328         pci_disable_link_state(h->pdev, PCIE_LINK_STATE_L0S |
4329                                 PCIE_LINK_STATE_L1 | PCIE_LINK_STATE_CLKPM);
4330
4331         err = pci_enable_device(h->pdev);
4332         if (err) {
4333                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4334                 return err;
4335         }
4336
4337         err = pci_request_regions(h->pdev, "cciss");
4338         if (err) {
4339                 dev_warn(&h->pdev->dev,
4340                         "Cannot obtain PCI resources, aborting\n");
4341                 return err;
4342         }
4343
4344         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4345         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4346
4347 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4348  * else we use the IO-APIC interrupt assigned to us by system ROM.
4349  */
4350         cciss_interrupt_mode(h);
4351         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4352         if (err)
4353                 goto err_out_free_res;
4354         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4355         if (!h->vaddr) {
4356                 err = -ENOMEM;
4357                 goto err_out_free_res;
4358         }
4359         err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4360         if (err)
4361                 goto err_out_free_res;
4362         err = cciss_find_cfgtables(h);
4363         if (err)
4364                 goto err_out_free_res;
4365         print_cfg_table(h);
4366         cciss_find_board_params(h);
4367
4368         if (!CISS_signature_present(h)) {
4369                 err = -ENODEV;
4370                 goto err_out_free_res;
4371         }
4372         cciss_enable_scsi_prefetch(h);
4373         cciss_p600_dma_prefetch_quirk(h);
4374         err = cciss_enter_simple_mode(h);
4375         if (err)
4376                 goto err_out_free_res;
4377         cciss_put_controller_into_performant_mode(h);
4378         return 0;
4379
4380 err_out_free_res:
4381         /*
4382          * Deliberately omit pci_disable_device(): it does something nasty to
4383          * Smart Array controllers that pci_enable_device does not undo
4384          */
4385         if (h->transtable)
4386                 iounmap(h->transtable);
4387         if (h->cfgtable)
4388                 iounmap(h->cfgtable);
4389         if (h->vaddr)
4390                 iounmap(h->vaddr);
4391         pci_release_regions(h->pdev);
4392         return err;
4393 }
4394
4395 /* Function to find the first free pointer into our hba[] array
4396  * Returns -1 if no free entries are left.
4397  */
4398 static int alloc_cciss_hba(struct pci_dev *pdev)
4399 {
4400         int i;
4401
4402         for (i = 0; i < MAX_CTLR; i++) {
4403                 if (!hba[i]) {
4404                         ctlr_info_t *h;
4405
4406                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4407                         if (!h)
4408                                 goto Enomem;
4409                         hba[i] = h;
4410                         return i;
4411                 }
4412         }
4413         dev_warn(&pdev->dev, "This driver supports a maximum"
4414                " of %d controllers.\n", MAX_CTLR);
4415         return -1;
4416 Enomem:
4417         dev_warn(&pdev->dev, "out of memory.\n");
4418         return -1;
4419 }
4420
4421 static void free_hba(ctlr_info_t *h)
4422 {
4423         int i;
4424
4425         hba[h->ctlr] = NULL;
4426         for (i = 0; i < h->highest_lun + 1; i++)
4427                 if (h->gendisk[i] != NULL)
4428                         put_disk(h->gendisk[i]);
4429         kfree(h);
4430 }
4431
4432 /* Send a message CDB to the firmware. */
4433 static int cciss_message(struct pci_dev *pdev, unsigned char opcode,
4434                          unsigned char type)
4435 {
4436         typedef struct {
4437                 CommandListHeader_struct CommandHeader;
4438                 RequestBlock_struct Request;
4439                 ErrDescriptor_struct ErrorDescriptor;
4440         } Command;
4441         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4442         Command *cmd;
4443         dma_addr_t paddr64;
4444         uint32_t paddr32, tag;
4445         void __iomem *vaddr;
4446         int i, err;
4447
4448         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4449         if (vaddr == NULL)
4450                 return -ENOMEM;
4451
4452         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4453            CCISS commands, so they must be allocated from the lower 4GiB of
4454            memory. */
4455         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4456         if (err) {
4457                 iounmap(vaddr);
4458                 return -ENOMEM;
4459         }
4460
4461         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4462         if (cmd == NULL) {
4463                 iounmap(vaddr);
4464                 return -ENOMEM;
4465         }
4466
4467         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4468            although there's no guarantee, we assume that the address is at
4469            least 4-byte aligned (most likely, it's page-aligned). */
4470         paddr32 = paddr64;
4471
4472         cmd->CommandHeader.ReplyQueue = 0;
4473         cmd->CommandHeader.SGList = 0;
4474         cmd->CommandHeader.SGTotal = 0;
4475         cmd->CommandHeader.Tag.lower = paddr32;
4476         cmd->CommandHeader.Tag.upper = 0;
4477         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4478
4479         cmd->Request.CDBLen = 16;
4480         cmd->Request.Type.Type = TYPE_MSG;
4481         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4482         cmd->Request.Type.Direction = XFER_NONE;
4483         cmd->Request.Timeout = 0; /* Don't time out */
4484         cmd->Request.CDB[0] = opcode;
4485         cmd->Request.CDB[1] = type;
4486         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4487
4488         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4489         cmd->ErrorDescriptor.Addr.upper = 0;
4490         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4491
4492         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4493
4494         for (i = 0; i < 10; i++) {
4495                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4496                 if ((tag & ~3) == paddr32)
4497                         break;
4498                 msleep(CCISS_POST_RESET_NOOP_TIMEOUT_MSECS);
4499         }
4500
4501         iounmap(vaddr);
4502
4503         /* we leak the DMA buffer here ... no choice since the controller could
4504            still complete the command. */
4505         if (i == 10) {
4506                 dev_err(&pdev->dev,
4507                         "controller message %02x:%02x timed out\n",
4508                         opcode, type);
4509                 return -ETIMEDOUT;
4510         }
4511
4512         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4513
4514         if (tag & 2) {
4515                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4516                         opcode, type);
4517                 return -EIO;
4518         }
4519
4520         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4521                 opcode, type);
4522         return 0;
4523 }
4524
4525 #define cciss_noop(p) cciss_message(p, 3, 0)
4526
4527 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4528         void * __iomem vaddr, u32 use_doorbell)
4529 {
4530         u16 pmcsr;
4531         int pos;
4532
4533         if (use_doorbell) {
4534                 /* For everything after the P600, the PCI power state method
4535                  * of resetting the controller doesn't work, so we have this
4536                  * other way using the doorbell register.
4537                  */
4538                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4539                 writel(use_doorbell, vaddr + SA5_DOORBELL);
4540         } else { /* Try to do it the PCI power state way */
4541
4542                 /* Quoting from the Open CISS Specification: "The Power
4543                  * Management Control/Status Register (CSR) controls the power
4544                  * state of the device.  The normal operating state is D0,
4545                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4546                  * the controller, place the interface device in D3 then to D0,
4547                  * this causes a secondary PCI reset which will reset the
4548                  * controller." */
4549
4550                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4551                 if (pos == 0) {
4552                         dev_err(&pdev->dev,
4553                                 "cciss_controller_hard_reset: "
4554                                 "PCI PM not supported\n");
4555                         return -ENODEV;
4556                 }
4557                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4558                 /* enter the D3hot power management state */
4559                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4560                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4561                 pmcsr |= PCI_D3hot;
4562                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4563
4564                 msleep(500);
4565
4566                 /* enter the D0 power management state */
4567                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4568                 pmcsr |= PCI_D0;
4569                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4570
4571                 /*
4572                  * The P600 requires a small delay when changing states.
4573                  * Otherwise we may think the board did not reset and we bail.
4574                  * This for kdump only and is particular to the P600.
4575                  */
4576                 msleep(500);
4577         }
4578         return 0;
4579 }
4580
4581 static void init_driver_version(char *driver_version, int len)
4582 {
4583         memset(driver_version, 0, len);
4584         strncpy(driver_version, "cciss " DRIVER_NAME, len - 1);
4585 }
4586
4587 static int write_driver_ver_to_cfgtable(CfgTable_struct __iomem *cfgtable)
4588 {
4589         char *driver_version;
4590         int i, size = sizeof(cfgtable->driver_version);
4591
4592         driver_version = kmalloc(size, GFP_KERNEL);
4593         if (!driver_version)
4594                 return -ENOMEM;
4595
4596         init_driver_version(driver_version, size);
4597         for (i = 0; i < size; i++)
4598                 writeb(driver_version[i], &cfgtable->driver_version[i]);
4599         kfree(driver_version);
4600         return 0;
4601 }
4602
4603 static void read_driver_ver_from_cfgtable(CfgTable_struct __iomem *cfgtable,
4604                                           unsigned char *driver_ver)
4605 {
4606         int i;
4607
4608         for (i = 0; i < sizeof(cfgtable->driver_version); i++)
4609                 driver_ver[i] = readb(&cfgtable->driver_version[i]);
4610 }
4611
4612 static int controller_reset_failed(CfgTable_struct __iomem *cfgtable)
4613 {
4614
4615         char *driver_ver, *old_driver_ver;
4616         int rc, size = sizeof(cfgtable->driver_version);
4617
4618         old_driver_ver = kmalloc(2 * size, GFP_KERNEL);
4619         if (!old_driver_ver)
4620                 return -ENOMEM;
4621         driver_ver = old_driver_ver + size;
4622
4623         /* After a reset, the 32 bytes of "driver version" in the cfgtable
4624          * should have been changed, otherwise we know the reset failed.
4625          */
4626         init_driver_version(old_driver_ver, size);
4627         read_driver_ver_from_cfgtable(cfgtable, driver_ver);
4628         rc = !memcmp(driver_ver, old_driver_ver, size);
4629         kfree(old_driver_ver);
4630         return rc;
4631 }
4632
4633 /* This does a hard reset of the controller using PCI power management
4634  * states or using the doorbell register. */
4635 static int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4636 {
4637         u64 cfg_offset;
4638         u32 cfg_base_addr;
4639         u64 cfg_base_addr_index;
4640         void __iomem *vaddr;
4641         unsigned long paddr;
4642         u32 misc_fw_support;
4643         int rc;
4644         CfgTable_struct __iomem *cfgtable;
4645         u32 use_doorbell;
4646         u32 board_id;
4647         u16 command_register;
4648
4649         /* For controllers as old a the p600, this is very nearly
4650          * the same thing as
4651          *
4652          * pci_save_state(pci_dev);
4653          * pci_set_power_state(pci_dev, PCI_D3hot);
4654          * pci_set_power_state(pci_dev, PCI_D0);
4655          * pci_restore_state(pci_dev);
4656          *
4657          * For controllers newer than the P600, the pci power state
4658          * method of resetting doesn't work so we have another way
4659          * using the doorbell register.
4660          */
4661
4662         /* Exclude 640x boards.  These are two pci devices in one slot
4663          * which share a battery backed cache module.  One controls the
4664          * cache, the other accesses the cache through the one that controls
4665          * it.  If we reset the one controlling the cache, the other will
4666          * likely not be happy.  Just forbid resetting this conjoined mess.
4667          */
4668         cciss_lookup_board_id(pdev, &board_id);
4669         if (!ctlr_is_resettable(board_id)) {
4670                 dev_warn(&pdev->dev, "Controller not resettable\n");
4671                 return -ENODEV;
4672         }
4673
4674         /* if controller is soft- but not hard resettable... */
4675         if (!ctlr_is_hard_resettable(board_id))
4676                 return -ENOTSUPP; /* try soft reset later. */
4677
4678         /* Save the PCI command register */
4679         pci_read_config_word(pdev, 4, &command_register);
4680         /* Turn the board off.  This is so that later pci_restore_state()
4681          * won't turn the board on before the rest of config space is ready.
4682          */
4683         pci_disable_device(pdev);
4684         pci_save_state(pdev);
4685
4686         /* find the first memory BAR, so we can find the cfg table */
4687         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4688         if (rc)
4689                 return rc;
4690         vaddr = remap_pci_mem(paddr, 0x250);
4691         if (!vaddr)
4692                 return -ENOMEM;
4693
4694         /* find cfgtable in order to check if reset via doorbell is supported */
4695         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4696                                         &cfg_base_addr_index, &cfg_offset);
4697         if (rc)
4698                 goto unmap_vaddr;
4699         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4700                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4701         if (!cfgtable) {
4702                 rc = -ENOMEM;
4703                 goto unmap_vaddr;
4704         }
4705         rc = write_driver_ver_to_cfgtable(cfgtable);
4706         if (rc)
4707                 goto unmap_vaddr;
4708
4709         /* If reset via doorbell register is supported, use that.
4710          * There are two such methods.  Favor the newest method.
4711          */
4712         misc_fw_support = readl(&cfgtable->misc_fw_support);
4713         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET2;
4714         if (use_doorbell) {
4715                 use_doorbell = DOORBELL_CTLR_RESET2;
4716         } else {
4717                 use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4718                 if (use_doorbell) {
4719                         dev_warn(&pdev->dev, "Controller claims that "
4720                                 "'Bit 2 doorbell reset' is "
4721                                 "supported, but not 'bit 5 doorbell reset'.  "
4722                                 "Firmware update is recommended.\n");
4723                         rc = -ENOTSUPP; /* use the soft reset */
4724                         goto unmap_cfgtable;
4725                 }
4726         }
4727
4728         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4729         if (rc)
4730                 goto unmap_cfgtable;
4731         pci_restore_state(pdev);
4732         rc = pci_enable_device(pdev);
4733         if (rc) {
4734                 dev_warn(&pdev->dev, "failed to enable device.\n");
4735                 goto unmap_cfgtable;
4736         }
4737         pci_write_config_word(pdev, 4, command_register);
4738
4739         /* Some devices (notably the HP Smart Array 5i Controller)
4740            need a little pause here */
4741         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4742
4743         /* Wait for board to become not ready, then ready. */
4744         dev_info(&pdev->dev, "Waiting for board to reset.\n");
4745         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4746         if (rc) {
4747                 dev_warn(&pdev->dev, "Failed waiting for board to hard reset."
4748                                 "  Will try soft reset.\n");
4749                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4750                 goto unmap_cfgtable;
4751         }
4752         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4753         if (rc) {
4754                 dev_warn(&pdev->dev,
4755                         "failed waiting for board to become ready "
4756                         "after hard reset\n");
4757                 goto unmap_cfgtable;
4758         }
4759
4760         rc = controller_reset_failed(vaddr);
4761         if (rc < 0)
4762                 goto unmap_cfgtable;
4763         if (rc) {
4764                 dev_warn(&pdev->dev, "Unable to successfully hard reset "
4765                         "controller. Will try soft reset.\n");
4766                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4767         } else {
4768                 dev_info(&pdev->dev, "Board ready after hard reset.\n");
4769         }
4770
4771 unmap_cfgtable:
4772         iounmap(cfgtable);
4773
4774 unmap_vaddr:
4775         iounmap(vaddr);
4776         return rc;
4777 }
4778
4779 static int cciss_init_reset_devices(struct pci_dev *pdev)
4780 {
4781         int rc, i;
4782
4783         if (!reset_devices)
4784                 return 0;
4785
4786         /* Reset the controller with a PCI power-cycle or via doorbell */
4787         rc = cciss_kdump_hard_reset_controller(pdev);
4788
4789         /* -ENOTSUPP here means we cannot reset the controller
4790          * but it's already (and still) up and running in
4791          * "performant mode".  Or, it might be 640x, which can't reset
4792          * due to concerns about shared bbwc between 6402/6404 pair.
4793          */
4794         if (rc == -ENOTSUPP)
4795                 return rc; /* just try to do the kdump anyhow. */
4796         if (rc)
4797                 return -ENODEV;
4798
4799         /* Now try to get the controller to respond to a no-op */
4800         dev_warn(&pdev->dev, "Waiting for controller to respond to no-op\n");
4801         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4802                 if (cciss_noop(pdev) == 0)
4803                         break;
4804                 else
4805                         dev_warn(&pdev->dev, "no-op failed%s\n",
4806                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4807                                         "; re-trying" : ""));
4808                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4809         }
4810         return 0;
4811 }
4812
4813 static int cciss_allocate_cmd_pool(ctlr_info_t *h)
4814 {
4815         h->cmd_pool_bits = kmalloc(BITS_TO_LONGS(h->nr_cmds) *
4816                 sizeof(unsigned long), GFP_KERNEL);
4817         h->cmd_pool = pci_alloc_consistent(h->pdev,
4818                 h->nr_cmds * sizeof(CommandList_struct),
4819                 &(h->cmd_pool_dhandle));
4820         h->errinfo_pool = pci_alloc_consistent(h->pdev,
4821                 h->nr_cmds * sizeof(ErrorInfo_struct),
4822                 &(h->errinfo_pool_dhandle));
4823         if ((h->cmd_pool_bits == NULL)
4824                 || (h->cmd_pool == NULL)
4825                 || (h->errinfo_pool == NULL)) {
4826                 dev_err(&h->pdev->dev, "out of memory");
4827                 return -ENOMEM;
4828         }
4829         return 0;
4830 }
4831
4832 static int cciss_allocate_scatterlists(ctlr_info_t *h)
4833 {
4834         int i;
4835
4836         /* zero it, so that on free we need not know how many were alloc'ed */
4837         h->scatter_list = kzalloc(h->max_commands *
4838                                 sizeof(struct scatterlist *), GFP_KERNEL);
4839         if (!h->scatter_list)
4840                 return -ENOMEM;
4841
4842         for (i = 0; i < h->nr_cmds; i++) {
4843                 h->scatter_list[i] = kmalloc(sizeof(struct scatterlist) *
4844                                                 h->maxsgentries, GFP_KERNEL);
4845                 if (h->scatter_list[i] == NULL) {
4846                         dev_err(&h->pdev->dev, "could not allocate "
4847                                 "s/g lists\n");
4848                         return -ENOMEM;
4849                 }
4850         }
4851         return 0;
4852 }
4853
4854 static void cciss_free_scatterlists(ctlr_info_t *h)
4855 {
4856         int i;
4857
4858         if (h->scatter_list) {
4859                 for (i = 0; i < h->nr_cmds; i++)
4860                         kfree(h->scatter_list[i]);
4861                 kfree(h->scatter_list);
4862         }
4863 }
4864
4865 static void cciss_free_cmd_pool(ctlr_info_t *h)
4866 {
4867         kfree(h->cmd_pool_bits);
4868         if (h->cmd_pool)
4869                 pci_free_consistent(h->pdev,
4870                         h->nr_cmds * sizeof(CommandList_struct),
4871                         h->cmd_pool, h->cmd_pool_dhandle);
4872         if (h->errinfo_pool)
4873                 pci_free_consistent(h->pdev,
4874                         h->nr_cmds * sizeof(ErrorInfo_struct),
4875                         h->errinfo_pool, h->errinfo_pool_dhandle);
4876 }
4877
4878 static int cciss_request_irq(ctlr_info_t *h,
4879         irqreturn_t (*msixhandler)(int, void *),
4880         irqreturn_t (*intxhandler)(int, void *))
4881 {
4882         if (h->pdev->msi_enabled || h->pdev->msix_enabled) {
4883                 if (!request_irq(h->intr[h->intr_mode], msixhandler,
4884                                 0, h->devname, h))
4885                         return 0;
4886                 dev_err(&h->pdev->dev, "Unable to get msi irq %d"
4887                         " for %s\n", h->intr[h->intr_mode],
4888                         h->devname);
4889                 return -1;
4890         }
4891
4892         if (!request_irq(h->intr[h->intr_mode], intxhandler,
4893                         IRQF_SHARED, h->devname, h))
4894                 return 0;
4895         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4896                 h->intr[h->intr_mode], h->devname);
4897         return -1;
4898 }
4899
4900 static int cciss_kdump_soft_reset(ctlr_info_t *h)
4901 {
4902         if (cciss_send_reset(h, CTLR_LUNID, CCISS_RESET_TYPE_CONTROLLER)) {
4903                 dev_warn(&h->pdev->dev, "Resetting array controller failed.\n");
4904                 return -EIO;
4905         }
4906
4907         dev_info(&h->pdev->dev, "Waiting for board to soft reset.\n");
4908         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_NOT_READY)) {
4909                 dev_warn(&h->pdev->dev, "Soft reset had no effect.\n");
4910                 return -1;
4911         }
4912
4913         dev_info(&h->pdev->dev, "Board reset, awaiting READY status.\n");
4914         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY)) {
4915                 dev_warn(&h->pdev->dev, "Board failed to become ready "
4916                         "after soft reset.\n");
4917                 return -1;
4918         }
4919
4920         return 0;
4921 }
4922
4923 static void cciss_undo_allocations_after_kdump_soft_reset(ctlr_info_t *h)
4924 {
4925         int ctlr = h->ctlr;
4926
4927         free_irq(h->intr[h->intr_mode], h);
4928         pci_free_irq_vectors(h->pdev);
4929         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4930         cciss_free_scatterlists(h);
4931         cciss_free_cmd_pool(h);
4932         kfree(h->blockFetchTable);
4933         if (h->reply_pool)
4934                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
4935                                 h->reply_pool, h->reply_pool_dhandle);
4936         if (h->transtable)
4937                 iounmap(h->transtable);
4938         if (h->cfgtable)
4939                 iounmap(h->cfgtable);
4940         if (h->vaddr)
4941                 iounmap(h->vaddr);
4942         unregister_blkdev(h->major, h->devname);
4943         cciss_destroy_hba_sysfs_entry(h);
4944         pci_release_regions(h->pdev);
4945         kfree(h);
4946         hba[ctlr] = NULL;
4947 }
4948
4949 /*
4950  *  This is it.  Find all the controllers and register them.  I really hate
4951  *  stealing all these major device numbers.
4952  *  returns the number of block devices registered.
4953  */
4954 static int cciss_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4955 {
4956         int i;
4957         int j = 0;
4958         int rc;
4959         int try_soft_reset = 0;
4960         int dac, return_code;
4961         InquiryData_struct *inq_buff;
4962         ctlr_info_t *h;
4963         unsigned long flags;
4964
4965         /*
4966          * By default the cciss driver is used for all older HP Smart Array
4967          * controllers. There are module paramaters that allow a user to
4968          * override this behavior and instead use the hpsa SCSI driver. If
4969          * this is the case cciss may be loaded first from the kdump initrd
4970          * image and cause a kernel panic. So if reset_devices is true and
4971          * cciss_allow_hpsa is set just bail.
4972          */
4973         if ((reset_devices) && (cciss_allow_hpsa == 1))
4974                 return -ENODEV;
4975         rc = cciss_init_reset_devices(pdev);
4976         if (rc) {
4977                 if (rc != -ENOTSUPP)
4978                         return rc;
4979                 /* If the reset fails in a particular way (it has no way to do
4980                  * a proper hard reset, so returns -ENOTSUPP) we can try to do
4981                  * a soft reset once we get the controller configured up to the
4982                  * point that it can accept a command.
4983                  */
4984                 try_soft_reset = 1;
4985                 rc = 0;
4986         }
4987
4988 reinit_after_soft_reset:
4989
4990         i = alloc_cciss_hba(pdev);
4991         if (i < 0)
4992                 return -ENOMEM;
4993
4994         h = hba[i];
4995         h->pdev = pdev;
4996         h->busy_initializing = 1;
4997         h->intr_mode = cciss_simple_mode ? SIMPLE_MODE_INT : PERF_MODE_INT;
4998         INIT_LIST_HEAD(&h->cmpQ);
4999         INIT_LIST_HEAD(&h->reqQ);
5000         mutex_init(&h->busy_shutting_down);
5001
5002         if (cciss_pci_init(h) != 0)
5003                 goto clean_no_release_regions;
5004
5005         sprintf(h->devname, "cciss%d", i);
5006         h->ctlr = i;
5007
5008         if (cciss_tape_cmds < 2)
5009                 cciss_tape_cmds = 2;
5010         if (cciss_tape_cmds > 16)
5011                 cciss_tape_cmds = 16;
5012
5013         init_completion(&h->scan_wait);
5014
5015         if (cciss_create_hba_sysfs_entry(h))
5016                 goto clean0;
5017
5018         /* configure PCI DMA stuff */
5019         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
5020                 dac = 1;
5021         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
5022                 dac = 0;
5023         else {
5024                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
5025                 goto clean1;
5026         }
5027
5028         /*
5029          * register with the major number, or get a dynamic major number
5030          * by passing 0 as argument.  This is done for greater than
5031          * 8 controller support.
5032          */
5033         if (i < MAX_CTLR_ORIG)
5034                 h->major = COMPAQ_CISS_MAJOR + i;
5035         rc = register_blkdev(h->major, h->devname);
5036         if (rc == -EBUSY || rc == -EINVAL) {
5037                 dev_err(&h->pdev->dev,
5038                        "Unable to get major number %d for %s "
5039                        "on hba %d\n", h->major, h->devname, i);
5040                 goto clean1;
5041         } else {
5042                 if (i >= MAX_CTLR_ORIG)
5043                         h->major = rc;
5044         }
5045
5046         /* make sure the board interrupts are off */
5047         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5048         rc = cciss_request_irq(h, do_cciss_msix_intr, do_cciss_intx);
5049         if (rc)
5050                 goto clean2;
5051
5052         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
5053                h->devname, pdev->device, pci_name(pdev),
5054                h->intr[h->intr_mode], dac ? "" : " not");
5055
5056         if (cciss_allocate_cmd_pool(h))
5057                 goto clean4;
5058
5059         if (cciss_allocate_scatterlists(h))
5060                 goto clean4;
5061
5062         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
5063                 h->chainsize, h->nr_cmds);
5064         if (!h->cmd_sg_list && h->chainsize > 0)
5065                 goto clean4;
5066
5067         spin_lock_init(&h->lock);
5068
5069         /* Initialize the pdev driver private data.
5070            have it point to h.  */
5071         pci_set_drvdata(pdev, h);
5072         /* command and error info recs zeroed out before
5073            they are used */
5074         bitmap_zero(h->cmd_pool_bits, h->nr_cmds);
5075
5076         h->num_luns = 0;
5077         h->highest_lun = -1;
5078         for (j = 0; j < CISS_MAX_LUN; j++) {
5079                 h->drv[j] = NULL;
5080                 h->gendisk[j] = NULL;
5081         }
5082
5083         /* At this point, the controller is ready to take commands.
5084          * Now, if reset_devices and the hard reset didn't work, try
5085          * the soft reset and see if that works.
5086          */
5087         if (try_soft_reset) {
5088
5089                 /* This is kind of gross.  We may or may not get a completion
5090                  * from the soft reset command, and if we do, then the value
5091                  * from the fifo may or may not be valid.  So, we wait 10 secs
5092                  * after the reset throwing away any completions we get during
5093                  * that time.  Unregister the interrupt handler and register
5094                  * fake ones to scoop up any residual completions.
5095                  */
5096                 spin_lock_irqsave(&h->lock, flags);
5097                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5098                 spin_unlock_irqrestore(&h->lock, flags);
5099                 free_irq(h->intr[h->intr_mode], h);
5100                 rc = cciss_request_irq(h, cciss_msix_discard_completions,
5101                                         cciss_intx_discard_completions);
5102                 if (rc) {
5103                         dev_warn(&h->pdev->dev, "Failed to request_irq after "
5104                                 "soft reset.\n");
5105                         goto clean4;
5106                 }
5107
5108                 rc = cciss_kdump_soft_reset(h);
5109                 if (rc) {
5110                         dev_warn(&h->pdev->dev, "Soft reset failed.\n");
5111                         goto clean4;
5112                 }
5113
5114                 dev_info(&h->pdev->dev, "Board READY.\n");
5115                 dev_info(&h->pdev->dev,
5116                         "Waiting for stale completions to drain.\n");
5117                 h->access.set_intr_mask(h, CCISS_INTR_ON);
5118                 msleep(10000);
5119                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5120
5121                 rc = controller_reset_failed(h->cfgtable);
5122                 if (rc)
5123                         dev_info(&h->pdev->dev,
5124                                 "Soft reset appears to have failed.\n");
5125
5126                 /* since the controller's reset, we have to go back and re-init
5127                  * everything.  Easiest to just forget what we've done and do it
5128                  * all over again.
5129                  */
5130                 cciss_undo_allocations_after_kdump_soft_reset(h);
5131                 try_soft_reset = 0;
5132                 if (rc)
5133                         /* don't go to clean4, we already unallocated */
5134                         return -ENODEV;
5135
5136                 goto reinit_after_soft_reset;
5137         }
5138
5139         cciss_scsi_setup(h);
5140
5141         /* Turn the interrupts on so we can service requests */
5142         h->access.set_intr_mask(h, CCISS_INTR_ON);
5143
5144         /* Get the firmware version */
5145         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
5146         if (inq_buff == NULL) {
5147                 dev_err(&h->pdev->dev, "out of memory\n");
5148                 goto clean4;
5149         }
5150
5151         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
5152                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
5153         if (return_code == IO_OK) {
5154                 h->firm_ver[0] = inq_buff->data_byte[32];
5155                 h->firm_ver[1] = inq_buff->data_byte[33];
5156                 h->firm_ver[2] = inq_buff->data_byte[34];
5157                 h->firm_ver[3] = inq_buff->data_byte[35];
5158         } else {         /* send command failed */
5159                 dev_warn(&h->pdev->dev, "unable to determine firmware"
5160                         " version of controller\n");
5161         }
5162         kfree(inq_buff);
5163
5164         cciss_procinit(h);
5165
5166         h->cciss_max_sectors = 8192;
5167
5168         rebuild_lun_table(h, 1, 0);
5169         cciss_engage_scsi(h);
5170         h->busy_initializing = 0;
5171         return 0;
5172
5173 clean4:
5174         cciss_free_cmd_pool(h);
5175         cciss_free_scatterlists(h);
5176         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5177         free_irq(h->intr[h->intr_mode], h);
5178 clean2:
5179         unregister_blkdev(h->major, h->devname);
5180 clean1:
5181         cciss_destroy_hba_sysfs_entry(h);
5182 clean0:
5183         pci_release_regions(pdev);
5184 clean_no_release_regions:
5185         h->busy_initializing = 0;
5186
5187         /*
5188          * Deliberately omit pci_disable_device(): it does something nasty to
5189          * Smart Array controllers that pci_enable_device does not undo
5190          */
5191         pci_set_drvdata(pdev, NULL);
5192         free_hba(h);
5193         return -ENODEV;
5194 }
5195
5196 static void cciss_shutdown(struct pci_dev *pdev)
5197 {
5198         ctlr_info_t *h;
5199         char *flush_buf;
5200         int return_code;
5201
5202         h = pci_get_drvdata(pdev);
5203         flush_buf = kzalloc(4, GFP_KERNEL);
5204         if (!flush_buf) {
5205                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
5206                 return;
5207         }
5208         /* write all data in the battery backed cache to disk */
5209         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
5210                 4, 0, CTLR_LUNID, TYPE_CMD);
5211         kfree(flush_buf);
5212         if (return_code != IO_OK)
5213                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
5214         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5215         free_irq(h->intr[h->intr_mode], h);
5216 }
5217
5218 static int cciss_enter_simple_mode(struct ctlr_info *h)
5219 {
5220         u32 trans_support;
5221
5222         trans_support = readl(&(h->cfgtable->TransportSupport));
5223         if (!(trans_support & SIMPLE_MODE))
5224                 return -ENOTSUPP;
5225
5226         h->max_commands = readl(&(h->cfgtable->CmdsOutMax));
5227         writel(CFGTBL_Trans_Simple, &(h->cfgtable->HostWrite.TransportRequest));
5228         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
5229         cciss_wait_for_mode_change_ack(h);
5230         print_cfg_table(h);
5231         if (!(readl(&(h->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
5232                 dev_warn(&h->pdev->dev, "unable to get board into simple mode\n");
5233                 return -ENODEV;
5234         }
5235         h->transMethod = CFGTBL_Trans_Simple;
5236         return 0;
5237 }
5238
5239
5240 static void cciss_remove_one(struct pci_dev *pdev)
5241 {
5242         ctlr_info_t *h;
5243         int i, j;
5244
5245         if (pci_get_drvdata(pdev) == NULL) {
5246                 dev_err(&pdev->dev, "Unable to remove device\n");
5247                 return;
5248         }
5249
5250         h = pci_get_drvdata(pdev);
5251         i = h->ctlr;
5252         if (hba[i] == NULL) {
5253                 dev_err(&pdev->dev, "device appears to already be removed\n");
5254                 return;
5255         }
5256
5257         mutex_lock(&h->busy_shutting_down);
5258
5259         remove_from_scan_list(h);
5260         remove_proc_entry(h->devname, proc_cciss);
5261         unregister_blkdev(h->major, h->devname);
5262
5263         /* remove it from the disk list */
5264         for (j = 0; j < CISS_MAX_LUN; j++) {
5265                 struct gendisk *disk = h->gendisk[j];
5266                 if (disk) {
5267                         struct request_queue *q = disk->queue;
5268
5269                         if (disk->flags & GENHD_FL_UP) {
5270                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
5271                                 del_gendisk(disk);
5272                         }
5273                         if (q)
5274                                 blk_cleanup_queue(q);
5275                 }
5276         }
5277
5278 #ifdef CONFIG_CISS_SCSI_TAPE
5279         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
5280 #endif
5281
5282         cciss_shutdown(pdev);
5283
5284         pci_free_irq_vectors(h->pdev);
5285
5286         iounmap(h->transtable);
5287         iounmap(h->cfgtable);
5288         iounmap(h->vaddr);
5289
5290         cciss_free_cmd_pool(h);
5291         /* Free up sg elements */
5292         for (j = 0; j < h->nr_cmds; j++)
5293                 kfree(h->scatter_list[j]);
5294         kfree(h->scatter_list);
5295         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5296         kfree(h->blockFetchTable);
5297         if (h->reply_pool)
5298                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
5299                                 h->reply_pool, h->reply_pool_dhandle);
5300         /*
5301          * Deliberately omit pci_disable_device(): it does something nasty to
5302          * Smart Array controllers that pci_enable_device does not undo
5303          */
5304         pci_release_regions(pdev);
5305         pci_set_drvdata(pdev, NULL);
5306         cciss_destroy_hba_sysfs_entry(h);
5307         mutex_unlock(&h->busy_shutting_down);
5308         free_hba(h);
5309 }
5310
5311 static struct pci_driver cciss_pci_driver = {
5312         .name = "cciss",
5313         .probe = cciss_init_one,
5314         .remove = cciss_remove_one,
5315         .id_table = cciss_pci_device_id,        /* id_table */
5316         .shutdown = cciss_shutdown,
5317 };
5318
5319 /*
5320  *  This is it.  Register the PCI driver information for the cards we control
5321  *  the OS will call our registered routines when it finds one of our cards.
5322  */
5323 static int __init cciss_init(void)
5324 {
5325         int err;
5326
5327         /*
5328          * The hardware requires that commands are aligned on a 64-bit
5329          * boundary. Given that we use pci_alloc_consistent() to allocate an
5330          * array of them, the size must be a multiple of 8 bytes.
5331          */
5332         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
5333         printk(KERN_INFO DRIVER_NAME "\n");
5334
5335         err = bus_register(&cciss_bus_type);
5336         if (err)
5337                 return err;
5338
5339         /* Start the scan thread */
5340         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
5341         if (IS_ERR(cciss_scan_thread)) {
5342                 err = PTR_ERR(cciss_scan_thread);
5343                 goto err_bus_unregister;
5344         }
5345
5346         /* Register for our PCI devices */
5347         err = pci_register_driver(&cciss_pci_driver);
5348         if (err)
5349                 goto err_thread_stop;
5350
5351         return err;
5352
5353 err_thread_stop:
5354         kthread_stop(cciss_scan_thread);
5355 err_bus_unregister:
5356         bus_unregister(&cciss_bus_type);
5357
5358         return err;
5359 }
5360
5361 static void __exit cciss_cleanup(void)
5362 {
5363         int i;
5364
5365         pci_unregister_driver(&cciss_pci_driver);
5366         /* double check that all controller entrys have been removed */
5367         for (i = 0; i < MAX_CTLR; i++) {
5368                 if (hba[i] != NULL) {
5369                         dev_warn(&hba[i]->pdev->dev,
5370                                 "had to remove controller\n");
5371                         cciss_remove_one(hba[i]->pdev);
5372                 }
5373         }
5374         kthread_stop(cciss_scan_thread);
5375         if (proc_cciss)
5376                 remove_proc_entry("driver/cciss", NULL);
5377         bus_unregister(&cciss_bus_type);
5378 }
5379
5380 module_init(cciss_init);
5381 module_exit(cciss_cleanup);