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 <asm/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 <linux/cdrom.h>
  56 #include <linux/scatterlist.h>
  57 #include <linux/kthread.h>
  58
  59 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  60 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  61 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  62
  63 /* Embedded module documentation macros - see modules.h */
  64 MODULE_AUTHOR("Hewlett-Packard Company");
  65 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  66 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  67 MODULE_VERSION("3.6.26");
  68 MODULE_LICENSE("GPL");
  69 static int cciss_tape_cmds = 6;
  70 module_param(cciss_tape_cmds, int, 0644);
  71 MODULE_PARM_DESC(cciss_tape_cmds,
  72         "number of commands to allocate for tape devices (default: 6)");
  73 static int cciss_simple_mode;
  74 module_param(cciss_simple_mode, int, S_IRUGO|S_IWUSR);
  75 MODULE_PARM_DESC(cciss_simple_mode,
  76         "Use 'simple mode' rather than 'performant mode'");
  77
  78 static int cciss_allow_hpsa;
  79 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  80 MODULE_PARM_DESC(cciss_allow_hpsa,
  81         "Prevent cciss driver from accessing hardware known to be "
  82         " supported by the hpsa driver");
  83
  84 static DEFINE_MUTEX(cciss_mutex);
  85 static struct proc_dir_entry *proc_cciss;
  86
  87 #include "cciss_cmd.h"
  88 #include "cciss.h"
  89 #include <linux/cciss_ioctl.h>
  90
  91 /* define the PCI info for the cards we can control */
  92 static const struct pci_device_id cciss_pci_device_id[] = {
  93         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  94         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  95         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  96         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  97         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  98         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  99         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
 100         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
 101         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
 110         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
 111         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 112         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 113         {0,}
 114 };
 115
 116 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 117
 118 /*  board_id = Subsystem Device ID & Vendor ID
 119  *  product = Marketing Name for the board
 120  *  access = Address of the struct of function pointers
 121  */
 122 static struct board_type products[] = {
 123         {0x40700E11, "Smart Array 5300", &SA5_access},
 124         {0x40800E11, "Smart Array 5i", &SA5B_access},
 125         {0x40820E11, "Smart Array 532", &SA5B_access},
 126         {0x40830E11, "Smart Array 5312", &SA5B_access},
 127         {0x409A0E11, "Smart Array 641", &SA5_access},
 128         {0x409B0E11, "Smart Array 642", &SA5_access},
 129         {0x409C0E11, "Smart Array 6400", &SA5_access},
 130         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 131         {0x40910E11, "Smart Array 6i", &SA5_access},
 132         {0x3225103C, "Smart Array P600", &SA5_access},
 133         {0x3223103C, "Smart Array P800", &SA5_access},
 134         {0x3234103C, "Smart Array P400", &SA5_access},
 135         {0x3235103C, "Smart Array P400i", &SA5_access},
 136         {0x3211103C, "Smart Array E200i", &SA5_access},
 137         {0x3212103C, "Smart Array E200", &SA5_access},
 138         {0x3213103C, "Smart Array E200i", &SA5_access},
 139         {0x3214103C, "Smart Array E200i", &SA5_access},
 140         {0x3215103C, "Smart Array E200i", &SA5_access},
 141         {0x3237103C, "Smart Array E500", &SA5_access},
 142         {0x323D103C, "Smart Array P700m", &SA5_access},
 143 };
 144
 145 /* How long to wait (in milliseconds) for board to go into simple mode */
 146 #define MAX_CONFIG_WAIT 30000
 147 #define MAX_IOCTL_CONFIG_WAIT 1000
 148
 149 /*define how many times we will try a command because of bus resets */
 150 #define MAX_CMD_RETRIES 3
 151
 152 #define MAX_CTLR        32
 153
 154 /* Originally cciss driver only supports 8 major numbers */
 155 #define MAX_CTLR_ORIG   8
 156
 157 static ctlr_info_t *hba[MAX_CTLR];
 158
 159 static struct task_struct *cciss_scan_thread;
 160 static DEFINE_MUTEX(scan_mutex);
 161 static LIST_HEAD(scan_q);
 162
 163 static void do_cciss_request(struct request_queue *q);
 164 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 165 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 166 static int cciss_open(struct block_device *bdev, fmode_t mode);
 167 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 168 static void cciss_release(struct gendisk *disk, fmode_t mode);
 169 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 170                        unsigned int cmd, unsigned long arg);
 171 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 172
 173 static int cciss_revalidate(struct gendisk *disk);
 174 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 175 static int deregister_disk(ctlr_info_t *h, int drv_index,
 176                            int clear_all, int via_ioctl);
 177
 178 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 179                         sector_t *total_size, unsigned int *block_size);
 180 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 181                         sector_t *total_size, unsigned int *block_size);
 182 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 183                         sector_t total_size,
 184                         unsigned int block_size, InquiryData_struct *inq_buff,
 185                                    drive_info_struct *drv);
 186 static void cciss_interrupt_mode(ctlr_info_t *);
 187 static int cciss_enter_simple_mode(struct ctlr_info *h);
 188 static void start_io(ctlr_info_t *h);
 189 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 190                         __u8 page_code, unsigned char scsi3addr[],
 191                         int cmd_type);
 192 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 193         int attempt_retry);
 194 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 195
 196 static int add_to_scan_list(struct ctlr_info *h);
 197 static int scan_thread(void *data);
 198 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 199 static void cciss_hba_release(struct device *dev);
 200 static void cciss_device_release(struct device *dev);
 201 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 202 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 203 static inline u32 next_command(ctlr_info_t *h);
 204 static int cciss_find_cfg_addrs(struct pci_dev *pdev, void __iomem *vaddr,
 205                                 u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 206                                 u64 *cfg_offset);
 207 static int cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 208                                      unsigned long *memory_bar);
 209 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag);
 210 static int write_driver_ver_to_cfgtable(CfgTable_struct __iomem *cfgtable);
 211
 212 /* performant mode helper functions */
 213 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 214                                 int *bucket_map);
 215 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 216
 217 #ifdef CONFIG_PROC_FS
 218 static void cciss_procinit(ctlr_info_t *h);
 219 #else
 220 static void cciss_procinit(ctlr_info_t *h)
 221 {
 222 }
 223 #endif                          /* CONFIG_PROC_FS */
 224
 225 #ifdef CONFIG_COMPAT
 226 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 227                               unsigned, unsigned long);
 228 #endif
 229
 230 static const struct block_device_operations cciss_fops = {
 231         .owner = THIS_MODULE,
 232         .open = cciss_unlocked_open,
 233         .release = cciss_release,
 234         .ioctl = cciss_ioctl,
 235         .getgeo = cciss_getgeo,
 236 #ifdef CONFIG_COMPAT
 237         .compat_ioctl = cciss_compat_ioctl,
 238 #endif
 239         .revalidate_disk = cciss_revalidate,
 240 };
 241
 242 /* set_performant_mode: Modify the tag for cciss performant
 243  * set bit 0 for pull model, bits 3-1 for block fetch
 244  * register number
 245  */
 246 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 247 {
 248         if (likely(h->transMethod & CFGTBL_Trans_Performant))
 249                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 250 }
 251
 252 /*
 253  * Enqueuing and dequeuing functions for cmdlists.
 254  */
 255 static inline void addQ(struct list_head *list, CommandList_struct *c)
 256 {
 257         list_add_tail(&c->list, list);
 258 }
 259
 260 static inline void removeQ(CommandList_struct *c)
 261 {
 262         /*
 263          * After kexec/dump some commands might still
 264          * be in flight, which the firmware will try
 265          * to complete. Resetting the firmware doesn't work
 266          * with old fw revisions, so we have to mark
 267          * them off as 'stale' to prevent the driver from
 268          * falling over.
 269          */
 270         if (WARN_ON(list_empty(&c->list))) {
 271                 c->cmd_type = CMD_MSG_STALE;
 272                 return;
 273         }
 274
 275         list_del_init(&c->list);
 276 }
 277
 278 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 279         CommandList_struct *c)
 280 {
 281         unsigned long flags;
 282         set_performant_mode(h, c);
 283         spin_lock_irqsave(&h->lock, flags);
 284         addQ(&h->reqQ, c);
 285         h->Qdepth++;
 286         if (h->Qdepth > h->maxQsinceinit)
 287                 h->maxQsinceinit = h->Qdepth;
 288         start_io(h);
 289         spin_unlock_irqrestore(&h->lock, flags);
 290 }
 291
 292 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 293         int nr_cmds)
 294 {
 295         int i;
 296
 297         if (!cmd_sg_list)
 298                 return;
 299         for (i = 0; i < nr_cmds; i++) {
 300                 kfree(cmd_sg_list[i]);
 301                 cmd_sg_list[i] = NULL;
 302         }
 303         kfree(cmd_sg_list);
 304 }
 305
 306 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 307         ctlr_info_t *h, int chainsize, int nr_cmds)
 308 {
 309         int j;
 310         SGDescriptor_struct **cmd_sg_list;
 311
 312         if (chainsize <= 0)
 313                 return NULL;
 314
 315         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 316         if (!cmd_sg_list)
 317                 return NULL;
 318
 319         /* Build up chain blocks for each command */
 320         for (j = 0; j < nr_cmds; j++) {
 321                 /* Need a block of chainsized s/g elements. */
 322                 cmd_sg_list[j] = kmalloc((chainsize *
 323                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 324                 if (!cmd_sg_list[j]) {
 325                         dev_err(&h->pdev->dev, "Cannot get memory "
 326                                 "for s/g chains.\n");
 327                         goto clean;
 328                 }
 329         }
 330         return cmd_sg_list;
 331 clean:
 332         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 333         return NULL;
 334 }
 335
 336 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 337 {
 338         SGDescriptor_struct *chain_sg;
 339         u64bit temp64;
 340
 341         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 342                 return;
 343
 344         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 345         temp64.val32.lower = chain_sg->Addr.lower;
 346         temp64.val32.upper = chain_sg->Addr.upper;
 347         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 348 }
 349
 350 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 351         SGDescriptor_struct *chain_block, int len)
 352 {
 353         SGDescriptor_struct *chain_sg;
 354         u64bit temp64;
 355
 356         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 357         chain_sg->Ext = CCISS_SG_CHAIN;
 358         chain_sg->Len = len;
 359         temp64.val = pci_map_single(h->pdev, chain_block, len,
 360                                 PCI_DMA_TODEVICE);
 361         chain_sg->Addr.lower = temp64.val32.lower;
 362         chain_sg->Addr.upper = temp64.val32.upper;
 363 }
 364
 365 #include "cciss_scsi.c"         /* For SCSI tape support */
 366
 367 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 368         "UNKNOWN"
 369 };
 370 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 371
 372 #ifdef CONFIG_PROC_FS
 373
 374 /*
 375  * Report information about this controller.
 376  */
 377 #define ENG_GIG 1000000000
 378 #define ENG_GIG_FACTOR (ENG_GIG/512)
 379 #define ENGAGE_SCSI     "engage scsi"
 380
 381 static void cciss_seq_show_header(struct seq_file *seq)
 382 {
 383         ctlr_info_t *h = seq->private;
 384
 385         seq_printf(seq, "%s: HP %s Controller\n"
 386                 "Board ID: 0x%08lx\n"
 387                 "Firmware Version: %c%c%c%c\n"
 388                 "IRQ: %d\n"
 389                 "Logical drives: %d\n"
 390                 "Current Q depth: %d\n"
 391                 "Current # commands on controller: %d\n"
 392                 "Max Q depth since init: %d\n"
 393                 "Max # commands on controller since init: %d\n"
 394                 "Max SG entries since init: %d\n",
 395                 h->devname,
 396                 h->product_name,
 397                 (unsigned long)h->board_id,
 398                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 399                 h->firm_ver[3], (unsigned int)h->intr[h->intr_mode],
 400                 h->num_luns,
 401                 h->Qdepth, h->commands_outstanding,
 402                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 403
 404 #ifdef CONFIG_CISS_SCSI_TAPE
 405         cciss_seq_tape_report(seq, h);
 406 #endif /* CONFIG_CISS_SCSI_TAPE */
 407 }
 408
 409 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 410 {
 411         ctlr_info_t *h = seq->private;
 412         unsigned long flags;
 413
 414         /* prevent displaying bogus info during configuration
 415          * or deconfiguration of a logical volume
 416          */
 417         spin_lock_irqsave(&h->lock, flags);
 418         if (h->busy_configuring) {
 419                 spin_unlock_irqrestore(&h->lock, flags);
 420                 return ERR_PTR(-EBUSY);
 421         }
 422         h->busy_configuring = 1;
 423         spin_unlock_irqrestore(&h->lock, flags);
 424
 425         if (*pos == 0)
 426                 cciss_seq_show_header(seq);
 427
 428         return pos;
 429 }
 430
 431 static int cciss_seq_show(struct seq_file *seq, void *v)
 432 {
 433         sector_t vol_sz, vol_sz_frac;
 434         ctlr_info_t *h = seq->private;
 435         unsigned ctlr = h->ctlr;
 436         loff_t *pos = v;
 437         drive_info_struct *drv = h->drv[*pos];
 438
 439         if (*pos > h->highest_lun)
 440                 return 0;
 441
 442         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 443                 return 0;
 444
 445         if (drv->heads == 0)
 446                 return 0;
 447
 448         vol_sz = drv->nr_blocks;
 449         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 450         vol_sz_frac *= 100;
 451         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 452
 453         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 454                 drv->raid_level = RAID_UNKNOWN;
 455         seq_printf(seq, "cciss/c%dd%d:"
 456                         "\t%4u.%02uGB\tRAID %s\n",
 457                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 458                         raid_label[drv->raid_level]);
 459         return 0;
 460 }
 461
 462 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 463 {
 464         ctlr_info_t *h = seq->private;
 465
 466         if (*pos > h->highest_lun)
 467                 return NULL;
 468         *pos += 1;
 469
 470         return pos;
 471 }
 472
 473 static void cciss_seq_stop(struct seq_file *seq, void *v)
 474 {
 475         ctlr_info_t *h = seq->private;
 476
 477         /* Only reset h->busy_configuring if we succeeded in setting
 478          * it during cciss_seq_start. */
 479         if (v == ERR_PTR(-EBUSY))
 480                 return;
 481
 482         h->busy_configuring = 0;
 483 }
 484
 485 static const struct seq_operations cciss_seq_ops = {
 486         .start = cciss_seq_start,
 487         .show  = cciss_seq_show,
 488         .next  = cciss_seq_next,
 489         .stop  = cciss_seq_stop,
 490 };
 491
 492 static int cciss_seq_open(struct inode *inode, struct file *file)
 493 {
 494         int ret = seq_open(file, &cciss_seq_ops);
 495         struct seq_file *seq = file->private_data;
 496
 497         if (!ret)
 498                 seq->private = PDE_DATA(inode);
 499
 500         return ret;
 501 }
 502
 503 static ssize_t
 504 cciss_proc_write(struct file *file, const char __user *buf,
 505                  size_t length, loff_t *ppos)
 506 {
 507         int err;
 508         char *buffer;
 509
 510 #ifndef CONFIG_CISS_SCSI_TAPE
 511         return -EINVAL;
 512 #endif
 513
 514         if (!buf || length > PAGE_SIZE - 1)
 515                 return -EINVAL;
 516
 517         buffer = (char *)__get_free_page(GFP_KERNEL);
 518         if (!buffer)
 519                 return -ENOMEM;
 520
 521         err = -EFAULT;
 522         if (copy_from_user(buffer, buf, length))
 523                 goto out;
 524         buffer[length] = '\0';
 525
 526 #ifdef CONFIG_CISS_SCSI_TAPE
 527         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 528                 struct seq_file *seq = file->private_data;
 529                 ctlr_info_t *h = seq->private;
 530
 531                 err = cciss_engage_scsi(h);
 532                 if (err == 0)
 533                         err = length;
 534         } else
 535 #endif /* CONFIG_CISS_SCSI_TAPE */
 536                 err = -EINVAL;
 537         /* might be nice to have "disengage" too, but it's not
 538            safely possible. (only 1 module use count, lock issues.) */
 539
 540 out:
 541         free_page((unsigned long)buffer);
 542         return err;
 543 }
 544
 545 static const struct file_operations cciss_proc_fops = {
 546         .owner   = THIS_MODULE,
 547         .open    = cciss_seq_open,
 548         .read    = seq_read,
 549         .llseek  = seq_lseek,
 550         .release = seq_release,
 551         .write   = cciss_proc_write,
 552 };
 553
 554 static void cciss_procinit(ctlr_info_t *h)
 555 {
 556         struct proc_dir_entry *pde;
 557
 558         if (proc_cciss == NULL)
 559                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 560         if (!proc_cciss)
 561                 return;
 562         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 563                                         S_IROTH, proc_cciss,
 564                                         &cciss_proc_fops, h);
 565 }
 566 #endif                          /* CONFIG_PROC_FS */
 567
 568 #define MAX_PRODUCT_NAME_LEN 19
 569
 570 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 571 #define to_drv(n) container_of(n, drive_info_struct, dev)
 572
 573 /* List of controllers which cannot be hard reset on kexec with reset_devices */
 574 static u32 unresettable_controller[] = {
 575         0x3223103C, /* Smart Array P800 */
 576         0x3234103C, /* Smart Array P400 */
 577         0x3235103C, /* Smart Array P400i */
 578         0x3211103C, /* Smart Array E200i */
 579         0x3212103C, /* Smart Array E200 */
 580         0x3213103C, /* Smart Array E200i */
 581         0x3214103C, /* Smart Array E200i */
 582         0x3215103C, /* Smart Array E200i */
 583         0x3237103C, /* Smart Array E500 */
 584         0x323D103C, /* Smart Array P700m */
 585         0x40800E11, /* Smart Array 5i */
 586         0x409C0E11, /* Smart Array 6400 */
 587         0x409D0E11, /* Smart Array 6400 EM */
 588         0x40700E11, /* Smart Array 5300 */
 589         0x40820E11, /* Smart Array 532 */
 590         0x40830E11, /* Smart Array 5312 */
 591         0x409A0E11, /* Smart Array 641 */
 592         0x409B0E11, /* Smart Array 642 */
 593         0x40910E11, /* Smart Array 6i */
 594 };
 595
 596 /* List of controllers which cannot even be soft reset */
 597 static u32 soft_unresettable_controller[] = {
 598         0x40800E11, /* Smart Array 5i */
 599         0x40700E11, /* Smart Array 5300 */
 600         0x40820E11, /* Smart Array 532 */
 601         0x40830E11, /* Smart Array 5312 */
 602         0x409A0E11, /* Smart Array 641 */
 603         0x409B0E11, /* Smart Array 642 */
 604         0x40910E11, /* Smart Array 6i */
 605         /* Exclude 640x boards.  These are two pci devices in one slot
 606          * which share a battery backed cache module.  One controls the
 607          * cache, the other accesses the cache through the one that controls
 608          * it.  If we reset the one controlling the cache, the other will
 609          * likely not be happy.  Just forbid resetting this conjoined mess.
 610          */
 611         0x409C0E11, /* Smart Array 6400 */
 612         0x409D0E11, /* Smart Array 6400 EM */
 613 };
 614
 615 static int ctlr_is_hard_resettable(u32 board_id)
 616 {
 617         int i;
 618
 619         for (i = 0; i < ARRAY_SIZE(unresettable_controller); i++)
 620                 if (unresettable_controller[i] == board_id)
 621                         return 0;
 622         return 1;
 623 }
 624
 625 static int ctlr_is_soft_resettable(u32 board_id)
 626 {
 627         int i;
 628
 629         for (i = 0; i < ARRAY_SIZE(soft_unresettable_controller); i++)
 630                 if (soft_unresettable_controller[i] == board_id)
 631                         return 0;
 632         return 1;
 633 }
 634
 635 static int ctlr_is_resettable(u32 board_id)
 636 {
 637         return ctlr_is_hard_resettable(board_id) ||
 638                 ctlr_is_soft_resettable(board_id);
 639 }
 640
 641 static ssize_t host_show_resettable(struct device *dev,
 642                                     struct device_attribute *attr,
 643                                     char *buf)
 644 {
 645         struct ctlr_info *h = to_hba(dev);
 646
 647         return snprintf(buf, 20, "%d\n", ctlr_is_resettable(h->board_id));
 648 }
 649 static DEVICE_ATTR(resettable, S_IRUGO, host_show_resettable, NULL);
 650
 651 static ssize_t host_store_rescan(struct device *dev,
 652                                  struct device_attribute *attr,
 653                                  const char *buf, size_t count)
 654 {
 655         struct ctlr_info *h = to_hba(dev);
 656
 657         add_to_scan_list(h);
 658         wake_up_process(cciss_scan_thread);
 659         wait_for_completion_interruptible(&h->scan_wait);
 660
 661         return count;
 662 }
 663 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 664
 665 static ssize_t host_show_transport_mode(struct device *dev,
 666                                  struct device_attribute *attr,
 667                                  char *buf)
 668 {
 669         struct ctlr_info *h = to_hba(dev);
 670
 671         return snprintf(buf, 20, "%s\n",
 672                 h->transMethod & CFGTBL_Trans_Performant ?
 673                         "performant" : "simple");
 674 }
 675 static DEVICE_ATTR(transport_mode, S_IRUGO, host_show_transport_mode, NULL);
 676
 677 static ssize_t dev_show_unique_id(struct device *dev,
 678                                  struct device_attribute *attr,
 679                                  char *buf)
 680 {
 681         drive_info_struct *drv = to_drv(dev);
 682         struct ctlr_info *h = to_hba(drv->dev.parent);
 683         __u8 sn[16];
 684         unsigned long flags;
 685         int ret = 0;
 686
 687         spin_lock_irqsave(&h->lock, flags);
 688         if (h->busy_configuring)
 689                 ret = -EBUSY;
 690         else
 691                 memcpy(sn, drv->serial_no, sizeof(sn));
 692         spin_unlock_irqrestore(&h->lock, flags);
 693
 694         if (ret)
 695                 return ret;
 696         else
 697                 return snprintf(buf, 16 * 2 + 2,
 698                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 699                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 700                                 sn[0], sn[1], sn[2], sn[3],
 701                                 sn[4], sn[5], sn[6], sn[7],
 702                                 sn[8], sn[9], sn[10], sn[11],
 703                                 sn[12], sn[13], sn[14], sn[15]);
 704 }
 705 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 706
 707 static ssize_t dev_show_vendor(struct device *dev,
 708                                struct device_attribute *attr,
 709                                char *buf)
 710 {
 711         drive_info_struct *drv = to_drv(dev);
 712         struct ctlr_info *h = to_hba(drv->dev.parent);
 713         char vendor[VENDOR_LEN + 1];
 714         unsigned long flags;
 715         int ret = 0;
 716
 717         spin_lock_irqsave(&h->lock, flags);
 718         if (h->busy_configuring)
 719                 ret = -EBUSY;
 720         else
 721                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 722         spin_unlock_irqrestore(&h->lock, flags);
 723
 724         if (ret)
 725                 return ret;
 726         else
 727                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 728 }
 729 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 730
 731 static ssize_t dev_show_model(struct device *dev,
 732                               struct device_attribute *attr,
 733                               char *buf)
 734 {
 735         drive_info_struct *drv = to_drv(dev);
 736         struct ctlr_info *h = to_hba(drv->dev.parent);
 737         char model[MODEL_LEN + 1];
 738         unsigned long flags;
 739         int ret = 0;
 740
 741         spin_lock_irqsave(&h->lock, flags);
 742         if (h->busy_configuring)
 743                 ret = -EBUSY;
 744         else
 745                 memcpy(model, drv->model, MODEL_LEN + 1);
 746         spin_unlock_irqrestore(&h->lock, flags);
 747
 748         if (ret)
 749                 return ret;
 750         else
 751                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 752 }
 753 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 754
 755 static ssize_t dev_show_rev(struct device *dev,
 756                             struct device_attribute *attr,
 757                             char *buf)
 758 {
 759         drive_info_struct *drv = to_drv(dev);
 760         struct ctlr_info *h = to_hba(drv->dev.parent);
 761         char rev[REV_LEN + 1];
 762         unsigned long flags;
 763         int ret = 0;
 764
 765         spin_lock_irqsave(&h->lock, flags);
 766         if (h->busy_configuring)
 767                 ret = -EBUSY;
 768         else
 769                 memcpy(rev, drv->rev, REV_LEN + 1);
 770         spin_unlock_irqrestore(&h->lock, flags);
 771
 772         if (ret)
 773                 return ret;
 774         else
 775                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 776 }
 777 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 778
 779 static ssize_t cciss_show_lunid(struct device *dev,
 780                                 struct device_attribute *attr, char *buf)
 781 {
 782         drive_info_struct *drv = to_drv(dev);
 783         struct ctlr_info *h = to_hba(drv->dev.parent);
 784         unsigned long flags;
 785         unsigned char lunid[8];
 786
 787         spin_lock_irqsave(&h->lock, flags);
 788         if (h->busy_configuring) {
 789                 spin_unlock_irqrestore(&h->lock, flags);
 790                 return -EBUSY;
 791         }
 792         if (!drv->heads) {
 793                 spin_unlock_irqrestore(&h->lock, flags);
 794                 return -ENOTTY;
 795         }
 796         memcpy(lunid, drv->LunID, sizeof(lunid));
 797         spin_unlock_irqrestore(&h->lock, flags);
 798         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 799                 lunid[0], lunid[1], lunid[2], lunid[3],
 800                 lunid[4], lunid[5], lunid[6], lunid[7]);
 801 }
 802 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 803
 804 static ssize_t cciss_show_raid_level(struct device *dev,
 805                                      struct device_attribute *attr, char *buf)
 806 {
 807         drive_info_struct *drv = to_drv(dev);
 808         struct ctlr_info *h = to_hba(drv->dev.parent);
 809         int raid;
 810         unsigned long flags;
 811
 812         spin_lock_irqsave(&h->lock, flags);
 813         if (h->busy_configuring) {
 814                 spin_unlock_irqrestore(&h->lock, flags);
 815                 return -EBUSY;
 816         }
 817         raid = drv->raid_level;
 818         spin_unlock_irqrestore(&h->lock, flags);
 819         if (raid < 0 || raid > RAID_UNKNOWN)
 820                 raid = RAID_UNKNOWN;
 821
 822         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 823                         raid_label[raid]);
 824 }
 825 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 826
 827 static ssize_t cciss_show_usage_count(struct device *dev,
 828                                       struct device_attribute *attr, char *buf)
 829 {
 830         drive_info_struct *drv = to_drv(dev);
 831         struct ctlr_info *h = to_hba(drv->dev.parent);
 832         unsigned long flags;
 833         int count;
 834
 835         spin_lock_irqsave(&h->lock, flags);
 836         if (h->busy_configuring) {
 837                 spin_unlock_irqrestore(&h->lock, flags);
 838                 return -EBUSY;
 839         }
 840         count = drv->usage_count;
 841         spin_unlock_irqrestore(&h->lock, flags);
 842         return snprintf(buf, 20, "%d\n", count);
 843 }
 844 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 845
 846 static struct attribute *cciss_host_attrs[] = {
 847         &dev_attr_rescan.attr,
 848         &dev_attr_resettable.attr,
 849         &dev_attr_transport_mode.attr,
 850         NULL
 851 };
 852
 853 static struct attribute_group cciss_host_attr_group = {
 854         .attrs = cciss_host_attrs,
 855 };
 856
 857 static const struct attribute_group *cciss_host_attr_groups[] = {
 858         &cciss_host_attr_group,
 859         NULL
 860 };
 861
 862 static struct device_type cciss_host_type = {
 863         .name           = "cciss_host",
 864         .groups         = cciss_host_attr_groups,
 865         .release        = cciss_hba_release,
 866 };
 867
 868 static struct attribute *cciss_dev_attrs[] = {
 869         &dev_attr_unique_id.attr,
 870         &dev_attr_model.attr,
 871         &dev_attr_vendor.attr,
 872         &dev_attr_rev.attr,
 873         &dev_attr_lunid.attr,
 874         &dev_attr_raid_level.attr,
 875         &dev_attr_usage_count.attr,
 876         NULL
 877 };
 878
 879 static struct attribute_group cciss_dev_attr_group = {
 880         .attrs = cciss_dev_attrs,
 881 };
 882
 883 static const struct attribute_group *cciss_dev_attr_groups[] = {
 884         &cciss_dev_attr_group,
 885         NULL
 886 };
 887
 888 static struct device_type cciss_dev_type = {
 889         .name           = "cciss_device",
 890         .groups         = cciss_dev_attr_groups,
 891         .release        = cciss_device_release,
 892 };
 893
 894 static struct bus_type cciss_bus_type = {
 895         .name           = "cciss",
 896 };
 897
 898 /*
 899  * cciss_hba_release is called when the reference count
 900  * of h->dev goes to zero.
 901  */
 902 static void cciss_hba_release(struct device *dev)
 903 {
 904         /*
 905          * nothing to do, but need this to avoid a warning
 906          * about not having a release handler from lib/kref.c.
 907          */
 908 }
 909
 910 /*
 911  * Initialize sysfs entry for each controller.  This sets up and registers
 912  * the 'cciss#' directory for each individual controller under
 913  * /sys/bus/pci/devices/<dev>/.
 914  */
 915 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 916 {
 917         device_initialize(&h->dev);
 918         h->dev.type = &cciss_host_type;
 919         h->dev.bus = &cciss_bus_type;
 920         dev_set_name(&h->dev, "%s", h->devname);
 921         h->dev.parent = &h->pdev->dev;
 922
 923         return device_add(&h->dev);
 924 }
 925
 926 /*
 927  * Remove sysfs entries for an hba.
 928  */
 929 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 930 {
 931         device_del(&h->dev);
 932         put_device(&h->dev); /* final put. */
 933 }
 934
 935 /* cciss_device_release is called when the reference count
 936  * of h->drv[x]dev goes to zero.
 937  */
 938 static void cciss_device_release(struct device *dev)
 939 {
 940         drive_info_struct *drv = to_drv(dev);
 941         kfree(drv);
 942 }
 943
 944 /*
 945  * Initialize sysfs for each logical drive.  This sets up and registers
 946  * the 'c#d#' directory for each individual logical drive under
 947  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 948  * /sys/block/cciss!c#d# to this entry.
 949  */
 950 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 951                                        int drv_index)
 952 {
 953         struct device *dev;
 954
 955         if (h->drv[drv_index]->device_initialized)
 956                 return 0;
 957
 958         dev = &h->drv[drv_index]->dev;
 959         device_initialize(dev);
 960         dev->type = &cciss_dev_type;
 961         dev->bus = &cciss_bus_type;
 962         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 963         dev->parent = &h->dev;
 964         h->drv[drv_index]->device_initialized = 1;
 965         return device_add(dev);
 966 }
 967
 968 /*
 969  * Remove sysfs entries for a logical drive.
 970  */
 971 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 972         int ctlr_exiting)
 973 {
 974         struct device *dev = &h->drv[drv_index]->dev;
 975
 976         /* special case for c*d0, we only destroy it on controller exit */
 977         if (drv_index == 0 && !ctlr_exiting)
 978                 return;
 979
 980         device_del(dev);
 981         put_device(dev); /* the "final" put. */
 982         h->drv[drv_index] = NULL;
 983 }
 984
 985 /*
 986  * For operations that cannot sleep, a command block is allocated at init,
 987  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 988  * which ones are free or in use.
 989  */
 990 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 991 {
 992         CommandList_struct *c;
 993         int i;
 994         u64bit temp64;
 995         dma_addr_t cmd_dma_handle, err_dma_handle;
 996
 997         do {
 998                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 999                 if (i == h->nr_cmds)
1000                         return NULL;
1001         } while (test_and_set_bit(i, h->cmd_pool_bits) != 0);
1002         c = h->cmd_pool + i;
1003         memset(c, 0, sizeof(CommandList_struct));
1004         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
1005         c->err_info = h->errinfo_pool + i;
1006         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
1007         err_dma_handle = h->errinfo_pool_dhandle
1008             + i * sizeof(ErrorInfo_struct);
1009         h->nr_allocs++;
1010
1011         c->cmdindex = i;
1012
1013         INIT_LIST_HEAD(&c->list);
1014         c->busaddr = (__u32) cmd_dma_handle;
1015         temp64.val = (__u64) err_dma_handle;
1016         c->ErrDesc.Addr.lower = temp64.val32.lower;
1017         c->ErrDesc.Addr.upper = temp64.val32.upper;
1018         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
1019
1020         c->ctlr = h->ctlr;
1021         return c;
1022 }
1023
1024 /* allocate a command using pci_alloc_consistent, used for ioctls,
1025  * etc., not for the main i/o path.
1026  */
1027 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
1028 {
1029         CommandList_struct *c;
1030         u64bit temp64;
1031         dma_addr_t cmd_dma_handle, err_dma_handle;
1032
1033         c = pci_zalloc_consistent(h->pdev, sizeof(CommandList_struct),
1034                                   &cmd_dma_handle);
1035         if (c == NULL)
1036                 return NULL;
1037
1038         c->cmdindex = -1;
1039
1040         c->err_info = pci_zalloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
1041                                             &err_dma_handle);
1042
1043         if (c->err_info == NULL) {
1044                 pci_free_consistent(h->pdev,
1045                         sizeof(CommandList_struct), c, cmd_dma_handle);
1046                 return NULL;
1047         }
1048
1049         INIT_LIST_HEAD(&c->list);
1050         c->busaddr = (__u32) cmd_dma_handle;
1051         temp64.val = (__u64) err_dma_handle;
1052         c->ErrDesc.Addr.lower = temp64.val32.lower;
1053         c->ErrDesc.Addr.upper = temp64.val32.upper;
1054         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
1055
1056         c->ctlr = h->ctlr;
1057         return c;
1058 }
1059
1060 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
1061 {
1062         int i;
1063
1064         i = c - h->cmd_pool;
1065         clear_bit(i, h->cmd_pool_bits);
1066         h->nr_frees++;
1067 }
1068
1069 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
1070 {
1071         u64bit temp64;
1072
1073         temp64.val32.lower = c->ErrDesc.Addr.lower;
1074         temp64.val32.upper = c->ErrDesc.Addr.upper;
1075         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1076                             c->err_info, (dma_addr_t) temp64.val);
1077         pci_free_consistent(h->pdev, sizeof(CommandList_struct), c,
1078                 (dma_addr_t) cciss_tag_discard_error_bits(h, (u32) c->busaddr));
1079 }
1080
1081 static inline ctlr_info_t *get_host(struct gendisk *disk)
1082 {
1083         return disk->queue->queuedata;
1084 }
1085
1086 static inline drive_info_struct *get_drv(struct gendisk *disk)
1087 {
1088         return disk->private_data;
1089 }
1090
1091 /*
1092  * Open.  Make sure the device is really there.
1093  */
1094 static int cciss_open(struct block_device *bdev, fmode_t mode)
1095 {
1096         ctlr_info_t *h = get_host(bdev->bd_disk);
1097         drive_info_struct *drv = get_drv(bdev->bd_disk);
1098
1099         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1100         if (drv->busy_configuring)
1101                 return -EBUSY;
1102         /*
1103          * Root is allowed to open raw volume zero even if it's not configured
1104          * so array config can still work. Root is also allowed to open any
1105          * volume that has a LUN ID, so it can issue IOCTL to reread the
1106          * disk information.  I don't think I really like this
1107          * but I'm already using way to many device nodes to claim another one
1108          * for "raw controller".
1109          */
1110         if (drv->heads == 0) {
1111                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1112                         /* if not node 0 make sure it is a partition = 0 */
1113                         if (MINOR(bdev->bd_dev) & 0x0f) {
1114                                 return -ENXIO;
1115                                 /* if it is, make sure we have a LUN ID */
1116                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1117                                 sizeof(drv->LunID))) {
1118                                 return -ENXIO;
1119                         }
1120                 }
1121                 if (!capable(CAP_SYS_ADMIN))
1122                         return -EPERM;
1123         }
1124         drv->usage_count++;
1125         h->usage_count++;
1126         return 0;
1127 }
1128
1129 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1130 {
1131         int ret;
1132
1133         mutex_lock(&cciss_mutex);
1134         ret = cciss_open(bdev, mode);
1135         mutex_unlock(&cciss_mutex);
1136
1137         return ret;
1138 }
1139
1140 /*
1141  * Close.  Sync first.
1142  */
1143 static void cciss_release(struct gendisk *disk, fmode_t mode)
1144 {
1145         ctlr_info_t *h;
1146         drive_info_struct *drv;
1147
1148         mutex_lock(&cciss_mutex);
1149         h = get_host(disk);
1150         drv = get_drv(disk);
1151         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1152         drv->usage_count--;
1153         h->usage_count--;
1154         mutex_unlock(&cciss_mutex);
1155 }
1156
1157 #ifdef CONFIG_COMPAT
1158
1159 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1160                                   unsigned cmd, unsigned long arg);
1161 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1162                                       unsigned cmd, unsigned long arg);
1163
1164 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1165                               unsigned cmd, unsigned long arg)
1166 {
1167         switch (cmd) {
1168         case CCISS_GETPCIINFO:
1169         case CCISS_GETINTINFO:
1170         case CCISS_SETINTINFO:
1171         case CCISS_GETNODENAME:
1172         case CCISS_SETNODENAME:
1173         case CCISS_GETHEARTBEAT:
1174         case CCISS_GETBUSTYPES:
1175         case CCISS_GETFIRMVER:
1176         case CCISS_GETDRIVVER:
1177         case CCISS_REVALIDVOLS:
1178         case CCISS_DEREGDISK:
1179         case CCISS_REGNEWDISK:
1180         case CCISS_REGNEWD:
1181         case CCISS_RESCANDISK:
1182         case CCISS_GETLUNINFO:
1183                 return cciss_ioctl(bdev, mode, cmd, arg);
1184
1185         case CCISS_PASSTHRU32:
1186                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1187         case CCISS_BIG_PASSTHRU32:
1188                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1189
1190         default:
1191                 return -ENOIOCTLCMD;
1192         }
1193 }
1194
1195 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1196                                   unsigned cmd, unsigned long arg)
1197 {
1198         IOCTL32_Command_struct __user *arg32 =
1199             (IOCTL32_Command_struct __user *) arg;
1200         IOCTL_Command_struct arg64;
1201         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1202         int err;
1203         u32 cp;
1204
1205         memset(&arg64, 0, sizeof(arg64));
1206         err = 0;
1207         err |=
1208             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1209                            sizeof(arg64.LUN_info));
1210         err |=
1211             copy_from_user(&arg64.Request, &arg32->Request,
1212                            sizeof(arg64.Request));
1213         err |=
1214             copy_from_user(&arg64.error_info, &arg32->error_info,
1215                            sizeof(arg64.error_info));
1216         err |= get_user(arg64.buf_size, &arg32->buf_size);
1217         err |= get_user(cp, &arg32->buf);
1218         arg64.buf = compat_ptr(cp);
1219         err |= copy_to_user(p, &arg64, sizeof(arg64));
1220
1221         if (err)
1222                 return -EFAULT;
1223
1224         err = cciss_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1225         if (err)
1226                 return err;
1227         err |=
1228             copy_in_user(&arg32->error_info, &p->error_info,
1229                          sizeof(arg32->error_info));
1230         if (err)
1231                 return -EFAULT;
1232         return err;
1233 }
1234
1235 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1236                                       unsigned cmd, unsigned long arg)
1237 {
1238         BIG_IOCTL32_Command_struct __user *arg32 =
1239             (BIG_IOCTL32_Command_struct __user *) arg;
1240         BIG_IOCTL_Command_struct arg64;
1241         BIG_IOCTL_Command_struct __user *p =
1242             compat_alloc_user_space(sizeof(arg64));
1243         int err;
1244         u32 cp;
1245
1246         memset(&arg64, 0, sizeof(arg64));
1247         err = 0;
1248         err |=
1249             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1250                            sizeof(arg64.LUN_info));
1251         err |=
1252             copy_from_user(&arg64.Request, &arg32->Request,
1253                            sizeof(arg64.Request));
1254         err |=
1255             copy_from_user(&arg64.error_info, &arg32->error_info,
1256                            sizeof(arg64.error_info));
1257         err |= get_user(arg64.buf_size, &arg32->buf_size);
1258         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1259         err |= get_user(cp, &arg32->buf);
1260         arg64.buf = compat_ptr(cp);
1261         err |= copy_to_user(p, &arg64, sizeof(arg64));
1262
1263         if (err)
1264                 return -EFAULT;
1265
1266         err = cciss_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1267         if (err)
1268                 return err;
1269         err |=
1270             copy_in_user(&arg32->error_info, &p->error_info,
1271                          sizeof(arg32->error_info));
1272         if (err)
1273                 return -EFAULT;
1274         return err;
1275 }
1276 #endif
1277
1278 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1279 {
1280         drive_info_struct *drv = get_drv(bdev->bd_disk);
1281
1282         if (!drv->cylinders)
1283                 return -ENXIO;
1284
1285         geo->heads = drv->heads;
1286         geo->sectors = drv->sectors;
1287         geo->cylinders = drv->cylinders;
1288         return 0;
1289 }
1290
1291 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1292 {
1293         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1294                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1295                 (void)check_for_unit_attention(h, c);
1296 }
1297
1298 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1299 {
1300         cciss_pci_info_struct pciinfo;
1301
1302         if (!argp)
1303                 return -EINVAL;
1304         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1305         pciinfo.bus = h->pdev->bus->number;
1306         pciinfo.dev_fn = h->pdev->devfn;
1307         pciinfo.board_id = h->board_id;
1308         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1309                 return -EFAULT;
1310         return 0;
1311 }
1312
1313 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1314 {
1315         cciss_coalint_struct intinfo;
1316         unsigned long flags;
1317
1318         if (!argp)
1319                 return -EINVAL;
1320         spin_lock_irqsave(&h->lock, flags);
1321         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1322         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1323         spin_unlock_irqrestore(&h->lock, flags);
1324         if (copy_to_user
1325             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1326                 return -EFAULT;
1327         return 0;
1328 }
1329
1330 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1331 {
1332         cciss_coalint_struct intinfo;
1333         unsigned long flags;
1334         int i;
1335
1336         if (!argp)
1337                 return -EINVAL;
1338         if (!capable(CAP_SYS_ADMIN))
1339                 return -EPERM;
1340         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1341                 return -EFAULT;
1342         if ((intinfo.delay == 0) && (intinfo.count == 0))
1343                 return -EINVAL;
1344         spin_lock_irqsave(&h->lock, flags);
1345         /* Update the field, and then ring the doorbell */
1346         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1347         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1348         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1349
1350         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1351                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1352                         break;
1353                 udelay(1000); /* delay and try again */
1354         }
1355         spin_unlock_irqrestore(&h->lock, flags);
1356         if (i >= MAX_IOCTL_CONFIG_WAIT)
1357                 return -EAGAIN;
1358         return 0;
1359 }
1360
1361 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1362 {
1363         NodeName_type NodeName;
1364         unsigned long flags;
1365         int i;
1366
1367         if (!argp)
1368                 return -EINVAL;
1369         spin_lock_irqsave(&h->lock, flags);
1370         for (i = 0; i < 16; i++)
1371                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1372         spin_unlock_irqrestore(&h->lock, flags);
1373         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1374                 return -EFAULT;
1375         return 0;
1376 }
1377
1378 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1379 {
1380         NodeName_type NodeName;
1381         unsigned long flags;
1382         int i;
1383
1384         if (!argp)
1385                 return -EINVAL;
1386         if (!capable(CAP_SYS_ADMIN))
1387                 return -EPERM;
1388         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1389                 return -EFAULT;
1390         spin_lock_irqsave(&h->lock, flags);
1391         /* Update the field, and then ring the doorbell */
1392         for (i = 0; i < 16; i++)
1393                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1394         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1395         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1396                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1397                         break;
1398                 udelay(1000); /* delay and try again */
1399         }
1400         spin_unlock_irqrestore(&h->lock, flags);
1401         if (i >= MAX_IOCTL_CONFIG_WAIT)
1402                 return -EAGAIN;
1403         return 0;
1404 }
1405
1406 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1407 {
1408         Heartbeat_type heartbeat;
1409         unsigned long flags;
1410
1411         if (!argp)
1412                 return -EINVAL;
1413         spin_lock_irqsave(&h->lock, flags);
1414         heartbeat = readl(&h->cfgtable->HeartBeat);
1415         spin_unlock_irqrestore(&h->lock, flags);
1416         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1417                 return -EFAULT;
1418         return 0;
1419 }
1420
1421 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1422 {
1423         BusTypes_type BusTypes;
1424         unsigned long flags;
1425
1426         if (!argp)
1427                 return -EINVAL;
1428         spin_lock_irqsave(&h->lock, flags);
1429         BusTypes = readl(&h->cfgtable->BusTypes);
1430         spin_unlock_irqrestore(&h->lock, flags);
1431         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1432                 return -EFAULT;
1433         return 0;
1434 }
1435
1436 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1437 {
1438         FirmwareVer_type firmware;
1439
1440         if (!argp)
1441                 return -EINVAL;
1442         memcpy(firmware, h->firm_ver, 4);
1443
1444         if (copy_to_user
1445             (argp, firmware, sizeof(FirmwareVer_type)))
1446                 return -EFAULT;
1447         return 0;
1448 }
1449
1450 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1451 {
1452         DriverVer_type DriverVer = DRIVER_VERSION;
1453
1454         if (!argp)
1455                 return -EINVAL;
1456         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1457                 return -EFAULT;
1458         return 0;
1459 }
1460
1461 static int cciss_getluninfo(ctlr_info_t *h,
1462         struct gendisk *disk, void __user *argp)
1463 {
1464         LogvolInfo_struct luninfo;
1465         drive_info_struct *drv = get_drv(disk);
1466
1467         if (!argp)
1468                 return -EINVAL;
1469         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1470         luninfo.num_opens = drv->usage_count;
1471         luninfo.num_parts = 0;
1472         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1473                 return -EFAULT;
1474         return 0;
1475 }
1476
1477 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1478 {
1479         IOCTL_Command_struct iocommand;
1480         CommandList_struct *c;
1481         char *buff = NULL;
1482         u64bit temp64;
1483         DECLARE_COMPLETION_ONSTACK(wait);
1484
1485         if (!argp)
1486                 return -EINVAL;
1487
1488         if (!capable(CAP_SYS_RAWIO))
1489                 return -EPERM;
1490
1491         if (copy_from_user
1492             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1493                 return -EFAULT;
1494         if ((iocommand.buf_size < 1) &&
1495             (iocommand.Request.Type.Direction != XFER_NONE)) {
1496                 return -EINVAL;
1497         }
1498         if (iocommand.buf_size > 0) {
1499                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1500                 if (buff == NULL)
1501                         return -EFAULT;
1502         }
1503         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1504                 /* Copy the data into the buffer we created */
1505                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1506                         kfree(buff);
1507                         return -EFAULT;
1508                 }
1509         } else {
1510                 memset(buff, 0, iocommand.buf_size);
1511         }
1512         c = cmd_special_alloc(h);
1513         if (!c) {
1514                 kfree(buff);
1515                 return -ENOMEM;
1516         }
1517         /* Fill in the command type */
1518         c->cmd_type = CMD_IOCTL_PEND;
1519         /* Fill in Command Header */
1520         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1521         if (iocommand.buf_size > 0) { /* buffer to fill */
1522                 c->Header.SGList = 1;
1523                 c->Header.SGTotal = 1;
1524         } else { /* no buffers to fill */
1525                 c->Header.SGList = 0;
1526                 c->Header.SGTotal = 0;
1527         }
1528         c->Header.LUN = iocommand.LUN_info;
1529         /* use the kernel address the cmd block for tag */
1530         c->Header.Tag.lower = c->busaddr;
1531
1532         /* Fill in Request block */
1533         c->Request = iocommand.Request;
1534
1535         /* Fill in the scatter gather information */
1536         if (iocommand.buf_size > 0) {
1537                 temp64.val = pci_map_single(h->pdev, buff,
1538                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1539                 c->SG[0].Addr.lower = temp64.val32.lower;
1540                 c->SG[0].Addr.upper = temp64.val32.upper;
1541                 c->SG[0].Len = iocommand.buf_size;
1542                 c->SG[0].Ext = 0;  /* we are not chaining */
1543         }
1544         c->waiting = &wait;
1545
1546         enqueue_cmd_and_start_io(h, c);
1547         wait_for_completion(&wait);
1548
1549         /* unlock the buffers from DMA */
1550         temp64.val32.lower = c->SG[0].Addr.lower;
1551         temp64.val32.upper = c->SG[0].Addr.upper;
1552         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1553                          PCI_DMA_BIDIRECTIONAL);
1554         check_ioctl_unit_attention(h, c);
1555
1556         /* Copy the error information out */
1557         iocommand.error_info = *(c->err_info);
1558         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1559                 kfree(buff);
1560                 cmd_special_free(h, c);
1561                 return -EFAULT;
1562         }
1563
1564         if (iocommand.Request.Type.Direction == XFER_READ) {
1565                 /* Copy the data out of the buffer we created */
1566                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1567                         kfree(buff);
1568                         cmd_special_free(h, c);
1569                         return -EFAULT;
1570                 }
1571         }
1572         kfree(buff);
1573         cmd_special_free(h, c);
1574         return 0;
1575 }
1576
1577 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1578 {
1579         BIG_IOCTL_Command_struct *ioc;
1580         CommandList_struct *c;
1581         unsigned char **buff = NULL;
1582         int *buff_size = NULL;
1583         u64bit temp64;
1584         BYTE sg_used = 0;
1585         int status = 0;
1586         int i;
1587         DECLARE_COMPLETION_ONSTACK(wait);
1588         __u32 left;
1589         __u32 sz;
1590         BYTE __user *data_ptr;
1591
1592         if (!argp)
1593                 return -EINVAL;
1594         if (!capable(CAP_SYS_RAWIO))
1595                 return -EPERM;
1596         ioc = kmalloc(sizeof(*ioc), GFP_KERNEL);
1597         if (!ioc) {
1598                 status = -ENOMEM;
1599                 goto cleanup1;
1600         }
1601         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1602                 status = -EFAULT;
1603                 goto cleanup1;
1604         }
1605         if ((ioc->buf_size < 1) &&
1606             (ioc->Request.Type.Direction != XFER_NONE)) {
1607                 status = -EINVAL;
1608                 goto cleanup1;
1609         }
1610         /* Check kmalloc limits  using all SGs */
1611         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1612                 status = -EINVAL;
1613                 goto cleanup1;
1614         }
1615         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1616                 status = -EINVAL;
1617                 goto cleanup1;
1618         }
1619         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1620         if (!buff) {
1621                 status = -ENOMEM;
1622                 goto cleanup1;
1623         }
1624         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1625         if (!buff_size) {
1626                 status = -ENOMEM;
1627                 goto cleanup1;
1628         }
1629         left = ioc->buf_size;
1630         data_ptr = ioc->buf;
1631         while (left) {
1632                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1633                 buff_size[sg_used] = sz;
1634                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1635                 if (buff[sg_used] == NULL) {
1636                         status = -ENOMEM;
1637                         goto cleanup1;
1638                 }
1639                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1640                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1641                                 status = -EFAULT;
1642                                 goto cleanup1;
1643                         }
1644                 } else {
1645                         memset(buff[sg_used], 0, sz);
1646                 }
1647                 left -= sz;
1648                 data_ptr += sz;
1649                 sg_used++;
1650         }
1651         c = cmd_special_alloc(h);
1652         if (!c) {
1653                 status = -ENOMEM;
1654                 goto cleanup1;
1655         }
1656         c->cmd_type = CMD_IOCTL_PEND;
1657         c->Header.ReplyQueue = 0;
1658         c->Header.SGList = sg_used;
1659         c->Header.SGTotal = sg_used;
1660         c->Header.LUN = ioc->LUN_info;
1661         c->Header.Tag.lower = c->busaddr;
1662
1663         c->Request = ioc->Request;
1664         for (i = 0; i < sg_used; i++) {
1665                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1666                                     PCI_DMA_BIDIRECTIONAL);
1667                 c->SG[i].Addr.lower = temp64.val32.lower;
1668                 c->SG[i].Addr.upper = temp64.val32.upper;
1669                 c->SG[i].Len = buff_size[i];
1670                 c->SG[i].Ext = 0;       /* we are not chaining */
1671         }
1672         c->waiting = &wait;
1673         enqueue_cmd_and_start_io(h, c);
1674         wait_for_completion(&wait);
1675         /* unlock the buffers from DMA */
1676         for (i = 0; i < sg_used; i++) {
1677                 temp64.val32.lower = c->SG[i].Addr.lower;
1678                 temp64.val32.upper = c->SG[i].Addr.upper;
1679                 pci_unmap_single(h->pdev,
1680                         (dma_addr_t) temp64.val, buff_size[i],
1681                         PCI_DMA_BIDIRECTIONAL);
1682         }
1683         check_ioctl_unit_attention(h, c);
1684         /* Copy the error information out */
1685         ioc->error_info = *(c->err_info);
1686         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1687                 cmd_special_free(h, c);
1688                 status = -EFAULT;
1689                 goto cleanup1;
1690         }
1691         if (ioc->Request.Type.Direction == XFER_READ) {
1692                 /* Copy the data out of the buffer we created */
1693                 BYTE __user *ptr = ioc->buf;
1694                 for (i = 0; i < sg_used; i++) {
1695                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1696                                 cmd_special_free(h, c);
1697                                 status = -EFAULT;
1698                                 goto cleanup1;
1699                         }
1700                         ptr += buff_size[i];
1701                 }
1702         }
1703         cmd_special_free(h, c);
1704         status = 0;
1705 cleanup1:
1706         if (buff) {
1707                 for (i = 0; i < sg_used; i++)
1708                         kfree(buff[i]);
1709                 kfree(buff);
1710         }
1711         kfree(buff_size);
1712         kfree(ioc);
1713         return status;
1714 }
1715
1716 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1717         unsigned int cmd, unsigned long arg)
1718 {
1719         struct gendisk *disk = bdev->bd_disk;
1720         ctlr_info_t *h = get_host(disk);
1721         void __user *argp = (void __user *)arg;
1722
1723         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1724                 cmd, arg);
1725         switch (cmd) {
1726         case CCISS_GETPCIINFO:
1727                 return cciss_getpciinfo(h, argp);
1728         case CCISS_GETINTINFO:
1729                 return cciss_getintinfo(h, argp);
1730         case CCISS_SETINTINFO:
1731                 return cciss_setintinfo(h, argp);
1732         case CCISS_GETNODENAME:
1733                 return cciss_getnodename(h, argp);
1734         case CCISS_SETNODENAME:
1735                 return cciss_setnodename(h, argp);
1736         case CCISS_GETHEARTBEAT:
1737                 return cciss_getheartbeat(h, argp);
1738         case CCISS_GETBUSTYPES:
1739                 return cciss_getbustypes(h, argp);
1740         case CCISS_GETFIRMVER:
1741                 return cciss_getfirmver(h, argp);
1742         case CCISS_GETDRIVVER:
1743                 return cciss_getdrivver(h, argp);
1744         case CCISS_DEREGDISK:
1745         case CCISS_REGNEWD:
1746         case CCISS_REVALIDVOLS:
1747                 return rebuild_lun_table(h, 0, 1);
1748         case CCISS_GETLUNINFO:
1749                 return cciss_getluninfo(h, disk, argp);
1750         case CCISS_PASSTHRU:
1751                 return cciss_passthru(h, argp);
1752         case CCISS_BIG_PASSTHRU:
1753                 return cciss_bigpassthru(h, argp);
1754
1755         /* scsi_cmd_blk_ioctl handles these, below, though some are not */
1756         /* very meaningful for cciss.  SG_IO is the main one people want. */
1757
1758         case SG_GET_VERSION_NUM:
1759         case SG_SET_TIMEOUT:
1760         case SG_GET_TIMEOUT:
1761         case SG_GET_RESERVED_SIZE:
1762         case SG_SET_RESERVED_SIZE:
1763         case SG_EMULATED_HOST:
1764         case SG_IO:
1765         case SCSI_IOCTL_SEND_COMMAND:
1766                 return scsi_cmd_blk_ioctl(bdev, mode, cmd, argp);
1767
1768         /* scsi_cmd_blk_ioctl would normally handle these, below, but */
1769         /* they aren't a good fit for cciss, as CD-ROMs are */
1770         /* not supported, and we don't have any bus/target/lun */
1771         /* which we present to the kernel. */
1772
1773         case CDROM_SEND_PACKET:
1774         case CDROMCLOSETRAY:
1775         case CDROMEJECT:
1776         case SCSI_IOCTL_GET_IDLUN:
1777         case SCSI_IOCTL_GET_BUS_NUMBER:
1778         default:
1779                 return -ENOTTY;
1780         }
1781 }
1782
1783 static void cciss_check_queues(ctlr_info_t *h)
1784 {
1785         int start_queue = h->next_to_run;
1786         int i;
1787
1788         /* check to see if we have maxed out the number of commands that can
1789          * be placed on the queue.  If so then exit.  We do this check here
1790          * in case the interrupt we serviced was from an ioctl and did not
1791          * free any new commands.
1792          */
1793         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1794                 return;
1795
1796         /* We have room on the queue for more commands.  Now we need to queue
1797          * them up.  We will also keep track of the next queue to run so
1798          * that every queue gets a chance to be started first.
1799          */
1800         for (i = 0; i < h->highest_lun + 1; i++) {
1801                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1802                 /* make sure the disk has been added and the drive is real
1803                  * because this can be called from the middle of init_one.
1804                  */
1805                 if (!h->drv[curr_queue])
1806                         continue;
1807                 if (!(h->drv[curr_queue]->queue) ||
1808                         !(h->drv[curr_queue]->heads))
1809                         continue;
1810                 blk_start_queue(h->gendisk[curr_queue]->queue);
1811
1812                 /* check to see if we have maxed out the number of commands
1813                  * that can be placed on the queue.
1814                  */
1815                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1816                         if (curr_queue == start_queue) {
1817                                 h->next_to_run =
1818                                     (start_queue + 1) % (h->highest_lun + 1);
1819                                 break;
1820                         } else {
1821                                 h->next_to_run = curr_queue;
1822                                 break;
1823                         }
1824                 }
1825         }
1826 }
1827
1828 static void cciss_softirq_done(struct request *rq)
1829 {
1830         CommandList_struct *c = rq->completion_data;
1831         ctlr_info_t *h = hba[c->ctlr];
1832         SGDescriptor_struct *curr_sg = c->SG;
1833         u64bit temp64;
1834         unsigned long flags;
1835         int i, ddir;
1836         int sg_index = 0;
1837
1838         if (c->Request.Type.Direction == XFER_READ)
1839                 ddir = PCI_DMA_FROMDEVICE;
1840         else
1841                 ddir = PCI_DMA_TODEVICE;
1842
1843         /* command did not need to be retried */
1844         /* unmap the DMA mapping for all the scatter gather elements */
1845         for (i = 0; i < c->Header.SGList; i++) {
1846                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1847                         cciss_unmap_sg_chain_block(h, c);
1848                         /* Point to the next block */
1849                         curr_sg = h->cmd_sg_list[c->cmdindex];
1850                         sg_index = 0;
1851                 }
1852                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1853                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1854                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1855                                 ddir);
1856                 ++sg_index;
1857         }
1858
1859         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1860
1861         /* set the residual count for pc requests */
1862         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1863                 rq->resid_len = c->err_info->ResidualCnt;
1864
1865         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1866
1867         spin_lock_irqsave(&h->lock, flags);
1868         cmd_free(h, c);
1869         cciss_check_queues(h);
1870         spin_unlock_irqrestore(&h->lock, flags);
1871 }
1872
1873 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1874         unsigned char scsi3addr[], uint32_t log_unit)
1875 {
1876         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1877                 sizeof(h->drv[log_unit]->LunID));
1878 }
1879
1880 /* This function gets the SCSI vendor, model, and revision of a logical drive
1881  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1882  * they cannot be read.
1883  */
1884 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1885                                    char *vendor, char *model, char *rev)
1886 {
1887         int rc;
1888         InquiryData_struct *inq_buf;
1889         unsigned char scsi3addr[8];
1890
1891         *vendor = '\0';
1892         *model = '\0';
1893         *rev = '\0';
1894
1895         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1896         if (!inq_buf)
1897                 return;
1898
1899         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1900         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1901                         scsi3addr, TYPE_CMD);
1902         if (rc == IO_OK) {
1903                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1904                 vendor[VENDOR_LEN] = '\0';
1905                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1906                 model[MODEL_LEN] = '\0';
1907                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1908                 rev[REV_LEN] = '\0';
1909         }
1910
1911         kfree(inq_buf);
1912         return;
1913 }
1914
1915 /* This function gets the serial number of a logical drive via
1916  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1917  * number cannot be had, for whatever reason, 16 bytes of 0xff
1918  * are returned instead.
1919  */
1920 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1921                                 unsigned char *serial_no, int buflen)
1922 {
1923 #define PAGE_83_INQ_BYTES 64
1924         int rc;
1925         unsigned char *buf;
1926         unsigned char scsi3addr[8];
1927
1928         if (buflen > 16)
1929                 buflen = 16;
1930         memset(serial_no, 0xff, buflen);
1931         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1932         if (!buf)
1933                 return;
1934         memset(serial_no, 0, buflen);
1935         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1936         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1937                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1938         if (rc == IO_OK)
1939                 memcpy(serial_no, &buf[8], buflen);
1940         kfree(buf);
1941         return;
1942 }
1943
1944 /*
1945  * cciss_add_disk sets up the block device queue for a logical drive
1946  */
1947 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1948                                 int drv_index)
1949 {
1950         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1951         if (!disk->queue)
1952                 goto init_queue_failure;
1953         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1954         disk->major = h->major;
1955         disk->first_minor = drv_index << NWD_SHIFT;
1956         disk->fops = &cciss_fops;
1957         if (cciss_create_ld_sysfs_entry(h, drv_index))
1958                 goto cleanup_queue;
1959         disk->private_data = h->drv[drv_index];
1960         disk->driverfs_dev = &h->drv[drv_index]->dev;
1961
1962         /* Set up queue information */
1963         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1964
1965         /* This is a hardware imposed limit. */
1966         blk_queue_max_segments(disk->queue, h->maxsgentries);
1967
1968         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1969
1970         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1971
1972         disk->queue->queuedata = h;
1973
1974         blk_queue_logical_block_size(disk->queue,
1975                                      h->drv[drv_index]->block_size);
1976
1977         /* Make sure all queue data is written out before */
1978         /* setting h->drv[drv_index]->queue, as setting this */
1979         /* allows the interrupt handler to start the queue */
1980         wmb();
1981         h->drv[drv_index]->queue = disk->queue;
1982         add_disk(disk);
1983         return 0;
1984
1985 cleanup_queue:
1986         blk_cleanup_queue(disk->queue);
1987         disk->queue = NULL;
1988 init_queue_failure:
1989         return -1;
1990 }
1991
1992 /* This function will check the usage_count of the drive to be updated/added.
1993  * If the usage_count is zero and it is a heretofore unknown drive, or,
1994  * the drive's capacity, geometry, or serial number has changed,
1995  * then the drive information will be updated and the disk will be
1996  * re-registered with the kernel.  If these conditions don't hold,
1997  * then it will be left alone for the next reboot.  The exception to this
1998  * is disk 0 which will always be left registered with the kernel since it
1999  * is also the controller node.  Any changes to disk 0 will show up on
2000  * the next reboot.
2001  */
2002 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
2003         int first_time, int via_ioctl)
2004 {
2005         struct gendisk *disk;
2006         InquiryData_struct *inq_buff = NULL;
2007         unsigned int block_size;
2008         sector_t total_size;
2009         unsigned long flags = 0;
2010         int ret = 0;
2011         drive_info_struct *drvinfo;
2012
2013         /* Get information about the disk and modify the driver structure */
2014         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2015         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
2016         if (inq_buff == NULL || drvinfo == NULL)
2017                 goto mem_msg;
2018
2019         /* testing to see if 16-byte CDBs are already being used */
2020         if (h->cciss_read == CCISS_READ_16) {
2021                 cciss_read_capacity_16(h, drv_index,
2022                         &total_size, &block_size);
2023
2024         } else {
2025                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
2026                 /* if read_capacity returns all F's this volume is >2TB */
2027                 /* in size so we switch to 16-byte CDB's for all */
2028                 /* read/write ops */
2029                 if (total_size == 0xFFFFFFFFULL) {
2030                         cciss_read_capacity_16(h, drv_index,
2031                         &total_size, &block_size);
2032                         h->cciss_read = CCISS_READ_16;
2033                         h->cciss_write = CCISS_WRITE_16;
2034                 } else {
2035                         h->cciss_read = CCISS_READ_10;
2036                         h->cciss_write = CCISS_WRITE_10;
2037                 }
2038         }
2039
2040         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
2041                                inq_buff, drvinfo);
2042         drvinfo->block_size = block_size;
2043         drvinfo->nr_blocks = total_size + 1;
2044
2045         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
2046                                 drvinfo->model, drvinfo->rev);
2047         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
2048                         sizeof(drvinfo->serial_no));
2049         /* Save the lunid in case we deregister the disk, below. */
2050         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
2051                 sizeof(drvinfo->LunID));
2052
2053         /* Is it the same disk we already know, and nothing's changed? */
2054         if (h->drv[drv_index]->raid_level != -1 &&
2055                 ((memcmp(drvinfo->serial_no,
2056                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2057                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2058                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2059                 drvinfo->heads == h->drv[drv_index]->heads &&
2060                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2061                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2062                         /* The disk is unchanged, nothing to update */
2063                         goto freeret;
2064
2065         /* If we get here it's not the same disk, or something's changed,
2066          * so we need to * deregister it, and re-register it, if it's not
2067          * in use.
2068          * If the disk already exists then deregister it before proceeding
2069          * (unless it's the first disk (for the controller node).
2070          */
2071         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2072                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2073                 spin_lock_irqsave(&h->lock, flags);
2074                 h->drv[drv_index]->busy_configuring = 1;
2075                 spin_unlock_irqrestore(&h->lock, flags);
2076
2077                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2078                  * which keeps the interrupt handler from starting
2079                  * the queue.
2080                  */
2081                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2082         }
2083
2084         /* If the disk is in use return */
2085         if (ret)
2086                 goto freeret;
2087
2088         /* Save the new information from cciss_geometry_inquiry
2089          * and serial number inquiry.  If the disk was deregistered
2090          * above, then h->drv[drv_index] will be NULL.
2091          */
2092         if (h->drv[drv_index] == NULL) {
2093                 drvinfo->device_initialized = 0;
2094                 h->drv[drv_index] = drvinfo;
2095                 drvinfo = NULL; /* so it won't be freed below. */
2096         } else {
2097                 /* special case for cxd0 */
2098                 h->drv[drv_index]->block_size = drvinfo->block_size;
2099                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2100                 h->drv[drv_index]->heads = drvinfo->heads;
2101                 h->drv[drv_index]->sectors = drvinfo->sectors;
2102                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2103                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2104                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2105                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2106                         VENDOR_LEN + 1);
2107                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2108                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2109         }
2110
2111         ++h->num_luns;
2112         disk = h->gendisk[drv_index];
2113         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2114
2115         /* If it's not disk 0 (drv_index != 0)
2116          * or if it was disk 0, but there was previously
2117          * no actual corresponding configured logical drive
2118          * (raid_leve == -1) then we want to update the
2119          * logical drive's information.
2120          */
2121         if (drv_index || first_time) {
2122                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2123                         cciss_free_gendisk(h, drv_index);
2124                         cciss_free_drive_info(h, drv_index);
2125                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2126                                 drv_index);
2127                         --h->num_luns;
2128                 }
2129         }
2130
2131 freeret:
2132         kfree(inq_buff);
2133         kfree(drvinfo);
2134         return;
2135 mem_msg:
2136         dev_err(&h->pdev->dev, "out of memory\n");
2137         goto freeret;
2138 }
2139
2140 /* This function will find the first index of the controllers drive array
2141  * that has a null drv pointer and allocate the drive info struct and
2142  * will return that index   This is where new drives will be added.
2143  * If the index to be returned is greater than the highest_lun index for
2144  * the controller then highest_lun is set * to this new index.
2145  * If there are no available indexes or if tha allocation fails, then -1
2146  * is returned.  * "controller_node" is used to know if this is a real
2147  * logical drive, or just the controller node, which determines if this
2148  * counts towards highest_lun.
2149  */
2150 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2151 {
2152         int i;
2153         drive_info_struct *drv;
2154
2155         /* Search for an empty slot for our drive info */
2156         for (i = 0; i < CISS_MAX_LUN; i++) {
2157
2158                 /* if not cxd0 case, and it's occupied, skip it. */
2159                 if (h->drv[i] && i != 0)
2160                         continue;
2161                 /*
2162                  * If it's cxd0 case, and drv is alloc'ed already, and a
2163                  * disk is configured there, skip it.
2164                  */
2165                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2166                         continue;
2167
2168                 /*
2169                  * We've found an empty slot.  Update highest_lun
2170                  * provided this isn't just the fake cxd0 controller node.
2171                  */
2172                 if (i > h->highest_lun && !controller_node)
2173                         h->highest_lun = i;
2174
2175                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2176                 if (i == 0 && h->drv[i] != NULL)
2177                         return i;
2178
2179                 /*
2180                  * Found an empty slot, not already alloc'ed.  Allocate it.
2181                  * Mark it with raid_level == -1, so we know it's new later on.
2182                  */
2183                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2184                 if (!drv)
2185                         return -1;
2186                 drv->raid_level = -1; /* so we know it's new */
2187                 h->drv[i] = drv;
2188                 return i;
2189         }
2190         return -1;
2191 }
2192
2193 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2194 {
2195         kfree(h->drv[drv_index]);
2196         h->drv[drv_index] = NULL;
2197 }
2198
2199 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2200 {
2201         put_disk(h->gendisk[drv_index]);
2202         h->gendisk[drv_index] = NULL;
2203 }
2204
2205 /* cciss_add_gendisk finds a free hba[]->drv structure
2206  * and allocates a gendisk if needed, and sets the lunid
2207  * in the drvinfo structure.   It returns the index into
2208  * the ->drv[] array, or -1 if none are free.
2209  * is_controller_node indicates whether highest_lun should
2210  * count this disk, or if it's only being added to provide
2211  * a means to talk to the controller in case no logical
2212  * drives have yet been configured.
2213  */
2214 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2215         int controller_node)
2216 {
2217         int drv_index;
2218
2219         drv_index = cciss_alloc_drive_info(h, controller_node);
2220         if (drv_index == -1)
2221                 return -1;
2222
2223         /*Check if the gendisk needs to be allocated */
2224         if (!h->gendisk[drv_index]) {
2225                 h->gendisk[drv_index] =
2226                         alloc_disk(1 << NWD_SHIFT);
2227                 if (!h->gendisk[drv_index]) {
2228                         dev_err(&h->pdev->dev,
2229                                 "could not allocate a new disk %d\n",
2230                                 drv_index);
2231                         goto err_free_drive_info;
2232                 }
2233         }
2234         memcpy(h->drv[drv_index]->LunID, lunid,
2235                 sizeof(h->drv[drv_index]->LunID));
2236         if (cciss_create_ld_sysfs_entry(h, drv_index))
2237                 goto err_free_disk;
2238         /* Don't need to mark this busy because nobody */
2239         /* else knows about this disk yet to contend */
2240         /* for access to it. */
2241         h->drv[drv_index]->busy_configuring = 0;
2242         wmb();
2243         return drv_index;
2244
2245 err_free_disk:
2246         cciss_free_gendisk(h, drv_index);
2247 err_free_drive_info:
2248         cciss_free_drive_info(h, drv_index);
2249         return -1;
2250 }
2251
2252 /* This is for the special case of a controller which
2253  * has no logical drives.  In this case, we still need
2254  * to register a disk so the controller can be accessed
2255  * by the Array Config Utility.
2256  */
2257 static void cciss_add_controller_node(ctlr_info_t *h)
2258 {
2259         struct gendisk *disk;
2260         int drv_index;
2261
2262         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2263                 return;
2264
2265         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2266         if (drv_index == -1)
2267                 goto error;
2268         h->drv[drv_index]->block_size = 512;
2269         h->drv[drv_index]->nr_blocks = 0;
2270         h->drv[drv_index]->heads = 0;
2271         h->drv[drv_index]->sectors = 0;
2272         h->drv[drv_index]->cylinders = 0;
2273         h->drv[drv_index]->raid_level = -1;
2274         memset(h->drv[drv_index]->serial_no, 0, 16);
2275         disk = h->gendisk[drv_index];
2276         if (cciss_add_disk(h, disk, drv_index) == 0)
2277                 return;
2278         cciss_free_gendisk(h, drv_index);
2279         cciss_free_drive_info(h, drv_index);
2280 error:
2281         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2282         return;
2283 }
2284
2285 /* This function will add and remove logical drives from the Logical
2286  * drive array of the controller and maintain persistency of ordering
2287  * so that mount points are preserved until the next reboot.  This allows
2288  * for the removal of logical drives in the middle of the drive array
2289  * without a re-ordering of those drives.
2290  * INPUT
2291  * h            = The controller to perform the operations on
2292  */
2293 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2294         int via_ioctl)
2295 {
2296         int num_luns;
2297         ReportLunData_struct *ld_buff = NULL;
2298         int return_code;
2299         int listlength = 0;
2300         int i;
2301         int drv_found;
2302         int drv_index = 0;
2303         unsigned char lunid[8] = CTLR_LUNID;
2304         unsigned long flags;
2305
2306         if (!capable(CAP_SYS_RAWIO))
2307                 return -EPERM;
2308
2309         /* Set busy_configuring flag for this operation */
2310         spin_lock_irqsave(&h->lock, flags);
2311         if (h->busy_configuring) {
2312                 spin_unlock_irqrestore(&h->lock, flags);
2313                 return -EBUSY;
2314         }
2315         h->busy_configuring = 1;
2316         spin_unlock_irqrestore(&h->lock, flags);
2317
2318         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2319         if (ld_buff == NULL)
2320                 goto mem_msg;
2321
2322         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2323                                       sizeof(ReportLunData_struct),
2324                                       0, CTLR_LUNID, TYPE_CMD);
2325
2326         if (return_code == IO_OK)
2327                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2328         else {  /* reading number of logical volumes failed */
2329                 dev_warn(&h->pdev->dev,
2330                         "report logical volume command failed\n");
2331                 listlength = 0;
2332                 goto freeret;
2333         }
2334
2335         num_luns = listlength / 8;      /* 8 bytes per entry */
2336         if (num_luns > CISS_MAX_LUN) {
2337                 num_luns = CISS_MAX_LUN;
2338                 dev_warn(&h->pdev->dev, "more luns configured"
2339                        " on controller than can be handled by"
2340                        " this driver.\n");
2341         }
2342
2343         if (num_luns == 0)
2344                 cciss_add_controller_node(h);
2345
2346         /* Compare controller drive array to driver's drive array
2347          * to see if any drives are missing on the controller due
2348          * to action of Array Config Utility (user deletes drive)
2349          * and deregister logical drives which have disappeared.
2350          */
2351         for (i = 0; i <= h->highest_lun; i++) {
2352                 int j;
2353                 drv_found = 0;
2354
2355                 /* skip holes in the array from already deleted drives */
2356                 if (h->drv[i] == NULL)
2357                         continue;
2358
2359                 for (j = 0; j < num_luns; j++) {
2360                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2361                         if (memcmp(h->drv[i]->LunID, lunid,
2362                                 sizeof(lunid)) == 0) {
2363                                 drv_found = 1;
2364                                 break;
2365                         }
2366                 }
2367                 if (!drv_found) {
2368                         /* Deregister it from the OS, it's gone. */
2369                         spin_lock_irqsave(&h->lock, flags);
2370                         h->drv[i]->busy_configuring = 1;
2371                         spin_unlock_irqrestore(&h->lock, flags);
2372                         return_code = deregister_disk(h, i, 1, via_ioctl);
2373                         if (h->drv[i] != NULL)
2374                                 h->drv[i]->busy_configuring = 0;
2375                 }
2376         }
2377
2378         /* Compare controller drive array to driver's drive array.
2379          * Check for updates in the drive information and any new drives
2380          * on the controller due to ACU adding logical drives, or changing
2381          * a logical drive's size, etc.  Reregister any new/changed drives
2382          */
2383         for (i = 0; i < num_luns; i++) {
2384                 int j;
2385
2386                 drv_found = 0;
2387
2388                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2389                 /* Find if the LUN is already in the drive array
2390                  * of the driver.  If so then update its info
2391                  * if not in use.  If it does not exist then find
2392                  * the first free index and add it.
2393                  */
2394                 for (j = 0; j <= h->highest_lun; j++) {
2395                         if (h->drv[j] != NULL &&
2396                                 memcmp(h->drv[j]->LunID, lunid,
2397                                         sizeof(h->drv[j]->LunID)) == 0) {
2398                                 drv_index = j;
2399                                 drv_found = 1;
2400                                 break;
2401                         }
2402                 }
2403
2404                 /* check if the drive was found already in the array */
2405                 if (!drv_found) {
2406                         drv_index = cciss_add_gendisk(h, lunid, 0);
2407                         if (drv_index == -1)
2408                                 goto freeret;
2409                 }
2410                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2411         }               /* end for */
2412
2413 freeret:
2414         kfree(ld_buff);
2415         h->busy_configuring = 0;
2416         /* We return -1 here to tell the ACU that we have registered/updated
2417          * all of the drives that we can and to keep it from calling us
2418          * additional times.
2419          */
2420         return -1;
2421 mem_msg:
2422         dev_err(&h->pdev->dev, "out of memory\n");
2423         h->busy_configuring = 0;
2424         goto freeret;
2425 }
2426
2427 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2428 {
2429         /* zero out the disk size info */
2430         drive_info->nr_blocks = 0;
2431         drive_info->block_size = 0;
2432         drive_info->heads = 0;
2433         drive_info->sectors = 0;
2434         drive_info->cylinders = 0;
2435         drive_info->raid_level = -1;
2436         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2437         memset(drive_info->model, 0, sizeof(drive_info->model));
2438         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2439         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2440         /*
2441          * don't clear the LUNID though, we need to remember which
2442          * one this one is.
2443          */
2444 }
2445
2446 /* This function will deregister the disk and it's queue from the
2447  * kernel.  It must be called with the controller lock held and the
2448  * drv structures busy_configuring flag set.  It's parameters are:
2449  *
2450  * disk = This is the disk to be deregistered
2451  * drv  = This is the drive_info_struct associated with the disk to be
2452  *        deregistered.  It contains information about the disk used
2453  *        by the driver.
2454  * clear_all = This flag determines whether or not the disk information
2455  *             is going to be completely cleared out and the highest_lun
2456  *             reset.  Sometimes we want to clear out information about
2457  *             the disk in preparation for re-adding it.  In this case
2458  *             the highest_lun should be left unchanged and the LunID
2459  *             should not be cleared.
2460  * via_ioctl
2461  *    This indicates whether we've reached this path via ioctl.
2462  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2463  *    If this path is reached via ioctl(), then the max_usage_count will
2464  *    be 1, as the process calling ioctl() has got to have the device open.
2465  *    If we get here via sysfs, then the max usage count will be zero.
2466 */
2467 static int deregister_disk(ctlr_info_t *h, int drv_index,
2468                            int clear_all, int via_ioctl)
2469 {
2470         int i;
2471         struct gendisk *disk;
2472         drive_info_struct *drv;
2473         int recalculate_highest_lun;
2474
2475         if (!capable(CAP_SYS_RAWIO))
2476                 return -EPERM;
2477
2478         drv = h->drv[drv_index];
2479         disk = h->gendisk[drv_index];
2480
2481         /* make sure logical volume is NOT is use */
2482         if (clear_all || (h->gendisk[0] == disk)) {
2483                 if (drv->usage_count > via_ioctl)
2484                         return -EBUSY;
2485         } else if (drv->usage_count > 0)
2486                 return -EBUSY;
2487
2488         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2489
2490         /* invalidate the devices and deregister the disk.  If it is disk
2491          * zero do not deregister it but just zero out it's values.  This
2492          * allows us to delete disk zero but keep the controller registered.
2493          */
2494         if (h->gendisk[0] != disk) {
2495                 struct request_queue *q = disk->queue;
2496                 if (disk->flags & GENHD_FL_UP) {
2497                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2498                         del_gendisk(disk);
2499                 }
2500                 if (q)
2501                         blk_cleanup_queue(q);
2502                 /* If clear_all is set then we are deleting the logical
2503                  * drive, not just refreshing its info.  For drives
2504                  * other than disk 0 we will call put_disk.  We do not
2505                  * do this for disk 0 as we need it to be able to
2506                  * configure the controller.
2507                  */
2508                 if (clear_all){
2509                         /* This isn't pretty, but we need to find the
2510                          * disk in our array and NULL our the pointer.
2511                          * This is so that we will call alloc_disk if
2512                          * this index is used again later.
2513                          */
2514                         for (i=0; i < CISS_MAX_LUN; i++){
2515                                 if (h->gendisk[i] == disk) {
2516                                         h->gendisk[i] = NULL;
2517                                         break;
2518                                 }
2519                         }
2520                         put_disk(disk);
2521                 }
2522         } else {
2523                 set_capacity(disk, 0);
2524                 cciss_clear_drive_info(drv);
2525         }
2526
2527         --h->num_luns;
2528
2529         /* if it was the last disk, find the new hightest lun */
2530         if (clear_all && recalculate_highest_lun) {
2531                 int newhighest = -1;
2532                 for (i = 0; i <= h->highest_lun; i++) {
2533                         /* if the disk has size > 0, it is available */
2534                         if (h->drv[i] && h->drv[i]->heads)
2535                                 newhighest = i;
2536                 }
2537                 h->highest_lun = newhighest;
2538         }
2539         return 0;
2540 }
2541
2542 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2543                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2544                 int cmd_type)
2545 {
2546         u64bit buff_dma_handle;
2547         int status = IO_OK;
2548
2549         c->cmd_type = CMD_IOCTL_PEND;
2550         c->Header.ReplyQueue = 0;
2551         if (buff != NULL) {
2552                 c->Header.SGList = 1;
2553                 c->Header.SGTotal = 1;
2554         } else {
2555                 c->Header.SGList = 0;
2556                 c->Header.SGTotal = 0;
2557         }
2558         c->Header.Tag.lower = c->busaddr;
2559         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2560
2561         c->Request.Type.Type = cmd_type;
2562         if (cmd_type == TYPE_CMD) {
2563                 switch (cmd) {
2564                 case CISS_INQUIRY:
2565                         /* are we trying to read a vital product page */
2566                         if (page_code != 0) {
2567                                 c->Request.CDB[1] = 0x01;
2568                                 c->Request.CDB[2] = page_code;
2569                         }
2570                         c->Request.CDBLen = 6;
2571                         c->Request.Type.Attribute = ATTR_SIMPLE;
2572                         c->Request.Type.Direction = XFER_READ;
2573                         c->Request.Timeout = 0;
2574                         c->Request.CDB[0] = CISS_INQUIRY;
2575                         c->Request.CDB[4] = size & 0xFF;
2576                         break;
2577                 case CISS_REPORT_LOG:
2578                 case CISS_REPORT_PHYS:
2579                         /* Talking to controller so It's a physical command
2580                            mode = 00 target = 0.  Nothing to write.
2581                          */
2582                         c->Request.CDBLen = 12;
2583                         c->Request.Type.Attribute = ATTR_SIMPLE;
2584                         c->Request.Type.Direction = XFER_READ;
2585                         c->Request.Timeout = 0;
2586                         c->Request.CDB[0] = cmd;
2587                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2588                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2589                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2590                         c->Request.CDB[9] = size & 0xFF;
2591                         break;
2592
2593                 case CCISS_READ_CAPACITY:
2594                         c->Request.CDBLen = 10;
2595                         c->Request.Type.Attribute = ATTR_SIMPLE;
2596                         c->Request.Type.Direction = XFER_READ;
2597                         c->Request.Timeout = 0;
2598                         c->Request.CDB[0] = cmd;
2599                         break;
2600                 case CCISS_READ_CAPACITY_16:
2601                         c->Request.CDBLen = 16;
2602                         c->Request.Type.Attribute = ATTR_SIMPLE;
2603                         c->Request.Type.Direction = XFER_READ;
2604                         c->Request.Timeout = 0;
2605                         c->Request.CDB[0] = cmd;
2606                         c->Request.CDB[1] = 0x10;
2607                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2608                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2609                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2610                         c->Request.CDB[13] = size & 0xFF;
2611                         c->Request.Timeout = 0;
2612                         c->Request.CDB[0] = cmd;
2613                         break;
2614                 case CCISS_CACHE_FLUSH:
2615                         c->Request.CDBLen = 12;
2616                         c->Request.Type.Attribute = ATTR_SIMPLE;
2617                         c->Request.Type.Direction = XFER_WRITE;
2618                         c->Request.Timeout = 0;
2619                         c->Request.CDB[0] = BMIC_WRITE;
2620                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2621                         c->Request.CDB[7] = (size >> 8) & 0xFF;
2622                         c->Request.CDB[8] = size & 0xFF;
2623                         break;
2624                 case TEST_UNIT_READY:
2625                         c->Request.CDBLen = 6;
2626                         c->Request.Type.Attribute = ATTR_SIMPLE;
2627                         c->Request.Type.Direction = XFER_NONE;
2628                         c->Request.Timeout = 0;
2629                         break;
2630                 default:
2631                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2632                         return IO_ERROR;
2633                 }
2634         } else if (cmd_type == TYPE_MSG) {
2635                 switch (cmd) {
2636                 case CCISS_ABORT_MSG:
2637                         c->Request.CDBLen = 12;
2638                         c->Request.Type.Attribute = ATTR_SIMPLE;
2639                         c->Request.Type.Direction = XFER_WRITE;
2640                         c->Request.Timeout = 0;
2641                         c->Request.CDB[0] = cmd;        /* abort */
2642                         c->Request.CDB[1] = 0;  /* abort a command */
2643                         /* buff contains the tag of the command to abort */
2644                         memcpy(&c->Request.CDB[4], buff, 8);
2645                         break;
2646                 case CCISS_RESET_MSG:
2647                         c->Request.CDBLen = 16;
2648                         c->Request.Type.Attribute = ATTR_SIMPLE;
2649                         c->Request.Type.Direction = XFER_NONE;
2650                         c->Request.Timeout = 0;
2651                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2652                         c->Request.CDB[0] = cmd;        /* reset */
2653                         c->Request.CDB[1] = CCISS_RESET_TYPE_TARGET;
2654                         break;
2655                 case CCISS_NOOP_MSG:
2656                         c->Request.CDBLen = 1;
2657                         c->Request.Type.Attribute = ATTR_SIMPLE;
2658                         c->Request.Type.Direction = XFER_WRITE;
2659                         c->Request.Timeout = 0;
2660                         c->Request.CDB[0] = cmd;
2661                         break;
2662                 default:
2663                         dev_warn(&h->pdev->dev,
2664                                 "unknown message type %d\n", cmd);
2665                         return IO_ERROR;
2666                 }
2667         } else {
2668                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2669                 return IO_ERROR;
2670         }
2671         /* Fill in the scatter gather information */
2672         if (size > 0) {
2673                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2674                                                              buff, size,
2675                                                              PCI_DMA_BIDIRECTIONAL);
2676                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2677                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2678                 c->SG[0].Len = size;
2679                 c->SG[0].Ext = 0;       /* we are not chaining */
2680         }
2681         return status;
2682 }
2683
2684 static int cciss_send_reset(ctlr_info_t *h, unsigned char *scsi3addr,
2685                             u8 reset_type)
2686 {
2687         CommandList_struct *c;
2688         int return_status;
2689
2690         c = cmd_alloc(h);
2691         if (!c)
2692                 return -ENOMEM;
2693         return_status = fill_cmd(h, c, CCISS_RESET_MSG, NULL, 0, 0,
2694                 CTLR_LUNID, TYPE_MSG);
2695         c->Request.CDB[1] = reset_type; /* fill_cmd defaults to target reset */
2696         if (return_status != IO_OK) {
2697                 cmd_special_free(h, c);
2698                 return return_status;
2699         }
2700         c->waiting = NULL;
2701         enqueue_cmd_and_start_io(h, c);
2702         /* Don't wait for completion, the reset won't complete.  Don't free
2703          * the command either.  This is the last command we will send before
2704          * re-initializing everything, so it doesn't matter and won't leak.
2705          */
2706         return 0;
2707 }
2708
2709 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2710 {
2711         switch (c->err_info->ScsiStatus) {
2712         case SAM_STAT_GOOD:
2713                 return IO_OK;
2714         case SAM_STAT_CHECK_CONDITION:
2715                 switch (0xf & c->err_info->SenseInfo[2]) {
2716                 case 0: return IO_OK; /* no sense */
2717                 case 1: return IO_OK; /* recovered error */
2718                 default:
2719                         if (check_for_unit_attention(h, c))
2720                                 return IO_NEEDS_RETRY;
2721                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2722                                 "check condition, sense key = 0x%02x\n",
2723                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2724                 }
2725                 break;
2726         default:
2727                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2728                         "scsi status = 0x%02x\n",
2729                         c->Request.CDB[0], c->err_info->ScsiStatus);
2730                 break;
2731         }
2732         return IO_ERROR;
2733 }
2734
2735 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2736 {
2737         int return_status = IO_OK;
2738
2739         if (c->err_info->CommandStatus == CMD_SUCCESS)
2740                 return IO_OK;
2741
2742         switch (c->err_info->CommandStatus) {
2743         case CMD_TARGET_STATUS:
2744                 return_status = check_target_status(h, c);
2745                 break;
2746         case CMD_DATA_UNDERRUN:
2747         case CMD_DATA_OVERRUN:
2748                 /* expected for inquiry and report lun commands */
2749                 break;
2750         case CMD_INVALID:
2751                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2752                        "reported invalid\n", c->Request.CDB[0]);
2753                 return_status = IO_ERROR;
2754                 break;
2755         case CMD_PROTOCOL_ERR:
2756                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2757                        "protocol error\n", c->Request.CDB[0]);
2758                 return_status = IO_ERROR;
2759                 break;
2760         case CMD_HARDWARE_ERR:
2761                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2762                        " hardware error\n", c->Request.CDB[0]);
2763                 return_status = IO_ERROR;
2764                 break;
2765         case CMD_CONNECTION_LOST:
2766                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2767                        "connection lost\n", c->Request.CDB[0]);
2768                 return_status = IO_ERROR;
2769                 break;
2770         case CMD_ABORTED:
2771                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2772                        "aborted\n", c->Request.CDB[0]);
2773                 return_status = IO_ERROR;
2774                 break;
2775         case CMD_ABORT_FAILED:
2776                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2777                        "abort failed\n", c->Request.CDB[0]);
2778                 return_status = IO_ERROR;
2779                 break;
2780         case CMD_UNSOLICITED_ABORT:
2781                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2782                         c->Request.CDB[0]);
2783                 return_status = IO_NEEDS_RETRY;
2784                 break;
2785         case CMD_UNABORTABLE:
2786                 dev_warn(&h->pdev->dev, "cmd unabortable\n");
2787                 return_status = IO_ERROR;
2788                 break;
2789         default:
2790                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2791                        "unknown status %x\n", c->Request.CDB[0],
2792                        c->err_info->CommandStatus);
2793                 return_status = IO_ERROR;
2794         }
2795         return return_status;
2796 }
2797
2798 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2799         int attempt_retry)
2800 {
2801         DECLARE_COMPLETION_ONSTACK(wait);
2802         u64bit buff_dma_handle;
2803         int return_status = IO_OK;
2804
2805 resend_cmd2:
2806         c->waiting = &wait;
2807         enqueue_cmd_and_start_io(h, c);
2808
2809         wait_for_completion(&wait);
2810
2811         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2812                 goto command_done;
2813
2814         return_status = process_sendcmd_error(h, c);
2815
2816         if (return_status == IO_NEEDS_RETRY &&
2817                 c->retry_count < MAX_CMD_RETRIES) {
2818                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2819                         c->Request.CDB[0]);
2820                 c->retry_count++;
2821                 /* erase the old error information */
2822                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2823                 return_status = IO_OK;
2824                 reinit_completion(&wait);
2825                 goto resend_cmd2;
2826         }
2827
2828 command_done:
2829         /* unlock the buffers from DMA */
2830         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2831         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2832         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2833                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2834         return return_status;
2835 }
2836
2837 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2838                            __u8 page_code, unsigned char scsi3addr[],
2839                         int cmd_type)
2840 {
2841         CommandList_struct *c;
2842         int return_status;
2843
2844         c = cmd_special_alloc(h);
2845         if (!c)
2846                 return -ENOMEM;
2847         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2848                 scsi3addr, cmd_type);
2849         if (return_status == IO_OK)
2850                 return_status = sendcmd_withirq_core(h, c, 1);
2851
2852         cmd_special_free(h, c);
2853         return return_status;
2854 }
2855
2856 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2857                                    sector_t total_size,
2858                                    unsigned int block_size,
2859                                    InquiryData_struct *inq_buff,
2860                                    drive_info_struct *drv)
2861 {
2862         int return_code;
2863         unsigned long t;
2864         unsigned char scsi3addr[8];
2865
2866         memset(inq_buff, 0, sizeof(InquiryData_struct));
2867         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2868         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2869                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2870         if (return_code == IO_OK) {
2871                 if (inq_buff->data_byte[8] == 0xFF) {
2872                         dev_warn(&h->pdev->dev,
2873                                "reading geometry failed, volume "
2874                                "does not support reading geometry\n");
2875                         drv->heads = 255;
2876                         drv->sectors = 32;      /* Sectors per track */
2877                         drv->cylinders = total_size + 1;
2878                         drv->raid_level = RAID_UNKNOWN;
2879                 } else {
2880                         drv->heads = inq_buff->data_byte[6];
2881                         drv->sectors = inq_buff->data_byte[7];
2882                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2883                         drv->cylinders += inq_buff->data_byte[5];
2884                         drv->raid_level = inq_buff->data_byte[8];
2885                 }
2886                 drv->block_size = block_size;
2887                 drv->nr_blocks = total_size + 1;
2888                 t = drv->heads * drv->sectors;
2889                 if (t > 1) {
2890                         sector_t real_size = total_size + 1;
2891                         unsigned long rem = sector_div(real_size, t);
2892                         if (rem)
2893                                 real_size++;
2894                         drv->cylinders = real_size;
2895                 }
2896         } else {                /* Get geometry failed */
2897                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2898         }
2899 }
2900
2901 static void
2902 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2903                     unsigned int *block_size)
2904 {
2905         ReadCapdata_struct *buf;
2906         int return_code;
2907         unsigned char scsi3addr[8];
2908
2909         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2910         if (!buf) {
2911                 dev_warn(&h->pdev->dev, "out of memory\n");
2912                 return;
2913         }
2914
2915         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2916         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2917                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2918         if (return_code == IO_OK) {
2919                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2920                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2921         } else {                /* read capacity command failed */
2922                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2923                 *total_size = 0;
2924                 *block_size = BLOCK_SIZE;
2925         }
2926         kfree(buf);
2927 }
2928
2929 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2930         sector_t *total_size, unsigned int *block_size)
2931 {
2932         ReadCapdata_struct_16 *buf;
2933         int return_code;
2934         unsigned char scsi3addr[8];
2935
2936         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2937         if (!buf) {
2938                 dev_warn(&h->pdev->dev, "out of memory\n");
2939                 return;
2940         }
2941
2942         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2943         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2944                 buf, sizeof(ReadCapdata_struct_16),
2945                         0, scsi3addr, TYPE_CMD);
2946         if (return_code == IO_OK) {
2947                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2948                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2949         } else {                /* read capacity command failed */
2950                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2951                 *total_size = 0;
2952                 *block_size = BLOCK_SIZE;
2953         }
2954         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2955                (unsigned long long)*total_size+1, *block_size);
2956         kfree(buf);
2957 }
2958
2959 static int cciss_revalidate(struct gendisk *disk)
2960 {
2961         ctlr_info_t *h = get_host(disk);
2962         drive_info_struct *drv = get_drv(disk);
2963         int logvol;
2964         int FOUND = 0;
2965         unsigned int block_size;
2966         sector_t total_size;
2967         InquiryData_struct *inq_buff = NULL;
2968
2969         for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2970                 if (!h->drv[logvol])
2971                         continue;
2972                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2973                         sizeof(drv->LunID)) == 0) {
2974                         FOUND = 1;
2975                         break;
2976                 }
2977         }
2978
2979         if (!FOUND)
2980                 return 1;
2981
2982         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2983         if (inq_buff == NULL) {
2984                 dev_warn(&h->pdev->dev, "out of memory\n");
2985                 return 1;
2986         }
2987         if (h->cciss_read == CCISS_READ_10) {
2988                 cciss_read_capacity(h, logvol,
2989                                         &total_size, &block_size);
2990         } else {
2991                 cciss_read_capacity_16(h, logvol,
2992                                         &total_size, &block_size);
2993         }
2994         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2995                                inq_buff, drv);
2996
2997         blk_queue_logical_block_size(drv->queue, drv->block_size);
2998         set_capacity(disk, drv->nr_blocks);
2999
3000         kfree(inq_buff);
3001         return 0;
3002 }
3003
3004 /*
3005  * Map (physical) PCI mem into (virtual) kernel space
3006  */
3007 static void __iomem *remap_pci_mem(ulong base, ulong size)
3008 {
3009         ulong page_base = ((ulong) base) & PAGE_MASK;
3010         ulong page_offs = ((ulong) base) - page_base;
3011         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
3012
3013         return page_remapped ? (page_remapped + page_offs) : NULL;
3014 }
3015
3016 /*
3017  * Takes jobs of the Q and sends them to the hardware, then puts it on
3018  * the Q to wait for completion.
3019  */
3020 static void start_io(ctlr_info_t *h)
3021 {
3022         CommandList_struct *c;
3023
3024         while (!list_empty(&h->reqQ)) {
3025                 c = list_entry(h->reqQ.next, CommandList_struct, list);
3026                 /* can't do anything if fifo is full */
3027                 if ((h->access.fifo_full(h))) {
3028                         dev_warn(&h->pdev->dev, "fifo full\n");
3029                         break;
3030                 }
3031
3032                 /* Get the first entry from the Request Q */
3033                 removeQ(c);
3034                 h->Qdepth--;
3035
3036                 /* Tell the controller execute command */
3037                 h->access.submit_command(h, c);
3038
3039                 /* Put job onto the completed Q */
3040                 addQ(&h->cmpQ, c);
3041         }
3042 }
3043
3044 /* Assumes that h->lock is held. */
3045 /* Zeros out the error record and then resends the command back */
3046 /* to the controller */
3047 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
3048 {
3049         /* erase the old error information */
3050         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
3051
3052         /* add it to software queue and then send it to the controller */
3053         addQ(&h->reqQ, c);
3054         h->Qdepth++;
3055         if (h->Qdepth > h->maxQsinceinit)
3056                 h->maxQsinceinit = h->Qdepth;
3057
3058         start_io(h);
3059 }
3060
3061 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
3062         unsigned int msg_byte, unsigned int host_byte,
3063         unsigned int driver_byte)
3064 {
3065         /* inverse of macros in scsi.h */
3066         return (scsi_status_byte & 0xff) |
3067                 ((msg_byte & 0xff) << 8) |
3068                 ((host_byte & 0xff) << 16) |
3069                 ((driver_byte & 0xff) << 24);
3070 }
3071
3072 static inline int evaluate_target_status(ctlr_info_t *h,
3073                         CommandList_struct *cmd, int *retry_cmd)
3074 {
3075         unsigned char sense_key;
3076         unsigned char status_byte, msg_byte, host_byte, driver_byte;
3077         int error_value;
3078
3079         *retry_cmd = 0;
3080         /* If we get in here, it means we got "target status", that is, scsi status */
3081         status_byte = cmd->err_info->ScsiStatus;
3082         driver_byte = DRIVER_OK;
3083         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3084
3085         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3086                 host_byte = DID_PASSTHROUGH;
3087         else
3088                 host_byte = DID_OK;
3089
3090         error_value = make_status_bytes(status_byte, msg_byte,
3091                 host_byte, driver_byte);
3092
3093         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3094                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3095                         dev_warn(&h->pdev->dev, "cmd %p "
3096                                "has SCSI Status 0x%x\n",
3097                                cmd, cmd->err_info->ScsiStatus);
3098                 return error_value;
3099         }
3100
3101         /* check the sense key */
3102         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3103         /* no status or recovered error */
3104         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3105             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3106                 error_value = 0;
3107
3108         if (check_for_unit_attention(h, cmd)) {
3109                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3110                 return 0;
3111         }
3112
3113         /* Not SG_IO or similar? */
3114         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3115                 if (error_value != 0)
3116                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3117                                " sense key = 0x%x\n", cmd, sense_key);
3118                 return error_value;
3119         }
3120
3121         /* SG_IO or similar, copy sense data back */
3122         if (cmd->rq->sense) {
3123                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3124                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3125                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3126                         cmd->rq->sense_len);
3127         } else
3128                 cmd->rq->sense_len = 0;
3129
3130         return error_value;
3131 }
3132
3133 /* checks the status of the job and calls complete buffers to mark all
3134  * buffers for the completed job. Note that this function does not need
3135  * to hold the hba/queue lock.
3136  */
3137 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3138                                     int timeout)
3139 {
3140         int retry_cmd = 0;
3141         struct request *rq = cmd->rq;
3142
3143         rq->errors = 0;
3144
3145         if (timeout)
3146                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3147
3148         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3149                 goto after_error_processing;
3150
3151         switch (cmd->err_info->CommandStatus) {
3152         case CMD_TARGET_STATUS:
3153                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3154                 break;
3155         case CMD_DATA_UNDERRUN:
3156                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3157                         dev_warn(&h->pdev->dev, "cmd %p has"
3158                                " completed with data underrun "
3159                                "reported\n", cmd);
3160                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3161                 }
3162                 break;
3163         case CMD_DATA_OVERRUN:
3164                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3165                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3166                                " completed with data overrun "
3167                                "reported\n", cmd);
3168                 break;
3169         case CMD_INVALID:
3170                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3171                        "reported invalid\n", cmd);
3172                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3173                         cmd->err_info->CommandStatus, DRIVER_OK,
3174                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3175                                 DID_PASSTHROUGH : DID_ERROR);
3176                 break;
3177         case CMD_PROTOCOL_ERR:
3178                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3179                        "protocol error\n", cmd);
3180                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3181                         cmd->err_info->CommandStatus, DRIVER_OK,
3182                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3183                                 DID_PASSTHROUGH : DID_ERROR);
3184                 break;
3185         case CMD_HARDWARE_ERR:
3186                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3187                        " hardware error\n", cmd);
3188                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3189                         cmd->err_info->CommandStatus, DRIVER_OK,
3190                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3191                                 DID_PASSTHROUGH : DID_ERROR);
3192                 break;
3193         case CMD_CONNECTION_LOST:
3194                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3195                        "connection lost\n", cmd);
3196                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3197                         cmd->err_info->CommandStatus, DRIVER_OK,
3198                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3199                                 DID_PASSTHROUGH : DID_ERROR);
3200                 break;
3201         case CMD_ABORTED:
3202                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3203                        "aborted\n", cmd);
3204                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3205                         cmd->err_info->CommandStatus, DRIVER_OK,
3206                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3207                                 DID_PASSTHROUGH : DID_ABORT);
3208                 break;
3209         case CMD_ABORT_FAILED:
3210                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3211                        "abort failed\n", cmd);
3212                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3213                         cmd->err_info->CommandStatus, DRIVER_OK,
3214                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3215                                 DID_PASSTHROUGH : DID_ERROR);
3216                 break;
3217         case CMD_UNSOLICITED_ABORT:
3218                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3219                        "abort %p\n", h->ctlr, cmd);
3220                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3221                         retry_cmd = 1;
3222                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3223                         cmd->retry_count++;
3224                 } else
3225                         dev_warn(&h->pdev->dev,
3226                                 "%p retried too many times\n", cmd);
3227                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3228                         cmd->err_info->CommandStatus, DRIVER_OK,
3229                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3230                                 DID_PASSTHROUGH : DID_ABORT);
3231                 break;
3232         case CMD_TIMEOUT:
3233                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3234                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3235                         cmd->err_info->CommandStatus, DRIVER_OK,
3236                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3237                                 DID_PASSTHROUGH : DID_ERROR);
3238                 break;
3239         case CMD_UNABORTABLE:
3240                 dev_warn(&h->pdev->dev, "cmd %p unabortable\n", cmd);
3241                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3242                         cmd->err_info->CommandStatus, DRIVER_OK,
3243                         cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC ?
3244                                 DID_PASSTHROUGH : DID_ERROR);
3245                 break;
3246         default:
3247                 dev_warn(&h->pdev->dev, "cmd %p returned "
3248                        "unknown status %x\n", cmd,
3249                        cmd->err_info->CommandStatus);
3250                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3251                         cmd->err_info->CommandStatus, DRIVER_OK,
3252                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3253                                 DID_PASSTHROUGH : DID_ERROR);
3254         }
3255
3256 after_error_processing:
3257
3258         /* We need to return this command */
3259         if (retry_cmd) {
3260                 resend_cciss_cmd(h, cmd);
3261                 return;
3262         }
3263         cmd->rq->completion_data = cmd;
3264         blk_complete_request(cmd->rq);
3265 }
3266
3267 static inline u32 cciss_tag_contains_index(u32 tag)
3268 {
3269 #define DIRECT_LOOKUP_BIT 0x10
3270         return tag & DIRECT_LOOKUP_BIT;
3271 }
3272
3273 static inline u32 cciss_tag_to_index(u32 tag)
3274 {
3275 #define DIRECT_LOOKUP_SHIFT 5
3276         return tag >> DIRECT_LOOKUP_SHIFT;
3277 }
3278
3279 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag)
3280 {
3281 #define CCISS_PERF_ERROR_BITS ((1 << DIRECT_LOOKUP_SHIFT) - 1)
3282 #define CCISS_SIMPLE_ERROR_BITS 0x03
3283         if (likely(h->transMethod & CFGTBL_Trans_Performant))
3284                 return tag & ~CCISS_PERF_ERROR_BITS;
3285         return tag & ~CCISS_SIMPLE_ERROR_BITS;
3286 }
3287
3288 static inline void cciss_mark_tag_indexed(u32 *tag)
3289 {
3290         *tag |= DIRECT_LOOKUP_BIT;
3291 }
3292
3293 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3294 {
3295         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3296 }
3297
3298 /*
3299  * Get a request and submit it to the controller.
3300  */
3301 static void do_cciss_request(struct request_queue *q)
3302 {
3303         ctlr_info_t *h = q->queuedata;
3304         CommandList_struct *c;
3305         sector_t start_blk;
3306         int seg;
3307         struct request *creq;
3308         u64bit temp64;
3309         struct scatterlist *tmp_sg;
3310         SGDescriptor_struct *curr_sg;
3311         drive_info_struct *drv;
3312         int i, dir;
3313         int sg_index = 0;
3314         int chained = 0;
3315
3316       queue:
3317         creq = blk_peek_request(q);
3318         if (!creq)
3319                 goto startio;
3320
3321         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3322
3323         c = cmd_alloc(h);
3324         if (!c)
3325                 goto full;
3326
3327         blk_start_request(creq);
3328
3329         tmp_sg = h->scatter_list[c->cmdindex];
3330         spin_unlock_irq(q->queue_lock);
3331
3332         c->cmd_type = CMD_RWREQ;
3333         c->rq = creq;
3334
3335         /* fill in the request */
3336         drv = creq->rq_disk->private_data;
3337         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3338         /* got command from pool, so use the command block index instead */
3339         /* for direct lookups. */
3340         /* The first 2 bits are reserved for controller error reporting. */
3341         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3342         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3343         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3344         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3345         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3346         c->Request.Type.Attribute = ATTR_SIMPLE;
3347         c->Request.Type.Direction =
3348             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3349         c->Request.Timeout = 0; /* Don't time out */
3350         c->Request.CDB[0] =
3351             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3352         start_blk = blk_rq_pos(creq);
3353         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3354                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3355         sg_init_table(tmp_sg, h->maxsgentries);
3356         seg = blk_rq_map_sg(q, creq, tmp_sg);
3357
3358         /* get the DMA records for the setup */
3359         if (c->Request.Type.Direction == XFER_READ)
3360                 dir = PCI_DMA_FROMDEVICE;
3361         else
3362                 dir = PCI_DMA_TODEVICE;
3363
3364         curr_sg = c->SG;
3365         sg_index = 0;
3366         chained = 0;
3367
3368         for (i = 0; i < seg; i++) {
3369                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3370                         !chained && ((seg - i) > 1)) {
3371                         /* Point to next chain block. */
3372                         curr_sg = h->cmd_sg_list[c->cmdindex];
3373                         sg_index = 0;
3374                         chained = 1;
3375                 }
3376                 curr_sg[sg_index].Len = tmp_sg[i].length;
3377                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3378                                                 tmp_sg[i].offset,
3379                                                 tmp_sg[i].length, dir);
3380                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3381                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3382                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3383                 ++sg_index;
3384         }
3385         if (chained)
3386                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3387                         (seg - (h->max_cmd_sgentries - 1)) *
3388                                 sizeof(SGDescriptor_struct));
3389
3390         /* track how many SG entries we are using */
3391         if (seg > h->maxSG)
3392                 h->maxSG = seg;
3393
3394         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3395                         "chained[%d]\n",
3396                         blk_rq_sectors(creq), seg, chained);
3397
3398         c->Header.SGTotal = seg + chained;
3399         if (seg <= h->max_cmd_sgentries)
3400                 c->Header.SGList = c->Header.SGTotal;
3401         else
3402                 c->Header.SGList = h->max_cmd_sgentries;
3403         set_performant_mode(h, c);
3404
3405         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3406                 if(h->cciss_read == CCISS_READ_10) {
3407                         c->Request.CDB[1] = 0;
3408                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3409                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3410                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3411                         c->Request.CDB[5] = start_blk & 0xff;
3412                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3413                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3414                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3415                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3416                 } else {
3417                         u32 upper32 = upper_32_bits(start_blk);
3418
3419                         c->Request.CDBLen = 16;
3420                         c->Request.CDB[1]= 0;
3421                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3422                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3423                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3424                         c->Request.CDB[5]= upper32 & 0xff;
3425                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3426                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3427                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3428                         c->Request.CDB[9]= start_blk & 0xff;
3429                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3430                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3431                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3432                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3433                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3434                 }
3435         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3436                 c->Request.CDBLen = creq->cmd_len;
3437                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3438         } else {
3439                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3440                         creq->cmd_type);
3441                 BUG();
3442         }
3443
3444         spin_lock_irq(q->queue_lock);
3445
3446         addQ(&h->reqQ, c);
3447         h->Qdepth++;
3448         if (h->Qdepth > h->maxQsinceinit)
3449                 h->maxQsinceinit = h->Qdepth;
3450
3451         goto queue;
3452 full:
3453         blk_stop_queue(q);
3454 startio:
3455         /* We will already have the driver lock here so not need
3456          * to lock it.
3457          */
3458         start_io(h);
3459 }
3460
3461 static inline unsigned long get_next_completion(ctlr_info_t *h)
3462 {
3463         return h->access.command_completed(h);
3464 }
3465
3466 static inline int interrupt_pending(ctlr_info_t *h)
3467 {
3468         return h->access.intr_pending(h);
3469 }
3470
3471 static inline long interrupt_not_for_us(ctlr_info_t *h)
3472 {
3473         return ((h->access.intr_pending(h) == 0) ||
3474                 (h->interrupts_enabled == 0));
3475 }
3476
3477 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3478                         u32 raw_tag)
3479 {
3480         if (unlikely(tag_index >= h->nr_cmds)) {
3481                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3482                 return 1;
3483         }
3484         return 0;
3485 }
3486
3487 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3488                                 u32 raw_tag)
3489 {
3490         removeQ(c);
3491         if (likely(c->cmd_type == CMD_RWREQ))
3492                 complete_command(h, c, 0);
3493         else if (c->cmd_type == CMD_IOCTL_PEND)
3494                 complete(c->waiting);
3495 #ifdef CONFIG_CISS_SCSI_TAPE
3496         else if (c->cmd_type == CMD_SCSI)
3497                 complete_scsi_command(c, 0, raw_tag);
3498 #endif
3499 }
3500
3501 static inline u32 next_command(ctlr_info_t *h)
3502 {
3503         u32 a;
3504
3505         if (unlikely(!(h->transMethod & CFGTBL_Trans_Performant)))
3506                 return h->access.command_completed(h);
3507
3508         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3509                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3510                 (h->reply_pool_head)++;
3511                 h->commands_outstanding--;
3512         } else {
3513                 a = FIFO_EMPTY;
3514         }
3515         /* Check for wraparound */
3516         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3517                 h->reply_pool_head = h->reply_pool;
3518                 h->reply_pool_wraparound ^= 1;
3519         }
3520         return a;
3521 }
3522
3523 /* process completion of an indexed ("direct lookup") command */
3524 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3525 {
3526         u32 tag_index;
3527         CommandList_struct *c;
3528
3529         tag_index = cciss_tag_to_index(raw_tag);
3530         if (bad_tag(h, tag_index, raw_tag))
3531                 return next_command(h);
3532         c = h->cmd_pool + tag_index;
3533         finish_cmd(h, c, raw_tag);
3534         return next_command(h);
3535 }
3536
3537 /* process completion of a non-indexed command */
3538 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3539 {
3540         CommandList_struct *c = NULL;
3541         __u32 busaddr_masked, tag_masked;
3542
3543         tag_masked = cciss_tag_discard_error_bits(h, raw_tag);
3544         list_for_each_entry(c, &h->cmpQ, list) {
3545                 busaddr_masked = cciss_tag_discard_error_bits(h, c->busaddr);
3546                 if (busaddr_masked == tag_masked) {
3547                         finish_cmd(h, c, raw_tag);
3548                         return next_command(h);
3549                 }
3550         }
3551         bad_tag(h, h->nr_cmds + 1, raw_tag);
3552         return next_command(h);
3553 }
3554
3555 /* Some controllers, like p400, will give us one interrupt
3556  * after a soft reset, even if we turned interrupts off.
3557  * Only need to check for this in the cciss_xxx_discard_completions
3558  * functions.
3559  */
3560 static int ignore_bogus_interrupt(ctlr_info_t *h)
3561 {
3562         if (likely(!reset_devices))
3563                 return 0;
3564
3565         if (likely(h->interrupts_enabled))
3566                 return 0;
3567
3568         dev_info(&h->pdev->dev, "Received interrupt while interrupts disabled "
3569                 "(known firmware bug.)  Ignoring.\n");
3570
3571         return 1;
3572 }
3573
3574 static irqreturn_t cciss_intx_discard_completions(int irq, void *dev_id)
3575 {
3576         ctlr_info_t *h = dev_id;
3577         unsigned long flags;
3578         u32 raw_tag;
3579
3580         if (ignore_bogus_interrupt(h))
3581                 return IRQ_NONE;
3582
3583         if (interrupt_not_for_us(h))
3584                 return IRQ_NONE;
3585         spin_lock_irqsave(&h->lock, flags);
3586         while (interrupt_pending(h)) {
3587                 raw_tag = get_next_completion(h);
3588                 while (raw_tag != FIFO_EMPTY)
3589                         raw_tag = next_command(h);
3590         }
3591         spin_unlock_irqrestore(&h->lock, flags);
3592         return IRQ_HANDLED;
3593 }
3594
3595 static irqreturn_t cciss_msix_discard_completions(int irq, void *dev_id)
3596 {
3597         ctlr_info_t *h = dev_id;
3598         unsigned long flags;
3599         u32 raw_tag;
3600
3601         if (ignore_bogus_interrupt(h))
3602                 return IRQ_NONE;
3603
3604         spin_lock_irqsave(&h->lock, flags);
3605         raw_tag = get_next_completion(h);
3606         while (raw_tag != FIFO_EMPTY)
3607                 raw_tag = next_command(h);
3608         spin_unlock_irqrestore(&h->lock, flags);
3609         return IRQ_HANDLED;
3610 }
3611
3612 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3613 {
3614         ctlr_info_t *h = dev_id;
3615         unsigned long flags;
3616         u32 raw_tag;
3617
3618         if (interrupt_not_for_us(h))
3619                 return IRQ_NONE;
3620         spin_lock_irqsave(&h->lock, flags);
3621         while (interrupt_pending(h)) {
3622                 raw_tag = get_next_completion(h);
3623                 while (raw_tag != FIFO_EMPTY) {
3624                         if (cciss_tag_contains_index(raw_tag))
3625                                 raw_tag = process_indexed_cmd(h, raw_tag);
3626                         else
3627                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3628                 }
3629         }
3630         spin_unlock_irqrestore(&h->lock, flags);
3631         return IRQ_HANDLED;
3632 }
3633
3634 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3635  * check the interrupt pending register because it is not set.
3636  */
3637 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3638 {
3639         ctlr_info_t *h = dev_id;
3640         unsigned long flags;
3641         u32 raw_tag;
3642
3643         spin_lock_irqsave(&h->lock, flags);
3644         raw_tag = get_next_completion(h);
3645         while (raw_tag != FIFO_EMPTY) {
3646                 if (cciss_tag_contains_index(raw_tag))
3647                         raw_tag = process_indexed_cmd(h, raw_tag);
3648                 else
3649                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3650         }
3651         spin_unlock_irqrestore(&h->lock, flags);
3652         return IRQ_HANDLED;
3653 }
3654
3655 /**
3656  * add_to_scan_list() - add controller to rescan queue
3657  * @h:                Pointer to the controller.
3658  *
3659  * Adds the controller to the rescan queue if not already on the queue.
3660  *
3661  * returns 1 if added to the queue, 0 if skipped (could be on the
3662  * queue already, or the controller could be initializing or shutting
3663  * down).
3664  **/
3665 static int add_to_scan_list(struct ctlr_info *h)
3666 {
3667         struct ctlr_info *test_h;
3668         int found = 0;
3669         int ret = 0;
3670
3671         if (h->busy_initializing)
3672                 return 0;
3673
3674         if (!mutex_trylock(&h->busy_shutting_down))
3675                 return 0;
3676
3677         mutex_lock(&scan_mutex);
3678         list_for_each_entry(test_h, &scan_q, scan_list) {
3679                 if (test_h == h) {
3680                         found = 1;
3681                         break;
3682                 }
3683         }
3684         if (!found && !h->busy_scanning) {
3685                 reinit_completion(&h->scan_wait);
3686                 list_add_tail(&h->scan_list, &scan_q);
3687                 ret = 1;
3688         }
3689         mutex_unlock(&scan_mutex);
3690         mutex_unlock(&h->busy_shutting_down);
3691
3692         return ret;
3693 }
3694
3695 /**
3696  * remove_from_scan_list() - remove controller from rescan queue
3697  * @h:                     Pointer to the controller.
3698  *
3699  * Removes the controller from the rescan queue if present. Blocks if
3700  * the controller is currently conducting a rescan.  The controller
3701  * can be in one of three states:
3702  * 1. Doesn't need a scan
3703  * 2. On the scan list, but not scanning yet (we remove it)
3704  * 3. Busy scanning (and not on the list). In this case we want to wait for
3705  *    the scan to complete to make sure the scanning thread for this
3706  *    controller is completely idle.
3707  **/
3708 static void remove_from_scan_list(struct ctlr_info *h)
3709 {
3710         struct ctlr_info *test_h, *tmp_h;
3711
3712         mutex_lock(&scan_mutex);
3713         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3714                 if (test_h == h) { /* state 2. */
3715                         list_del(&h->scan_list);
3716                         complete_all(&h->scan_wait);
3717                         mutex_unlock(&scan_mutex);
3718                         return;
3719                 }
3720         }
3721         if (h->busy_scanning) { /* state 3. */
3722                 mutex_unlock(&scan_mutex);
3723                 wait_for_completion(&h->scan_wait);
3724         } else { /* state 1, nothing to do. */
3725                 mutex_unlock(&scan_mutex);
3726         }
3727 }
3728
3729 /**
3730  * scan_thread() - kernel thread used to rescan controllers
3731  * @data:        Ignored.
3732  *
3733  * A kernel thread used scan for drive topology changes on
3734  * controllers. The thread processes only one controller at a time
3735  * using a queue.  Controllers are added to the queue using
3736  * add_to_scan_list() and removed from the queue either after done
3737  * processing or using remove_from_scan_list().
3738  *
3739  * returns 0.
3740  **/
3741 static int scan_thread(void *data)
3742 {
3743         struct ctlr_info *h;
3744
3745         while (1) {
3746                 set_current_state(TASK_INTERRUPTIBLE);
3747                 schedule();
3748                 if (kthread_should_stop())
3749                         break;
3750
3751                 while (1) {
3752                         mutex_lock(&scan_mutex);
3753                         if (list_empty(&scan_q)) {
3754                                 mutex_unlock(&scan_mutex);
3755                                 break;
3756                         }
3757
3758                         h = list_entry(scan_q.next,
3759                                        struct ctlr_info,
3760                                        scan_list);
3761                         list_del(&h->scan_list);
3762                         h->busy_scanning = 1;
3763                         mutex_unlock(&scan_mutex);
3764
3765                         rebuild_lun_table(h, 0, 0);
3766                         complete_all(&h->scan_wait);
3767                         mutex_lock(&scan_mutex);
3768                         h->busy_scanning = 0;
3769                         mutex_unlock(&scan_mutex);
3770                 }
3771         }
3772
3773         return 0;
3774 }
3775
3776 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3777 {
3778         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3779                 return 0;
3780
3781         switch (c->err_info->SenseInfo[12]) {
3782         case STATE_CHANGED:
3783                 dev_warn(&h->pdev->dev, "a state change "
3784                         "detected, command retried\n");
3785                 return 1;
3786         break;
3787         case LUN_FAILED:
3788                 dev_warn(&h->pdev->dev, "LUN failure "
3789                         "detected, action required\n");
3790                 return 1;
3791         break;
3792         case REPORT_LUNS_CHANGED:
3793                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3794         /*
3795          * Here, we could call add_to_scan_list and wake up the scan thread,
3796          * except that it's quite likely that we will get more than one
3797          * REPORT_LUNS_CHANGED condition in quick succession, which means
3798          * that those which occur after the first one will likely happen
3799          * *during* the scan_thread's rescan.  And the rescan code is not
3800          * robust enough to restart in the middle, undoing what it has already
3801          * done, and it's not clear that it's even possible to do this, since
3802          * part of what it does is notify the block layer, which starts
3803          * doing it's own i/o to read partition tables and so on, and the
3804          * driver doesn't have visibility to know what might need undoing.
3805          * In any event, if possible, it is horribly complicated to get right
3806          * so we just don't do it for now.
3807          *
3808          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3809          */
3810                 return 1;
3811         break;
3812         case POWER_OR_RESET:
3813                 dev_warn(&h->pdev->dev,
3814                         "a power on or device reset detected\n");
3815                 return 1;
3816         break;
3817         case UNIT_ATTENTION_CLEARED:
3818                 dev_warn(&h->pdev->dev,
3819                         "unit attention cleared by another initiator\n");
3820                 return 1;
3821         break;
3822         default:
3823                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3824                 return 1;
3825         }
3826 }
3827
3828 /*
3829  *  We cannot read the structure directly, for portability we must use
3830  *   the io functions.
3831  *   This is for debug only.
3832  */
3833 static void print_cfg_table(ctlr_info_t *h)
3834 {
3835         int i;
3836         char temp_name[17];
3837         CfgTable_struct *tb = h->cfgtable;
3838
3839         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3840         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3841         for (i = 0; i < 4; i++)
3842                 temp_name[i] = readb(&(tb->Signature[i]));
3843         temp_name[4] = '\0';
3844         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3845         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3846                 readl(&(tb->SpecValence)));
3847         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3848                readl(&(tb->TransportSupport)));
3849         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3850                readl(&(tb->TransportActive)));
3851         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3852                readl(&(tb->HostWrite.TransportRequest)));
3853         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3854                readl(&(tb->HostWrite.CoalIntDelay)));
3855         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3856                readl(&(tb->HostWrite.CoalIntCount)));
3857         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3858                readl(&(tb->CmdsOutMax)));
3859         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3860                 readl(&(tb->BusTypes)));
3861         for (i = 0; i < 16; i++)
3862                 temp_name[i] = readb(&(tb->ServerName[i]));
3863         temp_name[16] = '\0';
3864         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3865         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3866                 readl(&(tb->HeartBeat)));
3867 }
3868
3869 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3870 {
3871         int i, offset, mem_type, bar_type;
3872         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3873                 return 0;
3874         offset = 0;
3875         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3876                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3877                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3878                         offset += 4;
3879                 else {
3880                         mem_type = pci_resource_flags(pdev, i) &
3881                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3882                         switch (mem_type) {
3883                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3884                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3885                                 offset += 4;    /* 32 bit */
3886                                 break;
3887                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3888                                 offset += 8;
3889                                 break;
3890                         default:        /* reserved in PCI 2.2 */
3891                                 dev_warn(&pdev->dev,
3892                                        "Base address is invalid\n");
3893                                 return -1;
3894                                 break;
3895                         }
3896                 }
3897                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3898                         return i + 1;
3899         }
3900         return -1;
3901 }
3902
3903 /* Fill in bucket_map[], given nsgs (the max number of
3904  * scatter gather elements supported) and bucket[],
3905  * which is an array of 8 integers.  The bucket[] array
3906  * contains 8 different DMA transfer sizes (in 16
3907  * byte increments) which the controller uses to fetch
3908  * commands.  This function fills in bucket_map[], which
3909  * maps a given number of scatter gather elements to one of
3910  * the 8 DMA transfer sizes.  The point of it is to allow the
3911  * controller to only do as much DMA as needed to fetch the
3912  * command, with the DMA transfer size encoded in the lower
3913  * bits of the command address.
3914  */
3915 static void  calc_bucket_map(int bucket[], int num_buckets,
3916         int nsgs, int *bucket_map)
3917 {
3918         int i, j, b, size;
3919
3920         /* even a command with 0 SGs requires 4 blocks */
3921 #define MINIMUM_TRANSFER_BLOCKS 4
3922 #define NUM_BUCKETS 8
3923         /* Note, bucket_map must have nsgs+1 entries. */
3924         for (i = 0; i <= nsgs; i++) {
3925                 /* Compute size of a command with i SG entries */
3926                 size = i + MINIMUM_TRANSFER_BLOCKS;
3927                 b = num_buckets; /* Assume the biggest bucket */
3928                 /* Find the bucket that is just big enough */
3929                 for (j = 0; j < 8; j++) {
3930                         if (bucket[j] >= size) {
3931                                 b = j;
3932                                 break;
3933                         }
3934                 }
3935                 /* for a command with i SG entries, use bucket b. */
3936                 bucket_map[i] = b;
3937         }
3938 }
3939
3940 static void cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3941 {
3942         int i;
3943
3944         /* under certain very rare conditions, this can take awhile.
3945          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3946          * as we enter this code.) */
3947         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3948                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3949                         break;
3950                 usleep_range(10000, 20000);
3951         }
3952 }
3953
3954 static void cciss_enter_performant_mode(ctlr_info_t *h, u32 use_short_tags)
3955 {
3956         /* This is a bit complicated.  There are 8 registers on
3957          * the controller which we write to to tell it 8 different
3958          * sizes of commands which there may be.  It's a way of
3959          * reducing the DMA done to fetch each command.  Encoded into
3960          * each command's tag are 3 bits which communicate to the controller
3961          * which of the eight sizes that command fits within.  The size of
3962          * each command depends on how many scatter gather entries there are.
3963          * Each SG entry requires 16 bytes.  The eight registers are programmed
3964          * with the number of 16-byte blocks a command of that size requires.
3965          * The smallest command possible requires 5 such 16 byte blocks.
3966          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3967          * blocks.  Note, this only extends to the SG entries contained
3968          * within the command block, and does not extend to chained blocks
3969          * of SG elements.   bft[] contains the eight values we write to
3970          * the registers.  They are not evenly distributed, but have more
3971          * sizes for small commands, and fewer sizes for larger commands.
3972          */
3973         __u32 trans_offset;
3974         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3975                         /*
3976                          *  5 = 1 s/g entry or 4k
3977                          *  6 = 2 s/g entry or 8k
3978                          *  8 = 4 s/g entry or 16k
3979                          * 10 = 6 s/g entry or 24k
3980                          */
3981         unsigned long register_value;
3982         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3983
3984         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3985
3986         /* Controller spec: zero out this buffer. */
3987         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3988         h->reply_pool_head = h->reply_pool;
3989
3990         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3991         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3992                                 h->blockFetchTable);
3993         writel(bft[0], &h->transtable->BlockFetch0);
3994         writel(bft[1], &h->transtable->BlockFetch1);
3995         writel(bft[2], &h->transtable->BlockFetch2);
3996         writel(bft[3], &h->transtable->BlockFetch3);
3997         writel(bft[4], &h->transtable->BlockFetch4);
3998         writel(bft[5], &h->transtable->BlockFetch5);
3999         writel(bft[6], &h->transtable->BlockFetch6);
4000         writel(bft[7], &h->transtable->BlockFetch7);
4001
4002         /* size of controller ring buffer */
4003         writel(h->max_commands, &h->transtable->RepQSize);
4004         writel(1, &h->transtable->RepQCount);
4005         writel(0, &h->transtable->RepQCtrAddrLow32);
4006         writel(0, &h->transtable->RepQCtrAddrHigh32);
4007         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
4008         writel(0, &h->transtable->RepQAddr0High32);
4009         writel(CFGTBL_Trans_Performant | use_short_tags,
4010                         &(h->cfgtable->HostWrite.TransportRequest));
4011
4012         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
4013         cciss_wait_for_mode_change_ack(h);
4014         register_value = readl(&(h->cfgtable->TransportActive));
4015         if (!(register_value & CFGTBL_Trans_Performant))
4016                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
4017                                         " performant mode\n");
4018 }
4019
4020 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h)
4021 {
4022         __u32 trans_support;
4023
4024         if (cciss_simple_mode)
4025                 return;
4026
4027         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
4028         /* Attempt to put controller into performant mode if supported */
4029         /* Does board support performant mode? */
4030         trans_support = readl(&(h->cfgtable->TransportSupport));
4031         if (!(trans_support & PERFORMANT_MODE))
4032                 return;
4033
4034         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
4035         /* Performant mode demands commands on a 32 byte boundary
4036          * pci_alloc_consistent aligns on page boundarys already.
4037          * Just need to check if divisible by 32
4038          */
4039         if ((sizeof(CommandList_struct) % 32) != 0) {
4040                 dev_warn(&h->pdev->dev, "%s %d %s\n",
4041                         "cciss info: command size[",
4042                         (int)sizeof(CommandList_struct),
4043                         "] not divisible by 32, no performant mode..\n");
4044                 return;
4045         }
4046
4047         /* Performant mode ring buffer and supporting data structures */
4048         h->reply_pool = (__u64 *)pci_alloc_consistent(
4049                 h->pdev, h->max_commands * sizeof(__u64),
4050                 &(h->reply_pool_dhandle));
4051
4052         /* Need a block fetch table for performant mode */
4053         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
4054                 sizeof(__u32)), GFP_KERNEL);
4055
4056         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
4057                 goto clean_up;
4058
4059         cciss_enter_performant_mode(h,
4060                 trans_support & CFGTBL_Trans_use_short_tags);
4061
4062         /* Change the access methods to the performant access methods */
4063         h->access = SA5_performant_access;
4064         h->transMethod = CFGTBL_Trans_Performant;
4065
4066         return;
4067 clean_up:
4068         kfree(h->blockFetchTable);
4069         if (h->reply_pool)
4070                 pci_free_consistent(h->pdev,
4071                                 h->max_commands * sizeof(__u64),
4072                                 h->reply_pool,
4073                                 h->reply_pool_dhandle);
4074         return;
4075
4076 } /* cciss_put_controller_into_performant_mode */
4077
4078 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
4079  * controllers that are capable. If not, we use IO-APIC mode.
4080  */
4081
4082 static void cciss_interrupt_mode(ctlr_info_t *h)
4083 {
4084 #ifdef CONFIG_PCI_MSI
4085         int err;
4086         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
4087         {0, 2}, {0, 3}
4088         };
4089
4090         /* Some boards advertise MSI but don't really support it */
4091         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
4092             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
4093                 goto default_int_mode;
4094
4095         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
4096                 err = pci_enable_msix_exact(h->pdev, cciss_msix_entries, 4);
4097                 if (!err) {
4098                         h->intr[0] = cciss_msix_entries[0].vector;
4099                         h->intr[1] = cciss_msix_entries[1].vector;
4100                         h->intr[2] = cciss_msix_entries[2].vector;
4101                         h->intr[3] = cciss_msix_entries[3].vector;
4102                         h->msix_vector = 1;
4103                         return;
4104                 } else {
4105                         dev_warn(&h->pdev->dev,
4106                                 "MSI-X init failed %d\n", err);
4107                 }
4108         }
4109         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
4110                 if (!pci_enable_msi(h->pdev))
4111                         h->msi_vector = 1;
4112                 else
4113                         dev_warn(&h->pdev->dev, "MSI init failed\n");
4114         }
4115 default_int_mode:
4116 #endif                          /* CONFIG_PCI_MSI */
4117         /* if we get here we're going to use the default interrupt mode */
4118         h->intr[h->intr_mode] = h->pdev->irq;
4119         return;
4120 }
4121
4122 static int cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
4123 {
4124         int i;
4125         u32 subsystem_vendor_id, subsystem_device_id;
4126
4127         subsystem_vendor_id = pdev->subsystem_vendor;
4128         subsystem_device_id = pdev->subsystem_device;
4129         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
4130                         subsystem_vendor_id;
4131
4132         for (i = 0; i < ARRAY_SIZE(products); i++) {
4133                 /* Stand aside for hpsa driver on request */
4134                 if (cciss_allow_hpsa)
4135                         return -ENODEV;
4136                 if (*board_id == products[i].board_id)
4137                         return i;
4138         }
4139         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4140                 *board_id);
4141         return -ENODEV;
4142 }
4143
4144 static inline bool cciss_board_disabled(ctlr_info_t *h)
4145 {
4146         u16 command;
4147
4148         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4149         return ((command & PCI_COMMAND_MEMORY) == 0);
4150 }
4151
4152 static int cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4153                                      unsigned long *memory_bar)
4154 {
4155         int i;
4156
4157         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4158                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4159                         /* addressing mode bits already removed */
4160                         *memory_bar = pci_resource_start(pdev, i);
4161                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4162                                 *memory_bar);
4163                         return 0;
4164                 }
4165         dev_warn(&pdev->dev, "no memory BAR found\n");
4166         return -ENODEV;
4167 }
4168
4169 static int cciss_wait_for_board_state(struct pci_dev *pdev,
4170                                       void __iomem *vaddr, int wait_for_ready)
4171 #define BOARD_READY 1
4172 #define BOARD_NOT_READY 0
4173 {
4174         int i, iterations;
4175         u32 scratchpad;
4176
4177         if (wait_for_ready)
4178                 iterations = CCISS_BOARD_READY_ITERATIONS;
4179         else
4180                 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
4181
4182         for (i = 0; i < iterations; i++) {
4183                 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
4184                 if (wait_for_ready) {
4185                         if (scratchpad == CCISS_FIRMWARE_READY)
4186                                 return 0;
4187                 } else {
4188                         if (scratchpad != CCISS_FIRMWARE_READY)
4189                                 return 0;
4190                 }
4191                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4192         }
4193         dev_warn(&pdev->dev, "board not ready, timed out.\n");
4194         return -ENODEV;
4195 }
4196
4197 static int cciss_find_cfg_addrs(struct pci_dev *pdev, void __iomem *vaddr,
4198                                 u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4199                                 u64 *cfg_offset)
4200 {
4201         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4202         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4203         *cfg_base_addr &= (u32) 0x0000ffff;
4204         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4205         if (*cfg_base_addr_index == -1) {
4206                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4207                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4208                 return -ENODEV;
4209         }
4210         return 0;
4211 }
4212
4213 static int cciss_find_cfgtables(ctlr_info_t *h)
4214 {
4215         u64 cfg_offset;
4216         u32 cfg_base_addr;
4217         u64 cfg_base_addr_index;
4218         u32 trans_offset;
4219         int rc;
4220
4221         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4222                 &cfg_base_addr_index, &cfg_offset);
4223         if (rc)
4224                 return rc;
4225         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4226                 cfg_base_addr_index) + cfg_offset, sizeof(*h->cfgtable));
4227         if (!h->cfgtable)
4228                 return -ENOMEM;
4229         rc = write_driver_ver_to_cfgtable(h->cfgtable);
4230         if (rc)
4231                 return rc;
4232         /* Find performant mode table. */
4233         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4234         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4235                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4236                                 sizeof(*h->transtable));
4237         if (!h->transtable)
4238                 return -ENOMEM;
4239         return 0;
4240 }
4241
4242 static void cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4243 {
4244         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4245
4246         /* Limit commands in memory limited kdump scenario. */
4247         if (reset_devices && h->max_commands > 32)
4248                 h->max_commands = 32;
4249
4250         if (h->max_commands < 16) {
4251                 dev_warn(&h->pdev->dev, "Controller reports "
4252                         "max supported commands of %d, an obvious lie. "
4253                         "Using 16.  Ensure that firmware is up to date.\n",
4254                         h->max_commands);
4255                 h->max_commands = 16;
4256         }
4257 }
4258
4259 /* Interrogate the hardware for some limits:
4260  * max commands, max SG elements without chaining, and with chaining,
4261  * SG chain block size, etc.
4262  */
4263 static void cciss_find_board_params(ctlr_info_t *h)
4264 {
4265         cciss_get_max_perf_mode_cmds(h);
4266         h->nr_cmds = h->max_commands - 4 - cciss_tape_cmds;
4267         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4268         /*
4269          * The P600 may exhibit poor performnace under some workloads
4270          * if we use the value in the configuration table. Limit this
4271          * controller to MAXSGENTRIES (32) instead.
4272          */
4273         if (h->board_id == 0x3225103C)
4274                 h->maxsgentries = MAXSGENTRIES;
4275         /*
4276          * Limit in-command s/g elements to 32 save dma'able memory.
4277          * Howvever spec says if 0, use 31
4278          */
4279         h->max_cmd_sgentries = 31;
4280         if (h->maxsgentries > 512) {
4281                 h->max_cmd_sgentries = 32;
4282                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4283                 h->maxsgentries--; /* save one for chain pointer */
4284         } else {
4285                 h->maxsgentries = 31; /* default to traditional values */
4286                 h->chainsize = 0;
4287         }
4288 }
4289
4290 static inline bool CISS_signature_present(ctlr_info_t *h)
4291 {
4292         if (!check_signature(h->cfgtable->Signature, "CISS", 4)) {
4293                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4294                 return false;
4295         }
4296         return true;
4297 }
4298
4299 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4300 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4301 {
4302 #ifdef CONFIG_X86
4303         u32 prefetch;
4304
4305         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4306         prefetch |= 0x100;
4307         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4308 #endif
4309 }
4310
4311 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4312  * in a prefetch beyond physical memory.
4313  */
4314 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4315 {
4316         u32 dma_prefetch;
4317         __u32 dma_refetch;
4318
4319         if (h->board_id != 0x3225103C)
4320                 return;
4321         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4322         dma_prefetch |= 0x8000;
4323         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4324         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4325         dma_refetch |= 0x1;
4326         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4327 }
4328
4329 static int cciss_pci_init(ctlr_info_t *h)
4330 {
4331         int prod_index, err;
4332
4333         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4334         if (prod_index < 0)
4335                 return -ENODEV;
4336         h->product_name = products[prod_index].product_name;
4337         h->access = *(products[prod_index].access);
4338
4339         if (cciss_board_disabled(h)) {
4340                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4341                 return -ENODEV;
4342         }
4343
4344         pci_disable_link_state(h->pdev, PCIE_LINK_STATE_L0S |
4345                                 PCIE_LINK_STATE_L1 | PCIE_LINK_STATE_CLKPM);
4346
4347         err = pci_enable_device(h->pdev);
4348         if (err) {
4349                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4350                 return err;
4351         }
4352
4353         err = pci_request_regions(h->pdev, "cciss");
4354         if (err) {
4355                 dev_warn(&h->pdev->dev,
4356                         "Cannot obtain PCI resources, aborting\n");
4357                 return err;
4358         }
4359
4360         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4361         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4362
4363 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4364  * else we use the IO-APIC interrupt assigned to us by system ROM.
4365  */
4366         cciss_interrupt_mode(h);
4367         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4368         if (err)
4369                 goto err_out_free_res;
4370         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4371         if (!h->vaddr) {
4372                 err = -ENOMEM;
4373                 goto err_out_free_res;
4374         }
4375         err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4376         if (err)
4377                 goto err_out_free_res;
4378         err = cciss_find_cfgtables(h);
4379         if (err)
4380                 goto err_out_free_res;
4381         print_cfg_table(h);
4382         cciss_find_board_params(h);
4383
4384         if (!CISS_signature_present(h)) {
4385                 err = -ENODEV;
4386                 goto err_out_free_res;
4387         }
4388         cciss_enable_scsi_prefetch(h);
4389         cciss_p600_dma_prefetch_quirk(h);
4390         err = cciss_enter_simple_mode(h);
4391         if (err)
4392                 goto err_out_free_res;
4393         cciss_put_controller_into_performant_mode(h);
4394         return 0;
4395
4396 err_out_free_res:
4397         /*
4398          * Deliberately omit pci_disable_device(): it does something nasty to
4399          * Smart Array controllers that pci_enable_device does not undo
4400          */
4401         if (h->transtable)
4402                 iounmap(h->transtable);
4403         if (h->cfgtable)
4404                 iounmap(h->cfgtable);
4405         if (h->vaddr)
4406                 iounmap(h->vaddr);
4407         pci_release_regions(h->pdev);
4408         return err;
4409 }
4410
4411 /* Function to find the first free pointer into our hba[] array
4412  * Returns -1 if no free entries are left.
4413  */
4414 static int alloc_cciss_hba(struct pci_dev *pdev)
4415 {
4416         int i;
4417
4418         for (i = 0; i < MAX_CTLR; i++) {
4419                 if (!hba[i]) {
4420                         ctlr_info_t *h;
4421
4422                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4423                         if (!h)
4424                                 goto Enomem;
4425                         hba[i] = h;
4426                         return i;
4427                 }
4428         }
4429         dev_warn(&pdev->dev, "This driver supports a maximum"
4430                " of %d controllers.\n", MAX_CTLR);
4431         return -1;
4432 Enomem:
4433         dev_warn(&pdev->dev, "out of memory.\n");
4434         return -1;
4435 }
4436
4437 static void free_hba(ctlr_info_t *h)
4438 {
4439         int i;
4440
4441         hba[h->ctlr] = NULL;
4442         for (i = 0; i < h->highest_lun + 1; i++)
4443                 if (h->gendisk[i] != NULL)
4444                         put_disk(h->gendisk[i]);
4445         kfree(h);
4446 }
4447
4448 /* Send a message CDB to the firmware. */
4449 static int cciss_message(struct pci_dev *pdev, unsigned char opcode,
4450                          unsigned char type)
4451 {
4452         typedef struct {
4453                 CommandListHeader_struct CommandHeader;
4454                 RequestBlock_struct Request;
4455                 ErrDescriptor_struct ErrorDescriptor;
4456         } Command;
4457         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4458         Command *cmd;
4459         dma_addr_t paddr64;
4460         uint32_t paddr32, tag;
4461         void __iomem *vaddr;
4462         int i, err;
4463
4464         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4465         if (vaddr == NULL)
4466                 return -ENOMEM;
4467
4468         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4469            CCISS commands, so they must be allocated from the lower 4GiB of
4470            memory. */
4471         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4472         if (err) {
4473                 iounmap(vaddr);
4474                 return -ENOMEM;
4475         }
4476
4477         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4478         if (cmd == NULL) {
4479                 iounmap(vaddr);
4480                 return -ENOMEM;
4481         }
4482
4483         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4484            although there's no guarantee, we assume that the address is at
4485            least 4-byte aligned (most likely, it's page-aligned). */
4486         paddr32 = paddr64;
4487
4488         cmd->CommandHeader.ReplyQueue = 0;
4489         cmd->CommandHeader.SGList = 0;
4490         cmd->CommandHeader.SGTotal = 0;
4491         cmd->CommandHeader.Tag.lower = paddr32;
4492         cmd->CommandHeader.Tag.upper = 0;
4493         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4494
4495         cmd->Request.CDBLen = 16;
4496         cmd->Request.Type.Type = TYPE_MSG;
4497         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4498         cmd->Request.Type.Direction = XFER_NONE;
4499         cmd->Request.Timeout = 0; /* Don't time out */
4500         cmd->Request.CDB[0] = opcode;
4501         cmd->Request.CDB[1] = type;
4502         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4503
4504         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4505         cmd->ErrorDescriptor.Addr.upper = 0;
4506         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4507
4508         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4509
4510         for (i = 0; i < 10; i++) {
4511                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4512                 if ((tag & ~3) == paddr32)
4513                         break;
4514                 msleep(CCISS_POST_RESET_NOOP_TIMEOUT_MSECS);
4515         }
4516
4517         iounmap(vaddr);
4518
4519         /* we leak the DMA buffer here ... no choice since the controller could
4520            still complete the command. */
4521         if (i == 10) {
4522                 dev_err(&pdev->dev,
4523                         "controller message %02x:%02x timed out\n",
4524                         opcode, type);
4525                 return -ETIMEDOUT;
4526         }
4527
4528         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4529
4530         if (tag & 2) {
4531                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4532                         opcode, type);
4533                 return -EIO;
4534         }
4535
4536         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4537                 opcode, type);
4538         return 0;
4539 }
4540
4541 #define cciss_noop(p) cciss_message(p, 3, 0)
4542
4543 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4544         void * __iomem vaddr, u32 use_doorbell)
4545 {
4546         u16 pmcsr;
4547         int pos;
4548
4549         if (use_doorbell) {
4550                 /* For everything after the P600, the PCI power state method
4551                  * of resetting the controller doesn't work, so we have this
4552                  * other way using the doorbell register.
4553                  */
4554                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4555                 writel(use_doorbell, vaddr + SA5_DOORBELL);
4556         } else { /* Try to do it the PCI power state way */
4557
4558                 /* Quoting from the Open CISS Specification: "The Power
4559                  * Management Control/Status Register (CSR) controls the power
4560                  * state of the device.  The normal operating state is D0,
4561                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4562                  * the controller, place the interface device in D3 then to D0,
4563                  * this causes a secondary PCI reset which will reset the
4564                  * controller." */
4565
4566                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4567                 if (pos == 0) {
4568                         dev_err(&pdev->dev,
4569                                 "cciss_controller_hard_reset: "
4570                                 "PCI PM not supported\n");
4571                         return -ENODEV;
4572                 }
4573                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4574                 /* enter the D3hot power management state */
4575                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4576                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4577                 pmcsr |= PCI_D3hot;
4578                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4579
4580                 msleep(500);
4581
4582                 /* enter the D0 power management state */
4583                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4584                 pmcsr |= PCI_D0;
4585                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4586
4587                 /*
4588                  * The P600 requires a small delay when changing states.
4589                  * Otherwise we may think the board did not reset and we bail.
4590                  * This for kdump only and is particular to the P600.
4591                  */
4592                 msleep(500);
4593         }
4594         return 0;
4595 }
4596
4597 static void init_driver_version(char *driver_version, int len)
4598 {
4599         memset(driver_version, 0, len);
4600         strncpy(driver_version, "cciss " DRIVER_NAME, len - 1);
4601 }
4602
4603 static int write_driver_ver_to_cfgtable(CfgTable_struct __iomem *cfgtable)
4604 {
4605         char *driver_version;
4606         int i, size = sizeof(cfgtable->driver_version);
4607
4608         driver_version = kmalloc(size, GFP_KERNEL);
4609         if (!driver_version)
4610                 return -ENOMEM;
4611
4612         init_driver_version(driver_version, size);
4613         for (i = 0; i < size; i++)
4614                 writeb(driver_version[i], &cfgtable->driver_version[i]);
4615         kfree(driver_version);
4616         return 0;
4617 }
4618
4619 static void read_driver_ver_from_cfgtable(CfgTable_struct __iomem *cfgtable,
4620                                           unsigned char *driver_ver)
4621 {
4622         int i;
4623
4624         for (i = 0; i < sizeof(cfgtable->driver_version); i++)
4625                 driver_ver[i] = readb(&cfgtable->driver_version[i]);
4626 }
4627
4628 static int controller_reset_failed(CfgTable_struct __iomem *cfgtable)
4629 {
4630
4631         char *driver_ver, *old_driver_ver;
4632         int rc, size = sizeof(cfgtable->driver_version);
4633
4634         old_driver_ver = kmalloc(2 * size, GFP_KERNEL);
4635         if (!old_driver_ver)
4636                 return -ENOMEM;
4637         driver_ver = old_driver_ver + size;
4638
4639         /* After a reset, the 32 bytes of "driver version" in the cfgtable
4640          * should have been changed, otherwise we know the reset failed.
4641          */
4642         init_driver_version(old_driver_ver, size);
4643         read_driver_ver_from_cfgtable(cfgtable, driver_ver);
4644         rc = !memcmp(driver_ver, old_driver_ver, size);
4645         kfree(old_driver_ver);
4646         return rc;
4647 }
4648
4649 /* This does a hard reset of the controller using PCI power management
4650  * states or using the doorbell register. */
4651 static int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4652 {
4653         u64 cfg_offset;
4654         u32 cfg_base_addr;
4655         u64 cfg_base_addr_index;
4656         void __iomem *vaddr;
4657         unsigned long paddr;
4658         u32 misc_fw_support;
4659         int rc;
4660         CfgTable_struct __iomem *cfgtable;
4661         u32 use_doorbell;
4662         u32 board_id;
4663         u16 command_register;
4664
4665         /* For controllers as old a the p600, this is very nearly
4666          * the same thing as
4667          *
4668          * pci_save_state(pci_dev);
4669          * pci_set_power_state(pci_dev, PCI_D3hot);
4670          * pci_set_power_state(pci_dev, PCI_D0);
4671          * pci_restore_state(pci_dev);
4672          *
4673          * For controllers newer than the P600, the pci power state
4674          * method of resetting doesn't work so we have another way
4675          * using the doorbell register.
4676          */
4677
4678         /* Exclude 640x boards.  These are two pci devices in one slot
4679          * which share a battery backed cache module.  One controls the
4680          * cache, the other accesses the cache through the one that controls
4681          * it.  If we reset the one controlling the cache, the other will
4682          * likely not be happy.  Just forbid resetting this conjoined mess.
4683          */
4684         cciss_lookup_board_id(pdev, &board_id);
4685         if (!ctlr_is_resettable(board_id)) {
4686                 dev_warn(&pdev->dev, "Controller not resettable\n");
4687                 return -ENODEV;
4688         }
4689
4690         /* if controller is soft- but not hard resettable... */
4691         if (!ctlr_is_hard_resettable(board_id))
4692                 return -ENOTSUPP; /* try soft reset later. */
4693
4694         /* Save the PCI command register */
4695         pci_read_config_word(pdev, 4, &command_register);
4696         /* Turn the board off.  This is so that later pci_restore_state()
4697          * won't turn the board on before the rest of config space is ready.
4698          */
4699         pci_disable_device(pdev);
4700         pci_save_state(pdev);
4701
4702         /* find the first memory BAR, so we can find the cfg table */
4703         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4704         if (rc)
4705                 return rc;
4706         vaddr = remap_pci_mem(paddr, 0x250);
4707         if (!vaddr)
4708                 return -ENOMEM;
4709
4710         /* find cfgtable in order to check if reset via doorbell is supported */
4711         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4712                                         &cfg_base_addr_index, &cfg_offset);
4713         if (rc)
4714                 goto unmap_vaddr;
4715         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4716                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4717         if (!cfgtable) {
4718                 rc = -ENOMEM;
4719                 goto unmap_vaddr;
4720         }
4721         rc = write_driver_ver_to_cfgtable(cfgtable);
4722         if (rc)
4723                 goto unmap_vaddr;
4724
4725         /* If reset via doorbell register is supported, use that.
4726          * There are two such methods.  Favor the newest method.
4727          */
4728         misc_fw_support = readl(&cfgtable->misc_fw_support);
4729         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET2;
4730         if (use_doorbell) {
4731                 use_doorbell = DOORBELL_CTLR_RESET2;
4732         } else {
4733                 use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4734                 if (use_doorbell) {
4735                         dev_warn(&pdev->dev, "Controller claims that "
4736                                 "'Bit 2 doorbell reset' is "
4737                                 "supported, but not 'bit 5 doorbell reset'.  "
4738                                 "Firmware update is recommended.\n");
4739                         rc = -ENOTSUPP; /* use the soft reset */
4740                         goto unmap_cfgtable;
4741                 }
4742         }
4743
4744         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4745         if (rc)
4746                 goto unmap_cfgtable;
4747         pci_restore_state(pdev);
4748         rc = pci_enable_device(pdev);
4749         if (rc) {
4750                 dev_warn(&pdev->dev, "failed to enable device.\n");
4751                 goto unmap_cfgtable;
4752         }
4753         pci_write_config_word(pdev, 4, command_register);
4754
4755         /* Some devices (notably the HP Smart Array 5i Controller)
4756            need a little pause here */
4757         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4758
4759         /* Wait for board to become not ready, then ready. */
4760         dev_info(&pdev->dev, "Waiting for board to reset.\n");
4761         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4762         if (rc) {
4763                 dev_warn(&pdev->dev, "Failed waiting for board to hard reset."
4764                                 "  Will try soft reset.\n");
4765                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4766                 goto unmap_cfgtable;
4767         }
4768         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4769         if (rc) {
4770                 dev_warn(&pdev->dev,
4771                         "failed waiting for board to become ready "
4772                         "after hard reset\n");
4773                 goto unmap_cfgtable;
4774         }
4775
4776         rc = controller_reset_failed(vaddr);
4777         if (rc < 0)
4778                 goto unmap_cfgtable;
4779         if (rc) {
4780                 dev_warn(&pdev->dev, "Unable to successfully hard reset "
4781                         "controller. Will try soft reset.\n");
4782                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4783         } else {
4784                 dev_info(&pdev->dev, "Board ready after hard reset.\n");
4785         }
4786
4787 unmap_cfgtable:
4788         iounmap(cfgtable);
4789
4790 unmap_vaddr:
4791         iounmap(vaddr);
4792         return rc;
4793 }
4794
4795 static int cciss_init_reset_devices(struct pci_dev *pdev)
4796 {
4797         int rc, i;
4798
4799         if (!reset_devices)
4800                 return 0;
4801
4802         /* Reset the controller with a PCI power-cycle or via doorbell */
4803         rc = cciss_kdump_hard_reset_controller(pdev);
4804
4805         /* -ENOTSUPP here means we cannot reset the controller
4806          * but it's already (and still) up and running in
4807          * "performant mode".  Or, it might be 640x, which can't reset
4808          * due to concerns about shared bbwc between 6402/6404 pair.
4809          */
4810         if (rc == -ENOTSUPP)
4811                 return rc; /* just try to do the kdump anyhow. */
4812         if (rc)
4813                 return -ENODEV;
4814
4815         /* Now try to get the controller to respond to a no-op */
4816         dev_warn(&pdev->dev, "Waiting for controller to respond to no-op\n");
4817         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4818                 if (cciss_noop(pdev) == 0)
4819                         break;
4820                 else
4821                         dev_warn(&pdev->dev, "no-op failed%s\n",
4822                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4823                                         "; re-trying" : ""));
4824                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4825         }
4826         return 0;
4827 }
4828
4829 static int cciss_allocate_cmd_pool(ctlr_info_t *h)
4830 {
4831         h->cmd_pool_bits = kmalloc(BITS_TO_LONGS(h->nr_cmds) *
4832                 sizeof(unsigned long), GFP_KERNEL);
4833         h->cmd_pool = pci_alloc_consistent(h->pdev,
4834                 h->nr_cmds * sizeof(CommandList_struct),
4835                 &(h->cmd_pool_dhandle));
4836         h->errinfo_pool = pci_alloc_consistent(h->pdev,
4837                 h->nr_cmds * sizeof(ErrorInfo_struct),
4838                 &(h->errinfo_pool_dhandle));
4839         if ((h->cmd_pool_bits == NULL)
4840                 || (h->cmd_pool == NULL)
4841                 || (h->errinfo_pool == NULL)) {
4842                 dev_err(&h->pdev->dev, "out of memory");
4843                 return -ENOMEM;
4844         }
4845         return 0;
4846 }
4847
4848 static int cciss_allocate_scatterlists(ctlr_info_t *h)
4849 {
4850         int i;
4851
4852         /* zero it, so that on free we need not know how many were alloc'ed */
4853         h->scatter_list = kzalloc(h->max_commands *
4854                                 sizeof(struct scatterlist *), GFP_KERNEL);
4855         if (!h->scatter_list)
4856                 return -ENOMEM;
4857
4858         for (i = 0; i < h->nr_cmds; i++) {
4859                 h->scatter_list[i] = kmalloc(sizeof(struct scatterlist) *
4860                                                 h->maxsgentries, GFP_KERNEL);
4861                 if (h->scatter_list[i] == NULL) {
4862                         dev_err(&h->pdev->dev, "could not allocate "
4863                                 "s/g lists\n");
4864                         return -ENOMEM;
4865                 }
4866         }
4867         return 0;
4868 }
4869
4870 static void cciss_free_scatterlists(ctlr_info_t *h)
4871 {
4872         int i;
4873
4874         if (h->scatter_list) {
4875                 for (i = 0; i < h->nr_cmds; i++)
4876                         kfree(h->scatter_list[i]);
4877                 kfree(h->scatter_list);
4878         }
4879 }
4880
4881 static void cciss_free_cmd_pool(ctlr_info_t *h)
4882 {
4883         kfree(h->cmd_pool_bits);
4884         if (h->cmd_pool)
4885                 pci_free_consistent(h->pdev,
4886                         h->nr_cmds * sizeof(CommandList_struct),
4887                         h->cmd_pool, h->cmd_pool_dhandle);
4888         if (h->errinfo_pool)
4889                 pci_free_consistent(h->pdev,
4890                         h->nr_cmds * sizeof(ErrorInfo_struct),
4891                         h->errinfo_pool, h->errinfo_pool_dhandle);
4892 }
4893
4894 static int cciss_request_irq(ctlr_info_t *h,
4895         irqreturn_t (*msixhandler)(int, void *),
4896         irqreturn_t (*intxhandler)(int, void *))
4897 {
4898         if (h->msix_vector || h->msi_vector) {
4899                 if (!request_irq(h->intr[h->intr_mode], msixhandler,
4900                                 0, h->devname, h))
4901                         return 0;
4902                 dev_err(&h->pdev->dev, "Unable to get msi irq %d"
4903                         " for %s\n", h->intr[h->intr_mode],
4904                         h->devname);
4905                 return -1;
4906         }
4907
4908         if (!request_irq(h->intr[h->intr_mode], intxhandler,
4909                         IRQF_SHARED, h->devname, h))
4910                 return 0;
4911         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4912                 h->intr[h->intr_mode], h->devname);
4913         return -1;
4914 }
4915
4916 static int cciss_kdump_soft_reset(ctlr_info_t *h)
4917 {
4918         if (cciss_send_reset(h, CTLR_LUNID, CCISS_RESET_TYPE_CONTROLLER)) {
4919                 dev_warn(&h->pdev->dev, "Resetting array controller failed.\n");
4920                 return -EIO;
4921         }
4922
4923         dev_info(&h->pdev->dev, "Waiting for board to soft reset.\n");
4924         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_NOT_READY)) {
4925                 dev_warn(&h->pdev->dev, "Soft reset had no effect.\n");
4926                 return -1;
4927         }
4928
4929         dev_info(&h->pdev->dev, "Board reset, awaiting READY status.\n");
4930         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY)) {
4931                 dev_warn(&h->pdev->dev, "Board failed to become ready "
4932                         "after soft reset.\n");
4933                 return -1;
4934         }
4935
4936         return 0;
4937 }
4938
4939 static void cciss_undo_allocations_after_kdump_soft_reset(ctlr_info_t *h)
4940 {
4941         int ctlr = h->ctlr;
4942
4943         free_irq(h->intr[h->intr_mode], h);
4944 #ifdef CONFIG_PCI_MSI
4945         if (h->msix_vector)
4946                 pci_disable_msix(h->pdev);
4947         else if (h->msi_vector)
4948                 pci_disable_msi(h->pdev);
4949 #endif /* CONFIG_PCI_MSI */
4950         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4951         cciss_free_scatterlists(h);
4952         cciss_free_cmd_pool(h);
4953         kfree(h->blockFetchTable);
4954         if (h->reply_pool)
4955                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
4956                                 h->reply_pool, h->reply_pool_dhandle);
4957         if (h->transtable)
4958                 iounmap(h->transtable);
4959         if (h->cfgtable)
4960                 iounmap(h->cfgtable);
4961         if (h->vaddr)
4962                 iounmap(h->vaddr);
4963         unregister_blkdev(h->major, h->devname);
4964         cciss_destroy_hba_sysfs_entry(h);
4965         pci_release_regions(h->pdev);
4966         kfree(h);
4967         hba[ctlr] = NULL;
4968 }
4969
4970 /*
4971  *  This is it.  Find all the controllers and register them.  I really hate
4972  *  stealing all these major device numbers.
4973  *  returns the number of block devices registered.
4974  */
4975 static int cciss_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4976 {
4977         int i;
4978         int j = 0;
4979         int rc;
4980         int try_soft_reset = 0;
4981         int dac, return_code;
4982         InquiryData_struct *inq_buff;
4983         ctlr_info_t *h;
4984         unsigned long flags;
4985
4986         /*
4987          * By default the cciss driver is used for all older HP Smart Array
4988          * controllers. There are module paramaters that allow a user to
4989          * override this behavior and instead use the hpsa SCSI driver. If
4990          * this is the case cciss may be loaded first from the kdump initrd
4991          * image and cause a kernel panic. So if reset_devices is true and
4992          * cciss_allow_hpsa is set just bail.
4993          */
4994         if ((reset_devices) && (cciss_allow_hpsa == 1))
4995                 return -ENODEV;
4996         rc = cciss_init_reset_devices(pdev);
4997         if (rc) {
4998                 if (rc != -ENOTSUPP)
4999                         return rc;
5000                 /* If the reset fails in a particular way (it has no way to do
5001                  * a proper hard reset, so returns -ENOTSUPP) we can try to do
5002                  * a soft reset once we get the controller configured up to the
5003                  * point that it can accept a command.
5004                  */
5005                 try_soft_reset = 1;
5006                 rc = 0;
5007         }
5008
5009 reinit_after_soft_reset:
5010
5011         i = alloc_cciss_hba(pdev);
5012         if (i < 0)
5013                 return -ENOMEM;
5014
5015         h = hba[i];
5016         h->pdev = pdev;
5017         h->busy_initializing = 1;
5018         h->intr_mode = cciss_simple_mode ? SIMPLE_MODE_INT : PERF_MODE_INT;
5019         INIT_LIST_HEAD(&h->cmpQ);
5020         INIT_LIST_HEAD(&h->reqQ);
5021         mutex_init(&h->busy_shutting_down);
5022
5023         if (cciss_pci_init(h) != 0)
5024                 goto clean_no_release_regions;
5025
5026         sprintf(h->devname, "cciss%d", i);
5027         h->ctlr = i;
5028
5029         if (cciss_tape_cmds < 2)
5030                 cciss_tape_cmds = 2;
5031         if (cciss_tape_cmds > 16)
5032                 cciss_tape_cmds = 16;
5033
5034         init_completion(&h->scan_wait);
5035
5036         if (cciss_create_hba_sysfs_entry(h))
5037                 goto clean0;
5038
5039         /* configure PCI DMA stuff */
5040         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
5041                 dac = 1;
5042         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
5043                 dac = 0;
5044         else {
5045                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
5046                 goto clean1;
5047         }
5048
5049         /*
5050          * register with the major number, or get a dynamic major number
5051          * by passing 0 as argument.  This is done for greater than
5052          * 8 controller support.
5053          */
5054         if (i < MAX_CTLR_ORIG)
5055                 h->major = COMPAQ_CISS_MAJOR + i;
5056         rc = register_blkdev(h->major, h->devname);
5057         if (rc == -EBUSY || rc == -EINVAL) {
5058                 dev_err(&h->pdev->dev,
5059                        "Unable to get major number %d for %s "
5060                        "on hba %d\n", h->major, h->devname, i);
5061                 goto clean1;
5062         } else {
5063                 if (i >= MAX_CTLR_ORIG)
5064                         h->major = rc;
5065         }
5066
5067         /* make sure the board interrupts are off */
5068         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5069         rc = cciss_request_irq(h, do_cciss_msix_intr, do_cciss_intx);
5070         if (rc)
5071                 goto clean2;
5072
5073         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
5074                h->devname, pdev->device, pci_name(pdev),
5075                h->intr[h->intr_mode], dac ? "" : " not");
5076
5077         if (cciss_allocate_cmd_pool(h))
5078                 goto clean4;
5079
5080         if (cciss_allocate_scatterlists(h))
5081                 goto clean4;
5082
5083         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
5084                 h->chainsize, h->nr_cmds);
5085         if (!h->cmd_sg_list && h->chainsize > 0)
5086                 goto clean4;
5087
5088         spin_lock_init(&h->lock);
5089
5090         /* Initialize the pdev driver private data.
5091            have it point to h.  */
5092         pci_set_drvdata(pdev, h);
5093         /* command and error info recs zeroed out before
5094            they are used */
5095         bitmap_zero(h->cmd_pool_bits, h->nr_cmds);
5096
5097         h->num_luns = 0;
5098         h->highest_lun = -1;
5099         for (j = 0; j < CISS_MAX_LUN; j++) {
5100                 h->drv[j] = NULL;
5101                 h->gendisk[j] = NULL;
5102         }
5103
5104         /* At this point, the controller is ready to take commands.
5105          * Now, if reset_devices and the hard reset didn't work, try
5106          * the soft reset and see if that works.
5107          */
5108         if (try_soft_reset) {
5109
5110                 /* This is kind of gross.  We may or may not get a completion
5111                  * from the soft reset command, and if we do, then the value
5112                  * from the fifo may or may not be valid.  So, we wait 10 secs
5113                  * after the reset throwing away any completions we get during
5114                  * that time.  Unregister the interrupt handler and register
5115                  * fake ones to scoop up any residual completions.
5116                  */
5117                 spin_lock_irqsave(&h->lock, flags);
5118                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5119                 spin_unlock_irqrestore(&h->lock, flags);
5120                 free_irq(h->intr[h->intr_mode], h);
5121                 rc = cciss_request_irq(h, cciss_msix_discard_completions,
5122                                         cciss_intx_discard_completions);
5123                 if (rc) {
5124                         dev_warn(&h->pdev->dev, "Failed to request_irq after "
5125                                 "soft reset.\n");
5126                         goto clean4;
5127                 }
5128
5129                 rc = cciss_kdump_soft_reset(h);
5130                 if (rc) {
5131                         dev_warn(&h->pdev->dev, "Soft reset failed.\n");
5132                         goto clean4;
5133                 }
5134
5135                 dev_info(&h->pdev->dev, "Board READY.\n");
5136                 dev_info(&h->pdev->dev,
5137                         "Waiting for stale completions to drain.\n");
5138                 h->access.set_intr_mask(h, CCISS_INTR_ON);
5139                 msleep(10000);
5140                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5141
5142                 rc = controller_reset_failed(h->cfgtable);
5143                 if (rc)
5144                         dev_info(&h->pdev->dev,
5145                                 "Soft reset appears to have failed.\n");
5146
5147                 /* since the controller's reset, we have to go back and re-init
5148                  * everything.  Easiest to just forget what we've done and do it
5149                  * all over again.
5150                  */
5151                 cciss_undo_allocations_after_kdump_soft_reset(h);
5152                 try_soft_reset = 0;
5153                 if (rc)
5154                         /* don't go to clean4, we already unallocated */
5155                         return -ENODEV;
5156
5157                 goto reinit_after_soft_reset;
5158         }
5159
5160         cciss_scsi_setup(h);
5161
5162         /* Turn the interrupts on so we can service requests */
5163         h->access.set_intr_mask(h, CCISS_INTR_ON);
5164
5165         /* Get the firmware version */
5166         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
5167         if (inq_buff == NULL) {
5168                 dev_err(&h->pdev->dev, "out of memory\n");
5169                 goto clean4;
5170         }
5171
5172         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
5173                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
5174         if (return_code == IO_OK) {
5175                 h->firm_ver[0] = inq_buff->data_byte[32];
5176                 h->firm_ver[1] = inq_buff->data_byte[33];
5177                 h->firm_ver[2] = inq_buff->data_byte[34];
5178                 h->firm_ver[3] = inq_buff->data_byte[35];
5179         } else {         /* send command failed */
5180                 dev_warn(&h->pdev->dev, "unable to determine firmware"
5181                         " version of controller\n");
5182         }
5183         kfree(inq_buff);
5184
5185         cciss_procinit(h);
5186
5187         h->cciss_max_sectors = 8192;
5188
5189         rebuild_lun_table(h, 1, 0);
5190         cciss_engage_scsi(h);
5191         h->busy_initializing = 0;
5192         return 0;
5193
5194 clean4:
5195         cciss_free_cmd_pool(h);
5196         cciss_free_scatterlists(h);
5197         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5198         free_irq(h->intr[h->intr_mode], h);
5199 clean2:
5200         unregister_blkdev(h->major, h->devname);
5201 clean1:
5202         cciss_destroy_hba_sysfs_entry(h);
5203 clean0:
5204         pci_release_regions(pdev);
5205 clean_no_release_regions:
5206         h->busy_initializing = 0;
5207
5208         /*
5209          * Deliberately omit pci_disable_device(): it does something nasty to
5210          * Smart Array controllers that pci_enable_device does not undo
5211          */
5212         pci_set_drvdata(pdev, NULL);
5213         free_hba(h);
5214         return -ENODEV;
5215 }
5216
5217 static void cciss_shutdown(struct pci_dev *pdev)
5218 {
5219         ctlr_info_t *h;
5220         char *flush_buf;
5221         int return_code;
5222
5223         h = pci_get_drvdata(pdev);
5224         flush_buf = kzalloc(4, GFP_KERNEL);
5225         if (!flush_buf) {
5226                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
5227                 return;
5228         }
5229         /* write all data in the battery backed cache to disk */
5230         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
5231                 4, 0, CTLR_LUNID, TYPE_CMD);
5232         kfree(flush_buf);
5233         if (return_code != IO_OK)
5234                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
5235         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5236         free_irq(h->intr[h->intr_mode], h);
5237 }
5238
5239 static int cciss_enter_simple_mode(struct ctlr_info *h)
5240 {
5241         u32 trans_support;
5242
5243         trans_support = readl(&(h->cfgtable->TransportSupport));
5244         if (!(trans_support & SIMPLE_MODE))
5245                 return -ENOTSUPP;
5246
5247         h->max_commands = readl(&(h->cfgtable->CmdsOutMax));
5248         writel(CFGTBL_Trans_Simple, &(h->cfgtable->HostWrite.TransportRequest));
5249         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
5250         cciss_wait_for_mode_change_ack(h);
5251         print_cfg_table(h);
5252         if (!(readl(&(h->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
5253                 dev_warn(&h->pdev->dev, "unable to get board into simple mode\n");
5254                 return -ENODEV;
5255         }
5256         h->transMethod = CFGTBL_Trans_Simple;
5257         return 0;
5258 }
5259
5260
5261 static void cciss_remove_one(struct pci_dev *pdev)
5262 {
5263         ctlr_info_t *h;
5264         int i, j;
5265
5266         if (pci_get_drvdata(pdev) == NULL) {
5267                 dev_err(&pdev->dev, "Unable to remove device\n");
5268                 return;
5269         }
5270
5271         h = pci_get_drvdata(pdev);
5272         i = h->ctlr;
5273         if (hba[i] == NULL) {
5274                 dev_err(&pdev->dev, "device appears to already be removed\n");
5275                 return;
5276         }
5277
5278         mutex_lock(&h->busy_shutting_down);
5279
5280         remove_from_scan_list(h);
5281         remove_proc_entry(h->devname, proc_cciss);
5282         unregister_blkdev(h->major, h->devname);
5283
5284         /* remove it from the disk list */
5285         for (j = 0; j < CISS_MAX_LUN; j++) {
5286                 struct gendisk *disk = h->gendisk[j];
5287                 if (disk) {
5288                         struct request_queue *q = disk->queue;
5289
5290                         if (disk->flags & GENHD_FL_UP) {
5291                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
5292                                 del_gendisk(disk);
5293                         }
5294                         if (q)
5295                                 blk_cleanup_queue(q);
5296                 }
5297         }
5298
5299 #ifdef CONFIG_CISS_SCSI_TAPE
5300         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
5301 #endif
5302
5303         cciss_shutdown(pdev);
5304
5305 #ifdef CONFIG_PCI_MSI
5306         if (h->msix_vector)
5307                 pci_disable_msix(h->pdev);
5308         else if (h->msi_vector)
5309                 pci_disable_msi(h->pdev);
5310 #endif                          /* CONFIG_PCI_MSI */
5311
5312         iounmap(h->transtable);
5313         iounmap(h->cfgtable);
5314         iounmap(h->vaddr);
5315
5316         cciss_free_cmd_pool(h);
5317         /* Free up sg elements */
5318         for (j = 0; j < h->nr_cmds; j++)
5319                 kfree(h->scatter_list[j]);
5320         kfree(h->scatter_list);
5321         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5322         kfree(h->blockFetchTable);
5323         if (h->reply_pool)
5324                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
5325                                 h->reply_pool, h->reply_pool_dhandle);
5326         /*
5327          * Deliberately omit pci_disable_device(): it does something nasty to
5328          * Smart Array controllers that pci_enable_device does not undo
5329          */
5330         pci_release_regions(pdev);
5331         pci_set_drvdata(pdev, NULL);
5332         cciss_destroy_hba_sysfs_entry(h);
5333         mutex_unlock(&h->busy_shutting_down);
5334         free_hba(h);
5335 }
5336
5337 static struct pci_driver cciss_pci_driver = {
5338         .name = "cciss",
5339         .probe = cciss_init_one,
5340         .remove = cciss_remove_one,
5341         .id_table = cciss_pci_device_id,        /* id_table */
5342         .shutdown = cciss_shutdown,
5343 };
5344
5345 /*
5346  *  This is it.  Register the PCI driver information for the cards we control
5347  *  the OS will call our registered routines when it finds one of our cards.
5348  */
5349 static int __init cciss_init(void)
5350 {
5351         int err;
5352
5353         /*
5354          * The hardware requires that commands are aligned on a 64-bit
5355          * boundary. Given that we use pci_alloc_consistent() to allocate an
5356          * array of them, the size must be a multiple of 8 bytes.
5357          */
5358         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
5359         printk(KERN_INFO DRIVER_NAME "\n");
5360
5361         err = bus_register(&cciss_bus_type);
5362         if (err)
5363                 return err;
5364
5365         /* Start the scan thread */
5366         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
5367         if (IS_ERR(cciss_scan_thread)) {
5368                 err = PTR_ERR(cciss_scan_thread);
5369                 goto err_bus_unregister;
5370         }
5371
5372         /* Register for our PCI devices */
5373         err = pci_register_driver(&cciss_pci_driver);
5374         if (err)
5375                 goto err_thread_stop;
5376
5377         return err;
5378
5379 err_thread_stop:
5380         kthread_stop(cciss_scan_thread);
5381 err_bus_unregister:
5382         bus_unregister(&cciss_bus_type);
5383
5384         return err;
5385 }
5386
5387 static void __exit cciss_cleanup(void)
5388 {
5389         int i;
5390
5391         pci_unregister_driver(&cciss_pci_driver);
5392         /* double check that all controller entrys have been removed */
5393         for (i = 0; i < MAX_CTLR; i++) {
5394                 if (hba[i] != NULL) {
5395                         dev_warn(&hba[i]->pdev->dev,
5396                                 "had to remove controller\n");
5397                         cciss_remove_one(hba[i]->pdev);
5398                 }
5399         }
5400         kthread_stop(cciss_scan_thread);
5401         if (proc_cciss)
5402                 remove_proc_entry("driver/cciss", NULL);
5403         bus_unregister(&cciss_bus_type);
5404 }
5405
5406 module_init(cciss_init);
5407 module_exit(cciss_cleanup);