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