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         disk->driverfs_dev = &h->drv[drv_index]->dev;
1955
1956         /* Set up queue information */
1957         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1958
1959         /* This is a hardware imposed limit. */
1960         blk_queue_max_segments(disk->queue, h->maxsgentries);
1961
1962         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1963
1964         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1965
1966         disk->queue->queuedata = h;
1967
1968         blk_queue_logical_block_size(disk->queue,
1969                                      h->drv[drv_index]->block_size);
1970
1971         /* Make sure all queue data is written out before */
1972         /* setting h->drv[drv_index]->queue, as setting this */
1973         /* allows the interrupt handler to start the queue */
1974         wmb();
1975         h->drv[drv_index]->queue = disk->queue;
1976         add_disk(disk);
1977         return 0;
1978
1979 cleanup_queue:
1980         blk_cleanup_queue(disk->queue);
1981         disk->queue = NULL;
1982 init_queue_failure:
1983         return -1;
1984 }
1985
1986 /* This function will check the usage_count of the drive to be updated/added.
1987  * If the usage_count is zero and it is a heretofore unknown drive, or,
1988  * the drive's capacity, geometry, or serial number has changed,
1989  * then the drive information will be updated and the disk will be
1990  * re-registered with the kernel.  If these conditions don't hold,
1991  * then it will be left alone for the next reboot.  The exception to this
1992  * is disk 0 which will always be left registered with the kernel since it
1993  * is also the controller node.  Any changes to disk 0 will show up on
1994  * the next reboot.
1995  */
1996 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1997         int first_time, int via_ioctl)
1998 {
1999         struct gendisk *disk;
2000         InquiryData_struct *inq_buff = NULL;
2001         unsigned int block_size;
2002         sector_t total_size;
2003         unsigned long flags = 0;
2004         int ret = 0;
2005         drive_info_struct *drvinfo;
2006
2007         /* Get information about the disk and modify the driver structure */
2008         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2009         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
2010         if (inq_buff == NULL || drvinfo == NULL)
2011                 goto mem_msg;
2012
2013         /* testing to see if 16-byte CDBs are already being used */
2014         if (h->cciss_read == CCISS_READ_16) {
2015                 cciss_read_capacity_16(h, drv_index,
2016                         &total_size, &block_size);
2017
2018         } else {
2019                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
2020                 /* if read_capacity returns all F's this volume is >2TB */
2021                 /* in size so we switch to 16-byte CDB's for all */
2022                 /* read/write ops */
2023                 if (total_size == 0xFFFFFFFFULL) {
2024                         cciss_read_capacity_16(h, drv_index,
2025                         &total_size, &block_size);
2026                         h->cciss_read = CCISS_READ_16;
2027                         h->cciss_write = CCISS_WRITE_16;
2028                 } else {
2029                         h->cciss_read = CCISS_READ_10;
2030                         h->cciss_write = CCISS_WRITE_10;
2031                 }
2032         }
2033
2034         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
2035                                inq_buff, drvinfo);
2036         drvinfo->block_size = block_size;
2037         drvinfo->nr_blocks = total_size + 1;
2038
2039         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
2040                                 drvinfo->model, drvinfo->rev);
2041         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
2042                         sizeof(drvinfo->serial_no));
2043         /* Save the lunid in case we deregister the disk, below. */
2044         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
2045                 sizeof(drvinfo->LunID));
2046
2047         /* Is it the same disk we already know, and nothing's changed? */
2048         if (h->drv[drv_index]->raid_level != -1 &&
2049                 ((memcmp(drvinfo->serial_no,
2050                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2051                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2052                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2053                 drvinfo->heads == h->drv[drv_index]->heads &&
2054                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2055                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2056                         /* The disk is unchanged, nothing to update */
2057                         goto freeret;
2058
2059         /* If we get here it's not the same disk, or something's changed,
2060          * so we need to * deregister it, and re-register it, if it's not
2061          * in use.
2062          * If the disk already exists then deregister it before proceeding
2063          * (unless it's the first disk (for the controller node).
2064          */
2065         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2066                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2067                 spin_lock_irqsave(&h->lock, flags);
2068                 h->drv[drv_index]->busy_configuring = 1;
2069                 spin_unlock_irqrestore(&h->lock, flags);
2070
2071                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2072                  * which keeps the interrupt handler from starting
2073                  * the queue.
2074                  */
2075                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2076         }
2077
2078         /* If the disk is in use return */
2079         if (ret)
2080                 goto freeret;
2081
2082         /* Save the new information from cciss_geometry_inquiry
2083          * and serial number inquiry.  If the disk was deregistered
2084          * above, then h->drv[drv_index] will be NULL.
2085          */
2086         if (h->drv[drv_index] == NULL) {
2087                 drvinfo->device_initialized = 0;
2088                 h->drv[drv_index] = drvinfo;
2089                 drvinfo = NULL; /* so it won't be freed below. */
2090         } else {
2091                 /* special case for cxd0 */
2092                 h->drv[drv_index]->block_size = drvinfo->block_size;
2093                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2094                 h->drv[drv_index]->heads = drvinfo->heads;
2095                 h->drv[drv_index]->sectors = drvinfo->sectors;
2096                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2097                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2098                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2099                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2100                         VENDOR_LEN + 1);
2101                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2102                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2103         }
2104
2105         ++h->num_luns;
2106         disk = h->gendisk[drv_index];
2107         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2108
2109         /* If it's not disk 0 (drv_index != 0)
2110          * or if it was disk 0, but there was previously
2111          * no actual corresponding configured logical drive
2112          * (raid_leve == -1) then we want to update the
2113          * logical drive's information.
2114          */
2115         if (drv_index || first_time) {
2116                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2117                         cciss_free_gendisk(h, drv_index);
2118                         cciss_free_drive_info(h, drv_index);
2119                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2120                                 drv_index);
2121                         --h->num_luns;
2122                 }
2123         }
2124
2125 freeret:
2126         kfree(inq_buff);
2127         kfree(drvinfo);
2128         return;
2129 mem_msg:
2130         dev_err(&h->pdev->dev, "out of memory\n");
2131         goto freeret;
2132 }
2133
2134 /* This function will find the first index of the controllers drive array
2135  * that has a null drv pointer and allocate the drive info struct and
2136  * will return that index   This is where new drives will be added.
2137  * If the index to be returned is greater than the highest_lun index for
2138  * the controller then highest_lun is set * to this new index.
2139  * If there are no available indexes or if tha allocation fails, then -1
2140  * is returned.  * "controller_node" is used to know if this is a real
2141  * logical drive, or just the controller node, which determines if this
2142  * counts towards highest_lun.
2143  */
2144 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2145 {
2146         int i;
2147         drive_info_struct *drv;
2148
2149         /* Search for an empty slot for our drive info */
2150         for (i = 0; i < CISS_MAX_LUN; i++) {
2151
2152                 /* if not cxd0 case, and it's occupied, skip it. */
2153                 if (h->drv[i] && i != 0)
2154                         continue;
2155                 /*
2156                  * If it's cxd0 case, and drv is alloc'ed already, and a
2157                  * disk is configured there, skip it.
2158                  */
2159                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2160                         continue;
2161
2162                 /*
2163                  * We've found an empty slot.  Update highest_lun
2164                  * provided this isn't just the fake cxd0 controller node.
2165                  */
2166                 if (i > h->highest_lun && !controller_node)
2167                         h->highest_lun = i;
2168
2169                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2170                 if (i == 0 && h->drv[i] != NULL)
2171                         return i;
2172
2173                 /*
2174                  * Found an empty slot, not already alloc'ed.  Allocate it.
2175                  * Mark it with raid_level == -1, so we know it's new later on.
2176                  */
2177                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2178                 if (!drv)
2179                         return -1;
2180                 drv->raid_level = -1; /* so we know it's new */
2181                 h->drv[i] = drv;
2182                 return i;
2183         }
2184         return -1;
2185 }
2186
2187 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2188 {
2189         kfree(h->drv[drv_index]);
2190         h->drv[drv_index] = NULL;
2191 }
2192
2193 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2194 {
2195         put_disk(h->gendisk[drv_index]);
2196         h->gendisk[drv_index] = NULL;
2197 }
2198
2199 /* cciss_add_gendisk finds a free hba[]->drv structure
2200  * and allocates a gendisk if needed, and sets the lunid
2201  * in the drvinfo structure.   It returns the index into
2202  * the ->drv[] array, or -1 if none are free.
2203  * is_controller_node indicates whether highest_lun should
2204  * count this disk, or if it's only being added to provide
2205  * a means to talk to the controller in case no logical
2206  * drives have yet been configured.
2207  */
2208 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2209         int controller_node)
2210 {
2211         int drv_index;
2212
2213         drv_index = cciss_alloc_drive_info(h, controller_node);
2214         if (drv_index == -1)
2215                 return -1;
2216
2217         /*Check if the gendisk needs to be allocated */
2218         if (!h->gendisk[drv_index]) {
2219                 h->gendisk[drv_index] =
2220                         alloc_disk(1 << NWD_SHIFT);
2221                 if (!h->gendisk[drv_index]) {
2222                         dev_err(&h->pdev->dev,
2223                                 "could not allocate a new disk %d\n",
2224                                 drv_index);
2225                         goto err_free_drive_info;
2226                 }
2227         }
2228         memcpy(h->drv[drv_index]->LunID, lunid,
2229                 sizeof(h->drv[drv_index]->LunID));
2230         if (cciss_create_ld_sysfs_entry(h, drv_index))
2231                 goto err_free_disk;
2232         /* Don't need to mark this busy because nobody */
2233         /* else knows about this disk yet to contend */
2234         /* for access to it. */
2235         h->drv[drv_index]->busy_configuring = 0;
2236         wmb();
2237         return drv_index;
2238
2239 err_free_disk:
2240         cciss_free_gendisk(h, drv_index);
2241 err_free_drive_info:
2242         cciss_free_drive_info(h, drv_index);
2243         return -1;
2244 }
2245
2246 /* This is for the special case of a controller which
2247  * has no logical drives.  In this case, we still need
2248  * to register a disk so the controller can be accessed
2249  * by the Array Config Utility.
2250  */
2251 static void cciss_add_controller_node(ctlr_info_t *h)
2252 {
2253         struct gendisk *disk;
2254         int drv_index;
2255
2256         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2257                 return;
2258
2259         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2260         if (drv_index == -1)
2261                 goto error;
2262         h->drv[drv_index]->block_size = 512;
2263         h->drv[drv_index]->nr_blocks = 0;
2264         h->drv[drv_index]->heads = 0;
2265         h->drv[drv_index]->sectors = 0;
2266         h->drv[drv_index]->cylinders = 0;
2267         h->drv[drv_index]->raid_level = -1;
2268         memset(h->drv[drv_index]->serial_no, 0, 16);
2269         disk = h->gendisk[drv_index];
2270         if (cciss_add_disk(h, disk, drv_index) == 0)
2271                 return;
2272         cciss_free_gendisk(h, drv_index);
2273         cciss_free_drive_info(h, drv_index);
2274 error:
2275         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2276         return;
2277 }
2278
2279 /* This function will add and remove logical drives from the Logical
2280  * drive array of the controller and maintain persistency of ordering
2281  * so that mount points are preserved until the next reboot.  This allows
2282  * for the removal of logical drives in the middle of the drive array
2283  * without a re-ordering of those drives.
2284  * INPUT
2285  * h            = The controller to perform the operations on
2286  */
2287 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2288         int via_ioctl)
2289 {
2290         int num_luns;
2291         ReportLunData_struct *ld_buff = NULL;
2292         int return_code;
2293         int listlength = 0;
2294         int i;
2295         int drv_found;
2296         int drv_index = 0;
2297         unsigned char lunid[8] = CTLR_LUNID;
2298         unsigned long flags;
2299
2300         if (!capable(CAP_SYS_RAWIO))
2301                 return -EPERM;
2302
2303         /* Set busy_configuring flag for this operation */
2304         spin_lock_irqsave(&h->lock, flags);
2305         if (h->busy_configuring) {
2306                 spin_unlock_irqrestore(&h->lock, flags);
2307                 return -EBUSY;
2308         }
2309         h->busy_configuring = 1;
2310         spin_unlock_irqrestore(&h->lock, flags);
2311
2312         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2313         if (ld_buff == NULL)
2314                 goto mem_msg;
2315
2316         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2317                                       sizeof(ReportLunData_struct),
2318                                       0, CTLR_LUNID, TYPE_CMD);
2319
2320         if (return_code == IO_OK)
2321                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2322         else {  /* reading number of logical volumes failed */
2323                 dev_warn(&h->pdev->dev,
2324                         "report logical volume command failed\n");
2325                 listlength = 0;
2326                 goto freeret;
2327         }
2328
2329         num_luns = listlength / 8;      /* 8 bytes per entry */
2330         if (num_luns > CISS_MAX_LUN) {
2331                 num_luns = CISS_MAX_LUN;
2332                 dev_warn(&h->pdev->dev, "more luns configured"
2333                        " on controller than can be handled by"
2334                        " this driver.\n");
2335         }
2336
2337         if (num_luns == 0)
2338                 cciss_add_controller_node(h);
2339
2340         /* Compare controller drive array to driver's drive array
2341          * to see if any drives are missing on the controller due
2342          * to action of Array Config Utility (user deletes drive)
2343          * and deregister logical drives which have disappeared.
2344          */
2345         for (i = 0; i <= h->highest_lun; i++) {
2346                 int j;
2347                 drv_found = 0;
2348
2349                 /* skip holes in the array from already deleted drives */
2350                 if (h->drv[i] == NULL)
2351                         continue;
2352
2353                 for (j = 0; j < num_luns; j++) {
2354                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2355                         if (memcmp(h->drv[i]->LunID, lunid,
2356                                 sizeof(lunid)) == 0) {
2357                                 drv_found = 1;
2358                                 break;
2359                         }
2360                 }
2361                 if (!drv_found) {
2362                         /* Deregister it from the OS, it's gone. */
2363                         spin_lock_irqsave(&h->lock, flags);
2364                         h->drv[i]->busy_configuring = 1;
2365                         spin_unlock_irqrestore(&h->lock, flags);
2366                         return_code = deregister_disk(h, i, 1, via_ioctl);
2367                         if (h->drv[i] != NULL)
2368                                 h->drv[i]->busy_configuring = 0;
2369                 }
2370         }
2371
2372         /* Compare controller drive array to driver's drive array.
2373          * Check for updates in the drive information and any new drives
2374          * on the controller due to ACU adding logical drives, or changing
2375          * a logical drive's size, etc.  Reregister any new/changed drives
2376          */
2377         for (i = 0; i < num_luns; i++) {
2378                 int j;
2379
2380                 drv_found = 0;
2381
2382                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2383                 /* Find if the LUN is already in the drive array
2384                  * of the driver.  If so then update its info
2385                  * if not in use.  If it does not exist then find
2386                  * the first free index and add it.
2387                  */
2388                 for (j = 0; j <= h->highest_lun; j++) {
2389                         if (h->drv[j] != NULL &&
2390                                 memcmp(h->drv[j]->LunID, lunid,
2391                                         sizeof(h->drv[j]->LunID)) == 0) {
2392                                 drv_index = j;
2393                                 drv_found = 1;
2394                                 break;
2395                         }
2396                 }
2397
2398                 /* check if the drive was found already in the array */
2399                 if (!drv_found) {
2400                         drv_index = cciss_add_gendisk(h, lunid, 0);
2401                         if (drv_index == -1)
2402                                 goto freeret;
2403                 }
2404                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2405         }               /* end for */
2406
2407 freeret:
2408         kfree(ld_buff);
2409         h->busy_configuring = 0;
2410         /* We return -1 here to tell the ACU that we have registered/updated
2411          * all of the drives that we can and to keep it from calling us
2412          * additional times.
2413          */
2414         return -1;
2415 mem_msg:
2416         dev_err(&h->pdev->dev, "out of memory\n");
2417         h->busy_configuring = 0;
2418         goto freeret;
2419 }
2420
2421 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2422 {
2423         /* zero out the disk size info */
2424         drive_info->nr_blocks = 0;
2425         drive_info->block_size = 0;
2426         drive_info->heads = 0;
2427         drive_info->sectors = 0;
2428         drive_info->cylinders = 0;
2429         drive_info->raid_level = -1;
2430         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2431         memset(drive_info->model, 0, sizeof(drive_info->model));
2432         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2433         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2434         /*
2435          * don't clear the LUNID though, we need to remember which
2436          * one this one is.
2437          */
2438 }
2439
2440 /* This function will deregister the disk and it's queue from the
2441  * kernel.  It must be called with the controller lock held and the
2442  * drv structures busy_configuring flag set.  It's parameters are:
2443  *
2444  * disk = This is the disk to be deregistered
2445  * drv  = This is the drive_info_struct associated with the disk to be
2446  *        deregistered.  It contains information about the disk used
2447  *        by the driver.
2448  * clear_all = This flag determines whether or not the disk information
2449  *             is going to be completely cleared out and the highest_lun
2450  *             reset.  Sometimes we want to clear out information about
2451  *             the disk in preparation for re-adding it.  In this case
2452  *             the highest_lun should be left unchanged and the LunID
2453  *             should not be cleared.
2454  * via_ioctl
2455  *    This indicates whether we've reached this path via ioctl.
2456  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2457  *    If this path is reached via ioctl(), then the max_usage_count will
2458  *    be 1, as the process calling ioctl() has got to have the device open.
2459  *    If we get here via sysfs, then the max usage count will be zero.
2460 */
2461 static int deregister_disk(ctlr_info_t *h, int drv_index,
2462                            int clear_all, int via_ioctl)
2463 {
2464         int i;
2465         struct gendisk *disk;
2466         drive_info_struct *drv;
2467         int recalculate_highest_lun;
2468
2469         if (!capable(CAP_SYS_RAWIO))
2470                 return -EPERM;
2471
2472         drv = h->drv[drv_index];
2473         disk = h->gendisk[drv_index];
2474
2475         /* make sure logical volume is NOT is use */
2476         if (clear_all || (h->gendisk[0] == disk)) {
2477                 if (drv->usage_count > via_ioctl)
2478                         return -EBUSY;
2479         } else if (drv->usage_count > 0)
2480                 return -EBUSY;
2481
2482         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2483
2484         /* invalidate the devices and deregister the disk.  If it is disk
2485          * zero do not deregister it but just zero out it's values.  This
2486          * allows us to delete disk zero but keep the controller registered.
2487          */
2488         if (h->gendisk[0] != disk) {
2489                 struct request_queue *q = disk->queue;
2490                 if (disk->flags & GENHD_FL_UP) {
2491                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2492                         del_gendisk(disk);
2493                 }
2494                 if (q)
2495                         blk_cleanup_queue(q);
2496                 /* If clear_all is set then we are deleting the logical
2497                  * drive, not just refreshing its info.  For drives
2498                  * other than disk 0 we will call put_disk.  We do not
2499                  * do this for disk 0 as we need it to be able to
2500                  * configure the controller.
2501                  */
2502                 if (clear_all){
2503                         /* This isn't pretty, but we need to find the
2504                          * disk in our array and NULL our the pointer.
2505                          * This is so that we will call alloc_disk if
2506                          * this index is used again later.
2507                          */
2508                         for (i=0; i < CISS_MAX_LUN; i++){
2509                                 if (h->gendisk[i] == disk) {
2510                                         h->gendisk[i] = NULL;
2511                                         break;
2512                                 }
2513                         }
2514                         put_disk(disk);
2515                 }
2516         } else {
2517                 set_capacity(disk, 0);
2518                 cciss_clear_drive_info(drv);
2519         }
2520
2521         --h->num_luns;
2522
2523         /* if it was the last disk, find the new hightest lun */
2524         if (clear_all && recalculate_highest_lun) {
2525                 int newhighest = -1;
2526                 for (i = 0; i <= h->highest_lun; i++) {
2527                         /* if the disk has size > 0, it is available */
2528                         if (h->drv[i] && h->drv[i]->heads)
2529                                 newhighest = i;
2530                 }
2531                 h->highest_lun = newhighest;
2532         }
2533         return 0;
2534 }
2535
2536 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2537                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2538                 int cmd_type)
2539 {
2540         u64bit buff_dma_handle;
2541         int status = IO_OK;
2542
2543         c->cmd_type = CMD_IOCTL_PEND;
2544         c->Header.ReplyQueue = 0;
2545         if (buff != NULL) {
2546                 c->Header.SGList = 1;
2547                 c->Header.SGTotal = 1;
2548         } else {
2549                 c->Header.SGList = 0;
2550                 c->Header.SGTotal = 0;
2551         }
2552         c->Header.Tag.lower = c->busaddr;
2553         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2554
2555         c->Request.Type.Type = cmd_type;
2556         if (cmd_type == TYPE_CMD) {
2557                 switch (cmd) {
2558                 case CISS_INQUIRY:
2559                         /* are we trying to read a vital product page */
2560                         if (page_code != 0) {
2561                                 c->Request.CDB[1] = 0x01;
2562                                 c->Request.CDB[2] = page_code;
2563                         }
2564                         c->Request.CDBLen = 6;
2565                         c->Request.Type.Attribute = ATTR_SIMPLE;
2566                         c->Request.Type.Direction = XFER_READ;
2567                         c->Request.Timeout = 0;
2568                         c->Request.CDB[0] = CISS_INQUIRY;
2569                         c->Request.CDB[4] = size & 0xFF;
2570                         break;
2571                 case CISS_REPORT_LOG:
2572                 case CISS_REPORT_PHYS:
2573                         /* Talking to controller so It's a physical command
2574                            mode = 00 target = 0.  Nothing to write.
2575                          */
2576                         c->Request.CDBLen = 12;
2577                         c->Request.Type.Attribute = ATTR_SIMPLE;
2578                         c->Request.Type.Direction = XFER_READ;
2579                         c->Request.Timeout = 0;
2580                         c->Request.CDB[0] = cmd;
2581                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2582                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2583                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2584                         c->Request.CDB[9] = size & 0xFF;
2585                         break;
2586
2587                 case CCISS_READ_CAPACITY:
2588                         c->Request.CDBLen = 10;
2589                         c->Request.Type.Attribute = ATTR_SIMPLE;
2590                         c->Request.Type.Direction = XFER_READ;
2591                         c->Request.Timeout = 0;
2592                         c->Request.CDB[0] = cmd;
2593                         break;
2594                 case CCISS_READ_CAPACITY_16:
2595                         c->Request.CDBLen = 16;
2596                         c->Request.Type.Attribute = ATTR_SIMPLE;
2597                         c->Request.Type.Direction = XFER_READ;
2598                         c->Request.Timeout = 0;
2599                         c->Request.CDB[0] = cmd;
2600                         c->Request.CDB[1] = 0x10;
2601                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2602                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2603                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2604                         c->Request.CDB[13] = size & 0xFF;
2605                         c->Request.Timeout = 0;
2606                         c->Request.CDB[0] = cmd;
2607                         break;
2608                 case CCISS_CACHE_FLUSH:
2609                         c->Request.CDBLen = 12;
2610                         c->Request.Type.Attribute = ATTR_SIMPLE;
2611                         c->Request.Type.Direction = XFER_WRITE;
2612                         c->Request.Timeout = 0;
2613                         c->Request.CDB[0] = BMIC_WRITE;
2614                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2615                         c->Request.CDB[7] = (size >> 8) & 0xFF;
2616                         c->Request.CDB[8] = size & 0xFF;
2617                         break;
2618                 case TEST_UNIT_READY:
2619                         c->Request.CDBLen = 6;
2620                         c->Request.Type.Attribute = ATTR_SIMPLE;
2621                         c->Request.Type.Direction = XFER_NONE;
2622                         c->Request.Timeout = 0;
2623                         break;
2624                 default:
2625                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2626                         return IO_ERROR;
2627                 }
2628         } else if (cmd_type == TYPE_MSG) {
2629                 switch (cmd) {
2630                 case CCISS_ABORT_MSG:
2631                         c->Request.CDBLen = 12;
2632                         c->Request.Type.Attribute = ATTR_SIMPLE;
2633                         c->Request.Type.Direction = XFER_WRITE;
2634                         c->Request.Timeout = 0;
2635                         c->Request.CDB[0] = cmd;        /* abort */
2636                         c->Request.CDB[1] = 0;  /* abort a command */
2637                         /* buff contains the tag of the command to abort */
2638                         memcpy(&c->Request.CDB[4], buff, 8);
2639                         break;
2640                 case CCISS_RESET_MSG:
2641                         c->Request.CDBLen = 16;
2642                         c->Request.Type.Attribute = ATTR_SIMPLE;
2643                         c->Request.Type.Direction = XFER_NONE;
2644                         c->Request.Timeout = 0;
2645                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2646                         c->Request.CDB[0] = cmd;        /* reset */
2647                         c->Request.CDB[1] = CCISS_RESET_TYPE_TARGET;
2648                         break;
2649                 case CCISS_NOOP_MSG:
2650                         c->Request.CDBLen = 1;
2651                         c->Request.Type.Attribute = ATTR_SIMPLE;
2652                         c->Request.Type.Direction = XFER_WRITE;
2653                         c->Request.Timeout = 0;
2654                         c->Request.CDB[0] = cmd;
2655                         break;
2656                 default:
2657                         dev_warn(&h->pdev->dev,
2658                                 "unknown message type %d\n", cmd);
2659                         return IO_ERROR;
2660                 }
2661         } else {
2662                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2663                 return IO_ERROR;
2664         }
2665         /* Fill in the scatter gather information */
2666         if (size > 0) {
2667                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2668                                                              buff, size,
2669                                                              PCI_DMA_BIDIRECTIONAL);
2670                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2671                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2672                 c->SG[0].Len = size;
2673                 c->SG[0].Ext = 0;       /* we are not chaining */
2674         }
2675         return status;
2676 }
2677
2678 static int cciss_send_reset(ctlr_info_t *h, unsigned char *scsi3addr,
2679                             u8 reset_type)
2680 {
2681         CommandList_struct *c;
2682         int return_status;
2683
2684         c = cmd_alloc(h);
2685         if (!c)
2686                 return -ENOMEM;
2687         return_status = fill_cmd(h, c, CCISS_RESET_MSG, NULL, 0, 0,
2688                 CTLR_LUNID, TYPE_MSG);
2689         c->Request.CDB[1] = reset_type; /* fill_cmd defaults to target reset */
2690         if (return_status != IO_OK) {
2691                 cmd_special_free(h, c);
2692                 return return_status;
2693         }
2694         c->waiting = NULL;
2695         enqueue_cmd_and_start_io(h, c);
2696         /* Don't wait for completion, the reset won't complete.  Don't free
2697          * the command either.  This is the last command we will send before
2698          * re-initializing everything, so it doesn't matter and won't leak.
2699          */
2700         return 0;
2701 }
2702
2703 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2704 {
2705         switch (c->err_info->ScsiStatus) {
2706         case SAM_STAT_GOOD:
2707                 return IO_OK;
2708         case SAM_STAT_CHECK_CONDITION:
2709                 switch (0xf & c->err_info->SenseInfo[2]) {
2710                 case 0: return IO_OK; /* no sense */
2711                 case 1: return IO_OK; /* recovered error */
2712                 default:
2713                         if (check_for_unit_attention(h, c))
2714                                 return IO_NEEDS_RETRY;
2715                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2716                                 "check condition, sense key = 0x%02x\n",
2717                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2718                 }
2719                 break;
2720         default:
2721                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2722                         "scsi status = 0x%02x\n",
2723                         c->Request.CDB[0], c->err_info->ScsiStatus);
2724                 break;
2725         }
2726         return IO_ERROR;
2727 }
2728
2729 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2730 {
2731         int return_status = IO_OK;
2732
2733         if (c->err_info->CommandStatus == CMD_SUCCESS)
2734                 return IO_OK;
2735
2736         switch (c->err_info->CommandStatus) {
2737         case CMD_TARGET_STATUS:
2738                 return_status = check_target_status(h, c);
2739                 break;
2740         case CMD_DATA_UNDERRUN:
2741         case CMD_DATA_OVERRUN:
2742                 /* expected for inquiry and report lun commands */
2743                 break;
2744         case CMD_INVALID:
2745                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2746                        "reported invalid\n", c->Request.CDB[0]);
2747                 return_status = IO_ERROR;
2748                 break;
2749         case CMD_PROTOCOL_ERR:
2750                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2751                        "protocol error\n", c->Request.CDB[0]);
2752                 return_status = IO_ERROR;
2753                 break;
2754         case CMD_HARDWARE_ERR:
2755                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2756                        " hardware error\n", c->Request.CDB[0]);
2757                 return_status = IO_ERROR;
2758                 break;
2759         case CMD_CONNECTION_LOST:
2760                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2761                        "connection lost\n", c->Request.CDB[0]);
2762                 return_status = IO_ERROR;
2763                 break;
2764         case CMD_ABORTED:
2765                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2766                        "aborted\n", c->Request.CDB[0]);
2767                 return_status = IO_ERROR;
2768                 break;
2769         case CMD_ABORT_FAILED:
2770                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2771                        "abort failed\n", c->Request.CDB[0]);
2772                 return_status = IO_ERROR;
2773                 break;
2774         case CMD_UNSOLICITED_ABORT:
2775                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2776                         c->Request.CDB[0]);
2777                 return_status = IO_NEEDS_RETRY;
2778                 break;
2779         case CMD_UNABORTABLE:
2780                 dev_warn(&h->pdev->dev, "cmd unabortable\n");
2781                 return_status = IO_ERROR;
2782                 break;
2783         default:
2784                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2785                        "unknown status %x\n", c->Request.CDB[0],
2786                        c->err_info->CommandStatus);
2787                 return_status = IO_ERROR;
2788         }
2789         return return_status;
2790 }
2791
2792 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2793         int attempt_retry)
2794 {
2795         DECLARE_COMPLETION_ONSTACK(wait);
2796         u64bit buff_dma_handle;
2797         int return_status = IO_OK;
2798
2799 resend_cmd2:
2800         c->waiting = &wait;
2801         enqueue_cmd_and_start_io(h, c);
2802
2803         wait_for_completion(&wait);
2804
2805         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2806                 goto command_done;
2807
2808         return_status = process_sendcmd_error(h, c);
2809
2810         if (return_status == IO_NEEDS_RETRY &&
2811                 c->retry_count < MAX_CMD_RETRIES) {
2812                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2813                         c->Request.CDB[0]);
2814                 c->retry_count++;
2815                 /* erase the old error information */
2816                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2817                 return_status = IO_OK;
2818                 reinit_completion(&wait);
2819                 goto resend_cmd2;
2820         }
2821
2822 command_done:
2823         /* unlock the buffers from DMA */
2824         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2825         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2826         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2827                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2828         return return_status;
2829 }
2830
2831 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2832                            __u8 page_code, unsigned char scsi3addr[],
2833                         int cmd_type)
2834 {
2835         CommandList_struct *c;
2836         int return_status;
2837
2838         c = cmd_special_alloc(h);
2839         if (!c)
2840                 return -ENOMEM;
2841         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2842                 scsi3addr, cmd_type);
2843         if (return_status == IO_OK)
2844                 return_status = sendcmd_withirq_core(h, c, 1);
2845
2846         cmd_special_free(h, c);
2847         return return_status;
2848 }
2849
2850 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2851                                    sector_t total_size,
2852                                    unsigned int block_size,
2853                                    InquiryData_struct *inq_buff,
2854                                    drive_info_struct *drv)
2855 {
2856         int return_code;
2857         unsigned long t;
2858         unsigned char scsi3addr[8];
2859
2860         memset(inq_buff, 0, sizeof(InquiryData_struct));
2861         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2862         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2863                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2864         if (return_code == IO_OK) {
2865                 if (inq_buff->data_byte[8] == 0xFF) {
2866                         dev_warn(&h->pdev->dev,
2867                                "reading geometry failed, volume "
2868                                "does not support reading geometry\n");
2869                         drv->heads = 255;
2870                         drv->sectors = 32;      /* Sectors per track */
2871                         drv->cylinders = total_size + 1;
2872                         drv->raid_level = RAID_UNKNOWN;
2873                 } else {
2874                         drv->heads = inq_buff->data_byte[6];
2875                         drv->sectors = inq_buff->data_byte[7];
2876                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2877                         drv->cylinders += inq_buff->data_byte[5];
2878                         drv->raid_level = inq_buff->data_byte[8];
2879                 }
2880                 drv->block_size = block_size;
2881                 drv->nr_blocks = total_size + 1;
2882                 t = drv->heads * drv->sectors;
2883                 if (t > 1) {
2884                         sector_t real_size = total_size + 1;
2885                         unsigned long rem = sector_div(real_size, t);
2886                         if (rem)
2887                                 real_size++;
2888                         drv->cylinders = real_size;
2889                 }
2890         } else {                /* Get geometry failed */
2891                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2892         }
2893 }
2894
2895 static void
2896 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2897                     unsigned int *block_size)
2898 {
2899         ReadCapdata_struct *buf;
2900         int return_code;
2901         unsigned char scsi3addr[8];
2902
2903         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2904         if (!buf) {
2905                 dev_warn(&h->pdev->dev, "out of memory\n");
2906                 return;
2907         }
2908
2909         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2910         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2911                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2912         if (return_code == IO_OK) {
2913                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2914                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2915         } else {                /* read capacity command failed */
2916                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2917                 *total_size = 0;
2918                 *block_size = BLOCK_SIZE;
2919         }
2920         kfree(buf);
2921 }
2922
2923 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2924         sector_t *total_size, unsigned int *block_size)
2925 {
2926         ReadCapdata_struct_16 *buf;
2927         int return_code;
2928         unsigned char scsi3addr[8];
2929
2930         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2931         if (!buf) {
2932                 dev_warn(&h->pdev->dev, "out of memory\n");
2933                 return;
2934         }
2935
2936         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2937         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2938                 buf, sizeof(ReadCapdata_struct_16),
2939                         0, scsi3addr, TYPE_CMD);
2940         if (return_code == IO_OK) {
2941                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2942                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2943         } else {                /* read capacity command failed */
2944                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2945                 *total_size = 0;
2946                 *block_size = BLOCK_SIZE;
2947         }
2948         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2949                (unsigned long long)*total_size+1, *block_size);
2950         kfree(buf);
2951 }
2952
2953 static int cciss_revalidate(struct gendisk *disk)
2954 {
2955         ctlr_info_t *h = get_host(disk);
2956         drive_info_struct *drv = get_drv(disk);
2957         int logvol;
2958         int FOUND = 0;
2959         unsigned int block_size;
2960         sector_t total_size;
2961         InquiryData_struct *inq_buff = NULL;
2962
2963         for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2964                 if (!h->drv[logvol])
2965                         continue;
2966                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2967                         sizeof(drv->LunID)) == 0) {
2968                         FOUND = 1;
2969                         break;
2970                 }
2971         }
2972
2973         if (!FOUND)
2974                 return 1;
2975
2976         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2977         if (inq_buff == NULL) {
2978                 dev_warn(&h->pdev->dev, "out of memory\n");
2979                 return 1;
2980         }
2981         if (h->cciss_read == CCISS_READ_10) {
2982                 cciss_read_capacity(h, logvol,
2983                                         &total_size, &block_size);
2984         } else {
2985                 cciss_read_capacity_16(h, logvol,
2986                                         &total_size, &block_size);
2987         }
2988         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2989                                inq_buff, drv);
2990
2991         blk_queue_logical_block_size(drv->queue, drv->block_size);
2992         set_capacity(disk, drv->nr_blocks);
2993
2994         kfree(inq_buff);
2995         return 0;
2996 }
2997
2998 /*
2999  * Map (physical) PCI mem into (virtual) kernel space
3000  */
3001 static void __iomem *remap_pci_mem(ulong base, ulong size)
3002 {
3003         ulong page_base = ((ulong) base) & PAGE_MASK;
3004         ulong page_offs = ((ulong) base) - page_base;
3005         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
3006
3007         return page_remapped ? (page_remapped + page_offs) : NULL;
3008 }
3009
3010 /*
3011  * Takes jobs of the Q and sends them to the hardware, then puts it on
3012  * the Q to wait for completion.
3013  */
3014 static void start_io(ctlr_info_t *h)
3015 {
3016         CommandList_struct *c;
3017
3018         while (!list_empty(&h->reqQ)) {
3019                 c = list_entry(h->reqQ.next, CommandList_struct, list);
3020                 /* can't do anything if fifo is full */
3021                 if ((h->access.fifo_full(h))) {
3022                         dev_warn(&h->pdev->dev, "fifo full\n");
3023                         break;
3024                 }
3025
3026                 /* Get the first entry from the Request Q */
3027                 removeQ(c);
3028                 h->Qdepth--;
3029
3030                 /* Tell the controller execute command */
3031                 h->access.submit_command(h, c);
3032
3033                 /* Put job onto the completed Q */
3034                 addQ(&h->cmpQ, c);
3035         }
3036 }
3037
3038 /* Assumes that h->lock is held. */
3039 /* Zeros out the error record and then resends the command back */
3040 /* to the controller */
3041 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
3042 {
3043         /* erase the old error information */
3044         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
3045
3046         /* add it to software queue and then send it to the controller */
3047         addQ(&h->reqQ, c);
3048         h->Qdepth++;
3049         if (h->Qdepth > h->maxQsinceinit)
3050                 h->maxQsinceinit = h->Qdepth;
3051
3052         start_io(h);
3053 }
3054
3055 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
3056         unsigned int msg_byte, unsigned int host_byte,
3057         unsigned int driver_byte)
3058 {
3059         /* inverse of macros in scsi.h */
3060         return (scsi_status_byte & 0xff) |
3061                 ((msg_byte & 0xff) << 8) |
3062                 ((host_byte & 0xff) << 16) |
3063                 ((driver_byte & 0xff) << 24);
3064 }
3065
3066 static inline int evaluate_target_status(ctlr_info_t *h,
3067                         CommandList_struct *cmd, int *retry_cmd)
3068 {
3069         unsigned char sense_key;
3070         unsigned char status_byte, msg_byte, host_byte, driver_byte;
3071         int error_value;
3072
3073         *retry_cmd = 0;
3074         /* If we get in here, it means we got "target status", that is, scsi status */
3075         status_byte = cmd->err_info->ScsiStatus;
3076         driver_byte = DRIVER_OK;
3077         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3078
3079         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3080                 host_byte = DID_PASSTHROUGH;
3081         else
3082                 host_byte = DID_OK;
3083
3084         error_value = make_status_bytes(status_byte, msg_byte,
3085                 host_byte, driver_byte);
3086
3087         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3088                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3089                         dev_warn(&h->pdev->dev, "cmd %p "
3090                                "has SCSI Status 0x%x\n",
3091                                cmd, cmd->err_info->ScsiStatus);
3092                 return error_value;
3093         }
3094
3095         /* check the sense key */
3096         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3097         /* no status or recovered error */
3098         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3099             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3100                 error_value = 0;
3101
3102         if (check_for_unit_attention(h, cmd)) {
3103                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3104                 return 0;
3105         }
3106
3107         /* Not SG_IO or similar? */
3108         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3109                 if (error_value != 0)
3110                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3111                                " sense key = 0x%x\n", cmd, sense_key);
3112                 return error_value;
3113         }
3114
3115         /* SG_IO or similar, copy sense data back */
3116         if (cmd->rq->sense) {
3117                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3118                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3119                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3120                         cmd->rq->sense_len);
3121         } else
3122                 cmd->rq->sense_len = 0;
3123
3124         return error_value;
3125 }
3126
3127 /* checks the status of the job and calls complete buffers to mark all
3128  * buffers for the completed job. Note that this function does not need
3129  * to hold the hba/queue lock.
3130  */
3131 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3132                                     int timeout)
3133 {
3134         int retry_cmd = 0;
3135         struct request *rq = cmd->rq;
3136
3137         rq->errors = 0;
3138
3139         if (timeout)
3140                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3141
3142         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3143                 goto after_error_processing;
3144
3145         switch (cmd->err_info->CommandStatus) {
3146         case CMD_TARGET_STATUS:
3147                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3148                 break;
3149         case CMD_DATA_UNDERRUN:
3150                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3151                         dev_warn(&h->pdev->dev, "cmd %p has"
3152                                " completed with data underrun "
3153                                "reported\n", cmd);
3154                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3155                 }
3156                 break;
3157         case CMD_DATA_OVERRUN:
3158                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3159                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3160                                " completed with data overrun "
3161                                "reported\n", cmd);
3162                 break;
3163         case CMD_INVALID:
3164                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3165                        "reported invalid\n", cmd);
3166                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3167                         cmd->err_info->CommandStatus, DRIVER_OK,
3168                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3169                                 DID_PASSTHROUGH : DID_ERROR);
3170                 break;
3171         case CMD_PROTOCOL_ERR:
3172                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3173                        "protocol error\n", cmd);
3174                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3175                         cmd->err_info->CommandStatus, DRIVER_OK,
3176                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3177                                 DID_PASSTHROUGH : DID_ERROR);
3178                 break;
3179         case CMD_HARDWARE_ERR:
3180                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3181                        " hardware error\n", cmd);
3182                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3183                         cmd->err_info->CommandStatus, DRIVER_OK,
3184                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3185                                 DID_PASSTHROUGH : DID_ERROR);
3186                 break;
3187         case CMD_CONNECTION_LOST:
3188                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3189                        "connection lost\n", cmd);
3190                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3191                         cmd->err_info->CommandStatus, DRIVER_OK,
3192                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3193                                 DID_PASSTHROUGH : DID_ERROR);
3194                 break;
3195         case CMD_ABORTED:
3196                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3197                        "aborted\n", cmd);
3198                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3199                         cmd->err_info->CommandStatus, DRIVER_OK,
3200                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3201                                 DID_PASSTHROUGH : DID_ABORT);
3202                 break;
3203         case CMD_ABORT_FAILED:
3204                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3205                        "abort failed\n", cmd);
3206                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3207                         cmd->err_info->CommandStatus, DRIVER_OK,
3208                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3209                                 DID_PASSTHROUGH : DID_ERROR);
3210                 break;
3211         case CMD_UNSOLICITED_ABORT:
3212                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3213                        "abort %p\n", h->ctlr, cmd);
3214                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3215                         retry_cmd = 1;
3216                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3217                         cmd->retry_count++;
3218                 } else
3219                         dev_warn(&h->pdev->dev,
3220                                 "%p retried too many times\n", cmd);
3221                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3222                         cmd->err_info->CommandStatus, DRIVER_OK,
3223                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3224                                 DID_PASSTHROUGH : DID_ABORT);
3225                 break;
3226         case CMD_TIMEOUT:
3227                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3228                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3229                         cmd->err_info->CommandStatus, DRIVER_OK,
3230                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3231                                 DID_PASSTHROUGH : DID_ERROR);
3232                 break;
3233         case CMD_UNABORTABLE:
3234                 dev_warn(&h->pdev->dev, "cmd %p unabortable\n", cmd);
3235                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3236                         cmd->err_info->CommandStatus, DRIVER_OK,
3237                         cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC ?
3238                                 DID_PASSTHROUGH : DID_ERROR);
3239                 break;
3240         default:
3241                 dev_warn(&h->pdev->dev, "cmd %p returned "
3242                        "unknown status %x\n", cmd,
3243                        cmd->err_info->CommandStatus);
3244                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3245                         cmd->err_info->CommandStatus, DRIVER_OK,
3246                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3247                                 DID_PASSTHROUGH : DID_ERROR);
3248         }
3249
3250 after_error_processing:
3251
3252         /* We need to return this command */
3253         if (retry_cmd) {
3254                 resend_cciss_cmd(h, cmd);
3255                 return;
3256         }
3257         cmd->rq->completion_data = cmd;
3258         blk_complete_request(cmd->rq);
3259 }
3260
3261 static inline u32 cciss_tag_contains_index(u32 tag)
3262 {
3263 #define DIRECT_LOOKUP_BIT 0x10
3264         return tag & DIRECT_LOOKUP_BIT;
3265 }
3266
3267 static inline u32 cciss_tag_to_index(u32 tag)
3268 {
3269 #define DIRECT_LOOKUP_SHIFT 5
3270         return tag >> DIRECT_LOOKUP_SHIFT;
3271 }
3272
3273 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag)
3274 {
3275 #define CCISS_PERF_ERROR_BITS ((1 << DIRECT_LOOKUP_SHIFT) - 1)
3276 #define CCISS_SIMPLE_ERROR_BITS 0x03
3277         if (likely(h->transMethod & CFGTBL_Trans_Performant))
3278                 return tag & ~CCISS_PERF_ERROR_BITS;
3279         return tag & ~CCISS_SIMPLE_ERROR_BITS;
3280 }
3281
3282 static inline void cciss_mark_tag_indexed(u32 *tag)
3283 {
3284         *tag |= DIRECT_LOOKUP_BIT;
3285 }
3286
3287 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3288 {
3289         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3290 }
3291
3292 /*
3293  * Get a request and submit it to the controller.
3294  */
3295 static void do_cciss_request(struct request_queue *q)
3296 {
3297         ctlr_info_t *h = q->queuedata;
3298         CommandList_struct *c;
3299         sector_t start_blk;
3300         int seg;
3301         struct request *creq;
3302         u64bit temp64;
3303         struct scatterlist *tmp_sg;
3304         SGDescriptor_struct *curr_sg;
3305         drive_info_struct *drv;
3306         int i, dir;
3307         int sg_index = 0;
3308         int chained = 0;
3309
3310       queue:
3311         creq = blk_peek_request(q);
3312         if (!creq)
3313                 goto startio;
3314
3315         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3316
3317         c = cmd_alloc(h);
3318         if (!c)
3319                 goto full;
3320
3321         blk_start_request(creq);
3322
3323         tmp_sg = h->scatter_list[c->cmdindex];
3324         spin_unlock_irq(q->queue_lock);
3325
3326         c->cmd_type = CMD_RWREQ;
3327         c->rq = creq;
3328
3329         /* fill in the request */
3330         drv = creq->rq_disk->private_data;
3331         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3332         /* got command from pool, so use the command block index instead */
3333         /* for direct lookups. */
3334         /* The first 2 bits are reserved for controller error reporting. */
3335         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3336         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3337         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3338         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3339         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3340         c->Request.Type.Attribute = ATTR_SIMPLE;
3341         c->Request.Type.Direction =
3342             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3343         c->Request.Timeout = 0; /* Don't time out */
3344         c->Request.CDB[0] =
3345             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3346         start_blk = blk_rq_pos(creq);
3347         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3348                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3349         sg_init_table(tmp_sg, h->maxsgentries);
3350         seg = blk_rq_map_sg(q, creq, tmp_sg);
3351
3352         /* get the DMA records for the setup */
3353         if (c->Request.Type.Direction == XFER_READ)
3354                 dir = PCI_DMA_FROMDEVICE;
3355         else
3356                 dir = PCI_DMA_TODEVICE;
3357
3358         curr_sg = c->SG;
3359         sg_index = 0;
3360         chained = 0;
3361
3362         for (i = 0; i < seg; i++) {
3363                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3364                         !chained && ((seg - i) > 1)) {
3365                         /* Point to next chain block. */
3366                         curr_sg = h->cmd_sg_list[c->cmdindex];
3367                         sg_index = 0;
3368                         chained = 1;
3369                 }
3370                 curr_sg[sg_index].Len = tmp_sg[i].length;
3371                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3372                                                 tmp_sg[i].offset,
3373                                                 tmp_sg[i].length, dir);
3374                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3375                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3376                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3377                 ++sg_index;
3378         }
3379         if (chained)
3380                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3381                         (seg - (h->max_cmd_sgentries - 1)) *
3382                                 sizeof(SGDescriptor_struct));
3383
3384         /* track how many SG entries we are using */
3385         if (seg > h->maxSG)
3386                 h->maxSG = seg;
3387
3388         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3389                         "chained[%d]\n",
3390                         blk_rq_sectors(creq), seg, chained);
3391
3392         c->Header.SGTotal = seg + chained;
3393         if (seg <= h->max_cmd_sgentries)
3394                 c->Header.SGList = c->Header.SGTotal;
3395         else
3396                 c->Header.SGList = h->max_cmd_sgentries;
3397         set_performant_mode(h, c);
3398
3399         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3400                 if(h->cciss_read == CCISS_READ_10) {
3401                         c->Request.CDB[1] = 0;
3402                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3403                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3404                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3405                         c->Request.CDB[5] = start_blk & 0xff;
3406                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3407                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3408                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3409                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3410                 } else {
3411                         u32 upper32 = upper_32_bits(start_blk);
3412
3413                         c->Request.CDBLen = 16;
3414                         c->Request.CDB[1]= 0;
3415                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3416                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3417                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3418                         c->Request.CDB[5]= upper32 & 0xff;
3419                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3420                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3421                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3422                         c->Request.CDB[9]= start_blk & 0xff;
3423                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3424                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3425                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3426                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3427                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3428                 }
3429         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3430                 c->Request.CDBLen = creq->cmd_len;
3431                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3432         } else {
3433                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3434                         creq->cmd_type);
3435                 BUG();
3436         }
3437
3438         spin_lock_irq(q->queue_lock);
3439
3440         addQ(&h->reqQ, c);
3441         h->Qdepth++;
3442         if (h->Qdepth > h->maxQsinceinit)
3443                 h->maxQsinceinit = h->Qdepth;
3444
3445         goto queue;
3446 full:
3447         blk_stop_queue(q);
3448 startio:
3449         /* We will already have the driver lock here so not need
3450          * to lock it.
3451          */
3452         start_io(h);
3453 }
3454
3455 static inline unsigned long get_next_completion(ctlr_info_t *h)
3456 {
3457         return h->access.command_completed(h);
3458 }
3459
3460 static inline int interrupt_pending(ctlr_info_t *h)
3461 {
3462         return h->access.intr_pending(h);
3463 }
3464
3465 static inline long interrupt_not_for_us(ctlr_info_t *h)
3466 {
3467         return ((h->access.intr_pending(h) == 0) ||
3468                 (h->interrupts_enabled == 0));
3469 }
3470
3471 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3472                         u32 raw_tag)
3473 {
3474         if (unlikely(tag_index >= h->nr_cmds)) {
3475                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3476                 return 1;
3477         }
3478         return 0;
3479 }
3480
3481 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3482                                 u32 raw_tag)
3483 {
3484         removeQ(c);
3485         if (likely(c->cmd_type == CMD_RWREQ))
3486                 complete_command(h, c, 0);
3487         else if (c->cmd_type == CMD_IOCTL_PEND)
3488                 complete(c->waiting);
3489 #ifdef CONFIG_CISS_SCSI_TAPE
3490         else if (c->cmd_type == CMD_SCSI)
3491                 complete_scsi_command(c, 0, raw_tag);
3492 #endif
3493 }
3494
3495 static inline u32 next_command(ctlr_info_t *h)
3496 {
3497         u32 a;
3498
3499         if (unlikely(!(h->transMethod & CFGTBL_Trans_Performant)))
3500                 return h->access.command_completed(h);
3501
3502         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3503                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3504                 (h->reply_pool_head)++;
3505                 h->commands_outstanding--;
3506         } else {
3507                 a = FIFO_EMPTY;
3508         }
3509         /* Check for wraparound */
3510         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3511                 h->reply_pool_head = h->reply_pool;
3512                 h->reply_pool_wraparound ^= 1;
3513         }
3514         return a;
3515 }
3516
3517 /* process completion of an indexed ("direct lookup") command */
3518 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3519 {
3520         u32 tag_index;
3521         CommandList_struct *c;
3522
3523         tag_index = cciss_tag_to_index(raw_tag);
3524         if (bad_tag(h, tag_index, raw_tag))
3525                 return next_command(h);
3526         c = h->cmd_pool + tag_index;
3527         finish_cmd(h, c, raw_tag);
3528         return next_command(h);
3529 }
3530
3531 /* process completion of a non-indexed command */
3532 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3533 {
3534         CommandList_struct *c = NULL;
3535         __u32 busaddr_masked, tag_masked;
3536
3537         tag_masked = cciss_tag_discard_error_bits(h, raw_tag);
3538         list_for_each_entry(c, &h->cmpQ, list) {
3539                 busaddr_masked = cciss_tag_discard_error_bits(h, c->busaddr);
3540                 if (busaddr_masked == tag_masked) {
3541                         finish_cmd(h, c, raw_tag);
3542                         return next_command(h);
3543                 }
3544         }
3545         bad_tag(h, h->nr_cmds + 1, raw_tag);
3546         return next_command(h);
3547 }
3548
3549 /* Some controllers, like p400, will give us one interrupt
3550  * after a soft reset, even if we turned interrupts off.
3551  * Only need to check for this in the cciss_xxx_discard_completions
3552  * functions.
3553  */
3554 static int ignore_bogus_interrupt(ctlr_info_t *h)
3555 {
3556         if (likely(!reset_devices))
3557                 return 0;
3558
3559         if (likely(h->interrupts_enabled))
3560                 return 0;
3561
3562         dev_info(&h->pdev->dev, "Received interrupt while interrupts disabled "
3563                 "(known firmware bug.)  Ignoring.\n");
3564
3565         return 1;
3566 }
3567
3568 static irqreturn_t cciss_intx_discard_completions(int irq, void *dev_id)
3569 {
3570         ctlr_info_t *h = dev_id;
3571         unsigned long flags;
3572         u32 raw_tag;
3573
3574         if (ignore_bogus_interrupt(h))
3575                 return IRQ_NONE;
3576
3577         if (interrupt_not_for_us(h))
3578                 return IRQ_NONE;
3579         spin_lock_irqsave(&h->lock, flags);
3580         while (interrupt_pending(h)) {
3581                 raw_tag = get_next_completion(h);
3582                 while (raw_tag != FIFO_EMPTY)
3583                         raw_tag = next_command(h);
3584         }
3585         spin_unlock_irqrestore(&h->lock, flags);
3586         return IRQ_HANDLED;
3587 }
3588
3589 static irqreturn_t cciss_msix_discard_completions(int irq, void *dev_id)
3590 {
3591         ctlr_info_t *h = dev_id;
3592         unsigned long flags;
3593         u32 raw_tag;
3594
3595         if (ignore_bogus_interrupt(h))
3596                 return IRQ_NONE;
3597
3598         spin_lock_irqsave(&h->lock, flags);
3599         raw_tag = get_next_completion(h);
3600         while (raw_tag != FIFO_EMPTY)
3601                 raw_tag = next_command(h);
3602         spin_unlock_irqrestore(&h->lock, flags);
3603         return IRQ_HANDLED;
3604 }
3605
3606 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3607 {
3608         ctlr_info_t *h = dev_id;
3609         unsigned long flags;
3610         u32 raw_tag;
3611
3612         if (interrupt_not_for_us(h))
3613                 return IRQ_NONE;
3614         spin_lock_irqsave(&h->lock, flags);
3615         while (interrupt_pending(h)) {
3616                 raw_tag = get_next_completion(h);
3617                 while (raw_tag != FIFO_EMPTY) {
3618                         if (cciss_tag_contains_index(raw_tag))
3619                                 raw_tag = process_indexed_cmd(h, raw_tag);
3620                         else
3621                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3622                 }
3623         }
3624         spin_unlock_irqrestore(&h->lock, flags);
3625         return IRQ_HANDLED;
3626 }
3627
3628 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3629  * check the interrupt pending register because it is not set.
3630  */
3631 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3632 {
3633         ctlr_info_t *h = dev_id;
3634         unsigned long flags;
3635         u32 raw_tag;
3636
3637         spin_lock_irqsave(&h->lock, flags);
3638         raw_tag = get_next_completion(h);
3639         while (raw_tag != FIFO_EMPTY) {
3640                 if (cciss_tag_contains_index(raw_tag))
3641                         raw_tag = process_indexed_cmd(h, raw_tag);
3642                 else
3643                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3644         }
3645         spin_unlock_irqrestore(&h->lock, flags);
3646         return IRQ_HANDLED;
3647 }
3648
3649 /**
3650  * add_to_scan_list() - add controller to rescan queue
3651  * @h:                Pointer to the controller.
3652  *
3653  * Adds the controller to the rescan queue if not already on the queue.
3654  *
3655  * returns 1 if added to the queue, 0 if skipped (could be on the
3656  * queue already, or the controller could be initializing or shutting
3657  * down).
3658  **/
3659 static int add_to_scan_list(struct ctlr_info *h)
3660 {
3661         struct ctlr_info *test_h;
3662         int found = 0;
3663         int ret = 0;
3664
3665         if (h->busy_initializing)
3666                 return 0;
3667
3668         if (!mutex_trylock(&h->busy_shutting_down))
3669                 return 0;
3670
3671         mutex_lock(&scan_mutex);
3672         list_for_each_entry(test_h, &scan_q, scan_list) {
3673                 if (test_h == h) {
3674                         found = 1;
3675                         break;
3676                 }
3677         }
3678         if (!found && !h->busy_scanning) {
3679                 reinit_completion(&h->scan_wait);
3680                 list_add_tail(&h->scan_list, &scan_q);
3681                 ret = 1;
3682         }
3683         mutex_unlock(&scan_mutex);
3684         mutex_unlock(&h->busy_shutting_down);
3685
3686         return ret;
3687 }
3688
3689 /**
3690  * remove_from_scan_list() - remove controller from rescan queue
3691  * @h:                     Pointer to the controller.
3692  *
3693  * Removes the controller from the rescan queue if present. Blocks if
3694  * the controller is currently conducting a rescan.  The controller
3695  * can be in one of three states:
3696  * 1. Doesn't need a scan
3697  * 2. On the scan list, but not scanning yet (we remove it)
3698  * 3. Busy scanning (and not on the list). In this case we want to wait for
3699  *    the scan to complete to make sure the scanning thread for this
3700  *    controller is completely idle.
3701  **/
3702 static void remove_from_scan_list(struct ctlr_info *h)
3703 {
3704         struct ctlr_info *test_h, *tmp_h;
3705
3706         mutex_lock(&scan_mutex);
3707         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3708                 if (test_h == h) { /* state 2. */
3709                         list_del(&h->scan_list);
3710                         complete_all(&h->scan_wait);
3711                         mutex_unlock(&scan_mutex);
3712                         return;
3713                 }
3714         }
3715         if (h->busy_scanning) { /* state 3. */
3716                 mutex_unlock(&scan_mutex);
3717                 wait_for_completion(&h->scan_wait);
3718         } else { /* state 1, nothing to do. */
3719                 mutex_unlock(&scan_mutex);
3720         }
3721 }
3722
3723 /**
3724  * scan_thread() - kernel thread used to rescan controllers
3725  * @data:        Ignored.
3726  *
3727  * A kernel thread used scan for drive topology changes on
3728  * controllers. The thread processes only one controller at a time
3729  * using a queue.  Controllers are added to the queue using
3730  * add_to_scan_list() and removed from the queue either after done
3731  * processing or using remove_from_scan_list().
3732  *
3733  * returns 0.
3734  **/
3735 static int scan_thread(void *data)
3736 {
3737         struct ctlr_info *h;
3738
3739         while (1) {
3740                 set_current_state(TASK_INTERRUPTIBLE);
3741                 schedule();
3742                 if (kthread_should_stop())
3743                         break;
3744
3745                 while (1) {
3746                         mutex_lock(&scan_mutex);
3747                         if (list_empty(&scan_q)) {
3748                                 mutex_unlock(&scan_mutex);
3749                                 break;
3750                         }
3751
3752                         h = list_entry(scan_q.next,
3753                                        struct ctlr_info,
3754                                        scan_list);
3755                         list_del(&h->scan_list);
3756                         h->busy_scanning = 1;
3757                         mutex_unlock(&scan_mutex);
3758
3759                         rebuild_lun_table(h, 0, 0);
3760                         complete_all(&h->scan_wait);
3761                         mutex_lock(&scan_mutex);
3762                         h->busy_scanning = 0;
3763                         mutex_unlock(&scan_mutex);
3764                 }
3765         }
3766
3767         return 0;
3768 }
3769
3770 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3771 {
3772         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3773                 return 0;
3774
3775         switch (c->err_info->SenseInfo[12]) {
3776         case STATE_CHANGED:
3777                 dev_warn(&h->pdev->dev, "a state change "
3778                         "detected, command retried\n");
3779                 return 1;
3780         break;
3781         case LUN_FAILED:
3782                 dev_warn(&h->pdev->dev, "LUN failure "
3783                         "detected, action required\n");
3784                 return 1;
3785         break;
3786         case REPORT_LUNS_CHANGED:
3787                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3788         /*
3789          * Here, we could call add_to_scan_list and wake up the scan thread,
3790          * except that it's quite likely that we will get more than one
3791          * REPORT_LUNS_CHANGED condition in quick succession, which means
3792          * that those which occur after the first one will likely happen
3793          * *during* the scan_thread's rescan.  And the rescan code is not
3794          * robust enough to restart in the middle, undoing what it has already
3795          * done, and it's not clear that it's even possible to do this, since
3796          * part of what it does is notify the block layer, which starts
3797          * doing it's own i/o to read partition tables and so on, and the
3798          * driver doesn't have visibility to know what might need undoing.
3799          * In any event, if possible, it is horribly complicated to get right
3800          * so we just don't do it for now.
3801          *
3802          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3803          */
3804                 return 1;
3805         break;
3806         case POWER_OR_RESET:
3807                 dev_warn(&h->pdev->dev,
3808                         "a power on or device reset detected\n");
3809                 return 1;
3810         break;
3811         case UNIT_ATTENTION_CLEARED:
3812                 dev_warn(&h->pdev->dev,
3813                         "unit attention cleared by another initiator\n");
3814                 return 1;
3815         break;
3816         default:
3817                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3818                 return 1;
3819         }
3820 }
3821
3822 /*
3823  *  We cannot read the structure directly, for portability we must use
3824  *   the io functions.
3825  *   This is for debug only.
3826  */
3827 static void print_cfg_table(ctlr_info_t *h)
3828 {
3829         int i;
3830         char temp_name[17];
3831         CfgTable_struct *tb = h->cfgtable;
3832
3833         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3834         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3835         for (i = 0; i < 4; i++)
3836                 temp_name[i] = readb(&(tb->Signature[i]));
3837         temp_name[4] = '\0';
3838         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3839         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3840                 readl(&(tb->SpecValence)));
3841         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3842                readl(&(tb->TransportSupport)));
3843         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3844                readl(&(tb->TransportActive)));
3845         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3846                readl(&(tb->HostWrite.TransportRequest)));
3847         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3848                readl(&(tb->HostWrite.CoalIntDelay)));
3849         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3850                readl(&(tb->HostWrite.CoalIntCount)));
3851         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3852                readl(&(tb->CmdsOutMax)));
3853         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3854                 readl(&(tb->BusTypes)));
3855         for (i = 0; i < 16; i++)
3856                 temp_name[i] = readb(&(tb->ServerName[i]));
3857         temp_name[16] = '\0';
3858         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3859         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3860                 readl(&(tb->HeartBeat)));
3861 }
3862
3863 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3864 {
3865         int i, offset, mem_type, bar_type;
3866         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3867                 return 0;
3868         offset = 0;
3869         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3870                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3871                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3872                         offset += 4;
3873                 else {
3874                         mem_type = pci_resource_flags(pdev, i) &
3875                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3876                         switch (mem_type) {
3877                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3878                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3879                                 offset += 4;    /* 32 bit */
3880                                 break;
3881                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3882                                 offset += 8;
3883                                 break;
3884                         default:        /* reserved in PCI 2.2 */
3885                                 dev_warn(&pdev->dev,
3886                                        "Base address is invalid\n");
3887                                 return -1;
3888                                 break;
3889                         }
3890                 }
3891                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3892                         return i + 1;
3893         }
3894         return -1;
3895 }
3896
3897 /* Fill in bucket_map[], given nsgs (the max number of
3898  * scatter gather elements supported) and bucket[],
3899  * which is an array of 8 integers.  The bucket[] array
3900  * contains 8 different DMA transfer sizes (in 16
3901  * byte increments) which the controller uses to fetch
3902  * commands.  This function fills in bucket_map[], which
3903  * maps a given number of scatter gather elements to one of
3904  * the 8 DMA transfer sizes.  The point of it is to allow the
3905  * controller to only do as much DMA as needed to fetch the
3906  * command, with the DMA transfer size encoded in the lower
3907  * bits of the command address.
3908  */
3909 static void  calc_bucket_map(int bucket[], int num_buckets,
3910         int nsgs, int *bucket_map)
3911 {
3912         int i, j, b, size;
3913
3914         /* even a command with 0 SGs requires 4 blocks */
3915 #define MINIMUM_TRANSFER_BLOCKS 4
3916 #define NUM_BUCKETS 8
3917         /* Note, bucket_map must have nsgs+1 entries. */
3918         for (i = 0; i <= nsgs; i++) {
3919                 /* Compute size of a command with i SG entries */
3920                 size = i + MINIMUM_TRANSFER_BLOCKS;
3921                 b = num_buckets; /* Assume the biggest bucket */
3922                 /* Find the bucket that is just big enough */
3923                 for (j = 0; j < 8; j++) {
3924                         if (bucket[j] >= size) {
3925                                 b = j;
3926                                 break;
3927                         }
3928                 }
3929                 /* for a command with i SG entries, use bucket b. */
3930                 bucket_map[i] = b;
3931         }
3932 }
3933
3934 static void cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3935 {
3936         int i;
3937
3938         /* under certain very rare conditions, this can take awhile.
3939          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3940          * as we enter this code.) */
3941         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3942                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3943                         break;
3944                 usleep_range(10000, 20000);
3945         }
3946 }
3947
3948 static void cciss_enter_performant_mode(ctlr_info_t *h, u32 use_short_tags)
3949 {
3950         /* This is a bit complicated.  There are 8 registers on
3951          * the controller which we write to to tell it 8 different
3952          * sizes of commands which there may be.  It's a way of
3953          * reducing the DMA done to fetch each command.  Encoded into
3954          * each command's tag are 3 bits which communicate to the controller
3955          * which of the eight sizes that command fits within.  The size of
3956          * each command depends on how many scatter gather entries there are.
3957          * Each SG entry requires 16 bytes.  The eight registers are programmed
3958          * with the number of 16-byte blocks a command of that size requires.
3959          * The smallest command possible requires 5 such 16 byte blocks.
3960          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3961          * blocks.  Note, this only extends to the SG entries contained
3962          * within the command block, and does not extend to chained blocks
3963          * of SG elements.   bft[] contains the eight values we write to
3964          * the registers.  They are not evenly distributed, but have more
3965          * sizes for small commands, and fewer sizes for larger commands.
3966          */
3967         __u32 trans_offset;
3968         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3969                         /*
3970                          *  5 = 1 s/g entry or 4k
3971                          *  6 = 2 s/g entry or 8k
3972                          *  8 = 4 s/g entry or 16k
3973                          * 10 = 6 s/g entry or 24k
3974                          */
3975         unsigned long register_value;
3976         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3977
3978         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3979
3980         /* Controller spec: zero out this buffer. */
3981         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3982         h->reply_pool_head = h->reply_pool;
3983
3984         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3985         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3986                                 h->blockFetchTable);
3987         writel(bft[0], &h->transtable->BlockFetch0);
3988         writel(bft[1], &h->transtable->BlockFetch1);
3989         writel(bft[2], &h->transtable->BlockFetch2);
3990         writel(bft[3], &h->transtable->BlockFetch3);
3991         writel(bft[4], &h->transtable->BlockFetch4);
3992         writel(bft[5], &h->transtable->BlockFetch5);
3993         writel(bft[6], &h->transtable->BlockFetch6);
3994         writel(bft[7], &h->transtable->BlockFetch7);
3995
3996         /* size of controller ring buffer */
3997         writel(h->max_commands, &h->transtable->RepQSize);
3998         writel(1, &h->transtable->RepQCount);
3999         writel(0, &h->transtable->RepQCtrAddrLow32);
4000         writel(0, &h->transtable->RepQCtrAddrHigh32);
4001         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
4002         writel(0, &h->transtable->RepQAddr0High32);
4003         writel(CFGTBL_Trans_Performant | use_short_tags,
4004                         &(h->cfgtable->HostWrite.TransportRequest));
4005
4006         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
4007         cciss_wait_for_mode_change_ack(h);
4008         register_value = readl(&(h->cfgtable->TransportActive));
4009         if (!(register_value & CFGTBL_Trans_Performant))
4010                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
4011                                         " performant mode\n");
4012 }
4013
4014 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h)
4015 {
4016         __u32 trans_support;
4017
4018         if (cciss_simple_mode)
4019                 return;
4020
4021         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
4022         /* Attempt to put controller into performant mode if supported */
4023         /* Does board support performant mode? */
4024         trans_support = readl(&(h->cfgtable->TransportSupport));
4025         if (!(trans_support & PERFORMANT_MODE))
4026                 return;
4027
4028         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
4029         /* Performant mode demands commands on a 32 byte boundary
4030          * pci_alloc_consistent aligns on page boundarys already.
4031          * Just need to check if divisible by 32
4032          */
4033         if ((sizeof(CommandList_struct) % 32) != 0) {
4034                 dev_warn(&h->pdev->dev, "%s %d %s\n",
4035                         "cciss info: command size[",
4036                         (int)sizeof(CommandList_struct),
4037                         "] not divisible by 32, no performant mode..\n");
4038                 return;
4039         }
4040
4041         /* Performant mode ring buffer and supporting data structures */
4042         h->reply_pool = (__u64 *)pci_alloc_consistent(
4043                 h->pdev, h->max_commands * sizeof(__u64),
4044                 &(h->reply_pool_dhandle));
4045
4046         /* Need a block fetch table for performant mode */
4047         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
4048                 sizeof(__u32)), GFP_KERNEL);
4049
4050         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
4051                 goto clean_up;
4052
4053         cciss_enter_performant_mode(h,
4054                 trans_support & CFGTBL_Trans_use_short_tags);
4055
4056         /* Change the access methods to the performant access methods */
4057         h->access = SA5_performant_access;
4058         h->transMethod = CFGTBL_Trans_Performant;
4059
4060         return;
4061 clean_up:
4062         kfree(h->blockFetchTable);
4063         if (h->reply_pool)
4064                 pci_free_consistent(h->pdev,
4065                                 h->max_commands * sizeof(__u64),
4066                                 h->reply_pool,
4067                                 h->reply_pool_dhandle);
4068         return;
4069
4070 } /* cciss_put_controller_into_performant_mode */
4071
4072 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
4073  * controllers that are capable. If not, we use IO-APIC mode.
4074  */
4075
4076 static void cciss_interrupt_mode(ctlr_info_t *h)
4077 {
4078 #ifdef CONFIG_PCI_MSI
4079         int err;
4080         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
4081         {0, 2}, {0, 3}
4082         };
4083
4084         /* Some boards advertise MSI but don't really support it */
4085         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
4086             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
4087                 goto default_int_mode;
4088
4089         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
4090                 err = pci_enable_msix_exact(h->pdev, cciss_msix_entries, 4);
4091                 if (!err) {
4092                         h->intr[0] = cciss_msix_entries[0].vector;
4093                         h->intr[1] = cciss_msix_entries[1].vector;
4094                         h->intr[2] = cciss_msix_entries[2].vector;
4095                         h->intr[3] = cciss_msix_entries[3].vector;
4096                         h->msix_vector = 1;
4097                         return;
4098                 } else {
4099                         dev_warn(&h->pdev->dev,
4100                                 "MSI-X init failed %d\n", err);
4101                 }
4102         }
4103         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
4104                 if (!pci_enable_msi(h->pdev))
4105                         h->msi_vector = 1;
4106                 else
4107                         dev_warn(&h->pdev->dev, "MSI init failed\n");
4108         }
4109 default_int_mode:
4110 #endif                          /* CONFIG_PCI_MSI */
4111         /* if we get here we're going to use the default interrupt mode */
4112         h->intr[h->intr_mode] = h->pdev->irq;
4113         return;
4114 }
4115
4116 static int cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
4117 {
4118         int i;
4119         u32 subsystem_vendor_id, subsystem_device_id;
4120
4121         subsystem_vendor_id = pdev->subsystem_vendor;
4122         subsystem_device_id = pdev->subsystem_device;
4123         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
4124                         subsystem_vendor_id;
4125
4126         for (i = 0; i < ARRAY_SIZE(products); i++) {
4127                 /* Stand aside for hpsa driver on request */
4128                 if (cciss_allow_hpsa)
4129                         return -ENODEV;
4130                 if (*board_id == products[i].board_id)
4131                         return i;
4132         }
4133         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4134                 *board_id);
4135         return -ENODEV;
4136 }
4137
4138 static inline bool cciss_board_disabled(ctlr_info_t *h)
4139 {
4140         u16 command;
4141
4142         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4143         return ((command & PCI_COMMAND_MEMORY) == 0);
4144 }
4145
4146 static int cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4147                                      unsigned long *memory_bar)
4148 {
4149         int i;
4150
4151         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4152                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4153                         /* addressing mode bits already removed */
4154                         *memory_bar = pci_resource_start(pdev, i);
4155                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4156                                 *memory_bar);
4157                         return 0;
4158                 }
4159         dev_warn(&pdev->dev, "no memory BAR found\n");
4160         return -ENODEV;
4161 }
4162
4163 static int cciss_wait_for_board_state(struct pci_dev *pdev,
4164                                       void __iomem *vaddr, int wait_for_ready)
4165 #define BOARD_READY 1
4166 #define BOARD_NOT_READY 0
4167 {
4168         int i, iterations;
4169         u32 scratchpad;
4170
4171         if (wait_for_ready)
4172                 iterations = CCISS_BOARD_READY_ITERATIONS;
4173         else
4174                 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
4175
4176         for (i = 0; i < iterations; i++) {
4177                 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
4178                 if (wait_for_ready) {
4179                         if (scratchpad == CCISS_FIRMWARE_READY)
4180                                 return 0;
4181                 } else {
4182                         if (scratchpad != CCISS_FIRMWARE_READY)
4183                                 return 0;
4184                 }
4185                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4186         }
4187         dev_warn(&pdev->dev, "board not ready, timed out.\n");
4188         return -ENODEV;
4189 }
4190
4191 static int cciss_find_cfg_addrs(struct pci_dev *pdev, void __iomem *vaddr,
4192                                 u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4193                                 u64 *cfg_offset)
4194 {
4195         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4196         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4197         *cfg_base_addr &= (u32) 0x0000ffff;
4198         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4199         if (*cfg_base_addr_index == -1) {
4200                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4201                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4202                 return -ENODEV;
4203         }
4204         return 0;
4205 }
4206
4207 static int cciss_find_cfgtables(ctlr_info_t *h)
4208 {
4209         u64 cfg_offset;
4210         u32 cfg_base_addr;
4211         u64 cfg_base_addr_index;
4212         u32 trans_offset;
4213         int rc;
4214
4215         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4216                 &cfg_base_addr_index, &cfg_offset);
4217         if (rc)
4218                 return rc;
4219         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4220                 cfg_base_addr_index) + cfg_offset, sizeof(*h->cfgtable));
4221         if (!h->cfgtable)
4222                 return -ENOMEM;
4223         rc = write_driver_ver_to_cfgtable(h->cfgtable);
4224         if (rc)
4225                 return rc;
4226         /* Find performant mode table. */
4227         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4228         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4229                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4230                                 sizeof(*h->transtable));
4231         if (!h->transtable)
4232                 return -ENOMEM;
4233         return 0;
4234 }
4235
4236 static void cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4237 {
4238         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4239
4240         /* Limit commands in memory limited kdump scenario. */
4241         if (reset_devices && h->max_commands > 32)
4242                 h->max_commands = 32;
4243
4244         if (h->max_commands < 16) {
4245                 dev_warn(&h->pdev->dev, "Controller reports "
4246                         "max supported commands of %d, an obvious lie. "
4247                         "Using 16.  Ensure that firmware is up to date.\n",
4248                         h->max_commands);
4249                 h->max_commands = 16;
4250         }
4251 }
4252
4253 /* Interrogate the hardware for some limits:
4254  * max commands, max SG elements without chaining, and with chaining,
4255  * SG chain block size, etc.
4256  */
4257 static void cciss_find_board_params(ctlr_info_t *h)
4258 {
4259         cciss_get_max_perf_mode_cmds(h);
4260         h->nr_cmds = h->max_commands - 4 - cciss_tape_cmds;
4261         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4262         /*
4263          * The P600 may exhibit poor performnace under some workloads
4264          * if we use the value in the configuration table. Limit this
4265          * controller to MAXSGENTRIES (32) instead.
4266          */
4267         if (h->board_id == 0x3225103C)
4268                 h->maxsgentries = MAXSGENTRIES;
4269         /*
4270          * Limit in-command s/g elements to 32 save dma'able memory.
4271          * Howvever spec says if 0, use 31
4272          */
4273         h->max_cmd_sgentries = 31;
4274         if (h->maxsgentries > 512) {
4275                 h->max_cmd_sgentries = 32;
4276                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4277                 h->maxsgentries--; /* save one for chain pointer */
4278         } else {
4279                 h->maxsgentries = 31; /* default to traditional values */
4280                 h->chainsize = 0;
4281         }
4282 }
4283
4284 static inline bool CISS_signature_present(ctlr_info_t *h)
4285 {
4286         if (!check_signature(h->cfgtable->Signature, "CISS", 4)) {
4287                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4288                 return false;
4289         }
4290         return true;
4291 }
4292
4293 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4294 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4295 {
4296 #ifdef CONFIG_X86
4297         u32 prefetch;
4298
4299         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4300         prefetch |= 0x100;
4301         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4302 #endif
4303 }
4304
4305 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4306  * in a prefetch beyond physical memory.
4307  */
4308 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4309 {
4310         u32 dma_prefetch;
4311         __u32 dma_refetch;
4312
4313         if (h->board_id != 0x3225103C)
4314                 return;
4315         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4316         dma_prefetch |= 0x8000;
4317         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4318         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4319         dma_refetch |= 0x1;
4320         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4321 }
4322
4323 static int cciss_pci_init(ctlr_info_t *h)
4324 {
4325         int prod_index, err;
4326
4327         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4328         if (prod_index < 0)
4329                 return -ENODEV;
4330         h->product_name = products[prod_index].product_name;
4331         h->access = *(products[prod_index].access);
4332
4333         if (cciss_board_disabled(h)) {
4334                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4335                 return -ENODEV;
4336         }
4337
4338         pci_disable_link_state(h->pdev, PCIE_LINK_STATE_L0S |
4339                                 PCIE_LINK_STATE_L1 | PCIE_LINK_STATE_CLKPM);
4340
4341         err = pci_enable_device(h->pdev);
4342         if (err) {
4343                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4344                 return err;
4345         }
4346
4347         err = pci_request_regions(h->pdev, "cciss");
4348         if (err) {
4349                 dev_warn(&h->pdev->dev,
4350                         "Cannot obtain PCI resources, aborting\n");
4351                 return err;
4352         }
4353
4354         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4355         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4356
4357 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4358  * else we use the IO-APIC interrupt assigned to us by system ROM.
4359  */
4360         cciss_interrupt_mode(h);
4361         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4362         if (err)
4363                 goto err_out_free_res;
4364         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4365         if (!h->vaddr) {
4366                 err = -ENOMEM;
4367                 goto err_out_free_res;
4368         }
4369         err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4370         if (err)
4371                 goto err_out_free_res;
4372         err = cciss_find_cfgtables(h);
4373         if (err)
4374                 goto err_out_free_res;
4375         print_cfg_table(h);
4376         cciss_find_board_params(h);
4377
4378         if (!CISS_signature_present(h)) {
4379                 err = -ENODEV;
4380                 goto err_out_free_res;
4381         }
4382         cciss_enable_scsi_prefetch(h);
4383         cciss_p600_dma_prefetch_quirk(h);
4384         err = cciss_enter_simple_mode(h);
4385         if (err)
4386                 goto err_out_free_res;
4387         cciss_put_controller_into_performant_mode(h);
4388         return 0;
4389
4390 err_out_free_res:
4391         /*
4392          * Deliberately omit pci_disable_device(): it does something nasty to
4393          * Smart Array controllers that pci_enable_device does not undo
4394          */
4395         if (h->transtable)
4396                 iounmap(h->transtable);
4397         if (h->cfgtable)
4398                 iounmap(h->cfgtable);
4399         if (h->vaddr)
4400                 iounmap(h->vaddr);
4401         pci_release_regions(h->pdev);
4402         return err;
4403 }
4404
4405 /* Function to find the first free pointer into our hba[] array
4406  * Returns -1 if no free entries are left.
4407  */
4408 static int alloc_cciss_hba(struct pci_dev *pdev)
4409 {
4410         int i;
4411
4412         for (i = 0; i < MAX_CTLR; i++) {
4413                 if (!hba[i]) {
4414                         ctlr_info_t *h;
4415
4416                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4417                         if (!h)
4418                                 goto Enomem;
4419                         hba[i] = h;
4420                         return i;
4421                 }
4422         }
4423         dev_warn(&pdev->dev, "This driver supports a maximum"
4424                " of %d controllers.\n", MAX_CTLR);
4425         return -1;
4426 Enomem:
4427         dev_warn(&pdev->dev, "out of memory.\n");
4428         return -1;
4429 }
4430
4431 static void free_hba(ctlr_info_t *h)
4432 {
4433         int i;
4434
4435         hba[h->ctlr] = NULL;
4436         for (i = 0; i < h->highest_lun + 1; i++)
4437                 if (h->gendisk[i] != NULL)
4438                         put_disk(h->gendisk[i]);
4439         kfree(h);
4440 }
4441
4442 /* Send a message CDB to the firmware. */
4443 static int cciss_message(struct pci_dev *pdev, unsigned char opcode,
4444                          unsigned char type)
4445 {
4446         typedef struct {
4447                 CommandListHeader_struct CommandHeader;
4448                 RequestBlock_struct Request;
4449                 ErrDescriptor_struct ErrorDescriptor;
4450         } Command;
4451         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4452         Command *cmd;
4453         dma_addr_t paddr64;
4454         uint32_t paddr32, tag;
4455         void __iomem *vaddr;
4456         int i, err;
4457
4458         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4459         if (vaddr == NULL)
4460                 return -ENOMEM;
4461
4462         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4463            CCISS commands, so they must be allocated from the lower 4GiB of
4464            memory. */
4465         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4466         if (err) {
4467                 iounmap(vaddr);
4468                 return -ENOMEM;
4469         }
4470
4471         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4472         if (cmd == NULL) {
4473                 iounmap(vaddr);
4474                 return -ENOMEM;
4475         }
4476
4477         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4478            although there's no guarantee, we assume that the address is at
4479            least 4-byte aligned (most likely, it's page-aligned). */
4480         paddr32 = paddr64;
4481
4482         cmd->CommandHeader.ReplyQueue = 0;
4483         cmd->CommandHeader.SGList = 0;
4484         cmd->CommandHeader.SGTotal = 0;
4485         cmd->CommandHeader.Tag.lower = paddr32;
4486         cmd->CommandHeader.Tag.upper = 0;
4487         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4488
4489         cmd->Request.CDBLen = 16;
4490         cmd->Request.Type.Type = TYPE_MSG;
4491         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4492         cmd->Request.Type.Direction = XFER_NONE;
4493         cmd->Request.Timeout = 0; /* Don't time out */
4494         cmd->Request.CDB[0] = opcode;
4495         cmd->Request.CDB[1] = type;
4496         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4497
4498         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4499         cmd->ErrorDescriptor.Addr.upper = 0;
4500         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4501
4502         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4503
4504         for (i = 0; i < 10; i++) {
4505                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4506                 if ((tag & ~3) == paddr32)
4507                         break;
4508                 msleep(CCISS_POST_RESET_NOOP_TIMEOUT_MSECS);
4509         }
4510
4511         iounmap(vaddr);
4512
4513         /* we leak the DMA buffer here ... no choice since the controller could
4514            still complete the command. */
4515         if (i == 10) {
4516                 dev_err(&pdev->dev,
4517                         "controller message %02x:%02x timed out\n",
4518                         opcode, type);
4519                 return -ETIMEDOUT;
4520         }
4521
4522         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4523
4524         if (tag & 2) {
4525                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4526                         opcode, type);
4527                 return -EIO;
4528         }
4529
4530         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4531                 opcode, type);
4532         return 0;
4533 }
4534
4535 #define cciss_noop(p) cciss_message(p, 3, 0)
4536
4537 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4538         void * __iomem vaddr, u32 use_doorbell)
4539 {
4540         u16 pmcsr;
4541         int pos;
4542
4543         if (use_doorbell) {
4544                 /* For everything after the P600, the PCI power state method
4545                  * of resetting the controller doesn't work, so we have this
4546                  * other way using the doorbell register.
4547                  */
4548                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4549                 writel(use_doorbell, vaddr + SA5_DOORBELL);
4550         } else { /* Try to do it the PCI power state way */
4551
4552                 /* Quoting from the Open CISS Specification: "The Power
4553                  * Management Control/Status Register (CSR) controls the power
4554                  * state of the device.  The normal operating state is D0,
4555                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4556                  * the controller, place the interface device in D3 then to D0,
4557                  * this causes a secondary PCI reset which will reset the
4558                  * controller." */
4559
4560                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4561                 if (pos == 0) {
4562                         dev_err(&pdev->dev,
4563                                 "cciss_controller_hard_reset: "
4564                                 "PCI PM not supported\n");
4565                         return -ENODEV;
4566                 }
4567                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4568                 /* enter the D3hot power management state */
4569                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4570                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4571                 pmcsr |= PCI_D3hot;
4572                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4573
4574                 msleep(500);
4575
4576                 /* enter the D0 power management state */
4577                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4578                 pmcsr |= PCI_D0;
4579                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4580
4581                 /*
4582                  * The P600 requires a small delay when changing states.
4583                  * Otherwise we may think the board did not reset and we bail.
4584                  * This for kdump only and is particular to the P600.
4585                  */
4586                 msleep(500);
4587         }
4588         return 0;
4589 }
4590
4591 static void init_driver_version(char *driver_version, int len)
4592 {
4593         memset(driver_version, 0, len);
4594         strncpy(driver_version, "cciss " DRIVER_NAME, len - 1);
4595 }
4596
4597 static int write_driver_ver_to_cfgtable(CfgTable_struct __iomem *cfgtable)
4598 {
4599         char *driver_version;
4600         int i, size = sizeof(cfgtable->driver_version);
4601
4602         driver_version = kmalloc(size, GFP_KERNEL);
4603         if (!driver_version)
4604                 return -ENOMEM;
4605
4606         init_driver_version(driver_version, size);
4607         for (i = 0; i < size; i++)
4608                 writeb(driver_version[i], &cfgtable->driver_version[i]);
4609         kfree(driver_version);
4610         return 0;
4611 }
4612
4613 static void read_driver_ver_from_cfgtable(CfgTable_struct __iomem *cfgtable,
4614                                           unsigned char *driver_ver)
4615 {
4616         int i;
4617
4618         for (i = 0; i < sizeof(cfgtable->driver_version); i++)
4619                 driver_ver[i] = readb(&cfgtable->driver_version[i]);
4620 }
4621
4622 static int controller_reset_failed(CfgTable_struct __iomem *cfgtable)
4623 {
4624
4625         char *driver_ver, *old_driver_ver;
4626         int rc, size = sizeof(cfgtable->driver_version);
4627
4628         old_driver_ver = kmalloc(2 * size, GFP_KERNEL);
4629         if (!old_driver_ver)
4630                 return -ENOMEM;
4631         driver_ver = old_driver_ver + size;
4632
4633         /* After a reset, the 32 bytes of "driver version" in the cfgtable
4634          * should have been changed, otherwise we know the reset failed.
4635          */
4636         init_driver_version(old_driver_ver, size);
4637         read_driver_ver_from_cfgtable(cfgtable, driver_ver);
4638         rc = !memcmp(driver_ver, old_driver_ver, size);
4639         kfree(old_driver_ver);
4640         return rc;
4641 }
4642
4643 /* This does a hard reset of the controller using PCI power management
4644  * states or using the doorbell register. */
4645 static int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4646 {
4647         u64 cfg_offset;
4648         u32 cfg_base_addr;
4649         u64 cfg_base_addr_index;
4650         void __iomem *vaddr;
4651         unsigned long paddr;
4652         u32 misc_fw_support;
4653         int rc;
4654         CfgTable_struct __iomem *cfgtable;
4655         u32 use_doorbell;
4656         u32 board_id;
4657         u16 command_register;
4658
4659         /* For controllers as old a the p600, this is very nearly
4660          * the same thing as
4661          *
4662          * pci_save_state(pci_dev);
4663          * pci_set_power_state(pci_dev, PCI_D3hot);
4664          * pci_set_power_state(pci_dev, PCI_D0);
4665          * pci_restore_state(pci_dev);
4666          *
4667          * For controllers newer than the P600, the pci power state
4668          * method of resetting doesn't work so we have another way
4669          * using the doorbell register.
4670          */
4671
4672         /* Exclude 640x boards.  These are two pci devices in one slot
4673          * which share a battery backed cache module.  One controls the
4674          * cache, the other accesses the cache through the one that controls
4675          * it.  If we reset the one controlling the cache, the other will
4676          * likely not be happy.  Just forbid resetting this conjoined mess.
4677          */
4678         cciss_lookup_board_id(pdev, &board_id);
4679         if (!ctlr_is_resettable(board_id)) {
4680                 dev_warn(&pdev->dev, "Controller not resettable\n");
4681                 return -ENODEV;
4682         }
4683
4684         /* if controller is soft- but not hard resettable... */
4685         if (!ctlr_is_hard_resettable(board_id))
4686                 return -ENOTSUPP; /* try soft reset later. */
4687
4688         /* Save the PCI command register */
4689         pci_read_config_word(pdev, 4, &command_register);
4690         /* Turn the board off.  This is so that later pci_restore_state()
4691          * won't turn the board on before the rest of config space is ready.
4692          */
4693         pci_disable_device(pdev);
4694         pci_save_state(pdev);
4695
4696         /* find the first memory BAR, so we can find the cfg table */
4697         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4698         if (rc)
4699                 return rc;
4700         vaddr = remap_pci_mem(paddr, 0x250);
4701         if (!vaddr)
4702                 return -ENOMEM;
4703
4704         /* find cfgtable in order to check if reset via doorbell is supported */
4705         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4706                                         &cfg_base_addr_index, &cfg_offset);
4707         if (rc)
4708                 goto unmap_vaddr;
4709         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4710                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4711         if (!cfgtable) {
4712                 rc = -ENOMEM;
4713                 goto unmap_vaddr;
4714         }
4715         rc = write_driver_ver_to_cfgtable(cfgtable);
4716         if (rc)
4717                 goto unmap_vaddr;
4718
4719         /* If reset via doorbell register is supported, use that.
4720          * There are two such methods.  Favor the newest method.
4721          */
4722         misc_fw_support = readl(&cfgtable->misc_fw_support);
4723         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET2;
4724         if (use_doorbell) {
4725                 use_doorbell = DOORBELL_CTLR_RESET2;
4726         } else {
4727                 use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4728                 if (use_doorbell) {
4729                         dev_warn(&pdev->dev, "Controller claims that "
4730                                 "'Bit 2 doorbell reset' is "
4731                                 "supported, but not 'bit 5 doorbell reset'.  "
4732                                 "Firmware update is recommended.\n");
4733                         rc = -ENOTSUPP; /* use the soft reset */
4734                         goto unmap_cfgtable;
4735                 }
4736         }
4737
4738         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4739         if (rc)
4740                 goto unmap_cfgtable;
4741         pci_restore_state(pdev);
4742         rc = pci_enable_device(pdev);
4743         if (rc) {
4744                 dev_warn(&pdev->dev, "failed to enable device.\n");
4745                 goto unmap_cfgtable;
4746         }
4747         pci_write_config_word(pdev, 4, command_register);
4748
4749         /* Some devices (notably the HP Smart Array 5i Controller)
4750            need a little pause here */
4751         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4752
4753         /* Wait for board to become not ready, then ready. */
4754         dev_info(&pdev->dev, "Waiting for board to reset.\n");
4755         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4756         if (rc) {
4757                 dev_warn(&pdev->dev, "Failed waiting for board to hard reset."
4758                                 "  Will try soft reset.\n");
4759                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4760                 goto unmap_cfgtable;
4761         }
4762         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4763         if (rc) {
4764                 dev_warn(&pdev->dev,
4765                         "failed waiting for board to become ready "
4766                         "after hard reset\n");
4767                 goto unmap_cfgtable;
4768         }
4769
4770         rc = controller_reset_failed(vaddr);
4771         if (rc < 0)
4772                 goto unmap_cfgtable;
4773         if (rc) {
4774                 dev_warn(&pdev->dev, "Unable to successfully hard reset "
4775                         "controller. Will try soft reset.\n");
4776                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4777         } else {
4778                 dev_info(&pdev->dev, "Board ready after hard reset.\n");
4779         }
4780
4781 unmap_cfgtable:
4782         iounmap(cfgtable);
4783
4784 unmap_vaddr:
4785         iounmap(vaddr);
4786         return rc;
4787 }
4788
4789 static int cciss_init_reset_devices(struct pci_dev *pdev)
4790 {
4791         int rc, i;
4792
4793         if (!reset_devices)
4794                 return 0;
4795
4796         /* Reset the controller with a PCI power-cycle or via doorbell */
4797         rc = cciss_kdump_hard_reset_controller(pdev);
4798
4799         /* -ENOTSUPP here means we cannot reset the controller
4800          * but it's already (and still) up and running in
4801          * "performant mode".  Or, it might be 640x, which can't reset
4802          * due to concerns about shared bbwc between 6402/6404 pair.
4803          */
4804         if (rc == -ENOTSUPP)
4805                 return rc; /* just try to do the kdump anyhow. */
4806         if (rc)
4807                 return -ENODEV;
4808
4809         /* Now try to get the controller to respond to a no-op */
4810         dev_warn(&pdev->dev, "Waiting for controller to respond to no-op\n");
4811         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4812                 if (cciss_noop(pdev) == 0)
4813                         break;
4814                 else
4815                         dev_warn(&pdev->dev, "no-op failed%s\n",
4816                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4817                                         "; re-trying" : ""));
4818                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4819         }
4820         return 0;
4821 }
4822
4823 static int cciss_allocate_cmd_pool(ctlr_info_t *h)
4824 {
4825         h->cmd_pool_bits = kmalloc(BITS_TO_LONGS(h->nr_cmds) *
4826                 sizeof(unsigned long), GFP_KERNEL);
4827         h->cmd_pool = pci_alloc_consistent(h->pdev,
4828                 h->nr_cmds * sizeof(CommandList_struct),
4829                 &(h->cmd_pool_dhandle));
4830         h->errinfo_pool = pci_alloc_consistent(h->pdev,
4831                 h->nr_cmds * sizeof(ErrorInfo_struct),
4832                 &(h->errinfo_pool_dhandle));
4833         if ((h->cmd_pool_bits == NULL)
4834                 || (h->cmd_pool == NULL)
4835                 || (h->errinfo_pool == NULL)) {
4836                 dev_err(&h->pdev->dev, "out of memory");
4837                 return -ENOMEM;
4838         }
4839         return 0;
4840 }
4841
4842 static int cciss_allocate_scatterlists(ctlr_info_t *h)
4843 {
4844         int i;
4845
4846         /* zero it, so that on free we need not know how many were alloc'ed */
4847         h->scatter_list = kzalloc(h->max_commands *
4848                                 sizeof(struct scatterlist *), GFP_KERNEL);
4849         if (!h->scatter_list)
4850                 return -ENOMEM;
4851
4852         for (i = 0; i < h->nr_cmds; i++) {
4853                 h->scatter_list[i] = kmalloc(sizeof(struct scatterlist) *
4854                                                 h->maxsgentries, GFP_KERNEL);
4855                 if (h->scatter_list[i] == NULL) {
4856                         dev_err(&h->pdev->dev, "could not allocate "
4857                                 "s/g lists\n");
4858                         return -ENOMEM;
4859                 }
4860         }
4861         return 0;
4862 }
4863
4864 static void cciss_free_scatterlists(ctlr_info_t *h)
4865 {
4866         int i;
4867
4868         if (h->scatter_list) {
4869                 for (i = 0; i < h->nr_cmds; i++)
4870                         kfree(h->scatter_list[i]);
4871                 kfree(h->scatter_list);
4872         }
4873 }
4874
4875 static void cciss_free_cmd_pool(ctlr_info_t *h)
4876 {
4877         kfree(h->cmd_pool_bits);
4878         if (h->cmd_pool)
4879                 pci_free_consistent(h->pdev,
4880                         h->nr_cmds * sizeof(CommandList_struct),
4881                         h->cmd_pool, h->cmd_pool_dhandle);
4882         if (h->errinfo_pool)
4883                 pci_free_consistent(h->pdev,
4884                         h->nr_cmds * sizeof(ErrorInfo_struct),
4885                         h->errinfo_pool, h->errinfo_pool_dhandle);
4886 }
4887
4888 static int cciss_request_irq(ctlr_info_t *h,
4889         irqreturn_t (*msixhandler)(int, void *),
4890         irqreturn_t (*intxhandler)(int, void *))
4891 {
4892         if (h->msix_vector || h->msi_vector) {
4893                 if (!request_irq(h->intr[h->intr_mode], msixhandler,
4894                                 0, h->devname, h))
4895                         return 0;
4896                 dev_err(&h->pdev->dev, "Unable to get msi irq %d"
4897                         " for %s\n", h->intr[h->intr_mode],
4898                         h->devname);
4899                 return -1;
4900         }
4901
4902         if (!request_irq(h->intr[h->intr_mode], intxhandler,
4903                         IRQF_SHARED, h->devname, h))
4904                 return 0;
4905         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4906                 h->intr[h->intr_mode], h->devname);
4907         return -1;
4908 }
4909
4910 static int cciss_kdump_soft_reset(ctlr_info_t *h)
4911 {
4912         if (cciss_send_reset(h, CTLR_LUNID, CCISS_RESET_TYPE_CONTROLLER)) {
4913                 dev_warn(&h->pdev->dev, "Resetting array controller failed.\n");
4914                 return -EIO;
4915         }
4916
4917         dev_info(&h->pdev->dev, "Waiting for board to soft reset.\n");
4918         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_NOT_READY)) {
4919                 dev_warn(&h->pdev->dev, "Soft reset had no effect.\n");
4920                 return -1;
4921         }
4922
4923         dev_info(&h->pdev->dev, "Board reset, awaiting READY status.\n");
4924         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY)) {
4925                 dev_warn(&h->pdev->dev, "Board failed to become ready "
4926                         "after soft reset.\n");
4927                 return -1;
4928         }
4929
4930         return 0;
4931 }
4932
4933 static void cciss_undo_allocations_after_kdump_soft_reset(ctlr_info_t *h)
4934 {
4935         int ctlr = h->ctlr;
4936
4937         free_irq(h->intr[h->intr_mode], h);
4938 #ifdef CONFIG_PCI_MSI
4939         if (h->msix_vector)
4940                 pci_disable_msix(h->pdev);
4941         else if (h->msi_vector)
4942                 pci_disable_msi(h->pdev);
4943 #endif /* CONFIG_PCI_MSI */
4944         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4945         cciss_free_scatterlists(h);
4946         cciss_free_cmd_pool(h);
4947         kfree(h->blockFetchTable);
4948         if (h->reply_pool)
4949                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
4950                                 h->reply_pool, h->reply_pool_dhandle);
4951         if (h->transtable)
4952                 iounmap(h->transtable);
4953         if (h->cfgtable)
4954                 iounmap(h->cfgtable);
4955         if (h->vaddr)
4956                 iounmap(h->vaddr);
4957         unregister_blkdev(h->major, h->devname);
4958         cciss_destroy_hba_sysfs_entry(h);
4959         pci_release_regions(h->pdev);
4960         kfree(h);
4961         hba[ctlr] = NULL;
4962 }
4963
4964 /*
4965  *  This is it.  Find all the controllers and register them.  I really hate
4966  *  stealing all these major device numbers.
4967  *  returns the number of block devices registered.
4968  */
4969 static int cciss_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4970 {
4971         int i;
4972         int j = 0;
4973         int rc;
4974         int try_soft_reset = 0;
4975         int dac, return_code;
4976         InquiryData_struct *inq_buff;
4977         ctlr_info_t *h;
4978         unsigned long flags;
4979
4980         /*
4981          * By default the cciss driver is used for all older HP Smart Array
4982          * controllers. There are module paramaters that allow a user to
4983          * override this behavior and instead use the hpsa SCSI driver. If
4984          * this is the case cciss may be loaded first from the kdump initrd
4985          * image and cause a kernel panic. So if reset_devices is true and
4986          * cciss_allow_hpsa is set just bail.
4987          */
4988         if ((reset_devices) && (cciss_allow_hpsa == 1))
4989                 return -ENODEV;
4990         rc = cciss_init_reset_devices(pdev);
4991         if (rc) {
4992                 if (rc != -ENOTSUPP)
4993                         return rc;
4994                 /* If the reset fails in a particular way (it has no way to do
4995                  * a proper hard reset, so returns -ENOTSUPP) we can try to do
4996                  * a soft reset once we get the controller configured up to the
4997                  * point that it can accept a command.
4998                  */
4999                 try_soft_reset = 1;
5000                 rc = 0;
5001         }
5002
5003 reinit_after_soft_reset:
5004
5005         i = alloc_cciss_hba(pdev);
5006         if (i < 0)
5007                 return -ENOMEM;
5008
5009         h = hba[i];
5010         h->pdev = pdev;
5011         h->busy_initializing = 1;
5012         h->intr_mode = cciss_simple_mode ? SIMPLE_MODE_INT : PERF_MODE_INT;
5013         INIT_LIST_HEAD(&h->cmpQ);
5014         INIT_LIST_HEAD(&h->reqQ);
5015         mutex_init(&h->busy_shutting_down);
5016
5017         if (cciss_pci_init(h) != 0)
5018                 goto clean_no_release_regions;
5019
5020         sprintf(h->devname, "cciss%d", i);
5021         h->ctlr = i;
5022
5023         if (cciss_tape_cmds < 2)
5024                 cciss_tape_cmds = 2;
5025         if (cciss_tape_cmds > 16)
5026                 cciss_tape_cmds = 16;
5027
5028         init_completion(&h->scan_wait);
5029
5030         if (cciss_create_hba_sysfs_entry(h))
5031                 goto clean0;
5032
5033         /* configure PCI DMA stuff */
5034         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
5035                 dac = 1;
5036         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
5037                 dac = 0;
5038         else {
5039                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
5040                 goto clean1;
5041         }
5042
5043         /*
5044          * register with the major number, or get a dynamic major number
5045          * by passing 0 as argument.  This is done for greater than
5046          * 8 controller support.
5047          */
5048         if (i < MAX_CTLR_ORIG)
5049                 h->major = COMPAQ_CISS_MAJOR + i;
5050         rc = register_blkdev(h->major, h->devname);
5051         if (rc == -EBUSY || rc == -EINVAL) {
5052                 dev_err(&h->pdev->dev,
5053                        "Unable to get major number %d for %s "
5054                        "on hba %d\n", h->major, h->devname, i);
5055                 goto clean1;
5056         } else {
5057                 if (i >= MAX_CTLR_ORIG)
5058                         h->major = rc;
5059         }
5060
5061         /* make sure the board interrupts are off */
5062         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5063         rc = cciss_request_irq(h, do_cciss_msix_intr, do_cciss_intx);
5064         if (rc)
5065                 goto clean2;
5066
5067         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
5068                h->devname, pdev->device, pci_name(pdev),
5069                h->intr[h->intr_mode], dac ? "" : " not");
5070
5071         if (cciss_allocate_cmd_pool(h))
5072                 goto clean4;
5073
5074         if (cciss_allocate_scatterlists(h))
5075                 goto clean4;
5076
5077         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
5078                 h->chainsize, h->nr_cmds);
5079         if (!h->cmd_sg_list && h->chainsize > 0)
5080                 goto clean4;
5081
5082         spin_lock_init(&h->lock);
5083
5084         /* Initialize the pdev driver private data.
5085            have it point to h.  */
5086         pci_set_drvdata(pdev, h);
5087         /* command and error info recs zeroed out before
5088            they are used */
5089         bitmap_zero(h->cmd_pool_bits, h->nr_cmds);
5090
5091         h->num_luns = 0;
5092         h->highest_lun = -1;
5093         for (j = 0; j < CISS_MAX_LUN; j++) {
5094                 h->drv[j] = NULL;
5095                 h->gendisk[j] = NULL;
5096         }
5097
5098         /* At this point, the controller is ready to take commands.
5099          * Now, if reset_devices and the hard reset didn't work, try
5100          * the soft reset and see if that works.
5101          */
5102         if (try_soft_reset) {
5103
5104                 /* This is kind of gross.  We may or may not get a completion
5105                  * from the soft reset command, and if we do, then the value
5106                  * from the fifo may or may not be valid.  So, we wait 10 secs
5107                  * after the reset throwing away any completions we get during
5108                  * that time.  Unregister the interrupt handler and register
5109                  * fake ones to scoop up any residual completions.
5110                  */
5111                 spin_lock_irqsave(&h->lock, flags);
5112                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5113                 spin_unlock_irqrestore(&h->lock, flags);
5114                 free_irq(h->intr[h->intr_mode], h);
5115                 rc = cciss_request_irq(h, cciss_msix_discard_completions,
5116                                         cciss_intx_discard_completions);
5117                 if (rc) {
5118                         dev_warn(&h->pdev->dev, "Failed to request_irq after "
5119                                 "soft reset.\n");
5120                         goto clean4;
5121                 }
5122
5123                 rc = cciss_kdump_soft_reset(h);
5124                 if (rc) {
5125                         dev_warn(&h->pdev->dev, "Soft reset failed.\n");
5126                         goto clean4;
5127                 }
5128
5129                 dev_info(&h->pdev->dev, "Board READY.\n");
5130                 dev_info(&h->pdev->dev,
5131                         "Waiting for stale completions to drain.\n");
5132                 h->access.set_intr_mask(h, CCISS_INTR_ON);
5133                 msleep(10000);
5134                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5135
5136                 rc = controller_reset_failed(h->cfgtable);
5137                 if (rc)
5138                         dev_info(&h->pdev->dev,
5139                                 "Soft reset appears to have failed.\n");
5140
5141                 /* since the controller's reset, we have to go back and re-init
5142                  * everything.  Easiest to just forget what we've done and do it
5143                  * all over again.
5144                  */
5145                 cciss_undo_allocations_after_kdump_soft_reset(h);
5146                 try_soft_reset = 0;
5147                 if (rc)
5148                         /* don't go to clean4, we already unallocated */
5149                         return -ENODEV;
5150
5151                 goto reinit_after_soft_reset;
5152         }
5153
5154         cciss_scsi_setup(h);
5155
5156         /* Turn the interrupts on so we can service requests */
5157         h->access.set_intr_mask(h, CCISS_INTR_ON);
5158
5159         /* Get the firmware version */
5160         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
5161         if (inq_buff == NULL) {
5162                 dev_err(&h->pdev->dev, "out of memory\n");
5163                 goto clean4;
5164         }
5165
5166         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
5167                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
5168         if (return_code == IO_OK) {
5169                 h->firm_ver[0] = inq_buff->data_byte[32];
5170                 h->firm_ver[1] = inq_buff->data_byte[33];
5171                 h->firm_ver[2] = inq_buff->data_byte[34];
5172                 h->firm_ver[3] = inq_buff->data_byte[35];
5173         } else {         /* send command failed */
5174                 dev_warn(&h->pdev->dev, "unable to determine firmware"
5175                         " version of controller\n");
5176         }
5177         kfree(inq_buff);
5178
5179         cciss_procinit(h);
5180
5181         h->cciss_max_sectors = 8192;
5182
5183         rebuild_lun_table(h, 1, 0);
5184         cciss_engage_scsi(h);
5185         h->busy_initializing = 0;
5186         return 0;
5187
5188 clean4:
5189         cciss_free_cmd_pool(h);
5190         cciss_free_scatterlists(h);
5191         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5192         free_irq(h->intr[h->intr_mode], h);
5193 clean2:
5194         unregister_blkdev(h->major, h->devname);
5195 clean1:
5196         cciss_destroy_hba_sysfs_entry(h);
5197 clean0:
5198         pci_release_regions(pdev);
5199 clean_no_release_regions:
5200         h->busy_initializing = 0;
5201
5202         /*
5203          * Deliberately omit pci_disable_device(): it does something nasty to
5204          * Smart Array controllers that pci_enable_device does not undo
5205          */
5206         pci_set_drvdata(pdev, NULL);
5207         free_hba(h);
5208         return -ENODEV;
5209 }
5210
5211 static void cciss_shutdown(struct pci_dev *pdev)
5212 {
5213         ctlr_info_t *h;
5214         char *flush_buf;
5215         int return_code;
5216
5217         h = pci_get_drvdata(pdev);
5218         flush_buf = kzalloc(4, GFP_KERNEL);
5219         if (!flush_buf) {
5220                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
5221                 return;
5222         }
5223         /* write all data in the battery backed cache to disk */
5224         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
5225                 4, 0, CTLR_LUNID, TYPE_CMD);
5226         kfree(flush_buf);
5227         if (return_code != IO_OK)
5228                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
5229         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5230         free_irq(h->intr[h->intr_mode], h);
5231 }
5232
5233 static int cciss_enter_simple_mode(struct ctlr_info *h)
5234 {
5235         u32 trans_support;
5236
5237         trans_support = readl(&(h->cfgtable->TransportSupport));
5238         if (!(trans_support & SIMPLE_MODE))
5239                 return -ENOTSUPP;
5240
5241         h->max_commands = readl(&(h->cfgtable->CmdsOutMax));
5242         writel(CFGTBL_Trans_Simple, &(h->cfgtable->HostWrite.TransportRequest));
5243         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
5244         cciss_wait_for_mode_change_ack(h);
5245         print_cfg_table(h);
5246         if (!(readl(&(h->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
5247                 dev_warn(&h->pdev->dev, "unable to get board into simple mode\n");
5248                 return -ENODEV;
5249         }
5250         h->transMethod = CFGTBL_Trans_Simple;
5251         return 0;
5252 }
5253
5254
5255 static void cciss_remove_one(struct pci_dev *pdev)
5256 {
5257         ctlr_info_t *h;
5258         int i, j;
5259
5260         if (pci_get_drvdata(pdev) == NULL) {
5261                 dev_err(&pdev->dev, "Unable to remove device\n");
5262                 return;
5263         }
5264
5265         h = pci_get_drvdata(pdev);
5266         i = h->ctlr;
5267         if (hba[i] == NULL) {
5268                 dev_err(&pdev->dev, "device appears to already be removed\n");
5269                 return;
5270         }
5271
5272         mutex_lock(&h->busy_shutting_down);
5273
5274         remove_from_scan_list(h);
5275         remove_proc_entry(h->devname, proc_cciss);
5276         unregister_blkdev(h->major, h->devname);
5277
5278         /* remove it from the disk list */
5279         for (j = 0; j < CISS_MAX_LUN; j++) {
5280                 struct gendisk *disk = h->gendisk[j];
5281                 if (disk) {
5282                         struct request_queue *q = disk->queue;
5283
5284                         if (disk->flags & GENHD_FL_UP) {
5285                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
5286                                 del_gendisk(disk);
5287                         }
5288                         if (q)
5289                                 blk_cleanup_queue(q);
5290                 }
5291         }
5292
5293 #ifdef CONFIG_CISS_SCSI_TAPE
5294         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
5295 #endif
5296
5297         cciss_shutdown(pdev);
5298
5299 #ifdef CONFIG_PCI_MSI
5300         if (h->msix_vector)
5301                 pci_disable_msix(h->pdev);
5302         else if (h->msi_vector)
5303                 pci_disable_msi(h->pdev);
5304 #endif                          /* CONFIG_PCI_MSI */
5305
5306         iounmap(h->transtable);
5307         iounmap(h->cfgtable);
5308         iounmap(h->vaddr);
5309
5310         cciss_free_cmd_pool(h);
5311         /* Free up sg elements */
5312         for (j = 0; j < h->nr_cmds; j++)
5313                 kfree(h->scatter_list[j]);
5314         kfree(h->scatter_list);
5315         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5316         kfree(h->blockFetchTable);
5317         if (h->reply_pool)
5318                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
5319                                 h->reply_pool, h->reply_pool_dhandle);
5320         /*
5321          * Deliberately omit pci_disable_device(): it does something nasty to
5322          * Smart Array controllers that pci_enable_device does not undo
5323          */
5324         pci_release_regions(pdev);
5325         pci_set_drvdata(pdev, NULL);
5326         cciss_destroy_hba_sysfs_entry(h);
5327         mutex_unlock(&h->busy_shutting_down);
5328         free_hba(h);
5329 }
5330
5331 static struct pci_driver cciss_pci_driver = {
5332         .name = "cciss",
5333         .probe = cciss_init_one,
5334         .remove = cciss_remove_one,
5335         .id_table = cciss_pci_device_id,        /* id_table */
5336         .shutdown = cciss_shutdown,
5337 };
5338
5339 /*
5340  *  This is it.  Register the PCI driver information for the cards we control
5341  *  the OS will call our registered routines when it finds one of our cards.
5342  */
5343 static int __init cciss_init(void)
5344 {
5345         int err;
5346
5347         /*
5348          * The hardware requires that commands are aligned on a 64-bit
5349          * boundary. Given that we use pci_alloc_consistent() to allocate an
5350          * array of them, the size must be a multiple of 8 bytes.
5351          */
5352         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
5353         printk(KERN_INFO DRIVER_NAME "\n");
5354
5355         err = bus_register(&cciss_bus_type);
5356         if (err)
5357                 return err;
5358
5359         /* Start the scan thread */
5360         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
5361         if (IS_ERR(cciss_scan_thread)) {
5362                 err = PTR_ERR(cciss_scan_thread);
5363                 goto err_bus_unregister;
5364         }
5365
5366         /* Register for our PCI devices */
5367         err = pci_register_driver(&cciss_pci_driver);
5368         if (err)
5369                 goto err_thread_stop;
5370
5371         return err;
5372
5373 err_thread_stop:
5374         kthread_stop(cciss_scan_thread);
5375 err_bus_unregister:
5376         bus_unregister(&cciss_bus_type);
5377
5378         return err;
5379 }
5380
5381 static void __exit cciss_cleanup(void)
5382 {
5383         int i;
5384
5385         pci_unregister_driver(&cciss_pci_driver);
5386         /* double check that all controller entrys have been removed */
5387         for (i = 0; i < MAX_CTLR; i++) {
5388                 if (hba[i] != NULL) {
5389                         dev_warn(&hba[i]->pdev->dev,
5390                                 "had to remove controller\n");
5391                         cciss_remove_one(hba[i]->pdev);
5392                 }
5393         }
5394         kthread_stop(cciss_scan_thread);
5395         if (proc_cciss)
5396                 remove_proc_entry("driver/cciss", NULL);
5397         bus_unregister(&cciss_bus_type);
5398 }
5399
5400 module_init(cciss_init);
5401 module_exit(cciss_cleanup);