drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/delay.h>
  30 #include <linux/major.h>
  31 #include <linux/fs.h>
  32 #include <linux/bio.h>
  33 #include <linux/blkpg.h>
  34 #include <linux/timer.h>
  35 #include <linux/proc_fs.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/init.h>
  38 #include <linux/jiffies.h>
  39 #include <linux/hdreg.h>
  40 #include <linux/spinlock.h>
  41 #include <linux/compat.h>
  42 #include <linux/mutex.h>
  43 #include <asm/uaccess.h>
  44 #include <asm/io.h>
  45
  46 #include <linux/dma-mapping.h>
  47 #include <linux/blkdev.h>
  48 #include <linux/genhd.h>
  49 #include <linux/completion.h>
  50 #include <scsi/scsi.h>
  51 #include <scsi/sg.h>
  52 #include <scsi/scsi_ioctl.h>
  53 #include <linux/cdrom.h>
  54 #include <linux/scatterlist.h>
  55 #include <linux/kthread.h>
  56
  57 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  58 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  59 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  60
  61 /* Embedded module documentation macros - see modules.h */
  62 MODULE_AUTHOR("Hewlett-Packard Company");
  63 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  64 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  65 MODULE_VERSION("3.6.26");
  66 MODULE_LICENSE("GPL");
  67
  68 static DEFINE_MUTEX(cciss_mutex);
  69
  70 #include "cciss_cmd.h"
  71 #include "cciss.h"
  72 #include <linux/cciss_ioctl.h>
  73
  74 /* define the PCI info for the cards we can control */
  75 static const struct pci_device_id cciss_pci_device_id[] = {
  76         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  77         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  78         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  79         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  80         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  85         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  86         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  87         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  88         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  89         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
  96         {0,}
  97 };
  98
  99 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 100
 101 /*  board_id = Subsystem Device ID & Vendor ID
 102  *  product = Marketing Name for the board
 103  *  access = Address of the struct of function pointers
 104  */
 105 static struct board_type products[] = {
 106         {0x40700E11, "Smart Array 5300", &SA5_access},
 107         {0x40800E11, "Smart Array 5i", &SA5B_access},
 108         {0x40820E11, "Smart Array 532", &SA5B_access},
 109         {0x40830E11, "Smart Array 5312", &SA5B_access},
 110         {0x409A0E11, "Smart Array 641", &SA5_access},
 111         {0x409B0E11, "Smart Array 642", &SA5_access},
 112         {0x409C0E11, "Smart Array 6400", &SA5_access},
 113         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 114         {0x40910E11, "Smart Array 6i", &SA5_access},
 115         {0x3225103C, "Smart Array P600", &SA5_access},
 116         {0x3235103C, "Smart Array P400i", &SA5_access},
 117         {0x3211103C, "Smart Array E200i", &SA5_access},
 118         {0x3212103C, "Smart Array E200", &SA5_access},
 119         {0x3213103C, "Smart Array E200i", &SA5_access},
 120         {0x3214103C, "Smart Array E200i", &SA5_access},
 121         {0x3215103C, "Smart Array E200i", &SA5_access},
 122         {0x3237103C, "Smart Array E500", &SA5_access},
 123         {0x3223103C, "Smart Array P800", &SA5_access},
 124         {0x3234103C, "Smart Array P400", &SA5_access},
 125         {0x323D103C, "Smart Array P700m", &SA5_access},
 126 };
 127
 128 /* How long to wait (in milliseconds) for board to go into simple mode */
 129 #define MAX_CONFIG_WAIT 30000
 130 #define MAX_IOCTL_CONFIG_WAIT 1000
 131
 132 /*define how many times we will try a command because of bus resets */
 133 #define MAX_CMD_RETRIES 3
 134
 135 #define MAX_CTLR        32
 136
 137 /* Originally cciss driver only supports 8 major numbers */
 138 #define MAX_CTLR_ORIG   8
 139
 140 static ctlr_info_t *hba[MAX_CTLR];
 141
 142 static struct task_struct *cciss_scan_thread;
 143 static DEFINE_MUTEX(scan_mutex);
 144 static LIST_HEAD(scan_q);
 145
 146 static void do_cciss_request(struct request_queue *q);
 147 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 148 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 149 static int cciss_open(struct block_device *bdev, fmode_t mode);
 150 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 151 static int cciss_release(struct gendisk *disk, fmode_t mode);
 152 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 153                     unsigned int cmd, unsigned long arg);
 154 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 155                        unsigned int cmd, unsigned long arg);
 156 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 157
 158 static int cciss_revalidate(struct gendisk *disk);
 159 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 160 static int deregister_disk(ctlr_info_t *h, int drv_index,
 161                            int clear_all, int via_ioctl);
 162
 163 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 164                         sector_t *total_size, unsigned int *block_size);
 165 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 166                         sector_t *total_size, unsigned int *block_size);
 167 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 168                         sector_t total_size,
 169                         unsigned int block_size, InquiryData_struct *inq_buff,
 170                                    drive_info_struct *drv);
 171 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
 172 static void start_io(ctlr_info_t *h);
 173 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 174                         __u8 page_code, unsigned char scsi3addr[],
 175                         int cmd_type);
 176 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 177         int attempt_retry);
 178 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 179
 180 static int add_to_scan_list(struct ctlr_info *h);
 181 static int scan_thread(void *data);
 182 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 183 static void cciss_hba_release(struct device *dev);
 184 static void cciss_device_release(struct device *dev);
 185 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 186 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 187 static inline u32 next_command(ctlr_info_t *h);
 188 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
 189         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 190         u64 *cfg_offset);
 191 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 192         unsigned long *memory_bar);
 193
 194
 195 /* performant mode helper functions */
 196 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 197                                 int *bucket_map);
 198 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 199
 200 #ifdef CONFIG_PROC_FS
 201 static void cciss_procinit(ctlr_info_t *h);
 202 #else
 203 static void cciss_procinit(ctlr_info_t *h)
 204 {
 205 }
 206 #endif                          /* CONFIG_PROC_FS */
 207
 208 #ifdef CONFIG_COMPAT
 209 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 210                               unsigned, unsigned long);
 211 #endif
 212
 213 static const struct block_device_operations cciss_fops = {
 214         .owner = THIS_MODULE,
 215         .open = cciss_unlocked_open,
 216         .release = cciss_release,
 217         .ioctl = do_ioctl,
 218         .getgeo = cciss_getgeo,
 219 #ifdef CONFIG_COMPAT
 220         .compat_ioctl = cciss_compat_ioctl,
 221 #endif
 222         .revalidate_disk = cciss_revalidate,
 223 };
 224
 225 /* set_performant_mode: Modify the tag for cciss performant
 226  * set bit 0 for pull model, bits 3-1 for block fetch
 227  * register number
 228  */
 229 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 230 {
 231         if (likely(h->transMethod == CFGTBL_Trans_Performant))
 232                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 233 }
 234
 235 /*
 236  * Enqueuing and dequeuing functions for cmdlists.
 237  */
 238 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 239 {
 240         hlist_add_head(&c->list, list);
 241 }
 242
 243 static inline void removeQ(CommandList_struct *c)
 244 {
 245         /*
 246          * After kexec/dump some commands might still
 247          * be in flight, which the firmware will try
 248          * to complete. Resetting the firmware doesn't work
 249          * with old fw revisions, so we have to mark
 250          * them off as 'stale' to prevent the driver from
 251          * falling over.
 252          */
 253         if (WARN_ON(hlist_unhashed(&c->list))) {
 254                 c->cmd_type = CMD_MSG_STALE;
 255                 return;
 256         }
 257
 258         hlist_del_init(&c->list);
 259 }
 260
 261 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 262         CommandList_struct *c)
 263 {
 264         unsigned long flags;
 265         set_performant_mode(h, c);
 266         spin_lock_irqsave(&h->lock, flags);
 267         addQ(&h->reqQ, c);
 268         h->Qdepth++;
 269         if (h->Qdepth > h->maxQsinceinit)
 270                 h->maxQsinceinit = h->Qdepth;
 271         start_io(h);
 272         spin_unlock_irqrestore(&h->lock, flags);
 273 }
 274
 275 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 276         int nr_cmds)
 277 {
 278         int i;
 279
 280         if (!cmd_sg_list)
 281                 return;
 282         for (i = 0; i < nr_cmds; i++) {
 283                 kfree(cmd_sg_list[i]);
 284                 cmd_sg_list[i] = NULL;
 285         }
 286         kfree(cmd_sg_list);
 287 }
 288
 289 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 290         ctlr_info_t *h, int chainsize, int nr_cmds)
 291 {
 292         int j;
 293         SGDescriptor_struct **cmd_sg_list;
 294
 295         if (chainsize <= 0)
 296                 return NULL;
 297
 298         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 299         if (!cmd_sg_list)
 300                 return NULL;
 301
 302         /* Build up chain blocks for each command */
 303         for (j = 0; j < nr_cmds; j++) {
 304                 /* Need a block of chainsized s/g elements. */
 305                 cmd_sg_list[j] = kmalloc((chainsize *
 306                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 307                 if (!cmd_sg_list[j]) {
 308                         dev_err(&h->pdev->dev, "Cannot get memory "
 309                                 "for s/g chains.\n");
 310                         goto clean;
 311                 }
 312         }
 313         return cmd_sg_list;
 314 clean:
 315         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 316         return NULL;
 317 }
 318
 319 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 320 {
 321         SGDescriptor_struct *chain_sg;
 322         u64bit temp64;
 323
 324         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 325                 return;
 326
 327         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 328         temp64.val32.lower = chain_sg->Addr.lower;
 329         temp64.val32.upper = chain_sg->Addr.upper;
 330         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 331 }
 332
 333 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 334         SGDescriptor_struct *chain_block, int len)
 335 {
 336         SGDescriptor_struct *chain_sg;
 337         u64bit temp64;
 338
 339         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 340         chain_sg->Ext = CCISS_SG_CHAIN;
 341         chain_sg->Len = len;
 342         temp64.val = pci_map_single(h->pdev, chain_block, len,
 343                                 PCI_DMA_TODEVICE);
 344         chain_sg->Addr.lower = temp64.val32.lower;
 345         chain_sg->Addr.upper = temp64.val32.upper;
 346 }
 347
 348 #include "cciss_scsi.c"         /* For SCSI tape support */
 349
 350 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 351         "UNKNOWN"
 352 };
 353 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 354
 355 #ifdef CONFIG_PROC_FS
 356
 357 /*
 358  * Report information about this controller.
 359  */
 360 #define ENG_GIG 1000000000
 361 #define ENG_GIG_FACTOR (ENG_GIG/512)
 362 #define ENGAGE_SCSI     "engage scsi"
 363
 364 static struct proc_dir_entry *proc_cciss;
 365
 366 static void cciss_seq_show_header(struct seq_file *seq)
 367 {
 368         ctlr_info_t *h = seq->private;
 369
 370         seq_printf(seq, "%s: HP %s Controller\n"
 371                 "Board ID: 0x%08lx\n"
 372                 "Firmware Version: %c%c%c%c\n"
 373                 "IRQ: %d\n"
 374                 "Logical drives: %d\n"
 375                 "Current Q depth: %d\n"
 376                 "Current # commands on controller: %d\n"
 377                 "Max Q depth since init: %d\n"
 378                 "Max # commands on controller since init: %d\n"
 379                 "Max SG entries since init: %d\n",
 380                 h->devname,
 381                 h->product_name,
 382                 (unsigned long)h->board_id,
 383                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 384                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 385                 h->num_luns,
 386                 h->Qdepth, h->commands_outstanding,
 387                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 388
 389 #ifdef CONFIG_CISS_SCSI_TAPE
 390         cciss_seq_tape_report(seq, h);
 391 #endif /* CONFIG_CISS_SCSI_TAPE */
 392 }
 393
 394 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 395 {
 396         ctlr_info_t *h = seq->private;
 397         unsigned long flags;
 398
 399         /* prevent displaying bogus info during configuration
 400          * or deconfiguration of a logical volume
 401          */
 402         spin_lock_irqsave(&h->lock, flags);
 403         if (h->busy_configuring) {
 404                 spin_unlock_irqrestore(&h->lock, flags);
 405                 return ERR_PTR(-EBUSY);
 406         }
 407         h->busy_configuring = 1;
 408         spin_unlock_irqrestore(&h->lock, flags);
 409
 410         if (*pos == 0)
 411                 cciss_seq_show_header(seq);
 412
 413         return pos;
 414 }
 415
 416 static int cciss_seq_show(struct seq_file *seq, void *v)
 417 {
 418         sector_t vol_sz, vol_sz_frac;
 419         ctlr_info_t *h = seq->private;
 420         unsigned ctlr = h->ctlr;
 421         loff_t *pos = v;
 422         drive_info_struct *drv = h->drv[*pos];
 423
 424         if (*pos > h->highest_lun)
 425                 return 0;
 426
 427         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 428                 return 0;
 429
 430         if (drv->heads == 0)
 431                 return 0;
 432
 433         vol_sz = drv->nr_blocks;
 434         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 435         vol_sz_frac *= 100;
 436         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 437
 438         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 439                 drv->raid_level = RAID_UNKNOWN;
 440         seq_printf(seq, "cciss/c%dd%d:"
 441                         "\t%4u.%02uGB\tRAID %s\n",
 442                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 443                         raid_label[drv->raid_level]);
 444         return 0;
 445 }
 446
 447 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 448 {
 449         ctlr_info_t *h = seq->private;
 450
 451         if (*pos > h->highest_lun)
 452                 return NULL;
 453         *pos += 1;
 454
 455         return pos;
 456 }
 457
 458 static void cciss_seq_stop(struct seq_file *seq, void *v)
 459 {
 460         ctlr_info_t *h = seq->private;
 461
 462         /* Only reset h->busy_configuring if we succeeded in setting
 463          * it during cciss_seq_start. */
 464         if (v == ERR_PTR(-EBUSY))
 465                 return;
 466
 467         h->busy_configuring = 0;
 468 }
 469
 470 static const struct seq_operations cciss_seq_ops = {
 471         .start = cciss_seq_start,
 472         .show  = cciss_seq_show,
 473         .next  = cciss_seq_next,
 474         .stop  = cciss_seq_stop,
 475 };
 476
 477 static int cciss_seq_open(struct inode *inode, struct file *file)
 478 {
 479         int ret = seq_open(file, &cciss_seq_ops);
 480         struct seq_file *seq = file->private_data;
 481
 482         if (!ret)
 483                 seq->private = PDE(inode)->data;
 484
 485         return ret;
 486 }
 487
 488 static ssize_t
 489 cciss_proc_write(struct file *file, const char __user *buf,
 490                  size_t length, loff_t *ppos)
 491 {
 492         int err;
 493         char *buffer;
 494
 495 #ifndef CONFIG_CISS_SCSI_TAPE
 496         return -EINVAL;
 497 #endif
 498
 499         if (!buf || length > PAGE_SIZE - 1)
 500                 return -EINVAL;
 501
 502         buffer = (char *)__get_free_page(GFP_KERNEL);
 503         if (!buffer)
 504                 return -ENOMEM;
 505
 506         err = -EFAULT;
 507         if (copy_from_user(buffer, buf, length))
 508                 goto out;
 509         buffer[length] = '\0';
 510
 511 #ifdef CONFIG_CISS_SCSI_TAPE
 512         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 513                 struct seq_file *seq = file->private_data;
 514                 ctlr_info_t *h = seq->private;
 515
 516                 err = cciss_engage_scsi(h);
 517                 if (err == 0)
 518                         err = length;
 519         } else
 520 #endif /* CONFIG_CISS_SCSI_TAPE */
 521                 err = -EINVAL;
 522         /* might be nice to have "disengage" too, but it's not
 523            safely possible. (only 1 module use count, lock issues.) */
 524
 525 out:
 526         free_page((unsigned long)buffer);
 527         return err;
 528 }
 529
 530 static const struct file_operations cciss_proc_fops = {
 531         .owner   = THIS_MODULE,
 532         .open    = cciss_seq_open,
 533         .read    = seq_read,
 534         .llseek  = seq_lseek,
 535         .release = seq_release,
 536         .write   = cciss_proc_write,
 537 };
 538
 539 static void __devinit cciss_procinit(ctlr_info_t *h)
 540 {
 541         struct proc_dir_entry *pde;
 542
 543         if (proc_cciss == NULL)
 544                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 545         if (!proc_cciss)
 546                 return;
 547         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 548                                         S_IROTH, proc_cciss,
 549                                         &cciss_proc_fops, h);
 550 }
 551 #endif                          /* CONFIG_PROC_FS */
 552
 553 #define MAX_PRODUCT_NAME_LEN 19
 554
 555 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 556 #define to_drv(n) container_of(n, drive_info_struct, dev)
 557
 558 static ssize_t host_store_rescan(struct device *dev,
 559                                  struct device_attribute *attr,
 560                                  const char *buf, size_t count)
 561 {
 562         struct ctlr_info *h = to_hba(dev);
 563
 564         add_to_scan_list(h);
 565         wake_up_process(cciss_scan_thread);
 566         wait_for_completion_interruptible(&h->scan_wait);
 567
 568         return count;
 569 }
 570 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 571
 572 static ssize_t dev_show_unique_id(struct device *dev,
 573                                  struct device_attribute *attr,
 574                                  char *buf)
 575 {
 576         drive_info_struct *drv = to_drv(dev);
 577         struct ctlr_info *h = to_hba(drv->dev.parent);
 578         __u8 sn[16];
 579         unsigned long flags;
 580         int ret = 0;
 581
 582         spin_lock_irqsave(&h->lock, flags);
 583         if (h->busy_configuring)
 584                 ret = -EBUSY;
 585         else
 586                 memcpy(sn, drv->serial_no, sizeof(sn));
 587         spin_unlock_irqrestore(&h->lock, flags);
 588
 589         if (ret)
 590                 return ret;
 591         else
 592                 return snprintf(buf, 16 * 2 + 2,
 593                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 594                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 595                                 sn[0], sn[1], sn[2], sn[3],
 596                                 sn[4], sn[5], sn[6], sn[7],
 597                                 sn[8], sn[9], sn[10], sn[11],
 598                                 sn[12], sn[13], sn[14], sn[15]);
 599 }
 600 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 601
 602 static ssize_t dev_show_vendor(struct device *dev,
 603                                struct device_attribute *attr,
 604                                char *buf)
 605 {
 606         drive_info_struct *drv = to_drv(dev);
 607         struct ctlr_info *h = to_hba(drv->dev.parent);
 608         char vendor[VENDOR_LEN + 1];
 609         unsigned long flags;
 610         int ret = 0;
 611
 612         spin_lock_irqsave(&h->lock, flags);
 613         if (h->busy_configuring)
 614                 ret = -EBUSY;
 615         else
 616                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 617         spin_unlock_irqrestore(&h->lock, flags);
 618
 619         if (ret)
 620                 return ret;
 621         else
 622                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 623 }
 624 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 625
 626 static ssize_t dev_show_model(struct device *dev,
 627                               struct device_attribute *attr,
 628                               char *buf)
 629 {
 630         drive_info_struct *drv = to_drv(dev);
 631         struct ctlr_info *h = to_hba(drv->dev.parent);
 632         char model[MODEL_LEN + 1];
 633         unsigned long flags;
 634         int ret = 0;
 635
 636         spin_lock_irqsave(&h->lock, flags);
 637         if (h->busy_configuring)
 638                 ret = -EBUSY;
 639         else
 640                 memcpy(model, drv->model, MODEL_LEN + 1);
 641         spin_unlock_irqrestore(&h->lock, flags);
 642
 643         if (ret)
 644                 return ret;
 645         else
 646                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 647 }
 648 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 649
 650 static ssize_t dev_show_rev(struct device *dev,
 651                             struct device_attribute *attr,
 652                             char *buf)
 653 {
 654         drive_info_struct *drv = to_drv(dev);
 655         struct ctlr_info *h = to_hba(drv->dev.parent);
 656         char rev[REV_LEN + 1];
 657         unsigned long flags;
 658         int ret = 0;
 659
 660         spin_lock_irqsave(&h->lock, flags);
 661         if (h->busy_configuring)
 662                 ret = -EBUSY;
 663         else
 664                 memcpy(rev, drv->rev, REV_LEN + 1);
 665         spin_unlock_irqrestore(&h->lock, flags);
 666
 667         if (ret)
 668                 return ret;
 669         else
 670                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 671 }
 672 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 673
 674 static ssize_t cciss_show_lunid(struct device *dev,
 675                                 struct device_attribute *attr, char *buf)
 676 {
 677         drive_info_struct *drv = to_drv(dev);
 678         struct ctlr_info *h = to_hba(drv->dev.parent);
 679         unsigned long flags;
 680         unsigned char lunid[8];
 681
 682         spin_lock_irqsave(&h->lock, flags);
 683         if (h->busy_configuring) {
 684                 spin_unlock_irqrestore(&h->lock, flags);
 685                 return -EBUSY;
 686         }
 687         if (!drv->heads) {
 688                 spin_unlock_irqrestore(&h->lock, flags);
 689                 return -ENOTTY;
 690         }
 691         memcpy(lunid, drv->LunID, sizeof(lunid));
 692         spin_unlock_irqrestore(&h->lock, flags);
 693         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 694                 lunid[0], lunid[1], lunid[2], lunid[3],
 695                 lunid[4], lunid[5], lunid[6], lunid[7]);
 696 }
 697 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 698
 699 static ssize_t cciss_show_raid_level(struct device *dev,
 700                                      struct device_attribute *attr, char *buf)
 701 {
 702         drive_info_struct *drv = to_drv(dev);
 703         struct ctlr_info *h = to_hba(drv->dev.parent);
 704         int raid;
 705         unsigned long flags;
 706
 707         spin_lock_irqsave(&h->lock, flags);
 708         if (h->busy_configuring) {
 709                 spin_unlock_irqrestore(&h->lock, flags);
 710                 return -EBUSY;
 711         }
 712         raid = drv->raid_level;
 713         spin_unlock_irqrestore(&h->lock, flags);
 714         if (raid < 0 || raid > RAID_UNKNOWN)
 715                 raid = RAID_UNKNOWN;
 716
 717         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 718                         raid_label[raid]);
 719 }
 720 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 721
 722 static ssize_t cciss_show_usage_count(struct device *dev,
 723                                       struct device_attribute *attr, char *buf)
 724 {
 725         drive_info_struct *drv = to_drv(dev);
 726         struct ctlr_info *h = to_hba(drv->dev.parent);
 727         unsigned long flags;
 728         int count;
 729
 730         spin_lock_irqsave(&h->lock, flags);
 731         if (h->busy_configuring) {
 732                 spin_unlock_irqrestore(&h->lock, flags);
 733                 return -EBUSY;
 734         }
 735         count = drv->usage_count;
 736         spin_unlock_irqrestore(&h->lock, flags);
 737         return snprintf(buf, 20, "%d\n", count);
 738 }
 739 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 740
 741 static struct attribute *cciss_host_attrs[] = {
 742         &dev_attr_rescan.attr,
 743         NULL
 744 };
 745
 746 static struct attribute_group cciss_host_attr_group = {
 747         .attrs = cciss_host_attrs,
 748 };
 749
 750 static const struct attribute_group *cciss_host_attr_groups[] = {
 751         &cciss_host_attr_group,
 752         NULL
 753 };
 754
 755 static struct device_type cciss_host_type = {
 756         .name           = "cciss_host",
 757         .groups         = cciss_host_attr_groups,
 758         .release        = cciss_hba_release,
 759 };
 760
 761 static struct attribute *cciss_dev_attrs[] = {
 762         &dev_attr_unique_id.attr,
 763         &dev_attr_model.attr,
 764         &dev_attr_vendor.attr,
 765         &dev_attr_rev.attr,
 766         &dev_attr_lunid.attr,
 767         &dev_attr_raid_level.attr,
 768         &dev_attr_usage_count.attr,
 769         NULL
 770 };
 771
 772 static struct attribute_group cciss_dev_attr_group = {
 773         .attrs = cciss_dev_attrs,
 774 };
 775
 776 static const struct attribute_group *cciss_dev_attr_groups[] = {
 777         &cciss_dev_attr_group,
 778         NULL
 779 };
 780
 781 static struct device_type cciss_dev_type = {
 782         .name           = "cciss_device",
 783         .groups         = cciss_dev_attr_groups,
 784         .release        = cciss_device_release,
 785 };
 786
 787 static struct bus_type cciss_bus_type = {
 788         .name           = "cciss",
 789 };
 790
 791 /*
 792  * cciss_hba_release is called when the reference count
 793  * of h->dev goes to zero.
 794  */
 795 static void cciss_hba_release(struct device *dev)
 796 {
 797         /*
 798          * nothing to do, but need this to avoid a warning
 799          * about not having a release handler from lib/kref.c.
 800          */
 801 }
 802
 803 /*
 804  * Initialize sysfs entry for each controller.  This sets up and registers
 805  * the 'cciss#' directory for each individual controller under
 806  * /sys/bus/pci/devices/<dev>/.
 807  */
 808 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 809 {
 810         device_initialize(&h->dev);
 811         h->dev.type = &cciss_host_type;
 812         h->dev.bus = &cciss_bus_type;
 813         dev_set_name(&h->dev, "%s", h->devname);
 814         h->dev.parent = &h->pdev->dev;
 815
 816         return device_add(&h->dev);
 817 }
 818
 819 /*
 820  * Remove sysfs entries for an hba.
 821  */
 822 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 823 {
 824         device_del(&h->dev);
 825         put_device(&h->dev); /* final put. */
 826 }
 827
 828 /* cciss_device_release is called when the reference count
 829  * of h->drv[x]dev goes to zero.
 830  */
 831 static void cciss_device_release(struct device *dev)
 832 {
 833         drive_info_struct *drv = to_drv(dev);
 834         kfree(drv);
 835 }
 836
 837 /*
 838  * Initialize sysfs for each logical drive.  This sets up and registers
 839  * the 'c#d#' directory for each individual logical drive under
 840  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 841  * /sys/block/cciss!c#d# to this entry.
 842  */
 843 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 844                                        int drv_index)
 845 {
 846         struct device *dev;
 847
 848         if (h->drv[drv_index]->device_initialized)
 849                 return 0;
 850
 851         dev = &h->drv[drv_index]->dev;
 852         device_initialize(dev);
 853         dev->type = &cciss_dev_type;
 854         dev->bus = &cciss_bus_type;
 855         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 856         dev->parent = &h->dev;
 857         h->drv[drv_index]->device_initialized = 1;
 858         return device_add(dev);
 859 }
 860
 861 /*
 862  * Remove sysfs entries for a logical drive.
 863  */
 864 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 865         int ctlr_exiting)
 866 {
 867         struct device *dev = &h->drv[drv_index]->dev;
 868
 869         /* special case for c*d0, we only destroy it on controller exit */
 870         if (drv_index == 0 && !ctlr_exiting)
 871                 return;
 872
 873         device_del(dev);
 874         put_device(dev); /* the "final" put. */
 875         h->drv[drv_index] = NULL;
 876 }
 877
 878 /*
 879  * For operations that cannot sleep, a command block is allocated at init,
 880  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 881  * which ones are free or in use.
 882  */
 883 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 884 {
 885         CommandList_struct *c;
 886         int i;
 887         u64bit temp64;
 888         dma_addr_t cmd_dma_handle, err_dma_handle;
 889
 890         do {
 891                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 892                 if (i == h->nr_cmds)
 893                         return NULL;
 894         } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
 895                   h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 896         c = h->cmd_pool + i;
 897         memset(c, 0, sizeof(CommandList_struct));
 898         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
 899         c->err_info = h->errinfo_pool + i;
 900         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 901         err_dma_handle = h->errinfo_pool_dhandle
 902             + i * sizeof(ErrorInfo_struct);
 903         h->nr_allocs++;
 904
 905         c->cmdindex = i;
 906
 907         INIT_HLIST_NODE(&c->list);
 908         c->busaddr = (__u32) cmd_dma_handle;
 909         temp64.val = (__u64) err_dma_handle;
 910         c->ErrDesc.Addr.lower = temp64.val32.lower;
 911         c->ErrDesc.Addr.upper = temp64.val32.upper;
 912         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 913
 914         c->ctlr = h->ctlr;
 915         return c;
 916 }
 917
 918 /* allocate a command using pci_alloc_consistent, used for ioctls,
 919  * etc., not for the main i/o path.
 920  */
 921 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
 922 {
 923         CommandList_struct *c;
 924         u64bit temp64;
 925         dma_addr_t cmd_dma_handle, err_dma_handle;
 926
 927         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 928                 sizeof(CommandList_struct), &cmd_dma_handle);
 929         if (c == NULL)
 930                 return NULL;
 931         memset(c, 0, sizeof(CommandList_struct));
 932
 933         c->cmdindex = -1;
 934
 935         c->err_info = (ErrorInfo_struct *)
 936             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 937                     &err_dma_handle);
 938
 939         if (c->err_info == NULL) {
 940                 pci_free_consistent(h->pdev,
 941                         sizeof(CommandList_struct), c, cmd_dma_handle);
 942                 return NULL;
 943         }
 944         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 945
 946         INIT_HLIST_NODE(&c->list);
 947         c->busaddr = (__u32) cmd_dma_handle;
 948         temp64.val = (__u64) err_dma_handle;
 949         c->ErrDesc.Addr.lower = temp64.val32.lower;
 950         c->ErrDesc.Addr.upper = temp64.val32.upper;
 951         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 952
 953         c->ctlr = h->ctlr;
 954         return c;
 955 }
 956
 957 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
 958 {
 959         int i;
 960
 961         i = c - h->cmd_pool;
 962         clear_bit(i & (BITS_PER_LONG - 1),
 963                   h->cmd_pool_bits + (i / BITS_PER_LONG));
 964         h->nr_frees++;
 965 }
 966
 967 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
 968 {
 969         u64bit temp64;
 970
 971         temp64.val32.lower = c->ErrDesc.Addr.lower;
 972         temp64.val32.upper = c->ErrDesc.Addr.upper;
 973         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 974                             c->err_info, (dma_addr_t) temp64.val);
 975         pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 976                             c, (dma_addr_t) c->busaddr);
 977 }
 978
 979 static inline ctlr_info_t *get_host(struct gendisk *disk)
 980 {
 981         return disk->queue->queuedata;
 982 }
 983
 984 static inline drive_info_struct *get_drv(struct gendisk *disk)
 985 {
 986         return disk->private_data;
 987 }
 988
 989 /*
 990  * Open.  Make sure the device is really there.
 991  */
 992 static int cciss_open(struct block_device *bdev, fmode_t mode)
 993 {
 994         ctlr_info_t *h = get_host(bdev->bd_disk);
 995         drive_info_struct *drv = get_drv(bdev->bd_disk);
 996
 997         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
 998         if (drv->busy_configuring)
 999                 return -EBUSY;
1000         /*
1001          * Root is allowed to open raw volume zero even if it's not configured
1002          * so array config can still work. Root is also allowed to open any
1003          * volume that has a LUN ID, so it can issue IOCTL to reread the
1004          * disk information.  I don't think I really like this
1005          * but I'm already using way to many device nodes to claim another one
1006          * for "raw controller".
1007          */
1008         if (drv->heads == 0) {
1009                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1010                         /* if not node 0 make sure it is a partition = 0 */
1011                         if (MINOR(bdev->bd_dev) & 0x0f) {
1012                                 return -ENXIO;
1013                                 /* if it is, make sure we have a LUN ID */
1014                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1015                                 sizeof(drv->LunID))) {
1016                                 return -ENXIO;
1017                         }
1018                 }
1019                 if (!capable(CAP_SYS_ADMIN))
1020                         return -EPERM;
1021         }
1022         drv->usage_count++;
1023         h->usage_count++;
1024         return 0;
1025 }
1026
1027 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1028 {
1029         int ret;
1030
1031         mutex_lock(&cciss_mutex);
1032         ret = cciss_open(bdev, mode);
1033         mutex_unlock(&cciss_mutex);
1034
1035         return ret;
1036 }
1037
1038 /*
1039  * Close.  Sync first.
1040  */
1041 static int cciss_release(struct gendisk *disk, fmode_t mode)
1042 {
1043         ctlr_info_t *h;
1044         drive_info_struct *drv;
1045
1046         mutex_lock(&cciss_mutex);
1047         h = get_host(disk);
1048         drv = get_drv(disk);
1049         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1050         drv->usage_count--;
1051         h->usage_count--;
1052         mutex_unlock(&cciss_mutex);
1053         return 0;
1054 }
1055
1056 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1057                     unsigned cmd, unsigned long arg)
1058 {
1059         int ret;
1060         mutex_lock(&cciss_mutex);
1061         ret = cciss_ioctl(bdev, mode, cmd, arg);
1062         mutex_unlock(&cciss_mutex);
1063         return ret;
1064 }
1065
1066 #ifdef CONFIG_COMPAT
1067
1068 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1069                                   unsigned cmd, unsigned long arg);
1070 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1071                                       unsigned cmd, unsigned long arg);
1072
1073 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1074                               unsigned cmd, unsigned long arg)
1075 {
1076         switch (cmd) {
1077         case CCISS_GETPCIINFO:
1078         case CCISS_GETINTINFO:
1079         case CCISS_SETINTINFO:
1080         case CCISS_GETNODENAME:
1081         case CCISS_SETNODENAME:
1082         case CCISS_GETHEARTBEAT:
1083         case CCISS_GETBUSTYPES:
1084         case CCISS_GETFIRMVER:
1085         case CCISS_GETDRIVVER:
1086         case CCISS_REVALIDVOLS:
1087         case CCISS_DEREGDISK:
1088         case CCISS_REGNEWDISK:
1089         case CCISS_REGNEWD:
1090         case CCISS_RESCANDISK:
1091         case CCISS_GETLUNINFO:
1092                 return do_ioctl(bdev, mode, cmd, arg);
1093
1094         case CCISS_PASSTHRU32:
1095                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1096         case CCISS_BIG_PASSTHRU32:
1097                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1098
1099         default:
1100                 return -ENOIOCTLCMD;
1101         }
1102 }
1103
1104 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1105                                   unsigned cmd, unsigned long arg)
1106 {
1107         IOCTL32_Command_struct __user *arg32 =
1108             (IOCTL32_Command_struct __user *) arg;
1109         IOCTL_Command_struct arg64;
1110         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1111         int err;
1112         u32 cp;
1113
1114         err = 0;
1115         err |=
1116             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1117                            sizeof(arg64.LUN_info));
1118         err |=
1119             copy_from_user(&arg64.Request, &arg32->Request,
1120                            sizeof(arg64.Request));
1121         err |=
1122             copy_from_user(&arg64.error_info, &arg32->error_info,
1123                            sizeof(arg64.error_info));
1124         err |= get_user(arg64.buf_size, &arg32->buf_size);
1125         err |= get_user(cp, &arg32->buf);
1126         arg64.buf = compat_ptr(cp);
1127         err |= copy_to_user(p, &arg64, sizeof(arg64));
1128
1129         if (err)
1130                 return -EFAULT;
1131
1132         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1133         if (err)
1134                 return err;
1135         err |=
1136             copy_in_user(&arg32->error_info, &p->error_info,
1137                          sizeof(arg32->error_info));
1138         if (err)
1139                 return -EFAULT;
1140         return err;
1141 }
1142
1143 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1144                                       unsigned cmd, unsigned long arg)
1145 {
1146         BIG_IOCTL32_Command_struct __user *arg32 =
1147             (BIG_IOCTL32_Command_struct __user *) arg;
1148         BIG_IOCTL_Command_struct arg64;
1149         BIG_IOCTL_Command_struct __user *p =
1150             compat_alloc_user_space(sizeof(arg64));
1151         int err;
1152         u32 cp;
1153
1154         memset(&arg64, 0, sizeof(arg64));
1155         err = 0;
1156         err |=
1157             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1158                            sizeof(arg64.LUN_info));
1159         err |=
1160             copy_from_user(&arg64.Request, &arg32->Request,
1161                            sizeof(arg64.Request));
1162         err |=
1163             copy_from_user(&arg64.error_info, &arg32->error_info,
1164                            sizeof(arg64.error_info));
1165         err |= get_user(arg64.buf_size, &arg32->buf_size);
1166         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1167         err |= get_user(cp, &arg32->buf);
1168         arg64.buf = compat_ptr(cp);
1169         err |= copy_to_user(p, &arg64, sizeof(arg64));
1170
1171         if (err)
1172                 return -EFAULT;
1173
1174         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1175         if (err)
1176                 return err;
1177         err |=
1178             copy_in_user(&arg32->error_info, &p->error_info,
1179                          sizeof(arg32->error_info));
1180         if (err)
1181                 return -EFAULT;
1182         return err;
1183 }
1184 #endif
1185
1186 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1187 {
1188         drive_info_struct *drv = get_drv(bdev->bd_disk);
1189
1190         if (!drv->cylinders)
1191                 return -ENXIO;
1192
1193         geo->heads = drv->heads;
1194         geo->sectors = drv->sectors;
1195         geo->cylinders = drv->cylinders;
1196         return 0;
1197 }
1198
1199 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1200 {
1201         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1202                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1203                 (void)check_for_unit_attention(h, c);
1204 }
1205
1206 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1207 {
1208         cciss_pci_info_struct pciinfo;
1209
1210         if (!argp)
1211                 return -EINVAL;
1212         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1213         pciinfo.bus = h->pdev->bus->number;
1214         pciinfo.dev_fn = h->pdev->devfn;
1215         pciinfo.board_id = h->board_id;
1216         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1217                 return -EFAULT;
1218         return 0;
1219 }
1220
1221 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1222 {
1223         cciss_coalint_struct intinfo;
1224
1225         if (!argp)
1226                 return -EINVAL;
1227         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1228         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1229         if (copy_to_user
1230             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1231                 return -EFAULT;
1232         return 0;
1233 }
1234
1235 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1236 {
1237         cciss_coalint_struct intinfo;
1238         unsigned long flags;
1239         int i;
1240
1241         if (!argp)
1242                 return -EINVAL;
1243         if (!capable(CAP_SYS_ADMIN))
1244                 return -EPERM;
1245         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1246                 return -EFAULT;
1247         if ((intinfo.delay == 0) && (intinfo.count == 0))
1248                 return -EINVAL;
1249         spin_lock_irqsave(&h->lock, flags);
1250         /* Update the field, and then ring the doorbell */
1251         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1252         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1253         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1254
1255         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1256                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1257                         break;
1258                 udelay(1000); /* delay and try again */
1259         }
1260         spin_unlock_irqrestore(&h->lock, flags);
1261         if (i >= MAX_IOCTL_CONFIG_WAIT)
1262                 return -EAGAIN;
1263         return 0;
1264 }
1265
1266 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1267 {
1268         NodeName_type NodeName;
1269         int i;
1270
1271         if (!argp)
1272                 return -EINVAL;
1273         for (i = 0; i < 16; i++)
1274                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1275         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1276                 return -EFAULT;
1277         return 0;
1278 }
1279
1280 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1281 {
1282         NodeName_type NodeName;
1283         unsigned long flags;
1284         int i;
1285
1286         if (!argp)
1287                 return -EINVAL;
1288         if (!capable(CAP_SYS_ADMIN))
1289                 return -EPERM;
1290         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1291                 return -EFAULT;
1292         spin_lock_irqsave(&h->lock, flags);
1293         /* Update the field, and then ring the doorbell */
1294         for (i = 0; i < 16; i++)
1295                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1296         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1297         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1298                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1299                         break;
1300                 udelay(1000); /* delay and try again */
1301         }
1302         spin_unlock_irqrestore(&h->lock, flags);
1303         if (i >= MAX_IOCTL_CONFIG_WAIT)
1304                 return -EAGAIN;
1305         return 0;
1306 }
1307
1308 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1309 {
1310         Heartbeat_type heartbeat;
1311
1312         if (!argp)
1313                 return -EINVAL;
1314         heartbeat = readl(&h->cfgtable->HeartBeat);
1315         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1316                 return -EFAULT;
1317         return 0;
1318 }
1319
1320 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1321 {
1322         BusTypes_type BusTypes;
1323
1324         if (!argp)
1325                 return -EINVAL;
1326         BusTypes = readl(&h->cfgtable->BusTypes);
1327         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1328                 return -EFAULT;
1329         return 0;
1330 }
1331
1332 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1333 {
1334         FirmwareVer_type firmware;
1335
1336         if (!argp)
1337                 return -EINVAL;
1338         memcpy(firmware, h->firm_ver, 4);
1339
1340         if (copy_to_user
1341             (argp, firmware, sizeof(FirmwareVer_type)))
1342                 return -EFAULT;
1343         return 0;
1344 }
1345
1346 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1347 {
1348         DriverVer_type DriverVer = DRIVER_VERSION;
1349
1350         if (!argp)
1351                 return -EINVAL;
1352         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1353                 return -EFAULT;
1354         return 0;
1355 }
1356
1357 static int cciss_getluninfo(ctlr_info_t *h,
1358         struct gendisk *disk, void __user *argp)
1359 {
1360         LogvolInfo_struct luninfo;
1361         drive_info_struct *drv = get_drv(disk);
1362
1363         if (!argp)
1364                 return -EINVAL;
1365         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1366         luninfo.num_opens = drv->usage_count;
1367         luninfo.num_parts = 0;
1368         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1369                 return -EFAULT;
1370         return 0;
1371 }
1372
1373 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1374 {
1375         IOCTL_Command_struct iocommand;
1376         CommandList_struct *c;
1377         char *buff = NULL;
1378         u64bit temp64;
1379         DECLARE_COMPLETION_ONSTACK(wait);
1380
1381         if (!argp)
1382                 return -EINVAL;
1383
1384         if (!capable(CAP_SYS_RAWIO))
1385                 return -EPERM;
1386
1387         if (copy_from_user
1388             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1389                 return -EFAULT;
1390         if ((iocommand.buf_size < 1) &&
1391             (iocommand.Request.Type.Direction != XFER_NONE)) {
1392                 return -EINVAL;
1393         }
1394         if (iocommand.buf_size > 0) {
1395                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1396                 if (buff == NULL)
1397                         return -EFAULT;
1398         }
1399         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1400                 /* Copy the data into the buffer we created */
1401                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1402                         kfree(buff);
1403                         return -EFAULT;
1404                 }
1405         } else {
1406                 memset(buff, 0, iocommand.buf_size);
1407         }
1408         c = cmd_special_alloc(h);
1409         if (!c) {
1410                 kfree(buff);
1411                 return -ENOMEM;
1412         }
1413         /* Fill in the command type */
1414         c->cmd_type = CMD_IOCTL_PEND;
1415         /* Fill in Command Header */
1416         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1417         if (iocommand.buf_size > 0) { /* buffer to fill */
1418                 c->Header.SGList = 1;
1419                 c->Header.SGTotal = 1;
1420         } else { /* no buffers to fill */
1421                 c->Header.SGList = 0;
1422                 c->Header.SGTotal = 0;
1423         }
1424         c->Header.LUN = iocommand.LUN_info;
1425         /* use the kernel address the cmd block for tag */
1426         c->Header.Tag.lower = c->busaddr;
1427
1428         /* Fill in Request block */
1429         c->Request = iocommand.Request;
1430
1431         /* Fill in the scatter gather information */
1432         if (iocommand.buf_size > 0) {
1433                 temp64.val = pci_map_single(h->pdev, buff,
1434                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1435                 c->SG[0].Addr.lower = temp64.val32.lower;
1436                 c->SG[0].Addr.upper = temp64.val32.upper;
1437                 c->SG[0].Len = iocommand.buf_size;
1438                 c->SG[0].Ext = 0;  /* we are not chaining */
1439         }
1440         c->waiting = &wait;
1441
1442         enqueue_cmd_and_start_io(h, c);
1443         wait_for_completion(&wait);
1444
1445         /* unlock the buffers from DMA */
1446         temp64.val32.lower = c->SG[0].Addr.lower;
1447         temp64.val32.upper = c->SG[0].Addr.upper;
1448         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1449                          PCI_DMA_BIDIRECTIONAL);
1450         check_ioctl_unit_attention(h, c);
1451
1452         /* Copy the error information out */
1453         iocommand.error_info = *(c->err_info);
1454         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1455                 kfree(buff);
1456                 cmd_special_free(h, c);
1457                 return -EFAULT;
1458         }
1459
1460         if (iocommand.Request.Type.Direction == XFER_READ) {
1461                 /* Copy the data out of the buffer we created */
1462                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1463                         kfree(buff);
1464                         cmd_special_free(h, c);
1465                         return -EFAULT;
1466                 }
1467         }
1468         kfree(buff);
1469         cmd_special_free(h, c);
1470         return 0;
1471 }
1472
1473 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1474 {
1475         BIG_IOCTL_Command_struct *ioc;
1476         CommandList_struct *c;
1477         unsigned char **buff = NULL;
1478         int *buff_size = NULL;
1479         u64bit temp64;
1480         BYTE sg_used = 0;
1481         int status = 0;
1482         int i;
1483         DECLARE_COMPLETION_ONSTACK(wait);
1484         __u32 left;
1485         __u32 sz;
1486         BYTE __user *data_ptr;
1487
1488         if (!argp)
1489                 return -EINVAL;
1490         if (!capable(CAP_SYS_RAWIO))
1491                 return -EPERM;
1492         ioc = (BIG_IOCTL_Command_struct *)
1493             kmalloc(sizeof(*ioc), GFP_KERNEL);
1494         if (!ioc) {
1495                 status = -ENOMEM;
1496                 goto cleanup1;
1497         }
1498         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1499                 status = -EFAULT;
1500                 goto cleanup1;
1501         }
1502         if ((ioc->buf_size < 1) &&
1503             (ioc->Request.Type.Direction != XFER_NONE)) {
1504                 status = -EINVAL;
1505                 goto cleanup1;
1506         }
1507         /* Check kmalloc limits  using all SGs */
1508         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1509                 status = -EINVAL;
1510                 goto cleanup1;
1511         }
1512         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1513                 status = -EINVAL;
1514                 goto cleanup1;
1515         }
1516         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1517         if (!buff) {
1518                 status = -ENOMEM;
1519                 goto cleanup1;
1520         }
1521         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1522         if (!buff_size) {
1523                 status = -ENOMEM;
1524                 goto cleanup1;
1525         }
1526         left = ioc->buf_size;
1527         data_ptr = ioc->buf;
1528         while (left) {
1529                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1530                 buff_size[sg_used] = sz;
1531                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1532                 if (buff[sg_used] == NULL) {
1533                         status = -ENOMEM;
1534                         goto cleanup1;
1535                 }
1536                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1537                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1538                                 status = -EFAULT;
1539                                 goto cleanup1;
1540                         }
1541                 } else {
1542                         memset(buff[sg_used], 0, sz);
1543                 }
1544                 left -= sz;
1545                 data_ptr += sz;
1546                 sg_used++;
1547         }
1548         c = cmd_special_alloc(h);
1549         if (!c) {
1550                 status = -ENOMEM;
1551                 goto cleanup1;
1552         }
1553         c->cmd_type = CMD_IOCTL_PEND;
1554         c->Header.ReplyQueue = 0;
1555         c->Header.SGList = sg_used;
1556         c->Header.SGTotal = sg_used;
1557         c->Header.LUN = ioc->LUN_info;
1558         c->Header.Tag.lower = c->busaddr;
1559
1560         c->Request = ioc->Request;
1561         for (i = 0; i < sg_used; i++) {
1562                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1563                                     PCI_DMA_BIDIRECTIONAL);
1564                 c->SG[i].Addr.lower = temp64.val32.lower;
1565                 c->SG[i].Addr.upper = temp64.val32.upper;
1566                 c->SG[i].Len = buff_size[i];
1567                 c->SG[i].Ext = 0;       /* we are not chaining */
1568         }
1569         c->waiting = &wait;
1570         enqueue_cmd_and_start_io(h, c);
1571         wait_for_completion(&wait);
1572         /* unlock the buffers from DMA */
1573         for (i = 0; i < sg_used; i++) {
1574                 temp64.val32.lower = c->SG[i].Addr.lower;
1575                 temp64.val32.upper = c->SG[i].Addr.upper;
1576                 pci_unmap_single(h->pdev,
1577                         (dma_addr_t) temp64.val, buff_size[i],
1578                         PCI_DMA_BIDIRECTIONAL);
1579         }
1580         check_ioctl_unit_attention(h, c);
1581         /* Copy the error information out */
1582         ioc->error_info = *(c->err_info);
1583         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1584                 cmd_special_free(h, c);
1585                 status = -EFAULT;
1586                 goto cleanup1;
1587         }
1588         if (ioc->Request.Type.Direction == XFER_READ) {
1589                 /* Copy the data out of the buffer we created */
1590                 BYTE __user *ptr = ioc->buf;
1591                 for (i = 0; i < sg_used; i++) {
1592                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1593                                 cmd_special_free(h, c);
1594                                 status = -EFAULT;
1595                                 goto cleanup1;
1596                         }
1597                         ptr += buff_size[i];
1598                 }
1599         }
1600         cmd_special_free(h, c);
1601         status = 0;
1602 cleanup1:
1603         if (buff) {
1604                 for (i = 0; i < sg_used; i++)
1605                         kfree(buff[i]);
1606                 kfree(buff);
1607         }
1608         kfree(buff_size);
1609         kfree(ioc);
1610         return status;
1611 }
1612
1613 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1614         unsigned int cmd, unsigned long arg)
1615 {
1616         struct gendisk *disk = bdev->bd_disk;
1617         ctlr_info_t *h = get_host(disk);
1618         void __user *argp = (void __user *)arg;
1619
1620         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1621                 cmd, arg);
1622         switch (cmd) {
1623         case CCISS_GETPCIINFO:
1624                 return cciss_getpciinfo(h, argp);
1625         case CCISS_GETINTINFO:
1626                 return cciss_getintinfo(h, argp);
1627         case CCISS_SETINTINFO:
1628                 return cciss_setintinfo(h, argp);
1629         case CCISS_GETNODENAME:
1630                 return cciss_getnodename(h, argp);
1631         case CCISS_SETNODENAME:
1632                 return cciss_setnodename(h, argp);
1633         case CCISS_GETHEARTBEAT:
1634                 return cciss_getheartbeat(h, argp);
1635         case CCISS_GETBUSTYPES:
1636                 return cciss_getbustypes(h, argp);
1637         case CCISS_GETFIRMVER:
1638                 return cciss_getfirmver(h, argp);
1639         case CCISS_GETDRIVVER:
1640                 return cciss_getdrivver(h, argp);
1641         case CCISS_DEREGDISK:
1642         case CCISS_REGNEWD:
1643         case CCISS_REVALIDVOLS:
1644                 return rebuild_lun_table(h, 0, 1);
1645         case CCISS_GETLUNINFO:
1646                 return cciss_getluninfo(h, disk, argp);
1647         case CCISS_PASSTHRU:
1648                 return cciss_passthru(h, argp);
1649         case CCISS_BIG_PASSTHRU:
1650                 return cciss_bigpassthru(h, argp);
1651
1652         /* scsi_cmd_ioctl handles these, below, though some are not */
1653         /* very meaningful for cciss.  SG_IO is the main one people want. */
1654
1655         case SG_GET_VERSION_NUM:
1656         case SG_SET_TIMEOUT:
1657         case SG_GET_TIMEOUT:
1658         case SG_GET_RESERVED_SIZE:
1659         case SG_SET_RESERVED_SIZE:
1660         case SG_EMULATED_HOST:
1661         case SG_IO:
1662         case SCSI_IOCTL_SEND_COMMAND:
1663                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1664
1665         /* scsi_cmd_ioctl would normally handle these, below, but */
1666         /* they aren't a good fit for cciss, as CD-ROMs are */
1667         /* not supported, and we don't have any bus/target/lun */
1668         /* which we present to the kernel. */
1669
1670         case CDROM_SEND_PACKET:
1671         case CDROMCLOSETRAY:
1672         case CDROMEJECT:
1673         case SCSI_IOCTL_GET_IDLUN:
1674         case SCSI_IOCTL_GET_BUS_NUMBER:
1675         default:
1676                 return -ENOTTY;
1677         }
1678 }
1679
1680 static void cciss_check_queues(ctlr_info_t *h)
1681 {
1682         int start_queue = h->next_to_run;
1683         int i;
1684
1685         /* check to see if we have maxed out the number of commands that can
1686          * be placed on the queue.  If so then exit.  We do this check here
1687          * in case the interrupt we serviced was from an ioctl and did not
1688          * free any new commands.
1689          */
1690         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1691                 return;
1692
1693         /* We have room on the queue for more commands.  Now we need to queue
1694          * them up.  We will also keep track of the next queue to run so
1695          * that every queue gets a chance to be started first.
1696          */
1697         for (i = 0; i < h->highest_lun + 1; i++) {
1698                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1699                 /* make sure the disk has been added and the drive is real
1700                  * because this can be called from the middle of init_one.
1701                  */
1702                 if (!h->drv[curr_queue])
1703                         continue;
1704                 if (!(h->drv[curr_queue]->queue) ||
1705                         !(h->drv[curr_queue]->heads))
1706                         continue;
1707                 blk_start_queue(h->gendisk[curr_queue]->queue);
1708
1709                 /* check to see if we have maxed out the number of commands
1710                  * that can be placed on the queue.
1711                  */
1712                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1713                         if (curr_queue == start_queue) {
1714                                 h->next_to_run =
1715                                     (start_queue + 1) % (h->highest_lun + 1);
1716                                 break;
1717                         } else {
1718                                 h->next_to_run = curr_queue;
1719                                 break;
1720                         }
1721                 }
1722         }
1723 }
1724
1725 static void cciss_softirq_done(struct request *rq)
1726 {
1727         CommandList_struct *c = rq->completion_data;
1728         ctlr_info_t *h = hba[c->ctlr];
1729         SGDescriptor_struct *curr_sg = c->SG;
1730         u64bit temp64;
1731         unsigned long flags;
1732         int i, ddir;
1733         int sg_index = 0;
1734
1735         if (c->Request.Type.Direction == XFER_READ)
1736                 ddir = PCI_DMA_FROMDEVICE;
1737         else
1738                 ddir = PCI_DMA_TODEVICE;
1739
1740         /* command did not need to be retried */
1741         /* unmap the DMA mapping for all the scatter gather elements */
1742         for (i = 0; i < c->Header.SGList; i++) {
1743                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1744                         cciss_unmap_sg_chain_block(h, c);
1745                         /* Point to the next block */
1746                         curr_sg = h->cmd_sg_list[c->cmdindex];
1747                         sg_index = 0;
1748                 }
1749                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1750                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1751                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1752                                 ddir);
1753                 ++sg_index;
1754         }
1755
1756         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1757
1758         /* set the residual count for pc requests */
1759         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1760                 rq->resid_len = c->err_info->ResidualCnt;
1761
1762         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1763
1764         spin_lock_irqsave(&h->lock, flags);
1765         cmd_free(h, c);
1766         cciss_check_queues(h);
1767         spin_unlock_irqrestore(&h->lock, flags);
1768 }
1769
1770 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1771         unsigned char scsi3addr[], uint32_t log_unit)
1772 {
1773         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1774                 sizeof(h->drv[log_unit]->LunID));
1775 }
1776
1777 /* This function gets the SCSI vendor, model, and revision of a logical drive
1778  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1779  * they cannot be read.
1780  */
1781 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1782                                    char *vendor, char *model, char *rev)
1783 {
1784         int rc;
1785         InquiryData_struct *inq_buf;
1786         unsigned char scsi3addr[8];
1787
1788         *vendor = '\0';
1789         *model = '\0';
1790         *rev = '\0';
1791
1792         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1793         if (!inq_buf)
1794                 return;
1795
1796         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1797         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1798                         scsi3addr, TYPE_CMD);
1799         if (rc == IO_OK) {
1800                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1801                 vendor[VENDOR_LEN] = '\0';
1802                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1803                 model[MODEL_LEN] = '\0';
1804                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1805                 rev[REV_LEN] = '\0';
1806         }
1807
1808         kfree(inq_buf);
1809         return;
1810 }
1811
1812 /* This function gets the serial number of a logical drive via
1813  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1814  * number cannot be had, for whatever reason, 16 bytes of 0xff
1815  * are returned instead.
1816  */
1817 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1818                                 unsigned char *serial_no, int buflen)
1819 {
1820 #define PAGE_83_INQ_BYTES 64
1821         int rc;
1822         unsigned char *buf;
1823         unsigned char scsi3addr[8];
1824
1825         if (buflen > 16)
1826                 buflen = 16;
1827         memset(serial_no, 0xff, buflen);
1828         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1829         if (!buf)
1830                 return;
1831         memset(serial_no, 0, buflen);
1832         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1833         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1834                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1835         if (rc == IO_OK)
1836                 memcpy(serial_no, &buf[8], buflen);
1837         kfree(buf);
1838         return;
1839 }
1840
1841 /*
1842  * cciss_add_disk sets up the block device queue for a logical drive
1843  */
1844 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1845                                 int drv_index)
1846 {
1847         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1848         if (!disk->queue)
1849                 goto init_queue_failure;
1850         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1851         disk->major = h->major;
1852         disk->first_minor = drv_index << NWD_SHIFT;
1853         disk->fops = &cciss_fops;
1854         if (cciss_create_ld_sysfs_entry(h, drv_index))
1855                 goto cleanup_queue;
1856         disk->private_data = h->drv[drv_index];
1857         disk->driverfs_dev = &h->drv[drv_index]->dev;
1858
1859         /* Set up queue information */
1860         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1861
1862         /* This is a hardware imposed limit. */
1863         blk_queue_max_segments(disk->queue, h->maxsgentries);
1864
1865         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1866
1867         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1868
1869         disk->queue->queuedata = h;
1870
1871         blk_queue_logical_block_size(disk->queue,
1872                                      h->drv[drv_index]->block_size);
1873
1874         /* Make sure all queue data is written out before */
1875         /* setting h->drv[drv_index]->queue, as setting this */
1876         /* allows the interrupt handler to start the queue */
1877         wmb();
1878         h->drv[drv_index]->queue = disk->queue;
1879         add_disk(disk);
1880         return 0;
1881
1882 cleanup_queue:
1883         blk_cleanup_queue(disk->queue);
1884         disk->queue = NULL;
1885 init_queue_failure:
1886         return -1;
1887 }
1888
1889 /* This function will check the usage_count of the drive to be updated/added.
1890  * If the usage_count is zero and it is a heretofore unknown drive, or,
1891  * the drive's capacity, geometry, or serial number has changed,
1892  * then the drive information will be updated and the disk will be
1893  * re-registered with the kernel.  If these conditions don't hold,
1894  * then it will be left alone for the next reboot.  The exception to this
1895  * is disk 0 which will always be left registered with the kernel since it
1896  * is also the controller node.  Any changes to disk 0 will show up on
1897  * the next reboot.
1898  */
1899 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1900         int first_time, int via_ioctl)
1901 {
1902         struct gendisk *disk;
1903         InquiryData_struct *inq_buff = NULL;
1904         unsigned int block_size;
1905         sector_t total_size;
1906         unsigned long flags = 0;
1907         int ret = 0;
1908         drive_info_struct *drvinfo;
1909
1910         /* Get information about the disk and modify the driver structure */
1911         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1912         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1913         if (inq_buff == NULL || drvinfo == NULL)
1914                 goto mem_msg;
1915
1916         /* testing to see if 16-byte CDBs are already being used */
1917         if (h->cciss_read == CCISS_READ_16) {
1918                 cciss_read_capacity_16(h, drv_index,
1919                         &total_size, &block_size);
1920
1921         } else {
1922                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1923                 /* if read_capacity returns all F's this volume is >2TB */
1924                 /* in size so we switch to 16-byte CDB's for all */
1925                 /* read/write ops */
1926                 if (total_size == 0xFFFFFFFFULL) {
1927                         cciss_read_capacity_16(h, drv_index,
1928                         &total_size, &block_size);
1929                         h->cciss_read = CCISS_READ_16;
1930                         h->cciss_write = CCISS_WRITE_16;
1931                 } else {
1932                         h->cciss_read = CCISS_READ_10;
1933                         h->cciss_write = CCISS_WRITE_10;
1934                 }
1935         }
1936
1937         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1938                                inq_buff, drvinfo);
1939         drvinfo->block_size = block_size;
1940         drvinfo->nr_blocks = total_size + 1;
1941
1942         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1943                                 drvinfo->model, drvinfo->rev);
1944         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1945                         sizeof(drvinfo->serial_no));
1946         /* Save the lunid in case we deregister the disk, below. */
1947         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1948                 sizeof(drvinfo->LunID));
1949
1950         /* Is it the same disk we already know, and nothing's changed? */
1951         if (h->drv[drv_index]->raid_level != -1 &&
1952                 ((memcmp(drvinfo->serial_no,
1953                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1954                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1955                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1956                 drvinfo->heads == h->drv[drv_index]->heads &&
1957                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1958                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1959                         /* The disk is unchanged, nothing to update */
1960                         goto freeret;
1961
1962         /* If we get here it's not the same disk, or something's changed,
1963          * so we need to * deregister it, and re-register it, if it's not
1964          * in use.
1965          * If the disk already exists then deregister it before proceeding
1966          * (unless it's the first disk (for the controller node).
1967          */
1968         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1969                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
1970                 spin_lock_irqsave(&h->lock, flags);
1971                 h->drv[drv_index]->busy_configuring = 1;
1972                 spin_unlock_irqrestore(&h->lock, flags);
1973
1974                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1975                  * which keeps the interrupt handler from starting
1976                  * the queue.
1977                  */
1978                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1979         }
1980
1981         /* If the disk is in use return */
1982         if (ret)
1983                 goto freeret;
1984
1985         /* Save the new information from cciss_geometry_inquiry
1986          * and serial number inquiry.  If the disk was deregistered
1987          * above, then h->drv[drv_index] will be NULL.
1988          */
1989         if (h->drv[drv_index] == NULL) {
1990                 drvinfo->device_initialized = 0;
1991                 h->drv[drv_index] = drvinfo;
1992                 drvinfo = NULL; /* so it won't be freed below. */
1993         } else {
1994                 /* special case for cxd0 */
1995                 h->drv[drv_index]->block_size = drvinfo->block_size;
1996                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
1997                 h->drv[drv_index]->heads = drvinfo->heads;
1998                 h->drv[drv_index]->sectors = drvinfo->sectors;
1999                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2000                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2001                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2002                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2003                         VENDOR_LEN + 1);
2004                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2005                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2006         }
2007
2008         ++h->num_luns;
2009         disk = h->gendisk[drv_index];
2010         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2011
2012         /* If it's not disk 0 (drv_index != 0)
2013          * or if it was disk 0, but there was previously
2014          * no actual corresponding configured logical drive
2015          * (raid_leve == -1) then we want to update the
2016          * logical drive's information.
2017          */
2018         if (drv_index || first_time) {
2019                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2020                         cciss_free_gendisk(h, drv_index);
2021                         cciss_free_drive_info(h, drv_index);
2022                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2023                                 drv_index);
2024                         --h->num_luns;
2025                 }
2026         }
2027
2028 freeret:
2029         kfree(inq_buff);
2030         kfree(drvinfo);
2031         return;
2032 mem_msg:
2033         dev_err(&h->pdev->dev, "out of memory\n");
2034         goto freeret;
2035 }
2036
2037 /* This function will find the first index of the controllers drive array
2038  * that has a null drv pointer and allocate the drive info struct and
2039  * will return that index   This is where new drives will be added.
2040  * If the index to be returned is greater than the highest_lun index for
2041  * the controller then highest_lun is set * to this new index.
2042  * If there are no available indexes or if tha allocation fails, then -1
2043  * is returned.  * "controller_node" is used to know if this is a real
2044  * logical drive, or just the controller node, which determines if this
2045  * counts towards highest_lun.
2046  */
2047 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2048 {
2049         int i;
2050         drive_info_struct *drv;
2051
2052         /* Search for an empty slot for our drive info */
2053         for (i = 0; i < CISS_MAX_LUN; i++) {
2054
2055                 /* if not cxd0 case, and it's occupied, skip it. */
2056                 if (h->drv[i] && i != 0)
2057                         continue;
2058                 /*
2059                  * If it's cxd0 case, and drv is alloc'ed already, and a
2060                  * disk is configured there, skip it.
2061                  */
2062                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2063                         continue;
2064
2065                 /*
2066                  * We've found an empty slot.  Update highest_lun
2067                  * provided this isn't just the fake cxd0 controller node.
2068                  */
2069                 if (i > h->highest_lun && !controller_node)
2070                         h->highest_lun = i;
2071
2072                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2073                 if (i == 0 && h->drv[i] != NULL)
2074                         return i;
2075
2076                 /*
2077                  * Found an empty slot, not already alloc'ed.  Allocate it.
2078                  * Mark it with raid_level == -1, so we know it's new later on.
2079                  */
2080                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2081                 if (!drv)
2082                         return -1;
2083                 drv->raid_level = -1; /* so we know it's new */
2084                 h->drv[i] = drv;
2085                 return i;
2086         }
2087         return -1;
2088 }
2089
2090 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2091 {
2092         kfree(h->drv[drv_index]);
2093         h->drv[drv_index] = NULL;
2094 }
2095
2096 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2097 {
2098         put_disk(h->gendisk[drv_index]);
2099         h->gendisk[drv_index] = NULL;
2100 }
2101
2102 /* cciss_add_gendisk finds a free hba[]->drv structure
2103  * and allocates a gendisk if needed, and sets the lunid
2104  * in the drvinfo structure.   It returns the index into
2105  * the ->drv[] array, or -1 if none are free.
2106  * is_controller_node indicates whether highest_lun should
2107  * count this disk, or if it's only being added to provide
2108  * a means to talk to the controller in case no logical
2109  * drives have yet been configured.
2110  */
2111 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2112         int controller_node)
2113 {
2114         int drv_index;
2115
2116         drv_index = cciss_alloc_drive_info(h, controller_node);
2117         if (drv_index == -1)
2118                 return -1;
2119
2120         /*Check if the gendisk needs to be allocated */
2121         if (!h->gendisk[drv_index]) {
2122                 h->gendisk[drv_index] =
2123                         alloc_disk(1 << NWD_SHIFT);
2124                 if (!h->gendisk[drv_index]) {
2125                         dev_err(&h->pdev->dev,
2126                                 "could not allocate a new disk %d\n",
2127                                 drv_index);
2128                         goto err_free_drive_info;
2129                 }
2130         }
2131         memcpy(h->drv[drv_index]->LunID, lunid,
2132                 sizeof(h->drv[drv_index]->LunID));
2133         if (cciss_create_ld_sysfs_entry(h, drv_index))
2134                 goto err_free_disk;
2135         /* Don't need to mark this busy because nobody */
2136         /* else knows about this disk yet to contend */
2137         /* for access to it. */
2138         h->drv[drv_index]->busy_configuring = 0;
2139         wmb();
2140         return drv_index;
2141
2142 err_free_disk:
2143         cciss_free_gendisk(h, drv_index);
2144 err_free_drive_info:
2145         cciss_free_drive_info(h, drv_index);
2146         return -1;
2147 }
2148
2149 /* This is for the special case of a controller which
2150  * has no logical drives.  In this case, we still need
2151  * to register a disk so the controller can be accessed
2152  * by the Array Config Utility.
2153  */
2154 static void cciss_add_controller_node(ctlr_info_t *h)
2155 {
2156         struct gendisk *disk;
2157         int drv_index;
2158
2159         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2160                 return;
2161
2162         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2163         if (drv_index == -1)
2164                 goto error;
2165         h->drv[drv_index]->block_size = 512;
2166         h->drv[drv_index]->nr_blocks = 0;
2167         h->drv[drv_index]->heads = 0;
2168         h->drv[drv_index]->sectors = 0;
2169         h->drv[drv_index]->cylinders = 0;
2170         h->drv[drv_index]->raid_level = -1;
2171         memset(h->drv[drv_index]->serial_no, 0, 16);
2172         disk = h->gendisk[drv_index];
2173         if (cciss_add_disk(h, disk, drv_index) == 0)
2174                 return;
2175         cciss_free_gendisk(h, drv_index);
2176         cciss_free_drive_info(h, drv_index);
2177 error:
2178         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2179         return;
2180 }
2181
2182 /* This function will add and remove logical drives from the Logical
2183  * drive array of the controller and maintain persistency of ordering
2184  * so that mount points are preserved until the next reboot.  This allows
2185  * for the removal of logical drives in the middle of the drive array
2186  * without a re-ordering of those drives.
2187  * INPUT
2188  * h            = The controller to perform the operations on
2189  */
2190 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2191         int via_ioctl)
2192 {
2193         int num_luns;
2194         ReportLunData_struct *ld_buff = NULL;
2195         int return_code;
2196         int listlength = 0;
2197         int i;
2198         int drv_found;
2199         int drv_index = 0;
2200         unsigned char lunid[8] = CTLR_LUNID;
2201         unsigned long flags;
2202
2203         if (!capable(CAP_SYS_RAWIO))
2204                 return -EPERM;
2205
2206         /* Set busy_configuring flag for this operation */
2207         spin_lock_irqsave(&h->lock, flags);
2208         if (h->busy_configuring) {
2209                 spin_unlock_irqrestore(&h->lock, flags);
2210                 return -EBUSY;
2211         }
2212         h->busy_configuring = 1;
2213         spin_unlock_irqrestore(&h->lock, flags);
2214
2215         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2216         if (ld_buff == NULL)
2217                 goto mem_msg;
2218
2219         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2220                                       sizeof(ReportLunData_struct),
2221                                       0, CTLR_LUNID, TYPE_CMD);
2222
2223         if (return_code == IO_OK)
2224                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2225         else {  /* reading number of logical volumes failed */
2226                 dev_warn(&h->pdev->dev,
2227                         "report logical volume command failed\n");
2228                 listlength = 0;
2229                 goto freeret;
2230         }
2231
2232         num_luns = listlength / 8;      /* 8 bytes per entry */
2233         if (num_luns > CISS_MAX_LUN) {
2234                 num_luns = CISS_MAX_LUN;
2235                 dev_warn(&h->pdev->dev, "more luns configured"
2236                        " on controller than can be handled by"
2237                        " this driver.\n");
2238         }
2239
2240         if (num_luns == 0)
2241                 cciss_add_controller_node(h);
2242
2243         /* Compare controller drive array to driver's drive array
2244          * to see if any drives are missing on the controller due
2245          * to action of Array Config Utility (user deletes drive)
2246          * and deregister logical drives which have disappeared.
2247          */
2248         for (i = 0; i <= h->highest_lun; i++) {
2249                 int j;
2250                 drv_found = 0;
2251
2252                 /* skip holes in the array from already deleted drives */
2253                 if (h->drv[i] == NULL)
2254                         continue;
2255
2256                 for (j = 0; j < num_luns; j++) {
2257                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2258                         if (memcmp(h->drv[i]->LunID, lunid,
2259                                 sizeof(lunid)) == 0) {
2260                                 drv_found = 1;
2261                                 break;
2262                         }
2263                 }
2264                 if (!drv_found) {
2265                         /* Deregister it from the OS, it's gone. */
2266                         spin_lock_irqsave(&h->lock, flags);
2267                         h->drv[i]->busy_configuring = 1;
2268                         spin_unlock_irqrestore(&h->lock, flags);
2269                         return_code = deregister_disk(h, i, 1, via_ioctl);
2270                         if (h->drv[i] != NULL)
2271                                 h->drv[i]->busy_configuring = 0;
2272                 }
2273         }
2274
2275         /* Compare controller drive array to driver's drive array.
2276          * Check for updates in the drive information and any new drives
2277          * on the controller due to ACU adding logical drives, or changing
2278          * a logical drive's size, etc.  Reregister any new/changed drives
2279          */
2280         for (i = 0; i < num_luns; i++) {
2281                 int j;
2282
2283                 drv_found = 0;
2284
2285                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2286                 /* Find if the LUN is already in the drive array
2287                  * of the driver.  If so then update its info
2288                  * if not in use.  If it does not exist then find
2289                  * the first free index and add it.
2290                  */
2291                 for (j = 0; j <= h->highest_lun; j++) {
2292                         if (h->drv[j] != NULL &&
2293                                 memcmp(h->drv[j]->LunID, lunid,
2294                                         sizeof(h->drv[j]->LunID)) == 0) {
2295                                 drv_index = j;
2296                                 drv_found = 1;
2297                                 break;
2298                         }
2299                 }
2300
2301                 /* check if the drive was found already in the array */
2302                 if (!drv_found) {
2303                         drv_index = cciss_add_gendisk(h, lunid, 0);
2304                         if (drv_index == -1)
2305                                 goto freeret;
2306                 }
2307                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2308         }               /* end for */
2309
2310 freeret:
2311         kfree(ld_buff);
2312         h->busy_configuring = 0;
2313         /* We return -1 here to tell the ACU that we have registered/updated
2314          * all of the drives that we can and to keep it from calling us
2315          * additional times.
2316          */
2317         return -1;
2318 mem_msg:
2319         dev_err(&h->pdev->dev, "out of memory\n");
2320         h->busy_configuring = 0;
2321         goto freeret;
2322 }
2323
2324 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2325 {
2326         /* zero out the disk size info */
2327         drive_info->nr_blocks = 0;
2328         drive_info->block_size = 0;
2329         drive_info->heads = 0;
2330         drive_info->sectors = 0;
2331         drive_info->cylinders = 0;
2332         drive_info->raid_level = -1;
2333         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2334         memset(drive_info->model, 0, sizeof(drive_info->model));
2335         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2336         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2337         /*
2338          * don't clear the LUNID though, we need to remember which
2339          * one this one is.
2340          */
2341 }
2342
2343 /* This function will deregister the disk and it's queue from the
2344  * kernel.  It must be called with the controller lock held and the
2345  * drv structures busy_configuring flag set.  It's parameters are:
2346  *
2347  * disk = This is the disk to be deregistered
2348  * drv  = This is the drive_info_struct associated with the disk to be
2349  *        deregistered.  It contains information about the disk used
2350  *        by the driver.
2351  * clear_all = This flag determines whether or not the disk information
2352  *             is going to be completely cleared out and the highest_lun
2353  *             reset.  Sometimes we want to clear out information about
2354  *             the disk in preparation for re-adding it.  In this case
2355  *             the highest_lun should be left unchanged and the LunID
2356  *             should not be cleared.
2357  * via_ioctl
2358  *    This indicates whether we've reached this path via ioctl.
2359  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2360  *    If this path is reached via ioctl(), then the max_usage_count will
2361  *    be 1, as the process calling ioctl() has got to have the device open.
2362  *    If we get here via sysfs, then the max usage count will be zero.
2363 */
2364 static int deregister_disk(ctlr_info_t *h, int drv_index,
2365                            int clear_all, int via_ioctl)
2366 {
2367         int i;
2368         struct gendisk *disk;
2369         drive_info_struct *drv;
2370         int recalculate_highest_lun;
2371
2372         if (!capable(CAP_SYS_RAWIO))
2373                 return -EPERM;
2374
2375         drv = h->drv[drv_index];
2376         disk = h->gendisk[drv_index];
2377
2378         /* make sure logical volume is NOT is use */
2379         if (clear_all || (h->gendisk[0] == disk)) {
2380                 if (drv->usage_count > via_ioctl)
2381                         return -EBUSY;
2382         } else if (drv->usage_count > 0)
2383                 return -EBUSY;
2384
2385         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2386
2387         /* invalidate the devices and deregister the disk.  If it is disk
2388          * zero do not deregister it but just zero out it's values.  This
2389          * allows us to delete disk zero but keep the controller registered.
2390          */
2391         if (h->gendisk[0] != disk) {
2392                 struct request_queue *q = disk->queue;
2393                 if (disk->flags & GENHD_FL_UP) {
2394                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2395                         del_gendisk(disk);
2396                 }
2397                 if (q)
2398                         blk_cleanup_queue(q);
2399                 /* If clear_all is set then we are deleting the logical
2400                  * drive, not just refreshing its info.  For drives
2401                  * other than disk 0 we will call put_disk.  We do not
2402                  * do this for disk 0 as we need it to be able to
2403                  * configure the controller.
2404                  */
2405                 if (clear_all){
2406                         /* This isn't pretty, but we need to find the
2407                          * disk in our array and NULL our the pointer.
2408                          * This is so that we will call alloc_disk if
2409                          * this index is used again later.
2410                          */
2411                         for (i=0; i < CISS_MAX_LUN; i++){
2412                                 if (h->gendisk[i] == disk) {
2413                                         h->gendisk[i] = NULL;
2414                                         break;
2415                                 }
2416                         }
2417                         put_disk(disk);
2418                 }
2419         } else {
2420                 set_capacity(disk, 0);
2421                 cciss_clear_drive_info(drv);
2422         }
2423
2424         --h->num_luns;
2425
2426         /* if it was the last disk, find the new hightest lun */
2427         if (clear_all && recalculate_highest_lun) {
2428                 int newhighest = -1;
2429                 for (i = 0; i <= h->highest_lun; i++) {
2430                         /* if the disk has size > 0, it is available */
2431                         if (h->drv[i] && h->drv[i]->heads)
2432                                 newhighest = i;
2433                 }
2434                 h->highest_lun = newhighest;
2435         }
2436         return 0;
2437 }
2438
2439 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2440                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2441                 int cmd_type)
2442 {
2443         u64bit buff_dma_handle;
2444         int status = IO_OK;
2445
2446         c->cmd_type = CMD_IOCTL_PEND;
2447         c->Header.ReplyQueue = 0;
2448         if (buff != NULL) {
2449                 c->Header.SGList = 1;
2450                 c->Header.SGTotal = 1;
2451         } else {
2452                 c->Header.SGList = 0;
2453                 c->Header.SGTotal = 0;
2454         }
2455         c->Header.Tag.lower = c->busaddr;
2456         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2457
2458         c->Request.Type.Type = cmd_type;
2459         if (cmd_type == TYPE_CMD) {
2460                 switch (cmd) {
2461                 case CISS_INQUIRY:
2462                         /* are we trying to read a vital product page */
2463                         if (page_code != 0) {
2464                                 c->Request.CDB[1] = 0x01;
2465                                 c->Request.CDB[2] = page_code;
2466                         }
2467                         c->Request.CDBLen = 6;
2468                         c->Request.Type.Attribute = ATTR_SIMPLE;
2469                         c->Request.Type.Direction = XFER_READ;
2470                         c->Request.Timeout = 0;
2471                         c->Request.CDB[0] = CISS_INQUIRY;
2472                         c->Request.CDB[4] = size & 0xFF;
2473                         break;
2474                 case CISS_REPORT_LOG:
2475                 case CISS_REPORT_PHYS:
2476                         /* Talking to controller so It's a physical command
2477                            mode = 00 target = 0.  Nothing to write.
2478                          */
2479                         c->Request.CDBLen = 12;
2480                         c->Request.Type.Attribute = ATTR_SIMPLE;
2481                         c->Request.Type.Direction = XFER_READ;
2482                         c->Request.Timeout = 0;
2483                         c->Request.CDB[0] = cmd;
2484                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2485                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2486                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2487                         c->Request.CDB[9] = size & 0xFF;
2488                         break;
2489
2490                 case CCISS_READ_CAPACITY:
2491                         c->Request.CDBLen = 10;
2492                         c->Request.Type.Attribute = ATTR_SIMPLE;
2493                         c->Request.Type.Direction = XFER_READ;
2494                         c->Request.Timeout = 0;
2495                         c->Request.CDB[0] = cmd;
2496                         break;
2497                 case CCISS_READ_CAPACITY_16:
2498                         c->Request.CDBLen = 16;
2499                         c->Request.Type.Attribute = ATTR_SIMPLE;
2500                         c->Request.Type.Direction = XFER_READ;
2501                         c->Request.Timeout = 0;
2502                         c->Request.CDB[0] = cmd;
2503                         c->Request.CDB[1] = 0x10;
2504                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2505                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2506                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2507                         c->Request.CDB[13] = size & 0xFF;
2508                         c->Request.Timeout = 0;
2509                         c->Request.CDB[0] = cmd;
2510                         break;
2511                 case CCISS_CACHE_FLUSH:
2512                         c->Request.CDBLen = 12;
2513                         c->Request.Type.Attribute = ATTR_SIMPLE;
2514                         c->Request.Type.Direction = XFER_WRITE;
2515                         c->Request.Timeout = 0;
2516                         c->Request.CDB[0] = BMIC_WRITE;
2517                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2518                         break;
2519                 case TEST_UNIT_READY:
2520                         c->Request.CDBLen = 6;
2521                         c->Request.Type.Attribute = ATTR_SIMPLE;
2522                         c->Request.Type.Direction = XFER_NONE;
2523                         c->Request.Timeout = 0;
2524                         break;
2525                 default:
2526                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2527                         return IO_ERROR;
2528                 }
2529         } else if (cmd_type == TYPE_MSG) {
2530                 switch (cmd) {
2531                 case 0: /* ABORT message */
2532                         c->Request.CDBLen = 12;
2533                         c->Request.Type.Attribute = ATTR_SIMPLE;
2534                         c->Request.Type.Direction = XFER_WRITE;
2535                         c->Request.Timeout = 0;
2536                         c->Request.CDB[0] = cmd;        /* abort */
2537                         c->Request.CDB[1] = 0;  /* abort a command */
2538                         /* buff contains the tag of the command to abort */
2539                         memcpy(&c->Request.CDB[4], buff, 8);
2540                         break;
2541                 case 1: /* RESET message */
2542                         c->Request.CDBLen = 16;
2543                         c->Request.Type.Attribute = ATTR_SIMPLE;
2544                         c->Request.Type.Direction = XFER_NONE;
2545                         c->Request.Timeout = 0;
2546                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2547                         c->Request.CDB[0] = cmd;        /* reset */
2548                         c->Request.CDB[1] = 0x03;       /* reset a target */
2549                         break;
2550                 case 3: /* No-Op message */
2551                         c->Request.CDBLen = 1;
2552                         c->Request.Type.Attribute = ATTR_SIMPLE;
2553                         c->Request.Type.Direction = XFER_WRITE;
2554                         c->Request.Timeout = 0;
2555                         c->Request.CDB[0] = cmd;
2556                         break;
2557                 default:
2558                         dev_warn(&h->pdev->dev,
2559                                 "unknown message type %d\n", cmd);
2560                         return IO_ERROR;
2561                 }
2562         } else {
2563                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2564                 return IO_ERROR;
2565         }
2566         /* Fill in the scatter gather information */
2567         if (size > 0) {
2568                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2569                                                              buff, size,
2570                                                              PCI_DMA_BIDIRECTIONAL);
2571                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2572                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2573                 c->SG[0].Len = size;
2574                 c->SG[0].Ext = 0;       /* we are not chaining */
2575         }
2576         return status;
2577 }
2578
2579 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2580 {
2581         switch (c->err_info->ScsiStatus) {
2582         case SAM_STAT_GOOD:
2583                 return IO_OK;
2584         case SAM_STAT_CHECK_CONDITION:
2585                 switch (0xf & c->err_info->SenseInfo[2]) {
2586                 case 0: return IO_OK; /* no sense */
2587                 case 1: return IO_OK; /* recovered error */
2588                 default:
2589                         if (check_for_unit_attention(h, c))
2590                                 return IO_NEEDS_RETRY;
2591                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2592                                 "check condition, sense key = 0x%02x\n",
2593                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2594                 }
2595                 break;
2596         default:
2597                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2598                         "scsi status = 0x%02x\n",
2599                         c->Request.CDB[0], c->err_info->ScsiStatus);
2600                 break;
2601         }
2602         return IO_ERROR;
2603 }
2604
2605 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2606 {
2607         int return_status = IO_OK;
2608
2609         if (c->err_info->CommandStatus == CMD_SUCCESS)
2610                 return IO_OK;
2611
2612         switch (c->err_info->CommandStatus) {
2613         case CMD_TARGET_STATUS:
2614                 return_status = check_target_status(h, c);
2615                 break;
2616         case CMD_DATA_UNDERRUN:
2617         case CMD_DATA_OVERRUN:
2618                 /* expected for inquiry and report lun commands */
2619                 break;
2620         case CMD_INVALID:
2621                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2622                        "reported invalid\n", c->Request.CDB[0]);
2623                 return_status = IO_ERROR;
2624                 break;
2625         case CMD_PROTOCOL_ERR:
2626                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2627                        "protocol error\n", c->Request.CDB[0]);
2628                 return_status = IO_ERROR;
2629                 break;
2630         case CMD_HARDWARE_ERR:
2631                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2632                        " hardware error\n", c->Request.CDB[0]);
2633                 return_status = IO_ERROR;
2634                 break;
2635         case CMD_CONNECTION_LOST:
2636                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2637                        "connection lost\n", c->Request.CDB[0]);
2638                 return_status = IO_ERROR;
2639                 break;
2640         case CMD_ABORTED:
2641                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2642                        "aborted\n", c->Request.CDB[0]);
2643                 return_status = IO_ERROR;
2644                 break;
2645         case CMD_ABORT_FAILED:
2646                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2647                        "abort failed\n", c->Request.CDB[0]);
2648                 return_status = IO_ERROR;
2649                 break;
2650         case CMD_UNSOLICITED_ABORT:
2651                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2652                         c->Request.CDB[0]);
2653                 return_status = IO_NEEDS_RETRY;
2654                 break;
2655         default:
2656                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2657                        "unknown status %x\n", c->Request.CDB[0],
2658                        c->err_info->CommandStatus);
2659                 return_status = IO_ERROR;
2660         }
2661         return return_status;
2662 }
2663
2664 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2665         int attempt_retry)
2666 {
2667         DECLARE_COMPLETION_ONSTACK(wait);
2668         u64bit buff_dma_handle;
2669         int return_status = IO_OK;
2670
2671 resend_cmd2:
2672         c->waiting = &wait;
2673         enqueue_cmd_and_start_io(h, c);
2674
2675         wait_for_completion(&wait);
2676
2677         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2678                 goto command_done;
2679
2680         return_status = process_sendcmd_error(h, c);
2681
2682         if (return_status == IO_NEEDS_RETRY &&
2683                 c->retry_count < MAX_CMD_RETRIES) {
2684                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2685                         c->Request.CDB[0]);
2686                 c->retry_count++;
2687                 /* erase the old error information */
2688                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2689                 return_status = IO_OK;
2690                 INIT_COMPLETION(wait);
2691                 goto resend_cmd2;
2692         }
2693
2694 command_done:
2695         /* unlock the buffers from DMA */
2696         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2697         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2698         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2699                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2700         return return_status;
2701 }
2702
2703 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2704                            __u8 page_code, unsigned char scsi3addr[],
2705                         int cmd_type)
2706 {
2707         CommandList_struct *c;
2708         int return_status;
2709
2710         c = cmd_special_alloc(h);
2711         if (!c)
2712                 return -ENOMEM;
2713         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2714                 scsi3addr, cmd_type);
2715         if (return_status == IO_OK)
2716                 return_status = sendcmd_withirq_core(h, c, 1);
2717
2718         cmd_special_free(h, c);
2719         return return_status;
2720 }
2721
2722 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2723                                    sector_t total_size,
2724                                    unsigned int block_size,
2725                                    InquiryData_struct *inq_buff,
2726                                    drive_info_struct *drv)
2727 {
2728         int return_code;
2729         unsigned long t;
2730         unsigned char scsi3addr[8];
2731
2732         memset(inq_buff, 0, sizeof(InquiryData_struct));
2733         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2734         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2735                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2736         if (return_code == IO_OK) {
2737                 if (inq_buff->data_byte[8] == 0xFF) {
2738                         dev_warn(&h->pdev->dev,
2739                                "reading geometry failed, volume "
2740                                "does not support reading geometry\n");
2741                         drv->heads = 255;
2742                         drv->sectors = 32;      /* Sectors per track */
2743                         drv->cylinders = total_size + 1;
2744                         drv->raid_level = RAID_UNKNOWN;
2745                 } else {
2746                         drv->heads = inq_buff->data_byte[6];
2747                         drv->sectors = inq_buff->data_byte[7];
2748                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2749                         drv->cylinders += inq_buff->data_byte[5];
2750                         drv->raid_level = inq_buff->data_byte[8];
2751                 }
2752                 drv->block_size = block_size;
2753                 drv->nr_blocks = total_size + 1;
2754                 t = drv->heads * drv->sectors;
2755                 if (t > 1) {
2756                         sector_t real_size = total_size + 1;
2757                         unsigned long rem = sector_div(real_size, t);
2758                         if (rem)
2759                                 real_size++;
2760                         drv->cylinders = real_size;
2761                 }
2762         } else {                /* Get geometry failed */
2763                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2764         }
2765 }
2766
2767 static void
2768 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2769                     unsigned int *block_size)
2770 {
2771         ReadCapdata_struct *buf;
2772         int return_code;
2773         unsigned char scsi3addr[8];
2774
2775         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2776         if (!buf) {
2777                 dev_warn(&h->pdev->dev, "out of memory\n");
2778                 return;
2779         }
2780
2781         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2782         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2783                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2784         if (return_code == IO_OK) {
2785                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2786                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2787         } else {                /* read capacity command failed */
2788                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2789                 *total_size = 0;
2790                 *block_size = BLOCK_SIZE;
2791         }
2792         kfree(buf);
2793 }
2794
2795 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2796         sector_t *total_size, unsigned int *block_size)
2797 {
2798         ReadCapdata_struct_16 *buf;
2799         int return_code;
2800         unsigned char scsi3addr[8];
2801
2802         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2803         if (!buf) {
2804                 dev_warn(&h->pdev->dev, "out of memory\n");
2805                 return;
2806         }
2807
2808         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2809         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2810                 buf, sizeof(ReadCapdata_struct_16),
2811                         0, scsi3addr, TYPE_CMD);
2812         if (return_code == IO_OK) {
2813                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2814                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2815         } else {                /* read capacity command failed */
2816                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2817                 *total_size = 0;
2818                 *block_size = BLOCK_SIZE;
2819         }
2820         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2821                (unsigned long long)*total_size+1, *block_size);
2822         kfree(buf);
2823 }
2824
2825 static int cciss_revalidate(struct gendisk *disk)
2826 {
2827         ctlr_info_t *h = get_host(disk);
2828         drive_info_struct *drv = get_drv(disk);
2829         int logvol;
2830         int FOUND = 0;
2831         unsigned int block_size;
2832         sector_t total_size;
2833         InquiryData_struct *inq_buff = NULL;
2834
2835         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2836                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2837                         sizeof(drv->LunID)) == 0) {
2838                         FOUND = 1;
2839                         break;
2840                 }
2841         }
2842
2843         if (!FOUND)
2844                 return 1;
2845
2846         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2847         if (inq_buff == NULL) {
2848                 dev_warn(&h->pdev->dev, "out of memory\n");
2849                 return 1;
2850         }
2851         if (h->cciss_read == CCISS_READ_10) {
2852                 cciss_read_capacity(h, logvol,
2853                                         &total_size, &block_size);
2854         } else {
2855                 cciss_read_capacity_16(h, logvol,
2856                                         &total_size, &block_size);
2857         }
2858         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2859                                inq_buff, drv);
2860
2861         blk_queue_logical_block_size(drv->queue, drv->block_size);
2862         set_capacity(disk, drv->nr_blocks);
2863
2864         kfree(inq_buff);
2865         return 0;
2866 }
2867
2868 /*
2869  * Map (physical) PCI mem into (virtual) kernel space
2870  */
2871 static void __iomem *remap_pci_mem(ulong base, ulong size)
2872 {
2873         ulong page_base = ((ulong) base) & PAGE_MASK;
2874         ulong page_offs = ((ulong) base) - page_base;
2875         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2876
2877         return page_remapped ? (page_remapped + page_offs) : NULL;
2878 }
2879
2880 /*
2881  * Takes jobs of the Q and sends them to the hardware, then puts it on
2882  * the Q to wait for completion.
2883  */
2884 static void start_io(ctlr_info_t *h)
2885 {
2886         CommandList_struct *c;
2887
2888         while (!hlist_empty(&h->reqQ)) {
2889                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2890                 /* can't do anything if fifo is full */
2891                 if ((h->access.fifo_full(h))) {
2892                         dev_warn(&h->pdev->dev, "fifo full\n");
2893                         break;
2894                 }
2895
2896                 /* Get the first entry from the Request Q */
2897                 removeQ(c);
2898                 h->Qdepth--;
2899
2900                 /* Tell the controller execute command */
2901                 h->access.submit_command(h, c);
2902
2903                 /* Put job onto the completed Q */
2904                 addQ(&h->cmpQ, c);
2905         }
2906 }
2907
2908 /* Assumes that h->lock is held. */
2909 /* Zeros out the error record and then resends the command back */
2910 /* to the controller */
2911 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2912 {
2913         /* erase the old error information */
2914         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2915
2916         /* add it to software queue and then send it to the controller */
2917         addQ(&h->reqQ, c);
2918         h->Qdepth++;
2919         if (h->Qdepth > h->maxQsinceinit)
2920                 h->maxQsinceinit = h->Qdepth;
2921
2922         start_io(h);
2923 }
2924
2925 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2926         unsigned int msg_byte, unsigned int host_byte,
2927         unsigned int driver_byte)
2928 {
2929         /* inverse of macros in scsi.h */
2930         return (scsi_status_byte & 0xff) |
2931                 ((msg_byte & 0xff) << 8) |
2932                 ((host_byte & 0xff) << 16) |
2933                 ((driver_byte & 0xff) << 24);
2934 }
2935
2936 static inline int evaluate_target_status(ctlr_info_t *h,
2937                         CommandList_struct *cmd, int *retry_cmd)
2938 {
2939         unsigned char sense_key;
2940         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2941         int error_value;
2942
2943         *retry_cmd = 0;
2944         /* If we get in here, it means we got "target status", that is, scsi status */
2945         status_byte = cmd->err_info->ScsiStatus;
2946         driver_byte = DRIVER_OK;
2947         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2948
2949         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2950                 host_byte = DID_PASSTHROUGH;
2951         else
2952                 host_byte = DID_OK;
2953
2954         error_value = make_status_bytes(status_byte, msg_byte,
2955                 host_byte, driver_byte);
2956
2957         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2958                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
2959                         dev_warn(&h->pdev->dev, "cmd %p "
2960                                "has SCSI Status 0x%x\n",
2961                                cmd, cmd->err_info->ScsiStatus);
2962                 return error_value;
2963         }
2964
2965         /* check the sense key */
2966         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2967         /* no status or recovered error */
2968         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
2969             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
2970                 error_value = 0;
2971
2972         if (check_for_unit_attention(h, cmd)) {
2973                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
2974                 return 0;
2975         }
2976
2977         /* Not SG_IO or similar? */
2978         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
2979                 if (error_value != 0)
2980                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
2981                                " sense key = 0x%x\n", cmd, sense_key);
2982                 return error_value;
2983         }
2984
2985         /* SG_IO or similar, copy sense data back */
2986         if (cmd->rq->sense) {
2987                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2988                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2989                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2990                         cmd->rq->sense_len);
2991         } else
2992                 cmd->rq->sense_len = 0;
2993
2994         return error_value;
2995 }
2996
2997 /* checks the status of the job and calls complete buffers to mark all
2998  * buffers for the completed job. Note that this function does not need
2999  * to hold the hba/queue lock.
3000  */
3001 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3002                                     int timeout)
3003 {
3004         int retry_cmd = 0;
3005         struct request *rq = cmd->rq;
3006
3007         rq->errors = 0;
3008
3009         if (timeout)
3010                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3011
3012         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3013                 goto after_error_processing;
3014
3015         switch (cmd->err_info->CommandStatus) {
3016         case CMD_TARGET_STATUS:
3017                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3018                 break;
3019         case CMD_DATA_UNDERRUN:
3020                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3021                         dev_warn(&h->pdev->dev, "cmd %p has"
3022                                " completed with data underrun "
3023                                "reported\n", cmd);
3024                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3025                 }
3026                 break;
3027         case CMD_DATA_OVERRUN:
3028                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3029                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3030                                " completed with data overrun "
3031                                "reported\n", cmd);
3032                 break;
3033         case CMD_INVALID:
3034                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3035                        "reported invalid\n", cmd);
3036                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3037                         cmd->err_info->CommandStatus, DRIVER_OK,
3038                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3039                                 DID_PASSTHROUGH : DID_ERROR);
3040                 break;
3041         case CMD_PROTOCOL_ERR:
3042                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3043                        "protocol error\n", cmd);
3044                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3045                         cmd->err_info->CommandStatus, DRIVER_OK,
3046                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3047                                 DID_PASSTHROUGH : DID_ERROR);
3048                 break;
3049         case CMD_HARDWARE_ERR:
3050                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3051                        " hardware error\n", cmd);
3052                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3053                         cmd->err_info->CommandStatus, DRIVER_OK,
3054                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3055                                 DID_PASSTHROUGH : DID_ERROR);
3056                 break;
3057         case CMD_CONNECTION_LOST:
3058                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3059                        "connection lost\n", cmd);
3060                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3061                         cmd->err_info->CommandStatus, DRIVER_OK,
3062                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3063                                 DID_PASSTHROUGH : DID_ERROR);
3064                 break;
3065         case CMD_ABORTED:
3066                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3067                        "aborted\n", cmd);
3068                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3069                         cmd->err_info->CommandStatus, DRIVER_OK,
3070                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3071                                 DID_PASSTHROUGH : DID_ABORT);
3072                 break;
3073         case CMD_ABORT_FAILED:
3074                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3075                        "abort failed\n", cmd);
3076                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3077                         cmd->err_info->CommandStatus, DRIVER_OK,
3078                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3079                                 DID_PASSTHROUGH : DID_ERROR);
3080                 break;
3081         case CMD_UNSOLICITED_ABORT:
3082                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3083                        "abort %p\n", h->ctlr, cmd);
3084                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3085                         retry_cmd = 1;
3086                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3087                         cmd->retry_count++;
3088                 } else
3089                         dev_warn(&h->pdev->dev,
3090                                 "%p retried too many times\n", cmd);
3091                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3092                         cmd->err_info->CommandStatus, DRIVER_OK,
3093                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3094                                 DID_PASSTHROUGH : DID_ABORT);
3095                 break;
3096         case CMD_TIMEOUT:
3097                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3098                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3099                         cmd->err_info->CommandStatus, DRIVER_OK,
3100                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3101                                 DID_PASSTHROUGH : DID_ERROR);
3102                 break;
3103         default:
3104                 dev_warn(&h->pdev->dev, "cmd %p returned "
3105                        "unknown status %x\n", cmd,
3106                        cmd->err_info->CommandStatus);
3107                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3108                         cmd->err_info->CommandStatus, DRIVER_OK,
3109                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3110                                 DID_PASSTHROUGH : DID_ERROR);
3111         }
3112
3113 after_error_processing:
3114
3115         /* We need to return this command */
3116         if (retry_cmd) {
3117                 resend_cciss_cmd(h, cmd);
3118                 return;
3119         }
3120         cmd->rq->completion_data = cmd;
3121         blk_complete_request(cmd->rq);
3122 }
3123
3124 static inline u32 cciss_tag_contains_index(u32 tag)
3125 {
3126 #define DIRECT_LOOKUP_BIT 0x10
3127         return tag & DIRECT_LOOKUP_BIT;
3128 }
3129
3130 static inline u32 cciss_tag_to_index(u32 tag)
3131 {
3132 #define DIRECT_LOOKUP_SHIFT 5
3133         return tag >> DIRECT_LOOKUP_SHIFT;
3134 }
3135
3136 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3137 {
3138 #define CCISS_ERROR_BITS 0x03
3139         return tag & ~CCISS_ERROR_BITS;
3140 }
3141
3142 static inline void cciss_mark_tag_indexed(u32 *tag)
3143 {
3144         *tag |= DIRECT_LOOKUP_BIT;
3145 }
3146
3147 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3148 {
3149         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3150 }
3151
3152 /*
3153  * Get a request and submit it to the controller.
3154  */
3155 static void do_cciss_request(struct request_queue *q)
3156 {
3157         ctlr_info_t *h = q->queuedata;
3158         CommandList_struct *c;
3159         sector_t start_blk;
3160         int seg;
3161         struct request *creq;
3162         u64bit temp64;
3163         struct scatterlist *tmp_sg;
3164         SGDescriptor_struct *curr_sg;
3165         drive_info_struct *drv;
3166         int i, dir;
3167         int sg_index = 0;
3168         int chained = 0;
3169
3170         /* We call start_io here in case there is a command waiting on the
3171          * queue that has not been sent.
3172          */
3173         if (blk_queue_plugged(q))
3174                 goto startio;
3175
3176       queue:
3177         creq = blk_peek_request(q);
3178         if (!creq)
3179                 goto startio;
3180
3181         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3182
3183         c = cmd_alloc(h);
3184         if (!c)
3185                 goto full;
3186
3187         blk_start_request(creq);
3188
3189         tmp_sg = h->scatter_list[c->cmdindex];
3190         spin_unlock_irq(q->queue_lock);
3191
3192         c->cmd_type = CMD_RWREQ;
3193         c->rq = creq;
3194
3195         /* fill in the request */
3196         drv = creq->rq_disk->private_data;
3197         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3198         /* got command from pool, so use the command block index instead */
3199         /* for direct lookups. */
3200         /* The first 2 bits are reserved for controller error reporting. */
3201         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3202         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3203         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3204         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3205         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3206         c->Request.Type.Attribute = ATTR_SIMPLE;
3207         c->Request.Type.Direction =
3208             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3209         c->Request.Timeout = 0; /* Don't time out */
3210         c->Request.CDB[0] =
3211             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3212         start_blk = blk_rq_pos(creq);
3213         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3214                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3215         sg_init_table(tmp_sg, h->maxsgentries);
3216         seg = blk_rq_map_sg(q, creq, tmp_sg);
3217
3218         /* get the DMA records for the setup */
3219         if (c->Request.Type.Direction == XFER_READ)
3220                 dir = PCI_DMA_FROMDEVICE;
3221         else
3222                 dir = PCI_DMA_TODEVICE;
3223
3224         curr_sg = c->SG;
3225         sg_index = 0;
3226         chained = 0;
3227
3228         for (i = 0; i < seg; i++) {
3229                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3230                         !chained && ((seg - i) > 1)) {
3231                         /* Point to next chain block. */
3232                         curr_sg = h->cmd_sg_list[c->cmdindex];
3233                         sg_index = 0;
3234                         chained = 1;
3235                 }
3236                 curr_sg[sg_index].Len = tmp_sg[i].length;
3237                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3238                                                 tmp_sg[i].offset,
3239                                                 tmp_sg[i].length, dir);
3240                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3241                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3242                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3243                 ++sg_index;
3244         }
3245         if (chained)
3246                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3247                         (seg - (h->max_cmd_sgentries - 1)) *
3248                                 sizeof(SGDescriptor_struct));
3249
3250         /* track how many SG entries we are using */
3251         if (seg > h->maxSG)
3252                 h->maxSG = seg;
3253
3254         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3255                         "chained[%d]\n",
3256                         blk_rq_sectors(creq), seg, chained);
3257
3258         c->Header.SGTotal = seg + chained;
3259         if (seg <= h->max_cmd_sgentries)
3260                 c->Header.SGList = c->Header.SGTotal;
3261         else
3262                 c->Header.SGList = h->max_cmd_sgentries;
3263         set_performant_mode(h, c);
3264
3265         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3266                 if(h->cciss_read == CCISS_READ_10) {
3267                         c->Request.CDB[1] = 0;
3268                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3269                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3270                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3271                         c->Request.CDB[5] = start_blk & 0xff;
3272                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3273                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3274                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3275                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3276                 } else {
3277                         u32 upper32 = upper_32_bits(start_blk);
3278
3279                         c->Request.CDBLen = 16;
3280                         c->Request.CDB[1]= 0;
3281                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3282                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3283                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3284                         c->Request.CDB[5]= upper32 & 0xff;
3285                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3286                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3287                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3288                         c->Request.CDB[9]= start_blk & 0xff;
3289                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3290                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3291                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3292                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3293                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3294                 }
3295         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3296                 c->Request.CDBLen = creq->cmd_len;
3297                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3298         } else {
3299                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3300                         creq->cmd_type);
3301                 BUG();
3302         }
3303
3304         spin_lock_irq(q->queue_lock);
3305
3306         addQ(&h->reqQ, c);
3307         h->Qdepth++;
3308         if (h->Qdepth > h->maxQsinceinit)
3309                 h->maxQsinceinit = h->Qdepth;
3310
3311         goto queue;
3312 full:
3313         blk_stop_queue(q);
3314 startio:
3315         /* We will already have the driver lock here so not need
3316          * to lock it.
3317          */
3318         start_io(h);
3319 }
3320
3321 static inline unsigned long get_next_completion(ctlr_info_t *h)
3322 {
3323         return h->access.command_completed(h);
3324 }
3325
3326 static inline int interrupt_pending(ctlr_info_t *h)
3327 {
3328         return h->access.intr_pending(h);
3329 }
3330
3331 static inline long interrupt_not_for_us(ctlr_info_t *h)
3332 {
3333         return ((h->access.intr_pending(h) == 0) ||
3334                 (h->interrupts_enabled == 0));
3335 }
3336
3337 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3338                         u32 raw_tag)
3339 {
3340         if (unlikely(tag_index >= h->nr_cmds)) {
3341                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3342                 return 1;
3343         }
3344         return 0;
3345 }
3346
3347 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3348                                 u32 raw_tag)
3349 {
3350         removeQ(c);
3351         if (likely(c->cmd_type == CMD_RWREQ))
3352                 complete_command(h, c, 0);
3353         else if (c->cmd_type == CMD_IOCTL_PEND)
3354                 complete(c->waiting);
3355 #ifdef CONFIG_CISS_SCSI_TAPE
3356         else if (c->cmd_type == CMD_SCSI)
3357                 complete_scsi_command(c, 0, raw_tag);
3358 #endif
3359 }
3360
3361 static inline u32 next_command(ctlr_info_t *h)
3362 {
3363         u32 a;
3364
3365         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3366                 return h->access.command_completed(h);
3367
3368         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3369                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3370                 (h->reply_pool_head)++;
3371                 h->commands_outstanding--;
3372         } else {
3373                 a = FIFO_EMPTY;
3374         }
3375         /* Check for wraparound */
3376         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3377                 h->reply_pool_head = h->reply_pool;
3378                 h->reply_pool_wraparound ^= 1;
3379         }
3380         return a;
3381 }
3382
3383 /* process completion of an indexed ("direct lookup") command */
3384 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3385 {
3386         u32 tag_index;
3387         CommandList_struct *c;
3388
3389         tag_index = cciss_tag_to_index(raw_tag);
3390         if (bad_tag(h, tag_index, raw_tag))
3391                 return next_command(h);
3392         c = h->cmd_pool + tag_index;
3393         finish_cmd(h, c, raw_tag);
3394         return next_command(h);
3395 }
3396
3397 /* process completion of a non-indexed command */
3398 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3399 {
3400         u32 tag;
3401         CommandList_struct *c = NULL;
3402         struct hlist_node *tmp;
3403         __u32 busaddr_masked, tag_masked;
3404
3405         tag = cciss_tag_discard_error_bits(raw_tag);
3406         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3407                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3408                 tag_masked = cciss_tag_discard_error_bits(tag);
3409                 if (busaddr_masked == tag_masked) {
3410                         finish_cmd(h, c, raw_tag);
3411                         return next_command(h);
3412                 }
3413         }
3414         bad_tag(h, h->nr_cmds + 1, raw_tag);
3415         return next_command(h);
3416 }
3417
3418 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3419 {
3420         ctlr_info_t *h = dev_id;
3421         unsigned long flags;
3422         u32 raw_tag;
3423
3424         if (interrupt_not_for_us(h))
3425                 return IRQ_NONE;
3426         spin_lock_irqsave(&h->lock, flags);
3427         while (interrupt_pending(h)) {
3428                 raw_tag = get_next_completion(h);
3429                 while (raw_tag != FIFO_EMPTY) {
3430                         if (cciss_tag_contains_index(raw_tag))
3431                                 raw_tag = process_indexed_cmd(h, raw_tag);
3432                         else
3433                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3434                 }
3435         }
3436         spin_unlock_irqrestore(&h->lock, flags);
3437         return IRQ_HANDLED;
3438 }
3439
3440 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3441  * check the interrupt pending register because it is not set.
3442  */
3443 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3444 {
3445         ctlr_info_t *h = dev_id;
3446         unsigned long flags;
3447         u32 raw_tag;
3448
3449         spin_lock_irqsave(&h->lock, flags);
3450         raw_tag = get_next_completion(h);
3451         while (raw_tag != FIFO_EMPTY) {
3452                 if (cciss_tag_contains_index(raw_tag))
3453                         raw_tag = process_indexed_cmd(h, raw_tag);
3454                 else
3455                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3456         }
3457         spin_unlock_irqrestore(&h->lock, flags);
3458         return IRQ_HANDLED;
3459 }
3460
3461 /**
3462  * add_to_scan_list() - add controller to rescan queue
3463  * @h:                Pointer to the controller.
3464  *
3465  * Adds the controller to the rescan queue if not already on the queue.
3466  *
3467  * returns 1 if added to the queue, 0 if skipped (could be on the
3468  * queue already, or the controller could be initializing or shutting
3469  * down).
3470  **/
3471 static int add_to_scan_list(struct ctlr_info *h)
3472 {
3473         struct ctlr_info *test_h;
3474         int found = 0;
3475         int ret = 0;
3476
3477         if (h->busy_initializing)
3478                 return 0;
3479
3480         if (!mutex_trylock(&h->busy_shutting_down))
3481                 return 0;
3482
3483         mutex_lock(&scan_mutex);
3484         list_for_each_entry(test_h, &scan_q, scan_list) {
3485                 if (test_h == h) {
3486                         found = 1;
3487                         break;
3488                 }
3489         }
3490         if (!found && !h->busy_scanning) {
3491                 INIT_COMPLETION(h->scan_wait);
3492                 list_add_tail(&h->scan_list, &scan_q);
3493                 ret = 1;
3494         }
3495         mutex_unlock(&scan_mutex);
3496         mutex_unlock(&h->busy_shutting_down);
3497
3498         return ret;
3499 }
3500
3501 /**
3502  * remove_from_scan_list() - remove controller from rescan queue
3503  * @h:                     Pointer to the controller.
3504  *
3505  * Removes the controller from the rescan queue if present. Blocks if
3506  * the controller is currently conducting a rescan.  The controller
3507  * can be in one of three states:
3508  * 1. Doesn't need a scan
3509  * 2. On the scan list, but not scanning yet (we remove it)
3510  * 3. Busy scanning (and not on the list). In this case we want to wait for
3511  *    the scan to complete to make sure the scanning thread for this
3512  *    controller is completely idle.
3513  **/
3514 static void remove_from_scan_list(struct ctlr_info *h)
3515 {
3516         struct ctlr_info *test_h, *tmp_h;
3517
3518         mutex_lock(&scan_mutex);
3519         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3520                 if (test_h == h) { /* state 2. */
3521                         list_del(&h->scan_list);
3522                         complete_all(&h->scan_wait);
3523                         mutex_unlock(&scan_mutex);
3524                         return;
3525                 }
3526         }
3527         if (h->busy_scanning) { /* state 3. */
3528                 mutex_unlock(&scan_mutex);
3529                 wait_for_completion(&h->scan_wait);
3530         } else { /* state 1, nothing to do. */
3531                 mutex_unlock(&scan_mutex);
3532         }
3533 }
3534
3535 /**
3536  * scan_thread() - kernel thread used to rescan controllers
3537  * @data:        Ignored.
3538  *
3539  * A kernel thread used scan for drive topology changes on
3540  * controllers. The thread processes only one controller at a time
3541  * using a queue.  Controllers are added to the queue using
3542  * add_to_scan_list() and removed from the queue either after done
3543  * processing or using remove_from_scan_list().
3544  *
3545  * returns 0.
3546  **/
3547 static int scan_thread(void *data)
3548 {
3549         struct ctlr_info *h;
3550
3551         while (1) {
3552                 set_current_state(TASK_INTERRUPTIBLE);
3553                 schedule();
3554                 if (kthread_should_stop())
3555                         break;
3556
3557                 while (1) {
3558                         mutex_lock(&scan_mutex);
3559                         if (list_empty(&scan_q)) {
3560                                 mutex_unlock(&scan_mutex);
3561                                 break;
3562                         }
3563
3564                         h = list_entry(scan_q.next,
3565                                        struct ctlr_info,
3566                                        scan_list);
3567                         list_del(&h->scan_list);
3568                         h->busy_scanning = 1;
3569                         mutex_unlock(&scan_mutex);
3570
3571                         rebuild_lun_table(h, 0, 0);
3572                         complete_all(&h->scan_wait);
3573                         mutex_lock(&scan_mutex);
3574                         h->busy_scanning = 0;
3575                         mutex_unlock(&scan_mutex);
3576                 }
3577         }
3578
3579         return 0;
3580 }
3581
3582 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3583 {
3584         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3585                 return 0;
3586
3587         switch (c->err_info->SenseInfo[12]) {
3588         case STATE_CHANGED:
3589                 dev_warn(&h->pdev->dev, "a state change "
3590                         "detected, command retried\n");
3591                 return 1;
3592         break;
3593         case LUN_FAILED:
3594                 dev_warn(&h->pdev->dev, "LUN failure "
3595                         "detected, action required\n");
3596                 return 1;
3597         break;
3598         case REPORT_LUNS_CHANGED:
3599                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3600         /*
3601          * Here, we could call add_to_scan_list and wake up the scan thread,
3602          * except that it's quite likely that we will get more than one
3603          * REPORT_LUNS_CHANGED condition in quick succession, which means
3604          * that those which occur after the first one will likely happen
3605          * *during* the scan_thread's rescan.  And the rescan code is not
3606          * robust enough to restart in the middle, undoing what it has already
3607          * done, and it's not clear that it's even possible to do this, since
3608          * part of what it does is notify the block layer, which starts
3609          * doing it's own i/o to read partition tables and so on, and the
3610          * driver doesn't have visibility to know what might need undoing.
3611          * In any event, if possible, it is horribly complicated to get right
3612          * so we just don't do it for now.
3613          *
3614          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3615          */
3616                 return 1;
3617         break;
3618         case POWER_OR_RESET:
3619                 dev_warn(&h->pdev->dev,
3620                         "a power on or device reset detected\n");
3621                 return 1;
3622         break;
3623         case UNIT_ATTENTION_CLEARED:
3624                 dev_warn(&h->pdev->dev,
3625                         "unit attention cleared by another initiator\n");
3626                 return 1;
3627         break;
3628         default:
3629                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3630                 return 1;
3631         }
3632 }
3633
3634 /*
3635  *  We cannot read the structure directly, for portability we must use
3636  *   the io functions.
3637  *   This is for debug only.
3638  */
3639 static void print_cfg_table(ctlr_info_t *h)
3640 {
3641         int i;
3642         char temp_name[17];
3643         CfgTable_struct *tb = h->cfgtable;
3644
3645         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3646         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3647         for (i = 0; i < 4; i++)
3648                 temp_name[i] = readb(&(tb->Signature[i]));
3649         temp_name[4] = '\0';
3650         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3651         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3652                 readl(&(tb->SpecValence)));
3653         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3654                readl(&(tb->TransportSupport)));
3655         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3656                readl(&(tb->TransportActive)));
3657         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3658                readl(&(tb->HostWrite.TransportRequest)));
3659         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3660                readl(&(tb->HostWrite.CoalIntDelay)));
3661         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3662                readl(&(tb->HostWrite.CoalIntCount)));
3663         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3664                readl(&(tb->CmdsOutMax)));
3665         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3666                 readl(&(tb->BusTypes)));
3667         for (i = 0; i < 16; i++)
3668                 temp_name[i] = readb(&(tb->ServerName[i]));
3669         temp_name[16] = '\0';
3670         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3671         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3672                 readl(&(tb->HeartBeat)));
3673 }
3674
3675 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3676 {
3677         int i, offset, mem_type, bar_type;
3678         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3679                 return 0;
3680         offset = 0;
3681         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3682                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3683                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3684                         offset += 4;
3685                 else {
3686                         mem_type = pci_resource_flags(pdev, i) &
3687                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3688                         switch (mem_type) {
3689                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3690                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3691                                 offset += 4;    /* 32 bit */
3692                                 break;
3693                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3694                                 offset += 8;
3695                                 break;
3696                         default:        /* reserved in PCI 2.2 */
3697                                 dev_warn(&pdev->dev,
3698                                        "Base address is invalid\n");
3699                                 return -1;
3700                                 break;
3701                         }
3702                 }
3703                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3704                         return i + 1;
3705         }
3706         return -1;
3707 }
3708
3709 /* Fill in bucket_map[], given nsgs (the max number of
3710  * scatter gather elements supported) and bucket[],
3711  * which is an array of 8 integers.  The bucket[] array
3712  * contains 8 different DMA transfer sizes (in 16
3713  * byte increments) which the controller uses to fetch
3714  * commands.  This function fills in bucket_map[], which
3715  * maps a given number of scatter gather elements to one of
3716  * the 8 DMA transfer sizes.  The point of it is to allow the
3717  * controller to only do as much DMA as needed to fetch the
3718  * command, with the DMA transfer size encoded in the lower
3719  * bits of the command address.
3720  */
3721 static void  calc_bucket_map(int bucket[], int num_buckets,
3722         int nsgs, int *bucket_map)
3723 {
3724         int i, j, b, size;
3725
3726         /* even a command with 0 SGs requires 4 blocks */
3727 #define MINIMUM_TRANSFER_BLOCKS 4
3728 #define NUM_BUCKETS 8
3729         /* Note, bucket_map must have nsgs+1 entries. */
3730         for (i = 0; i <= nsgs; i++) {
3731                 /* Compute size of a command with i SG entries */
3732                 size = i + MINIMUM_TRANSFER_BLOCKS;
3733                 b = num_buckets; /* Assume the biggest bucket */
3734                 /* Find the bucket that is just big enough */
3735                 for (j = 0; j < 8; j++) {
3736                         if (bucket[j] >= size) {
3737                                 b = j;
3738                                 break;
3739                         }
3740                 }
3741                 /* for a command with i SG entries, use bucket b. */
3742                 bucket_map[i] = b;
3743         }
3744 }
3745
3746 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3747 {
3748         int i;
3749
3750         /* under certain very rare conditions, this can take awhile.
3751          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3752          * as we enter this code.) */
3753         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3754                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3755                         break;
3756                 msleep(10);
3757         }
3758 }
3759
3760 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3761 {
3762         /* This is a bit complicated.  There are 8 registers on
3763          * the controller which we write to to tell it 8 different
3764          * sizes of commands which there may be.  It's a way of
3765          * reducing the DMA done to fetch each command.  Encoded into
3766          * each command's tag are 3 bits which communicate to the controller
3767          * which of the eight sizes that command fits within.  The size of
3768          * each command depends on how many scatter gather entries there are.
3769          * Each SG entry requires 16 bytes.  The eight registers are programmed
3770          * with the number of 16-byte blocks a command of that size requires.
3771          * The smallest command possible requires 5 such 16 byte blocks.
3772          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3773          * blocks.  Note, this only extends to the SG entries contained
3774          * within the command block, and does not extend to chained blocks
3775          * of SG elements.   bft[] contains the eight values we write to
3776          * the registers.  They are not evenly distributed, but have more
3777          * sizes for small commands, and fewer sizes for larger commands.
3778          */
3779         __u32 trans_offset;
3780         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3781                         /*
3782                          *  5 = 1 s/g entry or 4k
3783                          *  6 = 2 s/g entry or 8k
3784                          *  8 = 4 s/g entry or 16k
3785                          * 10 = 6 s/g entry or 24k
3786                          */
3787         unsigned long register_value;
3788         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3789
3790         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3791
3792         /* Controller spec: zero out this buffer. */
3793         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3794         h->reply_pool_head = h->reply_pool;
3795
3796         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3797         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3798                                 h->blockFetchTable);
3799         writel(bft[0], &h->transtable->BlockFetch0);
3800         writel(bft[1], &h->transtable->BlockFetch1);
3801         writel(bft[2], &h->transtable->BlockFetch2);
3802         writel(bft[3], &h->transtable->BlockFetch3);
3803         writel(bft[4], &h->transtable->BlockFetch4);
3804         writel(bft[5], &h->transtable->BlockFetch5);
3805         writel(bft[6], &h->transtable->BlockFetch6);
3806         writel(bft[7], &h->transtable->BlockFetch7);
3807
3808         /* size of controller ring buffer */
3809         writel(h->max_commands, &h->transtable->RepQSize);
3810         writel(1, &h->transtable->RepQCount);
3811         writel(0, &h->transtable->RepQCtrAddrLow32);
3812         writel(0, &h->transtable->RepQCtrAddrHigh32);
3813         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3814         writel(0, &h->transtable->RepQAddr0High32);
3815         writel(CFGTBL_Trans_Performant,
3816                         &(h->cfgtable->HostWrite.TransportRequest));
3817
3818         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3819         cciss_wait_for_mode_change_ack(h);
3820         register_value = readl(&(h->cfgtable->TransportActive));
3821         if (!(register_value & CFGTBL_Trans_Performant))
3822                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3823                                         " performant mode\n");
3824 }
3825
3826 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3827 {
3828         __u32 trans_support;
3829
3830         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3831         /* Attempt to put controller into performant mode if supported */
3832         /* Does board support performant mode? */
3833         trans_support = readl(&(h->cfgtable->TransportSupport));
3834         if (!(trans_support & PERFORMANT_MODE))
3835                 return;
3836
3837         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3838         /* Performant mode demands commands on a 32 byte boundary
3839          * pci_alloc_consistent aligns on page boundarys already.
3840          * Just need to check if divisible by 32
3841          */
3842         if ((sizeof(CommandList_struct) % 32) != 0) {
3843                 dev_warn(&h->pdev->dev, "%s %d %s\n",
3844                         "cciss info: command size[",
3845                         (int)sizeof(CommandList_struct),
3846                         "] not divisible by 32, no performant mode..\n");
3847                 return;
3848         }
3849
3850         /* Performant mode ring buffer and supporting data structures */
3851         h->reply_pool = (__u64 *)pci_alloc_consistent(
3852                 h->pdev, h->max_commands * sizeof(__u64),
3853                 &(h->reply_pool_dhandle));
3854
3855         /* Need a block fetch table for performant mode */
3856         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3857                 sizeof(__u32)), GFP_KERNEL);
3858
3859         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3860                 goto clean_up;
3861
3862         cciss_enter_performant_mode(h);
3863
3864         /* Change the access methods to the performant access methods */
3865         h->access = SA5_performant_access;
3866         h->transMethod = CFGTBL_Trans_Performant;
3867
3868         return;
3869 clean_up:
3870         kfree(h->blockFetchTable);
3871         if (h->reply_pool)
3872                 pci_free_consistent(h->pdev,
3873                                 h->max_commands * sizeof(__u64),
3874                                 h->reply_pool,
3875                                 h->reply_pool_dhandle);
3876         return;
3877
3878 } /* cciss_put_controller_into_performant_mode */
3879
3880 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3881  * controllers that are capable. If not, we use IO-APIC mode.
3882  */
3883
3884 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3885 {
3886 #ifdef CONFIG_PCI_MSI
3887         int err;
3888         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3889         {0, 2}, {0, 3}
3890         };
3891
3892         /* Some boards advertise MSI but don't really support it */
3893         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3894             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3895                 goto default_int_mode;
3896
3897         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3898                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3899                 if (!err) {
3900                         h->intr[0] = cciss_msix_entries[0].vector;
3901                         h->intr[1] = cciss_msix_entries[1].vector;
3902                         h->intr[2] = cciss_msix_entries[2].vector;
3903                         h->intr[3] = cciss_msix_entries[3].vector;
3904                         h->msix_vector = 1;
3905                         return;
3906                 }
3907                 if (err > 0) {
3908                         dev_warn(&h->pdev->dev,
3909                                 "only %d MSI-X vectors available\n", err);
3910                         goto default_int_mode;
3911                 } else {
3912                         dev_warn(&h->pdev->dev,
3913                                 "MSI-X init failed %d\n", err);
3914                         goto default_int_mode;
3915                 }
3916         }
3917         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3918                 if (!pci_enable_msi(h->pdev))
3919                         h->msi_vector = 1;
3920                 else
3921                         dev_warn(&h->pdev->dev, "MSI init failed\n");
3922         }
3923 default_int_mode:
3924 #endif                          /* CONFIG_PCI_MSI */
3925         /* if we get here we're going to use the default interrupt mode */
3926         h->intr[PERF_MODE_INT] = h->pdev->irq;
3927         return;
3928 }
3929
3930 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3931 {
3932         int i;
3933         u32 subsystem_vendor_id, subsystem_device_id;
3934
3935         subsystem_vendor_id = pdev->subsystem_vendor;
3936         subsystem_device_id = pdev->subsystem_device;
3937         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3938                         subsystem_vendor_id;
3939
3940         for (i = 0; i < ARRAY_SIZE(products); i++) {
3941                 if (*board_id == products[i].board_id)
3942                         return i;
3943         }
3944         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
3945                 *board_id);
3946         return -ENODEV;
3947 }
3948
3949 static inline bool cciss_board_disabled(ctlr_info_t *h)
3950 {
3951         u16 command;
3952
3953         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
3954         return ((command & PCI_COMMAND_MEMORY) == 0);
3955 }
3956
3957 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
3958         unsigned long *memory_bar)
3959 {
3960         int i;
3961
3962         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
3963                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
3964                         /* addressing mode bits already removed */
3965                         *memory_bar = pci_resource_start(pdev, i);
3966                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
3967                                 *memory_bar);
3968                         return 0;
3969                 }
3970         dev_warn(&pdev->dev, "no memory BAR found\n");
3971         return -ENODEV;
3972 }
3973
3974 static int __devinit cciss_wait_for_board_ready(ctlr_info_t *h)
3975 {
3976         int i;
3977         u32 scratchpad;
3978
3979         for (i = 0; i < CCISS_BOARD_READY_ITERATIONS; i++) {
3980                 scratchpad = readl(h->vaddr + SA5_SCRATCHPAD_OFFSET);
3981                 if (scratchpad == CCISS_FIRMWARE_READY)
3982                         return 0;
3983                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
3984         }
3985         dev_warn(&h->pdev->dev, "board not ready, timed out.\n");
3986         return -ENODEV;
3987 }
3988
3989 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
3990         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
3991         u64 *cfg_offset)
3992 {
3993         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
3994         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
3995         *cfg_base_addr &= (u32) 0x0000ffff;
3996         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
3997         if (*cfg_base_addr_index == -1) {
3998                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
3999                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4000                 return -ENODEV;
4001         }
4002         return 0;
4003 }
4004
4005 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4006 {
4007         u64 cfg_offset;
4008         u32 cfg_base_addr;
4009         u64 cfg_base_addr_index;
4010         u32 trans_offset;
4011         int rc;
4012
4013         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4014                 &cfg_base_addr_index, &cfg_offset);
4015         if (rc)
4016                 return rc;
4017         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4018                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4019         if (!h->cfgtable)
4020                 return -ENOMEM;
4021         /* Find performant mode table. */
4022         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4023         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4024                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4025                                 sizeof(*h->transtable));
4026         if (!h->transtable)
4027                 return -ENOMEM;
4028         return 0;
4029 }
4030
4031 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4032 {
4033         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4034         if (h->max_commands < 16) {
4035                 dev_warn(&h->pdev->dev, "Controller reports "
4036                         "max supported commands of %d, an obvious lie. "
4037                         "Using 16.  Ensure that firmware is up to date.\n",
4038                         h->max_commands);
4039                 h->max_commands = 16;
4040         }
4041 }
4042
4043 /* Interrogate the hardware for some limits:
4044  * max commands, max SG elements without chaining, and with chaining,
4045  * SG chain block size, etc.
4046  */
4047 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4048 {
4049         cciss_get_max_perf_mode_cmds(h);
4050         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4051         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4052         /*
4053          * Limit in-command s/g elements to 32 save dma'able memory.
4054          * Howvever spec says if 0, use 31
4055          */
4056         h->max_cmd_sgentries = 31;
4057         if (h->maxsgentries > 512) {
4058                 h->max_cmd_sgentries = 32;
4059                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4060                 h->maxsgentries--; /* save one for chain pointer */
4061         } else {
4062                 h->maxsgentries = 31; /* default to traditional values */
4063                 h->chainsize = 0;
4064         }
4065 }
4066
4067 static inline bool CISS_signature_present(ctlr_info_t *h)
4068 {
4069         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4070             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4071             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4072             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4073                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4074                 return false;
4075         }
4076         return true;
4077 }
4078
4079 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4080 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4081 {
4082 #ifdef CONFIG_X86
4083         u32 prefetch;
4084
4085         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4086         prefetch |= 0x100;
4087         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4088 #endif
4089 }
4090
4091 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4092  * in a prefetch beyond physical memory.
4093  */
4094 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4095 {
4096         u32 dma_prefetch;
4097         __u32 dma_refetch;
4098
4099         if (h->board_id != 0x3225103C)
4100                 return;
4101         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4102         dma_prefetch |= 0x8000;
4103         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4104         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4105         dma_refetch |= 0x1;
4106         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4107 }
4108
4109 static int __devinit cciss_pci_init(ctlr_info_t *h)
4110 {
4111         int prod_index, err;
4112
4113         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4114         if (prod_index < 0)
4115                 return -ENODEV;
4116         h->product_name = products[prod_index].product_name;
4117         h->access = *(products[prod_index].access);
4118
4119         if (cciss_board_disabled(h)) {
4120                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4121                 return -ENODEV;
4122         }
4123         err = pci_enable_device(h->pdev);
4124         if (err) {
4125                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4126                 return err;
4127         }
4128
4129         err = pci_request_regions(h->pdev, "cciss");
4130         if (err) {
4131                 dev_warn(&h->pdev->dev,
4132                         "Cannot obtain PCI resources, aborting\n");
4133                 return err;
4134         }
4135
4136         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4137         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4138
4139 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4140  * else we use the IO-APIC interrupt assigned to us by system ROM.
4141  */
4142         cciss_interrupt_mode(h);
4143         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4144         if (err)
4145                 goto err_out_free_res;
4146         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4147         if (!h->vaddr) {
4148                 err = -ENOMEM;
4149                 goto err_out_free_res;
4150         }
4151         err = cciss_wait_for_board_ready(h);
4152         if (err)
4153                 goto err_out_free_res;
4154         err = cciss_find_cfgtables(h);
4155         if (err)
4156                 goto err_out_free_res;
4157         print_cfg_table(h);
4158         cciss_find_board_params(h);
4159
4160         if (!CISS_signature_present(h)) {
4161                 err = -ENODEV;
4162                 goto err_out_free_res;
4163         }
4164         cciss_enable_scsi_prefetch(h);
4165         cciss_p600_dma_prefetch_quirk(h);
4166         cciss_put_controller_into_performant_mode(h);
4167         return 0;
4168
4169 err_out_free_res:
4170         /*
4171          * Deliberately omit pci_disable_device(): it does something nasty to
4172          * Smart Array controllers that pci_enable_device does not undo
4173          */
4174         if (h->transtable)
4175                 iounmap(h->transtable);
4176         if (h->cfgtable)
4177                 iounmap(h->cfgtable);
4178         if (h->vaddr)
4179                 iounmap(h->vaddr);
4180         pci_release_regions(h->pdev);
4181         return err;
4182 }
4183
4184 /* Function to find the first free pointer into our hba[] array
4185  * Returns -1 if no free entries are left.
4186  */
4187 static int alloc_cciss_hba(struct pci_dev *pdev)
4188 {
4189         int i;
4190
4191         for (i = 0; i < MAX_CTLR; i++) {
4192                 if (!hba[i]) {
4193                         ctlr_info_t *h;
4194
4195                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4196                         if (!h)
4197                                 goto Enomem;
4198                         hba[i] = h;
4199                         return i;
4200                 }
4201         }
4202         dev_warn(&pdev->dev, "This driver supports a maximum"
4203                " of %d controllers.\n", MAX_CTLR);
4204         return -1;
4205 Enomem:
4206         dev_warn(&pdev->dev, "out of memory.\n");
4207         return -1;
4208 }
4209
4210 static void free_hba(ctlr_info_t *h)
4211 {
4212         int i;
4213
4214         hba[h->ctlr] = NULL;
4215         for (i = 0; i < h->highest_lun + 1; i++)
4216                 if (h->gendisk[i] != NULL)
4217                         put_disk(h->gendisk[i]);
4218         kfree(h);
4219 }
4220
4221 /* Send a message CDB to the firmware. */
4222 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4223 {
4224         typedef struct {
4225                 CommandListHeader_struct CommandHeader;
4226                 RequestBlock_struct Request;
4227                 ErrDescriptor_struct ErrorDescriptor;
4228         } Command;
4229         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4230         Command *cmd;
4231         dma_addr_t paddr64;
4232         uint32_t paddr32, tag;
4233         void __iomem *vaddr;
4234         int i, err;
4235
4236         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4237         if (vaddr == NULL)
4238                 return -ENOMEM;
4239
4240         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4241            CCISS commands, so they must be allocated from the lower 4GiB of
4242            memory. */
4243         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4244         if (err) {
4245                 iounmap(vaddr);
4246                 return -ENOMEM;
4247         }
4248
4249         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4250         if (cmd == NULL) {
4251                 iounmap(vaddr);
4252                 return -ENOMEM;
4253         }
4254
4255         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4256            although there's no guarantee, we assume that the address is at
4257            least 4-byte aligned (most likely, it's page-aligned). */
4258         paddr32 = paddr64;
4259
4260         cmd->CommandHeader.ReplyQueue = 0;
4261         cmd->CommandHeader.SGList = 0;
4262         cmd->CommandHeader.SGTotal = 0;
4263         cmd->CommandHeader.Tag.lower = paddr32;
4264         cmd->CommandHeader.Tag.upper = 0;
4265         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4266
4267         cmd->Request.CDBLen = 16;
4268         cmd->Request.Type.Type = TYPE_MSG;
4269         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4270         cmd->Request.Type.Direction = XFER_NONE;
4271         cmd->Request.Timeout = 0; /* Don't time out */
4272         cmd->Request.CDB[0] = opcode;
4273         cmd->Request.CDB[1] = type;
4274         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4275
4276         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4277         cmd->ErrorDescriptor.Addr.upper = 0;
4278         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4279
4280         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4281
4282         for (i = 0; i < 10; i++) {
4283                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4284                 if ((tag & ~3) == paddr32)
4285                         break;
4286                 schedule_timeout_uninterruptible(HZ);
4287         }
4288
4289         iounmap(vaddr);
4290
4291         /* we leak the DMA buffer here ... no choice since the controller could
4292            still complete the command. */
4293         if (i == 10) {
4294                 dev_err(&pdev->dev,
4295                         "controller message %02x:%02x timed out\n",
4296                         opcode, type);
4297                 return -ETIMEDOUT;
4298         }
4299
4300         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4301
4302         if (tag & 2) {
4303                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4304                         opcode, type);
4305                 return -EIO;
4306         }
4307
4308         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4309                 opcode, type);
4310         return 0;
4311 }
4312
4313 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4314 #define cciss_noop(p) cciss_message(p, 3, 0)
4315
4316 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4317 {
4318 /* the #defines are stolen from drivers/pci/msi.h. */
4319 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4320 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4321
4322         int pos;
4323         u16 control = 0;
4324
4325         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4326         if (pos) {
4327                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4328                 if (control & PCI_MSI_FLAGS_ENABLE) {
4329                         dev_info(&pdev->dev, "resetting MSI\n");
4330                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4331                 }
4332         }
4333
4334         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4335         if (pos) {
4336                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4337                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4338                         dev_info(&pdev->dev, "resetting MSI-X\n");
4339                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4340                 }
4341         }
4342
4343         return 0;
4344 }
4345
4346 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4347         void * __iomem vaddr, bool use_doorbell)
4348 {
4349         u16 pmcsr;
4350         int pos;
4351
4352         if (use_doorbell) {
4353                 /* For everything after the P600, the PCI power state method
4354                  * of resetting the controller doesn't work, so we have this
4355                  * other way using the doorbell register.
4356                  */
4357                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4358                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4359                 msleep(1000);
4360         } else { /* Try to do it the PCI power state way */
4361
4362                 /* Quoting from the Open CISS Specification: "The Power
4363                  * Management Control/Status Register (CSR) controls the power
4364                  * state of the device.  The normal operating state is D0,
4365                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4366                  * the controller, place the interface device in D3 then to D0,
4367                  * this causes a secondary PCI reset which will reset the
4368                  * controller." */
4369
4370                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4371                 if (pos == 0) {
4372                         dev_err(&pdev->dev,
4373                                 "cciss_controller_hard_reset: "
4374                                 "PCI PM not supported\n");
4375                         return -ENODEV;
4376                 }
4377                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4378                 /* enter the D3hot power management state */
4379                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4380                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4381                 pmcsr |= PCI_D3hot;
4382                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4383
4384                 msleep(500);
4385
4386                 /* enter the D0 power management state */
4387                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4388                 pmcsr |= PCI_D0;
4389                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4390
4391                 msleep(500);
4392         }
4393         return 0;
4394 }
4395
4396 /* This does a hard reset of the controller using PCI power management
4397  * states or using the doorbell register. */
4398 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4399 {
4400         u16 saved_config_space[32];
4401         u64 cfg_offset;
4402         u32 cfg_base_addr;
4403         u64 cfg_base_addr_index;
4404         void __iomem *vaddr;
4405         unsigned long paddr;
4406         u32 misc_fw_support, active_transport;
4407         int rc, i;
4408         CfgTable_struct __iomem *cfgtable;
4409         bool use_doorbell;
4410         u32 board_id;
4411
4412         /* For controllers as old a the p600, this is very nearly
4413          * the same thing as
4414          *
4415          * pci_save_state(pci_dev);
4416          * pci_set_power_state(pci_dev, PCI_D3hot);
4417          * pci_set_power_state(pci_dev, PCI_D0);
4418          * pci_restore_state(pci_dev);
4419          *
4420          * but we can't use these nice canned kernel routines on
4421          * kexec, because they also check the MSI/MSI-X state in PCI
4422          * configuration space and do the wrong thing when it is
4423          * set/cleared.  Also, the pci_save/restore_state functions
4424          * violate the ordering requirements for restoring the
4425          * configuration space from the CCISS document (see the
4426          * comment below).  So we roll our own ....
4427          *
4428          * For controllers newer than the P600, the pci power state
4429          * method of resetting doesn't work so we have another way
4430          * using the doorbell register.
4431          */
4432
4433         /* Exclude 640x boards.  These are two pci devices in one slot
4434          * which share a battery backed cache module.  One controls the
4435          * cache, the other accesses the cache through the one that controls
4436          * it.  If we reset the one controlling the cache, the other will
4437          * likely not be happy.  Just forbid resetting this conjoined mess.
4438          */
4439         cciss_lookup_board_id(pdev, &board_id);
4440         if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4441                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4442                                 "due to shared cache module.");
4443                 return -ENODEV;
4444         }
4445
4446         for (i = 0; i < 32; i++)
4447                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4448
4449         /* find the first memory BAR, so we can find the cfg table */
4450         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4451         if (rc)
4452                 return rc;
4453         vaddr = remap_pci_mem(paddr, 0x250);
4454         if (!vaddr)
4455                 return -ENOMEM;
4456
4457         /* find cfgtable in order to check if reset via doorbell is supported */
4458         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4459                                         &cfg_base_addr_index, &cfg_offset);
4460         if (rc)
4461                 goto unmap_vaddr;
4462         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4463                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4464         if (!cfgtable) {
4465                 rc = -ENOMEM;
4466                 goto unmap_vaddr;
4467         }
4468
4469         /* If reset via doorbell register is supported, use that. */
4470         misc_fw_support = readl(&cfgtable->misc_fw_support);
4471         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4472
4473         /* The doorbell reset seems to cause lockups on some Smart
4474          * Arrays (e.g. P410, P410i, maybe others).  Until this is
4475          * fixed or at least isolated, avoid the doorbell reset.
4476          */
4477         use_doorbell = 0;
4478
4479         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4480         if (rc)
4481                 goto unmap_cfgtable;
4482
4483         /* Restore the PCI configuration space.  The Open CISS
4484          * Specification says, "Restore the PCI Configuration
4485          * Registers, offsets 00h through 60h. It is important to
4486          * restore the command register, 16-bits at offset 04h,
4487          * last. Do not restore the configuration status register,
4488          * 16-bits at offset 06h."  Note that the offset is 2*i.
4489          */
4490         for (i = 0; i < 32; i++) {
4491                 if (i == 2 || i == 3)
4492                         continue;
4493                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4494         }
4495         wmb();
4496         pci_write_config_word(pdev, 4, saved_config_space[2]);
4497
4498         /* Some devices (notably the HP Smart Array 5i Controller)
4499            need a little pause here */
4500         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4501
4502         /* Controller should be in simple mode at this point.  If it's not,
4503          * It means we're on one of those controllers which doesn't support
4504          * the doorbell reset method and on which the PCI power management reset
4505          * method doesn't work (P800, for example.)
4506          * In those cases, don't try to proceed, as it generally doesn't work.
4507          */
4508         active_transport = readl(&cfgtable->TransportActive);
4509         if (active_transport & PERFORMANT_MODE) {
4510                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4511                         " Ignoring controller.\n");
4512                 rc = -ENODEV;
4513         }
4514
4515 unmap_cfgtable:
4516         iounmap(cfgtable);
4517
4518 unmap_vaddr:
4519         iounmap(vaddr);
4520         return rc;
4521 }
4522
4523 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4524 {
4525         int rc, i;
4526
4527         if (!reset_devices)
4528                 return 0;
4529
4530         /* Reset the controller with a PCI power-cycle or via doorbell */
4531         rc = cciss_kdump_hard_reset_controller(pdev);
4532
4533         /* -ENOTSUPP here means we cannot reset the controller
4534          * but it's already (and still) up and running in
4535          * "performant mode".  Or, it might be 640x, which can't reset
4536          * due to concerns about shared bbwc between 6402/6404 pair.
4537          */
4538         if (rc == -ENOTSUPP)
4539                 return 0; /* just try to do the kdump anyhow. */
4540         if (rc)
4541                 return -ENODEV;
4542         if (cciss_reset_msi(pdev))
4543                 return -ENODEV;
4544
4545         /* Now try to get the controller to respond to a no-op */
4546         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4547                 if (cciss_noop(pdev) == 0)
4548                         break;
4549                 else
4550                         dev_warn(&pdev->dev, "no-op failed%s\n",
4551                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4552                                         "; re-trying" : ""));
4553                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4554         }
4555         return 0;
4556 }
4557
4558 /*
4559  *  This is it.  Find all the controllers and register them.  I really hate
4560  *  stealing all these major device numbers.
4561  *  returns the number of block devices registered.
4562  */
4563 static int __devinit cciss_init_one(struct pci_dev *pdev,
4564                                     const struct pci_device_id *ent)
4565 {
4566         int i;
4567         int j = 0;
4568         int k = 0;
4569         int rc;
4570         int dac, return_code;
4571         InquiryData_struct *inq_buff;
4572         ctlr_info_t *h;
4573
4574         rc = cciss_init_reset_devices(pdev);
4575         if (rc)
4576                 return rc;
4577         i = alloc_cciss_hba(pdev);
4578         if (i < 0)
4579                 return -1;
4580
4581         h = hba[i];
4582         h->pdev = pdev;
4583         h->busy_initializing = 1;
4584         INIT_HLIST_HEAD(&h->cmpQ);
4585         INIT_HLIST_HEAD(&h->reqQ);
4586         mutex_init(&h->busy_shutting_down);
4587
4588         if (cciss_pci_init(h) != 0)
4589                 goto clean_no_release_regions;
4590
4591         sprintf(h->devname, "cciss%d", i);
4592         h->ctlr = i;
4593
4594         init_completion(&h->scan_wait);
4595
4596         if (cciss_create_hba_sysfs_entry(h))
4597                 goto clean0;
4598
4599         /* configure PCI DMA stuff */
4600         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4601                 dac = 1;
4602         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4603                 dac = 0;
4604         else {
4605                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4606                 goto clean1;
4607         }
4608
4609         /*
4610          * register with the major number, or get a dynamic major number
4611          * by passing 0 as argument.  This is done for greater than
4612          * 8 controller support.
4613          */
4614         if (i < MAX_CTLR_ORIG)
4615                 h->major = COMPAQ_CISS_MAJOR + i;
4616         rc = register_blkdev(h->major, h->devname);
4617         if (rc == -EBUSY || rc == -EINVAL) {
4618                 dev_err(&h->pdev->dev,
4619                        "Unable to get major number %d for %s "
4620                        "on hba %d\n", h->major, h->devname, i);
4621                 goto clean1;
4622         } else {
4623                 if (i >= MAX_CTLR_ORIG)
4624                         h->major = rc;
4625         }
4626
4627         /* make sure the board interrupts are off */
4628         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4629         if (h->msi_vector || h->msix_vector) {
4630                 if (request_irq(h->intr[PERF_MODE_INT],
4631                                 do_cciss_msix_intr,
4632                                 IRQF_DISABLED, h->devname, h)) {
4633                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4634                                h->intr[PERF_MODE_INT], h->devname);
4635                         goto clean2;
4636                 }
4637         } else {
4638                 if (request_irq(h->intr[PERF_MODE_INT], do_cciss_intx,
4639                                 IRQF_DISABLED, h->devname, h)) {
4640                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4641                                h->intr[PERF_MODE_INT], h->devname);
4642                         goto clean2;
4643                 }
4644         }
4645
4646         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4647                h->devname, pdev->device, pci_name(pdev),
4648                h->intr[PERF_MODE_INT], dac ? "" : " not");
4649
4650         h->cmd_pool_bits =
4651             kmalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4652                         * sizeof(unsigned long), GFP_KERNEL);
4653         h->cmd_pool = (CommandList_struct *)
4654             pci_alloc_consistent(h->pdev,
4655                     h->nr_cmds * sizeof(CommandList_struct),
4656                     &(h->cmd_pool_dhandle));
4657         h->errinfo_pool = (ErrorInfo_struct *)
4658             pci_alloc_consistent(h->pdev,
4659                     h->nr_cmds * sizeof(ErrorInfo_struct),
4660                     &(h->errinfo_pool_dhandle));
4661         if ((h->cmd_pool_bits == NULL)
4662             || (h->cmd_pool == NULL)
4663             || (h->errinfo_pool == NULL)) {
4664                 dev_err(&h->pdev->dev, "out of memory");
4665                 goto clean4;
4666         }
4667
4668         /* Need space for temp scatter list */
4669         h->scatter_list = kmalloc(h->max_commands *
4670                                                 sizeof(struct scatterlist *),
4671                                                 GFP_KERNEL);
4672         if (!h->scatter_list)
4673                 goto clean4;
4674
4675         for (k = 0; k < h->nr_cmds; k++) {
4676                 h->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4677                                                         h->maxsgentries,
4678                                                         GFP_KERNEL);
4679                 if (h->scatter_list[k] == NULL) {
4680                         dev_err(&h->pdev->dev,
4681                                 "could not allocate s/g lists\n");
4682                         goto clean4;
4683                 }
4684         }
4685         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4686                 h->chainsize, h->nr_cmds);
4687         if (!h->cmd_sg_list && h->chainsize > 0)
4688                 goto clean4;
4689
4690         spin_lock_init(&h->lock);
4691
4692         /* Initialize the pdev driver private data.
4693            have it point to h.  */
4694         pci_set_drvdata(pdev, h);
4695         /* command and error info recs zeroed out before
4696            they are used */
4697         memset(h->cmd_pool_bits, 0,
4698                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4699                         * sizeof(unsigned long));
4700
4701         h->num_luns = 0;
4702         h->highest_lun = -1;
4703         for (j = 0; j < CISS_MAX_LUN; j++) {
4704                 h->drv[j] = NULL;
4705                 h->gendisk[j] = NULL;
4706         }
4707
4708         cciss_scsi_setup(h);
4709
4710         /* Turn the interrupts on so we can service requests */
4711         h->access.set_intr_mask(h, CCISS_INTR_ON);
4712
4713         /* Get the firmware version */
4714         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4715         if (inq_buff == NULL) {
4716                 dev_err(&h->pdev->dev, "out of memory\n");
4717                 goto clean4;
4718         }
4719
4720         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4721                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4722         if (return_code == IO_OK) {
4723                 h->firm_ver[0] = inq_buff->data_byte[32];
4724                 h->firm_ver[1] = inq_buff->data_byte[33];
4725                 h->firm_ver[2] = inq_buff->data_byte[34];
4726                 h->firm_ver[3] = inq_buff->data_byte[35];
4727         } else {         /* send command failed */
4728                 dev_warn(&h->pdev->dev, "unable to determine firmware"
4729                         " version of controller\n");
4730         }
4731         kfree(inq_buff);
4732
4733         cciss_procinit(h);
4734
4735         h->cciss_max_sectors = 8192;
4736
4737         rebuild_lun_table(h, 1, 0);
4738         h->busy_initializing = 0;
4739         return 1;
4740
4741 clean4:
4742         kfree(h->cmd_pool_bits);
4743         /* Free up sg elements */
4744         for (k-- ; k >= 0; k--)
4745                 kfree(h->scatter_list[k]);
4746         kfree(h->scatter_list);
4747         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4748         if (h->cmd_pool)
4749                 pci_free_consistent(h->pdev,
4750                                     h->nr_cmds * sizeof(CommandList_struct),
4751                                     h->cmd_pool, h->cmd_pool_dhandle);
4752         if (h->errinfo_pool)
4753                 pci_free_consistent(h->pdev,
4754                                     h->nr_cmds * sizeof(ErrorInfo_struct),
4755                                     h->errinfo_pool,
4756                                     h->errinfo_pool_dhandle);
4757         free_irq(h->intr[PERF_MODE_INT], h);
4758 clean2:
4759         unregister_blkdev(h->major, h->devname);
4760 clean1:
4761         cciss_destroy_hba_sysfs_entry(h);
4762 clean0:
4763         pci_release_regions(pdev);
4764 clean_no_release_regions:
4765         h->busy_initializing = 0;
4766
4767         /*
4768          * Deliberately omit pci_disable_device(): it does something nasty to
4769          * Smart Array controllers that pci_enable_device does not undo
4770          */
4771         pci_set_drvdata(pdev, NULL);
4772         free_hba(h);
4773         return -1;
4774 }
4775
4776 static void cciss_shutdown(struct pci_dev *pdev)
4777 {
4778         ctlr_info_t *h;
4779         char *flush_buf;
4780         int return_code;
4781
4782         h = pci_get_drvdata(pdev);
4783         flush_buf = kzalloc(4, GFP_KERNEL);
4784         if (!flush_buf) {
4785                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
4786                 return;
4787         }
4788         /* write all data in the battery backed cache to disk */
4789         memset(flush_buf, 0, 4);
4790         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4791                 4, 0, CTLR_LUNID, TYPE_CMD);
4792         kfree(flush_buf);
4793         if (return_code != IO_OK)
4794                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
4795         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4796         free_irq(h->intr[PERF_MODE_INT], h);
4797 }
4798
4799 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4800 {
4801         ctlr_info_t *h;
4802         int i, j;
4803
4804         if (pci_get_drvdata(pdev) == NULL) {
4805                 dev_err(&pdev->dev, "Unable to remove device\n");
4806                 return;
4807         }
4808
4809         h = pci_get_drvdata(pdev);
4810         i = h->ctlr;
4811         if (hba[i] == NULL) {
4812                 dev_err(&pdev->dev, "device appears to already be removed\n");
4813                 return;
4814         }
4815
4816         mutex_lock(&h->busy_shutting_down);
4817
4818         remove_from_scan_list(h);
4819         remove_proc_entry(h->devname, proc_cciss);
4820         unregister_blkdev(h->major, h->devname);
4821
4822         /* remove it from the disk list */
4823         for (j = 0; j < CISS_MAX_LUN; j++) {
4824                 struct gendisk *disk = h->gendisk[j];
4825                 if (disk) {
4826                         struct request_queue *q = disk->queue;
4827
4828                         if (disk->flags & GENHD_FL_UP) {
4829                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
4830                                 del_gendisk(disk);
4831                         }
4832                         if (q)
4833                                 blk_cleanup_queue(q);
4834                 }
4835         }
4836
4837 #ifdef CONFIG_CISS_SCSI_TAPE
4838         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
4839 #endif
4840
4841         cciss_shutdown(pdev);
4842
4843 #ifdef CONFIG_PCI_MSI
4844         if (h->msix_vector)
4845                 pci_disable_msix(h->pdev);
4846         else if (h->msi_vector)
4847                 pci_disable_msi(h->pdev);
4848 #endif                          /* CONFIG_PCI_MSI */
4849
4850         iounmap(h->transtable);
4851         iounmap(h->cfgtable);
4852         iounmap(h->vaddr);
4853
4854         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(CommandList_struct),
4855                             h->cmd_pool, h->cmd_pool_dhandle);
4856         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(ErrorInfo_struct),
4857                             h->errinfo_pool, h->errinfo_pool_dhandle);
4858         kfree(h->cmd_pool_bits);
4859         /* Free up sg elements */
4860         for (j = 0; j < h->nr_cmds; j++)
4861                 kfree(h->scatter_list[j]);
4862         kfree(h->scatter_list);
4863         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4864         /*
4865          * Deliberately omit pci_disable_device(): it does something nasty to
4866          * Smart Array controllers that pci_enable_device does not undo
4867          */
4868         pci_release_regions(pdev);
4869         pci_set_drvdata(pdev, NULL);
4870         cciss_destroy_hba_sysfs_entry(h);
4871         mutex_unlock(&h->busy_shutting_down);
4872         free_hba(h);
4873 }
4874
4875 static struct pci_driver cciss_pci_driver = {
4876         .name = "cciss",
4877         .probe = cciss_init_one,
4878         .remove = __devexit_p(cciss_remove_one),
4879         .id_table = cciss_pci_device_id,        /* id_table */
4880         .shutdown = cciss_shutdown,
4881 };
4882
4883 /*
4884  *  This is it.  Register the PCI driver information for the cards we control
4885  *  the OS will call our registered routines when it finds one of our cards.
4886  */
4887 static int __init cciss_init(void)
4888 {
4889         int err;
4890
4891         /*
4892          * The hardware requires that commands are aligned on a 64-bit
4893          * boundary. Given that we use pci_alloc_consistent() to allocate an
4894          * array of them, the size must be a multiple of 8 bytes.
4895          */
4896         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4897         printk(KERN_INFO DRIVER_NAME "\n");
4898
4899         err = bus_register(&cciss_bus_type);
4900         if (err)
4901                 return err;
4902
4903         /* Start the scan thread */
4904         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4905         if (IS_ERR(cciss_scan_thread)) {
4906                 err = PTR_ERR(cciss_scan_thread);
4907                 goto err_bus_unregister;
4908         }
4909
4910         /* Register for our PCI devices */
4911         err = pci_register_driver(&cciss_pci_driver);
4912         if (err)
4913                 goto err_thread_stop;
4914
4915         return err;
4916
4917 err_thread_stop:
4918         kthread_stop(cciss_scan_thread);
4919 err_bus_unregister:
4920         bus_unregister(&cciss_bus_type);
4921
4922         return err;
4923 }
4924
4925 static void __exit cciss_cleanup(void)
4926 {
4927         int i;
4928
4929         pci_unregister_driver(&cciss_pci_driver);
4930         /* double check that all controller entrys have been removed */
4931         for (i = 0; i < MAX_CTLR; i++) {
4932                 if (hba[i] != NULL) {
4933                         dev_warn(&hba[i]->pdev->dev,
4934                                 "had to remove controller\n");
4935                         cciss_remove_one(hba[i]->pdev);
4936                 }
4937         }
4938         kthread_stop(cciss_scan_thread);
4939         remove_proc_entry("driver/cciss", NULL);
4940         bus_unregister(&cciss_bus_type);
4941 }
4942
4943 module_init(cciss_init);
4944 module_exit(cciss_cleanup);