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         {0x3223103C, "Smart Array P800", &SA5_access},
 117         {0x3234103C, "Smart Array P400", &SA5_access},
 118         {0x3235103C, "Smart Array P400i", &SA5_access},
 119         {0x3211103C, "Smart Array E200i", &SA5_access},
 120         {0x3212103C, "Smart Array E200", &SA5_access},
 121         {0x3213103C, "Smart Array E200i", &SA5_access},
 122         {0x3214103C, "Smart Array E200i", &SA5_access},
 123         {0x3215103C, "Smart Array E200i", &SA5_access},
 124         {0x3237103C, "Smart Array E500", &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         err = 0;
1155         err |=
1156             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1157                            sizeof(arg64.LUN_info));
1158         err |=
1159             copy_from_user(&arg64.Request, &arg32->Request,
1160                            sizeof(arg64.Request));
1161         err |=
1162             copy_from_user(&arg64.error_info, &arg32->error_info,
1163                            sizeof(arg64.error_info));
1164         err |= get_user(arg64.buf_size, &arg32->buf_size);
1165         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1166         err |= get_user(cp, &arg32->buf);
1167         arg64.buf = compat_ptr(cp);
1168         err |= copy_to_user(p, &arg64, sizeof(arg64));
1169
1170         if (err)
1171                 return -EFAULT;
1172
1173         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1174         if (err)
1175                 return err;
1176         err |=
1177             copy_in_user(&arg32->error_info, &p->error_info,
1178                          sizeof(arg32->error_info));
1179         if (err)
1180                 return -EFAULT;
1181         return err;
1182 }
1183 #endif
1184
1185 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1186 {
1187         drive_info_struct *drv = get_drv(bdev->bd_disk);
1188
1189         if (!drv->cylinders)
1190                 return -ENXIO;
1191
1192         geo->heads = drv->heads;
1193         geo->sectors = drv->sectors;
1194         geo->cylinders = drv->cylinders;
1195         return 0;
1196 }
1197
1198 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1199 {
1200         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1201                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1202                 (void)check_for_unit_attention(h, c);
1203 }
1204
1205 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1206 {
1207         cciss_pci_info_struct pciinfo;
1208
1209         if (!argp)
1210                 return -EINVAL;
1211         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1212         pciinfo.bus = h->pdev->bus->number;
1213         pciinfo.dev_fn = h->pdev->devfn;
1214         pciinfo.board_id = h->board_id;
1215         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1216                 return -EFAULT;
1217         return 0;
1218 }
1219
1220 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1221 {
1222         cciss_coalint_struct intinfo;
1223
1224         if (!argp)
1225                 return -EINVAL;
1226         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1227         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1228         if (copy_to_user
1229             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1230                 return -EFAULT;
1231         return 0;
1232 }
1233
1234 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1235 {
1236         cciss_coalint_struct intinfo;
1237         unsigned long flags;
1238         int i;
1239
1240         if (!argp)
1241                 return -EINVAL;
1242         if (!capable(CAP_SYS_ADMIN))
1243                 return -EPERM;
1244         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1245                 return -EFAULT;
1246         if ((intinfo.delay == 0) && (intinfo.count == 0))
1247                 return -EINVAL;
1248         spin_lock_irqsave(&h->lock, flags);
1249         /* Update the field, and then ring the doorbell */
1250         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1251         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1252         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1253
1254         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1255                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1256                         break;
1257                 udelay(1000); /* delay and try again */
1258         }
1259         spin_unlock_irqrestore(&h->lock, flags);
1260         if (i >= MAX_IOCTL_CONFIG_WAIT)
1261                 return -EAGAIN;
1262         return 0;
1263 }
1264
1265 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1266 {
1267         NodeName_type NodeName;
1268         int i;
1269
1270         if (!argp)
1271                 return -EINVAL;
1272         for (i = 0; i < 16; i++)
1273                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1274         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1275                 return -EFAULT;
1276         return 0;
1277 }
1278
1279 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1280 {
1281         NodeName_type NodeName;
1282         unsigned long flags;
1283         int i;
1284
1285         if (!argp)
1286                 return -EINVAL;
1287         if (!capable(CAP_SYS_ADMIN))
1288                 return -EPERM;
1289         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1290                 return -EFAULT;
1291         spin_lock_irqsave(&h->lock, flags);
1292         /* Update the field, and then ring the doorbell */
1293         for (i = 0; i < 16; i++)
1294                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1295         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1296         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1297                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1298                         break;
1299                 udelay(1000); /* delay and try again */
1300         }
1301         spin_unlock_irqrestore(&h->lock, flags);
1302         if (i >= MAX_IOCTL_CONFIG_WAIT)
1303                 return -EAGAIN;
1304         return 0;
1305 }
1306
1307 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1308 {
1309         Heartbeat_type heartbeat;
1310
1311         if (!argp)
1312                 return -EINVAL;
1313         heartbeat = readl(&h->cfgtable->HeartBeat);
1314         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1315                 return -EFAULT;
1316         return 0;
1317 }
1318
1319 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1320 {
1321         BusTypes_type BusTypes;
1322
1323         if (!argp)
1324                 return -EINVAL;
1325         BusTypes = readl(&h->cfgtable->BusTypes);
1326         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1327                 return -EFAULT;
1328         return 0;
1329 }
1330
1331 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1332 {
1333         FirmwareVer_type firmware;
1334
1335         if (!argp)
1336                 return -EINVAL;
1337         memcpy(firmware, h->firm_ver, 4);
1338
1339         if (copy_to_user
1340             (argp, firmware, sizeof(FirmwareVer_type)))
1341                 return -EFAULT;
1342         return 0;
1343 }
1344
1345 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1346 {
1347         DriverVer_type DriverVer = DRIVER_VERSION;
1348
1349         if (!argp)
1350                 return -EINVAL;
1351         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1352                 return -EFAULT;
1353         return 0;
1354 }
1355
1356 static int cciss_getluninfo(ctlr_info_t *h,
1357         struct gendisk *disk, void __user *argp)
1358 {
1359         LogvolInfo_struct luninfo;
1360         drive_info_struct *drv = get_drv(disk);
1361
1362         if (!argp)
1363                 return -EINVAL;
1364         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1365         luninfo.num_opens = drv->usage_count;
1366         luninfo.num_parts = 0;
1367         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1368                 return -EFAULT;
1369         return 0;
1370 }
1371
1372 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1373 {
1374         IOCTL_Command_struct iocommand;
1375         CommandList_struct *c;
1376         char *buff = NULL;
1377         u64bit temp64;
1378         DECLARE_COMPLETION_ONSTACK(wait);
1379
1380         if (!argp)
1381                 return -EINVAL;
1382
1383         if (!capable(CAP_SYS_RAWIO))
1384                 return -EPERM;
1385
1386         if (copy_from_user
1387             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1388                 return -EFAULT;
1389         if ((iocommand.buf_size < 1) &&
1390             (iocommand.Request.Type.Direction != XFER_NONE)) {
1391                 return -EINVAL;
1392         }
1393         if (iocommand.buf_size > 0) {
1394                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1395                 if (buff == NULL)
1396                         return -EFAULT;
1397         }
1398         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1399                 /* Copy the data into the buffer we created */
1400                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1401                         kfree(buff);
1402                         return -EFAULT;
1403                 }
1404         } else {
1405                 memset(buff, 0, iocommand.buf_size);
1406         }
1407         c = cmd_special_alloc(h);
1408         if (!c) {
1409                 kfree(buff);
1410                 return -ENOMEM;
1411         }
1412         /* Fill in the command type */
1413         c->cmd_type = CMD_IOCTL_PEND;
1414         /* Fill in Command Header */
1415         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1416         if (iocommand.buf_size > 0) { /* buffer to fill */
1417                 c->Header.SGList = 1;
1418                 c->Header.SGTotal = 1;
1419         } else { /* no buffers to fill */
1420                 c->Header.SGList = 0;
1421                 c->Header.SGTotal = 0;
1422         }
1423         c->Header.LUN = iocommand.LUN_info;
1424         /* use the kernel address the cmd block for tag */
1425         c->Header.Tag.lower = c->busaddr;
1426
1427         /* Fill in Request block */
1428         c->Request = iocommand.Request;
1429
1430         /* Fill in the scatter gather information */
1431         if (iocommand.buf_size > 0) {
1432                 temp64.val = pci_map_single(h->pdev, buff,
1433                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1434                 c->SG[0].Addr.lower = temp64.val32.lower;
1435                 c->SG[0].Addr.upper = temp64.val32.upper;
1436                 c->SG[0].Len = iocommand.buf_size;
1437                 c->SG[0].Ext = 0;  /* we are not chaining */
1438         }
1439         c->waiting = &wait;
1440
1441         enqueue_cmd_and_start_io(h, c);
1442         wait_for_completion(&wait);
1443
1444         /* unlock the buffers from DMA */
1445         temp64.val32.lower = c->SG[0].Addr.lower;
1446         temp64.val32.upper = c->SG[0].Addr.upper;
1447         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1448                          PCI_DMA_BIDIRECTIONAL);
1449         check_ioctl_unit_attention(h, c);
1450
1451         /* Copy the error information out */
1452         iocommand.error_info = *(c->err_info);
1453         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1454                 kfree(buff);
1455                 cmd_special_free(h, c);
1456                 return -EFAULT;
1457         }
1458
1459         if (iocommand.Request.Type.Direction == XFER_READ) {
1460                 /* Copy the data out of the buffer we created */
1461                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1462                         kfree(buff);
1463                         cmd_special_free(h, c);
1464                         return -EFAULT;
1465                 }
1466         }
1467         kfree(buff);
1468         cmd_special_free(h, c);
1469         return 0;
1470 }
1471
1472 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1473 {
1474         BIG_IOCTL_Command_struct *ioc;
1475         CommandList_struct *c;
1476         unsigned char **buff = NULL;
1477         int *buff_size = NULL;
1478         u64bit temp64;
1479         BYTE sg_used = 0;
1480         int status = 0;
1481         int i;
1482         DECLARE_COMPLETION_ONSTACK(wait);
1483         __u32 left;
1484         __u32 sz;
1485         BYTE __user *data_ptr;
1486
1487         if (!argp)
1488                 return -EINVAL;
1489         if (!capable(CAP_SYS_RAWIO))
1490                 return -EPERM;
1491         ioc = (BIG_IOCTL_Command_struct *)
1492             kmalloc(sizeof(*ioc), GFP_KERNEL);
1493         if (!ioc) {
1494                 status = -ENOMEM;
1495                 goto cleanup1;
1496         }
1497         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1498                 status = -EFAULT;
1499                 goto cleanup1;
1500         }
1501         if ((ioc->buf_size < 1) &&
1502             (ioc->Request.Type.Direction != XFER_NONE)) {
1503                 status = -EINVAL;
1504                 goto cleanup1;
1505         }
1506         /* Check kmalloc limits  using all SGs */
1507         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1508                 status = -EINVAL;
1509                 goto cleanup1;
1510         }
1511         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1512                 status = -EINVAL;
1513                 goto cleanup1;
1514         }
1515         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1516         if (!buff) {
1517                 status = -ENOMEM;
1518                 goto cleanup1;
1519         }
1520         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1521         if (!buff_size) {
1522                 status = -ENOMEM;
1523                 goto cleanup1;
1524         }
1525         left = ioc->buf_size;
1526         data_ptr = ioc->buf;
1527         while (left) {
1528                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1529                 buff_size[sg_used] = sz;
1530                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1531                 if (buff[sg_used] == NULL) {
1532                         status = -ENOMEM;
1533                         goto cleanup1;
1534                 }
1535                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1536                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1537                                 status = -EFAULT;
1538                                 goto cleanup1;
1539                         }
1540                 } else {
1541                         memset(buff[sg_used], 0, sz);
1542                 }
1543                 left -= sz;
1544                 data_ptr += sz;
1545                 sg_used++;
1546         }
1547         c = cmd_special_alloc(h);
1548         if (!c) {
1549                 status = -ENOMEM;
1550                 goto cleanup1;
1551         }
1552         c->cmd_type = CMD_IOCTL_PEND;
1553         c->Header.ReplyQueue = 0;
1554         c->Header.SGList = sg_used;
1555         c->Header.SGTotal = sg_used;
1556         c->Header.LUN = ioc->LUN_info;
1557         c->Header.Tag.lower = c->busaddr;
1558
1559         c->Request = ioc->Request;
1560         for (i = 0; i < sg_used; i++) {
1561                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1562                                     PCI_DMA_BIDIRECTIONAL);
1563                 c->SG[i].Addr.lower = temp64.val32.lower;
1564                 c->SG[i].Addr.upper = temp64.val32.upper;
1565                 c->SG[i].Len = buff_size[i];
1566                 c->SG[i].Ext = 0;       /* we are not chaining */
1567         }
1568         c->waiting = &wait;
1569         enqueue_cmd_and_start_io(h, c);
1570         wait_for_completion(&wait);
1571         /* unlock the buffers from DMA */
1572         for (i = 0; i < sg_used; i++) {
1573                 temp64.val32.lower = c->SG[i].Addr.lower;
1574                 temp64.val32.upper = c->SG[i].Addr.upper;
1575                 pci_unmap_single(h->pdev,
1576                         (dma_addr_t) temp64.val, buff_size[i],
1577                         PCI_DMA_BIDIRECTIONAL);
1578         }
1579         check_ioctl_unit_attention(h, c);
1580         /* Copy the error information out */
1581         ioc->error_info = *(c->err_info);
1582         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1583                 cmd_special_free(h, c);
1584                 status = -EFAULT;
1585                 goto cleanup1;
1586         }
1587         if (ioc->Request.Type.Direction == XFER_READ) {
1588                 /* Copy the data out of the buffer we created */
1589                 BYTE __user *ptr = ioc->buf;
1590                 for (i = 0; i < sg_used; i++) {
1591                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1592                                 cmd_special_free(h, c);
1593                                 status = -EFAULT;
1594                                 goto cleanup1;
1595                         }
1596                         ptr += buff_size[i];
1597                 }
1598         }
1599         cmd_special_free(h, c);
1600         status = 0;
1601 cleanup1:
1602         if (buff) {
1603                 for (i = 0; i < sg_used; i++)
1604                         kfree(buff[i]);
1605                 kfree(buff);
1606         }
1607         kfree(buff_size);
1608         kfree(ioc);
1609         return status;
1610 }
1611
1612 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1613         unsigned int cmd, unsigned long arg)
1614 {
1615         struct gendisk *disk = bdev->bd_disk;
1616         ctlr_info_t *h = get_host(disk);
1617         void __user *argp = (void __user *)arg;
1618
1619         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1620                 cmd, arg);
1621         switch (cmd) {
1622         case CCISS_GETPCIINFO:
1623                 return cciss_getpciinfo(h, argp);
1624         case CCISS_GETINTINFO:
1625                 return cciss_getintinfo(h, argp);
1626         case CCISS_SETINTINFO:
1627                 return cciss_setintinfo(h, argp);
1628         case CCISS_GETNODENAME:
1629                 return cciss_getnodename(h, argp);
1630         case CCISS_SETNODENAME:
1631                 return cciss_setnodename(h, argp);
1632         case CCISS_GETHEARTBEAT:
1633                 return cciss_getheartbeat(h, argp);
1634         case CCISS_GETBUSTYPES:
1635                 return cciss_getbustypes(h, argp);
1636         case CCISS_GETFIRMVER:
1637                 return cciss_getfirmver(h, argp);
1638         case CCISS_GETDRIVVER:
1639                 return cciss_getdrivver(h, argp);
1640         case CCISS_DEREGDISK:
1641         case CCISS_REGNEWD:
1642         case CCISS_REVALIDVOLS:
1643                 return rebuild_lun_table(h, 0, 1);
1644         case CCISS_GETLUNINFO:
1645                 return cciss_getluninfo(h, disk, argp);
1646         case CCISS_PASSTHRU:
1647                 return cciss_passthru(h, argp);
1648         case CCISS_BIG_PASSTHRU:
1649                 return cciss_bigpassthru(h, argp);
1650
1651         /* scsi_cmd_ioctl handles these, below, though some are not */
1652         /* very meaningful for cciss.  SG_IO is the main one people want. */
1653
1654         case SG_GET_VERSION_NUM:
1655         case SG_SET_TIMEOUT:
1656         case SG_GET_TIMEOUT:
1657         case SG_GET_RESERVED_SIZE:
1658         case SG_SET_RESERVED_SIZE:
1659         case SG_EMULATED_HOST:
1660         case SG_IO:
1661         case SCSI_IOCTL_SEND_COMMAND:
1662                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1663
1664         /* scsi_cmd_ioctl would normally handle these, below, but */
1665         /* they aren't a good fit for cciss, as CD-ROMs are */
1666         /* not supported, and we don't have any bus/target/lun */
1667         /* which we present to the kernel. */
1668
1669         case CDROM_SEND_PACKET:
1670         case CDROMCLOSETRAY:
1671         case CDROMEJECT:
1672         case SCSI_IOCTL_GET_IDLUN:
1673         case SCSI_IOCTL_GET_BUS_NUMBER:
1674         default:
1675                 return -ENOTTY;
1676         }
1677 }
1678
1679 static void cciss_check_queues(ctlr_info_t *h)
1680 {
1681         int start_queue = h->next_to_run;
1682         int i;
1683
1684         /* check to see if we have maxed out the number of commands that can
1685          * be placed on the queue.  If so then exit.  We do this check here
1686          * in case the interrupt we serviced was from an ioctl and did not
1687          * free any new commands.
1688          */
1689         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1690                 return;
1691
1692         /* We have room on the queue for more commands.  Now we need to queue
1693          * them up.  We will also keep track of the next queue to run so
1694          * that every queue gets a chance to be started first.
1695          */
1696         for (i = 0; i < h->highest_lun + 1; i++) {
1697                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1698                 /* make sure the disk has been added and the drive is real
1699                  * because this can be called from the middle of init_one.
1700                  */
1701                 if (!h->drv[curr_queue])
1702                         continue;
1703                 if (!(h->drv[curr_queue]->queue) ||
1704                         !(h->drv[curr_queue]->heads))
1705                         continue;
1706                 blk_start_queue(h->gendisk[curr_queue]->queue);
1707
1708                 /* check to see if we have maxed out the number of commands
1709                  * that can be placed on the queue.
1710                  */
1711                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1712                         if (curr_queue == start_queue) {
1713                                 h->next_to_run =
1714                                     (start_queue + 1) % (h->highest_lun + 1);
1715                                 break;
1716                         } else {
1717                                 h->next_to_run = curr_queue;
1718                                 break;
1719                         }
1720                 }
1721         }
1722 }
1723
1724 static void cciss_softirq_done(struct request *rq)
1725 {
1726         CommandList_struct *c = rq->completion_data;
1727         ctlr_info_t *h = hba[c->ctlr];
1728         SGDescriptor_struct *curr_sg = c->SG;
1729         u64bit temp64;
1730         unsigned long flags;
1731         int i, ddir;
1732         int sg_index = 0;
1733
1734         if (c->Request.Type.Direction == XFER_READ)
1735                 ddir = PCI_DMA_FROMDEVICE;
1736         else
1737                 ddir = PCI_DMA_TODEVICE;
1738
1739         /* command did not need to be retried */
1740         /* unmap the DMA mapping for all the scatter gather elements */
1741         for (i = 0; i < c->Header.SGList; i++) {
1742                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1743                         cciss_unmap_sg_chain_block(h, c);
1744                         /* Point to the next block */
1745                         curr_sg = h->cmd_sg_list[c->cmdindex];
1746                         sg_index = 0;
1747                 }
1748                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1749                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1750                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1751                                 ddir);
1752                 ++sg_index;
1753         }
1754
1755         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1756
1757         /* set the residual count for pc requests */
1758         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1759                 rq->resid_len = c->err_info->ResidualCnt;
1760
1761         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1762
1763         spin_lock_irqsave(&h->lock, flags);
1764         cmd_free(h, c);
1765         cciss_check_queues(h);
1766         spin_unlock_irqrestore(&h->lock, flags);
1767 }
1768
1769 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1770         unsigned char scsi3addr[], uint32_t log_unit)
1771 {
1772         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1773                 sizeof(h->drv[log_unit]->LunID));
1774 }
1775
1776 /* This function gets the SCSI vendor, model, and revision of a logical drive
1777  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1778  * they cannot be read.
1779  */
1780 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1781                                    char *vendor, char *model, char *rev)
1782 {
1783         int rc;
1784         InquiryData_struct *inq_buf;
1785         unsigned char scsi3addr[8];
1786
1787         *vendor = '\0';
1788         *model = '\0';
1789         *rev = '\0';
1790
1791         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1792         if (!inq_buf)
1793                 return;
1794
1795         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1796         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1797                         scsi3addr, TYPE_CMD);
1798         if (rc == IO_OK) {
1799                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1800                 vendor[VENDOR_LEN] = '\0';
1801                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1802                 model[MODEL_LEN] = '\0';
1803                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1804                 rev[REV_LEN] = '\0';
1805         }
1806
1807         kfree(inq_buf);
1808         return;
1809 }
1810
1811 /* This function gets the serial number of a logical drive via
1812  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1813  * number cannot be had, for whatever reason, 16 bytes of 0xff
1814  * are returned instead.
1815  */
1816 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1817                                 unsigned char *serial_no, int buflen)
1818 {
1819 #define PAGE_83_INQ_BYTES 64
1820         int rc;
1821         unsigned char *buf;
1822         unsigned char scsi3addr[8];
1823
1824         if (buflen > 16)
1825                 buflen = 16;
1826         memset(serial_no, 0xff, buflen);
1827         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1828         if (!buf)
1829                 return;
1830         memset(serial_no, 0, buflen);
1831         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1832         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1833                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1834         if (rc == IO_OK)
1835                 memcpy(serial_no, &buf[8], buflen);
1836         kfree(buf);
1837         return;
1838 }
1839
1840 /*
1841  * cciss_add_disk sets up the block device queue for a logical drive
1842  */
1843 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1844                                 int drv_index)
1845 {
1846         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1847         if (!disk->queue)
1848                 goto init_queue_failure;
1849         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1850         disk->major = h->major;
1851         disk->first_minor = drv_index << NWD_SHIFT;
1852         disk->fops = &cciss_fops;
1853         if (cciss_create_ld_sysfs_entry(h, drv_index))
1854                 goto cleanup_queue;
1855         disk->private_data = h->drv[drv_index];
1856         disk->driverfs_dev = &h->drv[drv_index]->dev;
1857
1858         /* Set up queue information */
1859         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1860
1861         /* This is a hardware imposed limit. */
1862         blk_queue_max_segments(disk->queue, h->maxsgentries);
1863
1864         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1865
1866         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1867
1868         disk->queue->queuedata = h;
1869
1870         blk_queue_logical_block_size(disk->queue,
1871                                      h->drv[drv_index]->block_size);
1872
1873         /* Make sure all queue data is written out before */
1874         /* setting h->drv[drv_index]->queue, as setting this */
1875         /* allows the interrupt handler to start the queue */
1876         wmb();
1877         h->drv[drv_index]->queue = disk->queue;
1878         add_disk(disk);
1879         return 0;
1880
1881 cleanup_queue:
1882         blk_cleanup_queue(disk->queue);
1883         disk->queue = NULL;
1884 init_queue_failure:
1885         return -1;
1886 }
1887
1888 /* This function will check the usage_count of the drive to be updated/added.
1889  * If the usage_count is zero and it is a heretofore unknown drive, or,
1890  * the drive's capacity, geometry, or serial number has changed,
1891  * then the drive information will be updated and the disk will be
1892  * re-registered with the kernel.  If these conditions don't hold,
1893  * then it will be left alone for the next reboot.  The exception to this
1894  * is disk 0 which will always be left registered with the kernel since it
1895  * is also the controller node.  Any changes to disk 0 will show up on
1896  * the next reboot.
1897  */
1898 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1899         int first_time, int via_ioctl)
1900 {
1901         struct gendisk *disk;
1902         InquiryData_struct *inq_buff = NULL;
1903         unsigned int block_size;
1904         sector_t total_size;
1905         unsigned long flags = 0;
1906         int ret = 0;
1907         drive_info_struct *drvinfo;
1908
1909         /* Get information about the disk and modify the driver structure */
1910         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1911         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1912         if (inq_buff == NULL || drvinfo == NULL)
1913                 goto mem_msg;
1914
1915         /* testing to see if 16-byte CDBs are already being used */
1916         if (h->cciss_read == CCISS_READ_16) {
1917                 cciss_read_capacity_16(h, drv_index,
1918                         &total_size, &block_size);
1919
1920         } else {
1921                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1922                 /* if read_capacity returns all F's this volume is >2TB */
1923                 /* in size so we switch to 16-byte CDB's for all */
1924                 /* read/write ops */
1925                 if (total_size == 0xFFFFFFFFULL) {
1926                         cciss_read_capacity_16(h, drv_index,
1927                         &total_size, &block_size);
1928                         h->cciss_read = CCISS_READ_16;
1929                         h->cciss_write = CCISS_WRITE_16;
1930                 } else {
1931                         h->cciss_read = CCISS_READ_10;
1932                         h->cciss_write = CCISS_WRITE_10;
1933                 }
1934         }
1935
1936         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1937                                inq_buff, drvinfo);
1938         drvinfo->block_size = block_size;
1939         drvinfo->nr_blocks = total_size + 1;
1940
1941         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1942                                 drvinfo->model, drvinfo->rev);
1943         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1944                         sizeof(drvinfo->serial_no));
1945         /* Save the lunid in case we deregister the disk, below. */
1946         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1947                 sizeof(drvinfo->LunID));
1948
1949         /* Is it the same disk we already know, and nothing's changed? */
1950         if (h->drv[drv_index]->raid_level != -1 &&
1951                 ((memcmp(drvinfo->serial_no,
1952                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1953                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1954                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1955                 drvinfo->heads == h->drv[drv_index]->heads &&
1956                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1957                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1958                         /* The disk is unchanged, nothing to update */
1959                         goto freeret;
1960
1961         /* If we get here it's not the same disk, or something's changed,
1962          * so we need to * deregister it, and re-register it, if it's not
1963          * in use.
1964          * If the disk already exists then deregister it before proceeding
1965          * (unless it's the first disk (for the controller node).
1966          */
1967         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1968                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
1969                 spin_lock_irqsave(&h->lock, flags);
1970                 h->drv[drv_index]->busy_configuring = 1;
1971                 spin_unlock_irqrestore(&h->lock, flags);
1972
1973                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
1974                  * which keeps the interrupt handler from starting
1975                  * the queue.
1976                  */
1977                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1978         }
1979
1980         /* If the disk is in use return */
1981         if (ret)
1982                 goto freeret;
1983
1984         /* Save the new information from cciss_geometry_inquiry
1985          * and serial number inquiry.  If the disk was deregistered
1986          * above, then h->drv[drv_index] will be NULL.
1987          */
1988         if (h->drv[drv_index] == NULL) {
1989                 drvinfo->device_initialized = 0;
1990                 h->drv[drv_index] = drvinfo;
1991                 drvinfo = NULL; /* so it won't be freed below. */
1992         } else {
1993                 /* special case for cxd0 */
1994                 h->drv[drv_index]->block_size = drvinfo->block_size;
1995                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
1996                 h->drv[drv_index]->heads = drvinfo->heads;
1997                 h->drv[drv_index]->sectors = drvinfo->sectors;
1998                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
1999                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2000                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2001                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2002                         VENDOR_LEN + 1);
2003                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2004                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2005         }
2006
2007         ++h->num_luns;
2008         disk = h->gendisk[drv_index];
2009         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2010
2011         /* If it's not disk 0 (drv_index != 0)
2012          * or if it was disk 0, but there was previously
2013          * no actual corresponding configured logical drive
2014          * (raid_leve == -1) then we want to update the
2015          * logical drive's information.
2016          */
2017         if (drv_index || first_time) {
2018                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2019                         cciss_free_gendisk(h, drv_index);
2020                         cciss_free_drive_info(h, drv_index);
2021                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2022                                 drv_index);
2023                         --h->num_luns;
2024                 }
2025         }
2026
2027 freeret:
2028         kfree(inq_buff);
2029         kfree(drvinfo);
2030         return;
2031 mem_msg:
2032         dev_err(&h->pdev->dev, "out of memory\n");
2033         goto freeret;
2034 }
2035
2036 /* This function will find the first index of the controllers drive array
2037  * that has a null drv pointer and allocate the drive info struct and
2038  * will return that index   This is where new drives will be added.
2039  * If the index to be returned is greater than the highest_lun index for
2040  * the controller then highest_lun is set * to this new index.
2041  * If there are no available indexes or if tha allocation fails, then -1
2042  * is returned.  * "controller_node" is used to know if this is a real
2043  * logical drive, or just the controller node, which determines if this
2044  * counts towards highest_lun.
2045  */
2046 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2047 {
2048         int i;
2049         drive_info_struct *drv;
2050
2051         /* Search for an empty slot for our drive info */
2052         for (i = 0; i < CISS_MAX_LUN; i++) {
2053
2054                 /* if not cxd0 case, and it's occupied, skip it. */
2055                 if (h->drv[i] && i != 0)
2056                         continue;
2057                 /*
2058                  * If it's cxd0 case, and drv is alloc'ed already, and a
2059                  * disk is configured there, skip it.
2060                  */
2061                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2062                         continue;
2063
2064                 /*
2065                  * We've found an empty slot.  Update highest_lun
2066                  * provided this isn't just the fake cxd0 controller node.
2067                  */
2068                 if (i > h->highest_lun && !controller_node)
2069                         h->highest_lun = i;
2070
2071                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2072                 if (i == 0 && h->drv[i] != NULL)
2073                         return i;
2074
2075                 /*
2076                  * Found an empty slot, not already alloc'ed.  Allocate it.
2077                  * Mark it with raid_level == -1, so we know it's new later on.
2078                  */
2079                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2080                 if (!drv)
2081                         return -1;
2082                 drv->raid_level = -1; /* so we know it's new */
2083                 h->drv[i] = drv;
2084                 return i;
2085         }
2086         return -1;
2087 }
2088
2089 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2090 {
2091         kfree(h->drv[drv_index]);
2092         h->drv[drv_index] = NULL;
2093 }
2094
2095 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2096 {
2097         put_disk(h->gendisk[drv_index]);
2098         h->gendisk[drv_index] = NULL;
2099 }
2100
2101 /* cciss_add_gendisk finds a free hba[]->drv structure
2102  * and allocates a gendisk if needed, and sets the lunid
2103  * in the drvinfo structure.   It returns the index into
2104  * the ->drv[] array, or -1 if none are free.
2105  * is_controller_node indicates whether highest_lun should
2106  * count this disk, or if it's only being added to provide
2107  * a means to talk to the controller in case no logical
2108  * drives have yet been configured.
2109  */
2110 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2111         int controller_node)
2112 {
2113         int drv_index;
2114
2115         drv_index = cciss_alloc_drive_info(h, controller_node);
2116         if (drv_index == -1)
2117                 return -1;
2118
2119         /*Check if the gendisk needs to be allocated */
2120         if (!h->gendisk[drv_index]) {
2121                 h->gendisk[drv_index] =
2122                         alloc_disk(1 << NWD_SHIFT);
2123                 if (!h->gendisk[drv_index]) {
2124                         dev_err(&h->pdev->dev,
2125                                 "could not allocate a new disk %d\n",
2126                                 drv_index);
2127                         goto err_free_drive_info;
2128                 }
2129         }
2130         memcpy(h->drv[drv_index]->LunID, lunid,
2131                 sizeof(h->drv[drv_index]->LunID));
2132         if (cciss_create_ld_sysfs_entry(h, drv_index))
2133                 goto err_free_disk;
2134         /* Don't need to mark this busy because nobody */
2135         /* else knows about this disk yet to contend */
2136         /* for access to it. */
2137         h->drv[drv_index]->busy_configuring = 0;
2138         wmb();
2139         return drv_index;
2140
2141 err_free_disk:
2142         cciss_free_gendisk(h, drv_index);
2143 err_free_drive_info:
2144         cciss_free_drive_info(h, drv_index);
2145         return -1;
2146 }
2147
2148 /* This is for the special case of a controller which
2149  * has no logical drives.  In this case, we still need
2150  * to register a disk so the controller can be accessed
2151  * by the Array Config Utility.
2152  */
2153 static void cciss_add_controller_node(ctlr_info_t *h)
2154 {
2155         struct gendisk *disk;
2156         int drv_index;
2157
2158         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2159                 return;
2160
2161         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2162         if (drv_index == -1)
2163                 goto error;
2164         h->drv[drv_index]->block_size = 512;
2165         h->drv[drv_index]->nr_blocks = 0;
2166         h->drv[drv_index]->heads = 0;
2167         h->drv[drv_index]->sectors = 0;
2168         h->drv[drv_index]->cylinders = 0;
2169         h->drv[drv_index]->raid_level = -1;
2170         memset(h->drv[drv_index]->serial_no, 0, 16);
2171         disk = h->gendisk[drv_index];
2172         if (cciss_add_disk(h, disk, drv_index) == 0)
2173                 return;
2174         cciss_free_gendisk(h, drv_index);
2175         cciss_free_drive_info(h, drv_index);
2176 error:
2177         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2178         return;
2179 }
2180
2181 /* This function will add and remove logical drives from the Logical
2182  * drive array of the controller and maintain persistency of ordering
2183  * so that mount points are preserved until the next reboot.  This allows
2184  * for the removal of logical drives in the middle of the drive array
2185  * without a re-ordering of those drives.
2186  * INPUT
2187  * h            = The controller to perform the operations on
2188  */
2189 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2190         int via_ioctl)
2191 {
2192         int num_luns;
2193         ReportLunData_struct *ld_buff = NULL;
2194         int return_code;
2195         int listlength = 0;
2196         int i;
2197         int drv_found;
2198         int drv_index = 0;
2199         unsigned char lunid[8] = CTLR_LUNID;
2200         unsigned long flags;
2201
2202         if (!capable(CAP_SYS_RAWIO))
2203                 return -EPERM;
2204
2205         /* Set busy_configuring flag for this operation */
2206         spin_lock_irqsave(&h->lock, flags);
2207         if (h->busy_configuring) {
2208                 spin_unlock_irqrestore(&h->lock, flags);
2209                 return -EBUSY;
2210         }
2211         h->busy_configuring = 1;
2212         spin_unlock_irqrestore(&h->lock, flags);
2213
2214         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2215         if (ld_buff == NULL)
2216                 goto mem_msg;
2217
2218         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2219                                       sizeof(ReportLunData_struct),
2220                                       0, CTLR_LUNID, TYPE_CMD);
2221
2222         if (return_code == IO_OK)
2223                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2224         else {  /* reading number of logical volumes failed */
2225                 dev_warn(&h->pdev->dev,
2226                         "report logical volume command failed\n");
2227                 listlength = 0;
2228                 goto freeret;
2229         }
2230
2231         num_luns = listlength / 8;      /* 8 bytes per entry */
2232         if (num_luns > CISS_MAX_LUN) {
2233                 num_luns = CISS_MAX_LUN;
2234                 dev_warn(&h->pdev->dev, "more luns configured"
2235                        " on controller than can be handled by"
2236                        " this driver.\n");
2237         }
2238
2239         if (num_luns == 0)
2240                 cciss_add_controller_node(h);
2241
2242         /* Compare controller drive array to driver's drive array
2243          * to see if any drives are missing on the controller due
2244          * to action of Array Config Utility (user deletes drive)
2245          * and deregister logical drives which have disappeared.
2246          */
2247         for (i = 0; i <= h->highest_lun; i++) {
2248                 int j;
2249                 drv_found = 0;
2250
2251                 /* skip holes in the array from already deleted drives */
2252                 if (h->drv[i] == NULL)
2253                         continue;
2254
2255                 for (j = 0; j < num_luns; j++) {
2256                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2257                         if (memcmp(h->drv[i]->LunID, lunid,
2258                                 sizeof(lunid)) == 0) {
2259                                 drv_found = 1;
2260                                 break;
2261                         }
2262                 }
2263                 if (!drv_found) {
2264                         /* Deregister it from the OS, it's gone. */
2265                         spin_lock_irqsave(&h->lock, flags);
2266                         h->drv[i]->busy_configuring = 1;
2267                         spin_unlock_irqrestore(&h->lock, flags);
2268                         return_code = deregister_disk(h, i, 1, via_ioctl);
2269                         if (h->drv[i] != NULL)
2270                                 h->drv[i]->busy_configuring = 0;
2271                 }
2272         }
2273
2274         /* Compare controller drive array to driver's drive array.
2275          * Check for updates in the drive information and any new drives
2276          * on the controller due to ACU adding logical drives, or changing
2277          * a logical drive's size, etc.  Reregister any new/changed drives
2278          */
2279         for (i = 0; i < num_luns; i++) {
2280                 int j;
2281
2282                 drv_found = 0;
2283
2284                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2285                 /* Find if the LUN is already in the drive array
2286                  * of the driver.  If so then update its info
2287                  * if not in use.  If it does not exist then find
2288                  * the first free index and add it.
2289                  */
2290                 for (j = 0; j <= h->highest_lun; j++) {
2291                         if (h->drv[j] != NULL &&
2292                                 memcmp(h->drv[j]->LunID, lunid,
2293                                         sizeof(h->drv[j]->LunID)) == 0) {
2294                                 drv_index = j;
2295                                 drv_found = 1;
2296                                 break;
2297                         }
2298                 }
2299
2300                 /* check if the drive was found already in the array */
2301                 if (!drv_found) {
2302                         drv_index = cciss_add_gendisk(h, lunid, 0);
2303                         if (drv_index == -1)
2304                                 goto freeret;
2305                 }
2306                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2307         }               /* end for */
2308
2309 freeret:
2310         kfree(ld_buff);
2311         h->busy_configuring = 0;
2312         /* We return -1 here to tell the ACU that we have registered/updated
2313          * all of the drives that we can and to keep it from calling us
2314          * additional times.
2315          */
2316         return -1;
2317 mem_msg:
2318         dev_err(&h->pdev->dev, "out of memory\n");
2319         h->busy_configuring = 0;
2320         goto freeret;
2321 }
2322
2323 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2324 {
2325         /* zero out the disk size info */
2326         drive_info->nr_blocks = 0;
2327         drive_info->block_size = 0;
2328         drive_info->heads = 0;
2329         drive_info->sectors = 0;
2330         drive_info->cylinders = 0;
2331         drive_info->raid_level = -1;
2332         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2333         memset(drive_info->model, 0, sizeof(drive_info->model));
2334         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2335         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2336         /*
2337          * don't clear the LUNID though, we need to remember which
2338          * one this one is.
2339          */
2340 }
2341
2342 /* This function will deregister the disk and it's queue from the
2343  * kernel.  It must be called with the controller lock held and the
2344  * drv structures busy_configuring flag set.  It's parameters are:
2345  *
2346  * disk = This is the disk to be deregistered
2347  * drv  = This is the drive_info_struct associated with the disk to be
2348  *        deregistered.  It contains information about the disk used
2349  *        by the driver.
2350  * clear_all = This flag determines whether or not the disk information
2351  *             is going to be completely cleared out and the highest_lun
2352  *             reset.  Sometimes we want to clear out information about
2353  *             the disk in preparation for re-adding it.  In this case
2354  *             the highest_lun should be left unchanged and the LunID
2355  *             should not be cleared.
2356  * via_ioctl
2357  *    This indicates whether we've reached this path via ioctl.
2358  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2359  *    If this path is reached via ioctl(), then the max_usage_count will
2360  *    be 1, as the process calling ioctl() has got to have the device open.
2361  *    If we get here via sysfs, then the max usage count will be zero.
2362 */
2363 static int deregister_disk(ctlr_info_t *h, int drv_index,
2364                            int clear_all, int via_ioctl)
2365 {
2366         int i;
2367         struct gendisk *disk;
2368         drive_info_struct *drv;
2369         int recalculate_highest_lun;
2370
2371         if (!capable(CAP_SYS_RAWIO))
2372                 return -EPERM;
2373
2374         drv = h->drv[drv_index];
2375         disk = h->gendisk[drv_index];
2376
2377         /* make sure logical volume is NOT is use */
2378         if (clear_all || (h->gendisk[0] == disk)) {
2379                 if (drv->usage_count > via_ioctl)
2380                         return -EBUSY;
2381         } else if (drv->usage_count > 0)
2382                 return -EBUSY;
2383
2384         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2385
2386         /* invalidate the devices and deregister the disk.  If it is disk
2387          * zero do not deregister it but just zero out it's values.  This
2388          * allows us to delete disk zero but keep the controller registered.
2389          */
2390         if (h->gendisk[0] != disk) {
2391                 struct request_queue *q = disk->queue;
2392                 if (disk->flags & GENHD_FL_UP) {
2393                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2394                         del_gendisk(disk);
2395                 }
2396                 if (q)
2397                         blk_cleanup_queue(q);
2398                 /* If clear_all is set then we are deleting the logical
2399                  * drive, not just refreshing its info.  For drives
2400                  * other than disk 0 we will call put_disk.  We do not
2401                  * do this for disk 0 as we need it to be able to
2402                  * configure the controller.
2403                  */
2404                 if (clear_all){
2405                         /* This isn't pretty, but we need to find the
2406                          * disk in our array and NULL our the pointer.
2407                          * This is so that we will call alloc_disk if
2408                          * this index is used again later.
2409                          */
2410                         for (i=0; i < CISS_MAX_LUN; i++){
2411                                 if (h->gendisk[i] == disk) {
2412                                         h->gendisk[i] = NULL;
2413                                         break;
2414                                 }
2415                         }
2416                         put_disk(disk);
2417                 }
2418         } else {
2419                 set_capacity(disk, 0);
2420                 cciss_clear_drive_info(drv);
2421         }
2422
2423         --h->num_luns;
2424
2425         /* if it was the last disk, find the new hightest lun */
2426         if (clear_all && recalculate_highest_lun) {
2427                 int newhighest = -1;
2428                 for (i = 0; i <= h->highest_lun; i++) {
2429                         /* if the disk has size > 0, it is available */
2430                         if (h->drv[i] && h->drv[i]->heads)
2431                                 newhighest = i;
2432                 }
2433                 h->highest_lun = newhighest;
2434         }
2435         return 0;
2436 }
2437
2438 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2439                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2440                 int cmd_type)
2441 {
2442         u64bit buff_dma_handle;
2443         int status = IO_OK;
2444
2445         c->cmd_type = CMD_IOCTL_PEND;
2446         c->Header.ReplyQueue = 0;
2447         if (buff != NULL) {
2448                 c->Header.SGList = 1;
2449                 c->Header.SGTotal = 1;
2450         } else {
2451                 c->Header.SGList = 0;
2452                 c->Header.SGTotal = 0;
2453         }
2454         c->Header.Tag.lower = c->busaddr;
2455         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2456
2457         c->Request.Type.Type = cmd_type;
2458         if (cmd_type == TYPE_CMD) {
2459                 switch (cmd) {
2460                 case CISS_INQUIRY:
2461                         /* are we trying to read a vital product page */
2462                         if (page_code != 0) {
2463                                 c->Request.CDB[1] = 0x01;
2464                                 c->Request.CDB[2] = page_code;
2465                         }
2466                         c->Request.CDBLen = 6;
2467                         c->Request.Type.Attribute = ATTR_SIMPLE;
2468                         c->Request.Type.Direction = XFER_READ;
2469                         c->Request.Timeout = 0;
2470                         c->Request.CDB[0] = CISS_INQUIRY;
2471                         c->Request.CDB[4] = size & 0xFF;
2472                         break;
2473                 case CISS_REPORT_LOG:
2474                 case CISS_REPORT_PHYS:
2475                         /* Talking to controller so It's a physical command
2476                            mode = 00 target = 0.  Nothing to write.
2477                          */
2478                         c->Request.CDBLen = 12;
2479                         c->Request.Type.Attribute = ATTR_SIMPLE;
2480                         c->Request.Type.Direction = XFER_READ;
2481                         c->Request.Timeout = 0;
2482                         c->Request.CDB[0] = cmd;
2483                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2484                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2485                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2486                         c->Request.CDB[9] = size & 0xFF;
2487                         break;
2488
2489                 case CCISS_READ_CAPACITY:
2490                         c->Request.CDBLen = 10;
2491                         c->Request.Type.Attribute = ATTR_SIMPLE;
2492                         c->Request.Type.Direction = XFER_READ;
2493                         c->Request.Timeout = 0;
2494                         c->Request.CDB[0] = cmd;
2495                         break;
2496                 case CCISS_READ_CAPACITY_16:
2497                         c->Request.CDBLen = 16;
2498                         c->Request.Type.Attribute = ATTR_SIMPLE;
2499                         c->Request.Type.Direction = XFER_READ;
2500                         c->Request.Timeout = 0;
2501                         c->Request.CDB[0] = cmd;
2502                         c->Request.CDB[1] = 0x10;
2503                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2504                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2505                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2506                         c->Request.CDB[13] = size & 0xFF;
2507                         c->Request.Timeout = 0;
2508                         c->Request.CDB[0] = cmd;
2509                         break;
2510                 case CCISS_CACHE_FLUSH:
2511                         c->Request.CDBLen = 12;
2512                         c->Request.Type.Attribute = ATTR_SIMPLE;
2513                         c->Request.Type.Direction = XFER_WRITE;
2514                         c->Request.Timeout = 0;
2515                         c->Request.CDB[0] = BMIC_WRITE;
2516                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2517                         break;
2518                 case TEST_UNIT_READY:
2519                         c->Request.CDBLen = 6;
2520                         c->Request.Type.Attribute = ATTR_SIMPLE;
2521                         c->Request.Type.Direction = XFER_NONE;
2522                         c->Request.Timeout = 0;
2523                         break;
2524                 default:
2525                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2526                         return IO_ERROR;
2527                 }
2528         } else if (cmd_type == TYPE_MSG) {
2529                 switch (cmd) {
2530                 case 0: /* ABORT message */
2531                         c->Request.CDBLen = 12;
2532                         c->Request.Type.Attribute = ATTR_SIMPLE;
2533                         c->Request.Type.Direction = XFER_WRITE;
2534                         c->Request.Timeout = 0;
2535                         c->Request.CDB[0] = cmd;        /* abort */
2536                         c->Request.CDB[1] = 0;  /* abort a command */
2537                         /* buff contains the tag of the command to abort */
2538                         memcpy(&c->Request.CDB[4], buff, 8);
2539                         break;
2540                 case 1: /* RESET message */
2541                         c->Request.CDBLen = 16;
2542                         c->Request.Type.Attribute = ATTR_SIMPLE;
2543                         c->Request.Type.Direction = XFER_NONE;
2544                         c->Request.Timeout = 0;
2545                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2546                         c->Request.CDB[0] = cmd;        /* reset */
2547                         c->Request.CDB[1] = 0x03;       /* reset a target */
2548                         break;
2549                 case 3: /* No-Op message */
2550                         c->Request.CDBLen = 1;
2551                         c->Request.Type.Attribute = ATTR_SIMPLE;
2552                         c->Request.Type.Direction = XFER_WRITE;
2553                         c->Request.Timeout = 0;
2554                         c->Request.CDB[0] = cmd;
2555                         break;
2556                 default:
2557                         dev_warn(&h->pdev->dev,
2558                                 "unknown message type %d\n", cmd);
2559                         return IO_ERROR;
2560                 }
2561         } else {
2562                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2563                 return IO_ERROR;
2564         }
2565         /* Fill in the scatter gather information */
2566         if (size > 0) {
2567                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2568                                                              buff, size,
2569                                                              PCI_DMA_BIDIRECTIONAL);
2570                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2571                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2572                 c->SG[0].Len = size;
2573                 c->SG[0].Ext = 0;       /* we are not chaining */
2574         }
2575         return status;
2576 }
2577
2578 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2579 {
2580         switch (c->err_info->ScsiStatus) {
2581         case SAM_STAT_GOOD:
2582                 return IO_OK;
2583         case SAM_STAT_CHECK_CONDITION:
2584                 switch (0xf & c->err_info->SenseInfo[2]) {
2585                 case 0: return IO_OK; /* no sense */
2586                 case 1: return IO_OK; /* recovered error */
2587                 default:
2588                         if (check_for_unit_attention(h, c))
2589                                 return IO_NEEDS_RETRY;
2590                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2591                                 "check condition, sense key = 0x%02x\n",
2592                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2593                 }
2594                 break;
2595         default:
2596                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2597                         "scsi status = 0x%02x\n",
2598                         c->Request.CDB[0], c->err_info->ScsiStatus);
2599                 break;
2600         }
2601         return IO_ERROR;
2602 }
2603
2604 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2605 {
2606         int return_status = IO_OK;
2607
2608         if (c->err_info->CommandStatus == CMD_SUCCESS)
2609                 return IO_OK;
2610
2611         switch (c->err_info->CommandStatus) {
2612         case CMD_TARGET_STATUS:
2613                 return_status = check_target_status(h, c);
2614                 break;
2615         case CMD_DATA_UNDERRUN:
2616         case CMD_DATA_OVERRUN:
2617                 /* expected for inquiry and report lun commands */
2618                 break;
2619         case CMD_INVALID:
2620                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2621                        "reported invalid\n", c->Request.CDB[0]);
2622                 return_status = IO_ERROR;
2623                 break;
2624         case CMD_PROTOCOL_ERR:
2625                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2626                        "protocol error\n", c->Request.CDB[0]);
2627                 return_status = IO_ERROR;
2628                 break;
2629         case CMD_HARDWARE_ERR:
2630                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2631                        " hardware error\n", c->Request.CDB[0]);
2632                 return_status = IO_ERROR;
2633                 break;
2634         case CMD_CONNECTION_LOST:
2635                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2636                        "connection lost\n", c->Request.CDB[0]);
2637                 return_status = IO_ERROR;
2638                 break;
2639         case CMD_ABORTED:
2640                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2641                        "aborted\n", c->Request.CDB[0]);
2642                 return_status = IO_ERROR;
2643                 break;
2644         case CMD_ABORT_FAILED:
2645                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2646                        "abort failed\n", c->Request.CDB[0]);
2647                 return_status = IO_ERROR;
2648                 break;
2649         case CMD_UNSOLICITED_ABORT:
2650                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2651                         c->Request.CDB[0]);
2652                 return_status = IO_NEEDS_RETRY;
2653                 break;
2654         default:
2655                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2656                        "unknown status %x\n", c->Request.CDB[0],
2657                        c->err_info->CommandStatus);
2658                 return_status = IO_ERROR;
2659         }
2660         return return_status;
2661 }
2662
2663 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2664         int attempt_retry)
2665 {
2666         DECLARE_COMPLETION_ONSTACK(wait);
2667         u64bit buff_dma_handle;
2668         int return_status = IO_OK;
2669
2670 resend_cmd2:
2671         c->waiting = &wait;
2672         enqueue_cmd_and_start_io(h, c);
2673
2674         wait_for_completion(&wait);
2675
2676         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2677                 goto command_done;
2678
2679         return_status = process_sendcmd_error(h, c);
2680
2681         if (return_status == IO_NEEDS_RETRY &&
2682                 c->retry_count < MAX_CMD_RETRIES) {
2683                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2684                         c->Request.CDB[0]);
2685                 c->retry_count++;
2686                 /* erase the old error information */
2687                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2688                 return_status = IO_OK;
2689                 INIT_COMPLETION(wait);
2690                 goto resend_cmd2;
2691         }
2692
2693 command_done:
2694         /* unlock the buffers from DMA */
2695         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2696         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2697         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2698                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2699         return return_status;
2700 }
2701
2702 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2703                            __u8 page_code, unsigned char scsi3addr[],
2704                         int cmd_type)
2705 {
2706         CommandList_struct *c;
2707         int return_status;
2708
2709         c = cmd_special_alloc(h);
2710         if (!c)
2711                 return -ENOMEM;
2712         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2713                 scsi3addr, cmd_type);
2714         if (return_status == IO_OK)
2715                 return_status = sendcmd_withirq_core(h, c, 1);
2716
2717         cmd_special_free(h, c);
2718         return return_status;
2719 }
2720
2721 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2722                                    sector_t total_size,
2723                                    unsigned int block_size,
2724                                    InquiryData_struct *inq_buff,
2725                                    drive_info_struct *drv)
2726 {
2727         int return_code;
2728         unsigned long t;
2729         unsigned char scsi3addr[8];
2730
2731         memset(inq_buff, 0, sizeof(InquiryData_struct));
2732         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2733         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2734                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2735         if (return_code == IO_OK) {
2736                 if (inq_buff->data_byte[8] == 0xFF) {
2737                         dev_warn(&h->pdev->dev,
2738                                "reading geometry failed, volume "
2739                                "does not support reading geometry\n");
2740                         drv->heads = 255;
2741                         drv->sectors = 32;      /* Sectors per track */
2742                         drv->cylinders = total_size + 1;
2743                         drv->raid_level = RAID_UNKNOWN;
2744                 } else {
2745                         drv->heads = inq_buff->data_byte[6];
2746                         drv->sectors = inq_buff->data_byte[7];
2747                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2748                         drv->cylinders += inq_buff->data_byte[5];
2749                         drv->raid_level = inq_buff->data_byte[8];
2750                 }
2751                 drv->block_size = block_size;
2752                 drv->nr_blocks = total_size + 1;
2753                 t = drv->heads * drv->sectors;
2754                 if (t > 1) {
2755                         sector_t real_size = total_size + 1;
2756                         unsigned long rem = sector_div(real_size, t);
2757                         if (rem)
2758                                 real_size++;
2759                         drv->cylinders = real_size;
2760                 }
2761         } else {                /* Get geometry failed */
2762                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2763         }
2764 }
2765
2766 static void
2767 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2768                     unsigned int *block_size)
2769 {
2770         ReadCapdata_struct *buf;
2771         int return_code;
2772         unsigned char scsi3addr[8];
2773
2774         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2775         if (!buf) {
2776                 dev_warn(&h->pdev->dev, "out of memory\n");
2777                 return;
2778         }
2779
2780         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2781         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2782                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2783         if (return_code == IO_OK) {
2784                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2785                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2786         } else {                /* read capacity command failed */
2787                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2788                 *total_size = 0;
2789                 *block_size = BLOCK_SIZE;
2790         }
2791         kfree(buf);
2792 }
2793
2794 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2795         sector_t *total_size, unsigned int *block_size)
2796 {
2797         ReadCapdata_struct_16 *buf;
2798         int return_code;
2799         unsigned char scsi3addr[8];
2800
2801         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2802         if (!buf) {
2803                 dev_warn(&h->pdev->dev, "out of memory\n");
2804                 return;
2805         }
2806
2807         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2808         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2809                 buf, sizeof(ReadCapdata_struct_16),
2810                         0, scsi3addr, TYPE_CMD);
2811         if (return_code == IO_OK) {
2812                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2813                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2814         } else {                /* read capacity command failed */
2815                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2816                 *total_size = 0;
2817                 *block_size = BLOCK_SIZE;
2818         }
2819         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2820                (unsigned long long)*total_size+1, *block_size);
2821         kfree(buf);
2822 }
2823
2824 static int cciss_revalidate(struct gendisk *disk)
2825 {
2826         ctlr_info_t *h = get_host(disk);
2827         drive_info_struct *drv = get_drv(disk);
2828         int logvol;
2829         int FOUND = 0;
2830         unsigned int block_size;
2831         sector_t total_size;
2832         InquiryData_struct *inq_buff = NULL;
2833
2834         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2835                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2836                         sizeof(drv->LunID)) == 0) {
2837                         FOUND = 1;
2838                         break;
2839                 }
2840         }
2841
2842         if (!FOUND)
2843                 return 1;
2844
2845         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2846         if (inq_buff == NULL) {
2847                 dev_warn(&h->pdev->dev, "out of memory\n");
2848                 return 1;
2849         }
2850         if (h->cciss_read == CCISS_READ_10) {
2851                 cciss_read_capacity(h, logvol,
2852                                         &total_size, &block_size);
2853         } else {
2854                 cciss_read_capacity_16(h, logvol,
2855                                         &total_size, &block_size);
2856         }
2857         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2858                                inq_buff, drv);
2859
2860         blk_queue_logical_block_size(drv->queue, drv->block_size);
2861         set_capacity(disk, drv->nr_blocks);
2862
2863         kfree(inq_buff);
2864         return 0;
2865 }
2866
2867 /*
2868  * Map (physical) PCI mem into (virtual) kernel space
2869  */
2870 static void __iomem *remap_pci_mem(ulong base, ulong size)
2871 {
2872         ulong page_base = ((ulong) base) & PAGE_MASK;
2873         ulong page_offs = ((ulong) base) - page_base;
2874         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2875
2876         return page_remapped ? (page_remapped + page_offs) : NULL;
2877 }
2878
2879 /*
2880  * Takes jobs of the Q and sends them to the hardware, then puts it on
2881  * the Q to wait for completion.
2882  */
2883 static void start_io(ctlr_info_t *h)
2884 {
2885         CommandList_struct *c;
2886
2887         while (!hlist_empty(&h->reqQ)) {
2888                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2889                 /* can't do anything if fifo is full */
2890                 if ((h->access.fifo_full(h))) {
2891                         dev_warn(&h->pdev->dev, "fifo full\n");
2892                         break;
2893                 }
2894
2895                 /* Get the first entry from the Request Q */
2896                 removeQ(c);
2897                 h->Qdepth--;
2898
2899                 /* Tell the controller execute command */
2900                 h->access.submit_command(h, c);
2901
2902                 /* Put job onto the completed Q */
2903                 addQ(&h->cmpQ, c);
2904         }
2905 }
2906
2907 /* Assumes that h->lock is held. */
2908 /* Zeros out the error record and then resends the command back */
2909 /* to the controller */
2910 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2911 {
2912         /* erase the old error information */
2913         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2914
2915         /* add it to software queue and then send it to the controller */
2916         addQ(&h->reqQ, c);
2917         h->Qdepth++;
2918         if (h->Qdepth > h->maxQsinceinit)
2919                 h->maxQsinceinit = h->Qdepth;
2920
2921         start_io(h);
2922 }
2923
2924 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2925         unsigned int msg_byte, unsigned int host_byte,
2926         unsigned int driver_byte)
2927 {
2928         /* inverse of macros in scsi.h */
2929         return (scsi_status_byte & 0xff) |
2930                 ((msg_byte & 0xff) << 8) |
2931                 ((host_byte & 0xff) << 16) |
2932                 ((driver_byte & 0xff) << 24);
2933 }
2934
2935 static inline int evaluate_target_status(ctlr_info_t *h,
2936                         CommandList_struct *cmd, int *retry_cmd)
2937 {
2938         unsigned char sense_key;
2939         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2940         int error_value;
2941
2942         *retry_cmd = 0;
2943         /* If we get in here, it means we got "target status", that is, scsi status */
2944         status_byte = cmd->err_info->ScsiStatus;
2945         driver_byte = DRIVER_OK;
2946         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2947
2948         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2949                 host_byte = DID_PASSTHROUGH;
2950         else
2951                 host_byte = DID_OK;
2952
2953         error_value = make_status_bytes(status_byte, msg_byte,
2954                 host_byte, driver_byte);
2955
2956         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2957                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
2958                         dev_warn(&h->pdev->dev, "cmd %p "
2959                                "has SCSI Status 0x%x\n",
2960                                cmd, cmd->err_info->ScsiStatus);
2961                 return error_value;
2962         }
2963
2964         /* check the sense key */
2965         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2966         /* no status or recovered error */
2967         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
2968             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
2969                 error_value = 0;
2970
2971         if (check_for_unit_attention(h, cmd)) {
2972                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
2973                 return 0;
2974         }
2975
2976         /* Not SG_IO or similar? */
2977         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
2978                 if (error_value != 0)
2979                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
2980                                " sense key = 0x%x\n", cmd, sense_key);
2981                 return error_value;
2982         }
2983
2984         /* SG_IO or similar, copy sense data back */
2985         if (cmd->rq->sense) {
2986                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
2987                         cmd->rq->sense_len = cmd->err_info->SenseLen;
2988                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
2989                         cmd->rq->sense_len);
2990         } else
2991                 cmd->rq->sense_len = 0;
2992
2993         return error_value;
2994 }
2995
2996 /* checks the status of the job and calls complete buffers to mark all
2997  * buffers for the completed job. Note that this function does not need
2998  * to hold the hba/queue lock.
2999  */
3000 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3001                                     int timeout)
3002 {
3003         int retry_cmd = 0;
3004         struct request *rq = cmd->rq;
3005
3006         rq->errors = 0;
3007
3008         if (timeout)
3009                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3010
3011         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3012                 goto after_error_processing;
3013
3014         switch (cmd->err_info->CommandStatus) {
3015         case CMD_TARGET_STATUS:
3016                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3017                 break;
3018         case CMD_DATA_UNDERRUN:
3019                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3020                         dev_warn(&h->pdev->dev, "cmd %p has"
3021                                " completed with data underrun "
3022                                "reported\n", cmd);
3023                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3024                 }
3025                 break;
3026         case CMD_DATA_OVERRUN:
3027                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3028                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3029                                " completed with data overrun "
3030                                "reported\n", cmd);
3031                 break;
3032         case CMD_INVALID:
3033                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3034                        "reported invalid\n", cmd);
3035                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3036                         cmd->err_info->CommandStatus, DRIVER_OK,
3037                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3038                                 DID_PASSTHROUGH : DID_ERROR);
3039                 break;
3040         case CMD_PROTOCOL_ERR:
3041                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3042                        "protocol error\n", cmd);
3043                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3044                         cmd->err_info->CommandStatus, DRIVER_OK,
3045                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3046                                 DID_PASSTHROUGH : DID_ERROR);
3047                 break;
3048         case CMD_HARDWARE_ERR:
3049                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3050                        " hardware error\n", cmd);
3051                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3052                         cmd->err_info->CommandStatus, DRIVER_OK,
3053                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3054                                 DID_PASSTHROUGH : DID_ERROR);
3055                 break;
3056         case CMD_CONNECTION_LOST:
3057                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3058                        "connection lost\n", cmd);
3059                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3060                         cmd->err_info->CommandStatus, DRIVER_OK,
3061                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3062                                 DID_PASSTHROUGH : DID_ERROR);
3063                 break;
3064         case CMD_ABORTED:
3065                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3066                        "aborted\n", cmd);
3067                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3068                         cmd->err_info->CommandStatus, DRIVER_OK,
3069                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3070                                 DID_PASSTHROUGH : DID_ABORT);
3071                 break;
3072         case CMD_ABORT_FAILED:
3073                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3074                        "abort failed\n", cmd);
3075                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3076                         cmd->err_info->CommandStatus, DRIVER_OK,
3077                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3078                                 DID_PASSTHROUGH : DID_ERROR);
3079                 break;
3080         case CMD_UNSOLICITED_ABORT:
3081                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3082                        "abort %p\n", h->ctlr, cmd);
3083                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3084                         retry_cmd = 1;
3085                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3086                         cmd->retry_count++;
3087                 } else
3088                         dev_warn(&h->pdev->dev,
3089                                 "%p retried too many times\n", cmd);
3090                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3091                         cmd->err_info->CommandStatus, DRIVER_OK,
3092                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3093                                 DID_PASSTHROUGH : DID_ABORT);
3094                 break;
3095         case CMD_TIMEOUT:
3096                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3097                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3098                         cmd->err_info->CommandStatus, DRIVER_OK,
3099                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3100                                 DID_PASSTHROUGH : DID_ERROR);
3101                 break;
3102         default:
3103                 dev_warn(&h->pdev->dev, "cmd %p returned "
3104                        "unknown status %x\n", cmd,
3105                        cmd->err_info->CommandStatus);
3106                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3107                         cmd->err_info->CommandStatus, DRIVER_OK,
3108                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3109                                 DID_PASSTHROUGH : DID_ERROR);
3110         }
3111
3112 after_error_processing:
3113
3114         /* We need to return this command */
3115         if (retry_cmd) {
3116                 resend_cciss_cmd(h, cmd);
3117                 return;
3118         }
3119         cmd->rq->completion_data = cmd;
3120         blk_complete_request(cmd->rq);
3121 }
3122
3123 static inline u32 cciss_tag_contains_index(u32 tag)
3124 {
3125 #define DIRECT_LOOKUP_BIT 0x10
3126         return tag & DIRECT_LOOKUP_BIT;
3127 }
3128
3129 static inline u32 cciss_tag_to_index(u32 tag)
3130 {
3131 #define DIRECT_LOOKUP_SHIFT 5
3132         return tag >> DIRECT_LOOKUP_SHIFT;
3133 }
3134
3135 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3136 {
3137 #define CCISS_ERROR_BITS 0x03
3138         return tag & ~CCISS_ERROR_BITS;
3139 }
3140
3141 static inline void cciss_mark_tag_indexed(u32 *tag)
3142 {
3143         *tag |= DIRECT_LOOKUP_BIT;
3144 }
3145
3146 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3147 {
3148         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3149 }
3150
3151 /*
3152  * Get a request and submit it to the controller.
3153  */
3154 static void do_cciss_request(struct request_queue *q)
3155 {
3156         ctlr_info_t *h = q->queuedata;
3157         CommandList_struct *c;
3158         sector_t start_blk;
3159         int seg;
3160         struct request *creq;
3161         u64bit temp64;
3162         struct scatterlist *tmp_sg;
3163         SGDescriptor_struct *curr_sg;
3164         drive_info_struct *drv;
3165         int i, dir;
3166         int sg_index = 0;
3167         int chained = 0;
3168
3169         /* We call start_io here in case there is a command waiting on the
3170          * queue that has not been sent.
3171          */
3172         if (blk_queue_plugged(q))
3173                 goto startio;
3174
3175       queue:
3176         creq = blk_peek_request(q);
3177         if (!creq)
3178                 goto startio;
3179
3180         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3181
3182         c = cmd_alloc(h);
3183         if (!c)
3184                 goto full;
3185
3186         blk_start_request(creq);
3187
3188         tmp_sg = h->scatter_list[c->cmdindex];
3189         spin_unlock_irq(q->queue_lock);
3190
3191         c->cmd_type = CMD_RWREQ;
3192         c->rq = creq;
3193
3194         /* fill in the request */
3195         drv = creq->rq_disk->private_data;
3196         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3197         /* got command from pool, so use the command block index instead */
3198         /* for direct lookups. */
3199         /* The first 2 bits are reserved for controller error reporting. */
3200         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3201         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3202         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3203         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3204         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3205         c->Request.Type.Attribute = ATTR_SIMPLE;
3206         c->Request.Type.Direction =
3207             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3208         c->Request.Timeout = 0; /* Don't time out */
3209         c->Request.CDB[0] =
3210             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3211         start_blk = blk_rq_pos(creq);
3212         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3213                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3214         sg_init_table(tmp_sg, h->maxsgentries);
3215         seg = blk_rq_map_sg(q, creq, tmp_sg);
3216
3217         /* get the DMA records for the setup */
3218         if (c->Request.Type.Direction == XFER_READ)
3219                 dir = PCI_DMA_FROMDEVICE;
3220         else
3221                 dir = PCI_DMA_TODEVICE;
3222
3223         curr_sg = c->SG;
3224         sg_index = 0;
3225         chained = 0;
3226
3227         for (i = 0; i < seg; i++) {
3228                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3229                         !chained && ((seg - i) > 1)) {
3230                         /* Point to next chain block. */
3231                         curr_sg = h->cmd_sg_list[c->cmdindex];
3232                         sg_index = 0;
3233                         chained = 1;
3234                 }
3235                 curr_sg[sg_index].Len = tmp_sg[i].length;
3236                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3237                                                 tmp_sg[i].offset,
3238                                                 tmp_sg[i].length, dir);
3239                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3240                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3241                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3242                 ++sg_index;
3243         }
3244         if (chained)
3245                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3246                         (seg - (h->max_cmd_sgentries - 1)) *
3247                                 sizeof(SGDescriptor_struct));
3248
3249         /* track how many SG entries we are using */
3250         if (seg > h->maxSG)
3251                 h->maxSG = seg;
3252
3253         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3254                         "chained[%d]\n",
3255                         blk_rq_sectors(creq), seg, chained);
3256
3257         c->Header.SGTotal = seg + chained;
3258         if (seg <= h->max_cmd_sgentries)
3259                 c->Header.SGList = c->Header.SGTotal;
3260         else
3261                 c->Header.SGList = h->max_cmd_sgentries;
3262         set_performant_mode(h, c);
3263
3264         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3265                 if(h->cciss_read == CCISS_READ_10) {
3266                         c->Request.CDB[1] = 0;
3267                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3268                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3269                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3270                         c->Request.CDB[5] = start_blk & 0xff;
3271                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3272                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3273                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3274                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3275                 } else {
3276                         u32 upper32 = upper_32_bits(start_blk);
3277
3278                         c->Request.CDBLen = 16;
3279                         c->Request.CDB[1]= 0;
3280                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3281                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3282                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3283                         c->Request.CDB[5]= upper32 & 0xff;
3284                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3285                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3286                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3287                         c->Request.CDB[9]= start_blk & 0xff;
3288                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3289                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3290                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3291                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3292                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3293                 }
3294         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3295                 c->Request.CDBLen = creq->cmd_len;
3296                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3297         } else {
3298                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3299                         creq->cmd_type);
3300                 BUG();
3301         }
3302
3303         spin_lock_irq(q->queue_lock);
3304
3305         addQ(&h->reqQ, c);
3306         h->Qdepth++;
3307         if (h->Qdepth > h->maxQsinceinit)
3308                 h->maxQsinceinit = h->Qdepth;
3309
3310         goto queue;
3311 full:
3312         blk_stop_queue(q);
3313 startio:
3314         /* We will already have the driver lock here so not need
3315          * to lock it.
3316          */
3317         start_io(h);
3318 }
3319
3320 static inline unsigned long get_next_completion(ctlr_info_t *h)
3321 {
3322         return h->access.command_completed(h);
3323 }
3324
3325 static inline int interrupt_pending(ctlr_info_t *h)
3326 {
3327         return h->access.intr_pending(h);
3328 }
3329
3330 static inline long interrupt_not_for_us(ctlr_info_t *h)
3331 {
3332         return ((h->access.intr_pending(h) == 0) ||
3333                 (h->interrupts_enabled == 0));
3334 }
3335
3336 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3337                         u32 raw_tag)
3338 {
3339         if (unlikely(tag_index >= h->nr_cmds)) {
3340                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3341                 return 1;
3342         }
3343         return 0;
3344 }
3345
3346 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3347                                 u32 raw_tag)
3348 {
3349         removeQ(c);
3350         if (likely(c->cmd_type == CMD_RWREQ))
3351                 complete_command(h, c, 0);
3352         else if (c->cmd_type == CMD_IOCTL_PEND)
3353                 complete(c->waiting);
3354 #ifdef CONFIG_CISS_SCSI_TAPE
3355         else if (c->cmd_type == CMD_SCSI)
3356                 complete_scsi_command(c, 0, raw_tag);
3357 #endif
3358 }
3359
3360 static inline u32 next_command(ctlr_info_t *h)
3361 {
3362         u32 a;
3363
3364         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3365                 return h->access.command_completed(h);
3366
3367         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3368                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3369                 (h->reply_pool_head)++;
3370                 h->commands_outstanding--;
3371         } else {
3372                 a = FIFO_EMPTY;
3373         }
3374         /* Check for wraparound */
3375         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3376                 h->reply_pool_head = h->reply_pool;
3377                 h->reply_pool_wraparound ^= 1;
3378         }
3379         return a;
3380 }
3381
3382 /* process completion of an indexed ("direct lookup") command */
3383 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3384 {
3385         u32 tag_index;
3386         CommandList_struct *c;
3387
3388         tag_index = cciss_tag_to_index(raw_tag);
3389         if (bad_tag(h, tag_index, raw_tag))
3390                 return next_command(h);
3391         c = h->cmd_pool + tag_index;
3392         finish_cmd(h, c, raw_tag);
3393         return next_command(h);
3394 }
3395
3396 /* process completion of a non-indexed command */
3397 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3398 {
3399         u32 tag;
3400         CommandList_struct *c = NULL;
3401         struct hlist_node *tmp;
3402         __u32 busaddr_masked, tag_masked;
3403
3404         tag = cciss_tag_discard_error_bits(raw_tag);
3405         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3406                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3407                 tag_masked = cciss_tag_discard_error_bits(tag);
3408                 if (busaddr_masked == tag_masked) {
3409                         finish_cmd(h, c, raw_tag);
3410                         return next_command(h);
3411                 }
3412         }
3413         bad_tag(h, h->nr_cmds + 1, raw_tag);
3414         return next_command(h);
3415 }
3416
3417 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3418 {
3419         ctlr_info_t *h = dev_id;
3420         unsigned long flags;
3421         u32 raw_tag;
3422
3423         if (interrupt_not_for_us(h))
3424                 return IRQ_NONE;
3425         spin_lock_irqsave(&h->lock, flags);
3426         while (interrupt_pending(h)) {
3427                 raw_tag = get_next_completion(h);
3428                 while (raw_tag != FIFO_EMPTY) {
3429                         if (cciss_tag_contains_index(raw_tag))
3430                                 raw_tag = process_indexed_cmd(h, raw_tag);
3431                         else
3432                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3433                 }
3434         }
3435         spin_unlock_irqrestore(&h->lock, flags);
3436         return IRQ_HANDLED;
3437 }
3438
3439 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3440  * check the interrupt pending register because it is not set.
3441  */
3442 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3443 {
3444         ctlr_info_t *h = dev_id;
3445         unsigned long flags;
3446         u32 raw_tag;
3447
3448         spin_lock_irqsave(&h->lock, flags);
3449         raw_tag = get_next_completion(h);
3450         while (raw_tag != FIFO_EMPTY) {
3451                 if (cciss_tag_contains_index(raw_tag))
3452                         raw_tag = process_indexed_cmd(h, raw_tag);
3453                 else
3454                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3455         }
3456         spin_unlock_irqrestore(&h->lock, flags);
3457         return IRQ_HANDLED;
3458 }
3459
3460 /**
3461  * add_to_scan_list() - add controller to rescan queue
3462  * @h:                Pointer to the controller.
3463  *
3464  * Adds the controller to the rescan queue if not already on the queue.
3465  *
3466  * returns 1 if added to the queue, 0 if skipped (could be on the
3467  * queue already, or the controller could be initializing or shutting
3468  * down).
3469  **/
3470 static int add_to_scan_list(struct ctlr_info *h)
3471 {
3472         struct ctlr_info *test_h;
3473         int found = 0;
3474         int ret = 0;
3475
3476         if (h->busy_initializing)
3477                 return 0;
3478
3479         if (!mutex_trylock(&h->busy_shutting_down))
3480                 return 0;
3481
3482         mutex_lock(&scan_mutex);
3483         list_for_each_entry(test_h, &scan_q, scan_list) {
3484                 if (test_h == h) {
3485                         found = 1;
3486                         break;
3487                 }
3488         }
3489         if (!found && !h->busy_scanning) {
3490                 INIT_COMPLETION(h->scan_wait);
3491                 list_add_tail(&h->scan_list, &scan_q);
3492                 ret = 1;
3493         }
3494         mutex_unlock(&scan_mutex);
3495         mutex_unlock(&h->busy_shutting_down);
3496
3497         return ret;
3498 }
3499
3500 /**
3501  * remove_from_scan_list() - remove controller from rescan queue
3502  * @h:                     Pointer to the controller.
3503  *
3504  * Removes the controller from the rescan queue if present. Blocks if
3505  * the controller is currently conducting a rescan.  The controller
3506  * can be in one of three states:
3507  * 1. Doesn't need a scan
3508  * 2. On the scan list, but not scanning yet (we remove it)
3509  * 3. Busy scanning (and not on the list). In this case we want to wait for
3510  *    the scan to complete to make sure the scanning thread for this
3511  *    controller is completely idle.
3512  **/
3513 static void remove_from_scan_list(struct ctlr_info *h)
3514 {
3515         struct ctlr_info *test_h, *tmp_h;
3516
3517         mutex_lock(&scan_mutex);
3518         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3519                 if (test_h == h) { /* state 2. */
3520                         list_del(&h->scan_list);
3521                         complete_all(&h->scan_wait);
3522                         mutex_unlock(&scan_mutex);
3523                         return;
3524                 }
3525         }
3526         if (h->busy_scanning) { /* state 3. */
3527                 mutex_unlock(&scan_mutex);
3528                 wait_for_completion(&h->scan_wait);
3529         } else { /* state 1, nothing to do. */
3530                 mutex_unlock(&scan_mutex);
3531         }
3532 }
3533
3534 /**
3535  * scan_thread() - kernel thread used to rescan controllers
3536  * @data:        Ignored.
3537  *
3538  * A kernel thread used scan for drive topology changes on
3539  * controllers. The thread processes only one controller at a time
3540  * using a queue.  Controllers are added to the queue using
3541  * add_to_scan_list() and removed from the queue either after done
3542  * processing or using remove_from_scan_list().
3543  *
3544  * returns 0.
3545  **/
3546 static int scan_thread(void *data)
3547 {
3548         struct ctlr_info *h;
3549
3550         while (1) {
3551                 set_current_state(TASK_INTERRUPTIBLE);
3552                 schedule();
3553                 if (kthread_should_stop())
3554                         break;
3555
3556                 while (1) {
3557                         mutex_lock(&scan_mutex);
3558                         if (list_empty(&scan_q)) {
3559                                 mutex_unlock(&scan_mutex);
3560                                 break;
3561                         }
3562
3563                         h = list_entry(scan_q.next,
3564                                        struct ctlr_info,
3565                                        scan_list);
3566                         list_del(&h->scan_list);
3567                         h->busy_scanning = 1;
3568                         mutex_unlock(&scan_mutex);
3569
3570                         rebuild_lun_table(h, 0, 0);
3571                         complete_all(&h->scan_wait);
3572                         mutex_lock(&scan_mutex);
3573                         h->busy_scanning = 0;
3574                         mutex_unlock(&scan_mutex);
3575                 }
3576         }
3577
3578         return 0;
3579 }
3580
3581 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3582 {
3583         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3584                 return 0;
3585
3586         switch (c->err_info->SenseInfo[12]) {
3587         case STATE_CHANGED:
3588                 dev_warn(&h->pdev->dev, "a state change "
3589                         "detected, command retried\n");
3590                 return 1;
3591         break;
3592         case LUN_FAILED:
3593                 dev_warn(&h->pdev->dev, "LUN failure "
3594                         "detected, action required\n");
3595                 return 1;
3596         break;
3597         case REPORT_LUNS_CHANGED:
3598                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3599         /*
3600          * Here, we could call add_to_scan_list and wake up the scan thread,
3601          * except that it's quite likely that we will get more than one
3602          * REPORT_LUNS_CHANGED condition in quick succession, which means
3603          * that those which occur after the first one will likely happen
3604          * *during* the scan_thread's rescan.  And the rescan code is not
3605          * robust enough to restart in the middle, undoing what it has already
3606          * done, and it's not clear that it's even possible to do this, since
3607          * part of what it does is notify the block layer, which starts
3608          * doing it's own i/o to read partition tables and so on, and the
3609          * driver doesn't have visibility to know what might need undoing.
3610          * In any event, if possible, it is horribly complicated to get right
3611          * so we just don't do it for now.
3612          *
3613          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3614          */
3615                 return 1;
3616         break;
3617         case POWER_OR_RESET:
3618                 dev_warn(&h->pdev->dev,
3619                         "a power on or device reset detected\n");
3620                 return 1;
3621         break;
3622         case UNIT_ATTENTION_CLEARED:
3623                 dev_warn(&h->pdev->dev,
3624                         "unit attention cleared by another initiator\n");
3625                 return 1;
3626         break;
3627         default:
3628                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3629                 return 1;
3630         }
3631 }
3632
3633 /*
3634  *  We cannot read the structure directly, for portability we must use
3635  *   the io functions.
3636  *   This is for debug only.
3637  */
3638 static void print_cfg_table(ctlr_info_t *h)
3639 {
3640         int i;
3641         char temp_name[17];
3642         CfgTable_struct *tb = h->cfgtable;
3643
3644         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3645         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3646         for (i = 0; i < 4; i++)
3647                 temp_name[i] = readb(&(tb->Signature[i]));
3648         temp_name[4] = '\0';
3649         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3650         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3651                 readl(&(tb->SpecValence)));
3652         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3653                readl(&(tb->TransportSupport)));
3654         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3655                readl(&(tb->TransportActive)));
3656         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3657                readl(&(tb->HostWrite.TransportRequest)));
3658         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3659                readl(&(tb->HostWrite.CoalIntDelay)));
3660         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3661                readl(&(tb->HostWrite.CoalIntCount)));
3662         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3663                readl(&(tb->CmdsOutMax)));
3664         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3665                 readl(&(tb->BusTypes)));
3666         for (i = 0; i < 16; i++)
3667                 temp_name[i] = readb(&(tb->ServerName[i]));
3668         temp_name[16] = '\0';
3669         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3670         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3671                 readl(&(tb->HeartBeat)));
3672 }
3673
3674 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3675 {
3676         int i, offset, mem_type, bar_type;
3677         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3678                 return 0;
3679         offset = 0;
3680         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3681                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3682                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3683                         offset += 4;
3684                 else {
3685                         mem_type = pci_resource_flags(pdev, i) &
3686                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3687                         switch (mem_type) {
3688                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3689                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3690                                 offset += 4;    /* 32 bit */
3691                                 break;
3692                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3693                                 offset += 8;
3694                                 break;
3695                         default:        /* reserved in PCI 2.2 */
3696                                 dev_warn(&pdev->dev,
3697                                        "Base address is invalid\n");
3698                                 return -1;
3699                                 break;
3700                         }
3701                 }
3702                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3703                         return i + 1;
3704         }
3705         return -1;
3706 }
3707
3708 /* Fill in bucket_map[], given nsgs (the max number of
3709  * scatter gather elements supported) and bucket[],
3710  * which is an array of 8 integers.  The bucket[] array
3711  * contains 8 different DMA transfer sizes (in 16
3712  * byte increments) which the controller uses to fetch
3713  * commands.  This function fills in bucket_map[], which
3714  * maps a given number of scatter gather elements to one of
3715  * the 8 DMA transfer sizes.  The point of it is to allow the
3716  * controller to only do as much DMA as needed to fetch the
3717  * command, with the DMA transfer size encoded in the lower
3718  * bits of the command address.
3719  */
3720 static void  calc_bucket_map(int bucket[], int num_buckets,
3721         int nsgs, int *bucket_map)
3722 {
3723         int i, j, b, size;
3724
3725         /* even a command with 0 SGs requires 4 blocks */
3726 #define MINIMUM_TRANSFER_BLOCKS 4
3727 #define NUM_BUCKETS 8
3728         /* Note, bucket_map must have nsgs+1 entries. */
3729         for (i = 0; i <= nsgs; i++) {
3730                 /* Compute size of a command with i SG entries */
3731                 size = i + MINIMUM_TRANSFER_BLOCKS;
3732                 b = num_buckets; /* Assume the biggest bucket */
3733                 /* Find the bucket that is just big enough */
3734                 for (j = 0; j < 8; j++) {
3735                         if (bucket[j] >= size) {
3736                                 b = j;
3737                                 break;
3738                         }
3739                 }
3740                 /* for a command with i SG entries, use bucket b. */
3741                 bucket_map[i] = b;
3742         }
3743 }
3744
3745 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3746 {
3747         int i;
3748
3749         /* under certain very rare conditions, this can take awhile.
3750          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3751          * as we enter this code.) */
3752         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3753                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3754                         break;
3755                 usleep_range(10000, 20000);
3756         }
3757 }
3758
3759 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3760 {
3761         /* This is a bit complicated.  There are 8 registers on
3762          * the controller which we write to to tell it 8 different
3763          * sizes of commands which there may be.  It's a way of
3764          * reducing the DMA done to fetch each command.  Encoded into
3765          * each command's tag are 3 bits which communicate to the controller
3766          * which of the eight sizes that command fits within.  The size of
3767          * each command depends on how many scatter gather entries there are.
3768          * Each SG entry requires 16 bytes.  The eight registers are programmed
3769          * with the number of 16-byte blocks a command of that size requires.
3770          * The smallest command possible requires 5 such 16 byte blocks.
3771          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3772          * blocks.  Note, this only extends to the SG entries contained
3773          * within the command block, and does not extend to chained blocks
3774          * of SG elements.   bft[] contains the eight values we write to
3775          * the registers.  They are not evenly distributed, but have more
3776          * sizes for small commands, and fewer sizes for larger commands.
3777          */
3778         __u32 trans_offset;
3779         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3780                         /*
3781                          *  5 = 1 s/g entry or 4k
3782                          *  6 = 2 s/g entry or 8k
3783                          *  8 = 4 s/g entry or 16k
3784                          * 10 = 6 s/g entry or 24k
3785                          */
3786         unsigned long register_value;
3787         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3788
3789         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3790
3791         /* Controller spec: zero out this buffer. */
3792         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3793         h->reply_pool_head = h->reply_pool;
3794
3795         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3796         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3797                                 h->blockFetchTable);
3798         writel(bft[0], &h->transtable->BlockFetch0);
3799         writel(bft[1], &h->transtable->BlockFetch1);
3800         writel(bft[2], &h->transtable->BlockFetch2);
3801         writel(bft[3], &h->transtable->BlockFetch3);
3802         writel(bft[4], &h->transtable->BlockFetch4);
3803         writel(bft[5], &h->transtable->BlockFetch5);
3804         writel(bft[6], &h->transtable->BlockFetch6);
3805         writel(bft[7], &h->transtable->BlockFetch7);
3806
3807         /* size of controller ring buffer */
3808         writel(h->max_commands, &h->transtable->RepQSize);
3809         writel(1, &h->transtable->RepQCount);
3810         writel(0, &h->transtable->RepQCtrAddrLow32);
3811         writel(0, &h->transtable->RepQCtrAddrHigh32);
3812         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3813         writel(0, &h->transtable->RepQAddr0High32);
3814         writel(CFGTBL_Trans_Performant,
3815                         &(h->cfgtable->HostWrite.TransportRequest));
3816
3817         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3818         cciss_wait_for_mode_change_ack(h);
3819         register_value = readl(&(h->cfgtable->TransportActive));
3820         if (!(register_value & CFGTBL_Trans_Performant))
3821                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3822                                         " performant mode\n");
3823 }
3824
3825 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3826 {
3827         __u32 trans_support;
3828
3829         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3830         /* Attempt to put controller into performant mode if supported */
3831         /* Does board support performant mode? */
3832         trans_support = readl(&(h->cfgtable->TransportSupport));
3833         if (!(trans_support & PERFORMANT_MODE))
3834                 return;
3835
3836         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3837         /* Performant mode demands commands on a 32 byte boundary
3838          * pci_alloc_consistent aligns on page boundarys already.
3839          * Just need to check if divisible by 32
3840          */
3841         if ((sizeof(CommandList_struct) % 32) != 0) {
3842                 dev_warn(&h->pdev->dev, "%s %d %s\n",
3843                         "cciss info: command size[",
3844                         (int)sizeof(CommandList_struct),
3845                         "] not divisible by 32, no performant mode..\n");
3846                 return;
3847         }
3848
3849         /* Performant mode ring buffer and supporting data structures */
3850         h->reply_pool = (__u64 *)pci_alloc_consistent(
3851                 h->pdev, h->max_commands * sizeof(__u64),
3852                 &(h->reply_pool_dhandle));
3853
3854         /* Need a block fetch table for performant mode */
3855         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3856                 sizeof(__u32)), GFP_KERNEL);
3857
3858         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3859                 goto clean_up;
3860
3861         cciss_enter_performant_mode(h);
3862
3863         /* Change the access methods to the performant access methods */
3864         h->access = SA5_performant_access;
3865         h->transMethod = CFGTBL_Trans_Performant;
3866
3867         return;
3868 clean_up:
3869         kfree(h->blockFetchTable);
3870         if (h->reply_pool)
3871                 pci_free_consistent(h->pdev,
3872                                 h->max_commands * sizeof(__u64),
3873                                 h->reply_pool,
3874                                 h->reply_pool_dhandle);
3875         return;
3876
3877 } /* cciss_put_controller_into_performant_mode */
3878
3879 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3880  * controllers that are capable. If not, we use IO-APIC mode.
3881  */
3882
3883 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3884 {
3885 #ifdef CONFIG_PCI_MSI
3886         int err;
3887         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3888         {0, 2}, {0, 3}
3889         };
3890
3891         /* Some boards advertise MSI but don't really support it */
3892         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3893             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3894                 goto default_int_mode;
3895
3896         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3897                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3898                 if (!err) {
3899                         h->intr[0] = cciss_msix_entries[0].vector;
3900                         h->intr[1] = cciss_msix_entries[1].vector;
3901                         h->intr[2] = cciss_msix_entries[2].vector;
3902                         h->intr[3] = cciss_msix_entries[3].vector;
3903                         h->msix_vector = 1;
3904                         return;
3905                 }
3906                 if (err > 0) {
3907                         dev_warn(&h->pdev->dev,
3908                                 "only %d MSI-X vectors available\n", err);
3909                         goto default_int_mode;
3910                 } else {
3911                         dev_warn(&h->pdev->dev,
3912                                 "MSI-X init failed %d\n", err);
3913                         goto default_int_mode;
3914                 }
3915         }
3916         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3917                 if (!pci_enable_msi(h->pdev))
3918                         h->msi_vector = 1;
3919                 else
3920                         dev_warn(&h->pdev->dev, "MSI init failed\n");
3921         }
3922 default_int_mode:
3923 #endif                          /* CONFIG_PCI_MSI */
3924         /* if we get here we're going to use the default interrupt mode */
3925         h->intr[PERF_MODE_INT] = h->pdev->irq;
3926         return;
3927 }
3928
3929 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3930 {
3931         int i;
3932         u32 subsystem_vendor_id, subsystem_device_id;
3933
3934         subsystem_vendor_id = pdev->subsystem_vendor;
3935         subsystem_device_id = pdev->subsystem_device;
3936         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3937                         subsystem_vendor_id;
3938
3939         for (i = 0; i < ARRAY_SIZE(products); i++)
3940                 if (*board_id == products[i].board_id)
3941                         return i;
3942         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
3943                 *board_id);
3944         return -ENODEV;
3945 }
3946
3947 static inline bool cciss_board_disabled(ctlr_info_t *h)
3948 {
3949         u16 command;
3950
3951         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
3952         return ((command & PCI_COMMAND_MEMORY) == 0);
3953 }
3954
3955 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
3956         unsigned long *memory_bar)
3957 {
3958         int i;
3959
3960         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
3961                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
3962                         /* addressing mode bits already removed */
3963                         *memory_bar = pci_resource_start(pdev, i);
3964                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
3965                                 *memory_bar);
3966                         return 0;
3967                 }
3968         dev_warn(&pdev->dev, "no memory BAR found\n");
3969         return -ENODEV;
3970 }
3971
3972 static int __devinit cciss_wait_for_board_state(struct pci_dev *pdev,
3973         void __iomem *vaddr, int wait_for_ready)
3974 #define BOARD_READY 1
3975 #define BOARD_NOT_READY 0
3976 {
3977         int i, iterations;
3978         u32 scratchpad;
3979
3980         if (wait_for_ready)
3981                 iterations = CCISS_BOARD_READY_ITERATIONS;
3982         else
3983                 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
3984
3985         for (i = 0; i < iterations; i++) {
3986                 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
3987                 if (wait_for_ready) {
3988                         if (scratchpad == CCISS_FIRMWARE_READY)
3989                                 return 0;
3990                 } else {
3991                         if (scratchpad != CCISS_FIRMWARE_READY)
3992                                 return 0;
3993                 }
3994                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
3995         }
3996         dev_warn(&pdev->dev, "board not ready, timed out.\n");
3997         return -ENODEV;
3998 }
3999
4000 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4001         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4002         u64 *cfg_offset)
4003 {
4004         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4005         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4006         *cfg_base_addr &= (u32) 0x0000ffff;
4007         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4008         if (*cfg_base_addr_index == -1) {
4009                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4010                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4011                 return -ENODEV;
4012         }
4013         return 0;
4014 }
4015
4016 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4017 {
4018         u64 cfg_offset;
4019         u32 cfg_base_addr;
4020         u64 cfg_base_addr_index;
4021         u32 trans_offset;
4022         int rc;
4023
4024         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4025                 &cfg_base_addr_index, &cfg_offset);
4026         if (rc)
4027                 return rc;
4028         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4029                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4030         if (!h->cfgtable)
4031                 return -ENOMEM;
4032         /* Find performant mode table. */
4033         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4034         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4035                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4036                                 sizeof(*h->transtable));
4037         if (!h->transtable)
4038                 return -ENOMEM;
4039         return 0;
4040 }
4041
4042 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4043 {
4044         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4045
4046         /* Limit commands in memory limited kdump scenario. */
4047         if (reset_devices && h->max_commands > 32)
4048                 h->max_commands = 32;
4049
4050         if (h->max_commands < 16) {
4051                 dev_warn(&h->pdev->dev, "Controller reports "
4052                         "max supported commands of %d, an obvious lie. "
4053                         "Using 16.  Ensure that firmware is up to date.\n",
4054                         h->max_commands);
4055                 h->max_commands = 16;
4056         }
4057 }
4058
4059 /* Interrogate the hardware for some limits:
4060  * max commands, max SG elements without chaining, and with chaining,
4061  * SG chain block size, etc.
4062  */
4063 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4064 {
4065         cciss_get_max_perf_mode_cmds(h);
4066         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4067         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4068         /*
4069          * Limit in-command s/g elements to 32 save dma'able memory.
4070          * Howvever spec says if 0, use 31
4071          */
4072         h->max_cmd_sgentries = 31;
4073         if (h->maxsgentries > 512) {
4074                 h->max_cmd_sgentries = 32;
4075                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4076                 h->maxsgentries--; /* save one for chain pointer */
4077         } else {
4078                 h->maxsgentries = 31; /* default to traditional values */
4079                 h->chainsize = 0;
4080         }
4081 }
4082
4083 static inline bool CISS_signature_present(ctlr_info_t *h)
4084 {
4085         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4086             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4087             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4088             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4089                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4090                 return false;
4091         }
4092         return true;
4093 }
4094
4095 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4096 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4097 {
4098 #ifdef CONFIG_X86
4099         u32 prefetch;
4100
4101         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4102         prefetch |= 0x100;
4103         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4104 #endif
4105 }
4106
4107 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4108  * in a prefetch beyond physical memory.
4109  */
4110 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4111 {
4112         u32 dma_prefetch;
4113         __u32 dma_refetch;
4114
4115         if (h->board_id != 0x3225103C)
4116                 return;
4117         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4118         dma_prefetch |= 0x8000;
4119         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4120         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4121         dma_refetch |= 0x1;
4122         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4123 }
4124
4125 static int __devinit cciss_pci_init(ctlr_info_t *h)
4126 {
4127         int prod_index, err;
4128
4129         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4130         if (prod_index < 0)
4131                 return -ENODEV;
4132         h->product_name = products[prod_index].product_name;
4133         h->access = *(products[prod_index].access);
4134
4135         if (cciss_board_disabled(h)) {
4136                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4137                 return -ENODEV;
4138         }
4139         err = pci_enable_device(h->pdev);
4140         if (err) {
4141                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4142                 return err;
4143         }
4144
4145         err = pci_request_regions(h->pdev, "cciss");
4146         if (err) {
4147                 dev_warn(&h->pdev->dev,
4148                         "Cannot obtain PCI resources, aborting\n");
4149                 return err;
4150         }
4151
4152         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4153         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4154
4155 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4156  * else we use the IO-APIC interrupt assigned to us by system ROM.
4157  */
4158         cciss_interrupt_mode(h);
4159         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4160         if (err)
4161                 goto err_out_free_res;
4162         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4163         if (!h->vaddr) {
4164                 err = -ENOMEM;
4165                 goto err_out_free_res;
4166         }
4167         err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4168         if (err)
4169                 goto err_out_free_res;
4170         err = cciss_find_cfgtables(h);
4171         if (err)
4172                 goto err_out_free_res;
4173         print_cfg_table(h);
4174         cciss_find_board_params(h);
4175
4176         if (!CISS_signature_present(h)) {
4177                 err = -ENODEV;
4178                 goto err_out_free_res;
4179         }
4180         cciss_enable_scsi_prefetch(h);
4181         cciss_p600_dma_prefetch_quirk(h);
4182         cciss_put_controller_into_performant_mode(h);
4183         return 0;
4184
4185 err_out_free_res:
4186         /*
4187          * Deliberately omit pci_disable_device(): it does something nasty to
4188          * Smart Array controllers that pci_enable_device does not undo
4189          */
4190         if (h->transtable)
4191                 iounmap(h->transtable);
4192         if (h->cfgtable)
4193                 iounmap(h->cfgtable);
4194         if (h->vaddr)
4195                 iounmap(h->vaddr);
4196         pci_release_regions(h->pdev);
4197         return err;
4198 }
4199
4200 /* Function to find the first free pointer into our hba[] array
4201  * Returns -1 if no free entries are left.
4202  */
4203 static int alloc_cciss_hba(struct pci_dev *pdev)
4204 {
4205         int i;
4206
4207         for (i = 0; i < MAX_CTLR; i++) {
4208                 if (!hba[i]) {
4209                         ctlr_info_t *h;
4210
4211                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4212                         if (!h)
4213                                 goto Enomem;
4214                         hba[i] = h;
4215                         return i;
4216                 }
4217         }
4218         dev_warn(&pdev->dev, "This driver supports a maximum"
4219                " of %d controllers.\n", MAX_CTLR);
4220         return -1;
4221 Enomem:
4222         dev_warn(&pdev->dev, "out of memory.\n");
4223         return -1;
4224 }
4225
4226 static void free_hba(ctlr_info_t *h)
4227 {
4228         int i;
4229
4230         hba[h->ctlr] = NULL;
4231         for (i = 0; i < h->highest_lun + 1; i++)
4232                 if (h->gendisk[i] != NULL)
4233                         put_disk(h->gendisk[i]);
4234         kfree(h);
4235 }
4236
4237 /* Send a message CDB to the firmware. */
4238 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4239 {
4240         typedef struct {
4241                 CommandListHeader_struct CommandHeader;
4242                 RequestBlock_struct Request;
4243                 ErrDescriptor_struct ErrorDescriptor;
4244         } Command;
4245         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4246         Command *cmd;
4247         dma_addr_t paddr64;
4248         uint32_t paddr32, tag;
4249         void __iomem *vaddr;
4250         int i, err;
4251
4252         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4253         if (vaddr == NULL)
4254                 return -ENOMEM;
4255
4256         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4257            CCISS commands, so they must be allocated from the lower 4GiB of
4258            memory. */
4259         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4260         if (err) {
4261                 iounmap(vaddr);
4262                 return -ENOMEM;
4263         }
4264
4265         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4266         if (cmd == NULL) {
4267                 iounmap(vaddr);
4268                 return -ENOMEM;
4269         }
4270
4271         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4272            although there's no guarantee, we assume that the address is at
4273            least 4-byte aligned (most likely, it's page-aligned). */
4274         paddr32 = paddr64;
4275
4276         cmd->CommandHeader.ReplyQueue = 0;
4277         cmd->CommandHeader.SGList = 0;
4278         cmd->CommandHeader.SGTotal = 0;
4279         cmd->CommandHeader.Tag.lower = paddr32;
4280         cmd->CommandHeader.Tag.upper = 0;
4281         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4282
4283         cmd->Request.CDBLen = 16;
4284         cmd->Request.Type.Type = TYPE_MSG;
4285         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4286         cmd->Request.Type.Direction = XFER_NONE;
4287         cmd->Request.Timeout = 0; /* Don't time out */
4288         cmd->Request.CDB[0] = opcode;
4289         cmd->Request.CDB[1] = type;
4290         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4291
4292         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4293         cmd->ErrorDescriptor.Addr.upper = 0;
4294         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4295
4296         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4297
4298         for (i = 0; i < 10; i++) {
4299                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4300                 if ((tag & ~3) == paddr32)
4301                         break;
4302                 schedule_timeout_uninterruptible(HZ);
4303         }
4304
4305         iounmap(vaddr);
4306
4307         /* we leak the DMA buffer here ... no choice since the controller could
4308            still complete the command. */
4309         if (i == 10) {
4310                 dev_err(&pdev->dev,
4311                         "controller message %02x:%02x timed out\n",
4312                         opcode, type);
4313                 return -ETIMEDOUT;
4314         }
4315
4316         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4317
4318         if (tag & 2) {
4319                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4320                         opcode, type);
4321                 return -EIO;
4322         }
4323
4324         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4325                 opcode, type);
4326         return 0;
4327 }
4328
4329 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4330 #define cciss_noop(p) cciss_message(p, 3, 0)
4331
4332 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4333         void * __iomem vaddr, bool use_doorbell)
4334 {
4335         u16 pmcsr;
4336         int pos;
4337
4338         if (use_doorbell) {
4339                 /* For everything after the P600, the PCI power state method
4340                  * of resetting the controller doesn't work, so we have this
4341                  * other way using the doorbell register.
4342                  */
4343                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4344                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4345                 msleep(1000);
4346         } else { /* Try to do it the PCI power state way */
4347
4348                 /* Quoting from the Open CISS Specification: "The Power
4349                  * Management Control/Status Register (CSR) controls the power
4350                  * state of the device.  The normal operating state is D0,
4351                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4352                  * the controller, place the interface device in D3 then to D0,
4353                  * this causes a secondary PCI reset which will reset the
4354                  * controller." */
4355
4356                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4357                 if (pos == 0) {
4358                         dev_err(&pdev->dev,
4359                                 "cciss_controller_hard_reset: "
4360                                 "PCI PM not supported\n");
4361                         return -ENODEV;
4362                 }
4363                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4364                 /* enter the D3hot power management state */
4365                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4366                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4367                 pmcsr |= PCI_D3hot;
4368                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4369
4370                 msleep(500);
4371
4372                 /* enter the D0 power management state */
4373                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4374                 pmcsr |= PCI_D0;
4375                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4376
4377                 msleep(500);
4378         }
4379         return 0;
4380 }
4381
4382 /* This does a hard reset of the controller using PCI power management
4383  * states or using the doorbell register. */
4384 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4385 {
4386         u64 cfg_offset;
4387         u32 cfg_base_addr;
4388         u64 cfg_base_addr_index;
4389         void __iomem *vaddr;
4390         unsigned long paddr;
4391         u32 misc_fw_support, active_transport;
4392         int rc;
4393         CfgTable_struct __iomem *cfgtable;
4394         bool use_doorbell;
4395         u32 board_id;
4396         u16 command_register;
4397
4398         /* For controllers as old a the p600, this is very nearly
4399          * the same thing as
4400          *
4401          * pci_save_state(pci_dev);
4402          * pci_set_power_state(pci_dev, PCI_D3hot);
4403          * pci_set_power_state(pci_dev, PCI_D0);
4404          * pci_restore_state(pci_dev);
4405          *
4406          * For controllers newer than the P600, the pci power state
4407          * method of resetting doesn't work so we have another way
4408          * using the doorbell register.
4409          */
4410
4411         /* Exclude 640x boards.  These are two pci devices in one slot
4412          * which share a battery backed cache module.  One controls the
4413          * cache, the other accesses the cache through the one that controls
4414          * it.  If we reset the one controlling the cache, the other will
4415          * likely not be happy.  Just forbid resetting this conjoined mess.
4416          */
4417         cciss_lookup_board_id(pdev, &board_id);
4418         if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4419                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4420                                 "due to shared cache module.");
4421                 return -ENODEV;
4422         }
4423
4424         /* Save the PCI command register */
4425         pci_read_config_word(pdev, 4, &command_register);
4426         /* Turn the board off.  This is so that later pci_restore_state()
4427          * won't turn the board on before the rest of config space is ready.
4428          */
4429         pci_disable_device(pdev);
4430         pci_save_state(pdev);
4431
4432         /* find the first memory BAR, so we can find the cfg table */
4433         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4434         if (rc)
4435                 return rc;
4436         vaddr = remap_pci_mem(paddr, 0x250);
4437         if (!vaddr)
4438                 return -ENOMEM;
4439
4440         /* find cfgtable in order to check if reset via doorbell is supported */
4441         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4442                                         &cfg_base_addr_index, &cfg_offset);
4443         if (rc)
4444                 goto unmap_vaddr;
4445         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4446                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4447         if (!cfgtable) {
4448                 rc = -ENOMEM;
4449                 goto unmap_vaddr;
4450         }
4451
4452         /* If reset via doorbell register is supported, use that. */
4453         misc_fw_support = readl(&cfgtable->misc_fw_support);
4454         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4455
4456         /* The doorbell reset seems to cause lockups on some Smart
4457          * Arrays (e.g. P410, P410i, maybe others).  Until this is
4458          * fixed or at least isolated, avoid the doorbell reset.
4459          */
4460         use_doorbell = 0;
4461
4462         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4463         if (rc)
4464                 goto unmap_cfgtable;
4465         pci_restore_state(pdev);
4466         rc = pci_enable_device(pdev);
4467         if (rc) {
4468                 dev_warn(&pdev->dev, "failed to enable device.\n");
4469                 goto unmap_cfgtable;
4470         }
4471         pci_write_config_word(pdev, 4, command_register);
4472
4473         /* Some devices (notably the HP Smart Array 5i Controller)
4474            need a little pause here */
4475         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4476
4477         /* Wait for board to become not ready, then ready. */
4478         dev_info(&pdev->dev, "Waiting for board to become ready.\n");
4479         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4480         if (rc) /* Don't bail, might be E500, etc. which can't be reset */
4481                 dev_warn(&pdev->dev,
4482                         "failed waiting for board to become not ready\n");
4483         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4484         if (rc) {
4485                 dev_warn(&pdev->dev,
4486                         "failed waiting for board to become ready\n");
4487                 goto unmap_cfgtable;
4488         }
4489         dev_info(&pdev->dev, "board ready.\n");
4490
4491         /* Controller should be in simple mode at this point.  If it's not,
4492          * It means we're on one of those controllers which doesn't support
4493          * the doorbell reset method and on which the PCI power management reset
4494          * method doesn't work (P800, for example.)
4495          * In those cases, don't try to proceed, as it generally doesn't work.
4496          */
4497         active_transport = readl(&cfgtable->TransportActive);
4498         if (active_transport & PERFORMANT_MODE) {
4499                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4500                         " Ignoring controller.\n");
4501                 rc = -ENODEV;
4502         }
4503
4504 unmap_cfgtable:
4505         iounmap(cfgtable);
4506
4507 unmap_vaddr:
4508         iounmap(vaddr);
4509         return rc;
4510 }
4511
4512 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4513 {
4514         int rc, i;
4515
4516         if (!reset_devices)
4517                 return 0;
4518
4519         /* Reset the controller with a PCI power-cycle or via doorbell */
4520         rc = cciss_kdump_hard_reset_controller(pdev);
4521
4522         /* -ENOTSUPP here means we cannot reset the controller
4523          * but it's already (and still) up and running in
4524          * "performant mode".  Or, it might be 640x, which can't reset
4525          * due to concerns about shared bbwc between 6402/6404 pair.
4526          */
4527         if (rc == -ENOTSUPP)
4528                 return 0; /* just try to do the kdump anyhow. */
4529         if (rc)
4530                 return -ENODEV;
4531
4532         /* Now try to get the controller to respond to a no-op */
4533         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4534                 if (cciss_noop(pdev) == 0)
4535                         break;
4536                 else
4537                         dev_warn(&pdev->dev, "no-op failed%s\n",
4538                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4539                                         "; re-trying" : ""));
4540                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4541         }
4542         return 0;
4543 }
4544
4545 /*
4546  *  This is it.  Find all the controllers and register them.  I really hate
4547  *  stealing all these major device numbers.
4548  *  returns the number of block devices registered.
4549  */
4550 static int __devinit cciss_init_one(struct pci_dev *pdev,
4551                                     const struct pci_device_id *ent)
4552 {
4553         int i;
4554         int j = 0;
4555         int k = 0;
4556         int rc;
4557         int dac, return_code;
4558         InquiryData_struct *inq_buff;
4559         ctlr_info_t *h;
4560
4561         rc = cciss_init_reset_devices(pdev);
4562         if (rc)
4563                 return rc;
4564         i = alloc_cciss_hba(pdev);
4565         if (i < 0)
4566                 return -1;
4567
4568         h = hba[i];
4569         h->pdev = pdev;
4570         h->busy_initializing = 1;
4571         INIT_HLIST_HEAD(&h->cmpQ);
4572         INIT_HLIST_HEAD(&h->reqQ);
4573         mutex_init(&h->busy_shutting_down);
4574
4575         if (cciss_pci_init(h) != 0)
4576                 goto clean_no_release_regions;
4577
4578         sprintf(h->devname, "cciss%d", i);
4579         h->ctlr = i;
4580
4581         init_completion(&h->scan_wait);
4582
4583         if (cciss_create_hba_sysfs_entry(h))
4584                 goto clean0;
4585
4586         /* configure PCI DMA stuff */
4587         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4588                 dac = 1;
4589         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4590                 dac = 0;
4591         else {
4592                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4593                 goto clean1;
4594         }
4595
4596         /*
4597          * register with the major number, or get a dynamic major number
4598          * by passing 0 as argument.  This is done for greater than
4599          * 8 controller support.
4600          */
4601         if (i < MAX_CTLR_ORIG)
4602                 h->major = COMPAQ_CISS_MAJOR + i;
4603         rc = register_blkdev(h->major, h->devname);
4604         if (rc == -EBUSY || rc == -EINVAL) {
4605                 dev_err(&h->pdev->dev,
4606                        "Unable to get major number %d for %s "
4607                        "on hba %d\n", h->major, h->devname, i);
4608                 goto clean1;
4609         } else {
4610                 if (i >= MAX_CTLR_ORIG)
4611                         h->major = rc;
4612         }
4613
4614         /* make sure the board interrupts are off */
4615         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4616         if (h->msi_vector || h->msix_vector) {
4617                 if (request_irq(h->intr[PERF_MODE_INT],
4618                                 do_cciss_msix_intr,
4619                                 IRQF_DISABLED, h->devname, h)) {
4620                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4621                                h->intr[PERF_MODE_INT], h->devname);
4622                         goto clean2;
4623                 }
4624         } else {
4625                 if (request_irq(h->intr[PERF_MODE_INT], do_cciss_intx,
4626                                 IRQF_DISABLED, h->devname, h)) {
4627                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4628                                h->intr[PERF_MODE_INT], h->devname);
4629                         goto clean2;
4630                 }
4631         }
4632
4633         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4634                h->devname, pdev->device, pci_name(pdev),
4635                h->intr[PERF_MODE_INT], dac ? "" : " not");
4636
4637         h->cmd_pool_bits =
4638             kmalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4639                         * sizeof(unsigned long), GFP_KERNEL);
4640         h->cmd_pool = (CommandList_struct *)
4641             pci_alloc_consistent(h->pdev,
4642                     h->nr_cmds * sizeof(CommandList_struct),
4643                     &(h->cmd_pool_dhandle));
4644         h->errinfo_pool = (ErrorInfo_struct *)
4645             pci_alloc_consistent(h->pdev,
4646                     h->nr_cmds * sizeof(ErrorInfo_struct),
4647                     &(h->errinfo_pool_dhandle));
4648         if ((h->cmd_pool_bits == NULL)
4649             || (h->cmd_pool == NULL)
4650             || (h->errinfo_pool == NULL)) {
4651                 dev_err(&h->pdev->dev, "out of memory");
4652                 goto clean4;
4653         }
4654
4655         /* Need space for temp scatter list */
4656         h->scatter_list = kmalloc(h->max_commands *
4657                                                 sizeof(struct scatterlist *),
4658                                                 GFP_KERNEL);
4659         if (!h->scatter_list)
4660                 goto clean4;
4661
4662         for (k = 0; k < h->nr_cmds; k++) {
4663                 h->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4664                                                         h->maxsgentries,
4665                                                         GFP_KERNEL);
4666                 if (h->scatter_list[k] == NULL) {
4667                         dev_err(&h->pdev->dev,
4668                                 "could not allocate s/g lists\n");
4669                         goto clean4;
4670                 }
4671         }
4672         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4673                 h->chainsize, h->nr_cmds);
4674         if (!h->cmd_sg_list && h->chainsize > 0)
4675                 goto clean4;
4676
4677         spin_lock_init(&h->lock);
4678
4679         /* Initialize the pdev driver private data.
4680            have it point to h.  */
4681         pci_set_drvdata(pdev, h);
4682         /* command and error info recs zeroed out before
4683            they are used */
4684         memset(h->cmd_pool_bits, 0,
4685                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4686                         * sizeof(unsigned long));
4687
4688         h->num_luns = 0;
4689         h->highest_lun = -1;
4690         for (j = 0; j < CISS_MAX_LUN; j++) {
4691                 h->drv[j] = NULL;
4692                 h->gendisk[j] = NULL;
4693         }
4694
4695         cciss_scsi_setup(h);
4696
4697         /* Turn the interrupts on so we can service requests */
4698         h->access.set_intr_mask(h, CCISS_INTR_ON);
4699
4700         /* Get the firmware version */
4701         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4702         if (inq_buff == NULL) {
4703                 dev_err(&h->pdev->dev, "out of memory\n");
4704                 goto clean4;
4705         }
4706
4707         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4708                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4709         if (return_code == IO_OK) {
4710                 h->firm_ver[0] = inq_buff->data_byte[32];
4711                 h->firm_ver[1] = inq_buff->data_byte[33];
4712                 h->firm_ver[2] = inq_buff->data_byte[34];
4713                 h->firm_ver[3] = inq_buff->data_byte[35];
4714         } else {         /* send command failed */
4715                 dev_warn(&h->pdev->dev, "unable to determine firmware"
4716                         " version of controller\n");
4717         }
4718         kfree(inq_buff);
4719
4720         cciss_procinit(h);
4721
4722         h->cciss_max_sectors = 8192;
4723
4724         rebuild_lun_table(h, 1, 0);
4725         h->busy_initializing = 0;
4726         return 1;
4727
4728 clean4:
4729         kfree(h->cmd_pool_bits);
4730         /* Free up sg elements */
4731         for (k-- ; k >= 0; k--)
4732                 kfree(h->scatter_list[k]);
4733         kfree(h->scatter_list);
4734         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4735         if (h->cmd_pool)
4736                 pci_free_consistent(h->pdev,
4737                                     h->nr_cmds * sizeof(CommandList_struct),
4738                                     h->cmd_pool, h->cmd_pool_dhandle);
4739         if (h->errinfo_pool)
4740                 pci_free_consistent(h->pdev,
4741                                     h->nr_cmds * sizeof(ErrorInfo_struct),
4742                                     h->errinfo_pool,
4743                                     h->errinfo_pool_dhandle);
4744         free_irq(h->intr[PERF_MODE_INT], h);
4745 clean2:
4746         unregister_blkdev(h->major, h->devname);
4747 clean1:
4748         cciss_destroy_hba_sysfs_entry(h);
4749 clean0:
4750         pci_release_regions(pdev);
4751 clean_no_release_regions:
4752         h->busy_initializing = 0;
4753
4754         /*
4755          * Deliberately omit pci_disable_device(): it does something nasty to
4756          * Smart Array controllers that pci_enable_device does not undo
4757          */
4758         pci_set_drvdata(pdev, NULL);
4759         free_hba(h);
4760         return -1;
4761 }
4762
4763 static void cciss_shutdown(struct pci_dev *pdev)
4764 {
4765         ctlr_info_t *h;
4766         char *flush_buf;
4767         int return_code;
4768
4769         h = pci_get_drvdata(pdev);
4770         flush_buf = kzalloc(4, GFP_KERNEL);
4771         if (!flush_buf) {
4772                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
4773                 return;
4774         }
4775         /* write all data in the battery backed cache to disk */
4776         memset(flush_buf, 0, 4);
4777         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4778                 4, 0, CTLR_LUNID, TYPE_CMD);
4779         kfree(flush_buf);
4780         if (return_code != IO_OK)
4781                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
4782         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4783         free_irq(h->intr[PERF_MODE_INT], h);
4784 }
4785
4786 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4787 {
4788         ctlr_info_t *h;
4789         int i, j;
4790
4791         if (pci_get_drvdata(pdev) == NULL) {
4792                 dev_err(&pdev->dev, "Unable to remove device\n");
4793                 return;
4794         }
4795
4796         h = pci_get_drvdata(pdev);
4797         i = h->ctlr;
4798         if (hba[i] == NULL) {
4799                 dev_err(&pdev->dev, "device appears to already be removed\n");
4800                 return;
4801         }
4802
4803         mutex_lock(&h->busy_shutting_down);
4804
4805         remove_from_scan_list(h);
4806         remove_proc_entry(h->devname, proc_cciss);
4807         unregister_blkdev(h->major, h->devname);
4808
4809         /* remove it from the disk list */
4810         for (j = 0; j < CISS_MAX_LUN; j++) {
4811                 struct gendisk *disk = h->gendisk[j];
4812                 if (disk) {
4813                         struct request_queue *q = disk->queue;
4814
4815                         if (disk->flags & GENHD_FL_UP) {
4816                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
4817                                 del_gendisk(disk);
4818                         }
4819                         if (q)
4820                                 blk_cleanup_queue(q);
4821                 }
4822         }
4823
4824 #ifdef CONFIG_CISS_SCSI_TAPE
4825         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
4826 #endif
4827
4828         cciss_shutdown(pdev);
4829
4830 #ifdef CONFIG_PCI_MSI
4831         if (h->msix_vector)
4832                 pci_disable_msix(h->pdev);
4833         else if (h->msi_vector)
4834                 pci_disable_msi(h->pdev);
4835 #endif                          /* CONFIG_PCI_MSI */
4836
4837         iounmap(h->transtable);
4838         iounmap(h->cfgtable);
4839         iounmap(h->vaddr);
4840
4841         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(CommandList_struct),
4842                             h->cmd_pool, h->cmd_pool_dhandle);
4843         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(ErrorInfo_struct),
4844                             h->errinfo_pool, h->errinfo_pool_dhandle);
4845         kfree(h->cmd_pool_bits);
4846         /* Free up sg elements */
4847         for (j = 0; j < h->nr_cmds; j++)
4848                 kfree(h->scatter_list[j]);
4849         kfree(h->scatter_list);
4850         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4851         /*
4852          * Deliberately omit pci_disable_device(): it does something nasty to
4853          * Smart Array controllers that pci_enable_device does not undo
4854          */
4855         pci_release_regions(pdev);
4856         pci_set_drvdata(pdev, NULL);
4857         cciss_destroy_hba_sysfs_entry(h);
4858         mutex_unlock(&h->busy_shutting_down);
4859         free_hba(h);
4860 }
4861
4862 static struct pci_driver cciss_pci_driver = {
4863         .name = "cciss",
4864         .probe = cciss_init_one,
4865         .remove = __devexit_p(cciss_remove_one),
4866         .id_table = cciss_pci_device_id,        /* id_table */
4867         .shutdown = cciss_shutdown,
4868 };
4869
4870 /*
4871  *  This is it.  Register the PCI driver information for the cards we control
4872  *  the OS will call our registered routines when it finds one of our cards.
4873  */
4874 static int __init cciss_init(void)
4875 {
4876         int err;
4877
4878         /*
4879          * The hardware requires that commands are aligned on a 64-bit
4880          * boundary. Given that we use pci_alloc_consistent() to allocate an
4881          * array of them, the size must be a multiple of 8 bytes.
4882          */
4883         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4884         printk(KERN_INFO DRIVER_NAME "\n");
4885
4886         err = bus_register(&cciss_bus_type);
4887         if (err)
4888                 return err;
4889
4890         /* Start the scan thread */
4891         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4892         if (IS_ERR(cciss_scan_thread)) {
4893                 err = PTR_ERR(cciss_scan_thread);
4894                 goto err_bus_unregister;
4895         }
4896
4897         /* Register for our PCI devices */
4898         err = pci_register_driver(&cciss_pci_driver);
4899         if (err)
4900                 goto err_thread_stop;
4901
4902         return err;
4903
4904 err_thread_stop:
4905         kthread_stop(cciss_scan_thread);
4906 err_bus_unregister:
4907         bus_unregister(&cciss_bus_type);
4908
4909         return err;
4910 }
4911
4912 static void __exit cciss_cleanup(void)
4913 {
4914         int i;
4915
4916         pci_unregister_driver(&cciss_pci_driver);
4917         /* double check that all controller entrys have been removed */
4918         for (i = 0; i < MAX_CTLR; i++) {
4919                 if (hba[i] != NULL) {
4920                         dev_warn(&hba[i]->pdev->dev,
4921                                 "had to remove controller\n");
4922                         cciss_remove_one(hba[i]->pdev);
4923                 }
4924         }
4925         kthread_stop(cciss_scan_thread);
4926         remove_proc_entry("driver/cciss", NULL);
4927         bus_unregister(&cciss_bus_type);
4928 }
4929
4930 module_init(cciss_init);
4931 module_exit(cciss_cleanup);