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/smp_lock.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/compat.h>
  43 #include <linux/mutex.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/io.h>
  46
  47 #include <linux/dma-mapping.h>
  48 #include <linux/blkdev.h>
  49 #include <linux/genhd.h>
  50 #include <linux/completion.h>
  51 #include <scsi/scsi.h>
  52 #include <scsi/sg.h>
  53 #include <scsi/scsi_ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/scatterlist.h>
  56 #include <linux/kthread.h>
  57
  58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  59 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  61
  62 /* Embedded module documentation macros - see modules.h */
  63 MODULE_AUTHOR("Hewlett-Packard Company");
  64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  65 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  66 MODULE_VERSION("3.6.26");
  67 MODULE_LICENSE("GPL");
  68
  69 static int cciss_allow_hpsa;
  70 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  71 MODULE_PARM_DESC(cciss_allow_hpsa,
  72         "Prevent cciss driver from accessing hardware known to be "
  73         " supported by the hpsa driver");
  74
  75 #include "cciss_cmd.h"
  76 #include "cciss.h"
  77 #include <linux/cciss_ioctl.h>
  78
  79 /* define the PCI info for the cards we can control */
  80 static const struct pci_device_id cciss_pci_device_id[] = {
  81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3250},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3251},
 110         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3252},
 111         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3253},
 112         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3254},
 113         {0,}
 114 };
 115
 116 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 117
 118 /*  board_id = Subsystem Device ID & Vendor ID
 119  *  product = Marketing Name for the board
 120  *  access = Address of the struct of function pointers
 121  */
 122 static struct board_type products[] = {
 123         {0x40700E11, "Smart Array 5300", &SA5_access},
 124         {0x40800E11, "Smart Array 5i", &SA5B_access},
 125         {0x40820E11, "Smart Array 532", &SA5B_access},
 126         {0x40830E11, "Smart Array 5312", &SA5B_access},
 127         {0x409A0E11, "Smart Array 641", &SA5_access},
 128         {0x409B0E11, "Smart Array 642", &SA5_access},
 129         {0x409C0E11, "Smart Array 6400", &SA5_access},
 130         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 131         {0x40910E11, "Smart Array 6i", &SA5_access},
 132         {0x3225103C, "Smart Array P600", &SA5_access},
 133         {0x3235103C, "Smart Array P400i", &SA5_access},
 134         {0x3211103C, "Smart Array E200i", &SA5_access},
 135         {0x3212103C, "Smart Array E200", &SA5_access},
 136         {0x3213103C, "Smart Array E200i", &SA5_access},
 137         {0x3214103C, "Smart Array E200i", &SA5_access},
 138         {0x3215103C, "Smart Array E200i", &SA5_access},
 139         {0x3237103C, "Smart Array E500", &SA5_access},
 140 /* controllers below this line are also supported by the hpsa driver. */
 141 #define HPSA_BOUNDARY 0x3223103C
 142         {0x3223103C, "Smart Array P800", &SA5_access},
 143         {0x3234103C, "Smart Array P400", &SA5_access},
 144         {0x323D103C, "Smart Array P700m", &SA5_access},
 145         {0x3241103C, "Smart Array P212", &SA5_access},
 146         {0x3243103C, "Smart Array P410", &SA5_access},
 147         {0x3245103C, "Smart Array P410i", &SA5_access},
 148         {0x3247103C, "Smart Array P411", &SA5_access},
 149         {0x3249103C, "Smart Array P812", &SA5_access},
 150         {0x324A103C, "Smart Array P712m", &SA5_access},
 151         {0x324B103C, "Smart Array P711m", &SA5_access},
 152         {0x3250103C, "Smart Array", &SA5_access},
 153         {0x3251103C, "Smart Array", &SA5_access},
 154         {0x3252103C, "Smart Array", &SA5_access},
 155         {0x3253103C, "Smart Array", &SA5_access},
 156         {0x3254103C, "Smart Array", &SA5_access},
 157 };
 158
 159 /* How long to wait (in milliseconds) for board to go into simple mode */
 160 #define MAX_CONFIG_WAIT 30000
 161 #define MAX_IOCTL_CONFIG_WAIT 1000
 162
 163 /*define how many times we will try a command because of bus resets */
 164 #define MAX_CMD_RETRIES 3
 165
 166 #define MAX_CTLR        32
 167
 168 /* Originally cciss driver only supports 8 major numbers */
 169 #define MAX_CTLR_ORIG   8
 170
 171 static ctlr_info_t *hba[MAX_CTLR];
 172
 173 static struct task_struct *cciss_scan_thread;
 174 static DEFINE_MUTEX(scan_mutex);
 175 static LIST_HEAD(scan_q);
 176
 177 static void do_cciss_request(struct request_queue *q);
 178 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 179 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 180 static int cciss_open(struct block_device *bdev, fmode_t mode);
 181 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 182 static int cciss_release(struct gendisk *disk, fmode_t mode);
 183 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 184                     unsigned int cmd, unsigned long arg);
 185 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 186                        unsigned int cmd, unsigned long arg);
 187 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 188
 189 static int cciss_revalidate(struct gendisk *disk);
 190 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 191 static int deregister_disk(ctlr_info_t *h, int drv_index,
 192                            int clear_all, int via_ioctl);
 193
 194 static void cciss_read_capacity(int ctlr, int logvol,
 195                         sector_t *total_size, unsigned int *block_size);
 196 static void cciss_read_capacity_16(int ctlr, int logvol,
 197                         sector_t *total_size, unsigned int *block_size);
 198 static void cciss_geometry_inquiry(int ctlr, int logvol,
 199                         sector_t total_size,
 200                         unsigned int block_size, InquiryData_struct *inq_buff,
 201                                    drive_info_struct *drv);
 202 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
 203 static void start_io(ctlr_info_t *h);
 204 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 205                         __u8 page_code, unsigned char scsi3addr[],
 206                         int cmd_type);
 207 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 208         int attempt_retry);
 209 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 210
 211 static int add_to_scan_list(struct ctlr_info *h);
 212 static int scan_thread(void *data);
 213 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 214 static void cciss_hba_release(struct device *dev);
 215 static void cciss_device_release(struct device *dev);
 216 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 217 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 218 static inline u32 next_command(ctlr_info_t *h);
 219
 220 /* performant mode helper functions */
 221 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 222                                 int *bucket_map);
 223 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 224
 225 #ifdef CONFIG_PROC_FS
 226 static void cciss_procinit(int i);
 227 #else
 228 static void cciss_procinit(int i)
 229 {
 230 }
 231 #endif                          /* CONFIG_PROC_FS */
 232
 233 #ifdef CONFIG_COMPAT
 234 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 235                               unsigned, unsigned long);
 236 #endif
 237
 238 static const struct block_device_operations cciss_fops = {
 239         .owner = THIS_MODULE,
 240         .open = cciss_unlocked_open,
 241         .release = cciss_release,
 242         .ioctl = do_ioctl,
 243         .getgeo = cciss_getgeo,
 244 #ifdef CONFIG_COMPAT
 245         .compat_ioctl = cciss_compat_ioctl,
 246 #endif
 247         .revalidate_disk = cciss_revalidate,
 248 };
 249
 250 /* set_performant_mode: Modify the tag for cciss performant
 251  * set bit 0 for pull model, bits 3-1 for block fetch
 252  * register number
 253  */
 254 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 255 {
 256         if (likely(h->transMethod == CFGTBL_Trans_Performant))
 257                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 258 }
 259
 260 /*
 261  * Enqueuing and dequeuing functions for cmdlists.
 262  */
 263 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 264 {
 265         hlist_add_head(&c->list, list);
 266 }
 267
 268 static inline void removeQ(CommandList_struct *c)
 269 {
 270         /*
 271          * After kexec/dump some commands might still
 272          * be in flight, which the firmware will try
 273          * to complete. Resetting the firmware doesn't work
 274          * with old fw revisions, so we have to mark
 275          * them off as 'stale' to prevent the driver from
 276          * falling over.
 277          */
 278         if (WARN_ON(hlist_unhashed(&c->list))) {
 279                 c->cmd_type = CMD_MSG_STALE;
 280                 return;
 281         }
 282
 283         hlist_del_init(&c->list);
 284 }
 285
 286 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 287         CommandList_struct *c)
 288 {
 289         unsigned long flags;
 290         set_performant_mode(h, c);
 291         spin_lock_irqsave(&h->lock, flags);
 292         addQ(&h->reqQ, c);
 293         h->Qdepth++;
 294         start_io(h);
 295         spin_unlock_irqrestore(&h->lock, flags);
 296 }
 297
 298 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 299         int nr_cmds)
 300 {
 301         int i;
 302
 303         if (!cmd_sg_list)
 304                 return;
 305         for (i = 0; i < nr_cmds; i++) {
 306                 kfree(cmd_sg_list[i]);
 307                 cmd_sg_list[i] = NULL;
 308         }
 309         kfree(cmd_sg_list);
 310 }
 311
 312 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 313         ctlr_info_t *h, int chainsize, int nr_cmds)
 314 {
 315         int j;
 316         SGDescriptor_struct **cmd_sg_list;
 317
 318         if (chainsize <= 0)
 319                 return NULL;
 320
 321         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 322         if (!cmd_sg_list)
 323                 return NULL;
 324
 325         /* Build up chain blocks for each command */
 326         for (j = 0; j < nr_cmds; j++) {
 327                 /* Need a block of chainsized s/g elements. */
 328                 cmd_sg_list[j] = kmalloc((chainsize *
 329                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 330                 if (!cmd_sg_list[j]) {
 331                         dev_err(&h->pdev->dev, "Cannot get memory "
 332                                 "for s/g chains.\n");
 333                         goto clean;
 334                 }
 335         }
 336         return cmd_sg_list;
 337 clean:
 338         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 339         return NULL;
 340 }
 341
 342 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 343 {
 344         SGDescriptor_struct *chain_sg;
 345         u64bit temp64;
 346
 347         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 348                 return;
 349
 350         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 351         temp64.val32.lower = chain_sg->Addr.lower;
 352         temp64.val32.upper = chain_sg->Addr.upper;
 353         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 354 }
 355
 356 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 357         SGDescriptor_struct *chain_block, int len)
 358 {
 359         SGDescriptor_struct *chain_sg;
 360         u64bit temp64;
 361
 362         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 363         chain_sg->Ext = CCISS_SG_CHAIN;
 364         chain_sg->Len = len;
 365         temp64.val = pci_map_single(h->pdev, chain_block, len,
 366                                 PCI_DMA_TODEVICE);
 367         chain_sg->Addr.lower = temp64.val32.lower;
 368         chain_sg->Addr.upper = temp64.val32.upper;
 369 }
 370
 371 #include "cciss_scsi.c"         /* For SCSI tape support */
 372
 373 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 374         "UNKNOWN"
 375 };
 376 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
 377
 378 #ifdef CONFIG_PROC_FS
 379
 380 /*
 381  * Report information about this controller.
 382  */
 383 #define ENG_GIG 1000000000
 384 #define ENG_GIG_FACTOR (ENG_GIG/512)
 385 #define ENGAGE_SCSI     "engage scsi"
 386
 387 static struct proc_dir_entry *proc_cciss;
 388
 389 static void cciss_seq_show_header(struct seq_file *seq)
 390 {
 391         ctlr_info_t *h = seq->private;
 392
 393         seq_printf(seq, "%s: HP %s Controller\n"
 394                 "Board ID: 0x%08lx\n"
 395                 "Firmware Version: %c%c%c%c\n"
 396                 "IRQ: %d\n"
 397                 "Logical drives: %d\n"
 398                 "Current Q depth: %d\n"
 399                 "Current # commands on controller: %d\n"
 400                 "Max Q depth since init: %d\n"
 401                 "Max # commands on controller since init: %d\n"
 402                 "Max SG entries since init: %d\n",
 403                 h->devname,
 404                 h->product_name,
 405                 (unsigned long)h->board_id,
 406                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 407                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 408                 h->num_luns,
 409                 h->Qdepth, h->commands_outstanding,
 410                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 411
 412 #ifdef CONFIG_CISS_SCSI_TAPE
 413         cciss_seq_tape_report(seq, h->ctlr);
 414 #endif /* CONFIG_CISS_SCSI_TAPE */
 415 }
 416
 417 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 418 {
 419         ctlr_info_t *h = seq->private;
 420         unsigned ctlr = h->ctlr;
 421         unsigned long flags;
 422
 423         /* prevent displaying bogus info during configuration
 424          * or deconfiguration of a logical volume
 425          */
 426         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 427         if (h->busy_configuring) {
 428                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 429                 return ERR_PTR(-EBUSY);
 430         }
 431         h->busy_configuring = 1;
 432         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 433
 434         if (*pos == 0)
 435                 cciss_seq_show_header(seq);
 436
 437         return pos;
 438 }
 439
 440 static int cciss_seq_show(struct seq_file *seq, void *v)
 441 {
 442         sector_t vol_sz, vol_sz_frac;
 443         ctlr_info_t *h = seq->private;
 444         unsigned ctlr = h->ctlr;
 445         loff_t *pos = v;
 446         drive_info_struct *drv = h->drv[*pos];
 447
 448         if (*pos > h->highest_lun)
 449                 return 0;
 450
 451         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 452                 return 0;
 453
 454         if (drv->heads == 0)
 455                 return 0;
 456
 457         vol_sz = drv->nr_blocks;
 458         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 459         vol_sz_frac *= 100;
 460         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 461
 462         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 463                 drv->raid_level = RAID_UNKNOWN;
 464         seq_printf(seq, "cciss/c%dd%d:"
 465                         "\t%4u.%02uGB\tRAID %s\n",
 466                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 467                         raid_label[drv->raid_level]);
 468         return 0;
 469 }
 470
 471 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 472 {
 473         ctlr_info_t *h = seq->private;
 474
 475         if (*pos > h->highest_lun)
 476                 return NULL;
 477         *pos += 1;
 478
 479         return pos;
 480 }
 481
 482 static void cciss_seq_stop(struct seq_file *seq, void *v)
 483 {
 484         ctlr_info_t *h = seq->private;
 485
 486         /* Only reset h->busy_configuring if we succeeded in setting
 487          * it during cciss_seq_start. */
 488         if (v == ERR_PTR(-EBUSY))
 489                 return;
 490
 491         h->busy_configuring = 0;
 492 }
 493
 494 static const struct seq_operations cciss_seq_ops = {
 495         .start = cciss_seq_start,
 496         .show  = cciss_seq_show,
 497         .next  = cciss_seq_next,
 498         .stop  = cciss_seq_stop,
 499 };
 500
 501 static int cciss_seq_open(struct inode *inode, struct file *file)
 502 {
 503         int ret = seq_open(file, &cciss_seq_ops);
 504         struct seq_file *seq = file->private_data;
 505
 506         if (!ret)
 507                 seq->private = PDE(inode)->data;
 508
 509         return ret;
 510 }
 511
 512 static ssize_t
 513 cciss_proc_write(struct file *file, const char __user *buf,
 514                  size_t length, loff_t *ppos)
 515 {
 516         int err;
 517         char *buffer;
 518
 519 #ifndef CONFIG_CISS_SCSI_TAPE
 520         return -EINVAL;
 521 #endif
 522
 523         if (!buf || length > PAGE_SIZE - 1)
 524                 return -EINVAL;
 525
 526         buffer = (char *)__get_free_page(GFP_KERNEL);
 527         if (!buffer)
 528                 return -ENOMEM;
 529
 530         err = -EFAULT;
 531         if (copy_from_user(buffer, buf, length))
 532                 goto out;
 533         buffer[length] = '\0';
 534
 535 #ifdef CONFIG_CISS_SCSI_TAPE
 536         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 537                 struct seq_file *seq = file->private_data;
 538                 ctlr_info_t *h = seq->private;
 539
 540                 err = cciss_engage_scsi(h->ctlr);
 541                 if (err == 0)
 542                         err = length;
 543         } else
 544 #endif /* CONFIG_CISS_SCSI_TAPE */
 545                 err = -EINVAL;
 546         /* might be nice to have "disengage" too, but it's not
 547            safely possible. (only 1 module use count, lock issues.) */
 548
 549 out:
 550         free_page((unsigned long)buffer);
 551         return err;
 552 }
 553
 554 static const struct file_operations cciss_proc_fops = {
 555         .owner   = THIS_MODULE,
 556         .open    = cciss_seq_open,
 557         .read    = seq_read,
 558         .llseek  = seq_lseek,
 559         .release = seq_release,
 560         .write   = cciss_proc_write,
 561 };
 562
 563 static void __devinit cciss_procinit(int i)
 564 {
 565         struct proc_dir_entry *pde;
 566
 567         if (proc_cciss == NULL)
 568                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 569         if (!proc_cciss)
 570                 return;
 571         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 572                                         S_IROTH, proc_cciss,
 573                                         &cciss_proc_fops, hba[i]);
 574 }
 575 #endif                          /* CONFIG_PROC_FS */
 576
 577 #define MAX_PRODUCT_NAME_LEN 19
 578
 579 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 580 #define to_drv(n) container_of(n, drive_info_struct, dev)
 581
 582 static ssize_t host_store_rescan(struct device *dev,
 583                                  struct device_attribute *attr,
 584                                  const char *buf, size_t count)
 585 {
 586         struct ctlr_info *h = to_hba(dev);
 587
 588         add_to_scan_list(h);
 589         wake_up_process(cciss_scan_thread);
 590         wait_for_completion_interruptible(&h->scan_wait);
 591
 592         return count;
 593 }
 594 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 595
 596 static ssize_t dev_show_unique_id(struct device *dev,
 597                                  struct device_attribute *attr,
 598                                  char *buf)
 599 {
 600         drive_info_struct *drv = to_drv(dev);
 601         struct ctlr_info *h = to_hba(drv->dev.parent);
 602         __u8 sn[16];
 603         unsigned long flags;
 604         int ret = 0;
 605
 606         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 607         if (h->busy_configuring)
 608                 ret = -EBUSY;
 609         else
 610                 memcpy(sn, drv->serial_no, sizeof(sn));
 611         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 612
 613         if (ret)
 614                 return ret;
 615         else
 616                 return snprintf(buf, 16 * 2 + 2,
 617                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 618                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 619                                 sn[0], sn[1], sn[2], sn[3],
 620                                 sn[4], sn[5], sn[6], sn[7],
 621                                 sn[8], sn[9], sn[10], sn[11],
 622                                 sn[12], sn[13], sn[14], sn[15]);
 623 }
 624 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 625
 626 static ssize_t dev_show_vendor(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 vendor[VENDOR_LEN + 1];
 633         unsigned long flags;
 634         int ret = 0;
 635
 636         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 637         if (h->busy_configuring)
 638                 ret = -EBUSY;
 639         else
 640                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 641         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 642
 643         if (ret)
 644                 return ret;
 645         else
 646                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 647 }
 648 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 649
 650 static ssize_t dev_show_model(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 model[MODEL_LEN + 1];
 657         unsigned long flags;
 658         int ret = 0;
 659
 660         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 661         if (h->busy_configuring)
 662                 ret = -EBUSY;
 663         else
 664                 memcpy(model, drv->model, MODEL_LEN + 1);
 665         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 666
 667         if (ret)
 668                 return ret;
 669         else
 670                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 671 }
 672 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 673
 674 static ssize_t dev_show_rev(struct device *dev,
 675                             struct device_attribute *attr,
 676                             char *buf)
 677 {
 678         drive_info_struct *drv = to_drv(dev);
 679         struct ctlr_info *h = to_hba(drv->dev.parent);
 680         char rev[REV_LEN + 1];
 681         unsigned long flags;
 682         int ret = 0;
 683
 684         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 685         if (h->busy_configuring)
 686                 ret = -EBUSY;
 687         else
 688                 memcpy(rev, drv->rev, REV_LEN + 1);
 689         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 690
 691         if (ret)
 692                 return ret;
 693         else
 694                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 695 }
 696 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 697
 698 static ssize_t cciss_show_lunid(struct device *dev,
 699                                 struct device_attribute *attr, char *buf)
 700 {
 701         drive_info_struct *drv = to_drv(dev);
 702         struct ctlr_info *h = to_hba(drv->dev.parent);
 703         unsigned long flags;
 704         unsigned char lunid[8];
 705
 706         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 707         if (h->busy_configuring) {
 708                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 709                 return -EBUSY;
 710         }
 711         if (!drv->heads) {
 712                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 713                 return -ENOTTY;
 714         }
 715         memcpy(lunid, drv->LunID, sizeof(lunid));
 716         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 717         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 718                 lunid[0], lunid[1], lunid[2], lunid[3],
 719                 lunid[4], lunid[5], lunid[6], lunid[7]);
 720 }
 721 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 722
 723 static ssize_t cciss_show_raid_level(struct device *dev,
 724                                      struct device_attribute *attr, char *buf)
 725 {
 726         drive_info_struct *drv = to_drv(dev);
 727         struct ctlr_info *h = to_hba(drv->dev.parent);
 728         int raid;
 729         unsigned long flags;
 730
 731         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 732         if (h->busy_configuring) {
 733                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 734                 return -EBUSY;
 735         }
 736         raid = drv->raid_level;
 737         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 738         if (raid < 0 || raid > RAID_UNKNOWN)
 739                 raid = RAID_UNKNOWN;
 740
 741         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 742                         raid_label[raid]);
 743 }
 744 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 745
 746 static ssize_t cciss_show_usage_count(struct device *dev,
 747                                       struct device_attribute *attr, char *buf)
 748 {
 749         drive_info_struct *drv = to_drv(dev);
 750         struct ctlr_info *h = to_hba(drv->dev.parent);
 751         unsigned long flags;
 752         int count;
 753
 754         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 755         if (h->busy_configuring) {
 756                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 757                 return -EBUSY;
 758         }
 759         count = drv->usage_count;
 760         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 761         return snprintf(buf, 20, "%d\n", count);
 762 }
 763 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 764
 765 static struct attribute *cciss_host_attrs[] = {
 766         &dev_attr_rescan.attr,
 767         NULL
 768 };
 769
 770 static struct attribute_group cciss_host_attr_group = {
 771         .attrs = cciss_host_attrs,
 772 };
 773
 774 static const struct attribute_group *cciss_host_attr_groups[] = {
 775         &cciss_host_attr_group,
 776         NULL
 777 };
 778
 779 static struct device_type cciss_host_type = {
 780         .name           = "cciss_host",
 781         .groups         = cciss_host_attr_groups,
 782         .release        = cciss_hba_release,
 783 };
 784
 785 static struct attribute *cciss_dev_attrs[] = {
 786         &dev_attr_unique_id.attr,
 787         &dev_attr_model.attr,
 788         &dev_attr_vendor.attr,
 789         &dev_attr_rev.attr,
 790         &dev_attr_lunid.attr,
 791         &dev_attr_raid_level.attr,
 792         &dev_attr_usage_count.attr,
 793         NULL
 794 };
 795
 796 static struct attribute_group cciss_dev_attr_group = {
 797         .attrs = cciss_dev_attrs,
 798 };
 799
 800 static const struct attribute_group *cciss_dev_attr_groups[] = {
 801         &cciss_dev_attr_group,
 802         NULL
 803 };
 804
 805 static struct device_type cciss_dev_type = {
 806         .name           = "cciss_device",
 807         .groups         = cciss_dev_attr_groups,
 808         .release        = cciss_device_release,
 809 };
 810
 811 static struct bus_type cciss_bus_type = {
 812         .name           = "cciss",
 813 };
 814
 815 /*
 816  * cciss_hba_release is called when the reference count
 817  * of h->dev goes to zero.
 818  */
 819 static void cciss_hba_release(struct device *dev)
 820 {
 821         /*
 822          * nothing to do, but need this to avoid a warning
 823          * about not having a release handler from lib/kref.c.
 824          */
 825 }
 826
 827 /*
 828  * Initialize sysfs entry for each controller.  This sets up and registers
 829  * the 'cciss#' directory for each individual controller under
 830  * /sys/bus/pci/devices/<dev>/.
 831  */
 832 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 833 {
 834         device_initialize(&h->dev);
 835         h->dev.type = &cciss_host_type;
 836         h->dev.bus = &cciss_bus_type;
 837         dev_set_name(&h->dev, "%s", h->devname);
 838         h->dev.parent = &h->pdev->dev;
 839
 840         return device_add(&h->dev);
 841 }
 842
 843 /*
 844  * Remove sysfs entries for an hba.
 845  */
 846 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 847 {
 848         device_del(&h->dev);
 849         put_device(&h->dev); /* final put. */
 850 }
 851
 852 /* cciss_device_release is called when the reference count
 853  * of h->drv[x]dev goes to zero.
 854  */
 855 static void cciss_device_release(struct device *dev)
 856 {
 857         drive_info_struct *drv = to_drv(dev);
 858         kfree(drv);
 859 }
 860
 861 /*
 862  * Initialize sysfs for each logical drive.  This sets up and registers
 863  * the 'c#d#' directory for each individual logical drive under
 864  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 865  * /sys/block/cciss!c#d# to this entry.
 866  */
 867 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 868                                        int drv_index)
 869 {
 870         struct device *dev;
 871
 872         if (h->drv[drv_index]->device_initialized)
 873                 return 0;
 874
 875         dev = &h->drv[drv_index]->dev;
 876         device_initialize(dev);
 877         dev->type = &cciss_dev_type;
 878         dev->bus = &cciss_bus_type;
 879         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 880         dev->parent = &h->dev;
 881         h->drv[drv_index]->device_initialized = 1;
 882         return device_add(dev);
 883 }
 884
 885 /*
 886  * Remove sysfs entries for a logical drive.
 887  */
 888 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 889         int ctlr_exiting)
 890 {
 891         struct device *dev = &h->drv[drv_index]->dev;
 892
 893         /* special case for c*d0, we only destroy it on controller exit */
 894         if (drv_index == 0 && !ctlr_exiting)
 895                 return;
 896
 897         device_del(dev);
 898         put_device(dev); /* the "final" put. */
 899         h->drv[drv_index] = NULL;
 900 }
 901
 902 /*
 903  * For operations that cannot sleep, a command block is allocated at init,
 904  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 905  * which ones are free or in use.  For operations that can wait for kmalloc
 906  * to possible sleep, this routine can be called with get_from_pool set to 0.
 907  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 908  */
 909 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 910 {
 911         CommandList_struct *c;
 912         int i;
 913         u64bit temp64;
 914         dma_addr_t cmd_dma_handle, err_dma_handle;
 915
 916         if (!get_from_pool) {
 917                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 918                         sizeof(CommandList_struct), &cmd_dma_handle);
 919                 if (c == NULL)
 920                         return NULL;
 921                 memset(c, 0, sizeof(CommandList_struct));
 922
 923                 c->cmdindex = -1;
 924
 925                 c->err_info = (ErrorInfo_struct *)
 926                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 927                             &err_dma_handle);
 928
 929                 if (c->err_info == NULL) {
 930                         pci_free_consistent(h->pdev,
 931                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 932                         return NULL;
 933                 }
 934                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 935         } else {                /* get it out of the controllers pool */
 936
 937                 do {
 938                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 939                         if (i == h->nr_cmds)
 940                                 return NULL;
 941                 } while (test_and_set_bit
 942                          (i & (BITS_PER_LONG - 1),
 943                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 944 #ifdef CCISS_DEBUG
 945                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 946 #endif
 947                 c = h->cmd_pool + i;
 948                 memset(c, 0, sizeof(CommandList_struct));
 949                 cmd_dma_handle = h->cmd_pool_dhandle
 950                     + i * sizeof(CommandList_struct);
 951                 c->err_info = h->errinfo_pool + i;
 952                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 953                 err_dma_handle = h->errinfo_pool_dhandle
 954                     + i * sizeof(ErrorInfo_struct);
 955                 h->nr_allocs++;
 956
 957                 c->cmdindex = i;
 958         }
 959
 960         INIT_HLIST_NODE(&c->list);
 961         c->busaddr = (__u32) cmd_dma_handle;
 962         temp64.val = (__u64) err_dma_handle;
 963         c->ErrDesc.Addr.lower = temp64.val32.lower;
 964         c->ErrDesc.Addr.upper = temp64.val32.upper;
 965         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 966
 967         c->ctlr = h->ctlr;
 968         return c;
 969 }
 970
 971 /*
 972  * Frees a command block that was previously allocated with cmd_alloc().
 973  */
 974 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 975 {
 976         int i;
 977         u64bit temp64;
 978
 979         if (!got_from_pool) {
 980                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 981                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 982                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 983                                     c->err_info, (dma_addr_t) temp64.val);
 984                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 985                                     c, (dma_addr_t) c->busaddr);
 986         } else {
 987                 i = c - h->cmd_pool;
 988                 clear_bit(i & (BITS_PER_LONG - 1),
 989                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 990                 h->nr_frees++;
 991         }
 992 }
 993
 994 static inline ctlr_info_t *get_host(struct gendisk *disk)
 995 {
 996         return disk->queue->queuedata;
 997 }
 998
 999 static inline drive_info_struct *get_drv(struct gendisk *disk)
1000 {
1001         return disk->private_data;
1002 }
1003
1004 /*
1005  * Open.  Make sure the device is really there.
1006  */
1007 static int cciss_open(struct block_device *bdev, fmode_t mode)
1008 {
1009         ctlr_info_t *host = get_host(bdev->bd_disk);
1010         drive_info_struct *drv = get_drv(bdev->bd_disk);
1011
1012 #ifdef CCISS_DEBUG
1013         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
1014 #endif                          /* CCISS_DEBUG */
1015
1016         if (drv->busy_configuring)
1017                 return -EBUSY;
1018         /*
1019          * Root is allowed to open raw volume zero even if it's not configured
1020          * so array config can still work. Root is also allowed to open any
1021          * volume that has a LUN ID, so it can issue IOCTL to reread the
1022          * disk information.  I don't think I really like this
1023          * but I'm already using way to many device nodes to claim another one
1024          * for "raw controller".
1025          */
1026         if (drv->heads == 0) {
1027                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1028                         /* if not node 0 make sure it is a partition = 0 */
1029                         if (MINOR(bdev->bd_dev) & 0x0f) {
1030                                 return -ENXIO;
1031                                 /* if it is, make sure we have a LUN ID */
1032                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1033                                 sizeof(drv->LunID))) {
1034                                 return -ENXIO;
1035                         }
1036                 }
1037                 if (!capable(CAP_SYS_ADMIN))
1038                         return -EPERM;
1039         }
1040         drv->usage_count++;
1041         host->usage_count++;
1042         return 0;
1043 }
1044
1045 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1046 {
1047         int ret;
1048
1049         lock_kernel();
1050         ret = cciss_open(bdev, mode);
1051         unlock_kernel();
1052
1053         return ret;
1054 }
1055
1056 /*
1057  * Close.  Sync first.
1058  */
1059 static int cciss_release(struct gendisk *disk, fmode_t mode)
1060 {
1061         ctlr_info_t *host;
1062         drive_info_struct *drv;
1063
1064         lock_kernel();
1065         host = get_host(disk);
1066         drv = get_drv(disk);
1067
1068 #ifdef CCISS_DEBUG
1069         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
1070 #endif                          /* CCISS_DEBUG */
1071
1072         drv->usage_count--;
1073         host->usage_count--;
1074         unlock_kernel();
1075         return 0;
1076 }
1077
1078 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1079                     unsigned cmd, unsigned long arg)
1080 {
1081         int ret;
1082         lock_kernel();
1083         ret = cciss_ioctl(bdev, mode, cmd, arg);
1084         unlock_kernel();
1085         return ret;
1086 }
1087
1088 #ifdef CONFIG_COMPAT
1089
1090 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1091                                   unsigned cmd, unsigned long arg);
1092 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1093                                       unsigned cmd, unsigned long arg);
1094
1095 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1096                               unsigned cmd, unsigned long arg)
1097 {
1098         switch (cmd) {
1099         case CCISS_GETPCIINFO:
1100         case CCISS_GETINTINFO:
1101         case CCISS_SETINTINFO:
1102         case CCISS_GETNODENAME:
1103         case CCISS_SETNODENAME:
1104         case CCISS_GETHEARTBEAT:
1105         case CCISS_GETBUSTYPES:
1106         case CCISS_GETFIRMVER:
1107         case CCISS_GETDRIVVER:
1108         case CCISS_REVALIDVOLS:
1109         case CCISS_DEREGDISK:
1110         case CCISS_REGNEWDISK:
1111         case CCISS_REGNEWD:
1112         case CCISS_RESCANDISK:
1113         case CCISS_GETLUNINFO:
1114                 return do_ioctl(bdev, mode, cmd, arg);
1115
1116         case CCISS_PASSTHRU32:
1117                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1118         case CCISS_BIG_PASSTHRU32:
1119                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1120
1121         default:
1122                 return -ENOIOCTLCMD;
1123         }
1124 }
1125
1126 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1127                                   unsigned cmd, unsigned long arg)
1128 {
1129         IOCTL32_Command_struct __user *arg32 =
1130             (IOCTL32_Command_struct __user *) arg;
1131         IOCTL_Command_struct arg64;
1132         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1133         int err;
1134         u32 cp;
1135
1136         err = 0;
1137         err |=
1138             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1139                            sizeof(arg64.LUN_info));
1140         err |=
1141             copy_from_user(&arg64.Request, &arg32->Request,
1142                            sizeof(arg64.Request));
1143         err |=
1144             copy_from_user(&arg64.error_info, &arg32->error_info,
1145                            sizeof(arg64.error_info));
1146         err |= get_user(arg64.buf_size, &arg32->buf_size);
1147         err |= get_user(cp, &arg32->buf);
1148         arg64.buf = compat_ptr(cp);
1149         err |= copy_to_user(p, &arg64, sizeof(arg64));
1150
1151         if (err)
1152                 return -EFAULT;
1153
1154         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1155         if (err)
1156                 return err;
1157         err |=
1158             copy_in_user(&arg32->error_info, &p->error_info,
1159                          sizeof(arg32->error_info));
1160         if (err)
1161                 return -EFAULT;
1162         return err;
1163 }
1164
1165 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1166                                       unsigned cmd, unsigned long arg)
1167 {
1168         BIG_IOCTL32_Command_struct __user *arg32 =
1169             (BIG_IOCTL32_Command_struct __user *) arg;
1170         BIG_IOCTL_Command_struct arg64;
1171         BIG_IOCTL_Command_struct __user *p =
1172             compat_alloc_user_space(sizeof(arg64));
1173         int err;
1174         u32 cp;
1175
1176         err = 0;
1177         err |=
1178             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1179                            sizeof(arg64.LUN_info));
1180         err |=
1181             copy_from_user(&arg64.Request, &arg32->Request,
1182                            sizeof(arg64.Request));
1183         err |=
1184             copy_from_user(&arg64.error_info, &arg32->error_info,
1185                            sizeof(arg64.error_info));
1186         err |= get_user(arg64.buf_size, &arg32->buf_size);
1187         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1188         err |= get_user(cp, &arg32->buf);
1189         arg64.buf = compat_ptr(cp);
1190         err |= copy_to_user(p, &arg64, sizeof(arg64));
1191
1192         if (err)
1193                 return -EFAULT;
1194
1195         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1196         if (err)
1197                 return err;
1198         err |=
1199             copy_in_user(&arg32->error_info, &p->error_info,
1200                          sizeof(arg32->error_info));
1201         if (err)
1202                 return -EFAULT;
1203         return err;
1204 }
1205 #endif
1206
1207 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1208 {
1209         drive_info_struct *drv = get_drv(bdev->bd_disk);
1210
1211         if (!drv->cylinders)
1212                 return -ENXIO;
1213
1214         geo->heads = drv->heads;
1215         geo->sectors = drv->sectors;
1216         geo->cylinders = drv->cylinders;
1217         return 0;
1218 }
1219
1220 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1221 {
1222         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1223                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1224                 (void)check_for_unit_attention(host, c);
1225 }
1226 /*
1227  * ioctl
1228  */
1229 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1230                        unsigned int cmd, unsigned long arg)
1231 {
1232         struct gendisk *disk = bdev->bd_disk;
1233         ctlr_info_t *host = get_host(disk);
1234         drive_info_struct *drv = get_drv(disk);
1235         int ctlr = host->ctlr;
1236         void __user *argp = (void __user *)arg;
1237
1238 #ifdef CCISS_DEBUG
1239         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1240 #endif                          /* CCISS_DEBUG */
1241
1242         switch (cmd) {
1243         case CCISS_GETPCIINFO:
1244                 {
1245                         cciss_pci_info_struct pciinfo;
1246
1247                         if (!arg)
1248                                 return -EINVAL;
1249                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1250                         pciinfo.bus = host->pdev->bus->number;
1251                         pciinfo.dev_fn = host->pdev->devfn;
1252                         pciinfo.board_id = host->board_id;
1253                         if (copy_to_user
1254                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1255                                 return -EFAULT;
1256                         return 0;
1257                 }
1258         case CCISS_GETINTINFO:
1259                 {
1260                         cciss_coalint_struct intinfo;
1261                         if (!arg)
1262                                 return -EINVAL;
1263                         intinfo.delay =
1264                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1265                         intinfo.count =
1266                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1267                         if (copy_to_user
1268                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1269                                 return -EFAULT;
1270                         return 0;
1271                 }
1272         case CCISS_SETINTINFO:
1273                 {
1274                         cciss_coalint_struct intinfo;
1275                         unsigned long flags;
1276                         int i;
1277
1278                         if (!arg)
1279                                 return -EINVAL;
1280                         if (!capable(CAP_SYS_ADMIN))
1281                                 return -EPERM;
1282                         if (copy_from_user
1283                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1284                                 return -EFAULT;
1285                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1286                         {
1287 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1288                                 return -EINVAL;
1289                         }
1290                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1291                         /* Update the field, and then ring the doorbell */
1292                         writel(intinfo.delay,
1293                                &(host->cfgtable->HostWrite.CoalIntDelay));
1294                         writel(intinfo.count,
1295                                &(host->cfgtable->HostWrite.CoalIntCount));
1296                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1297
1298                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1299                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1300                                       & CFGTBL_ChangeReq))
1301                                         break;
1302                                 /* delay and try again */
1303                                 udelay(1000);
1304                         }
1305                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1306                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1307                                 return -EAGAIN;
1308                         return 0;
1309                 }
1310         case CCISS_GETNODENAME:
1311                 {
1312                         NodeName_type NodeName;
1313                         int i;
1314
1315                         if (!arg)
1316                                 return -EINVAL;
1317                         for (i = 0; i < 16; i++)
1318                                 NodeName[i] =
1319                                     readb(&host->cfgtable->ServerName[i]);
1320                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1321                                 return -EFAULT;
1322                         return 0;
1323                 }
1324         case CCISS_SETNODENAME:
1325                 {
1326                         NodeName_type NodeName;
1327                         unsigned long flags;
1328                         int i;
1329
1330                         if (!arg)
1331                                 return -EINVAL;
1332                         if (!capable(CAP_SYS_ADMIN))
1333                                 return -EPERM;
1334
1335                         if (copy_from_user
1336                             (NodeName, argp, sizeof(NodeName_type)))
1337                                 return -EFAULT;
1338
1339                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1340
1341                         /* Update the field, and then ring the doorbell */
1342                         for (i = 0; i < 16; i++)
1343                                 writeb(NodeName[i],
1344                                        &host->cfgtable->ServerName[i]);
1345
1346                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1347
1348                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1349                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1350                                       & CFGTBL_ChangeReq))
1351                                         break;
1352                                 /* delay and try again */
1353                                 udelay(1000);
1354                         }
1355                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1356                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1357                                 return -EAGAIN;
1358                         return 0;
1359                 }
1360
1361         case CCISS_GETHEARTBEAT:
1362                 {
1363                         Heartbeat_type heartbeat;
1364
1365                         if (!arg)
1366                                 return -EINVAL;
1367                         heartbeat = readl(&host->cfgtable->HeartBeat);
1368                         if (copy_to_user
1369                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1370                                 return -EFAULT;
1371                         return 0;
1372                 }
1373         case CCISS_GETBUSTYPES:
1374                 {
1375                         BusTypes_type BusTypes;
1376
1377                         if (!arg)
1378                                 return -EINVAL;
1379                         BusTypes = readl(&host->cfgtable->BusTypes);
1380                         if (copy_to_user
1381                             (argp, &BusTypes, sizeof(BusTypes_type)))
1382                                 return -EFAULT;
1383                         return 0;
1384                 }
1385         case CCISS_GETFIRMVER:
1386                 {
1387                         FirmwareVer_type firmware;
1388
1389                         if (!arg)
1390                                 return -EINVAL;
1391                         memcpy(firmware, host->firm_ver, 4);
1392
1393                         if (copy_to_user
1394                             (argp, firmware, sizeof(FirmwareVer_type)))
1395                                 return -EFAULT;
1396                         return 0;
1397                 }
1398         case CCISS_GETDRIVVER:
1399                 {
1400                         DriverVer_type DriverVer = DRIVER_VERSION;
1401
1402                         if (!arg)
1403                                 return -EINVAL;
1404
1405                         if (copy_to_user
1406                             (argp, &DriverVer, sizeof(DriverVer_type)))
1407                                 return -EFAULT;
1408                         return 0;
1409                 }
1410
1411         case CCISS_DEREGDISK:
1412         case CCISS_REGNEWD:
1413         case CCISS_REVALIDVOLS:
1414                 return rebuild_lun_table(host, 0, 1);
1415
1416         case CCISS_GETLUNINFO:{
1417                         LogvolInfo_struct luninfo;
1418
1419                         memcpy(&luninfo.LunID, drv->LunID,
1420                                 sizeof(luninfo.LunID));
1421                         luninfo.num_opens = drv->usage_count;
1422                         luninfo.num_parts = 0;
1423                         if (copy_to_user(argp, &luninfo,
1424                                          sizeof(LogvolInfo_struct)))
1425                                 return -EFAULT;
1426                         return 0;
1427                 }
1428         case CCISS_PASSTHRU:
1429                 {
1430                         IOCTL_Command_struct iocommand;
1431                         CommandList_struct *c;
1432                         char *buff = NULL;
1433                         u64bit temp64;
1434                         DECLARE_COMPLETION_ONSTACK(wait);
1435
1436                         if (!arg)
1437                                 return -EINVAL;
1438
1439                         if (!capable(CAP_SYS_RAWIO))
1440                                 return -EPERM;
1441
1442                         if (copy_from_user
1443                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1444                                 return -EFAULT;
1445                         if ((iocommand.buf_size < 1) &&
1446                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1447                                 return -EINVAL;
1448                         }
1449 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1450                         /* Check kmalloc limits */
1451                         if (iocommand.buf_size > 128000)
1452                                 return -EINVAL;
1453 #endif
1454                         if (iocommand.buf_size > 0) {
1455                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1456                                 if (buff == NULL)
1457                                         return -EFAULT;
1458                         }
1459                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1460                                 /* Copy the data into the buffer we created */
1461                                 if (copy_from_user
1462                                     (buff, iocommand.buf, iocommand.buf_size)) {
1463                                         kfree(buff);
1464                                         return -EFAULT;
1465                                 }
1466                         } else {
1467                                 memset(buff, 0, iocommand.buf_size);
1468                         }
1469                         if ((c = cmd_alloc(host, 0)) == NULL) {
1470                                 kfree(buff);
1471                                 return -ENOMEM;
1472                         }
1473                         /* Fill in the command type */
1474                         c->cmd_type = CMD_IOCTL_PEND;
1475                         /* Fill in Command Header */
1476                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1477                         if (iocommand.buf_size > 0) /* buffer to fill */
1478                         {
1479                                 c->Header.SGList = 1;
1480                                 c->Header.SGTotal = 1;
1481                         } else /* no buffers to fill */
1482                         {
1483                                 c->Header.SGList = 0;
1484                                 c->Header.SGTotal = 0;
1485                         }
1486                         c->Header.LUN = iocommand.LUN_info;
1487                         /* use the kernel address the cmd block for tag */
1488                         c->Header.Tag.lower = c->busaddr;
1489
1490                         /* Fill in Request block */
1491                         c->Request = iocommand.Request;
1492
1493                         /* Fill in the scatter gather information */
1494                         if (iocommand.buf_size > 0) {
1495                                 temp64.val = pci_map_single(host->pdev, buff,
1496                                         iocommand.buf_size,
1497                                         PCI_DMA_BIDIRECTIONAL);
1498                                 c->SG[0].Addr.lower = temp64.val32.lower;
1499                                 c->SG[0].Addr.upper = temp64.val32.upper;
1500                                 c->SG[0].Len = iocommand.buf_size;
1501                                 c->SG[0].Ext = 0;  /* we are not chaining */
1502                         }
1503                         c->waiting = &wait;
1504
1505                         enqueue_cmd_and_start_io(host, c);
1506                         wait_for_completion(&wait);
1507
1508                         /* unlock the buffers from DMA */
1509                         temp64.val32.lower = c->SG[0].Addr.lower;
1510                         temp64.val32.upper = c->SG[0].Addr.upper;
1511                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1512                                          iocommand.buf_size,
1513                                          PCI_DMA_BIDIRECTIONAL);
1514
1515                         check_ioctl_unit_attention(host, c);
1516
1517                         /* Copy the error information out */
1518                         iocommand.error_info = *(c->err_info);
1519                         if (copy_to_user
1520                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1521                                 kfree(buff);
1522                                 cmd_free(host, c, 0);
1523                                 return -EFAULT;
1524                         }
1525
1526                         if (iocommand.Request.Type.Direction == XFER_READ) {
1527                                 /* Copy the data out of the buffer we created */
1528                                 if (copy_to_user
1529                                     (iocommand.buf, buff, iocommand.buf_size)) {
1530                                         kfree(buff);
1531                                         cmd_free(host, c, 0);
1532                                         return -EFAULT;
1533                                 }
1534                         }
1535                         kfree(buff);
1536                         cmd_free(host, c, 0);
1537                         return 0;
1538                 }
1539         case CCISS_BIG_PASSTHRU:{
1540                         BIG_IOCTL_Command_struct *ioc;
1541                         CommandList_struct *c;
1542                         unsigned char **buff = NULL;
1543                         int *buff_size = NULL;
1544                         u64bit temp64;
1545                         BYTE sg_used = 0;
1546                         int status = 0;
1547                         int i;
1548                         DECLARE_COMPLETION_ONSTACK(wait);
1549                         __u32 left;
1550                         __u32 sz;
1551                         BYTE __user *data_ptr;
1552
1553                         if (!arg)
1554                                 return -EINVAL;
1555                         if (!capable(CAP_SYS_RAWIO))
1556                                 return -EPERM;
1557                         ioc = (BIG_IOCTL_Command_struct *)
1558                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1559                         if (!ioc) {
1560                                 status = -ENOMEM;
1561                                 goto cleanup1;
1562                         }
1563                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1564                                 status = -EFAULT;
1565                                 goto cleanup1;
1566                         }
1567                         if ((ioc->buf_size < 1) &&
1568                             (ioc->Request.Type.Direction != XFER_NONE)) {
1569                                 status = -EINVAL;
1570                                 goto cleanup1;
1571                         }
1572                         /* Check kmalloc limits  using all SGs */
1573                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1574                                 status = -EINVAL;
1575                                 goto cleanup1;
1576                         }
1577                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1578                                 status = -EINVAL;
1579                                 goto cleanup1;
1580                         }
1581                         buff =
1582                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1583                         if (!buff) {
1584                                 status = -ENOMEM;
1585                                 goto cleanup1;
1586                         }
1587                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1588                                                    GFP_KERNEL);
1589                         if (!buff_size) {
1590                                 status = -ENOMEM;
1591                                 goto cleanup1;
1592                         }
1593                         left = ioc->buf_size;
1594                         data_ptr = ioc->buf;
1595                         while (left) {
1596                                 sz = (left >
1597                                       ioc->malloc_size) ? ioc->
1598                                     malloc_size : left;
1599                                 buff_size[sg_used] = sz;
1600                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1601                                 if (buff[sg_used] == NULL) {
1602                                         status = -ENOMEM;
1603                                         goto cleanup1;
1604                                 }
1605                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1606                                         if (copy_from_user
1607                                             (buff[sg_used], data_ptr, sz)) {
1608                                                 status = -EFAULT;
1609                                                 goto cleanup1;
1610                                         }
1611                                 } else {
1612                                         memset(buff[sg_used], 0, sz);
1613                                 }
1614                                 left -= sz;
1615                                 data_ptr += sz;
1616                                 sg_used++;
1617                         }
1618                         if ((c = cmd_alloc(host, 0)) == NULL) {
1619                                 status = -ENOMEM;
1620                                 goto cleanup1;
1621                         }
1622                         c->cmd_type = CMD_IOCTL_PEND;
1623                         c->Header.ReplyQueue = 0;
1624
1625                         if (ioc->buf_size > 0) {
1626                                 c->Header.SGList = sg_used;
1627                                 c->Header.SGTotal = sg_used;
1628                         } else {
1629                                 c->Header.SGList = 0;
1630                                 c->Header.SGTotal = 0;
1631                         }
1632                         c->Header.LUN = ioc->LUN_info;
1633                         c->Header.Tag.lower = c->busaddr;
1634
1635                         c->Request = ioc->Request;
1636                         if (ioc->buf_size > 0) {
1637                                 for (i = 0; i < sg_used; i++) {
1638                                         temp64.val =
1639                                             pci_map_single(host->pdev, buff[i],
1640                                                     buff_size[i],
1641                                                     PCI_DMA_BIDIRECTIONAL);
1642                                         c->SG[i].Addr.lower =
1643                                             temp64.val32.lower;
1644                                         c->SG[i].Addr.upper =
1645                                             temp64.val32.upper;
1646                                         c->SG[i].Len = buff_size[i];
1647                                         c->SG[i].Ext = 0;       /* we are not chaining */
1648                                 }
1649                         }
1650                         c->waiting = &wait;
1651                         enqueue_cmd_and_start_io(host, c);
1652                         wait_for_completion(&wait);
1653                         /* unlock the buffers from DMA */
1654                         for (i = 0; i < sg_used; i++) {
1655                                 temp64.val32.lower = c->SG[i].Addr.lower;
1656                                 temp64.val32.upper = c->SG[i].Addr.upper;
1657                                 pci_unmap_single(host->pdev,
1658                                         (dma_addr_t) temp64.val, buff_size[i],
1659                                         PCI_DMA_BIDIRECTIONAL);
1660                         }
1661                         check_ioctl_unit_attention(host, c);
1662                         /* Copy the error information out */
1663                         ioc->error_info = *(c->err_info);
1664                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1665                                 cmd_free(host, c, 0);
1666                                 status = -EFAULT;
1667                                 goto cleanup1;
1668                         }
1669                         if (ioc->Request.Type.Direction == XFER_READ) {
1670                                 /* Copy the data out of the buffer we created */
1671                                 BYTE __user *ptr = ioc->buf;
1672                                 for (i = 0; i < sg_used; i++) {
1673                                         if (copy_to_user
1674                                             (ptr, buff[i], buff_size[i])) {
1675                                                 cmd_free(host, c, 0);
1676                                                 status = -EFAULT;
1677                                                 goto cleanup1;
1678                                         }
1679                                         ptr += buff_size[i];
1680                                 }
1681                         }
1682                         cmd_free(host, c, 0);
1683                         status = 0;
1684                       cleanup1:
1685                         if (buff) {
1686                                 for (i = 0; i < sg_used; i++)
1687                                         kfree(buff[i]);
1688                                 kfree(buff);
1689                         }
1690                         kfree(buff_size);
1691                         kfree(ioc);
1692                         return status;
1693                 }
1694
1695         /* scsi_cmd_ioctl handles these, below, though some are not */
1696         /* very meaningful for cciss.  SG_IO is the main one people want. */
1697
1698         case SG_GET_VERSION_NUM:
1699         case SG_SET_TIMEOUT:
1700         case SG_GET_TIMEOUT:
1701         case SG_GET_RESERVED_SIZE:
1702         case SG_SET_RESERVED_SIZE:
1703         case SG_EMULATED_HOST:
1704         case SG_IO:
1705         case SCSI_IOCTL_SEND_COMMAND:
1706                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1707
1708         /* scsi_cmd_ioctl would normally handle these, below, but */
1709         /* they aren't a good fit for cciss, as CD-ROMs are */
1710         /* not supported, and we don't have any bus/target/lun */
1711         /* which we present to the kernel. */
1712
1713         case CDROM_SEND_PACKET:
1714         case CDROMCLOSETRAY:
1715         case CDROMEJECT:
1716         case SCSI_IOCTL_GET_IDLUN:
1717         case SCSI_IOCTL_GET_BUS_NUMBER:
1718         default:
1719                 return -ENOTTY;
1720         }
1721 }
1722
1723 static void cciss_check_queues(ctlr_info_t *h)
1724 {
1725         int start_queue = h->next_to_run;
1726         int i;
1727
1728         /* check to see if we have maxed out the number of commands that can
1729          * be placed on the queue.  If so then exit.  We do this check here
1730          * in case the interrupt we serviced was from an ioctl and did not
1731          * free any new commands.
1732          */
1733         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1734                 return;
1735
1736         /* We have room on the queue for more commands.  Now we need to queue
1737          * them up.  We will also keep track of the next queue to run so
1738          * that every queue gets a chance to be started first.
1739          */
1740         for (i = 0; i < h->highest_lun + 1; i++) {
1741                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1742                 /* make sure the disk has been added and the drive is real
1743                  * because this can be called from the middle of init_one.
1744                  */
1745                 if (!h->drv[curr_queue])
1746                         continue;
1747                 if (!(h->drv[curr_queue]->queue) ||
1748                         !(h->drv[curr_queue]->heads))
1749                         continue;
1750                 blk_start_queue(h->gendisk[curr_queue]->queue);
1751
1752                 /* check to see if we have maxed out the number of commands
1753                  * that can be placed on the queue.
1754                  */
1755                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1756                         if (curr_queue == start_queue) {
1757                                 h->next_to_run =
1758                                     (start_queue + 1) % (h->highest_lun + 1);
1759                                 break;
1760                         } else {
1761                                 h->next_to_run = curr_queue;
1762                                 break;
1763                         }
1764                 }
1765         }
1766 }
1767
1768 static void cciss_softirq_done(struct request *rq)
1769 {
1770         CommandList_struct *cmd = rq->completion_data;
1771         ctlr_info_t *h = hba[cmd->ctlr];
1772         SGDescriptor_struct *curr_sg = cmd->SG;
1773         u64bit temp64;
1774         unsigned long flags;
1775         int i, ddir;
1776         int sg_index = 0;
1777
1778         if (cmd->Request.Type.Direction == XFER_READ)
1779                 ddir = PCI_DMA_FROMDEVICE;
1780         else
1781                 ddir = PCI_DMA_TODEVICE;
1782
1783         /* command did not need to be retried */
1784         /* unmap the DMA mapping for all the scatter gather elements */
1785         for (i = 0; i < cmd->Header.SGList; i++) {
1786                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1787                         cciss_unmap_sg_chain_block(h, cmd);
1788                         /* Point to the next block */
1789                         curr_sg = h->cmd_sg_list[cmd->cmdindex];
1790                         sg_index = 0;
1791                 }
1792                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1793                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1794                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1795                                 ddir);
1796                 ++sg_index;
1797         }
1798
1799 #ifdef CCISS_DEBUG
1800         printk("Done with %p\n", rq);
1801 #endif                          /* CCISS_DEBUG */
1802
1803         /* set the residual count for pc requests */
1804         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1805                 rq->resid_len = cmd->err_info->ResidualCnt;
1806
1807         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1808
1809         spin_lock_irqsave(&h->lock, flags);
1810         cmd_free(h, cmd, 1);
1811         cciss_check_queues(h);
1812         spin_unlock_irqrestore(&h->lock, flags);
1813 }
1814
1815 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1816         unsigned char scsi3addr[], uint32_t log_unit)
1817 {
1818         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1819                 sizeof(h->drv[log_unit]->LunID));
1820 }
1821
1822 /* This function gets the SCSI vendor, model, and revision of a logical drive
1823  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1824  * they cannot be read.
1825  */
1826 static void cciss_get_device_descr(int ctlr, int logvol,
1827                                    char *vendor, char *model, char *rev)
1828 {
1829         int rc;
1830         InquiryData_struct *inq_buf;
1831         unsigned char scsi3addr[8];
1832
1833         *vendor = '\0';
1834         *model = '\0';
1835         *rev = '\0';
1836
1837         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1838         if (!inq_buf)
1839                 return;
1840
1841         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1842         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1843                         scsi3addr, TYPE_CMD);
1844         if (rc == IO_OK) {
1845                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1846                 vendor[VENDOR_LEN] = '\0';
1847                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1848                 model[MODEL_LEN] = '\0';
1849                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1850                 rev[REV_LEN] = '\0';
1851         }
1852
1853         kfree(inq_buf);
1854         return;
1855 }
1856
1857 /* This function gets the serial number of a logical drive via
1858  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1859  * number cannot be had, for whatever reason, 16 bytes of 0xff
1860  * are returned instead.
1861  */
1862 static void cciss_get_serial_no(int ctlr, int logvol,
1863                                 unsigned char *serial_no, int buflen)
1864 {
1865 #define PAGE_83_INQ_BYTES 64
1866         int rc;
1867         unsigned char *buf;
1868         unsigned char scsi3addr[8];
1869
1870         if (buflen > 16)
1871                 buflen = 16;
1872         memset(serial_no, 0xff, buflen);
1873         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1874         if (!buf)
1875                 return;
1876         memset(serial_no, 0, buflen);
1877         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1878         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1879                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1880         if (rc == IO_OK)
1881                 memcpy(serial_no, &buf[8], buflen);
1882         kfree(buf);
1883         return;
1884 }
1885
1886 /*
1887  * cciss_add_disk sets up the block device queue for a logical drive
1888  */
1889 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1890                                 int drv_index)
1891 {
1892         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1893         if (!disk->queue)
1894                 goto init_queue_failure;
1895         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1896         disk->major = h->major;
1897         disk->first_minor = drv_index << NWD_SHIFT;
1898         disk->fops = &cciss_fops;
1899         if (cciss_create_ld_sysfs_entry(h, drv_index))
1900                 goto cleanup_queue;
1901         disk->private_data = h->drv[drv_index];
1902         disk->driverfs_dev = &h->drv[drv_index]->dev;
1903
1904         /* Set up queue information */
1905         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1906
1907         /* This is a hardware imposed limit. */
1908         blk_queue_max_segments(disk->queue, h->maxsgentries);
1909
1910         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1911
1912         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1913
1914         disk->queue->queuedata = h;
1915
1916         blk_queue_logical_block_size(disk->queue,
1917                                      h->drv[drv_index]->block_size);
1918
1919         /* Make sure all queue data is written out before */
1920         /* setting h->drv[drv_index]->queue, as setting this */
1921         /* allows the interrupt handler to start the queue */
1922         wmb();
1923         h->drv[drv_index]->queue = disk->queue;
1924         add_disk(disk);
1925         return 0;
1926
1927 cleanup_queue:
1928         blk_cleanup_queue(disk->queue);
1929         disk->queue = NULL;
1930 init_queue_failure:
1931         return -1;
1932 }
1933
1934 /* This function will check the usage_count of the drive to be updated/added.
1935  * If the usage_count is zero and it is a heretofore unknown drive, or,
1936  * the drive's capacity, geometry, or serial number has changed,
1937  * then the drive information will be updated and the disk will be
1938  * re-registered with the kernel.  If these conditions don't hold,
1939  * then it will be left alone for the next reboot.  The exception to this
1940  * is disk 0 which will always be left registered with the kernel since it
1941  * is also the controller node.  Any changes to disk 0 will show up on
1942  * the next reboot.
1943  */
1944 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1945         int via_ioctl)
1946 {
1947         ctlr_info_t *h = hba[ctlr];
1948         struct gendisk *disk;
1949         InquiryData_struct *inq_buff = NULL;
1950         unsigned int block_size;
1951         sector_t total_size;
1952         unsigned long flags = 0;
1953         int ret = 0;
1954         drive_info_struct *drvinfo;
1955
1956         /* Get information about the disk and modify the driver structure */
1957         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1958         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1959         if (inq_buff == NULL || drvinfo == NULL)
1960                 goto mem_msg;
1961
1962         /* testing to see if 16-byte CDBs are already being used */
1963         if (h->cciss_read == CCISS_READ_16) {
1964                 cciss_read_capacity_16(h->ctlr, drv_index,
1965                         &total_size, &block_size);
1966
1967         } else {
1968                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1969                 /* if read_capacity returns all F's this volume is >2TB */
1970                 /* in size so we switch to 16-byte CDB's for all */
1971                 /* read/write ops */
1972                 if (total_size == 0xFFFFFFFFULL) {
1973                         cciss_read_capacity_16(ctlr, drv_index,
1974                         &total_size, &block_size);
1975                         h->cciss_read = CCISS_READ_16;
1976                         h->cciss_write = CCISS_WRITE_16;
1977                 } else {
1978                         h->cciss_read = CCISS_READ_10;
1979                         h->cciss_write = CCISS_WRITE_10;
1980                 }
1981         }
1982
1983         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1984                                inq_buff, drvinfo);
1985         drvinfo->block_size = block_size;
1986         drvinfo->nr_blocks = total_size + 1;
1987
1988         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1989                                 drvinfo->model, drvinfo->rev);
1990         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1991                         sizeof(drvinfo->serial_no));
1992         /* Save the lunid in case we deregister the disk, below. */
1993         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1994                 sizeof(drvinfo->LunID));
1995
1996         /* Is it the same disk we already know, and nothing's changed? */
1997         if (h->drv[drv_index]->raid_level != -1 &&
1998                 ((memcmp(drvinfo->serial_no,
1999                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2000                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2001                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2002                 drvinfo->heads == h->drv[drv_index]->heads &&
2003                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2004                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2005                         /* The disk is unchanged, nothing to update */
2006                         goto freeret;
2007
2008         /* If we get here it's not the same disk, or something's changed,
2009          * so we need to * deregister it, and re-register it, if it's not
2010          * in use.
2011          * If the disk already exists then deregister it before proceeding
2012          * (unless it's the first disk (for the controller node).
2013          */
2014         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2015                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
2016                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2017                 h->drv[drv_index]->busy_configuring = 1;
2018                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2019
2020                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2021                  * which keeps the interrupt handler from starting
2022                  * the queue.
2023                  */
2024                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2025         }
2026
2027         /* If the disk is in use return */
2028         if (ret)
2029                 goto freeret;
2030
2031         /* Save the new information from cciss_geometry_inquiry
2032          * and serial number inquiry.  If the disk was deregistered
2033          * above, then h->drv[drv_index] will be NULL.
2034          */
2035         if (h->drv[drv_index] == NULL) {
2036                 drvinfo->device_initialized = 0;
2037                 h->drv[drv_index] = drvinfo;
2038                 drvinfo = NULL; /* so it won't be freed below. */
2039         } else {
2040                 /* special case for cxd0 */
2041                 h->drv[drv_index]->block_size = drvinfo->block_size;
2042                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2043                 h->drv[drv_index]->heads = drvinfo->heads;
2044                 h->drv[drv_index]->sectors = drvinfo->sectors;
2045                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2046                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2047                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2048                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2049                         VENDOR_LEN + 1);
2050                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2051                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2052         }
2053
2054         ++h->num_luns;
2055         disk = h->gendisk[drv_index];
2056         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2057
2058         /* If it's not disk 0 (drv_index != 0)
2059          * or if it was disk 0, but there was previously
2060          * no actual corresponding configured logical drive
2061          * (raid_leve == -1) then we want to update the
2062          * logical drive's information.
2063          */
2064         if (drv_index || first_time) {
2065                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2066                         cciss_free_gendisk(h, drv_index);
2067                         cciss_free_drive_info(h, drv_index);
2068                         printk(KERN_WARNING "cciss:%d could not update "
2069                                 "disk %d\n", h->ctlr, drv_index);
2070                         --h->num_luns;
2071                 }
2072         }
2073
2074 freeret:
2075         kfree(inq_buff);
2076         kfree(drvinfo);
2077         return;
2078 mem_msg:
2079         printk(KERN_ERR "cciss: out of memory\n");
2080         goto freeret;
2081 }
2082
2083 /* This function will find the first index of the controllers drive array
2084  * that has a null drv pointer and allocate the drive info struct and
2085  * will return that index   This is where new drives will be added.
2086  * If the index to be returned is greater than the highest_lun index for
2087  * the controller then highest_lun is set * to this new index.
2088  * If there are no available indexes or if tha allocation fails, then -1
2089  * is returned.  * "controller_node" is used to know if this is a real
2090  * logical drive, or just the controller node, which determines if this
2091  * counts towards highest_lun.
2092  */
2093 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2094 {
2095         int i;
2096         drive_info_struct *drv;
2097
2098         /* Search for an empty slot for our drive info */
2099         for (i = 0; i < CISS_MAX_LUN; i++) {
2100
2101                 /* if not cxd0 case, and it's occupied, skip it. */
2102                 if (h->drv[i] && i != 0)
2103                         continue;
2104                 /*
2105                  * If it's cxd0 case, and drv is alloc'ed already, and a
2106                  * disk is configured there, skip it.
2107                  */
2108                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2109                         continue;
2110
2111                 /*
2112                  * We've found an empty slot.  Update highest_lun
2113                  * provided this isn't just the fake cxd0 controller node.
2114                  */
2115                 if (i > h->highest_lun && !controller_node)
2116                         h->highest_lun = i;
2117
2118                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2119                 if (i == 0 && h->drv[i] != NULL)
2120                         return i;
2121
2122                 /*
2123                  * Found an empty slot, not already alloc'ed.  Allocate it.
2124                  * Mark it with raid_level == -1, so we know it's new later on.
2125                  */
2126                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2127                 if (!drv)
2128                         return -1;
2129                 drv->raid_level = -1; /* so we know it's new */
2130                 h->drv[i] = drv;
2131                 return i;
2132         }
2133         return -1;
2134 }
2135
2136 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2137 {
2138         kfree(h->drv[drv_index]);
2139         h->drv[drv_index] = NULL;
2140 }
2141
2142 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2143 {
2144         put_disk(h->gendisk[drv_index]);
2145         h->gendisk[drv_index] = NULL;
2146 }
2147
2148 /* cciss_add_gendisk finds a free hba[]->drv structure
2149  * and allocates a gendisk if needed, and sets the lunid
2150  * in the drvinfo structure.   It returns the index into
2151  * the ->drv[] array, or -1 if none are free.
2152  * is_controller_node indicates whether highest_lun should
2153  * count this disk, or if it's only being added to provide
2154  * a means to talk to the controller in case no logical
2155  * drives have yet been configured.
2156  */
2157 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2158         int controller_node)
2159 {
2160         int drv_index;
2161
2162         drv_index = cciss_alloc_drive_info(h, controller_node);
2163         if (drv_index == -1)
2164                 return -1;
2165
2166         /*Check if the gendisk needs to be allocated */
2167         if (!h->gendisk[drv_index]) {
2168                 h->gendisk[drv_index] =
2169                         alloc_disk(1 << NWD_SHIFT);
2170                 if (!h->gendisk[drv_index]) {
2171                         printk(KERN_ERR "cciss%d: could not "
2172                                 "allocate a new disk %d\n",
2173                                 h->ctlr, drv_index);
2174                         goto err_free_drive_info;
2175                 }
2176         }
2177         memcpy(h->drv[drv_index]->LunID, lunid,
2178                 sizeof(h->drv[drv_index]->LunID));
2179         if (cciss_create_ld_sysfs_entry(h, drv_index))
2180                 goto err_free_disk;
2181         /* Don't need to mark this busy because nobody */
2182         /* else knows about this disk yet to contend */
2183         /* for access to it. */
2184         h->drv[drv_index]->busy_configuring = 0;
2185         wmb();
2186         return drv_index;
2187
2188 err_free_disk:
2189         cciss_free_gendisk(h, drv_index);
2190 err_free_drive_info:
2191         cciss_free_drive_info(h, drv_index);
2192         return -1;
2193 }
2194
2195 /* This is for the special case of a controller which
2196  * has no logical drives.  In this case, we still need
2197  * to register a disk so the controller can be accessed
2198  * by the Array Config Utility.
2199  */
2200 static void cciss_add_controller_node(ctlr_info_t *h)
2201 {
2202         struct gendisk *disk;
2203         int drv_index;
2204
2205         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2206                 return;
2207
2208         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2209         if (drv_index == -1)
2210                 goto error;
2211         h->drv[drv_index]->block_size = 512;
2212         h->drv[drv_index]->nr_blocks = 0;
2213         h->drv[drv_index]->heads = 0;
2214         h->drv[drv_index]->sectors = 0;
2215         h->drv[drv_index]->cylinders = 0;
2216         h->drv[drv_index]->raid_level = -1;
2217         memset(h->drv[drv_index]->serial_no, 0, 16);
2218         disk = h->gendisk[drv_index];
2219         if (cciss_add_disk(h, disk, drv_index) == 0)
2220                 return;
2221         cciss_free_gendisk(h, drv_index);
2222         cciss_free_drive_info(h, drv_index);
2223 error:
2224         printk(KERN_WARNING "cciss%d: could not "
2225                 "add disk 0.\n", h->ctlr);
2226         return;
2227 }
2228
2229 /* This function will add and remove logical drives from the Logical
2230  * drive array of the controller and maintain persistency of ordering
2231  * so that mount points are preserved until the next reboot.  This allows
2232  * for the removal of logical drives in the middle of the drive array
2233  * without a re-ordering of those drives.
2234  * INPUT
2235  * h            = The controller to perform the operations on
2236  */
2237 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2238         int via_ioctl)
2239 {
2240         int ctlr = h->ctlr;
2241         int num_luns;
2242         ReportLunData_struct *ld_buff = NULL;
2243         int return_code;
2244         int listlength = 0;
2245         int i;
2246         int drv_found;
2247         int drv_index = 0;
2248         unsigned char lunid[8] = CTLR_LUNID;
2249         unsigned long flags;
2250
2251         if (!capable(CAP_SYS_RAWIO))
2252                 return -EPERM;
2253
2254         /* Set busy_configuring flag for this operation */
2255         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2256         if (h->busy_configuring) {
2257                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2258                 return -EBUSY;
2259         }
2260         h->busy_configuring = 1;
2261         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2262
2263         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2264         if (ld_buff == NULL)
2265                 goto mem_msg;
2266
2267         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2268                                       sizeof(ReportLunData_struct),
2269                                       0, CTLR_LUNID, TYPE_CMD);
2270
2271         if (return_code == IO_OK)
2272                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2273         else {  /* reading number of logical volumes failed */
2274                 printk(KERN_WARNING "cciss: report logical volume"
2275                        " command failed\n");
2276                 listlength = 0;
2277                 goto freeret;
2278         }
2279
2280         num_luns = listlength / 8;      /* 8 bytes per entry */
2281         if (num_luns > CISS_MAX_LUN) {
2282                 num_luns = CISS_MAX_LUN;
2283                 printk(KERN_WARNING "cciss: more luns configured"
2284                        " on controller than can be handled by"
2285                        " this driver.\n");
2286         }
2287
2288         if (num_luns == 0)
2289                 cciss_add_controller_node(h);
2290
2291         /* Compare controller drive array to driver's drive array
2292          * to see if any drives are missing on the controller due
2293          * to action of Array Config Utility (user deletes drive)
2294          * and deregister logical drives which have disappeared.
2295          */
2296         for (i = 0; i <= h->highest_lun; i++) {
2297                 int j;
2298                 drv_found = 0;
2299
2300                 /* skip holes in the array from already deleted drives */
2301                 if (h->drv[i] == NULL)
2302                         continue;
2303
2304                 for (j = 0; j < num_luns; j++) {
2305                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2306                         if (memcmp(h->drv[i]->LunID, lunid,
2307                                 sizeof(lunid)) == 0) {
2308                                 drv_found = 1;
2309                                 break;
2310                         }
2311                 }
2312                 if (!drv_found) {
2313                         /* Deregister it from the OS, it's gone. */
2314                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2315                         h->drv[i]->busy_configuring = 1;
2316                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2317                         return_code = deregister_disk(h, i, 1, via_ioctl);
2318                         if (h->drv[i] != NULL)
2319                                 h->drv[i]->busy_configuring = 0;
2320                 }
2321         }
2322
2323         /* Compare controller drive array to driver's drive array.
2324          * Check for updates in the drive information and any new drives
2325          * on the controller due to ACU adding logical drives, or changing
2326          * a logical drive's size, etc.  Reregister any new/changed drives
2327          */
2328         for (i = 0; i < num_luns; i++) {
2329                 int j;
2330
2331                 drv_found = 0;
2332
2333                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2334                 /* Find if the LUN is already in the drive array
2335                  * of the driver.  If so then update its info
2336                  * if not in use.  If it does not exist then find
2337                  * the first free index and add it.
2338                  */
2339                 for (j = 0; j <= h->highest_lun; j++) {
2340                         if (h->drv[j] != NULL &&
2341                                 memcmp(h->drv[j]->LunID, lunid,
2342                                         sizeof(h->drv[j]->LunID)) == 0) {
2343                                 drv_index = j;
2344                                 drv_found = 1;
2345                                 break;
2346                         }
2347                 }
2348
2349                 /* check if the drive was found already in the array */
2350                 if (!drv_found) {
2351                         drv_index = cciss_add_gendisk(h, lunid, 0);
2352                         if (drv_index == -1)
2353                                 goto freeret;
2354                 }
2355                 cciss_update_drive_info(ctlr, drv_index, first_time,
2356                         via_ioctl);
2357         }               /* end for */
2358
2359 freeret:
2360         kfree(ld_buff);
2361         h->busy_configuring = 0;
2362         /* We return -1 here to tell the ACU that we have registered/updated
2363          * all of the drives that we can and to keep it from calling us
2364          * additional times.
2365          */
2366         return -1;
2367 mem_msg:
2368         printk(KERN_ERR "cciss: out of memory\n");
2369         h->busy_configuring = 0;
2370         goto freeret;
2371 }
2372
2373 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2374 {
2375         /* zero out the disk size info */
2376         drive_info->nr_blocks = 0;
2377         drive_info->block_size = 0;
2378         drive_info->heads = 0;
2379         drive_info->sectors = 0;
2380         drive_info->cylinders = 0;
2381         drive_info->raid_level = -1;
2382         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2383         memset(drive_info->model, 0, sizeof(drive_info->model));
2384         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2385         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2386         /*
2387          * don't clear the LUNID though, we need to remember which
2388          * one this one is.
2389          */
2390 }
2391
2392 /* This function will deregister the disk and it's queue from the
2393  * kernel.  It must be called with the controller lock held and the
2394  * drv structures busy_configuring flag set.  It's parameters are:
2395  *
2396  * disk = This is the disk to be deregistered
2397  * drv  = This is the drive_info_struct associated with the disk to be
2398  *        deregistered.  It contains information about the disk used
2399  *        by the driver.
2400  * clear_all = This flag determines whether or not the disk information
2401  *             is going to be completely cleared out and the highest_lun
2402  *             reset.  Sometimes we want to clear out information about
2403  *             the disk in preparation for re-adding it.  In this case
2404  *             the highest_lun should be left unchanged and the LunID
2405  *             should not be cleared.
2406  * via_ioctl
2407  *    This indicates whether we've reached this path via ioctl.
2408  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2409  *    If this path is reached via ioctl(), then the max_usage_count will
2410  *    be 1, as the process calling ioctl() has got to have the device open.
2411  *    If we get here via sysfs, then the max usage count will be zero.
2412 */
2413 static int deregister_disk(ctlr_info_t *h, int drv_index,
2414                            int clear_all, int via_ioctl)
2415 {
2416         int i;
2417         struct gendisk *disk;
2418         drive_info_struct *drv;
2419         int recalculate_highest_lun;
2420
2421         if (!capable(CAP_SYS_RAWIO))
2422                 return -EPERM;
2423
2424         drv = h->drv[drv_index];
2425         disk = h->gendisk[drv_index];
2426
2427         /* make sure logical volume is NOT is use */
2428         if (clear_all || (h->gendisk[0] == disk)) {
2429                 if (drv->usage_count > via_ioctl)
2430                         return -EBUSY;
2431         } else if (drv->usage_count > 0)
2432                 return -EBUSY;
2433
2434         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2435
2436         /* invalidate the devices and deregister the disk.  If it is disk
2437          * zero do not deregister it but just zero out it's values.  This
2438          * allows us to delete disk zero but keep the controller registered.
2439          */
2440         if (h->gendisk[0] != disk) {
2441                 struct request_queue *q = disk->queue;
2442                 if (disk->flags & GENHD_FL_UP) {
2443                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2444                         del_gendisk(disk);
2445                 }
2446                 if (q)
2447                         blk_cleanup_queue(q);
2448                 /* If clear_all is set then we are deleting the logical
2449                  * drive, not just refreshing its info.  For drives
2450                  * other than disk 0 we will call put_disk.  We do not
2451                  * do this for disk 0 as we need it to be able to
2452                  * configure the controller.
2453                  */
2454                 if (clear_all){
2455                         /* This isn't pretty, but we need to find the
2456                          * disk in our array and NULL our the pointer.
2457                          * This is so that we will call alloc_disk if
2458                          * this index is used again later.
2459                          */
2460                         for (i=0; i < CISS_MAX_LUN; i++){
2461                                 if (h->gendisk[i] == disk) {
2462                                         h->gendisk[i] = NULL;
2463                                         break;
2464                                 }
2465                         }
2466                         put_disk(disk);
2467                 }
2468         } else {
2469                 set_capacity(disk, 0);
2470                 cciss_clear_drive_info(drv);
2471         }
2472
2473         --h->num_luns;
2474
2475         /* if it was the last disk, find the new hightest lun */
2476         if (clear_all && recalculate_highest_lun) {
2477                 int newhighest = -1;
2478                 for (i = 0; i <= h->highest_lun; i++) {
2479                         /* if the disk has size > 0, it is available */
2480                         if (h->drv[i] && h->drv[i]->heads)
2481                                 newhighest = i;
2482                 }
2483                 h->highest_lun = newhighest;
2484         }
2485         return 0;
2486 }
2487
2488 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2489                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2490                 int cmd_type)
2491 {
2492         ctlr_info_t *h = hba[ctlr];
2493         u64bit buff_dma_handle;
2494         int status = IO_OK;
2495
2496         c->cmd_type = CMD_IOCTL_PEND;
2497         c->Header.ReplyQueue = 0;
2498         if (buff != NULL) {
2499                 c->Header.SGList = 1;
2500                 c->Header.SGTotal = 1;
2501         } else {
2502                 c->Header.SGList = 0;
2503                 c->Header.SGTotal = 0;
2504         }
2505         c->Header.Tag.lower = c->busaddr;
2506         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2507
2508         c->Request.Type.Type = cmd_type;
2509         if (cmd_type == TYPE_CMD) {
2510                 switch (cmd) {
2511                 case CISS_INQUIRY:
2512                         /* are we trying to read a vital product page */
2513                         if (page_code != 0) {
2514                                 c->Request.CDB[1] = 0x01;
2515                                 c->Request.CDB[2] = page_code;
2516                         }
2517                         c->Request.CDBLen = 6;
2518                         c->Request.Type.Attribute = ATTR_SIMPLE;
2519                         c->Request.Type.Direction = XFER_READ;
2520                         c->Request.Timeout = 0;
2521                         c->Request.CDB[0] = CISS_INQUIRY;
2522                         c->Request.CDB[4] = size & 0xFF;
2523                         break;
2524                 case CISS_REPORT_LOG:
2525                 case CISS_REPORT_PHYS:
2526                         /* Talking to controller so It's a physical command
2527                            mode = 00 target = 0.  Nothing to write.
2528                          */
2529                         c->Request.CDBLen = 12;
2530                         c->Request.Type.Attribute = ATTR_SIMPLE;
2531                         c->Request.Type.Direction = XFER_READ;
2532                         c->Request.Timeout = 0;
2533                         c->Request.CDB[0] = cmd;
2534                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2535                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2536                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2537                         c->Request.CDB[9] = size & 0xFF;
2538                         break;
2539
2540                 case CCISS_READ_CAPACITY:
2541                         c->Request.CDBLen = 10;
2542                         c->Request.Type.Attribute = ATTR_SIMPLE;
2543                         c->Request.Type.Direction = XFER_READ;
2544                         c->Request.Timeout = 0;
2545                         c->Request.CDB[0] = cmd;
2546                         break;
2547                 case CCISS_READ_CAPACITY_16:
2548                         c->Request.CDBLen = 16;
2549                         c->Request.Type.Attribute = ATTR_SIMPLE;
2550                         c->Request.Type.Direction = XFER_READ;
2551                         c->Request.Timeout = 0;
2552                         c->Request.CDB[0] = cmd;
2553                         c->Request.CDB[1] = 0x10;
2554                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2555                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2556                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2557                         c->Request.CDB[13] = size & 0xFF;
2558                         c->Request.Timeout = 0;
2559                         c->Request.CDB[0] = cmd;
2560                         break;
2561                 case CCISS_CACHE_FLUSH:
2562                         c->Request.CDBLen = 12;
2563                         c->Request.Type.Attribute = ATTR_SIMPLE;
2564                         c->Request.Type.Direction = XFER_WRITE;
2565                         c->Request.Timeout = 0;
2566                         c->Request.CDB[0] = BMIC_WRITE;
2567                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2568                         break;
2569                 case TEST_UNIT_READY:
2570                         c->Request.CDBLen = 6;
2571                         c->Request.Type.Attribute = ATTR_SIMPLE;
2572                         c->Request.Type.Direction = XFER_NONE;
2573                         c->Request.Timeout = 0;
2574                         break;
2575                 default:
2576                         printk(KERN_WARNING
2577                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2578                         return IO_ERROR;
2579                 }
2580         } else if (cmd_type == TYPE_MSG) {
2581                 switch (cmd) {
2582                 case 0: /* ABORT message */
2583                         c->Request.CDBLen = 12;
2584                         c->Request.Type.Attribute = ATTR_SIMPLE;
2585                         c->Request.Type.Direction = XFER_WRITE;
2586                         c->Request.Timeout = 0;
2587                         c->Request.CDB[0] = cmd;        /* abort */
2588                         c->Request.CDB[1] = 0;  /* abort a command */
2589                         /* buff contains the tag of the command to abort */
2590                         memcpy(&c->Request.CDB[4], buff, 8);
2591                         break;
2592                 case 1: /* RESET message */
2593                         c->Request.CDBLen = 16;
2594                         c->Request.Type.Attribute = ATTR_SIMPLE;
2595                         c->Request.Type.Direction = XFER_NONE;
2596                         c->Request.Timeout = 0;
2597                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2598                         c->Request.CDB[0] = cmd;        /* reset */
2599                         c->Request.CDB[1] = 0x03;       /* reset a target */
2600                         break;
2601                 case 3: /* No-Op message */
2602                         c->Request.CDBLen = 1;
2603                         c->Request.Type.Attribute = ATTR_SIMPLE;
2604                         c->Request.Type.Direction = XFER_WRITE;
2605                         c->Request.Timeout = 0;
2606                         c->Request.CDB[0] = cmd;
2607                         break;
2608                 default:
2609                         printk(KERN_WARNING
2610                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2611                         return IO_ERROR;
2612                 }
2613         } else {
2614                 printk(KERN_WARNING
2615                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2616                 return IO_ERROR;
2617         }
2618         /* Fill in the scatter gather information */
2619         if (size > 0) {
2620                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2621                                                              buff, size,
2622                                                              PCI_DMA_BIDIRECTIONAL);
2623                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2624                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2625                 c->SG[0].Len = size;
2626                 c->SG[0].Ext = 0;       /* we are not chaining */
2627         }
2628         return status;
2629 }
2630
2631 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2632 {
2633         switch (c->err_info->ScsiStatus) {
2634         case SAM_STAT_GOOD:
2635                 return IO_OK;
2636         case SAM_STAT_CHECK_CONDITION:
2637                 switch (0xf & c->err_info->SenseInfo[2]) {
2638                 case 0: return IO_OK; /* no sense */
2639                 case 1: return IO_OK; /* recovered error */
2640                 default:
2641                         if (check_for_unit_attention(h, c))
2642                                 return IO_NEEDS_RETRY;
2643                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2644                                 "check condition, sense key = 0x%02x\n",
2645                                 h->ctlr, c->Request.CDB[0],
2646                                 c->err_info->SenseInfo[2]);
2647                 }
2648                 break;
2649         default:
2650                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2651                         "scsi status = 0x%02x\n", h->ctlr,
2652                         c->Request.CDB[0], c->err_info->ScsiStatus);
2653                 break;
2654         }
2655         return IO_ERROR;
2656 }
2657
2658 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2659 {
2660         int return_status = IO_OK;
2661
2662         if (c->err_info->CommandStatus == CMD_SUCCESS)
2663                 return IO_OK;
2664
2665         switch (c->err_info->CommandStatus) {
2666         case CMD_TARGET_STATUS:
2667                 return_status = check_target_status(h, c);
2668                 break;
2669         case CMD_DATA_UNDERRUN:
2670         case CMD_DATA_OVERRUN:
2671                 /* expected for inquiry and report lun commands */
2672                 break;
2673         case CMD_INVALID:
2674                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2675                        "reported invalid\n", c->Request.CDB[0]);
2676                 return_status = IO_ERROR;
2677                 break;
2678         case CMD_PROTOCOL_ERR:
2679                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2680                        "protocol error \n", c->Request.CDB[0]);
2681                 return_status = IO_ERROR;
2682                 break;
2683         case CMD_HARDWARE_ERR:
2684                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2685                        " hardware error\n", c->Request.CDB[0]);
2686                 return_status = IO_ERROR;
2687                 break;
2688         case CMD_CONNECTION_LOST:
2689                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2690                        "connection lost\n", c->Request.CDB[0]);
2691                 return_status = IO_ERROR;
2692                 break;
2693         case CMD_ABORTED:
2694                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2695                        "aborted\n", c->Request.CDB[0]);
2696                 return_status = IO_ERROR;
2697                 break;
2698         case CMD_ABORT_FAILED:
2699                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2700                        "abort failed\n", c->Request.CDB[0]);
2701                 return_status = IO_ERROR;
2702                 break;
2703         case CMD_UNSOLICITED_ABORT:
2704                 printk(KERN_WARNING
2705                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2706                         c->Request.CDB[0]);
2707                 return_status = IO_NEEDS_RETRY;
2708                 break;
2709         default:
2710                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2711                        "unknown status %x\n", c->Request.CDB[0],
2712                        c->err_info->CommandStatus);
2713                 return_status = IO_ERROR;
2714         }
2715         return return_status;
2716 }
2717
2718 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2719         int attempt_retry)
2720 {
2721         DECLARE_COMPLETION_ONSTACK(wait);
2722         u64bit buff_dma_handle;
2723         int return_status = IO_OK;
2724
2725 resend_cmd2:
2726         c->waiting = &wait;
2727         enqueue_cmd_and_start_io(h, c);
2728
2729         wait_for_completion(&wait);
2730
2731         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2732                 goto command_done;
2733
2734         return_status = process_sendcmd_error(h, c);
2735
2736         if (return_status == IO_NEEDS_RETRY &&
2737                 c->retry_count < MAX_CMD_RETRIES) {
2738                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2739                         c->Request.CDB[0]);
2740                 c->retry_count++;
2741                 /* erase the old error information */
2742                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2743                 return_status = IO_OK;
2744                 INIT_COMPLETION(wait);
2745                 goto resend_cmd2;
2746         }
2747
2748 command_done:
2749         /* unlock the buffers from DMA */
2750         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2751         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2752         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2753                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2754         return return_status;
2755 }
2756
2757 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2758                            __u8 page_code, unsigned char scsi3addr[],
2759                         int cmd_type)
2760 {
2761         ctlr_info_t *h = hba[ctlr];
2762         CommandList_struct *c;
2763         int return_status;
2764
2765         c = cmd_alloc(h, 0);
2766         if (!c)
2767                 return -ENOMEM;
2768         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2769                 scsi3addr, cmd_type);
2770         if (return_status == IO_OK)
2771                 return_status = sendcmd_withirq_core(h, c, 1);
2772
2773         cmd_free(h, c, 0);
2774         return return_status;
2775 }
2776
2777 static void cciss_geometry_inquiry(int ctlr, int logvol,
2778                                    sector_t total_size,
2779                                    unsigned int block_size,
2780                                    InquiryData_struct *inq_buff,
2781                                    drive_info_struct *drv)
2782 {
2783         int return_code;
2784         unsigned long t;
2785         unsigned char scsi3addr[8];
2786
2787         memset(inq_buff, 0, sizeof(InquiryData_struct));
2788         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2789         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2790                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2791         if (return_code == IO_OK) {
2792                 if (inq_buff->data_byte[8] == 0xFF) {
2793                         printk(KERN_WARNING
2794                                "cciss: reading geometry failed, volume "
2795                                "does not support reading geometry\n");
2796                         drv->heads = 255;
2797                         drv->sectors = 32;      /* Sectors per track */
2798                         drv->cylinders = total_size + 1;
2799                         drv->raid_level = RAID_UNKNOWN;
2800                 } else {
2801                         drv->heads = inq_buff->data_byte[6];
2802                         drv->sectors = inq_buff->data_byte[7];
2803                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2804                         drv->cylinders += inq_buff->data_byte[5];
2805                         drv->raid_level = inq_buff->data_byte[8];
2806                 }
2807                 drv->block_size = block_size;
2808                 drv->nr_blocks = total_size + 1;
2809                 t = drv->heads * drv->sectors;
2810                 if (t > 1) {
2811                         sector_t real_size = total_size + 1;
2812                         unsigned long rem = sector_div(real_size, t);
2813                         if (rem)
2814                                 real_size++;
2815                         drv->cylinders = real_size;
2816                 }
2817         } else {                /* Get geometry failed */
2818                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2819         }
2820 }
2821
2822 static void
2823 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2824                     unsigned int *block_size)
2825 {
2826         ReadCapdata_struct *buf;
2827         int return_code;
2828         unsigned char scsi3addr[8];
2829
2830         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2831         if (!buf) {
2832                 printk(KERN_WARNING "cciss: out of memory\n");
2833                 return;
2834         }
2835
2836         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2837         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2838                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2839         if (return_code == IO_OK) {
2840                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2841                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2842         } else {                /* read capacity command failed */
2843                 printk(KERN_WARNING "cciss: read capacity failed\n");
2844                 *total_size = 0;
2845                 *block_size = BLOCK_SIZE;
2846         }
2847         kfree(buf);
2848 }
2849
2850 static void cciss_read_capacity_16(int ctlr, int logvol,
2851         sector_t *total_size, unsigned int *block_size)
2852 {
2853         ReadCapdata_struct_16 *buf;
2854         int return_code;
2855         unsigned char scsi3addr[8];
2856
2857         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2858         if (!buf) {
2859                 printk(KERN_WARNING "cciss: out of memory\n");
2860                 return;
2861         }
2862
2863         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2864         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2865                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2866                         0, scsi3addr, TYPE_CMD);
2867         if (return_code == IO_OK) {
2868                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2869                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2870         } else {                /* read capacity command failed */
2871                 printk(KERN_WARNING "cciss: read capacity failed\n");
2872                 *total_size = 0;
2873                 *block_size = BLOCK_SIZE;
2874         }
2875         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2876                (unsigned long long)*total_size+1, *block_size);
2877         kfree(buf);
2878 }
2879
2880 static int cciss_revalidate(struct gendisk *disk)
2881 {
2882         ctlr_info_t *h = get_host(disk);
2883         drive_info_struct *drv = get_drv(disk);
2884         int logvol;
2885         int FOUND = 0;
2886         unsigned int block_size;
2887         sector_t total_size;
2888         InquiryData_struct *inq_buff = NULL;
2889
2890         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2891                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2892                         sizeof(drv->LunID)) == 0) {
2893                         FOUND = 1;
2894                         break;
2895                 }
2896         }
2897
2898         if (!FOUND)
2899                 return 1;
2900
2901         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2902         if (inq_buff == NULL) {
2903                 printk(KERN_WARNING "cciss: out of memory\n");
2904                 return 1;
2905         }
2906         if (h->cciss_read == CCISS_READ_10) {
2907                 cciss_read_capacity(h->ctlr, logvol,
2908                                         &total_size, &block_size);
2909         } else {
2910                 cciss_read_capacity_16(h->ctlr, logvol,
2911                                         &total_size, &block_size);
2912         }
2913         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2914                                inq_buff, drv);
2915
2916         blk_queue_logical_block_size(drv->queue, drv->block_size);
2917         set_capacity(disk, drv->nr_blocks);
2918
2919         kfree(inq_buff);
2920         return 0;
2921 }
2922
2923 /*
2924  * Map (physical) PCI mem into (virtual) kernel space
2925  */
2926 static void __iomem *remap_pci_mem(ulong base, ulong size)
2927 {
2928         ulong page_base = ((ulong) base) & PAGE_MASK;
2929         ulong page_offs = ((ulong) base) - page_base;
2930         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2931
2932         return page_remapped ? (page_remapped + page_offs) : NULL;
2933 }
2934
2935 /*
2936  * Takes jobs of the Q and sends them to the hardware, then puts it on
2937  * the Q to wait for completion.
2938  */
2939 static void start_io(ctlr_info_t *h)
2940 {
2941         CommandList_struct *c;
2942
2943         while (!hlist_empty(&h->reqQ)) {
2944                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2945                 /* can't do anything if fifo is full */
2946                 if ((h->access.fifo_full(h))) {
2947                         printk(KERN_WARNING "cciss: fifo full\n");
2948                         break;
2949                 }
2950
2951                 /* Get the first entry from the Request Q */
2952                 removeQ(c);
2953                 h->Qdepth--;
2954
2955                 /* Tell the controller execute command */
2956                 h->access.submit_command(h, c);
2957
2958                 /* Put job onto the completed Q */
2959                 addQ(&h->cmpQ, c);
2960         }
2961 }
2962
2963 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2964 /* Zeros out the error record and then resends the command back */
2965 /* to the controller */
2966 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2967 {
2968         /* erase the old error information */
2969         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2970
2971         /* add it to software queue and then send it to the controller */
2972         addQ(&h->reqQ, c);
2973         h->Qdepth++;
2974         if (h->Qdepth > h->maxQsinceinit)
2975                 h->maxQsinceinit = h->Qdepth;
2976
2977         start_io(h);
2978 }
2979
2980 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2981         unsigned int msg_byte, unsigned int host_byte,
2982         unsigned int driver_byte)
2983 {
2984         /* inverse of macros in scsi.h */
2985         return (scsi_status_byte & 0xff) |
2986                 ((msg_byte & 0xff) << 8) |
2987                 ((host_byte & 0xff) << 16) |
2988                 ((driver_byte & 0xff) << 24);
2989 }
2990
2991 static inline int evaluate_target_status(ctlr_info_t *h,
2992                         CommandList_struct *cmd, int *retry_cmd)
2993 {
2994         unsigned char sense_key;
2995         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2996         int error_value;
2997
2998         *retry_cmd = 0;
2999         /* If we get in here, it means we got "target status", that is, scsi status */
3000         status_byte = cmd->err_info->ScsiStatus;
3001         driver_byte = DRIVER_OK;
3002         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3003
3004         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3005                 host_byte = DID_PASSTHROUGH;
3006         else
3007                 host_byte = DID_OK;
3008
3009         error_value = make_status_bytes(status_byte, msg_byte,
3010                 host_byte, driver_byte);
3011
3012         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3013                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3014                         printk(KERN_WARNING "cciss: cmd %p "
3015                                "has SCSI Status 0x%x\n",
3016                                cmd, cmd->err_info->ScsiStatus);
3017                 return error_value;
3018         }
3019
3020         /* check the sense key */
3021         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3022         /* no status or recovered error */
3023         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3024             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3025                 error_value = 0;
3026
3027         if (check_for_unit_attention(h, cmd)) {
3028                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3029                 return 0;
3030         }
3031
3032         /* Not SG_IO or similar? */
3033         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3034                 if (error_value != 0)
3035                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3036                                " sense key = 0x%x\n", cmd, sense_key);
3037                 return error_value;
3038         }
3039
3040         /* SG_IO or similar, copy sense data back */
3041         if (cmd->rq->sense) {
3042                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3043                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3044                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3045                         cmd->rq->sense_len);
3046         } else
3047                 cmd->rq->sense_len = 0;
3048
3049         return error_value;
3050 }
3051
3052 /* checks the status of the job and calls complete buffers to mark all
3053  * buffers for the completed job. Note that this function does not need
3054  * to hold the hba/queue lock.
3055  */
3056 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3057                                     int timeout)
3058 {
3059         int retry_cmd = 0;
3060         struct request *rq = cmd->rq;
3061
3062         rq->errors = 0;
3063
3064         if (timeout)
3065                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3066
3067         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3068                 goto after_error_processing;
3069
3070         switch (cmd->err_info->CommandStatus) {
3071         case CMD_TARGET_STATUS:
3072                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3073                 break;
3074         case CMD_DATA_UNDERRUN:
3075                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3076                         printk(KERN_WARNING "cciss: cmd %p has"
3077                                " completed with data underrun "
3078                                "reported\n", cmd);
3079                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3080                 }
3081                 break;
3082         case CMD_DATA_OVERRUN:
3083                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3084                         printk(KERN_WARNING "cciss: cmd %p has"
3085                                " completed with data overrun "
3086                                "reported\n", cmd);
3087                 break;
3088         case CMD_INVALID:
3089                 printk(KERN_WARNING "cciss: cmd %p is "
3090                        "reported invalid\n", cmd);
3091                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3092                         cmd->err_info->CommandStatus, DRIVER_OK,
3093                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3094                                 DID_PASSTHROUGH : DID_ERROR);
3095                 break;
3096         case CMD_PROTOCOL_ERR:
3097                 printk(KERN_WARNING "cciss: cmd %p has "
3098                        "protocol error \n", cmd);
3099                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3100                         cmd->err_info->CommandStatus, DRIVER_OK,
3101                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3102                                 DID_PASSTHROUGH : DID_ERROR);
3103                 break;
3104         case CMD_HARDWARE_ERR:
3105                 printk(KERN_WARNING "cciss: cmd %p had "
3106                        " hardware error\n", cmd);
3107                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3108                         cmd->err_info->CommandStatus, DRIVER_OK,
3109                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3110                                 DID_PASSTHROUGH : DID_ERROR);
3111                 break;
3112         case CMD_CONNECTION_LOST:
3113                 printk(KERN_WARNING "cciss: cmd %p had "
3114                        "connection lost\n", cmd);
3115                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3116                         cmd->err_info->CommandStatus, DRIVER_OK,
3117                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3118                                 DID_PASSTHROUGH : DID_ERROR);
3119                 break;
3120         case CMD_ABORTED:
3121                 printk(KERN_WARNING "cciss: cmd %p was "
3122                        "aborted\n", cmd);
3123                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3124                         cmd->err_info->CommandStatus, DRIVER_OK,
3125                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3126                                 DID_PASSTHROUGH : DID_ABORT);
3127                 break;
3128         case CMD_ABORT_FAILED:
3129                 printk(KERN_WARNING "cciss: cmd %p reports "
3130                        "abort failed\n", cmd);
3131                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3132                         cmd->err_info->CommandStatus, DRIVER_OK,
3133                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3134                                 DID_PASSTHROUGH : DID_ERROR);
3135                 break;
3136         case CMD_UNSOLICITED_ABORT:
3137                 printk(KERN_WARNING "cciss%d: unsolicited "
3138                        "abort %p\n", h->ctlr, cmd);
3139                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3140                         retry_cmd = 1;
3141                         printk(KERN_WARNING
3142                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3143                         cmd->retry_count++;
3144                 } else
3145                         printk(KERN_WARNING
3146                                "cciss%d: %p retried too "
3147                                "many times\n", h->ctlr, cmd);
3148                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3149                         cmd->err_info->CommandStatus, DRIVER_OK,
3150                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3151                                 DID_PASSTHROUGH : DID_ABORT);
3152                 break;
3153         case CMD_TIMEOUT:
3154                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3155                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3156                         cmd->err_info->CommandStatus, DRIVER_OK,
3157                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3158                                 DID_PASSTHROUGH : DID_ERROR);
3159                 break;
3160         default:
3161                 printk(KERN_WARNING "cciss: cmd %p returned "
3162                        "unknown status %x\n", cmd,
3163                        cmd->err_info->CommandStatus);
3164                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3165                         cmd->err_info->CommandStatus, DRIVER_OK,
3166                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3167                                 DID_PASSTHROUGH : DID_ERROR);
3168         }
3169
3170 after_error_processing:
3171
3172         /* We need to return this command */
3173         if (retry_cmd) {
3174                 resend_cciss_cmd(h, cmd);
3175                 return;
3176         }
3177         cmd->rq->completion_data = cmd;
3178         blk_complete_request(cmd->rq);
3179 }
3180
3181 static inline u32 cciss_tag_contains_index(u32 tag)
3182 {
3183 #define DIRECT_LOOKUP_BIT 0x10
3184         return tag & DIRECT_LOOKUP_BIT;
3185 }
3186
3187 static inline u32 cciss_tag_to_index(u32 tag)
3188 {
3189 #define DIRECT_LOOKUP_SHIFT 5
3190         return tag >> DIRECT_LOOKUP_SHIFT;
3191 }
3192
3193 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3194 {
3195 #define CCISS_ERROR_BITS 0x03
3196         return tag & ~CCISS_ERROR_BITS;
3197 }
3198
3199 static inline void cciss_mark_tag_indexed(u32 *tag)
3200 {
3201         *tag |= DIRECT_LOOKUP_BIT;
3202 }
3203
3204 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3205 {
3206         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3207 }
3208
3209 /*
3210  * Get a request and submit it to the controller.
3211  */
3212 static void do_cciss_request(struct request_queue *q)
3213 {
3214         ctlr_info_t *h = q->queuedata;
3215         CommandList_struct *c;
3216         sector_t start_blk;
3217         int seg;
3218         struct request *creq;
3219         u64bit temp64;
3220         struct scatterlist *tmp_sg;
3221         SGDescriptor_struct *curr_sg;
3222         drive_info_struct *drv;
3223         int i, dir;
3224         int sg_index = 0;
3225         int chained = 0;
3226
3227         /* We call start_io here in case there is a command waiting on the
3228          * queue that has not been sent.
3229          */
3230         if (blk_queue_plugged(q))
3231                 goto startio;
3232
3233       queue:
3234         creq = blk_peek_request(q);
3235         if (!creq)
3236                 goto startio;
3237
3238         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3239
3240         if ((c = cmd_alloc(h, 1)) == NULL)
3241                 goto full;
3242
3243         blk_start_request(creq);
3244
3245         tmp_sg = h->scatter_list[c->cmdindex];
3246         spin_unlock_irq(q->queue_lock);
3247
3248         c->cmd_type = CMD_RWREQ;
3249         c->rq = creq;
3250
3251         /* fill in the request */
3252         drv = creq->rq_disk->private_data;
3253         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3254         /* got command from pool, so use the command block index instead */
3255         /* for direct lookups. */
3256         /* The first 2 bits are reserved for controller error reporting. */
3257         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3258         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3259         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3260         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3261         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3262         c->Request.Type.Attribute = ATTR_SIMPLE;
3263         c->Request.Type.Direction =
3264             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3265         c->Request.Timeout = 0; /* Don't time out */
3266         c->Request.CDB[0] =
3267             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3268         start_blk = blk_rq_pos(creq);
3269 #ifdef CCISS_DEBUG
3270         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3271                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3272 #endif                          /* CCISS_DEBUG */
3273
3274         sg_init_table(tmp_sg, h->maxsgentries);
3275         seg = blk_rq_map_sg(q, creq, tmp_sg);
3276
3277         /* get the DMA records for the setup */
3278         if (c->Request.Type.Direction == XFER_READ)
3279                 dir = PCI_DMA_FROMDEVICE;
3280         else
3281                 dir = PCI_DMA_TODEVICE;
3282
3283         curr_sg = c->SG;
3284         sg_index = 0;
3285         chained = 0;
3286
3287         for (i = 0; i < seg; i++) {
3288                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3289                         !chained && ((seg - i) > 1)) {
3290                         /* Point to next chain block. */
3291                         curr_sg = h->cmd_sg_list[c->cmdindex];
3292                         sg_index = 0;
3293                         chained = 1;
3294                 }
3295                 curr_sg[sg_index].Len = tmp_sg[i].length;
3296                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3297                                                 tmp_sg[i].offset,
3298                                                 tmp_sg[i].length, dir);
3299                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3300                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3301                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3302                 ++sg_index;
3303         }
3304         if (chained)
3305                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3306                         (seg - (h->max_cmd_sgentries - 1)) *
3307                                 sizeof(SGDescriptor_struct));
3308
3309         /* track how many SG entries we are using */
3310         if (seg > h->maxSG)
3311                 h->maxSG = seg;
3312
3313 #ifdef CCISS_DEBUG
3314         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3315                         "chained[%d]\n",
3316                         blk_rq_sectors(creq), seg, chained);
3317 #endif                          /* CCISS_DEBUG */
3318
3319         c->Header.SGTotal = seg + chained;
3320         if (seg <= h->max_cmd_sgentries)
3321                 c->Header.SGList = c->Header.SGTotal;
3322         else
3323                 c->Header.SGList = h->max_cmd_sgentries;
3324         set_performant_mode(h, c);
3325
3326         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3327                 if(h->cciss_read == CCISS_READ_10) {
3328                         c->Request.CDB[1] = 0;
3329                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3330                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3331                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3332                         c->Request.CDB[5] = start_blk & 0xff;
3333                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3334                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3335                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3336                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3337                 } else {
3338                         u32 upper32 = upper_32_bits(start_blk);
3339
3340                         c->Request.CDBLen = 16;
3341                         c->Request.CDB[1]= 0;
3342                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3343                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3344                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3345                         c->Request.CDB[5]= upper32 & 0xff;
3346                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3347                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3348                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3349                         c->Request.CDB[9]= start_blk & 0xff;
3350                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3351                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3352                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3353                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3354                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3355                 }
3356         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3357                 c->Request.CDBLen = creq->cmd_len;
3358                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3359         } else {
3360                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3361                 BUG();
3362         }
3363
3364         spin_lock_irq(q->queue_lock);
3365
3366         addQ(&h->reqQ, c);
3367         h->Qdepth++;
3368         if (h->Qdepth > h->maxQsinceinit)
3369                 h->maxQsinceinit = h->Qdepth;
3370
3371         goto queue;
3372 full:
3373         blk_stop_queue(q);
3374 startio:
3375         /* We will already have the driver lock here so not need
3376          * to lock it.
3377          */
3378         start_io(h);
3379 }
3380
3381 static inline unsigned long get_next_completion(ctlr_info_t *h)
3382 {
3383         return h->access.command_completed(h);
3384 }
3385
3386 static inline int interrupt_pending(ctlr_info_t *h)
3387 {
3388         return h->access.intr_pending(h);
3389 }
3390
3391 static inline long interrupt_not_for_us(ctlr_info_t *h)
3392 {
3393         return !(h->msi_vector || h->msix_vector) &&
3394                 ((h->access.intr_pending(h) == 0) ||
3395                 (h->interrupts_enabled == 0));
3396 }
3397
3398 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3399                         u32 raw_tag)
3400 {
3401         if (unlikely(tag_index >= h->nr_cmds)) {
3402                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3403                 return 1;
3404         }
3405         return 0;
3406 }
3407
3408 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3409                                 u32 raw_tag)
3410 {
3411         removeQ(c);
3412         if (likely(c->cmd_type == CMD_RWREQ))
3413                 complete_command(h, c, 0);
3414         else if (c->cmd_type == CMD_IOCTL_PEND)
3415                 complete(c->waiting);
3416 #ifdef CONFIG_CISS_SCSI_TAPE
3417         else if (c->cmd_type == CMD_SCSI)
3418                 complete_scsi_command(c, 0, raw_tag);
3419 #endif
3420 }
3421
3422 static inline u32 next_command(ctlr_info_t *h)
3423 {
3424         u32 a;
3425
3426         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3427                 return h->access.command_completed(h);
3428
3429         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3430                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3431                 (h->reply_pool_head)++;
3432                 h->commands_outstanding--;
3433         } else {
3434                 a = FIFO_EMPTY;
3435         }
3436         /* Check for wraparound */
3437         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3438                 h->reply_pool_head = h->reply_pool;
3439                 h->reply_pool_wraparound ^= 1;
3440         }
3441         return a;
3442 }
3443
3444 /* process completion of an indexed ("direct lookup") command */
3445 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3446 {
3447         u32 tag_index;
3448         CommandList_struct *c;
3449
3450         tag_index = cciss_tag_to_index(raw_tag);
3451         if (bad_tag(h, tag_index, raw_tag))
3452                 return next_command(h);
3453         c = h->cmd_pool + tag_index;
3454         finish_cmd(h, c, raw_tag);
3455         return next_command(h);
3456 }
3457
3458 /* process completion of a non-indexed command */
3459 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3460 {
3461         u32 tag;
3462         CommandList_struct *c = NULL;
3463         struct hlist_node *tmp;
3464         __u32 busaddr_masked, tag_masked;
3465
3466         tag = cciss_tag_discard_error_bits(raw_tag);
3467         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3468                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3469                 tag_masked = cciss_tag_discard_error_bits(tag);
3470                 if (busaddr_masked == tag_masked) {
3471                         finish_cmd(h, c, raw_tag);
3472                         return next_command(h);
3473                 }
3474         }
3475         bad_tag(h, h->nr_cmds + 1, raw_tag);
3476         return next_command(h);
3477 }
3478
3479 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3480 {
3481         ctlr_info_t *h = dev_id;
3482         unsigned long flags;
3483         u32 raw_tag;
3484
3485         if (interrupt_not_for_us(h))
3486                 return IRQ_NONE;
3487         /*
3488          * If there are completed commands in the completion queue,
3489          * we had better do something about it.
3490          */
3491         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3492         while (interrupt_pending(h)) {
3493                 raw_tag = get_next_completion(h);
3494                 while (raw_tag != FIFO_EMPTY) {
3495                         if (cciss_tag_contains_index(raw_tag))
3496                                 raw_tag = process_indexed_cmd(h, raw_tag);
3497                         else
3498                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3499                 }
3500         }
3501
3502         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3503         return IRQ_HANDLED;
3504 }
3505
3506 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3507  * check the interrupt pending register because it is not set.
3508  */
3509 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3510 {
3511         ctlr_info_t *h = dev_id;
3512         unsigned long flags;
3513         u32 raw_tag;
3514
3515         if (interrupt_not_for_us(h))
3516                 return IRQ_NONE;
3517         /*
3518          * If there are completed commands in the completion queue,
3519          * we had better do something about it.
3520          */
3521         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3522         raw_tag = get_next_completion(h);
3523         while (raw_tag != FIFO_EMPTY) {
3524                 if (cciss_tag_contains_index(raw_tag))
3525                         raw_tag = process_indexed_cmd(h, raw_tag);
3526                 else
3527                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3528         }
3529
3530         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3531         return IRQ_HANDLED;
3532 }
3533
3534 /**
3535  * add_to_scan_list() - add controller to rescan queue
3536  * @h:                Pointer to the controller.
3537  *
3538  * Adds the controller to the rescan queue if not already on the queue.
3539  *
3540  * returns 1 if added to the queue, 0 if skipped (could be on the
3541  * queue already, or the controller could be initializing or shutting
3542  * down).
3543  **/
3544 static int add_to_scan_list(struct ctlr_info *h)
3545 {
3546         struct ctlr_info *test_h;
3547         int found = 0;
3548         int ret = 0;
3549
3550         if (h->busy_initializing)
3551                 return 0;
3552
3553         if (!mutex_trylock(&h->busy_shutting_down))
3554                 return 0;
3555
3556         mutex_lock(&scan_mutex);
3557         list_for_each_entry(test_h, &scan_q, scan_list) {
3558                 if (test_h == h) {
3559                         found = 1;
3560                         break;
3561                 }
3562         }
3563         if (!found && !h->busy_scanning) {
3564                 INIT_COMPLETION(h->scan_wait);
3565                 list_add_tail(&h->scan_list, &scan_q);
3566                 ret = 1;
3567         }
3568         mutex_unlock(&scan_mutex);
3569         mutex_unlock(&h->busy_shutting_down);
3570
3571         return ret;
3572 }
3573
3574 /**
3575  * remove_from_scan_list() - remove controller from rescan queue
3576  * @h:                     Pointer to the controller.
3577  *
3578  * Removes the controller from the rescan queue if present. Blocks if
3579  * the controller is currently conducting a rescan.  The controller
3580  * can be in one of three states:
3581  * 1. Doesn't need a scan
3582  * 2. On the scan list, but not scanning yet (we remove it)
3583  * 3. Busy scanning (and not on the list). In this case we want to wait for
3584  *    the scan to complete to make sure the scanning thread for this
3585  *    controller is completely idle.
3586  **/
3587 static void remove_from_scan_list(struct ctlr_info *h)
3588 {
3589         struct ctlr_info *test_h, *tmp_h;
3590
3591         mutex_lock(&scan_mutex);
3592         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3593                 if (test_h == h) { /* state 2. */
3594                         list_del(&h->scan_list);
3595                         complete_all(&h->scan_wait);
3596                         mutex_unlock(&scan_mutex);
3597                         return;
3598                 }
3599         }
3600         if (h->busy_scanning) { /* state 3. */
3601                 mutex_unlock(&scan_mutex);
3602                 wait_for_completion(&h->scan_wait);
3603         } else { /* state 1, nothing to do. */
3604                 mutex_unlock(&scan_mutex);
3605         }
3606 }
3607
3608 /**
3609  * scan_thread() - kernel thread used to rescan controllers
3610  * @data:        Ignored.
3611  *
3612  * A kernel thread used scan for drive topology changes on
3613  * controllers. The thread processes only one controller at a time
3614  * using a queue.  Controllers are added to the queue using
3615  * add_to_scan_list() and removed from the queue either after done
3616  * processing or using remove_from_scan_list().
3617  *
3618  * returns 0.
3619  **/
3620 static int scan_thread(void *data)
3621 {
3622         struct ctlr_info *h;
3623
3624         while (1) {
3625                 set_current_state(TASK_INTERRUPTIBLE);
3626                 schedule();
3627                 if (kthread_should_stop())
3628                         break;
3629
3630                 while (1) {
3631                         mutex_lock(&scan_mutex);
3632                         if (list_empty(&scan_q)) {
3633                                 mutex_unlock(&scan_mutex);
3634                                 break;
3635                         }
3636
3637                         h = list_entry(scan_q.next,
3638                                        struct ctlr_info,
3639                                        scan_list);
3640                         list_del(&h->scan_list);
3641                         h->busy_scanning = 1;
3642                         mutex_unlock(&scan_mutex);
3643
3644                         rebuild_lun_table(h, 0, 0);
3645                         complete_all(&h->scan_wait);
3646                         mutex_lock(&scan_mutex);
3647                         h->busy_scanning = 0;
3648                         mutex_unlock(&scan_mutex);
3649                 }
3650         }
3651
3652         return 0;
3653 }
3654
3655 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3656 {
3657         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3658                 return 0;
3659
3660         switch (c->err_info->SenseInfo[12]) {
3661         case STATE_CHANGED:
3662                 printk(KERN_WARNING "cciss%d: a state change "
3663                         "detected, command retried\n", h->ctlr);
3664                 return 1;
3665         break;
3666         case LUN_FAILED:
3667                 printk(KERN_WARNING "cciss%d: LUN failure "
3668                         "detected, action required\n", h->ctlr);
3669                 return 1;
3670         break;
3671         case REPORT_LUNS_CHANGED:
3672                 printk(KERN_WARNING "cciss%d: report LUN data "
3673                         "changed\n", h->ctlr);
3674         /*
3675          * Here, we could call add_to_scan_list and wake up the scan thread,
3676          * except that it's quite likely that we will get more than one
3677          * REPORT_LUNS_CHANGED condition in quick succession, which means
3678          * that those which occur after the first one will likely happen
3679          * *during* the scan_thread's rescan.  And the rescan code is not
3680          * robust enough to restart in the middle, undoing what it has already
3681          * done, and it's not clear that it's even possible to do this, since
3682          * part of what it does is notify the block layer, which starts
3683          * doing it's own i/o to read partition tables and so on, and the
3684          * driver doesn't have visibility to know what might need undoing.
3685          * In any event, if possible, it is horribly complicated to get right
3686          * so we just don't do it for now.
3687          *
3688          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3689          */
3690                 return 1;
3691         break;
3692         case POWER_OR_RESET:
3693                 printk(KERN_WARNING "cciss%d: a power on "
3694                         "or device reset detected\n", h->ctlr);
3695                 return 1;
3696         break;
3697         case UNIT_ATTENTION_CLEARED:
3698                 printk(KERN_WARNING "cciss%d: unit attention "
3699                     "cleared by another initiator\n", h->ctlr);
3700                 return 1;
3701         break;
3702         default:
3703                 printk(KERN_WARNING "cciss%d: unknown "
3704                         "unit attention detected\n", h->ctlr);
3705                                 return 1;
3706         }
3707 }
3708
3709 /*
3710  *  We cannot read the structure directly, for portability we must use
3711  *   the io functions.
3712  *   This is for debug only.
3713  */
3714 #ifdef CCISS_DEBUG
3715 static void print_cfg_table(CfgTable_struct *tb)
3716 {
3717         int i;
3718         char temp_name[17];
3719
3720         printk("Controller Configuration information\n");
3721         printk("------------------------------------\n");
3722         for (i = 0; i < 4; i++)
3723                 temp_name[i] = readb(&(tb->Signature[i]));
3724         temp_name[4] = '\0';
3725         printk("   Signature = %s\n", temp_name);
3726         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3727         printk("   Transport methods supported = 0x%x\n",
3728                readl(&(tb->TransportSupport)));
3729         printk("   Transport methods active = 0x%x\n",
3730                readl(&(tb->TransportActive)));
3731         printk("   Requested transport Method = 0x%x\n",
3732                readl(&(tb->HostWrite.TransportRequest)));
3733         printk("   Coalesce Interrupt Delay = 0x%x\n",
3734                readl(&(tb->HostWrite.CoalIntDelay)));
3735         printk("   Coalesce Interrupt Count = 0x%x\n",
3736                readl(&(tb->HostWrite.CoalIntCount)));
3737         printk("   Max outstanding commands = 0x%d\n",
3738                readl(&(tb->CmdsOutMax)));
3739         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3740         for (i = 0; i < 16; i++)
3741                 temp_name[i] = readb(&(tb->ServerName[i]));
3742         temp_name[16] = '\0';
3743         printk("   Server Name = %s\n", temp_name);
3744         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3745 }
3746 #endif                          /* CCISS_DEBUG */
3747
3748 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3749 {
3750         int i, offset, mem_type, bar_type;
3751         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3752                 return 0;
3753         offset = 0;
3754         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3755                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3756                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3757                         offset += 4;
3758                 else {
3759                         mem_type = pci_resource_flags(pdev, i) &
3760                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3761                         switch (mem_type) {
3762                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3763                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3764                                 offset += 4;    /* 32 bit */
3765                                 break;
3766                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3767                                 offset += 8;
3768                                 break;
3769                         default:        /* reserved in PCI 2.2 */
3770                                 printk(KERN_WARNING
3771                                        "Base address is invalid\n");
3772                                 return -1;
3773                                 break;
3774                         }
3775                 }
3776                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3777                         return i + 1;
3778         }
3779         return -1;
3780 }
3781
3782 /* Fill in bucket_map[], given nsgs (the max number of
3783  * scatter gather elements supported) and bucket[],
3784  * which is an array of 8 integers.  The bucket[] array
3785  * contains 8 different DMA transfer sizes (in 16
3786  * byte increments) which the controller uses to fetch
3787  * commands.  This function fills in bucket_map[], which
3788  * maps a given number of scatter gather elements to one of
3789  * the 8 DMA transfer sizes.  The point of it is to allow the
3790  * controller to only do as much DMA as needed to fetch the
3791  * command, with the DMA transfer size encoded in the lower
3792  * bits of the command address.
3793  */
3794 static void  calc_bucket_map(int bucket[], int num_buckets,
3795         int nsgs, int *bucket_map)
3796 {
3797         int i, j, b, size;
3798
3799         /* even a command with 0 SGs requires 4 blocks */
3800 #define MINIMUM_TRANSFER_BLOCKS 4
3801 #define NUM_BUCKETS 8
3802         /* Note, bucket_map must have nsgs+1 entries. */
3803         for (i = 0; i <= nsgs; i++) {
3804                 /* Compute size of a command with i SG entries */
3805                 size = i + MINIMUM_TRANSFER_BLOCKS;
3806                 b = num_buckets; /* Assume the biggest bucket */
3807                 /* Find the bucket that is just big enough */
3808                 for (j = 0; j < 8; j++) {
3809                         if (bucket[j] >= size) {
3810                                 b = j;
3811                                 break;
3812                         }
3813                 }
3814                 /* for a command with i SG entries, use bucket b. */
3815                 bucket_map[i] = b;
3816         }
3817 }
3818
3819 static void
3820 cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3821 {
3822         int l = 0;
3823         __u32 trans_support;
3824         __u32 trans_offset;
3825                         /*
3826                          *  5 = 1 s/g entry or 4k
3827                          *  6 = 2 s/g entry or 8k
3828                          *  8 = 4 s/g entry or 16k
3829                          * 10 = 6 s/g entry or 24k
3830                          */
3831         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3832         unsigned long register_value;
3833
3834         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3835
3836         /* Attempt to put controller into performant mode if supported */
3837         /* Does board support performant mode? */
3838         trans_support = readl(&(h->cfgtable->TransportSupport));
3839         if (!(trans_support & PERFORMANT_MODE))
3840                 return;
3841
3842         printk(KERN_WARNING "cciss%d: Placing controller into "
3843                                 "performant mode\n", h->ctlr);
3844         /* Performant mode demands commands on a 32 byte boundary
3845          * pci_alloc_consistent aligns on page boundarys already.
3846          * Just need to check if divisible by 32
3847          */
3848         if ((sizeof(CommandList_struct) % 32) != 0) {
3849                 printk(KERN_WARNING "%s %d %s\n",
3850                         "cciss info: command size[",
3851                         (int)sizeof(CommandList_struct),
3852                         "] not divisible by 32, no performant mode..\n");
3853                 return;
3854         }
3855
3856         /* Performant mode ring buffer and supporting data structures */
3857         h->reply_pool = (__u64 *)pci_alloc_consistent(
3858                 h->pdev, h->max_commands * sizeof(__u64),
3859                 &(h->reply_pool_dhandle));
3860
3861         /* Need a block fetch table for performant mode */
3862         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3863                 sizeof(__u32)), GFP_KERNEL);
3864
3865         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3866                 goto clean_up;
3867
3868         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3869
3870         /* Controller spec: zero out this buffer. */
3871         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3872         h->reply_pool_head = h->reply_pool;
3873
3874         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3875         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3876                                 h->blockFetchTable);
3877         writel(bft[0], &h->transtable->BlockFetch0);
3878         writel(bft[1], &h->transtable->BlockFetch1);
3879         writel(bft[2], &h->transtable->BlockFetch2);
3880         writel(bft[3], &h->transtable->BlockFetch3);
3881         writel(bft[4], &h->transtable->BlockFetch4);
3882         writel(bft[5], &h->transtable->BlockFetch5);
3883         writel(bft[6], &h->transtable->BlockFetch6);
3884         writel(bft[7], &h->transtable->BlockFetch7);
3885
3886         /* size of controller ring buffer */
3887         writel(h->max_commands, &h->transtable->RepQSize);
3888         writel(1, &h->transtable->RepQCount);
3889         writel(0, &h->transtable->RepQCtrAddrLow32);
3890         writel(0, &h->transtable->RepQCtrAddrHigh32);
3891         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3892         writel(0, &h->transtable->RepQAddr0High32);
3893         writel(CFGTBL_Trans_Performant,
3894                         &(h->cfgtable->HostWrite.TransportRequest));
3895
3896         h->transMethod = CFGTBL_Trans_Performant;
3897         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3898         /* under certain very rare conditions, this can take awhile.
3899          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3900          * as we enter this code.) */
3901         for (l = 0; l < MAX_CONFIG_WAIT; l++) {
3902                 register_value = readl(h->vaddr + SA5_DOORBELL);
3903                 if (!(register_value & CFGTBL_ChangeReq))
3904                         break;
3905                 /* delay and try again */
3906                 set_current_state(TASK_INTERRUPTIBLE);
3907                 schedule_timeout(10);
3908         }
3909         register_value = readl(&(h->cfgtable->TransportActive));
3910         if (!(register_value & CFGTBL_Trans_Performant)) {
3911                 printk(KERN_WARNING "cciss: unable to get board into"
3912                                         " performant mode\n");
3913                 return;
3914         }
3915
3916         /* Change the access methods to the performant access methods */
3917         h->access = SA5_performant_access;
3918
3919         return;
3920 clean_up:
3921         kfree(h->blockFetchTable);
3922         if (h->reply_pool)
3923                 pci_free_consistent(h->pdev,
3924                                 h->max_commands * sizeof(__u64),
3925                                 h->reply_pool,
3926                                 h->reply_pool_dhandle);
3927         return;
3928
3929 } /* cciss_put_controller_into_performant_mode */
3930
3931 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3932  * controllers that are capable. If not, we use IO-APIC mode.
3933  */
3934
3935 static void __devinit cciss_interrupt_mode(ctlr_info_t *c)
3936 {
3937 #ifdef CONFIG_PCI_MSI
3938         int err;
3939         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3940         {0, 2}, {0, 3}
3941         };
3942
3943         /* Some boards advertise MSI but don't really support it */
3944         if ((c->board_id == 0x40700E11) || (c->board_id == 0x40800E11) ||
3945             (c->board_id == 0x40820E11) || (c->board_id == 0x40830E11))
3946                 goto default_int_mode;
3947
3948         if (pci_find_capability(c->pdev, PCI_CAP_ID_MSIX)) {
3949                 err = pci_enable_msix(c->pdev, cciss_msix_entries, 4);
3950                 if (!err) {
3951                         c->intr[0] = cciss_msix_entries[0].vector;
3952                         c->intr[1] = cciss_msix_entries[1].vector;
3953                         c->intr[2] = cciss_msix_entries[2].vector;
3954                         c->intr[3] = cciss_msix_entries[3].vector;
3955                         c->msix_vector = 1;
3956                         return;
3957                 }
3958                 if (err > 0) {
3959                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3960                                "available\n", err);
3961                         goto default_int_mode;
3962                 } else {
3963                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3964                                err);
3965                         goto default_int_mode;
3966                 }
3967         }
3968         if (pci_find_capability(c->pdev, PCI_CAP_ID_MSI)) {
3969                 if (!pci_enable_msi(c->pdev)) {
3970                         c->msi_vector = 1;
3971                 } else {
3972                         printk(KERN_WARNING "cciss: MSI init failed\n");
3973                 }
3974         }
3975 default_int_mode:
3976 #endif                          /* CONFIG_PCI_MSI */
3977         /* if we get here we're going to use the default interrupt mode */
3978         c->intr[PERF_MODE_INT] = c->pdev->irq;
3979         return;
3980 }
3981
3982 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3983 {
3984         int i;
3985         u32 subsystem_vendor_id, subsystem_device_id;
3986
3987         subsystem_vendor_id = pdev->subsystem_vendor;
3988         subsystem_device_id = pdev->subsystem_device;
3989         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3990                         subsystem_vendor_id;
3991
3992         for (i = 0; i < ARRAY_SIZE(products); i++) {
3993                 /* Stand aside for hpsa driver on request */
3994                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3995                         return -ENODEV;
3996                 if (*board_id == products[i].board_id)
3997                         return i;
3998         }
3999         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4000                 *board_id);
4001         return -ENODEV;
4002 }
4003
4004 static inline bool cciss_board_disabled(ctlr_info_t *h)
4005 {
4006         u16 command;
4007
4008         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4009         return ((command & PCI_COMMAND_MEMORY) == 0);
4010 }
4011
4012 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4013         unsigned long *memory_bar)
4014 {
4015         int i;
4016
4017         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4018                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4019                         /* addressing mode bits already removed */
4020                         *memory_bar = pci_resource_start(pdev, i);
4021                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4022                                 *memory_bar);
4023                         return 0;
4024                 }
4025         dev_warn(&pdev->dev, "no memory BAR found\n");
4026         return -ENODEV;
4027 }
4028
4029 static int __devinit cciss_wait_for_board_ready(ctlr_info_t *h)
4030 {
4031         int i;
4032         u32 scratchpad;
4033
4034         for (i = 0; i < CCISS_BOARD_READY_ITERATIONS; i++) {
4035                 scratchpad = readl(h->vaddr + SA5_SCRATCHPAD_OFFSET);
4036                 if (scratchpad == CCISS_FIRMWARE_READY)
4037                         return 0;
4038                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4039         }
4040         dev_warn(&h->pdev->dev, "board not ready, timed out.\n");
4041         return -ENODEV;
4042 }
4043
4044 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4045 {
4046         u64 cfg_offset;
4047         u32 cfg_base_addr;
4048         u64 cfg_base_addr_index;
4049         u32 trans_offset;
4050
4051         /* get the address index number */
4052         cfg_base_addr = readl(h->vaddr + SA5_CTCFG_OFFSET);
4053         cfg_base_addr &= (u32) 0x0000ffff;
4054         cfg_base_addr_index = find_PCI_BAR_index(h->pdev, cfg_base_addr);
4055         if (cfg_base_addr_index == -1) {
4056                 dev_warn(&h->pdev->dev, "cannot find cfg_base_addr_index\n");
4057                 return -ENODEV;
4058         }
4059         cfg_offset = readl(h->vaddr + SA5_CTMEM_OFFSET);
4060         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4061                                cfg_base_addr_index) + cfg_offset,
4062                                 sizeof(h->cfgtable));
4063         if (!h->cfgtable)
4064                 return -ENOMEM;
4065         /* Find performant mode table. */
4066         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
4067         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4068                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4069                                 sizeof(*h->transtable));
4070         if (!h->transtable)
4071                 return -ENOMEM;
4072         return 0;
4073 }
4074
4075 /* Interrogate the hardware for some limits:
4076  * max commands, max SG elements without chaining, and with chaining,
4077  * SG chain block size, etc.
4078  */
4079 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4080 {
4081         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4082         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4083         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4084         /*
4085          * Limit in-command s/g elements to 32 save dma'able memory.
4086          * Howvever spec says if 0, use 31
4087          */
4088         h->max_cmd_sgentries = 31;
4089         if (h->maxsgentries > 512) {
4090                 h->max_cmd_sgentries = 32;
4091                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4092                 h->maxsgentries--; /* save one for chain pointer */
4093         } else {
4094                 h->maxsgentries = 31; /* default to traditional values */
4095                 h->chainsize = 0;
4096         }
4097 }
4098
4099 static inline bool CISS_signature_present(ctlr_info_t *h)
4100 {
4101         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4102             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4103             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4104             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4105                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4106                 return false;
4107         }
4108         return true;
4109 }
4110
4111 static int __devinit cciss_pci_init(ctlr_info_t *c)
4112 {
4113         int prod_index, err;
4114
4115         prod_index = cciss_lookup_board_id(c->pdev, &c->board_id);
4116         if (prod_index < 0)
4117                 return -ENODEV;
4118         c->product_name = products[prod_index].product_name;
4119         c->access = *(products[prod_index].access);
4120
4121         if (cciss_board_disabled(c)) {
4122                 printk(KERN_WARNING
4123                        "cciss: controller appears to be disabled\n");
4124                 return -ENODEV;
4125         }
4126         err = pci_enable_device(c->pdev);
4127         if (err) {
4128                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
4129                 return err;
4130         }
4131
4132         err = pci_request_regions(c->pdev, "cciss");
4133         if (err) {
4134                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
4135                        "aborting\n");
4136                 return err;
4137         }
4138
4139 #ifdef CCISS_DEBUG
4140         printk(KERN_INFO "command = %x\n", command);
4141         printk(KERN_INFO "irq = %x\n", c->pdev->irq);
4142         printk(KERN_INFO "board_id = %x\n", c->board_id);
4143 #endif                          /* CCISS_DEBUG */
4144
4145 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4146  * else we use the IO-APIC interrupt assigned to us by system ROM.
4147  */
4148         cciss_interrupt_mode(c);
4149         err = cciss_pci_find_memory_BAR(c->pdev, &c->paddr);
4150         if (err)
4151                 goto err_out_free_res;
4152         c->vaddr = remap_pci_mem(c->paddr, 0x250);
4153         if (!c->vaddr) {
4154                 err = -ENOMEM;
4155                 goto err_out_free_res;
4156         }
4157         err = cciss_wait_for_board_ready(c);
4158         if (err)
4159                 goto err_out_free_res;
4160         err = cciss_find_cfgtables(c);
4161         if (err)
4162                 goto err_out_free_res;
4163 #ifdef CCISS_DEBUG
4164         print_cfg_table(c->cfgtable);
4165 #endif                          /* CCISS_DEBUG */
4166         cciss_find_board_params(c);
4167
4168         if (!CISS_signature_present(c)) {
4169                 err = -ENODEV;
4170                 goto err_out_free_res;
4171         }
4172 #ifdef CONFIG_X86
4173         {
4174                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4175                 __u32 prefetch;
4176                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
4177                 prefetch |= 0x100;
4178                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
4179         }
4180 #endif
4181
4182         /* Disabling DMA prefetch and refetch for the P600.
4183          * An ASIC bug may result in accesses to invalid memory addresses.
4184          * We've disabled prefetch for some time now. Testing with XEN
4185          * kernels revealed a bug in the refetch if dom0 resides on a P600.
4186          */
4187         if (c->board_id == 0x3225103C) {
4188                 __u32 dma_prefetch;
4189                 __u32 dma_refetch;
4190                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
4191                 dma_prefetch |= 0x8000;
4192                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
4193                 pci_read_config_dword(c->pdev, PCI_COMMAND_PARITY,
4194                         &dma_refetch);
4195                 dma_refetch |= 0x1;
4196                 pci_write_config_dword(c->pdev, PCI_COMMAND_PARITY,
4197                         dma_refetch);
4198         }
4199
4200 #ifdef CCISS_DEBUG
4201         printk(KERN_WARNING "Trying to put board into Performant mode\n");
4202 #endif                          /* CCISS_DEBUG */
4203         cciss_put_controller_into_performant_mode(c);
4204         return 0;
4205
4206 err_out_free_res:
4207         /*
4208          * Deliberately omit pci_disable_device(): it does something nasty to
4209          * Smart Array controllers that pci_enable_device does not undo
4210          */
4211         if (c->transtable)
4212                 iounmap(c->transtable);
4213         if (c->cfgtable)
4214                 iounmap(c->cfgtable);
4215         if (c->vaddr)
4216                 iounmap(c->vaddr);
4217         pci_release_regions(c->pdev);
4218         return err;
4219 }
4220
4221 /* Function to find the first free pointer into our hba[] array
4222  * Returns -1 if no free entries are left.
4223  */
4224 static int alloc_cciss_hba(void)
4225 {
4226         int i;
4227
4228         for (i = 0; i < MAX_CTLR; i++) {
4229                 if (!hba[i]) {
4230                         ctlr_info_t *p;
4231
4232                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4233                         if (!p)
4234                                 goto Enomem;
4235                         hba[i] = p;
4236                         return i;
4237                 }
4238         }
4239         printk(KERN_WARNING "cciss: This driver supports a maximum"
4240                " of %d controllers.\n", MAX_CTLR);
4241         return -1;
4242 Enomem:
4243         printk(KERN_ERR "cciss: out of memory.\n");
4244         return -1;
4245 }
4246
4247 static void free_hba(int n)
4248 {
4249         ctlr_info_t *h = hba[n];
4250         int i;
4251
4252         hba[n] = NULL;
4253         for (i = 0; i < h->highest_lun + 1; i++)
4254                 if (h->gendisk[i] != NULL)
4255                         put_disk(h->gendisk[i]);
4256         kfree(h);
4257 }
4258
4259 /* Send a message CDB to the firmware. */
4260 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4261 {
4262         typedef struct {
4263                 CommandListHeader_struct CommandHeader;
4264                 RequestBlock_struct Request;
4265                 ErrDescriptor_struct ErrorDescriptor;
4266         } Command;
4267         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4268         Command *cmd;
4269         dma_addr_t paddr64;
4270         uint32_t paddr32, tag;
4271         void __iomem *vaddr;
4272         int i, err;
4273
4274         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4275         if (vaddr == NULL)
4276                 return -ENOMEM;
4277
4278         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4279            CCISS commands, so they must be allocated from the lower 4GiB of
4280            memory. */
4281         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4282         if (err) {
4283                 iounmap(vaddr);
4284                 return -ENOMEM;
4285         }
4286
4287         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4288         if (cmd == NULL) {
4289                 iounmap(vaddr);
4290                 return -ENOMEM;
4291         }
4292
4293         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4294            although there's no guarantee, we assume that the address is at
4295            least 4-byte aligned (most likely, it's page-aligned). */
4296         paddr32 = paddr64;
4297
4298         cmd->CommandHeader.ReplyQueue = 0;
4299         cmd->CommandHeader.SGList = 0;
4300         cmd->CommandHeader.SGTotal = 0;
4301         cmd->CommandHeader.Tag.lower = paddr32;
4302         cmd->CommandHeader.Tag.upper = 0;
4303         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4304
4305         cmd->Request.CDBLen = 16;
4306         cmd->Request.Type.Type = TYPE_MSG;
4307         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4308         cmd->Request.Type.Direction = XFER_NONE;
4309         cmd->Request.Timeout = 0; /* Don't time out */
4310         cmd->Request.CDB[0] = opcode;
4311         cmd->Request.CDB[1] = type;
4312         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4313
4314         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4315         cmd->ErrorDescriptor.Addr.upper = 0;
4316         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4317
4318         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4319
4320         for (i = 0; i < 10; i++) {
4321                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4322                 if ((tag & ~3) == paddr32)
4323                         break;
4324                 schedule_timeout_uninterruptible(HZ);
4325         }
4326
4327         iounmap(vaddr);
4328
4329         /* we leak the DMA buffer here ... no choice since the controller could
4330            still complete the command. */
4331         if (i == 10) {
4332                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4333                         opcode, type);
4334                 return -ETIMEDOUT;
4335         }
4336
4337         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4338
4339         if (tag & 2) {
4340                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4341                         opcode, type);
4342                 return -EIO;
4343         }
4344
4345         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4346                 opcode, type);
4347         return 0;
4348 }
4349
4350 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4351 #define cciss_noop(p) cciss_message(p, 3, 0)
4352
4353 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4354 {
4355 /* the #defines are stolen from drivers/pci/msi.h. */
4356 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4357 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4358
4359         int pos;
4360         u16 control = 0;
4361
4362         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4363         if (pos) {
4364                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4365                 if (control & PCI_MSI_FLAGS_ENABLE) {
4366                         printk(KERN_INFO "cciss: resetting MSI\n");
4367                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4368                 }
4369         }
4370
4371         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4372         if (pos) {
4373                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4374                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4375                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4376                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4377                 }
4378         }
4379
4380         return 0;
4381 }
4382
4383 /* This does a hard reset of the controller using PCI power management
4384  * states. */
4385 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4386 {
4387         u16 pmcsr, saved_config_space[32];
4388         int i, pos;
4389
4390         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4391
4392         /* This is very nearly the same thing as
4393
4394            pci_save_state(pci_dev);
4395            pci_set_power_state(pci_dev, PCI_D3hot);
4396            pci_set_power_state(pci_dev, PCI_D0);
4397            pci_restore_state(pci_dev);
4398
4399            but we can't use these nice canned kernel routines on
4400            kexec, because they also check the MSI/MSI-X state in PCI
4401            configuration space and do the wrong thing when it is
4402            set/cleared.  Also, the pci_save/restore_state functions
4403            violate the ordering requirements for restoring the
4404            configuration space from the CCISS document (see the
4405            comment below).  So we roll our own .... */
4406
4407         for (i = 0; i < 32; i++)
4408                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4409
4410         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4411         if (pos == 0) {
4412                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4413                 return -ENODEV;
4414         }
4415
4416         /* Quoting from the Open CISS Specification: "The Power
4417          * Management Control/Status Register (CSR) controls the power
4418          * state of the device.  The normal operating state is D0,
4419          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4420          * the controller, place the interface device in D3 then to
4421          * D0, this causes a secondary PCI reset which will reset the
4422          * controller." */
4423
4424         /* enter the D3hot power management state */
4425         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4426         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4427         pmcsr |= PCI_D3hot;
4428         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4429
4430         schedule_timeout_uninterruptible(HZ >> 1);
4431
4432         /* enter the D0 power management state */
4433         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4434         pmcsr |= PCI_D0;
4435         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4436
4437         schedule_timeout_uninterruptible(HZ >> 1);
4438
4439         /* Restore the PCI configuration space.  The Open CISS
4440          * Specification says, "Restore the PCI Configuration
4441          * Registers, offsets 00h through 60h. It is important to
4442          * restore the command register, 16-bits at offset 04h,
4443          * last. Do not restore the configuration status register,
4444          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4445         for (i = 0; i < 32; i++) {
4446                 if (i == 2 || i == 3)
4447                         continue;
4448                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4449         }
4450         wmb();
4451         pci_write_config_word(pdev, 4, saved_config_space[2]);
4452
4453         return 0;
4454 }
4455
4456 /*
4457  *  This is it.  Find all the controllers and register them.  I really hate
4458  *  stealing all these major device numbers.
4459  *  returns the number of block devices registered.
4460  */
4461 static int __devinit cciss_init_one(struct pci_dev *pdev,
4462                                     const struct pci_device_id *ent)
4463 {
4464         int i;
4465         int j = 0;
4466         int k = 0;
4467         int rc;
4468         int dac, return_code;
4469         InquiryData_struct *inq_buff;
4470
4471         if (reset_devices) {
4472                 /* Reset the controller with a PCI power-cycle */
4473                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4474                         return -ENODEV;
4475
4476                 /* Now try to get the controller to respond to a no-op. Some
4477                    devices (notably the HP Smart Array 5i Controller) need
4478                    up to 30 seconds to respond. */
4479                 for (i=0; i<30; i++) {
4480                         if (cciss_noop(pdev) == 0)
4481                                 break;
4482
4483                         schedule_timeout_uninterruptible(HZ);
4484                 }
4485                 if (i == 30) {
4486                         printk(KERN_ERR "cciss: controller seems dead\n");
4487                         return -EBUSY;
4488                 }
4489         }
4490
4491         i = alloc_cciss_hba();
4492         if (i < 0)
4493                 return -1;
4494
4495         hba[i]->pdev = pdev;
4496         hba[i]->busy_initializing = 1;
4497         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4498         INIT_HLIST_HEAD(&hba[i]->reqQ);
4499         mutex_init(&hba[i]->busy_shutting_down);
4500
4501         if (cciss_pci_init(hba[i]) != 0)
4502                 goto clean_no_release_regions;
4503
4504         sprintf(hba[i]->devname, "cciss%d", i);
4505         hba[i]->ctlr = i;
4506
4507         init_completion(&hba[i]->scan_wait);
4508
4509         if (cciss_create_hba_sysfs_entry(hba[i]))
4510                 goto clean0;
4511
4512         /* configure PCI DMA stuff */
4513         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4514                 dac = 1;
4515         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4516                 dac = 0;
4517         else {
4518                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4519                 goto clean1;
4520         }
4521
4522         /*
4523          * register with the major number, or get a dynamic major number
4524          * by passing 0 as argument.  This is done for greater than
4525          * 8 controller support.
4526          */
4527         if (i < MAX_CTLR_ORIG)
4528                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4529         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4530         if (rc == -EBUSY || rc == -EINVAL) {
4531                 printk(KERN_ERR
4532                        "cciss:  Unable to get major number %d for %s "
4533                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4534                 goto clean1;
4535         } else {
4536                 if (i >= MAX_CTLR_ORIG)
4537                         hba[i]->major = rc;
4538         }
4539
4540         /* make sure the board interrupts are off */
4541         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4542         if (hba[i]->msi_vector || hba[i]->msix_vector) {
4543                 if (request_irq(hba[i]->intr[PERF_MODE_INT],
4544                                 do_cciss_msix_intr,
4545                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4546                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4547                                hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4548                         goto clean2;
4549                 }
4550         } else {
4551                 if (request_irq(hba[i]->intr[PERF_MODE_INT], do_cciss_intx,
4552                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4553                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4554                                hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4555                         goto clean2;
4556                 }
4557         }
4558
4559         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4560                hba[i]->devname, pdev->device, pci_name(pdev),
4561                hba[i]->intr[PERF_MODE_INT], dac ? "" : " not");
4562
4563         hba[i]->cmd_pool_bits =
4564             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4565                         * sizeof(unsigned long), GFP_KERNEL);
4566         hba[i]->cmd_pool = (CommandList_struct *)
4567             pci_alloc_consistent(hba[i]->pdev,
4568                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4569                     &(hba[i]->cmd_pool_dhandle));
4570         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4571             pci_alloc_consistent(hba[i]->pdev,
4572                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4573                     &(hba[i]->errinfo_pool_dhandle));
4574         if ((hba[i]->cmd_pool_bits == NULL)
4575             || (hba[i]->cmd_pool == NULL)
4576             || (hba[i]->errinfo_pool == NULL)) {
4577                 printk(KERN_ERR "cciss: out of memory");
4578                 goto clean4;
4579         }
4580
4581         /* Need space for temp scatter list */
4582         hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4583                                                 sizeof(struct scatterlist *),
4584                                                 GFP_KERNEL);
4585         for (k = 0; k < hba[i]->nr_cmds; k++) {
4586                 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4587                                                         hba[i]->maxsgentries,
4588                                                         GFP_KERNEL);
4589                 if (hba[i]->scatter_list[k] == NULL) {
4590                         printk(KERN_ERR "cciss%d: could not allocate "
4591                                 "s/g lists\n", i);
4592                         goto clean4;
4593                 }
4594         }
4595         hba[i]->cmd_sg_list = cciss_allocate_sg_chain_blocks(hba[i],
4596                 hba[i]->chainsize, hba[i]->nr_cmds);
4597         if (!hba[i]->cmd_sg_list && hba[i]->chainsize > 0)
4598                 goto clean4;
4599
4600         spin_lock_init(&hba[i]->lock);
4601
4602         /* Initialize the pdev driver private data.
4603            have it point to hba[i].  */
4604         pci_set_drvdata(pdev, hba[i]);
4605         /* command and error info recs zeroed out before
4606            they are used */
4607         memset(hba[i]->cmd_pool_bits, 0,
4608                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4609                         * sizeof(unsigned long));
4610
4611         hba[i]->num_luns = 0;
4612         hba[i]->highest_lun = -1;
4613         for (j = 0; j < CISS_MAX_LUN; j++) {
4614                 hba[i]->drv[j] = NULL;
4615                 hba[i]->gendisk[j] = NULL;
4616         }
4617
4618         cciss_scsi_setup(i);
4619
4620         /* Turn the interrupts on so we can service requests */
4621         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4622
4623         /* Get the firmware version */
4624         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4625         if (inq_buff == NULL) {
4626                 printk(KERN_ERR "cciss: out of memory\n");
4627                 goto clean4;
4628         }
4629
4630         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4631                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4632         if (return_code == IO_OK) {
4633                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4634                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4635                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4636                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4637         } else {         /* send command failed */
4638                 printk(KERN_WARNING "cciss: unable to determine firmware"
4639                         " version of controller\n");
4640         }
4641         kfree(inq_buff);
4642
4643         cciss_procinit(i);
4644
4645         hba[i]->cciss_max_sectors = 8192;
4646
4647         rebuild_lun_table(hba[i], 1, 0);
4648         hba[i]->busy_initializing = 0;
4649         return 1;
4650
4651 clean4:
4652         kfree(hba[i]->cmd_pool_bits);
4653         /* Free up sg elements */
4654         for (k = 0; k < hba[i]->nr_cmds; k++)
4655                 kfree(hba[i]->scatter_list[k]);
4656         kfree(hba[i]->scatter_list);
4657         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4658         if (hba[i]->cmd_pool)
4659                 pci_free_consistent(hba[i]->pdev,
4660                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4661                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4662         if (hba[i]->errinfo_pool)
4663                 pci_free_consistent(hba[i]->pdev,
4664                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4665                                     hba[i]->errinfo_pool,
4666                                     hba[i]->errinfo_pool_dhandle);
4667         free_irq(hba[i]->intr[PERF_MODE_INT], hba[i]);
4668 clean2:
4669         unregister_blkdev(hba[i]->major, hba[i]->devname);
4670 clean1:
4671         cciss_destroy_hba_sysfs_entry(hba[i]);
4672 clean0:
4673         pci_release_regions(pdev);
4674 clean_no_release_regions:
4675         hba[i]->busy_initializing = 0;
4676
4677         /*
4678          * Deliberately omit pci_disable_device(): it does something nasty to
4679          * Smart Array controllers that pci_enable_device does not undo
4680          */
4681         pci_set_drvdata(pdev, NULL);
4682         free_hba(i);
4683         return -1;
4684 }
4685
4686 static void cciss_shutdown(struct pci_dev *pdev)
4687 {
4688         ctlr_info_t *h;
4689         char *flush_buf;
4690         int return_code;
4691
4692         h = pci_get_drvdata(pdev);
4693         flush_buf = kzalloc(4, GFP_KERNEL);
4694         if (!flush_buf) {
4695                 printk(KERN_WARNING
4696                         "cciss:%d cache not flushed, out of memory.\n",
4697                         h->ctlr);
4698                 return;
4699         }
4700         /* write all data in the battery backed cache to disk */
4701         memset(flush_buf, 0, 4);
4702         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4703                 4, 0, CTLR_LUNID, TYPE_CMD);
4704         kfree(flush_buf);
4705         if (return_code != IO_OK)
4706                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4707                         h->ctlr);
4708         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4709         free_irq(h->intr[PERF_MODE_INT], h);
4710 }
4711
4712 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4713 {
4714         ctlr_info_t *tmp_ptr;
4715         int i, j;
4716
4717         if (pci_get_drvdata(pdev) == NULL) {
4718                 printk(KERN_ERR "cciss: Unable to remove device \n");
4719                 return;
4720         }
4721
4722         tmp_ptr = pci_get_drvdata(pdev);
4723         i = tmp_ptr->ctlr;
4724         if (hba[i] == NULL) {
4725                 printk(KERN_ERR "cciss: device appears to "
4726                        "already be removed \n");
4727                 return;
4728         }
4729
4730         mutex_lock(&hba[i]->busy_shutting_down);
4731
4732         remove_from_scan_list(hba[i]);
4733         remove_proc_entry(hba[i]->devname, proc_cciss);
4734         unregister_blkdev(hba[i]->major, hba[i]->devname);
4735
4736         /* remove it from the disk list */
4737         for (j = 0; j < CISS_MAX_LUN; j++) {
4738                 struct gendisk *disk = hba[i]->gendisk[j];
4739                 if (disk) {
4740                         struct request_queue *q = disk->queue;
4741
4742                         if (disk->flags & GENHD_FL_UP) {
4743                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4744                                 del_gendisk(disk);
4745                         }
4746                         if (q)
4747                                 blk_cleanup_queue(q);
4748                 }
4749         }
4750
4751 #ifdef CONFIG_CISS_SCSI_TAPE
4752         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4753 #endif
4754
4755         cciss_shutdown(pdev);
4756
4757 #ifdef CONFIG_PCI_MSI
4758         if (hba[i]->msix_vector)
4759                 pci_disable_msix(hba[i]->pdev);
4760         else if (hba[i]->msi_vector)
4761                 pci_disable_msi(hba[i]->pdev);
4762 #endif                          /* CONFIG_PCI_MSI */
4763
4764         iounmap(hba[i]->transtable);
4765         iounmap(hba[i]->cfgtable);
4766         iounmap(hba[i]->vaddr);
4767
4768         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4769                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4770         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4771                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4772         kfree(hba[i]->cmd_pool_bits);
4773         /* Free up sg elements */
4774         for (j = 0; j < hba[i]->nr_cmds; j++)
4775                 kfree(hba[i]->scatter_list[j]);
4776         kfree(hba[i]->scatter_list);
4777         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4778         /*
4779          * Deliberately omit pci_disable_device(): it does something nasty to
4780          * Smart Array controllers that pci_enable_device does not undo
4781          */
4782         pci_release_regions(pdev);
4783         pci_set_drvdata(pdev, NULL);
4784         cciss_destroy_hba_sysfs_entry(hba[i]);
4785         mutex_unlock(&hba[i]->busy_shutting_down);
4786         free_hba(i);
4787 }
4788
4789 static struct pci_driver cciss_pci_driver = {
4790         .name = "cciss",
4791         .probe = cciss_init_one,
4792         .remove = __devexit_p(cciss_remove_one),
4793         .id_table = cciss_pci_device_id,        /* id_table */
4794         .shutdown = cciss_shutdown,
4795 };
4796
4797 /*
4798  *  This is it.  Register the PCI driver information for the cards we control
4799  *  the OS will call our registered routines when it finds one of our cards.
4800  */
4801 static int __init cciss_init(void)
4802 {
4803         int err;
4804
4805         /*
4806          * The hardware requires that commands are aligned on a 64-bit
4807          * boundary. Given that we use pci_alloc_consistent() to allocate an
4808          * array of them, the size must be a multiple of 8 bytes.
4809          */
4810         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4811         printk(KERN_INFO DRIVER_NAME "\n");
4812
4813         err = bus_register(&cciss_bus_type);
4814         if (err)
4815                 return err;
4816
4817         /* Start the scan thread */
4818         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4819         if (IS_ERR(cciss_scan_thread)) {
4820                 err = PTR_ERR(cciss_scan_thread);
4821                 goto err_bus_unregister;
4822         }
4823
4824         /* Register for our PCI devices */
4825         err = pci_register_driver(&cciss_pci_driver);
4826         if (err)
4827                 goto err_thread_stop;
4828
4829         return err;
4830
4831 err_thread_stop:
4832         kthread_stop(cciss_scan_thread);
4833 err_bus_unregister:
4834         bus_unregister(&cciss_bus_type);
4835
4836         return err;
4837 }
4838
4839 static void __exit cciss_cleanup(void)
4840 {
4841         int i;
4842
4843         pci_unregister_driver(&cciss_pci_driver);
4844         /* double check that all controller entrys have been removed */
4845         for (i = 0; i < MAX_CTLR; i++) {
4846                 if (hba[i] != NULL) {
4847                         printk(KERN_WARNING "cciss: had to remove"
4848                                " controller %d\n", i);
4849                         cciss_remove_one(hba[i]->pdev);
4850                 }
4851         }
4852         kthread_stop(cciss_scan_thread);
4853         remove_proc_entry("driver/cciss", NULL);
4854         bus_unregister(&cciss_bus_type);
4855 }
4856
4857 module_init(cciss_init);
4858 module_exit(cciss_cleanup);