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_release(struct gendisk *disk, fmode_t mode);
 182 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 183                        unsigned int cmd, unsigned long arg);
 184 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 185
 186 static int cciss_revalidate(struct gendisk *disk);
 187 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 188 static int deregister_disk(ctlr_info_t *h, int drv_index,
 189                            int clear_all, int via_ioctl);
 190
 191 static void cciss_read_capacity(int ctlr, int logvol,
 192                         sector_t *total_size, unsigned int *block_size);
 193 static void cciss_read_capacity_16(int ctlr, int logvol,
 194                         sector_t *total_size, unsigned int *block_size);
 195 static void cciss_geometry_inquiry(int ctlr, int logvol,
 196                         sector_t total_size,
 197                         unsigned int block_size, InquiryData_struct *inq_buff,
 198                                    drive_info_struct *drv);
 199 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 200                                            __u32);
 201 static void start_io(ctlr_info_t *h);
 202 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 203                         __u8 page_code, unsigned char scsi3addr[],
 204                         int cmd_type);
 205 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 206         int attempt_retry);
 207 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 208
 209 static int add_to_scan_list(struct ctlr_info *h);
 210 static int scan_thread(void *data);
 211 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 212 static void cciss_hba_release(struct device *dev);
 213 static void cciss_device_release(struct device *dev);
 214 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 215 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 216 static inline u32 next_command(ctlr_info_t *h);
 217
 218 /* performant mode helper functions */
 219 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 220                                 int *bucket_map);
 221 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 222
 223 #ifdef CONFIG_PROC_FS
 224 static void cciss_procinit(int i);
 225 #else
 226 static void cciss_procinit(int i)
 227 {
 228 }
 229 #endif                          /* CONFIG_PROC_FS */
 230
 231 #ifdef CONFIG_COMPAT
 232 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 233                               unsigned, unsigned long);
 234 #endif
 235
 236 static const struct block_device_operations cciss_fops = {
 237         .owner = THIS_MODULE,
 238         .open = cciss_open,
 239         .release = cciss_release,
 240         .locked_ioctl = cciss_ioctl,
 241         .getgeo = cciss_getgeo,
 242 #ifdef CONFIG_COMPAT
 243         .compat_ioctl = cciss_compat_ioctl,
 244 #endif
 245         .revalidate_disk = cciss_revalidate,
 246 };
 247
 248 /* set_performant_mode: Modify the tag for cciss performant
 249  * set bit 0 for pull model, bits 3-1 for block fetch
 250  * register number
 251  */
 252 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 253 {
 254         if (likely(h->transMethod == CFGTBL_Trans_Performant))
 255                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 256 }
 257
 258 /*
 259  * Enqueuing and dequeuing functions for cmdlists.
 260  */
 261 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 262 {
 263         hlist_add_head(&c->list, list);
 264 }
 265
 266 static inline void removeQ(CommandList_struct *c)
 267 {
 268         /*
 269          * After kexec/dump some commands might still
 270          * be in flight, which the firmware will try
 271          * to complete. Resetting the firmware doesn't work
 272          * with old fw revisions, so we have to mark
 273          * them off as 'stale' to prevent the driver from
 274          * falling over.
 275          */
 276         if (WARN_ON(hlist_unhashed(&c->list))) {
 277                 c->cmd_type = CMD_MSG_STALE;
 278                 return;
 279         }
 280
 281         hlist_del_init(&c->list);
 282 }
 283
 284 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 285         CommandList_struct *c)
 286 {
 287         unsigned long flags;
 288         set_performant_mode(h, c);
 289         spin_lock_irqsave(&h->lock, flags);
 290         addQ(&h->reqQ, c);
 291         h->Qdepth++;
 292         start_io(h);
 293         spin_unlock_irqrestore(&h->lock, flags);
 294 }
 295
 296 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 297         int nr_cmds)
 298 {
 299         int i;
 300
 301         if (!cmd_sg_list)
 302                 return;
 303         for (i = 0; i < nr_cmds; i++) {
 304                 kfree(cmd_sg_list[i]);
 305                 cmd_sg_list[i] = NULL;
 306         }
 307         kfree(cmd_sg_list);
 308 }
 309
 310 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 311         ctlr_info_t *h, int chainsize, int nr_cmds)
 312 {
 313         int j;
 314         SGDescriptor_struct **cmd_sg_list;
 315
 316         if (chainsize <= 0)
 317                 return NULL;
 318
 319         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 320         if (!cmd_sg_list)
 321                 return NULL;
 322
 323         /* Build up chain blocks for each command */
 324         for (j = 0; j < nr_cmds; j++) {
 325                 /* Need a block of chainsized s/g elements. */
 326                 cmd_sg_list[j] = kmalloc((chainsize *
 327                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 328                 if (!cmd_sg_list[j]) {
 329                         dev_err(&h->pdev->dev, "Cannot get memory "
 330                                 "for s/g chains.\n");
 331                         goto clean;
 332                 }
 333         }
 334         return cmd_sg_list;
 335 clean:
 336         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 337         return NULL;
 338 }
 339
 340 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 341 {
 342         SGDescriptor_struct *chain_sg;
 343         u64bit temp64;
 344
 345         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 346                 return;
 347
 348         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 349         temp64.val32.lower = chain_sg->Addr.lower;
 350         temp64.val32.upper = chain_sg->Addr.upper;
 351         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 352 }
 353
 354 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 355         SGDescriptor_struct *chain_block, int len)
 356 {
 357         SGDescriptor_struct *chain_sg;
 358         u64bit temp64;
 359
 360         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 361         chain_sg->Ext = CCISS_SG_CHAIN;
 362         chain_sg->Len = len;
 363         temp64.val = pci_map_single(h->pdev, chain_block, len,
 364                                 PCI_DMA_TODEVICE);
 365         chain_sg->Addr.lower = temp64.val32.lower;
 366         chain_sg->Addr.upper = temp64.val32.upper;
 367 }
 368
 369 #include "cciss_scsi.c"         /* For SCSI tape support */
 370
 371 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 372         "UNKNOWN"
 373 };
 374 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
 375
 376 #ifdef CONFIG_PROC_FS
 377
 378 /*
 379  * Report information about this controller.
 380  */
 381 #define ENG_GIG 1000000000
 382 #define ENG_GIG_FACTOR (ENG_GIG/512)
 383 #define ENGAGE_SCSI     "engage scsi"
 384
 385 static struct proc_dir_entry *proc_cciss;
 386
 387 static void cciss_seq_show_header(struct seq_file *seq)
 388 {
 389         ctlr_info_t *h = seq->private;
 390
 391         seq_printf(seq, "%s: HP %s Controller\n"
 392                 "Board ID: 0x%08lx\n"
 393                 "Firmware Version: %c%c%c%c\n"
 394                 "IRQ: %d\n"
 395                 "Logical drives: %d\n"
 396                 "Current Q depth: %d\n"
 397                 "Current # commands on controller: %d\n"
 398                 "Max Q depth since init: %d\n"
 399                 "Max # commands on controller since init: %d\n"
 400                 "Max SG entries since init: %d\n",
 401                 h->devname,
 402                 h->product_name,
 403                 (unsigned long)h->board_id,
 404                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 405                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 406                 h->num_luns,
 407                 h->Qdepth, h->commands_outstanding,
 408                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 409
 410 #ifdef CONFIG_CISS_SCSI_TAPE
 411         cciss_seq_tape_report(seq, h->ctlr);
 412 #endif /* CONFIG_CISS_SCSI_TAPE */
 413 }
 414
 415 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 416 {
 417         ctlr_info_t *h = seq->private;
 418         unsigned ctlr = h->ctlr;
 419         unsigned long flags;
 420
 421         /* prevent displaying bogus info during configuration
 422          * or deconfiguration of a logical volume
 423          */
 424         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 425         if (h->busy_configuring) {
 426                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 427                 return ERR_PTR(-EBUSY);
 428         }
 429         h->busy_configuring = 1;
 430         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 431
 432         if (*pos == 0)
 433                 cciss_seq_show_header(seq);
 434
 435         return pos;
 436 }
 437
 438 static int cciss_seq_show(struct seq_file *seq, void *v)
 439 {
 440         sector_t vol_sz, vol_sz_frac;
 441         ctlr_info_t *h = seq->private;
 442         unsigned ctlr = h->ctlr;
 443         loff_t *pos = v;
 444         drive_info_struct *drv = h->drv[*pos];
 445
 446         if (*pos > h->highest_lun)
 447                 return 0;
 448
 449         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 450                 return 0;
 451
 452         if (drv->heads == 0)
 453                 return 0;
 454
 455         vol_sz = drv->nr_blocks;
 456         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 457         vol_sz_frac *= 100;
 458         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 459
 460         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 461                 drv->raid_level = RAID_UNKNOWN;
 462         seq_printf(seq, "cciss/c%dd%d:"
 463                         "\t%4u.%02uGB\tRAID %s\n",
 464                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 465                         raid_label[drv->raid_level]);
 466         return 0;
 467 }
 468
 469 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 470 {
 471         ctlr_info_t *h = seq->private;
 472
 473         if (*pos > h->highest_lun)
 474                 return NULL;
 475         *pos += 1;
 476
 477         return pos;
 478 }
 479
 480 static void cciss_seq_stop(struct seq_file *seq, void *v)
 481 {
 482         ctlr_info_t *h = seq->private;
 483
 484         /* Only reset h->busy_configuring if we succeeded in setting
 485          * it during cciss_seq_start. */
 486         if (v == ERR_PTR(-EBUSY))
 487                 return;
 488
 489         h->busy_configuring = 0;
 490 }
 491
 492 static const struct seq_operations cciss_seq_ops = {
 493         .start = cciss_seq_start,
 494         .show  = cciss_seq_show,
 495         .next  = cciss_seq_next,
 496         .stop  = cciss_seq_stop,
 497 };
 498
 499 static int cciss_seq_open(struct inode *inode, struct file *file)
 500 {
 501         int ret = seq_open(file, &cciss_seq_ops);
 502         struct seq_file *seq = file->private_data;
 503
 504         if (!ret)
 505                 seq->private = PDE(inode)->data;
 506
 507         return ret;
 508 }
 509
 510 static ssize_t
 511 cciss_proc_write(struct file *file, const char __user *buf,
 512                  size_t length, loff_t *ppos)
 513 {
 514         int err;
 515         char *buffer;
 516
 517 #ifndef CONFIG_CISS_SCSI_TAPE
 518         return -EINVAL;
 519 #endif
 520
 521         if (!buf || length > PAGE_SIZE - 1)
 522                 return -EINVAL;
 523
 524         buffer = (char *)__get_free_page(GFP_KERNEL);
 525         if (!buffer)
 526                 return -ENOMEM;
 527
 528         err = -EFAULT;
 529         if (copy_from_user(buffer, buf, length))
 530                 goto out;
 531         buffer[length] = '\0';
 532
 533 #ifdef CONFIG_CISS_SCSI_TAPE
 534         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 535                 struct seq_file *seq = file->private_data;
 536                 ctlr_info_t *h = seq->private;
 537
 538                 err = cciss_engage_scsi(h->ctlr);
 539                 if (err == 0)
 540                         err = length;
 541         } else
 542 #endif /* CONFIG_CISS_SCSI_TAPE */
 543                 err = -EINVAL;
 544         /* might be nice to have "disengage" too, but it's not
 545            safely possible. (only 1 module use count, lock issues.) */
 546
 547 out:
 548         free_page((unsigned long)buffer);
 549         return err;
 550 }
 551
 552 static const struct file_operations cciss_proc_fops = {
 553         .owner   = THIS_MODULE,
 554         .open    = cciss_seq_open,
 555         .read    = seq_read,
 556         .llseek  = seq_lseek,
 557         .release = seq_release,
 558         .write   = cciss_proc_write,
 559 };
 560
 561 static void __devinit cciss_procinit(int i)
 562 {
 563         struct proc_dir_entry *pde;
 564
 565         if (proc_cciss == NULL)
 566                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 567         if (!proc_cciss)
 568                 return;
 569         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 570                                         S_IROTH, proc_cciss,
 571                                         &cciss_proc_fops, hba[i]);
 572 }
 573 #endif                          /* CONFIG_PROC_FS */
 574
 575 #define MAX_PRODUCT_NAME_LEN 19
 576
 577 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 578 #define to_drv(n) container_of(n, drive_info_struct, dev)
 579
 580 static ssize_t host_store_rescan(struct device *dev,
 581                                  struct device_attribute *attr,
 582                                  const char *buf, size_t count)
 583 {
 584         struct ctlr_info *h = to_hba(dev);
 585
 586         add_to_scan_list(h);
 587         wake_up_process(cciss_scan_thread);
 588         wait_for_completion_interruptible(&h->scan_wait);
 589
 590         return count;
 591 }
 592 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 593
 594 static ssize_t dev_show_unique_id(struct device *dev,
 595                                  struct device_attribute *attr,
 596                                  char *buf)
 597 {
 598         drive_info_struct *drv = to_drv(dev);
 599         struct ctlr_info *h = to_hba(drv->dev.parent);
 600         __u8 sn[16];
 601         unsigned long flags;
 602         int ret = 0;
 603
 604         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 605         if (h->busy_configuring)
 606                 ret = -EBUSY;
 607         else
 608                 memcpy(sn, drv->serial_no, sizeof(sn));
 609         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 610
 611         if (ret)
 612                 return ret;
 613         else
 614                 return snprintf(buf, 16 * 2 + 2,
 615                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 616                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 617                                 sn[0], sn[1], sn[2], sn[3],
 618                                 sn[4], sn[5], sn[6], sn[7],
 619                                 sn[8], sn[9], sn[10], sn[11],
 620                                 sn[12], sn[13], sn[14], sn[15]);
 621 }
 622 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 623
 624 static ssize_t dev_show_vendor(struct device *dev,
 625                                struct device_attribute *attr,
 626                                char *buf)
 627 {
 628         drive_info_struct *drv = to_drv(dev);
 629         struct ctlr_info *h = to_hba(drv->dev.parent);
 630         char vendor[VENDOR_LEN + 1];
 631         unsigned long flags;
 632         int ret = 0;
 633
 634         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 635         if (h->busy_configuring)
 636                 ret = -EBUSY;
 637         else
 638                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 639         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 640
 641         if (ret)
 642                 return ret;
 643         else
 644                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 645 }
 646 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 647
 648 static ssize_t dev_show_model(struct device *dev,
 649                               struct device_attribute *attr,
 650                               char *buf)
 651 {
 652         drive_info_struct *drv = to_drv(dev);
 653         struct ctlr_info *h = to_hba(drv->dev.parent);
 654         char model[MODEL_LEN + 1];
 655         unsigned long flags;
 656         int ret = 0;
 657
 658         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 659         if (h->busy_configuring)
 660                 ret = -EBUSY;
 661         else
 662                 memcpy(model, drv->model, MODEL_LEN + 1);
 663         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 664
 665         if (ret)
 666                 return ret;
 667         else
 668                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 669 }
 670 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 671
 672 static ssize_t dev_show_rev(struct device *dev,
 673                             struct device_attribute *attr,
 674                             char *buf)
 675 {
 676         drive_info_struct *drv = to_drv(dev);
 677         struct ctlr_info *h = to_hba(drv->dev.parent);
 678         char rev[REV_LEN + 1];
 679         unsigned long flags;
 680         int ret = 0;
 681
 682         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 683         if (h->busy_configuring)
 684                 ret = -EBUSY;
 685         else
 686                 memcpy(rev, drv->rev, REV_LEN + 1);
 687         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 688
 689         if (ret)
 690                 return ret;
 691         else
 692                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 693 }
 694 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 695
 696 static ssize_t cciss_show_lunid(struct device *dev,
 697                                 struct device_attribute *attr, char *buf)
 698 {
 699         drive_info_struct *drv = to_drv(dev);
 700         struct ctlr_info *h = to_hba(drv->dev.parent);
 701         unsigned long flags;
 702         unsigned char lunid[8];
 703
 704         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 705         if (h->busy_configuring) {
 706                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 707                 return -EBUSY;
 708         }
 709         if (!drv->heads) {
 710                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 711                 return -ENOTTY;
 712         }
 713         memcpy(lunid, drv->LunID, sizeof(lunid));
 714         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 715         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 716                 lunid[0], lunid[1], lunid[2], lunid[3],
 717                 lunid[4], lunid[5], lunid[6], lunid[7]);
 718 }
 719 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 720
 721 static ssize_t cciss_show_raid_level(struct device *dev,
 722                                      struct device_attribute *attr, char *buf)
 723 {
 724         drive_info_struct *drv = to_drv(dev);
 725         struct ctlr_info *h = to_hba(drv->dev.parent);
 726         int raid;
 727         unsigned long flags;
 728
 729         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 730         if (h->busy_configuring) {
 731                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 732                 return -EBUSY;
 733         }
 734         raid = drv->raid_level;
 735         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 736         if (raid < 0 || raid > RAID_UNKNOWN)
 737                 raid = RAID_UNKNOWN;
 738
 739         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 740                         raid_label[raid]);
 741 }
 742 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 743
 744 static ssize_t cciss_show_usage_count(struct device *dev,
 745                                       struct device_attribute *attr, char *buf)
 746 {
 747         drive_info_struct *drv = to_drv(dev);
 748         struct ctlr_info *h = to_hba(drv->dev.parent);
 749         unsigned long flags;
 750         int count;
 751
 752         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 753         if (h->busy_configuring) {
 754                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 755                 return -EBUSY;
 756         }
 757         count = drv->usage_count;
 758         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 759         return snprintf(buf, 20, "%d\n", count);
 760 }
 761 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 762
 763 static struct attribute *cciss_host_attrs[] = {
 764         &dev_attr_rescan.attr,
 765         NULL
 766 };
 767
 768 static struct attribute_group cciss_host_attr_group = {
 769         .attrs = cciss_host_attrs,
 770 };
 771
 772 static const struct attribute_group *cciss_host_attr_groups[] = {
 773         &cciss_host_attr_group,
 774         NULL
 775 };
 776
 777 static struct device_type cciss_host_type = {
 778         .name           = "cciss_host",
 779         .groups         = cciss_host_attr_groups,
 780         .release        = cciss_hba_release,
 781 };
 782
 783 static struct attribute *cciss_dev_attrs[] = {
 784         &dev_attr_unique_id.attr,
 785         &dev_attr_model.attr,
 786         &dev_attr_vendor.attr,
 787         &dev_attr_rev.attr,
 788         &dev_attr_lunid.attr,
 789         &dev_attr_raid_level.attr,
 790         &dev_attr_usage_count.attr,
 791         NULL
 792 };
 793
 794 static struct attribute_group cciss_dev_attr_group = {
 795         .attrs = cciss_dev_attrs,
 796 };
 797
 798 static const struct attribute_group *cciss_dev_attr_groups[] = {
 799         &cciss_dev_attr_group,
 800         NULL
 801 };
 802
 803 static struct device_type cciss_dev_type = {
 804         .name           = "cciss_device",
 805         .groups         = cciss_dev_attr_groups,
 806         .release        = cciss_device_release,
 807 };
 808
 809 static struct bus_type cciss_bus_type = {
 810         .name           = "cciss",
 811 };
 812
 813 /*
 814  * cciss_hba_release is called when the reference count
 815  * of h->dev goes to zero.
 816  */
 817 static void cciss_hba_release(struct device *dev)
 818 {
 819         /*
 820          * nothing to do, but need this to avoid a warning
 821          * about not having a release handler from lib/kref.c.
 822          */
 823 }
 824
 825 /*
 826  * Initialize sysfs entry for each controller.  This sets up and registers
 827  * the 'cciss#' directory for each individual controller under
 828  * /sys/bus/pci/devices/<dev>/.
 829  */
 830 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 831 {
 832         device_initialize(&h->dev);
 833         h->dev.type = &cciss_host_type;
 834         h->dev.bus = &cciss_bus_type;
 835         dev_set_name(&h->dev, "%s", h->devname);
 836         h->dev.parent = &h->pdev->dev;
 837
 838         return device_add(&h->dev);
 839 }
 840
 841 /*
 842  * Remove sysfs entries for an hba.
 843  */
 844 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 845 {
 846         device_del(&h->dev);
 847         put_device(&h->dev); /* final put. */
 848 }
 849
 850 /* cciss_device_release is called when the reference count
 851  * of h->drv[x]dev goes to zero.
 852  */
 853 static void cciss_device_release(struct device *dev)
 854 {
 855         drive_info_struct *drv = to_drv(dev);
 856         kfree(drv);
 857 }
 858
 859 /*
 860  * Initialize sysfs for each logical drive.  This sets up and registers
 861  * the 'c#d#' directory for each individual logical drive under
 862  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 863  * /sys/block/cciss!c#d# to this entry.
 864  */
 865 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 866                                        int drv_index)
 867 {
 868         struct device *dev;
 869
 870         if (h->drv[drv_index]->device_initialized)
 871                 return 0;
 872
 873         dev = &h->drv[drv_index]->dev;
 874         device_initialize(dev);
 875         dev->type = &cciss_dev_type;
 876         dev->bus = &cciss_bus_type;
 877         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 878         dev->parent = &h->dev;
 879         h->drv[drv_index]->device_initialized = 1;
 880         return device_add(dev);
 881 }
 882
 883 /*
 884  * Remove sysfs entries for a logical drive.
 885  */
 886 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 887         int ctlr_exiting)
 888 {
 889         struct device *dev = &h->drv[drv_index]->dev;
 890
 891         /* special case for c*d0, we only destroy it on controller exit */
 892         if (drv_index == 0 && !ctlr_exiting)
 893                 return;
 894
 895         device_del(dev);
 896         put_device(dev); /* the "final" put. */
 897         h->drv[drv_index] = NULL;
 898 }
 899
 900 /*
 901  * For operations that cannot sleep, a command block is allocated at init,
 902  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 903  * which ones are free or in use.  For operations that can wait for kmalloc
 904  * to possible sleep, this routine can be called with get_from_pool set to 0.
 905  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 906  */
 907 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 908 {
 909         CommandList_struct *c;
 910         int i;
 911         u64bit temp64;
 912         dma_addr_t cmd_dma_handle, err_dma_handle;
 913
 914         if (!get_from_pool) {
 915                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 916                         sizeof(CommandList_struct), &cmd_dma_handle);
 917                 if (c == NULL)
 918                         return NULL;
 919                 memset(c, 0, sizeof(CommandList_struct));
 920
 921                 c->cmdindex = -1;
 922
 923                 c->err_info = (ErrorInfo_struct *)
 924                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 925                             &err_dma_handle);
 926
 927                 if (c->err_info == NULL) {
 928                         pci_free_consistent(h->pdev,
 929                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 930                         return NULL;
 931                 }
 932                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 933         } else {                /* get it out of the controllers pool */
 934
 935                 do {
 936                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 937                         if (i == h->nr_cmds)
 938                                 return NULL;
 939                 } while (test_and_set_bit
 940                          (i & (BITS_PER_LONG - 1),
 941                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 942 #ifdef CCISS_DEBUG
 943                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 944 #endif
 945                 c = h->cmd_pool + i;
 946                 memset(c, 0, sizeof(CommandList_struct));
 947                 cmd_dma_handle = h->cmd_pool_dhandle
 948                     + i * sizeof(CommandList_struct);
 949                 c->err_info = h->errinfo_pool + i;
 950                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 951                 err_dma_handle = h->errinfo_pool_dhandle
 952                     + i * sizeof(ErrorInfo_struct);
 953                 h->nr_allocs++;
 954
 955                 c->cmdindex = i;
 956         }
 957
 958         INIT_HLIST_NODE(&c->list);
 959         c->busaddr = (__u32) cmd_dma_handle;
 960         temp64.val = (__u64) err_dma_handle;
 961         c->ErrDesc.Addr.lower = temp64.val32.lower;
 962         c->ErrDesc.Addr.upper = temp64.val32.upper;
 963         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 964
 965         c->ctlr = h->ctlr;
 966         return c;
 967 }
 968
 969 /*
 970  * Frees a command block that was previously allocated with cmd_alloc().
 971  */
 972 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 973 {
 974         int i;
 975         u64bit temp64;
 976
 977         if (!got_from_pool) {
 978                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 979                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 980                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 981                                     c->err_info, (dma_addr_t) temp64.val);
 982                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 983                                     c, (dma_addr_t) c->busaddr);
 984         } else {
 985                 i = c - h->cmd_pool;
 986                 clear_bit(i & (BITS_PER_LONG - 1),
 987                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 988                 h->nr_frees++;
 989         }
 990 }
 991
 992 static inline ctlr_info_t *get_host(struct gendisk *disk)
 993 {
 994         return disk->queue->queuedata;
 995 }
 996
 997 static inline drive_info_struct *get_drv(struct gendisk *disk)
 998 {
 999         return disk->private_data;
1000 }
1001
1002 /*
1003  * Open.  Make sure the device is really there.
1004  */
1005 static int cciss_open(struct block_device *bdev, fmode_t mode)
1006 {
1007         ctlr_info_t *host = get_host(bdev->bd_disk);
1008         drive_info_struct *drv = get_drv(bdev->bd_disk);
1009
1010 #ifdef CCISS_DEBUG
1011         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
1012 #endif                          /* CCISS_DEBUG */
1013
1014         if (drv->busy_configuring)
1015                 return -EBUSY;
1016         /*
1017          * Root is allowed to open raw volume zero even if it's not configured
1018          * so array config can still work. Root is also allowed to open any
1019          * volume that has a LUN ID, so it can issue IOCTL to reread the
1020          * disk information.  I don't think I really like this
1021          * but I'm already using way to many device nodes to claim another one
1022          * for "raw controller".
1023          */
1024         if (drv->heads == 0) {
1025                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1026                         /* if not node 0 make sure it is a partition = 0 */
1027                         if (MINOR(bdev->bd_dev) & 0x0f) {
1028                                 return -ENXIO;
1029                                 /* if it is, make sure we have a LUN ID */
1030                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1031                                 sizeof(drv->LunID))) {
1032                                 return -ENXIO;
1033                         }
1034                 }
1035                 if (!capable(CAP_SYS_ADMIN))
1036                         return -EPERM;
1037         }
1038         drv->usage_count++;
1039         host->usage_count++;
1040         return 0;
1041 }
1042
1043 /*
1044  * Close.  Sync first.
1045  */
1046 static int cciss_release(struct gendisk *disk, fmode_t mode)
1047 {
1048         ctlr_info_t *host = get_host(disk);
1049         drive_info_struct *drv = get_drv(disk);
1050
1051 #ifdef CCISS_DEBUG
1052         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
1053 #endif                          /* CCISS_DEBUG */
1054
1055         drv->usage_count--;
1056         host->usage_count--;
1057         return 0;
1058 }
1059
1060 #ifdef CONFIG_COMPAT
1061
1062 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1063                     unsigned cmd, unsigned long arg)
1064 {
1065         int ret;
1066         lock_kernel();
1067         ret = cciss_ioctl(bdev, mode, cmd, arg);
1068         unlock_kernel();
1069         return ret;
1070 }
1071
1072 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1073                                   unsigned cmd, unsigned long arg);
1074 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1075                                       unsigned cmd, unsigned long arg);
1076
1077 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1078                               unsigned cmd, unsigned long arg)
1079 {
1080         switch (cmd) {
1081         case CCISS_GETPCIINFO:
1082         case CCISS_GETINTINFO:
1083         case CCISS_SETINTINFO:
1084         case CCISS_GETNODENAME:
1085         case CCISS_SETNODENAME:
1086         case CCISS_GETHEARTBEAT:
1087         case CCISS_GETBUSTYPES:
1088         case CCISS_GETFIRMVER:
1089         case CCISS_GETDRIVVER:
1090         case CCISS_REVALIDVOLS:
1091         case CCISS_DEREGDISK:
1092         case CCISS_REGNEWDISK:
1093         case CCISS_REGNEWD:
1094         case CCISS_RESCANDISK:
1095         case CCISS_GETLUNINFO:
1096                 return do_ioctl(bdev, mode, cmd, arg);
1097
1098         case CCISS_PASSTHRU32:
1099                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1100         case CCISS_BIG_PASSTHRU32:
1101                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1102
1103         default:
1104                 return -ENOIOCTLCMD;
1105         }
1106 }
1107
1108 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1109                                   unsigned cmd, unsigned long arg)
1110 {
1111         IOCTL32_Command_struct __user *arg32 =
1112             (IOCTL32_Command_struct __user *) arg;
1113         IOCTL_Command_struct arg64;
1114         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1115         int err;
1116         u32 cp;
1117
1118         err = 0;
1119         err |=
1120             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1121                            sizeof(arg64.LUN_info));
1122         err |=
1123             copy_from_user(&arg64.Request, &arg32->Request,
1124                            sizeof(arg64.Request));
1125         err |=
1126             copy_from_user(&arg64.error_info, &arg32->error_info,
1127                            sizeof(arg64.error_info));
1128         err |= get_user(arg64.buf_size, &arg32->buf_size);
1129         err |= get_user(cp, &arg32->buf);
1130         arg64.buf = compat_ptr(cp);
1131         err |= copy_to_user(p, &arg64, sizeof(arg64));
1132
1133         if (err)
1134                 return -EFAULT;
1135
1136         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1137         if (err)
1138                 return err;
1139         err |=
1140             copy_in_user(&arg32->error_info, &p->error_info,
1141                          sizeof(arg32->error_info));
1142         if (err)
1143                 return -EFAULT;
1144         return err;
1145 }
1146
1147 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1148                                       unsigned cmd, unsigned long arg)
1149 {
1150         BIG_IOCTL32_Command_struct __user *arg32 =
1151             (BIG_IOCTL32_Command_struct __user *) arg;
1152         BIG_IOCTL_Command_struct arg64;
1153         BIG_IOCTL_Command_struct __user *p =
1154             compat_alloc_user_space(sizeof(arg64));
1155         int err;
1156         u32 cp;
1157
1158         err = 0;
1159         err |=
1160             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1161                            sizeof(arg64.LUN_info));
1162         err |=
1163             copy_from_user(&arg64.Request, &arg32->Request,
1164                            sizeof(arg64.Request));
1165         err |=
1166             copy_from_user(&arg64.error_info, &arg32->error_info,
1167                            sizeof(arg64.error_info));
1168         err |= get_user(arg64.buf_size, &arg32->buf_size);
1169         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1170         err |= get_user(cp, &arg32->buf);
1171         arg64.buf = compat_ptr(cp);
1172         err |= copy_to_user(p, &arg64, sizeof(arg64));
1173
1174         if (err)
1175                 return -EFAULT;
1176
1177         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1178         if (err)
1179                 return err;
1180         err |=
1181             copy_in_user(&arg32->error_info, &p->error_info,
1182                          sizeof(arg32->error_info));
1183         if (err)
1184                 return -EFAULT;
1185         return err;
1186 }
1187 #endif
1188
1189 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1190 {
1191         drive_info_struct *drv = get_drv(bdev->bd_disk);
1192
1193         if (!drv->cylinders)
1194                 return -ENXIO;
1195
1196         geo->heads = drv->heads;
1197         geo->sectors = drv->sectors;
1198         geo->cylinders = drv->cylinders;
1199         return 0;
1200 }
1201
1202 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1203 {
1204         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1205                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1206                 (void)check_for_unit_attention(host, c);
1207 }
1208 /*
1209  * ioctl
1210  */
1211 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1212                        unsigned int cmd, unsigned long arg)
1213 {
1214         struct gendisk *disk = bdev->bd_disk;
1215         ctlr_info_t *host = get_host(disk);
1216         drive_info_struct *drv = get_drv(disk);
1217         int ctlr = host->ctlr;
1218         void __user *argp = (void __user *)arg;
1219
1220 #ifdef CCISS_DEBUG
1221         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1222 #endif                          /* CCISS_DEBUG */
1223
1224         switch (cmd) {
1225         case CCISS_GETPCIINFO:
1226                 {
1227                         cciss_pci_info_struct pciinfo;
1228
1229                         if (!arg)
1230                                 return -EINVAL;
1231                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1232                         pciinfo.bus = host->pdev->bus->number;
1233                         pciinfo.dev_fn = host->pdev->devfn;
1234                         pciinfo.board_id = host->board_id;
1235                         if (copy_to_user
1236                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1237                                 return -EFAULT;
1238                         return 0;
1239                 }
1240         case CCISS_GETINTINFO:
1241                 {
1242                         cciss_coalint_struct intinfo;
1243                         if (!arg)
1244                                 return -EINVAL;
1245                         intinfo.delay =
1246                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1247                         intinfo.count =
1248                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1249                         if (copy_to_user
1250                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1251                                 return -EFAULT;
1252                         return 0;
1253                 }
1254         case CCISS_SETINTINFO:
1255                 {
1256                         cciss_coalint_struct intinfo;
1257                         unsigned long flags;
1258                         int i;
1259
1260                         if (!arg)
1261                                 return -EINVAL;
1262                         if (!capable(CAP_SYS_ADMIN))
1263                                 return -EPERM;
1264                         if (copy_from_user
1265                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1266                                 return -EFAULT;
1267                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1268                         {
1269 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1270                                 return -EINVAL;
1271                         }
1272                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1273                         /* Update the field, and then ring the doorbell */
1274                         writel(intinfo.delay,
1275                                &(host->cfgtable->HostWrite.CoalIntDelay));
1276                         writel(intinfo.count,
1277                                &(host->cfgtable->HostWrite.CoalIntCount));
1278                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1279
1280                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1281                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1282                                       & CFGTBL_ChangeReq))
1283                                         break;
1284                                 /* delay and try again */
1285                                 udelay(1000);
1286                         }
1287                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1288                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1289                                 return -EAGAIN;
1290                         return 0;
1291                 }
1292         case CCISS_GETNODENAME:
1293                 {
1294                         NodeName_type NodeName;
1295                         int i;
1296
1297                         if (!arg)
1298                                 return -EINVAL;
1299                         for (i = 0; i < 16; i++)
1300                                 NodeName[i] =
1301                                     readb(&host->cfgtable->ServerName[i]);
1302                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1303                                 return -EFAULT;
1304                         return 0;
1305                 }
1306         case CCISS_SETNODENAME:
1307                 {
1308                         NodeName_type NodeName;
1309                         unsigned long flags;
1310                         int i;
1311
1312                         if (!arg)
1313                                 return -EINVAL;
1314                         if (!capable(CAP_SYS_ADMIN))
1315                                 return -EPERM;
1316
1317                         if (copy_from_user
1318                             (NodeName, argp, sizeof(NodeName_type)))
1319                                 return -EFAULT;
1320
1321                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1322
1323                         /* Update the field, and then ring the doorbell */
1324                         for (i = 0; i < 16; i++)
1325                                 writeb(NodeName[i],
1326                                        &host->cfgtable->ServerName[i]);
1327
1328                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1329
1330                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1331                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1332                                       & CFGTBL_ChangeReq))
1333                                         break;
1334                                 /* delay and try again */
1335                                 udelay(1000);
1336                         }
1337                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1338                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1339                                 return -EAGAIN;
1340                         return 0;
1341                 }
1342
1343         case CCISS_GETHEARTBEAT:
1344                 {
1345                         Heartbeat_type heartbeat;
1346
1347                         if (!arg)
1348                                 return -EINVAL;
1349                         heartbeat = readl(&host->cfgtable->HeartBeat);
1350                         if (copy_to_user
1351                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1352                                 return -EFAULT;
1353                         return 0;
1354                 }
1355         case CCISS_GETBUSTYPES:
1356                 {
1357                         BusTypes_type BusTypes;
1358
1359                         if (!arg)
1360                                 return -EINVAL;
1361                         BusTypes = readl(&host->cfgtable->BusTypes);
1362                         if (copy_to_user
1363                             (argp, &BusTypes, sizeof(BusTypes_type)))
1364                                 return -EFAULT;
1365                         return 0;
1366                 }
1367         case CCISS_GETFIRMVER:
1368                 {
1369                         FirmwareVer_type firmware;
1370
1371                         if (!arg)
1372                                 return -EINVAL;
1373                         memcpy(firmware, host->firm_ver, 4);
1374
1375                         if (copy_to_user
1376                             (argp, firmware, sizeof(FirmwareVer_type)))
1377                                 return -EFAULT;
1378                         return 0;
1379                 }
1380         case CCISS_GETDRIVVER:
1381                 {
1382                         DriverVer_type DriverVer = DRIVER_VERSION;
1383
1384                         if (!arg)
1385                                 return -EINVAL;
1386
1387                         if (copy_to_user
1388                             (argp, &DriverVer, sizeof(DriverVer_type)))
1389                                 return -EFAULT;
1390                         return 0;
1391                 }
1392
1393         case CCISS_DEREGDISK:
1394         case CCISS_REGNEWD:
1395         case CCISS_REVALIDVOLS:
1396                 return rebuild_lun_table(host, 0, 1);
1397
1398         case CCISS_GETLUNINFO:{
1399                         LogvolInfo_struct luninfo;
1400
1401                         memcpy(&luninfo.LunID, drv->LunID,
1402                                 sizeof(luninfo.LunID));
1403                         luninfo.num_opens = drv->usage_count;
1404                         luninfo.num_parts = 0;
1405                         if (copy_to_user(argp, &luninfo,
1406                                          sizeof(LogvolInfo_struct)))
1407                                 return -EFAULT;
1408                         return 0;
1409                 }
1410         case CCISS_PASSTHRU:
1411                 {
1412                         IOCTL_Command_struct iocommand;
1413                         CommandList_struct *c;
1414                         char *buff = NULL;
1415                         u64bit temp64;
1416                         DECLARE_COMPLETION_ONSTACK(wait);
1417
1418                         if (!arg)
1419                                 return -EINVAL;
1420
1421                         if (!capable(CAP_SYS_RAWIO))
1422                                 return -EPERM;
1423
1424                         if (copy_from_user
1425                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1426                                 return -EFAULT;
1427                         if ((iocommand.buf_size < 1) &&
1428                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1429                                 return -EINVAL;
1430                         }
1431 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1432                         /* Check kmalloc limits */
1433                         if (iocommand.buf_size > 128000)
1434                                 return -EINVAL;
1435 #endif
1436                         if (iocommand.buf_size > 0) {
1437                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1438                                 if (buff == NULL)
1439                                         return -EFAULT;
1440                         }
1441                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1442                                 /* Copy the data into the buffer we created */
1443                                 if (copy_from_user
1444                                     (buff, iocommand.buf, iocommand.buf_size)) {
1445                                         kfree(buff);
1446                                         return -EFAULT;
1447                                 }
1448                         } else {
1449                                 memset(buff, 0, iocommand.buf_size);
1450                         }
1451                         if ((c = cmd_alloc(host, 0)) == NULL) {
1452                                 kfree(buff);
1453                                 return -ENOMEM;
1454                         }
1455                         /* Fill in the command type */
1456                         c->cmd_type = CMD_IOCTL_PEND;
1457                         /* Fill in Command Header */
1458                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1459                         if (iocommand.buf_size > 0) /* buffer to fill */
1460                         {
1461                                 c->Header.SGList = 1;
1462                                 c->Header.SGTotal = 1;
1463                         } else /* no buffers to fill */
1464                         {
1465                                 c->Header.SGList = 0;
1466                                 c->Header.SGTotal = 0;
1467                         }
1468                         c->Header.LUN = iocommand.LUN_info;
1469                         /* use the kernel address the cmd block for tag */
1470                         c->Header.Tag.lower = c->busaddr;
1471
1472                         /* Fill in Request block */
1473                         c->Request = iocommand.Request;
1474
1475                         /* Fill in the scatter gather information */
1476                         if (iocommand.buf_size > 0) {
1477                                 temp64.val = pci_map_single(host->pdev, buff,
1478                                         iocommand.buf_size,
1479                                         PCI_DMA_BIDIRECTIONAL);
1480                                 c->SG[0].Addr.lower = temp64.val32.lower;
1481                                 c->SG[0].Addr.upper = temp64.val32.upper;
1482                                 c->SG[0].Len = iocommand.buf_size;
1483                                 c->SG[0].Ext = 0;  /* we are not chaining */
1484                         }
1485                         c->waiting = &wait;
1486
1487                         enqueue_cmd_and_start_io(host, c);
1488                         wait_for_completion(&wait);
1489
1490                         /* unlock the buffers from DMA */
1491                         temp64.val32.lower = c->SG[0].Addr.lower;
1492                         temp64.val32.upper = c->SG[0].Addr.upper;
1493                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1494                                          iocommand.buf_size,
1495                                          PCI_DMA_BIDIRECTIONAL);
1496
1497                         check_ioctl_unit_attention(host, c);
1498
1499                         /* Copy the error information out */
1500                         iocommand.error_info = *(c->err_info);
1501                         if (copy_to_user
1502                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1503                                 kfree(buff);
1504                                 cmd_free(host, c, 0);
1505                                 return -EFAULT;
1506                         }
1507
1508                         if (iocommand.Request.Type.Direction == XFER_READ) {
1509                                 /* Copy the data out of the buffer we created */
1510                                 if (copy_to_user
1511                                     (iocommand.buf, buff, iocommand.buf_size)) {
1512                                         kfree(buff);
1513                                         cmd_free(host, c, 0);
1514                                         return -EFAULT;
1515                                 }
1516                         }
1517                         kfree(buff);
1518                         cmd_free(host, c, 0);
1519                         return 0;
1520                 }
1521         case CCISS_BIG_PASSTHRU:{
1522                         BIG_IOCTL_Command_struct *ioc;
1523                         CommandList_struct *c;
1524                         unsigned char **buff = NULL;
1525                         int *buff_size = NULL;
1526                         u64bit temp64;
1527                         BYTE sg_used = 0;
1528                         int status = 0;
1529                         int i;
1530                         DECLARE_COMPLETION_ONSTACK(wait);
1531                         __u32 left;
1532                         __u32 sz;
1533                         BYTE __user *data_ptr;
1534
1535                         if (!arg)
1536                                 return -EINVAL;
1537                         if (!capable(CAP_SYS_RAWIO))
1538                                 return -EPERM;
1539                         ioc = (BIG_IOCTL_Command_struct *)
1540                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1541                         if (!ioc) {
1542                                 status = -ENOMEM;
1543                                 goto cleanup1;
1544                         }
1545                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1546                                 status = -EFAULT;
1547                                 goto cleanup1;
1548                         }
1549                         if ((ioc->buf_size < 1) &&
1550                             (ioc->Request.Type.Direction != XFER_NONE)) {
1551                                 status = -EINVAL;
1552                                 goto cleanup1;
1553                         }
1554                         /* Check kmalloc limits  using all SGs */
1555                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1556                                 status = -EINVAL;
1557                                 goto cleanup1;
1558                         }
1559                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1560                                 status = -EINVAL;
1561                                 goto cleanup1;
1562                         }
1563                         buff =
1564                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1565                         if (!buff) {
1566                                 status = -ENOMEM;
1567                                 goto cleanup1;
1568                         }
1569                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1570                                                    GFP_KERNEL);
1571                         if (!buff_size) {
1572                                 status = -ENOMEM;
1573                                 goto cleanup1;
1574                         }
1575                         left = ioc->buf_size;
1576                         data_ptr = ioc->buf;
1577                         while (left) {
1578                                 sz = (left >
1579                                       ioc->malloc_size) ? ioc->
1580                                     malloc_size : left;
1581                                 buff_size[sg_used] = sz;
1582                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1583                                 if (buff[sg_used] == NULL) {
1584                                         status = -ENOMEM;
1585                                         goto cleanup1;
1586                                 }
1587                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1588                                         if (copy_from_user
1589                                             (buff[sg_used], data_ptr, sz)) {
1590                                                 status = -EFAULT;
1591                                                 goto cleanup1;
1592                                         }
1593                                 } else {
1594                                         memset(buff[sg_used], 0, sz);
1595                                 }
1596                                 left -= sz;
1597                                 data_ptr += sz;
1598                                 sg_used++;
1599                         }
1600                         if ((c = cmd_alloc(host, 0)) == NULL) {
1601                                 status = -ENOMEM;
1602                                 goto cleanup1;
1603                         }
1604                         c->cmd_type = CMD_IOCTL_PEND;
1605                         c->Header.ReplyQueue = 0;
1606
1607                         if (ioc->buf_size > 0) {
1608                                 c->Header.SGList = sg_used;
1609                                 c->Header.SGTotal = sg_used;
1610                         } else {
1611                                 c->Header.SGList = 0;
1612                                 c->Header.SGTotal = 0;
1613                         }
1614                         c->Header.LUN = ioc->LUN_info;
1615                         c->Header.Tag.lower = c->busaddr;
1616
1617                         c->Request = ioc->Request;
1618                         if (ioc->buf_size > 0) {
1619                                 for (i = 0; i < sg_used; i++) {
1620                                         temp64.val =
1621                                             pci_map_single(host->pdev, buff[i],
1622                                                     buff_size[i],
1623                                                     PCI_DMA_BIDIRECTIONAL);
1624                                         c->SG[i].Addr.lower =
1625                                             temp64.val32.lower;
1626                                         c->SG[i].Addr.upper =
1627                                             temp64.val32.upper;
1628                                         c->SG[i].Len = buff_size[i];
1629                                         c->SG[i].Ext = 0;       /* we are not chaining */
1630                                 }
1631                         }
1632                         c->waiting = &wait;
1633                         enqueue_cmd_and_start_io(host, c);
1634                         wait_for_completion(&wait);
1635                         /* unlock the buffers from DMA */
1636                         for (i = 0; i < sg_used; i++) {
1637                                 temp64.val32.lower = c->SG[i].Addr.lower;
1638                                 temp64.val32.upper = c->SG[i].Addr.upper;
1639                                 pci_unmap_single(host->pdev,
1640                                         (dma_addr_t) temp64.val, buff_size[i],
1641                                         PCI_DMA_BIDIRECTIONAL);
1642                         }
1643                         check_ioctl_unit_attention(host, c);
1644                         /* Copy the error information out */
1645                         ioc->error_info = *(c->err_info);
1646                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1647                                 cmd_free(host, c, 0);
1648                                 status = -EFAULT;
1649                                 goto cleanup1;
1650                         }
1651                         if (ioc->Request.Type.Direction == XFER_READ) {
1652                                 /* Copy the data out of the buffer we created */
1653                                 BYTE __user *ptr = ioc->buf;
1654                                 for (i = 0; i < sg_used; i++) {
1655                                         if (copy_to_user
1656                                             (ptr, buff[i], buff_size[i])) {
1657                                                 cmd_free(host, c, 0);
1658                                                 status = -EFAULT;
1659                                                 goto cleanup1;
1660                                         }
1661                                         ptr += buff_size[i];
1662                                 }
1663                         }
1664                         cmd_free(host, c, 0);
1665                         status = 0;
1666                       cleanup1:
1667                         if (buff) {
1668                                 for (i = 0; i < sg_used; i++)
1669                                         kfree(buff[i]);
1670                                 kfree(buff);
1671                         }
1672                         kfree(buff_size);
1673                         kfree(ioc);
1674                         return status;
1675                 }
1676
1677         /* scsi_cmd_ioctl handles these, below, though some are not */
1678         /* very meaningful for cciss.  SG_IO is the main one people want. */
1679
1680         case SG_GET_VERSION_NUM:
1681         case SG_SET_TIMEOUT:
1682         case SG_GET_TIMEOUT:
1683         case SG_GET_RESERVED_SIZE:
1684         case SG_SET_RESERVED_SIZE:
1685         case SG_EMULATED_HOST:
1686         case SG_IO:
1687         case SCSI_IOCTL_SEND_COMMAND:
1688                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1689
1690         /* scsi_cmd_ioctl would normally handle these, below, but */
1691         /* they aren't a good fit for cciss, as CD-ROMs are */
1692         /* not supported, and we don't have any bus/target/lun */
1693         /* which we present to the kernel. */
1694
1695         case CDROM_SEND_PACKET:
1696         case CDROMCLOSETRAY:
1697         case CDROMEJECT:
1698         case SCSI_IOCTL_GET_IDLUN:
1699         case SCSI_IOCTL_GET_BUS_NUMBER:
1700         default:
1701                 return -ENOTTY;
1702         }
1703 }
1704
1705 static void cciss_check_queues(ctlr_info_t *h)
1706 {
1707         int start_queue = h->next_to_run;
1708         int i;
1709
1710         /* check to see if we have maxed out the number of commands that can
1711          * be placed on the queue.  If so then exit.  We do this check here
1712          * in case the interrupt we serviced was from an ioctl and did not
1713          * free any new commands.
1714          */
1715         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1716                 return;
1717
1718         /* We have room on the queue for more commands.  Now we need to queue
1719          * them up.  We will also keep track of the next queue to run so
1720          * that every queue gets a chance to be started first.
1721          */
1722         for (i = 0; i < h->highest_lun + 1; i++) {
1723                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1724                 /* make sure the disk has been added and the drive is real
1725                  * because this can be called from the middle of init_one.
1726                  */
1727                 if (!h->drv[curr_queue])
1728                         continue;
1729                 if (!(h->drv[curr_queue]->queue) ||
1730                         !(h->drv[curr_queue]->heads))
1731                         continue;
1732                 blk_start_queue(h->gendisk[curr_queue]->queue);
1733
1734                 /* check to see if we have maxed out the number of commands
1735                  * that can be placed on the queue.
1736                  */
1737                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1738                         if (curr_queue == start_queue) {
1739                                 h->next_to_run =
1740                                     (start_queue + 1) % (h->highest_lun + 1);
1741                                 break;
1742                         } else {
1743                                 h->next_to_run = curr_queue;
1744                                 break;
1745                         }
1746                 }
1747         }
1748 }
1749
1750 static void cciss_softirq_done(struct request *rq)
1751 {
1752         CommandList_struct *cmd = rq->completion_data;
1753         ctlr_info_t *h = hba[cmd->ctlr];
1754         SGDescriptor_struct *curr_sg = cmd->SG;
1755         u64bit temp64;
1756         unsigned long flags;
1757         int i, ddir;
1758         int sg_index = 0;
1759
1760         if (cmd->Request.Type.Direction == XFER_READ)
1761                 ddir = PCI_DMA_FROMDEVICE;
1762         else
1763                 ddir = PCI_DMA_TODEVICE;
1764
1765         /* command did not need to be retried */
1766         /* unmap the DMA mapping for all the scatter gather elements */
1767         for (i = 0; i < cmd->Header.SGList; i++) {
1768                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1769                         cciss_unmap_sg_chain_block(h, cmd);
1770                         /* Point to the next block */
1771                         curr_sg = h->cmd_sg_list[cmd->cmdindex];
1772                         sg_index = 0;
1773                 }
1774                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1775                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1776                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1777                                 ddir);
1778                 ++sg_index;
1779         }
1780
1781 #ifdef CCISS_DEBUG
1782         printk("Done with %p\n", rq);
1783 #endif                          /* CCISS_DEBUG */
1784
1785         /* set the residual count for pc requests */
1786         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1787                 rq->resid_len = cmd->err_info->ResidualCnt;
1788
1789         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1790
1791         spin_lock_irqsave(&h->lock, flags);
1792         cmd_free(h, cmd, 1);
1793         cciss_check_queues(h);
1794         spin_unlock_irqrestore(&h->lock, flags);
1795 }
1796
1797 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1798         unsigned char scsi3addr[], uint32_t log_unit)
1799 {
1800         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1801                 sizeof(h->drv[log_unit]->LunID));
1802 }
1803
1804 /* This function gets the SCSI vendor, model, and revision of a logical drive
1805  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1806  * they cannot be read.
1807  */
1808 static void cciss_get_device_descr(int ctlr, int logvol,
1809                                    char *vendor, char *model, char *rev)
1810 {
1811         int rc;
1812         InquiryData_struct *inq_buf;
1813         unsigned char scsi3addr[8];
1814
1815         *vendor = '\0';
1816         *model = '\0';
1817         *rev = '\0';
1818
1819         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1820         if (!inq_buf)
1821                 return;
1822
1823         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1824         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1825                         scsi3addr, TYPE_CMD);
1826         if (rc == IO_OK) {
1827                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1828                 vendor[VENDOR_LEN] = '\0';
1829                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1830                 model[MODEL_LEN] = '\0';
1831                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1832                 rev[REV_LEN] = '\0';
1833         }
1834
1835         kfree(inq_buf);
1836         return;
1837 }
1838
1839 /* This function gets the serial number of a logical drive via
1840  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1841  * number cannot be had, for whatever reason, 16 bytes of 0xff
1842  * are returned instead.
1843  */
1844 static void cciss_get_serial_no(int ctlr, int logvol,
1845                                 unsigned char *serial_no, int buflen)
1846 {
1847 #define PAGE_83_INQ_BYTES 64
1848         int rc;
1849         unsigned char *buf;
1850         unsigned char scsi3addr[8];
1851
1852         if (buflen > 16)
1853                 buflen = 16;
1854         memset(serial_no, 0xff, buflen);
1855         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1856         if (!buf)
1857                 return;
1858         memset(serial_no, 0, buflen);
1859         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1860         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1861                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1862         if (rc == IO_OK)
1863                 memcpy(serial_no, &buf[8], buflen);
1864         kfree(buf);
1865         return;
1866 }
1867
1868 /*
1869  * cciss_add_disk sets up the block device queue for a logical drive
1870  */
1871 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1872                                 int drv_index)
1873 {
1874         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1875         if (!disk->queue)
1876                 goto init_queue_failure;
1877         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1878         disk->major = h->major;
1879         disk->first_minor = drv_index << NWD_SHIFT;
1880         disk->fops = &cciss_fops;
1881         if (cciss_create_ld_sysfs_entry(h, drv_index))
1882                 goto cleanup_queue;
1883         disk->private_data = h->drv[drv_index];
1884         disk->driverfs_dev = &h->drv[drv_index]->dev;
1885
1886         /* Set up queue information */
1887         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1888
1889         /* This is a hardware imposed limit. */
1890         blk_queue_max_segments(disk->queue, h->maxsgentries);
1891
1892         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1893
1894         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1895
1896         disk->queue->queuedata = h;
1897
1898         blk_queue_logical_block_size(disk->queue,
1899                                      h->drv[drv_index]->block_size);
1900
1901         /* Make sure all queue data is written out before */
1902         /* setting h->drv[drv_index]->queue, as setting this */
1903         /* allows the interrupt handler to start the queue */
1904         wmb();
1905         h->drv[drv_index]->queue = disk->queue;
1906         add_disk(disk);
1907         return 0;
1908
1909 cleanup_queue:
1910         blk_cleanup_queue(disk->queue);
1911         disk->queue = NULL;
1912 init_queue_failure:
1913         return -1;
1914 }
1915
1916 /* This function will check the usage_count of the drive to be updated/added.
1917  * If the usage_count is zero and it is a heretofore unknown drive, or,
1918  * the drive's capacity, geometry, or serial number has changed,
1919  * then the drive information will be updated and the disk will be
1920  * re-registered with the kernel.  If these conditions don't hold,
1921  * then it will be left alone for the next reboot.  The exception to this
1922  * is disk 0 which will always be left registered with the kernel since it
1923  * is also the controller node.  Any changes to disk 0 will show up on
1924  * the next reboot.
1925  */
1926 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1927         int via_ioctl)
1928 {
1929         ctlr_info_t *h = hba[ctlr];
1930         struct gendisk *disk;
1931         InquiryData_struct *inq_buff = NULL;
1932         unsigned int block_size;
1933         sector_t total_size;
1934         unsigned long flags = 0;
1935         int ret = 0;
1936         drive_info_struct *drvinfo;
1937
1938         /* Get information about the disk and modify the driver structure */
1939         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1940         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1941         if (inq_buff == NULL || drvinfo == NULL)
1942                 goto mem_msg;
1943
1944         /* testing to see if 16-byte CDBs are already being used */
1945         if (h->cciss_read == CCISS_READ_16) {
1946                 cciss_read_capacity_16(h->ctlr, drv_index,
1947                         &total_size, &block_size);
1948
1949         } else {
1950                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1951                 /* if read_capacity returns all F's this volume is >2TB */
1952                 /* in size so we switch to 16-byte CDB's for all */
1953                 /* read/write ops */
1954                 if (total_size == 0xFFFFFFFFULL) {
1955                         cciss_read_capacity_16(ctlr, drv_index,
1956                         &total_size, &block_size);
1957                         h->cciss_read = CCISS_READ_16;
1958                         h->cciss_write = CCISS_WRITE_16;
1959                 } else {
1960                         h->cciss_read = CCISS_READ_10;
1961                         h->cciss_write = CCISS_WRITE_10;
1962                 }
1963         }
1964
1965         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1966                                inq_buff, drvinfo);
1967         drvinfo->block_size = block_size;
1968         drvinfo->nr_blocks = total_size + 1;
1969
1970         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1971                                 drvinfo->model, drvinfo->rev);
1972         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1973                         sizeof(drvinfo->serial_no));
1974         /* Save the lunid in case we deregister the disk, below. */
1975         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1976                 sizeof(drvinfo->LunID));
1977
1978         /* Is it the same disk we already know, and nothing's changed? */
1979         if (h->drv[drv_index]->raid_level != -1 &&
1980                 ((memcmp(drvinfo->serial_no,
1981                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1982                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1983                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1984                 drvinfo->heads == h->drv[drv_index]->heads &&
1985                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1986                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1987                         /* The disk is unchanged, nothing to update */
1988                         goto freeret;
1989
1990         /* If we get here it's not the same disk, or something's changed,
1991          * so we need to * deregister it, and re-register it, if it's not
1992          * in use.
1993          * If the disk already exists then deregister it before proceeding
1994          * (unless it's the first disk (for the controller node).
1995          */
1996         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
1997                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1998                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1999                 h->drv[drv_index]->busy_configuring = 1;
2000                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2001
2002                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2003                  * which keeps the interrupt handler from starting
2004                  * the queue.
2005                  */
2006                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2007         }
2008
2009         /* If the disk is in use return */
2010         if (ret)
2011                 goto freeret;
2012
2013         /* Save the new information from cciss_geometry_inquiry
2014          * and serial number inquiry.  If the disk was deregistered
2015          * above, then h->drv[drv_index] will be NULL.
2016          */
2017         if (h->drv[drv_index] == NULL) {
2018                 drvinfo->device_initialized = 0;
2019                 h->drv[drv_index] = drvinfo;
2020                 drvinfo = NULL; /* so it won't be freed below. */
2021         } else {
2022                 /* special case for cxd0 */
2023                 h->drv[drv_index]->block_size = drvinfo->block_size;
2024                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2025                 h->drv[drv_index]->heads = drvinfo->heads;
2026                 h->drv[drv_index]->sectors = drvinfo->sectors;
2027                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2028                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2029                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2030                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2031                         VENDOR_LEN + 1);
2032                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2033                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2034         }
2035
2036         ++h->num_luns;
2037         disk = h->gendisk[drv_index];
2038         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2039
2040         /* If it's not disk 0 (drv_index != 0)
2041          * or if it was disk 0, but there was previously
2042          * no actual corresponding configured logical drive
2043          * (raid_leve == -1) then we want to update the
2044          * logical drive's information.
2045          */
2046         if (drv_index || first_time) {
2047                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2048                         cciss_free_gendisk(h, drv_index);
2049                         cciss_free_drive_info(h, drv_index);
2050                         printk(KERN_WARNING "cciss:%d could not update "
2051                                 "disk %d\n", h->ctlr, drv_index);
2052                         --h->num_luns;
2053                 }
2054         }
2055
2056 freeret:
2057         kfree(inq_buff);
2058         kfree(drvinfo);
2059         return;
2060 mem_msg:
2061         printk(KERN_ERR "cciss: out of memory\n");
2062         goto freeret;
2063 }
2064
2065 /* This function will find the first index of the controllers drive array
2066  * that has a null drv pointer and allocate the drive info struct and
2067  * will return that index   This is where new drives will be added.
2068  * If the index to be returned is greater than the highest_lun index for
2069  * the controller then highest_lun is set * to this new index.
2070  * If there are no available indexes or if tha allocation fails, then -1
2071  * is returned.  * "controller_node" is used to know if this is a real
2072  * logical drive, or just the controller node, which determines if this
2073  * counts towards highest_lun.
2074  */
2075 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2076 {
2077         int i;
2078         drive_info_struct *drv;
2079
2080         /* Search for an empty slot for our drive info */
2081         for (i = 0; i < CISS_MAX_LUN; i++) {
2082
2083                 /* if not cxd0 case, and it's occupied, skip it. */
2084                 if (h->drv[i] && i != 0)
2085                         continue;
2086                 /*
2087                  * If it's cxd0 case, and drv is alloc'ed already, and a
2088                  * disk is configured there, skip it.
2089                  */
2090                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2091                         continue;
2092
2093                 /*
2094                  * We've found an empty slot.  Update highest_lun
2095                  * provided this isn't just the fake cxd0 controller node.
2096                  */
2097                 if (i > h->highest_lun && !controller_node)
2098                         h->highest_lun = i;
2099
2100                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2101                 if (i == 0 && h->drv[i] != NULL)
2102                         return i;
2103
2104                 /*
2105                  * Found an empty slot, not already alloc'ed.  Allocate it.
2106                  * Mark it with raid_level == -1, so we know it's new later on.
2107                  */
2108                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2109                 if (!drv)
2110                         return -1;
2111                 drv->raid_level = -1; /* so we know it's new */
2112                 h->drv[i] = drv;
2113                 return i;
2114         }
2115         return -1;
2116 }
2117
2118 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2119 {
2120         kfree(h->drv[drv_index]);
2121         h->drv[drv_index] = NULL;
2122 }
2123
2124 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2125 {
2126         put_disk(h->gendisk[drv_index]);
2127         h->gendisk[drv_index] = NULL;
2128 }
2129
2130 /* cciss_add_gendisk finds a free hba[]->drv structure
2131  * and allocates a gendisk if needed, and sets the lunid
2132  * in the drvinfo structure.   It returns the index into
2133  * the ->drv[] array, or -1 if none are free.
2134  * is_controller_node indicates whether highest_lun should
2135  * count this disk, or if it's only being added to provide
2136  * a means to talk to the controller in case no logical
2137  * drives have yet been configured.
2138  */
2139 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2140         int controller_node)
2141 {
2142         int drv_index;
2143
2144         drv_index = cciss_alloc_drive_info(h, controller_node);
2145         if (drv_index == -1)
2146                 return -1;
2147
2148         /*Check if the gendisk needs to be allocated */
2149         if (!h->gendisk[drv_index]) {
2150                 h->gendisk[drv_index] =
2151                         alloc_disk(1 << NWD_SHIFT);
2152                 if (!h->gendisk[drv_index]) {
2153                         printk(KERN_ERR "cciss%d: could not "
2154                                 "allocate a new disk %d\n",
2155                                 h->ctlr, drv_index);
2156                         goto err_free_drive_info;
2157                 }
2158         }
2159         memcpy(h->drv[drv_index]->LunID, lunid,
2160                 sizeof(h->drv[drv_index]->LunID));
2161         if (cciss_create_ld_sysfs_entry(h, drv_index))
2162                 goto err_free_disk;
2163         /* Don't need to mark this busy because nobody */
2164         /* else knows about this disk yet to contend */
2165         /* for access to it. */
2166         h->drv[drv_index]->busy_configuring = 0;
2167         wmb();
2168         return drv_index;
2169
2170 err_free_disk:
2171         cciss_free_gendisk(h, drv_index);
2172 err_free_drive_info:
2173         cciss_free_drive_info(h, drv_index);
2174         return -1;
2175 }
2176
2177 /* This is for the special case of a controller which
2178  * has no logical drives.  In this case, we still need
2179  * to register a disk so the controller can be accessed
2180  * by the Array Config Utility.
2181  */
2182 static void cciss_add_controller_node(ctlr_info_t *h)
2183 {
2184         struct gendisk *disk;
2185         int drv_index;
2186
2187         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2188                 return;
2189
2190         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2191         if (drv_index == -1)
2192                 goto error;
2193         h->drv[drv_index]->block_size = 512;
2194         h->drv[drv_index]->nr_blocks = 0;
2195         h->drv[drv_index]->heads = 0;
2196         h->drv[drv_index]->sectors = 0;
2197         h->drv[drv_index]->cylinders = 0;
2198         h->drv[drv_index]->raid_level = -1;
2199         memset(h->drv[drv_index]->serial_no, 0, 16);
2200         disk = h->gendisk[drv_index];
2201         if (cciss_add_disk(h, disk, drv_index) == 0)
2202                 return;
2203         cciss_free_gendisk(h, drv_index);
2204         cciss_free_drive_info(h, drv_index);
2205 error:
2206         printk(KERN_WARNING "cciss%d: could not "
2207                 "add disk 0.\n", h->ctlr);
2208         return;
2209 }
2210
2211 /* This function will add and remove logical drives from the Logical
2212  * drive array of the controller and maintain persistency of ordering
2213  * so that mount points are preserved until the next reboot.  This allows
2214  * for the removal of logical drives in the middle of the drive array
2215  * without a re-ordering of those drives.
2216  * INPUT
2217  * h            = The controller to perform the operations on
2218  */
2219 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2220         int via_ioctl)
2221 {
2222         int ctlr = h->ctlr;
2223         int num_luns;
2224         ReportLunData_struct *ld_buff = NULL;
2225         int return_code;
2226         int listlength = 0;
2227         int i;
2228         int drv_found;
2229         int drv_index = 0;
2230         unsigned char lunid[8] = CTLR_LUNID;
2231         unsigned long flags;
2232
2233         if (!capable(CAP_SYS_RAWIO))
2234                 return -EPERM;
2235
2236         /* Set busy_configuring flag for this operation */
2237         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2238         if (h->busy_configuring) {
2239                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2240                 return -EBUSY;
2241         }
2242         h->busy_configuring = 1;
2243         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2244
2245         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2246         if (ld_buff == NULL)
2247                 goto mem_msg;
2248
2249         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2250                                       sizeof(ReportLunData_struct),
2251                                       0, CTLR_LUNID, TYPE_CMD);
2252
2253         if (return_code == IO_OK)
2254                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2255         else {  /* reading number of logical volumes failed */
2256                 printk(KERN_WARNING "cciss: report logical volume"
2257                        " command failed\n");
2258                 listlength = 0;
2259                 goto freeret;
2260         }
2261
2262         num_luns = listlength / 8;      /* 8 bytes per entry */
2263         if (num_luns > CISS_MAX_LUN) {
2264                 num_luns = CISS_MAX_LUN;
2265                 printk(KERN_WARNING "cciss: more luns configured"
2266                        " on controller than can be handled by"
2267                        " this driver.\n");
2268         }
2269
2270         if (num_luns == 0)
2271                 cciss_add_controller_node(h);
2272
2273         /* Compare controller drive array to driver's drive array
2274          * to see if any drives are missing on the controller due
2275          * to action of Array Config Utility (user deletes drive)
2276          * and deregister logical drives which have disappeared.
2277          */
2278         for (i = 0; i <= h->highest_lun; i++) {
2279                 int j;
2280                 drv_found = 0;
2281
2282                 /* skip holes in the array from already deleted drives */
2283                 if (h->drv[i] == NULL)
2284                         continue;
2285
2286                 for (j = 0; j < num_luns; j++) {
2287                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2288                         if (memcmp(h->drv[i]->LunID, lunid,
2289                                 sizeof(lunid)) == 0) {
2290                                 drv_found = 1;
2291                                 break;
2292                         }
2293                 }
2294                 if (!drv_found) {
2295                         /* Deregister it from the OS, it's gone. */
2296                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2297                         h->drv[i]->busy_configuring = 1;
2298                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2299                         return_code = deregister_disk(h, i, 1, via_ioctl);
2300                         if (h->drv[i] != NULL)
2301                                 h->drv[i]->busy_configuring = 0;
2302                 }
2303         }
2304
2305         /* Compare controller drive array to driver's drive array.
2306          * Check for updates in the drive information and any new drives
2307          * on the controller due to ACU adding logical drives, or changing
2308          * a logical drive's size, etc.  Reregister any new/changed drives
2309          */
2310         for (i = 0; i < num_luns; i++) {
2311                 int j;
2312
2313                 drv_found = 0;
2314
2315                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2316                 /* Find if the LUN is already in the drive array
2317                  * of the driver.  If so then update its info
2318                  * if not in use.  If it does not exist then find
2319                  * the first free index and add it.
2320                  */
2321                 for (j = 0; j <= h->highest_lun; j++) {
2322                         if (h->drv[j] != NULL &&
2323                                 memcmp(h->drv[j]->LunID, lunid,
2324                                         sizeof(h->drv[j]->LunID)) == 0) {
2325                                 drv_index = j;
2326                                 drv_found = 1;
2327                                 break;
2328                         }
2329                 }
2330
2331                 /* check if the drive was found already in the array */
2332                 if (!drv_found) {
2333                         drv_index = cciss_add_gendisk(h, lunid, 0);
2334                         if (drv_index == -1)
2335                                 goto freeret;
2336                 }
2337                 cciss_update_drive_info(ctlr, drv_index, first_time,
2338                         via_ioctl);
2339         }               /* end for */
2340
2341 freeret:
2342         kfree(ld_buff);
2343         h->busy_configuring = 0;
2344         /* We return -1 here to tell the ACU that we have registered/updated
2345          * all of the drives that we can and to keep it from calling us
2346          * additional times.
2347          */
2348         return -1;
2349 mem_msg:
2350         printk(KERN_ERR "cciss: out of memory\n");
2351         h->busy_configuring = 0;
2352         goto freeret;
2353 }
2354
2355 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2356 {
2357         /* zero out the disk size info */
2358         drive_info->nr_blocks = 0;
2359         drive_info->block_size = 0;
2360         drive_info->heads = 0;
2361         drive_info->sectors = 0;
2362         drive_info->cylinders = 0;
2363         drive_info->raid_level = -1;
2364         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2365         memset(drive_info->model, 0, sizeof(drive_info->model));
2366         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2367         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2368         /*
2369          * don't clear the LUNID though, we need to remember which
2370          * one this one is.
2371          */
2372 }
2373
2374 /* This function will deregister the disk and it's queue from the
2375  * kernel.  It must be called with the controller lock held and the
2376  * drv structures busy_configuring flag set.  It's parameters are:
2377  *
2378  * disk = This is the disk to be deregistered
2379  * drv  = This is the drive_info_struct associated with the disk to be
2380  *        deregistered.  It contains information about the disk used
2381  *        by the driver.
2382  * clear_all = This flag determines whether or not the disk information
2383  *             is going to be completely cleared out and the highest_lun
2384  *             reset.  Sometimes we want to clear out information about
2385  *             the disk in preparation for re-adding it.  In this case
2386  *             the highest_lun should be left unchanged and the LunID
2387  *             should not be cleared.
2388  * via_ioctl
2389  *    This indicates whether we've reached this path via ioctl.
2390  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2391  *    If this path is reached via ioctl(), then the max_usage_count will
2392  *    be 1, as the process calling ioctl() has got to have the device open.
2393  *    If we get here via sysfs, then the max usage count will be zero.
2394 */
2395 static int deregister_disk(ctlr_info_t *h, int drv_index,
2396                            int clear_all, int via_ioctl)
2397 {
2398         int i;
2399         struct gendisk *disk;
2400         drive_info_struct *drv;
2401         int recalculate_highest_lun;
2402
2403         if (!capable(CAP_SYS_RAWIO))
2404                 return -EPERM;
2405
2406         drv = h->drv[drv_index];
2407         disk = h->gendisk[drv_index];
2408
2409         /* make sure logical volume is NOT is use */
2410         if (clear_all || (h->gendisk[0] == disk)) {
2411                 if (drv->usage_count > via_ioctl)
2412                         return -EBUSY;
2413         } else if (drv->usage_count > 0)
2414                 return -EBUSY;
2415
2416         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2417
2418         /* invalidate the devices and deregister the disk.  If it is disk
2419          * zero do not deregister it but just zero out it's values.  This
2420          * allows us to delete disk zero but keep the controller registered.
2421          */
2422         if (h->gendisk[0] != disk) {
2423                 struct request_queue *q = disk->queue;
2424                 if (disk->flags & GENHD_FL_UP) {
2425                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2426                         del_gendisk(disk);
2427                 }
2428                 if (q)
2429                         blk_cleanup_queue(q);
2430                 /* If clear_all is set then we are deleting the logical
2431                  * drive, not just refreshing its info.  For drives
2432                  * other than disk 0 we will call put_disk.  We do not
2433                  * do this for disk 0 as we need it to be able to
2434                  * configure the controller.
2435                  */
2436                 if (clear_all){
2437                         /* This isn't pretty, but we need to find the
2438                          * disk in our array and NULL our the pointer.
2439                          * This is so that we will call alloc_disk if
2440                          * this index is used again later.
2441                          */
2442                         for (i=0; i < CISS_MAX_LUN; i++){
2443                                 if (h->gendisk[i] == disk) {
2444                                         h->gendisk[i] = NULL;
2445                                         break;
2446                                 }
2447                         }
2448                         put_disk(disk);
2449                 }
2450         } else {
2451                 set_capacity(disk, 0);
2452                 cciss_clear_drive_info(drv);
2453         }
2454
2455         --h->num_luns;
2456
2457         /* if it was the last disk, find the new hightest lun */
2458         if (clear_all && recalculate_highest_lun) {
2459                 int newhighest = -1;
2460                 for (i = 0; i <= h->highest_lun; i++) {
2461                         /* if the disk has size > 0, it is available */
2462                         if (h->drv[i] && h->drv[i]->heads)
2463                                 newhighest = i;
2464                 }
2465                 h->highest_lun = newhighest;
2466         }
2467         return 0;
2468 }
2469
2470 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2471                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2472                 int cmd_type)
2473 {
2474         ctlr_info_t *h = hba[ctlr];
2475         u64bit buff_dma_handle;
2476         int status = IO_OK;
2477
2478         c->cmd_type = CMD_IOCTL_PEND;
2479         c->Header.ReplyQueue = 0;
2480         if (buff != NULL) {
2481                 c->Header.SGList = 1;
2482                 c->Header.SGTotal = 1;
2483         } else {
2484                 c->Header.SGList = 0;
2485                 c->Header.SGTotal = 0;
2486         }
2487         c->Header.Tag.lower = c->busaddr;
2488         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2489
2490         c->Request.Type.Type = cmd_type;
2491         if (cmd_type == TYPE_CMD) {
2492                 switch (cmd) {
2493                 case CISS_INQUIRY:
2494                         /* are we trying to read a vital product page */
2495                         if (page_code != 0) {
2496                                 c->Request.CDB[1] = 0x01;
2497                                 c->Request.CDB[2] = page_code;
2498                         }
2499                         c->Request.CDBLen = 6;
2500                         c->Request.Type.Attribute = ATTR_SIMPLE;
2501                         c->Request.Type.Direction = XFER_READ;
2502                         c->Request.Timeout = 0;
2503                         c->Request.CDB[0] = CISS_INQUIRY;
2504                         c->Request.CDB[4] = size & 0xFF;
2505                         break;
2506                 case CISS_REPORT_LOG:
2507                 case CISS_REPORT_PHYS:
2508                         /* Talking to controller so It's a physical command
2509                            mode = 00 target = 0.  Nothing to write.
2510                          */
2511                         c->Request.CDBLen = 12;
2512                         c->Request.Type.Attribute = ATTR_SIMPLE;
2513                         c->Request.Type.Direction = XFER_READ;
2514                         c->Request.Timeout = 0;
2515                         c->Request.CDB[0] = cmd;
2516                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2517                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2518                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2519                         c->Request.CDB[9] = size & 0xFF;
2520                         break;
2521
2522                 case CCISS_READ_CAPACITY:
2523                         c->Request.CDBLen = 10;
2524                         c->Request.Type.Attribute = ATTR_SIMPLE;
2525                         c->Request.Type.Direction = XFER_READ;
2526                         c->Request.Timeout = 0;
2527                         c->Request.CDB[0] = cmd;
2528                         break;
2529                 case CCISS_READ_CAPACITY_16:
2530                         c->Request.CDBLen = 16;
2531                         c->Request.Type.Attribute = ATTR_SIMPLE;
2532                         c->Request.Type.Direction = XFER_READ;
2533                         c->Request.Timeout = 0;
2534                         c->Request.CDB[0] = cmd;
2535                         c->Request.CDB[1] = 0x10;
2536                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2537                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2538                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2539                         c->Request.CDB[13] = size & 0xFF;
2540                         c->Request.Timeout = 0;
2541                         c->Request.CDB[0] = cmd;
2542                         break;
2543                 case CCISS_CACHE_FLUSH:
2544                         c->Request.CDBLen = 12;
2545                         c->Request.Type.Attribute = ATTR_SIMPLE;
2546                         c->Request.Type.Direction = XFER_WRITE;
2547                         c->Request.Timeout = 0;
2548                         c->Request.CDB[0] = BMIC_WRITE;
2549                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2550                         break;
2551                 case TEST_UNIT_READY:
2552                         c->Request.CDBLen = 6;
2553                         c->Request.Type.Attribute = ATTR_SIMPLE;
2554                         c->Request.Type.Direction = XFER_NONE;
2555                         c->Request.Timeout = 0;
2556                         break;
2557                 default:
2558                         printk(KERN_WARNING
2559                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2560                         return IO_ERROR;
2561                 }
2562         } else if (cmd_type == TYPE_MSG) {
2563                 switch (cmd) {
2564                 case 0: /* ABORT message */
2565                         c->Request.CDBLen = 12;
2566                         c->Request.Type.Attribute = ATTR_SIMPLE;
2567                         c->Request.Type.Direction = XFER_WRITE;
2568                         c->Request.Timeout = 0;
2569                         c->Request.CDB[0] = cmd;        /* abort */
2570                         c->Request.CDB[1] = 0;  /* abort a command */
2571                         /* buff contains the tag of the command to abort */
2572                         memcpy(&c->Request.CDB[4], buff, 8);
2573                         break;
2574                 case 1: /* RESET message */
2575                         c->Request.CDBLen = 16;
2576                         c->Request.Type.Attribute = ATTR_SIMPLE;
2577                         c->Request.Type.Direction = XFER_NONE;
2578                         c->Request.Timeout = 0;
2579                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2580                         c->Request.CDB[0] = cmd;        /* reset */
2581                         c->Request.CDB[1] = 0x03;       /* reset a target */
2582                         break;
2583                 case 3: /* No-Op message */
2584                         c->Request.CDBLen = 1;
2585                         c->Request.Type.Attribute = ATTR_SIMPLE;
2586                         c->Request.Type.Direction = XFER_WRITE;
2587                         c->Request.Timeout = 0;
2588                         c->Request.CDB[0] = cmd;
2589                         break;
2590                 default:
2591                         printk(KERN_WARNING
2592                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2593                         return IO_ERROR;
2594                 }
2595         } else {
2596                 printk(KERN_WARNING
2597                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2598                 return IO_ERROR;
2599         }
2600         /* Fill in the scatter gather information */
2601         if (size > 0) {
2602                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2603                                                              buff, size,
2604                                                              PCI_DMA_BIDIRECTIONAL);
2605                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2606                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2607                 c->SG[0].Len = size;
2608                 c->SG[0].Ext = 0;       /* we are not chaining */
2609         }
2610         return status;
2611 }
2612
2613 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2614 {
2615         switch (c->err_info->ScsiStatus) {
2616         case SAM_STAT_GOOD:
2617                 return IO_OK;
2618         case SAM_STAT_CHECK_CONDITION:
2619                 switch (0xf & c->err_info->SenseInfo[2]) {
2620                 case 0: return IO_OK; /* no sense */
2621                 case 1: return IO_OK; /* recovered error */
2622                 default:
2623                         if (check_for_unit_attention(h, c))
2624                                 return IO_NEEDS_RETRY;
2625                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2626                                 "check condition, sense key = 0x%02x\n",
2627                                 h->ctlr, c->Request.CDB[0],
2628                                 c->err_info->SenseInfo[2]);
2629                 }
2630                 break;
2631         default:
2632                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2633                         "scsi status = 0x%02x\n", h->ctlr,
2634                         c->Request.CDB[0], c->err_info->ScsiStatus);
2635                 break;
2636         }
2637         return IO_ERROR;
2638 }
2639
2640 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2641 {
2642         int return_status = IO_OK;
2643
2644         if (c->err_info->CommandStatus == CMD_SUCCESS)
2645                 return IO_OK;
2646
2647         switch (c->err_info->CommandStatus) {
2648         case CMD_TARGET_STATUS:
2649                 return_status = check_target_status(h, c);
2650                 break;
2651         case CMD_DATA_UNDERRUN:
2652         case CMD_DATA_OVERRUN:
2653                 /* expected for inquiry and report lun commands */
2654                 break;
2655         case CMD_INVALID:
2656                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2657                        "reported invalid\n", c->Request.CDB[0]);
2658                 return_status = IO_ERROR;
2659                 break;
2660         case CMD_PROTOCOL_ERR:
2661                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2662                        "protocol error \n", c->Request.CDB[0]);
2663                 return_status = IO_ERROR;
2664                 break;
2665         case CMD_HARDWARE_ERR:
2666                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2667                        " hardware error\n", c->Request.CDB[0]);
2668                 return_status = IO_ERROR;
2669                 break;
2670         case CMD_CONNECTION_LOST:
2671                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2672                        "connection lost\n", c->Request.CDB[0]);
2673                 return_status = IO_ERROR;
2674                 break;
2675         case CMD_ABORTED:
2676                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2677                        "aborted\n", c->Request.CDB[0]);
2678                 return_status = IO_ERROR;
2679                 break;
2680         case CMD_ABORT_FAILED:
2681                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2682                        "abort failed\n", c->Request.CDB[0]);
2683                 return_status = IO_ERROR;
2684                 break;
2685         case CMD_UNSOLICITED_ABORT:
2686                 printk(KERN_WARNING
2687                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2688                         c->Request.CDB[0]);
2689                 return_status = IO_NEEDS_RETRY;
2690                 break;
2691         default:
2692                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2693                        "unknown status %x\n", c->Request.CDB[0],
2694                        c->err_info->CommandStatus);
2695                 return_status = IO_ERROR;
2696         }
2697         return return_status;
2698 }
2699
2700 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2701         int attempt_retry)
2702 {
2703         DECLARE_COMPLETION_ONSTACK(wait);
2704         u64bit buff_dma_handle;
2705         int return_status = IO_OK;
2706
2707 resend_cmd2:
2708         c->waiting = &wait;
2709         enqueue_cmd_and_start_io(h, c);
2710
2711         wait_for_completion(&wait);
2712
2713         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2714                 goto command_done;
2715
2716         return_status = process_sendcmd_error(h, c);
2717
2718         if (return_status == IO_NEEDS_RETRY &&
2719                 c->retry_count < MAX_CMD_RETRIES) {
2720                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2721                         c->Request.CDB[0]);
2722                 c->retry_count++;
2723                 /* erase the old error information */
2724                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2725                 return_status = IO_OK;
2726                 INIT_COMPLETION(wait);
2727                 goto resend_cmd2;
2728         }
2729
2730 command_done:
2731         /* unlock the buffers from DMA */
2732         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2733         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2734         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2735                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2736         return return_status;
2737 }
2738
2739 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2740                            __u8 page_code, unsigned char scsi3addr[],
2741                         int cmd_type)
2742 {
2743         ctlr_info_t *h = hba[ctlr];
2744         CommandList_struct *c;
2745         int return_status;
2746
2747         c = cmd_alloc(h, 0);
2748         if (!c)
2749                 return -ENOMEM;
2750         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2751                 scsi3addr, cmd_type);
2752         if (return_status == IO_OK)
2753                 return_status = sendcmd_withirq_core(h, c, 1);
2754
2755         cmd_free(h, c, 0);
2756         return return_status;
2757 }
2758
2759 static void cciss_geometry_inquiry(int ctlr, int logvol,
2760                                    sector_t total_size,
2761                                    unsigned int block_size,
2762                                    InquiryData_struct *inq_buff,
2763                                    drive_info_struct *drv)
2764 {
2765         int return_code;
2766         unsigned long t;
2767         unsigned char scsi3addr[8];
2768
2769         memset(inq_buff, 0, sizeof(InquiryData_struct));
2770         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2771         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2772                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2773         if (return_code == IO_OK) {
2774                 if (inq_buff->data_byte[8] == 0xFF) {
2775                         printk(KERN_WARNING
2776                                "cciss: reading geometry failed, volume "
2777                                "does not support reading geometry\n");
2778                         drv->heads = 255;
2779                         drv->sectors = 32;      /* Sectors per track */
2780                         drv->cylinders = total_size + 1;
2781                         drv->raid_level = RAID_UNKNOWN;
2782                 } else {
2783                         drv->heads = inq_buff->data_byte[6];
2784                         drv->sectors = inq_buff->data_byte[7];
2785                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2786                         drv->cylinders += inq_buff->data_byte[5];
2787                         drv->raid_level = inq_buff->data_byte[8];
2788                 }
2789                 drv->block_size = block_size;
2790                 drv->nr_blocks = total_size + 1;
2791                 t = drv->heads * drv->sectors;
2792                 if (t > 1) {
2793                         sector_t real_size = total_size + 1;
2794                         unsigned long rem = sector_div(real_size, t);
2795                         if (rem)
2796                                 real_size++;
2797                         drv->cylinders = real_size;
2798                 }
2799         } else {                /* Get geometry failed */
2800                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2801         }
2802 }
2803
2804 static void
2805 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2806                     unsigned int *block_size)
2807 {
2808         ReadCapdata_struct *buf;
2809         int return_code;
2810         unsigned char scsi3addr[8];
2811
2812         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2813         if (!buf) {
2814                 printk(KERN_WARNING "cciss: out of memory\n");
2815                 return;
2816         }
2817
2818         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2819         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2820                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2821         if (return_code == IO_OK) {
2822                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2823                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2824         } else {                /* read capacity command failed */
2825                 printk(KERN_WARNING "cciss: read capacity failed\n");
2826                 *total_size = 0;
2827                 *block_size = BLOCK_SIZE;
2828         }
2829         kfree(buf);
2830 }
2831
2832 static void cciss_read_capacity_16(int ctlr, int logvol,
2833         sector_t *total_size, unsigned int *block_size)
2834 {
2835         ReadCapdata_struct_16 *buf;
2836         int return_code;
2837         unsigned char scsi3addr[8];
2838
2839         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2840         if (!buf) {
2841                 printk(KERN_WARNING "cciss: out of memory\n");
2842                 return;
2843         }
2844
2845         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2846         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2847                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2848                         0, scsi3addr, TYPE_CMD);
2849         if (return_code == IO_OK) {
2850                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2851                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2852         } else {                /* read capacity command failed */
2853                 printk(KERN_WARNING "cciss: read capacity failed\n");
2854                 *total_size = 0;
2855                 *block_size = BLOCK_SIZE;
2856         }
2857         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2858                (unsigned long long)*total_size+1, *block_size);
2859         kfree(buf);
2860 }
2861
2862 static int cciss_revalidate(struct gendisk *disk)
2863 {
2864         ctlr_info_t *h = get_host(disk);
2865         drive_info_struct *drv = get_drv(disk);
2866         int logvol;
2867         int FOUND = 0;
2868         unsigned int block_size;
2869         sector_t total_size;
2870         InquiryData_struct *inq_buff = NULL;
2871
2872         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2873                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2874                         sizeof(drv->LunID)) == 0) {
2875                         FOUND = 1;
2876                         break;
2877                 }
2878         }
2879
2880         if (!FOUND)
2881                 return 1;
2882
2883         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2884         if (inq_buff == NULL) {
2885                 printk(KERN_WARNING "cciss: out of memory\n");
2886                 return 1;
2887         }
2888         if (h->cciss_read == CCISS_READ_10) {
2889                 cciss_read_capacity(h->ctlr, logvol,
2890                                         &total_size, &block_size);
2891         } else {
2892                 cciss_read_capacity_16(h->ctlr, logvol,
2893                                         &total_size, &block_size);
2894         }
2895         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2896                                inq_buff, drv);
2897
2898         blk_queue_logical_block_size(drv->queue, drv->block_size);
2899         set_capacity(disk, drv->nr_blocks);
2900
2901         kfree(inq_buff);
2902         return 0;
2903 }
2904
2905 /*
2906  * Map (physical) PCI mem into (virtual) kernel space
2907  */
2908 static void __iomem *remap_pci_mem(ulong base, ulong size)
2909 {
2910         ulong page_base = ((ulong) base) & PAGE_MASK;
2911         ulong page_offs = ((ulong) base) - page_base;
2912         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2913
2914         return page_remapped ? (page_remapped + page_offs) : NULL;
2915 }
2916
2917 /*
2918  * Takes jobs of the Q and sends them to the hardware, then puts it on
2919  * the Q to wait for completion.
2920  */
2921 static void start_io(ctlr_info_t *h)
2922 {
2923         CommandList_struct *c;
2924
2925         while (!hlist_empty(&h->reqQ)) {
2926                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2927                 /* can't do anything if fifo is full */
2928                 if ((h->access.fifo_full(h))) {
2929                         printk(KERN_WARNING "cciss: fifo full\n");
2930                         break;
2931                 }
2932
2933                 /* Get the first entry from the Request Q */
2934                 removeQ(c);
2935                 h->Qdepth--;
2936
2937                 /* Tell the controller execute command */
2938                 h->access.submit_command(h, c);
2939
2940                 /* Put job onto the completed Q */
2941                 addQ(&h->cmpQ, c);
2942         }
2943 }
2944
2945 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2946 /* Zeros out the error record and then resends the command back */
2947 /* to the controller */
2948 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2949 {
2950         /* erase the old error information */
2951         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2952
2953         /* add it to software queue and then send it to the controller */
2954         addQ(&h->reqQ, c);
2955         h->Qdepth++;
2956         if (h->Qdepth > h->maxQsinceinit)
2957                 h->maxQsinceinit = h->Qdepth;
2958
2959         start_io(h);
2960 }
2961
2962 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2963         unsigned int msg_byte, unsigned int host_byte,
2964         unsigned int driver_byte)
2965 {
2966         /* inverse of macros in scsi.h */
2967         return (scsi_status_byte & 0xff) |
2968                 ((msg_byte & 0xff) << 8) |
2969                 ((host_byte & 0xff) << 16) |
2970                 ((driver_byte & 0xff) << 24);
2971 }
2972
2973 static inline int evaluate_target_status(ctlr_info_t *h,
2974                         CommandList_struct *cmd, int *retry_cmd)
2975 {
2976         unsigned char sense_key;
2977         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2978         int error_value;
2979
2980         *retry_cmd = 0;
2981         /* If we get in here, it means we got "target status", that is, scsi status */
2982         status_byte = cmd->err_info->ScsiStatus;
2983         driver_byte = DRIVER_OK;
2984         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2985
2986         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2987                 host_byte = DID_PASSTHROUGH;
2988         else
2989                 host_byte = DID_OK;
2990
2991         error_value = make_status_bytes(status_byte, msg_byte,
2992                 host_byte, driver_byte);
2993
2994         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2995                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
2996                         printk(KERN_WARNING "cciss: cmd %p "
2997                                "has SCSI Status 0x%x\n",
2998                                cmd, cmd->err_info->ScsiStatus);
2999                 return error_value;
3000         }
3001
3002         /* check the sense key */
3003         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3004         /* no status or recovered error */
3005         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3006             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3007                 error_value = 0;
3008
3009         if (check_for_unit_attention(h, cmd)) {
3010                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3011                 return 0;
3012         }
3013
3014         /* Not SG_IO or similar? */
3015         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3016                 if (error_value != 0)
3017                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3018                                " sense key = 0x%x\n", cmd, sense_key);
3019                 return error_value;
3020         }
3021
3022         /* SG_IO or similar, copy sense data back */
3023         if (cmd->rq->sense) {
3024                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3025                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3026                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3027                         cmd->rq->sense_len);
3028         } else
3029                 cmd->rq->sense_len = 0;
3030
3031         return error_value;
3032 }
3033
3034 /* checks the status of the job and calls complete buffers to mark all
3035  * buffers for the completed job. Note that this function does not need
3036  * to hold the hba/queue lock.
3037  */
3038 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3039                                     int timeout)
3040 {
3041         int retry_cmd = 0;
3042         struct request *rq = cmd->rq;
3043
3044         rq->errors = 0;
3045
3046         if (timeout)
3047                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3048
3049         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3050                 goto after_error_processing;
3051
3052         switch (cmd->err_info->CommandStatus) {
3053         case CMD_TARGET_STATUS:
3054                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3055                 break;
3056         case CMD_DATA_UNDERRUN:
3057                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3058                         printk(KERN_WARNING "cciss: cmd %p has"
3059                                " completed with data underrun "
3060                                "reported\n", cmd);
3061                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3062                 }
3063                 break;
3064         case CMD_DATA_OVERRUN:
3065                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3066                         printk(KERN_WARNING "cciss: cmd %p has"
3067                                " completed with data overrun "
3068                                "reported\n", cmd);
3069                 break;
3070         case CMD_INVALID:
3071                 printk(KERN_WARNING "cciss: cmd %p is "
3072                        "reported invalid\n", cmd);
3073                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3074                         cmd->err_info->CommandStatus, DRIVER_OK,
3075                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3076                                 DID_PASSTHROUGH : DID_ERROR);
3077                 break;
3078         case CMD_PROTOCOL_ERR:
3079                 printk(KERN_WARNING "cciss: cmd %p has "
3080                        "protocol error \n", cmd);
3081                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3082                         cmd->err_info->CommandStatus, DRIVER_OK,
3083                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3084                                 DID_PASSTHROUGH : DID_ERROR);
3085                 break;
3086         case CMD_HARDWARE_ERR:
3087                 printk(KERN_WARNING "cciss: cmd %p had "
3088                        " hardware error\n", cmd);
3089                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3090                         cmd->err_info->CommandStatus, DRIVER_OK,
3091                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3092                                 DID_PASSTHROUGH : DID_ERROR);
3093                 break;
3094         case CMD_CONNECTION_LOST:
3095                 printk(KERN_WARNING "cciss: cmd %p had "
3096                        "connection lost\n", cmd);
3097                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3098                         cmd->err_info->CommandStatus, DRIVER_OK,
3099                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3100                                 DID_PASSTHROUGH : DID_ERROR);
3101                 break;
3102         case CMD_ABORTED:
3103                 printk(KERN_WARNING "cciss: cmd %p was "
3104                        "aborted\n", cmd);
3105                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3106                         cmd->err_info->CommandStatus, DRIVER_OK,
3107                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3108                                 DID_PASSTHROUGH : DID_ABORT);
3109                 break;
3110         case CMD_ABORT_FAILED:
3111                 printk(KERN_WARNING "cciss: cmd %p reports "
3112                        "abort failed\n", cmd);
3113                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3114                         cmd->err_info->CommandStatus, DRIVER_OK,
3115                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3116                                 DID_PASSTHROUGH : DID_ERROR);
3117                 break;
3118         case CMD_UNSOLICITED_ABORT:
3119                 printk(KERN_WARNING "cciss%d: unsolicited "
3120                        "abort %p\n", h->ctlr, cmd);
3121                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3122                         retry_cmd = 1;
3123                         printk(KERN_WARNING
3124                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3125                         cmd->retry_count++;
3126                 } else
3127                         printk(KERN_WARNING
3128                                "cciss%d: %p retried too "
3129                                "many times\n", h->ctlr, cmd);
3130                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3131                         cmd->err_info->CommandStatus, DRIVER_OK,
3132                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3133                                 DID_PASSTHROUGH : DID_ABORT);
3134                 break;
3135         case CMD_TIMEOUT:
3136                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3137                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3138                         cmd->err_info->CommandStatus, DRIVER_OK,
3139                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3140                                 DID_PASSTHROUGH : DID_ERROR);
3141                 break;
3142         default:
3143                 printk(KERN_WARNING "cciss: cmd %p returned "
3144                        "unknown status %x\n", cmd,
3145                        cmd->err_info->CommandStatus);
3146                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3147                         cmd->err_info->CommandStatus, DRIVER_OK,
3148                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3149                                 DID_PASSTHROUGH : DID_ERROR);
3150         }
3151
3152 after_error_processing:
3153
3154         /* We need to return this command */
3155         if (retry_cmd) {
3156                 resend_cciss_cmd(h, cmd);
3157                 return;
3158         }
3159         cmd->rq->completion_data = cmd;
3160         blk_complete_request(cmd->rq);
3161 }
3162
3163 static inline u32 cciss_tag_contains_index(u32 tag)
3164 {
3165 #define DIRECT_LOOKUP_BIT 0x10
3166         return tag & DIRECT_LOOKUP_BIT;
3167 }
3168
3169 static inline u32 cciss_tag_to_index(u32 tag)
3170 {
3171 #define DIRECT_LOOKUP_SHIFT 5
3172         return tag >> DIRECT_LOOKUP_SHIFT;
3173 }
3174
3175 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3176 {
3177 #define CCISS_ERROR_BITS 0x03
3178         return tag & ~CCISS_ERROR_BITS;
3179 }
3180
3181 static inline void cciss_mark_tag_indexed(u32 *tag)
3182 {
3183         *tag |= DIRECT_LOOKUP_BIT;
3184 }
3185
3186 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3187 {
3188         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3189 }
3190
3191 /*
3192  * Get a request and submit it to the controller.
3193  */
3194 static void do_cciss_request(struct request_queue *q)
3195 {
3196         ctlr_info_t *h = q->queuedata;
3197         CommandList_struct *c;
3198         sector_t start_blk;
3199         int seg;
3200         struct request *creq;
3201         u64bit temp64;
3202         struct scatterlist *tmp_sg;
3203         SGDescriptor_struct *curr_sg;
3204         drive_info_struct *drv;
3205         int i, dir;
3206         int sg_index = 0;
3207         int chained = 0;
3208
3209         /* We call start_io here in case there is a command waiting on the
3210          * queue that has not been sent.
3211          */
3212         if (blk_queue_plugged(q))
3213                 goto startio;
3214
3215       queue:
3216         creq = blk_peek_request(q);
3217         if (!creq)
3218                 goto startio;
3219
3220         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3221
3222         if ((c = cmd_alloc(h, 1)) == NULL)
3223                 goto full;
3224
3225         blk_start_request(creq);
3226
3227         tmp_sg = h->scatter_list[c->cmdindex];
3228         spin_unlock_irq(q->queue_lock);
3229
3230         c->cmd_type = CMD_RWREQ;
3231         c->rq = creq;
3232
3233         /* fill in the request */
3234         drv = creq->rq_disk->private_data;
3235         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3236         /* got command from pool, so use the command block index instead */
3237         /* for direct lookups. */
3238         /* The first 2 bits are reserved for controller error reporting. */
3239         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3240         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3241         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3242         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3243         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3244         c->Request.Type.Attribute = ATTR_SIMPLE;
3245         c->Request.Type.Direction =
3246             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3247         c->Request.Timeout = 0; /* Don't time out */
3248         c->Request.CDB[0] =
3249             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3250         start_blk = blk_rq_pos(creq);
3251 #ifdef CCISS_DEBUG
3252         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3253                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3254 #endif                          /* CCISS_DEBUG */
3255
3256         sg_init_table(tmp_sg, h->maxsgentries);
3257         seg = blk_rq_map_sg(q, creq, tmp_sg);
3258
3259         /* get the DMA records for the setup */
3260         if (c->Request.Type.Direction == XFER_READ)
3261                 dir = PCI_DMA_FROMDEVICE;
3262         else
3263                 dir = PCI_DMA_TODEVICE;
3264
3265         curr_sg = c->SG;
3266         sg_index = 0;
3267         chained = 0;
3268
3269         for (i = 0; i < seg; i++) {
3270                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3271                         !chained && ((seg - i) > 1)) {
3272                         /* Point to next chain block. */
3273                         curr_sg = h->cmd_sg_list[c->cmdindex];
3274                         sg_index = 0;
3275                         chained = 1;
3276                 }
3277                 curr_sg[sg_index].Len = tmp_sg[i].length;
3278                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3279                                                 tmp_sg[i].offset,
3280                                                 tmp_sg[i].length, dir);
3281                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3282                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3283                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3284                 ++sg_index;
3285         }
3286         if (chained)
3287                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3288                         (seg - (h->max_cmd_sgentries - 1)) *
3289                                 sizeof(SGDescriptor_struct));
3290
3291         /* track how many SG entries we are using */
3292         if (seg > h->maxSG)
3293                 h->maxSG = seg;
3294
3295 #ifdef CCISS_DEBUG
3296         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3297                         "chained[%d]\n",
3298                         blk_rq_sectors(creq), seg, chained);
3299 #endif                          /* CCISS_DEBUG */
3300
3301         c->Header.SGTotal = seg + chained;
3302         if (seg <= h->max_cmd_sgentries)
3303                 c->Header.SGList = c->Header.SGTotal;
3304         else
3305                 c->Header.SGList = h->max_cmd_sgentries;
3306         set_performant_mode(h, c);
3307
3308         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3309                 if(h->cciss_read == CCISS_READ_10) {
3310                         c->Request.CDB[1] = 0;
3311                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3312                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3313                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3314                         c->Request.CDB[5] = start_blk & 0xff;
3315                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3316                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3317                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3318                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3319                 } else {
3320                         u32 upper32 = upper_32_bits(start_blk);
3321
3322                         c->Request.CDBLen = 16;
3323                         c->Request.CDB[1]= 0;
3324                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3325                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3326                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3327                         c->Request.CDB[5]= upper32 & 0xff;
3328                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3329                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3330                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3331                         c->Request.CDB[9]= start_blk & 0xff;
3332                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3333                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3334                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3335                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3336                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3337                 }
3338         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3339                 c->Request.CDBLen = creq->cmd_len;
3340                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3341         } else {
3342                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3343                 BUG();
3344         }
3345
3346         spin_lock_irq(q->queue_lock);
3347
3348         addQ(&h->reqQ, c);
3349         h->Qdepth++;
3350         if (h->Qdepth > h->maxQsinceinit)
3351                 h->maxQsinceinit = h->Qdepth;
3352
3353         goto queue;
3354 full:
3355         blk_stop_queue(q);
3356 startio:
3357         /* We will already have the driver lock here so not need
3358          * to lock it.
3359          */
3360         start_io(h);
3361 }
3362
3363 static inline unsigned long get_next_completion(ctlr_info_t *h)
3364 {
3365         return h->access.command_completed(h);
3366 }
3367
3368 static inline int interrupt_pending(ctlr_info_t *h)
3369 {
3370         return h->access.intr_pending(h);
3371 }
3372
3373 static inline long interrupt_not_for_us(ctlr_info_t *h)
3374 {
3375         return !(h->msi_vector || h->msix_vector) &&
3376                 ((h->access.intr_pending(h) == 0) ||
3377                 (h->interrupts_enabled == 0));
3378 }
3379
3380 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3381                         u32 raw_tag)
3382 {
3383         if (unlikely(tag_index >= h->nr_cmds)) {
3384                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3385                 return 1;
3386         }
3387         return 0;
3388 }
3389
3390 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3391                                 u32 raw_tag)
3392 {
3393         removeQ(c);
3394         if (likely(c->cmd_type == CMD_RWREQ))
3395                 complete_command(h, c, 0);
3396         else if (c->cmd_type == CMD_IOCTL_PEND)
3397                 complete(c->waiting);
3398 #ifdef CONFIG_CISS_SCSI_TAPE
3399         else if (c->cmd_type == CMD_SCSI)
3400                 complete_scsi_command(c, 0, raw_tag);
3401 #endif
3402 }
3403
3404 static inline u32 next_command(ctlr_info_t *h)
3405 {
3406         u32 a;
3407
3408         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3409                 return h->access.command_completed(h);
3410
3411         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3412                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3413                 (h->reply_pool_head)++;
3414                 h->commands_outstanding--;
3415         } else {
3416                 a = FIFO_EMPTY;
3417         }
3418         /* Check for wraparound */
3419         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3420                 h->reply_pool_head = h->reply_pool;
3421                 h->reply_pool_wraparound ^= 1;
3422         }
3423         return a;
3424 }
3425
3426 /* process completion of an indexed ("direct lookup") command */
3427 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3428 {
3429         u32 tag_index;
3430         CommandList_struct *c;
3431
3432         tag_index = cciss_tag_to_index(raw_tag);
3433         if (bad_tag(h, tag_index, raw_tag))
3434                 return next_command(h);
3435         c = h->cmd_pool + tag_index;
3436         finish_cmd(h, c, raw_tag);
3437         return next_command(h);
3438 }
3439
3440 /* process completion of a non-indexed command */
3441 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3442 {
3443         u32 tag;
3444         CommandList_struct *c = NULL;
3445         struct hlist_node *tmp;
3446         __u32 busaddr_masked, tag_masked;
3447
3448         tag = cciss_tag_discard_error_bits(raw_tag);
3449         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3450                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3451                 tag_masked = cciss_tag_discard_error_bits(tag);
3452                 if (busaddr_masked == tag_masked) {
3453                         finish_cmd(h, c, raw_tag);
3454                         return next_command(h);
3455                 }
3456         }
3457         bad_tag(h, h->nr_cmds + 1, raw_tag);
3458         return next_command(h);
3459 }
3460
3461 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3462 {
3463         ctlr_info_t *h = dev_id;
3464         unsigned long flags;
3465         u32 raw_tag;
3466
3467         if (interrupt_not_for_us(h))
3468                 return IRQ_NONE;
3469         /*
3470          * If there are completed commands in the completion queue,
3471          * we had better do something about it.
3472          */
3473         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3474         while (interrupt_pending(h)) {
3475                 raw_tag = get_next_completion(h);
3476                 while (raw_tag != FIFO_EMPTY) {
3477                         if (cciss_tag_contains_index(raw_tag))
3478                                 raw_tag = process_indexed_cmd(h, raw_tag);
3479                         else
3480                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3481                 }
3482         }
3483
3484         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3485         return IRQ_HANDLED;
3486 }
3487
3488 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3489  * check the interrupt pending register because it is not set.
3490  */
3491 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3492 {
3493         ctlr_info_t *h = dev_id;
3494         unsigned long flags;
3495         u32 raw_tag;
3496
3497         if (interrupt_not_for_us(h))
3498                 return IRQ_NONE;
3499         /*
3500          * If there are completed commands in the completion queue,
3501          * we had better do something about it.
3502          */
3503         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3504         raw_tag = get_next_completion(h);
3505         while (raw_tag != FIFO_EMPTY) {
3506                 if (cciss_tag_contains_index(raw_tag))
3507                         raw_tag = process_indexed_cmd(h, raw_tag);
3508                 else
3509                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3510         }
3511
3512         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3513         return IRQ_HANDLED;
3514 }
3515
3516 /**
3517  * add_to_scan_list() - add controller to rescan queue
3518  * @h:                Pointer to the controller.
3519  *
3520  * Adds the controller to the rescan queue if not already on the queue.
3521  *
3522  * returns 1 if added to the queue, 0 if skipped (could be on the
3523  * queue already, or the controller could be initializing or shutting
3524  * down).
3525  **/
3526 static int add_to_scan_list(struct ctlr_info *h)
3527 {
3528         struct ctlr_info *test_h;
3529         int found = 0;
3530         int ret = 0;
3531
3532         if (h->busy_initializing)
3533                 return 0;
3534
3535         if (!mutex_trylock(&h->busy_shutting_down))
3536                 return 0;
3537
3538         mutex_lock(&scan_mutex);
3539         list_for_each_entry(test_h, &scan_q, scan_list) {
3540                 if (test_h == h) {
3541                         found = 1;
3542                         break;
3543                 }
3544         }
3545         if (!found && !h->busy_scanning) {
3546                 INIT_COMPLETION(h->scan_wait);
3547                 list_add_tail(&h->scan_list, &scan_q);
3548                 ret = 1;
3549         }
3550         mutex_unlock(&scan_mutex);
3551         mutex_unlock(&h->busy_shutting_down);
3552
3553         return ret;
3554 }
3555
3556 /**
3557  * remove_from_scan_list() - remove controller from rescan queue
3558  * @h:                     Pointer to the controller.
3559  *
3560  * Removes the controller from the rescan queue if present. Blocks if
3561  * the controller is currently conducting a rescan.  The controller
3562  * can be in one of three states:
3563  * 1. Doesn't need a scan
3564  * 2. On the scan list, but not scanning yet (we remove it)
3565  * 3. Busy scanning (and not on the list). In this case we want to wait for
3566  *    the scan to complete to make sure the scanning thread for this
3567  *    controller is completely idle.
3568  **/
3569 static void remove_from_scan_list(struct ctlr_info *h)
3570 {
3571         struct ctlr_info *test_h, *tmp_h;
3572
3573         mutex_lock(&scan_mutex);
3574         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3575                 if (test_h == h) { /* state 2. */
3576                         list_del(&h->scan_list);
3577                         complete_all(&h->scan_wait);
3578                         mutex_unlock(&scan_mutex);
3579                         return;
3580                 }
3581         }
3582         if (h->busy_scanning) { /* state 3. */
3583                 mutex_unlock(&scan_mutex);
3584                 wait_for_completion(&h->scan_wait);
3585         } else { /* state 1, nothing to do. */
3586                 mutex_unlock(&scan_mutex);
3587         }
3588 }
3589
3590 /**
3591  * scan_thread() - kernel thread used to rescan controllers
3592  * @data:        Ignored.
3593  *
3594  * A kernel thread used scan for drive topology changes on
3595  * controllers. The thread processes only one controller at a time
3596  * using a queue.  Controllers are added to the queue using
3597  * add_to_scan_list() and removed from the queue either after done
3598  * processing or using remove_from_scan_list().
3599  *
3600  * returns 0.
3601  **/
3602 static int scan_thread(void *data)
3603 {
3604         struct ctlr_info *h;
3605
3606         while (1) {
3607                 set_current_state(TASK_INTERRUPTIBLE);
3608                 schedule();
3609                 if (kthread_should_stop())
3610                         break;
3611
3612                 while (1) {
3613                         mutex_lock(&scan_mutex);
3614                         if (list_empty(&scan_q)) {
3615                                 mutex_unlock(&scan_mutex);
3616                                 break;
3617                         }
3618
3619                         h = list_entry(scan_q.next,
3620                                        struct ctlr_info,
3621                                        scan_list);
3622                         list_del(&h->scan_list);
3623                         h->busy_scanning = 1;
3624                         mutex_unlock(&scan_mutex);
3625
3626                         rebuild_lun_table(h, 0, 0);
3627                         complete_all(&h->scan_wait);
3628                         mutex_lock(&scan_mutex);
3629                         h->busy_scanning = 0;
3630                         mutex_unlock(&scan_mutex);
3631                 }
3632         }
3633
3634         return 0;
3635 }
3636
3637 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3638 {
3639         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3640                 return 0;
3641
3642         switch (c->err_info->SenseInfo[12]) {
3643         case STATE_CHANGED:
3644                 printk(KERN_WARNING "cciss%d: a state change "
3645                         "detected, command retried\n", h->ctlr);
3646                 return 1;
3647         break;
3648         case LUN_FAILED:
3649                 printk(KERN_WARNING "cciss%d: LUN failure "
3650                         "detected, action required\n", h->ctlr);
3651                 return 1;
3652         break;
3653         case REPORT_LUNS_CHANGED:
3654                 printk(KERN_WARNING "cciss%d: report LUN data "
3655                         "changed\n", h->ctlr);
3656         /*
3657          * Here, we could call add_to_scan_list and wake up the scan thread,
3658          * except that it's quite likely that we will get more than one
3659          * REPORT_LUNS_CHANGED condition in quick succession, which means
3660          * that those which occur after the first one will likely happen
3661          * *during* the scan_thread's rescan.  And the rescan code is not
3662          * robust enough to restart in the middle, undoing what it has already
3663          * done, and it's not clear that it's even possible to do this, since
3664          * part of what it does is notify the block layer, which starts
3665          * doing it's own i/o to read partition tables and so on, and the
3666          * driver doesn't have visibility to know what might need undoing.
3667          * In any event, if possible, it is horribly complicated to get right
3668          * so we just don't do it for now.
3669          *
3670          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3671          */
3672                 return 1;
3673         break;
3674         case POWER_OR_RESET:
3675                 printk(KERN_WARNING "cciss%d: a power on "
3676                         "or device reset detected\n", h->ctlr);
3677                 return 1;
3678         break;
3679         case UNIT_ATTENTION_CLEARED:
3680                 printk(KERN_WARNING "cciss%d: unit attention "
3681                     "cleared by another initiator\n", h->ctlr);
3682                 return 1;
3683         break;
3684         default:
3685                 printk(KERN_WARNING "cciss%d: unknown "
3686                         "unit attention detected\n", h->ctlr);
3687                                 return 1;
3688         }
3689 }
3690
3691 /*
3692  *  We cannot read the structure directly, for portability we must use
3693  *   the io functions.
3694  *   This is for debug only.
3695  */
3696 #ifdef CCISS_DEBUG
3697 static void print_cfg_table(CfgTable_struct *tb)
3698 {
3699         int i;
3700         char temp_name[17];
3701
3702         printk("Controller Configuration information\n");
3703         printk("------------------------------------\n");
3704         for (i = 0; i < 4; i++)
3705                 temp_name[i] = readb(&(tb->Signature[i]));
3706         temp_name[4] = '\0';
3707         printk("   Signature = %s\n", temp_name);
3708         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3709         printk("   Transport methods supported = 0x%x\n",
3710                readl(&(tb->TransportSupport)));
3711         printk("   Transport methods active = 0x%x\n",
3712                readl(&(tb->TransportActive)));
3713         printk("   Requested transport Method = 0x%x\n",
3714                readl(&(tb->HostWrite.TransportRequest)));
3715         printk("   Coalesce Interrupt Delay = 0x%x\n",
3716                readl(&(tb->HostWrite.CoalIntDelay)));
3717         printk("   Coalesce Interrupt Count = 0x%x\n",
3718                readl(&(tb->HostWrite.CoalIntCount)));
3719         printk("   Max outstanding commands = 0x%d\n",
3720                readl(&(tb->CmdsOutMax)));
3721         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3722         for (i = 0; i < 16; i++)
3723                 temp_name[i] = readb(&(tb->ServerName[i]));
3724         temp_name[16] = '\0';
3725         printk("   Server Name = %s\n", temp_name);
3726         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3727 }
3728 #endif                          /* CCISS_DEBUG */
3729
3730 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3731 {
3732         int i, offset, mem_type, bar_type;
3733         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3734                 return 0;
3735         offset = 0;
3736         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3737                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3738                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3739                         offset += 4;
3740                 else {
3741                         mem_type = pci_resource_flags(pdev, i) &
3742                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3743                         switch (mem_type) {
3744                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3745                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3746                                 offset += 4;    /* 32 bit */
3747                                 break;
3748                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3749                                 offset += 8;
3750                                 break;
3751                         default:        /* reserved in PCI 2.2 */
3752                                 printk(KERN_WARNING
3753                                        "Base address is invalid\n");
3754                                 return -1;
3755                                 break;
3756                         }
3757                 }
3758                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3759                         return i + 1;
3760         }
3761         return -1;
3762 }
3763
3764 /* Fill in bucket_map[], given nsgs (the max number of
3765  * scatter gather elements supported) and bucket[],
3766  * which is an array of 8 integers.  The bucket[] array
3767  * contains 8 different DMA transfer sizes (in 16
3768  * byte increments) which the controller uses to fetch
3769  * commands.  This function fills in bucket_map[], which
3770  * maps a given number of scatter gather elements to one of
3771  * the 8 DMA transfer sizes.  The point of it is to allow the
3772  * controller to only do as much DMA as needed to fetch the
3773  * command, with the DMA transfer size encoded in the lower
3774  * bits of the command address.
3775  */
3776 static void  calc_bucket_map(int bucket[], int num_buckets,
3777         int nsgs, int *bucket_map)
3778 {
3779         int i, j, b, size;
3780
3781         /* even a command with 0 SGs requires 4 blocks */
3782 #define MINIMUM_TRANSFER_BLOCKS 4
3783 #define NUM_BUCKETS 8
3784         /* Note, bucket_map must have nsgs+1 entries. */
3785         for (i = 0; i <= nsgs; i++) {
3786                 /* Compute size of a command with i SG entries */
3787                 size = i + MINIMUM_TRANSFER_BLOCKS;
3788                 b = num_buckets; /* Assume the biggest bucket */
3789                 /* Find the bucket that is just big enough */
3790                 for (j = 0; j < 8; j++) {
3791                         if (bucket[j] >= size) {
3792                                 b = j;
3793                                 break;
3794                         }
3795                 }
3796                 /* for a command with i SG entries, use bucket b. */
3797                 bucket_map[i] = b;
3798         }
3799 }
3800
3801 static void
3802 cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3803 {
3804         int l = 0;
3805         __u32 trans_support;
3806         __u32 trans_offset;
3807                         /*
3808                          *  5 = 1 s/g entry or 4k
3809                          *  6 = 2 s/g entry or 8k
3810                          *  8 = 4 s/g entry or 16k
3811                          * 10 = 6 s/g entry or 24k
3812                          */
3813         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3814         unsigned long register_value;
3815
3816         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3817
3818         /* Attempt to put controller into performant mode if supported */
3819         /* Does board support performant mode? */
3820         trans_support = readl(&(h->cfgtable->TransportSupport));
3821         if (!(trans_support & PERFORMANT_MODE))
3822                 return;
3823
3824         printk(KERN_WARNING "cciss%d: Placing controller into "
3825                                 "performant mode\n", h->ctlr);
3826         /* Performant mode demands commands on a 32 byte boundary
3827          * pci_alloc_consistent aligns on page boundarys already.
3828          * Just need to check if divisible by 32
3829          */
3830         if ((sizeof(CommandList_struct) % 32) != 0) {
3831                 printk(KERN_WARNING "%s %d %s\n",
3832                         "cciss info: command size[",
3833                         (int)sizeof(CommandList_struct),
3834                         "] not divisible by 32, no performant mode..\n");
3835                 return;
3836         }
3837
3838         /* Performant mode ring buffer and supporting data structures */
3839         h->reply_pool = (__u64 *)pci_alloc_consistent(
3840                 h->pdev, h->max_commands * sizeof(__u64),
3841                 &(h->reply_pool_dhandle));
3842
3843         /* Need a block fetch table for performant mode */
3844         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3845                 sizeof(__u32)), GFP_KERNEL);
3846
3847         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3848                 goto clean_up;
3849
3850         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3851
3852         /* Controller spec: zero out this buffer. */
3853         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3854         h->reply_pool_head = h->reply_pool;
3855
3856         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3857         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3858                                 h->blockFetchTable);
3859         writel(bft[0], &h->transtable->BlockFetch0);
3860         writel(bft[1], &h->transtable->BlockFetch1);
3861         writel(bft[2], &h->transtable->BlockFetch2);
3862         writel(bft[3], &h->transtable->BlockFetch3);
3863         writel(bft[4], &h->transtable->BlockFetch4);
3864         writel(bft[5], &h->transtable->BlockFetch5);
3865         writel(bft[6], &h->transtable->BlockFetch6);
3866         writel(bft[7], &h->transtable->BlockFetch7);
3867
3868         /* size of controller ring buffer */
3869         writel(h->max_commands, &h->transtable->RepQSize);
3870         writel(1, &h->transtable->RepQCount);
3871         writel(0, &h->transtable->RepQCtrAddrLow32);
3872         writel(0, &h->transtable->RepQCtrAddrHigh32);
3873         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3874         writel(0, &h->transtable->RepQAddr0High32);
3875         writel(CFGTBL_Trans_Performant,
3876                         &(h->cfgtable->HostWrite.TransportRequest));
3877
3878         h->transMethod = CFGTBL_Trans_Performant;
3879         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3880         /* under certain very rare conditions, this can take awhile.
3881          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3882          * as we enter this code.) */
3883         for (l = 0; l < MAX_CONFIG_WAIT; l++) {
3884                 register_value = readl(h->vaddr + SA5_DOORBELL);
3885                 if (!(register_value & CFGTBL_ChangeReq))
3886                         break;
3887                 /* delay and try again */
3888                 set_current_state(TASK_INTERRUPTIBLE);
3889                 schedule_timeout(10);
3890         }
3891         register_value = readl(&(h->cfgtable->TransportActive));
3892         if (!(register_value & CFGTBL_Trans_Performant)) {
3893                 printk(KERN_WARNING "cciss: unable to get board into"
3894                                         " performant mode\n");
3895                 return;
3896         }
3897
3898         /* Change the access methods to the performant access methods */
3899         h->access = SA5_performant_access;
3900
3901         return;
3902 clean_up:
3903         kfree(h->blockFetchTable);
3904         if (h->reply_pool)
3905                 pci_free_consistent(h->pdev,
3906                                 h->max_commands * sizeof(__u64),
3907                                 h->reply_pool,
3908                                 h->reply_pool_dhandle);
3909         return;
3910
3911 } /* cciss_put_controller_into_performant_mode */
3912
3913 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3914  * controllers that are capable. If not, we use IO-APIC mode.
3915  */
3916
3917 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3918                                            struct pci_dev *pdev, __u32 board_id)
3919 {
3920 #ifdef CONFIG_PCI_MSI
3921         int err;
3922         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3923         {0, 2}, {0, 3}
3924         };
3925
3926         /* Some boards advertise MSI but don't really support it */
3927         if ((board_id == 0x40700E11) ||
3928             (board_id == 0x40800E11) ||
3929             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3930                 goto default_int_mode;
3931
3932         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3933                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3934                 if (!err) {
3935                         c->intr[0] = cciss_msix_entries[0].vector;
3936                         c->intr[1] = cciss_msix_entries[1].vector;
3937                         c->intr[2] = cciss_msix_entries[2].vector;
3938                         c->intr[3] = cciss_msix_entries[3].vector;
3939                         c->msix_vector = 1;
3940                         return;
3941                 }
3942                 if (err > 0) {
3943                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3944                                "available\n", err);
3945                         goto default_int_mode;
3946                 } else {
3947                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3948                                err);
3949                         goto default_int_mode;
3950                 }
3951         }
3952         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3953                 if (!pci_enable_msi(pdev)) {
3954                         c->msi_vector = 1;
3955                 } else {
3956                         printk(KERN_WARNING "cciss: MSI init failed\n");
3957                 }
3958         }
3959 default_int_mode:
3960 #endif                          /* CONFIG_PCI_MSI */
3961         /* if we get here we're going to use the default interrupt mode */
3962         c->intr[PERF_MODE_INT] = pdev->irq;
3963         return;
3964 }
3965
3966 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3967 {
3968         ushort subsystem_vendor_id, subsystem_device_id, command;
3969         __u32 board_id, scratchpad = 0;
3970         __u64 cfg_offset;
3971         __u32 cfg_base_addr;
3972         __u64 cfg_base_addr_index;
3973         int i, prod_index, err;
3974         __u32 trans_offset;
3975
3976         subsystem_vendor_id = pdev->subsystem_vendor;
3977         subsystem_device_id = pdev->subsystem_device;
3978         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3979                     subsystem_vendor_id);
3980
3981         for (i = 0; i < ARRAY_SIZE(products); i++) {
3982                 /* Stand aside for hpsa driver on request */
3983                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3984                         return -ENODEV;
3985                 if (board_id == products[i].board_id)
3986                         break;
3987         }
3988         prod_index = i;
3989         if (prod_index == ARRAY_SIZE(products)) {
3990                 dev_warn(&pdev->dev,
3991                         "unrecognized board ID: 0x%08lx, ignoring.\n",
3992                         (unsigned long) board_id);
3993                 return -ENODEV;
3994         }
3995
3996         /* check to see if controller has been disabled */
3997         /* BEFORE trying to enable it */
3998         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3999         if (!(command & 0x02)) {
4000                 printk(KERN_WARNING
4001                        "cciss: controller appears to be disabled\n");
4002                 return -ENODEV;
4003         }
4004
4005         err = pci_enable_device(pdev);
4006         if (err) {
4007                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
4008                 return err;
4009         }
4010
4011         err = pci_request_regions(pdev, "cciss");
4012         if (err) {
4013                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
4014                        "aborting\n");
4015                 return err;
4016         }
4017
4018 #ifdef CCISS_DEBUG
4019         printk("command = %x\n", command);
4020         printk("irq = %x\n", pdev->irq);
4021         printk("board_id = %x\n", board_id);
4022 #endif                          /* CCISS_DEBUG */
4023
4024 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4025  * else we use the IO-APIC interrupt assigned to us by system ROM.
4026  */
4027         cciss_interrupt_mode(c, pdev, board_id);
4028
4029         /* find the memory BAR */
4030         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
4031                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
4032                         break;
4033         }
4034         if (i == DEVICE_COUNT_RESOURCE) {
4035                 printk(KERN_WARNING "cciss: No memory BAR found\n");
4036                 err = -ENODEV;
4037                 goto err_out_free_res;
4038         }
4039
4040         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
4041                                                  * already removed
4042                                                  */
4043
4044 #ifdef CCISS_DEBUG
4045         printk("address 0 = %lx\n", c->paddr);
4046 #endif                          /* CCISS_DEBUG */
4047         c->vaddr = remap_pci_mem(c->paddr, 0x250);
4048
4049         /* Wait for the board to become ready.  (PCI hotplug needs this.)
4050          * We poll for up to 120 secs, once per 100ms. */
4051         for (i = 0; i < 1200; i++) {
4052                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
4053                 if (scratchpad == CCISS_FIRMWARE_READY)
4054                         break;
4055                 set_current_state(TASK_INTERRUPTIBLE);
4056                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
4057         }
4058         if (scratchpad != CCISS_FIRMWARE_READY) {
4059                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
4060                 err = -ENODEV;
4061                 goto err_out_free_res;
4062         }
4063
4064         /* get the address index number */
4065         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
4066         cfg_base_addr &= (__u32) 0x0000ffff;
4067 #ifdef CCISS_DEBUG
4068         printk("cfg base address = %x\n", cfg_base_addr);
4069 #endif                          /* CCISS_DEBUG */
4070         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
4071 #ifdef CCISS_DEBUG
4072         printk("cfg base address index = %llx\n",
4073                 (unsigned long long)cfg_base_addr_index);
4074 #endif                          /* CCISS_DEBUG */
4075         if (cfg_base_addr_index == -1) {
4076                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
4077                 err = -ENODEV;
4078                 goto err_out_free_res;
4079         }
4080
4081         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
4082 #ifdef CCISS_DEBUG
4083         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
4084 #endif                          /* CCISS_DEBUG */
4085         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
4086                                                        cfg_base_addr_index) +
4087                                     cfg_offset, sizeof(CfgTable_struct));
4088         /* Find performant mode table. */
4089         trans_offset = readl(&(c->cfgtable->TransMethodOffset));
4090         c->transtable = remap_pci_mem(pci_resource_start(pdev,
4091                 cfg_base_addr_index) + cfg_offset+trans_offset,
4092                 sizeof(*c->transtable));
4093         c->board_id = board_id;
4094
4095         #ifdef CCISS_DEBUG
4096                 print_cfg_table(c->cfgtable);
4097         #endif                          /* CCISS_DEBUG */
4098
4099         /* Some controllers support Zero Memory Raid (ZMR).
4100          * When configured in ZMR mode the number of supported
4101          * commands drops to 64. So instead of just setting an
4102          * arbitrary value we make the driver a little smarter.
4103          * We read the config table to tell us how many commands
4104          * are supported on the controller then subtract 4 to
4105          * leave a little room for ioctl calls.
4106          */
4107         c->max_commands = readl(&(c->cfgtable->MaxPerformantModeCommands));
4108         c->maxsgentries = readl(&(c->cfgtable->MaxSGElements));
4109
4110         /*
4111          * Limit native command to 32 s/g elements to save dma'able memory.
4112          * Howvever spec says if 0, use 31
4113          */
4114
4115         c->max_cmd_sgentries = 31;
4116         if (c->maxsgentries > 512) {
4117                 c->max_cmd_sgentries = 32;
4118                 c->chainsize = c->maxsgentries - c->max_cmd_sgentries + 1;
4119                 c->maxsgentries -= 1;   /* account for chain pointer */
4120         } else {
4121                 c->maxsgentries = 31;   /* Default to traditional value */
4122                 c->chainsize = 0;       /* traditional */
4123         }
4124
4125         c->product_name = products[prod_index].product_name;
4126         c->access = *(products[prod_index].access);
4127         c->nr_cmds = c->max_commands - 4;
4128         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
4129             (readb(&c->cfgtable->Signature[1]) != 'I') ||
4130             (readb(&c->cfgtable->Signature[2]) != 'S') ||
4131             (readb(&c->cfgtable->Signature[3]) != 'S')) {
4132                 printk("Does not appear to be a valid CISS config table\n");
4133                 err = -ENODEV;
4134                 goto err_out_free_res;
4135         }
4136 #ifdef CONFIG_X86
4137         {
4138                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4139                 __u32 prefetch;
4140                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
4141                 prefetch |= 0x100;
4142                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
4143         }
4144 #endif
4145
4146         /* Disabling DMA prefetch and refetch for the P600.
4147          * An ASIC bug may result in accesses to invalid memory addresses.
4148          * We've disabled prefetch for some time now. Testing with XEN
4149          * kernels revealed a bug in the refetch if dom0 resides on a P600.
4150          */
4151         if(board_id == 0x3225103C) {
4152                         __u32 dma_prefetch;
4153                 __u32 dma_refetch;
4154                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
4155                 dma_prefetch |= 0x8000;
4156                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
4157                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
4158                 dma_refetch |= 0x1;
4159                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
4160         }
4161
4162 #ifdef CCISS_DEBUG
4163         printk(KERN_WARNING "Trying to put board into Performant mode\n");
4164 #endif                          /* CCISS_DEBUG */
4165         cciss_put_controller_into_performant_mode(c);
4166         return 0;
4167
4168 err_out_free_res:
4169         /*
4170          * Deliberately omit pci_disable_device(): it does something nasty to
4171          * Smart Array controllers that pci_enable_device does not undo
4172          */
4173         pci_release_regions(pdev);
4174         return err;
4175 }
4176
4177 /* Function to find the first free pointer into our hba[] array
4178  * Returns -1 if no free entries are left.
4179  */
4180 static int alloc_cciss_hba(void)
4181 {
4182         int i;
4183
4184         for (i = 0; i < MAX_CTLR; i++) {
4185                 if (!hba[i]) {
4186                         ctlr_info_t *p;
4187
4188                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4189                         if (!p)
4190                                 goto Enomem;
4191                         hba[i] = p;
4192                         return i;
4193                 }
4194         }
4195         printk(KERN_WARNING "cciss: This driver supports a maximum"
4196                " of %d controllers.\n", MAX_CTLR);
4197         return -1;
4198 Enomem:
4199         printk(KERN_ERR "cciss: out of memory.\n");
4200         return -1;
4201 }
4202
4203 static void free_hba(int n)
4204 {
4205         ctlr_info_t *h = hba[n];
4206         int i;
4207
4208         hba[n] = NULL;
4209         for (i = 0; i < h->highest_lun + 1; i++)
4210                 if (h->gendisk[i] != NULL)
4211                         put_disk(h->gendisk[i]);
4212         kfree(h);
4213 }
4214
4215 /* Send a message CDB to the firmware. */
4216 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4217 {
4218         typedef struct {
4219                 CommandListHeader_struct CommandHeader;
4220                 RequestBlock_struct Request;
4221                 ErrDescriptor_struct ErrorDescriptor;
4222         } Command;
4223         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4224         Command *cmd;
4225         dma_addr_t paddr64;
4226         uint32_t paddr32, tag;
4227         void __iomem *vaddr;
4228         int i, err;
4229
4230         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4231         if (vaddr == NULL)
4232                 return -ENOMEM;
4233
4234         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4235            CCISS commands, so they must be allocated from the lower 4GiB of
4236            memory. */
4237         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4238         if (err) {
4239                 iounmap(vaddr);
4240                 return -ENOMEM;
4241         }
4242
4243         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4244         if (cmd == NULL) {
4245                 iounmap(vaddr);
4246                 return -ENOMEM;
4247         }
4248
4249         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4250            although there's no guarantee, we assume that the address is at
4251            least 4-byte aligned (most likely, it's page-aligned). */
4252         paddr32 = paddr64;
4253
4254         cmd->CommandHeader.ReplyQueue = 0;
4255         cmd->CommandHeader.SGList = 0;
4256         cmd->CommandHeader.SGTotal = 0;
4257         cmd->CommandHeader.Tag.lower = paddr32;
4258         cmd->CommandHeader.Tag.upper = 0;
4259         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4260
4261         cmd->Request.CDBLen = 16;
4262         cmd->Request.Type.Type = TYPE_MSG;
4263         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4264         cmd->Request.Type.Direction = XFER_NONE;
4265         cmd->Request.Timeout = 0; /* Don't time out */
4266         cmd->Request.CDB[0] = opcode;
4267         cmd->Request.CDB[1] = type;
4268         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4269
4270         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4271         cmd->ErrorDescriptor.Addr.upper = 0;
4272         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4273
4274         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4275
4276         for (i = 0; i < 10; i++) {
4277                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4278                 if ((tag & ~3) == paddr32)
4279                         break;
4280                 schedule_timeout_uninterruptible(HZ);
4281         }
4282
4283         iounmap(vaddr);
4284
4285         /* we leak the DMA buffer here ... no choice since the controller could
4286            still complete the command. */
4287         if (i == 10) {
4288                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4289                         opcode, type);
4290                 return -ETIMEDOUT;
4291         }
4292
4293         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4294
4295         if (tag & 2) {
4296                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4297                         opcode, type);
4298                 return -EIO;
4299         }
4300
4301         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4302                 opcode, type);
4303         return 0;
4304 }
4305
4306 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4307 #define cciss_noop(p) cciss_message(p, 3, 0)
4308
4309 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4310 {
4311 /* the #defines are stolen from drivers/pci/msi.h. */
4312 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4313 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4314
4315         int pos;
4316         u16 control = 0;
4317
4318         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4319         if (pos) {
4320                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4321                 if (control & PCI_MSI_FLAGS_ENABLE) {
4322                         printk(KERN_INFO "cciss: resetting MSI\n");
4323                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4324                 }
4325         }
4326
4327         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4328         if (pos) {
4329                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4330                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4331                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4332                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4333                 }
4334         }
4335
4336         return 0;
4337 }
4338
4339 /* This does a hard reset of the controller using PCI power management
4340  * states. */
4341 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4342 {
4343         u16 pmcsr, saved_config_space[32];
4344         int i, pos;
4345
4346         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4347
4348         /* This is very nearly the same thing as
4349
4350            pci_save_state(pci_dev);
4351            pci_set_power_state(pci_dev, PCI_D3hot);
4352            pci_set_power_state(pci_dev, PCI_D0);
4353            pci_restore_state(pci_dev);
4354
4355            but we can't use these nice canned kernel routines on
4356            kexec, because they also check the MSI/MSI-X state in PCI
4357            configuration space and do the wrong thing when it is
4358            set/cleared.  Also, the pci_save/restore_state functions
4359            violate the ordering requirements for restoring the
4360            configuration space from the CCISS document (see the
4361            comment below).  So we roll our own .... */
4362
4363         for (i = 0; i < 32; i++)
4364                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4365
4366         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4367         if (pos == 0) {
4368                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4369                 return -ENODEV;
4370         }
4371
4372         /* Quoting from the Open CISS Specification: "The Power
4373          * Management Control/Status Register (CSR) controls the power
4374          * state of the device.  The normal operating state is D0,
4375          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4376          * the controller, place the interface device in D3 then to
4377          * D0, this causes a secondary PCI reset which will reset the
4378          * controller." */
4379
4380         /* enter the D3hot power management state */
4381         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4382         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4383         pmcsr |= PCI_D3hot;
4384         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4385
4386         schedule_timeout_uninterruptible(HZ >> 1);
4387
4388         /* enter the D0 power management state */
4389         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4390         pmcsr |= PCI_D0;
4391         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4392
4393         schedule_timeout_uninterruptible(HZ >> 1);
4394
4395         /* Restore the PCI configuration space.  The Open CISS
4396          * Specification says, "Restore the PCI Configuration
4397          * Registers, offsets 00h through 60h. It is important to
4398          * restore the command register, 16-bits at offset 04h,
4399          * last. Do not restore the configuration status register,
4400          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4401         for (i = 0; i < 32; i++) {
4402                 if (i == 2 || i == 3)
4403                         continue;
4404                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4405         }
4406         wmb();
4407         pci_write_config_word(pdev, 4, saved_config_space[2]);
4408
4409         return 0;
4410 }
4411
4412 /*
4413  *  This is it.  Find all the controllers and register them.  I really hate
4414  *  stealing all these major device numbers.
4415  *  returns the number of block devices registered.
4416  */
4417 static int __devinit cciss_init_one(struct pci_dev *pdev,
4418                                     const struct pci_device_id *ent)
4419 {
4420         int i;
4421         int j = 0;
4422         int k = 0;
4423         int rc;
4424         int dac, return_code;
4425         InquiryData_struct *inq_buff;
4426
4427         if (reset_devices) {
4428                 /* Reset the controller with a PCI power-cycle */
4429                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4430                         return -ENODEV;
4431
4432                 /* Now try to get the controller to respond to a no-op. Some
4433                    devices (notably the HP Smart Array 5i Controller) need
4434                    up to 30 seconds to respond. */
4435                 for (i=0; i<30; i++) {
4436                         if (cciss_noop(pdev) == 0)
4437                                 break;
4438
4439                         schedule_timeout_uninterruptible(HZ);
4440                 }
4441                 if (i == 30) {
4442                         printk(KERN_ERR "cciss: controller seems dead\n");
4443                         return -EBUSY;
4444                 }
4445         }
4446
4447         i = alloc_cciss_hba();
4448         if (i < 0)
4449                 return -1;
4450         hba[i]->busy_initializing = 1;
4451         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4452         INIT_HLIST_HEAD(&hba[i]->reqQ);
4453         mutex_init(&hba[i]->busy_shutting_down);
4454
4455         if (cciss_pci_init(hba[i], pdev) != 0)
4456                 goto clean_no_release_regions;
4457
4458         sprintf(hba[i]->devname, "cciss%d", i);
4459         hba[i]->ctlr = i;
4460         hba[i]->pdev = pdev;
4461
4462         init_completion(&hba[i]->scan_wait);
4463
4464         if (cciss_create_hba_sysfs_entry(hba[i]))
4465                 goto clean0;
4466
4467         /* configure PCI DMA stuff */
4468         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4469                 dac = 1;
4470         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4471                 dac = 0;
4472         else {
4473                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4474                 goto clean1;
4475         }
4476
4477         /*
4478          * register with the major number, or get a dynamic major number
4479          * by passing 0 as argument.  This is done for greater than
4480          * 8 controller support.
4481          */
4482         if (i < MAX_CTLR_ORIG)
4483                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4484         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4485         if (rc == -EBUSY || rc == -EINVAL) {
4486                 printk(KERN_ERR
4487                        "cciss:  Unable to get major number %d for %s "
4488                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4489                 goto clean1;
4490         } else {
4491                 if (i >= MAX_CTLR_ORIG)
4492                         hba[i]->major = rc;
4493         }
4494
4495         /* make sure the board interrupts are off */
4496         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4497         if (hba[i]->msi_vector || hba[i]->msix_vector) {
4498                 if (request_irq(hba[i]->intr[PERF_MODE_INT],
4499                                 do_cciss_msix_intr,
4500                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4501                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4502                                hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4503                         goto clean2;
4504                 }
4505         } else {
4506                 if (request_irq(hba[i]->intr[PERF_MODE_INT], do_cciss_intx,
4507                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4508                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4509                                hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4510                         goto clean2;
4511                 }
4512         }
4513
4514         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4515                hba[i]->devname, pdev->device, pci_name(pdev),
4516                hba[i]->intr[PERF_MODE_INT], dac ? "" : " not");
4517
4518         hba[i]->cmd_pool_bits =
4519             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4520                         * sizeof(unsigned long), GFP_KERNEL);
4521         hba[i]->cmd_pool = (CommandList_struct *)
4522             pci_alloc_consistent(hba[i]->pdev,
4523                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4524                     &(hba[i]->cmd_pool_dhandle));
4525         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4526             pci_alloc_consistent(hba[i]->pdev,
4527                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4528                     &(hba[i]->errinfo_pool_dhandle));
4529         if ((hba[i]->cmd_pool_bits == NULL)
4530             || (hba[i]->cmd_pool == NULL)
4531             || (hba[i]->errinfo_pool == NULL)) {
4532                 printk(KERN_ERR "cciss: out of memory");
4533                 goto clean4;
4534         }
4535
4536         /* Need space for temp scatter list */
4537         hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4538                                                 sizeof(struct scatterlist *),
4539                                                 GFP_KERNEL);
4540         for (k = 0; k < hba[i]->nr_cmds; k++) {
4541                 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4542                                                         hba[i]->maxsgentries,
4543                                                         GFP_KERNEL);
4544                 if (hba[i]->scatter_list[k] == NULL) {
4545                         printk(KERN_ERR "cciss%d: could not allocate "
4546                                 "s/g lists\n", i);
4547                         goto clean4;
4548                 }
4549         }
4550         hba[i]->cmd_sg_list = cciss_allocate_sg_chain_blocks(hba[i],
4551                 hba[i]->chainsize, hba[i]->nr_cmds);
4552         if (!hba[i]->cmd_sg_list && hba[i]->chainsize > 0)
4553                 goto clean4;
4554
4555         spin_lock_init(&hba[i]->lock);
4556
4557         /* Initialize the pdev driver private data.
4558            have it point to hba[i].  */
4559         pci_set_drvdata(pdev, hba[i]);
4560         /* command and error info recs zeroed out before
4561            they are used */
4562         memset(hba[i]->cmd_pool_bits, 0,
4563                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4564                         * sizeof(unsigned long));
4565
4566         hba[i]->num_luns = 0;
4567         hba[i]->highest_lun = -1;
4568         for (j = 0; j < CISS_MAX_LUN; j++) {
4569                 hba[i]->drv[j] = NULL;
4570                 hba[i]->gendisk[j] = NULL;
4571         }
4572
4573         cciss_scsi_setup(i);
4574
4575         /* Turn the interrupts on so we can service requests */
4576         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4577
4578         /* Get the firmware version */
4579         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4580         if (inq_buff == NULL) {
4581                 printk(KERN_ERR "cciss: out of memory\n");
4582                 goto clean4;
4583         }
4584
4585         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4586                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4587         if (return_code == IO_OK) {
4588                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4589                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4590                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4591                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4592         } else {         /* send command failed */
4593                 printk(KERN_WARNING "cciss: unable to determine firmware"
4594                         " version of controller\n");
4595         }
4596         kfree(inq_buff);
4597
4598         cciss_procinit(i);
4599
4600         hba[i]->cciss_max_sectors = 8192;
4601
4602         rebuild_lun_table(hba[i], 1, 0);
4603         hba[i]->busy_initializing = 0;
4604         return 1;
4605
4606 clean4:
4607         kfree(hba[i]->cmd_pool_bits);
4608         /* Free up sg elements */
4609         for (k = 0; k < hba[i]->nr_cmds; k++)
4610                 kfree(hba[i]->scatter_list[k]);
4611         kfree(hba[i]->scatter_list);
4612         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4613         if (hba[i]->cmd_pool)
4614                 pci_free_consistent(hba[i]->pdev,
4615                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4616                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4617         if (hba[i]->errinfo_pool)
4618                 pci_free_consistent(hba[i]->pdev,
4619                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4620                                     hba[i]->errinfo_pool,
4621                                     hba[i]->errinfo_pool_dhandle);
4622         free_irq(hba[i]->intr[PERF_MODE_INT], hba[i]);
4623 clean2:
4624         unregister_blkdev(hba[i]->major, hba[i]->devname);
4625 clean1:
4626         cciss_destroy_hba_sysfs_entry(hba[i]);
4627 clean0:
4628         pci_release_regions(pdev);
4629 clean_no_release_regions:
4630         hba[i]->busy_initializing = 0;
4631
4632         /*
4633          * Deliberately omit pci_disable_device(): it does something nasty to
4634          * Smart Array controllers that pci_enable_device does not undo
4635          */
4636         pci_set_drvdata(pdev, NULL);
4637         free_hba(i);
4638         return -1;
4639 }
4640
4641 static void cciss_shutdown(struct pci_dev *pdev)
4642 {
4643         ctlr_info_t *h;
4644         char *flush_buf;
4645         int return_code;
4646
4647         h = pci_get_drvdata(pdev);
4648         flush_buf = kzalloc(4, GFP_KERNEL);
4649         if (!flush_buf) {
4650                 printk(KERN_WARNING
4651                         "cciss:%d cache not flushed, out of memory.\n",
4652                         h->ctlr);
4653                 return;
4654         }
4655         /* write all data in the battery backed cache to disk */
4656         memset(flush_buf, 0, 4);
4657         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4658                 4, 0, CTLR_LUNID, TYPE_CMD);
4659         kfree(flush_buf);
4660         if (return_code != IO_OK)
4661                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4662                         h->ctlr);
4663         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4664         free_irq(h->intr[PERF_MODE_INT], h);
4665 }
4666
4667 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4668 {
4669         ctlr_info_t *tmp_ptr;
4670         int i, j;
4671
4672         if (pci_get_drvdata(pdev) == NULL) {
4673                 printk(KERN_ERR "cciss: Unable to remove device \n");
4674                 return;
4675         }
4676
4677         tmp_ptr = pci_get_drvdata(pdev);
4678         i = tmp_ptr->ctlr;
4679         if (hba[i] == NULL) {
4680                 printk(KERN_ERR "cciss: device appears to "
4681                        "already be removed \n");
4682                 return;
4683         }
4684
4685         mutex_lock(&hba[i]->busy_shutting_down);
4686
4687         remove_from_scan_list(hba[i]);
4688         remove_proc_entry(hba[i]->devname, proc_cciss);
4689         unregister_blkdev(hba[i]->major, hba[i]->devname);
4690
4691         /* remove it from the disk list */
4692         for (j = 0; j < CISS_MAX_LUN; j++) {
4693                 struct gendisk *disk = hba[i]->gendisk[j];
4694                 if (disk) {
4695                         struct request_queue *q = disk->queue;
4696
4697                         if (disk->flags & GENHD_FL_UP) {
4698                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4699                                 del_gendisk(disk);
4700                         }
4701                         if (q)
4702                                 blk_cleanup_queue(q);
4703                 }
4704         }
4705
4706 #ifdef CONFIG_CISS_SCSI_TAPE
4707         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4708 #endif
4709
4710         cciss_shutdown(pdev);
4711
4712 #ifdef CONFIG_PCI_MSI
4713         if (hba[i]->msix_vector)
4714                 pci_disable_msix(hba[i]->pdev);
4715         else if (hba[i]->msi_vector)
4716                 pci_disable_msi(hba[i]->pdev);
4717 #endif                          /* CONFIG_PCI_MSI */
4718
4719         iounmap(hba[i]->vaddr);
4720
4721         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4722                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4723         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4724                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4725         kfree(hba[i]->cmd_pool_bits);
4726         /* Free up sg elements */
4727         for (j = 0; j < hba[i]->nr_cmds; j++)
4728                 kfree(hba[i]->scatter_list[j]);
4729         kfree(hba[i]->scatter_list);
4730         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4731         /*
4732          * Deliberately omit pci_disable_device(): it does something nasty to
4733          * Smart Array controllers that pci_enable_device does not undo
4734          */
4735         pci_release_regions(pdev);
4736         pci_set_drvdata(pdev, NULL);
4737         cciss_destroy_hba_sysfs_entry(hba[i]);
4738         mutex_unlock(&hba[i]->busy_shutting_down);
4739         free_hba(i);
4740 }
4741
4742 static struct pci_driver cciss_pci_driver = {
4743         .name = "cciss",
4744         .probe = cciss_init_one,
4745         .remove = __devexit_p(cciss_remove_one),
4746         .id_table = cciss_pci_device_id,        /* id_table */
4747         .shutdown = cciss_shutdown,
4748 };
4749
4750 /*
4751  *  This is it.  Register the PCI driver information for the cards we control
4752  *  the OS will call our registered routines when it finds one of our cards.
4753  */
4754 static int __init cciss_init(void)
4755 {
4756         int err;
4757
4758         /*
4759          * The hardware requires that commands are aligned on a 64-bit
4760          * boundary. Given that we use pci_alloc_consistent() to allocate an
4761          * array of them, the size must be a multiple of 8 bytes.
4762          */
4763         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4764         printk(KERN_INFO DRIVER_NAME "\n");
4765
4766         err = bus_register(&cciss_bus_type);
4767         if (err)
4768                 return err;
4769
4770         /* Start the scan thread */
4771         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4772         if (IS_ERR(cciss_scan_thread)) {
4773                 err = PTR_ERR(cciss_scan_thread);
4774                 goto err_bus_unregister;
4775         }
4776
4777         /* Register for our PCI devices */
4778         err = pci_register_driver(&cciss_pci_driver);
4779         if (err)
4780                 goto err_thread_stop;
4781
4782         return err;
4783
4784 err_thread_stop:
4785         kthread_stop(cciss_scan_thread);
4786 err_bus_unregister:
4787         bus_unregister(&cciss_bus_type);
4788
4789         return err;
4790 }
4791
4792 static void __exit cciss_cleanup(void)
4793 {
4794         int i;
4795
4796         pci_unregister_driver(&cciss_pci_driver);
4797         /* double check that all controller entrys have been removed */
4798         for (i = 0; i < MAX_CTLR; i++) {
4799                 if (hba[i] != NULL) {
4800                         printk(KERN_WARNING "cciss: had to remove"
4801                                " controller %d\n", i);
4802                         cciss_remove_one(hba[i]->pdev);
4803                 }
4804         }
4805         kthread_stop(cciss_scan_thread);
4806         remove_proc_entry("driver/cciss", NULL);
4807         bus_unregister(&cciss_bus_type);
4808 }
4809
4810 module_init(cciss_init);
4811 module_exit(cciss_cleanup);