drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP SA 5xxx and 6xxx Controllers
   3  *    Copyright 2000, 2005 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; either version 2 of the License, or
   8  *    (at your option) any later version.
   9  *
  10  *    This program is distributed in the hope that it will be useful,
  11  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  13  *    NON INFRINGEMENT.  See the GNU General Public License for more details.
  14  *
  15  *    You should have received a copy of the GNU General Public License
  16  *    along with this program; if not, write to the Free Software
  17  *    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/config.h>       /* CONFIG_PROC_FS */
  24 #include <linux/module.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/types.h>
  27 #include <linux/pci.h>
  28 #include <linux/kernel.h>
  29 #include <linux/slab.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/init.h>
  38 #include <linux/hdreg.h>
  39 #include <linux/spinlock.h>
  40 #include <linux/compat.h>
  41 #include <asm/uaccess.h>
  42 #include <asm/io.h>
  43
  44 #include <linux/dma-mapping.h>
  45 #include <linux/blkdev.h>
  46 #include <linux/genhd.h>
  47 #include <linux/completion.h>
  48
  49 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  50 #define DRIVER_NAME "HP CISS Driver (v 2.6.8)"
  51 #define DRIVER_VERSION CCISS_DRIVER_VERSION(2,6,8)
  52
  53 /* Embedded module documentation macros - see modules.h */
  54 MODULE_AUTHOR("Hewlett-Packard Company");
  55 MODULE_DESCRIPTION("Driver for HP Controller SA5xxx SA6xxx version 2.6.8");
  56 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
  57                         " SA6i P600 P800 P400 P400i E200 E200i");
  58 MODULE_LICENSE("GPL");
  59
  60 #include "cciss_cmd.h"
  61 #include "cciss.h"
  62 #include <linux/cciss_ioctl.h>
  63
  64 /* define the PCI info for the cards we can control */
  65 static const struct pci_device_id cciss_pci_device_id[] = {
  66         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,
  67                         0x0E11, 0x4070, 0, 0, 0},
  68         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB,
  69                         0x0E11, 0x4080, 0, 0, 0},
  70         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB,
  71                         0x0E11, 0x4082, 0, 0, 0},
  72         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB,
  73                         0x0E11, 0x4083, 0, 0, 0},
  74         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC,
  75                 0x0E11, 0x409A, 0, 0, 0},
  76         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC,
  77                 0x0E11, 0x409B, 0, 0, 0},
  78         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC,
  79                 0x0E11, 0x409C, 0, 0, 0},
  80         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC,
  81                 0x0E11, 0x409D, 0, 0, 0},
  82         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC,
  83                 0x0E11, 0x4091, 0, 0, 0},
  84         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSA,
  85                 0x103C, 0x3225, 0, 0, 0},
  86         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC,
  87                 0x103c, 0x3223, 0, 0, 0},
  88         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC,
  89                 0x103c, 0x3234, 0, 0, 0},
  90         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC,
  91                 0x103c, 0x3235, 0, 0, 0},
  92         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD,
  93                 0x103c, 0x3211, 0, 0, 0},
  94         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD,
  95                 0x103c, 0x3212, 0, 0, 0},
  96         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD,
  97                 0x103c, 0x3213, 0, 0, 0},
  98         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD,
  99                 0x103c, 0x3214, 0, 0, 0},
 100         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD,
 101                 0x103c, 0x3215, 0, 0, 0},
 102         {0,}
 103 };
 104 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 105
 106 #define NR_PRODUCTS (sizeof(products)/sizeof(struct board_type))
 107
 108 /*  board_id = Subsystem Device ID & Vendor ID
 109  *  product = Marketing Name for the board
 110  *  access = Address of the struct of function pointers
 111  */
 112 static struct board_type products[] = {
 113         { 0x40700E11, "Smart Array 5300", &SA5_access },
 114         { 0x40800E11, "Smart Array 5i", &SA5B_access},
 115         { 0x40820E11, "Smart Array 532", &SA5B_access},
 116         { 0x40830E11, "Smart Array 5312", &SA5B_access},
 117         { 0x409A0E11, "Smart Array 641", &SA5_access},
 118         { 0x409B0E11, "Smart Array 642", &SA5_access},
 119         { 0x409C0E11, "Smart Array 6400", &SA5_access},
 120         { 0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 121         { 0x40910E11, "Smart Array 6i", &SA5_access},
 122         { 0x3225103C, "Smart Array P600", &SA5_access},
 123         { 0x3223103C, "Smart Array P800", &SA5_access},
 124         { 0x3234103C, "Smart Array P400", &SA5_access},
 125         { 0x3235103C, "Smart Array P400i", &SA5_access},
 126         { 0x3211103C, "Smart Array E200i", &SA5_access},
 127         { 0x3212103C, "Smart Array E200", &SA5_access},
 128         { 0x3213103C, "Smart Array E200i", &SA5_access},
 129         { 0x3214103C, "Smart Array E200i", &SA5_access},
 130         { 0x3215103C, "Smart Array E200i", &SA5_access},
 131 };
 132
 133 /* How long to wait (in millesconds) for board to go into simple mode */
 134 #define MAX_CONFIG_WAIT 30000
 135 #define MAX_IOCTL_CONFIG_WAIT 1000
 136
 137 /*define how many times we will try a command because of bus resets */
 138 #define MAX_CMD_RETRIES 3
 139
 140 #define READ_AHEAD       1024
 141 #define NR_CMDS          384 /* #commands that can be outstanding */
 142 #define MAX_CTLR        32
 143
 144 /* Originally cciss driver only supports 8 major numbers */
 145 #define MAX_CTLR_ORIG   8
 146
 147
 148 static ctlr_info_t *hba[MAX_CTLR];
 149
 150 static void do_cciss_request(request_queue_t *q);
 151 static int cciss_open(struct inode *inode, struct file *filep);
 152 static int cciss_release(struct inode *inode, struct file *filep);
 153 static int cciss_ioctl(struct inode *inode, struct file *filep,
 154                 unsigned int cmd, unsigned long arg);
 155
 156 static int revalidate_allvol(ctlr_info_t *host);
 157 static int cciss_revalidate(struct gendisk *disk);
 158 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk);
 159 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv, int clear_all);
 160
 161 static void cciss_read_capacity(int ctlr, int logvol, ReadCapdata_struct *buf,
 162         int withirq, unsigned int *total_size, unsigned int *block_size);
 163 static void cciss_geometry_inquiry(int ctlr, int logvol,
 164                         int withirq, unsigned int total_size,
 165                         unsigned int block_size, InquiryData_struct *inq_buff,
 166                         drive_info_struct *drv);
 167 static void cciss_getgeometry(int cntl_num);
 168
 169 static void start_io( ctlr_info_t *h);
 170 static int sendcmd( __u8 cmd, int ctlr, void *buff, size_t size,
 171         unsigned int use_unit_num, unsigned int log_unit, __u8 page_code,
 172         unsigned char *scsi3addr, int cmd_type);
 173 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 174         unsigned int use_unit_num, unsigned int log_unit, __u8  page_code,
 175         int cmd_type);
 176
 177 static void fail_all_cmds(unsigned long ctlr);
 178
 179 #ifdef CONFIG_PROC_FS
 180 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
 181                 int length, int *eof, void *data);
 182 static void cciss_procinit(int i);
 183 #else
 184 static void cciss_procinit(int i) {}
 185 #endif /* CONFIG_PROC_FS */
 186
 187 #ifdef CONFIG_COMPAT
 188 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg);
 189 #endif
 190
 191 static struct block_device_operations cciss_fops  = {
 192         .owner          = THIS_MODULE,
 193         .open           = cciss_open,
 194         .release        = cciss_release,
 195         .ioctl          = cciss_ioctl,
 196 #ifdef CONFIG_COMPAT
 197         .compat_ioctl   = cciss_compat_ioctl,
 198 #endif
 199         .revalidate_disk= cciss_revalidate,
 200 };
 201
 202 /*
 203  * Enqueuing and dequeuing functions for cmdlists.
 204  */
 205 static inline void addQ(CommandList_struct **Qptr, CommandList_struct *c)
 206 {
 207         if (*Qptr == NULL) {
 208                 *Qptr = c;
 209                 c->next = c->prev = c;
 210         } else {
 211                 c->prev = (*Qptr)->prev;
 212                 c->next = (*Qptr);
 213                 (*Qptr)->prev->next = c;
 214                 (*Qptr)->prev = c;
 215         }
 216 }
 217
 218 static inline CommandList_struct *removeQ(CommandList_struct **Qptr,
 219                                                 CommandList_struct *c)
 220 {
 221         if (c && c->next != c) {
 222                 if (*Qptr == c) *Qptr = c->next;
 223                 c->prev->next = c->next;
 224                 c->next->prev = c->prev;
 225         } else {
 226                 *Qptr = NULL;
 227         }
 228         return c;
 229 }
 230
 231 #include "cciss_scsi.c"         /* For SCSI tape support */
 232
 233 #ifdef CONFIG_PROC_FS
 234
 235 /*
 236  * Report information about this controller.
 237  */
 238 #define ENG_GIG 1000000000
 239 #define ENG_GIG_FACTOR (ENG_GIG/512)
 240 #define RAID_UNKNOWN 6
 241 static const char *raid_label[] = {"0","4","1(1+0)","5","5+1","ADG",
 242                                            "UNKNOWN"};
 243
 244 static struct proc_dir_entry *proc_cciss;
 245
 246 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
 247                 int length, int *eof, void *data)
 248 {
 249         off_t pos = 0;
 250         off_t len = 0;
 251         int size, i, ctlr;
 252         ctlr_info_t *h = (ctlr_info_t*)data;
 253         drive_info_struct *drv;
 254         unsigned long flags;
 255         sector_t vol_sz, vol_sz_frac;
 256
 257         ctlr = h->ctlr;
 258
 259         /* prevent displaying bogus info during configuration
 260          * or deconfiguration of a logical volume
 261          */
 262         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 263         if (h->busy_configuring) {
 264                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 265         return -EBUSY;
 266         }
 267         h->busy_configuring = 1;
 268         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 269
 270         size = sprintf(buffer, "%s: HP %s Controller\n"
 271                 "Board ID: 0x%08lx\n"
 272                 "Firmware Version: %c%c%c%c\n"
 273                 "IRQ: %d\n"
 274                 "Logical drives: %d\n"
 275                 "Current Q depth: %d\n"
 276                 "Current # commands on controller: %d\n"
 277                 "Max Q depth since init: %d\n"
 278                 "Max # commands on controller since init: %d\n"
 279                 "Max SG entries since init: %d\n\n",
 280                 h->devname,
 281                 h->product_name,
 282                 (unsigned long)h->board_id,
 283                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2], h->firm_ver[3],
 284                 (unsigned int)h->intr,
 285                 h->num_luns,
 286                 h->Qdepth, h->commands_outstanding,
 287                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 288
 289         pos += size; len += size;
 290         cciss_proc_tape_report(ctlr, buffer, &pos, &len);
 291         for(i=0; i<=h->highest_lun; i++) {
 292
 293                 drv = &h->drv[i];
 294                 if (drv->heads == 0)
 295                         continue;
 296
 297                 vol_sz = drv->nr_blocks;
 298                 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 299                 vol_sz_frac *= 100;
 300                 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 301
 302                 if (drv->raid_level > 5)
 303                         drv->raid_level = RAID_UNKNOWN;
 304                 size = sprintf(buffer+len, "cciss/c%dd%d:"
 305                                 "\t%4u.%02uGB\tRAID %s\n",
 306                                 ctlr, i, (int)vol_sz, (int)vol_sz_frac,
 307                                 raid_label[drv->raid_level]);
 308                 pos += size; len += size;
 309         }
 310
 311         *eof = 1;
 312         *start = buffer+offset;
 313         len -= offset;
 314         if (len>length)
 315                 len = length;
 316         h->busy_configuring = 0;
 317         return len;
 318 }
 319
 320 static int
 321 cciss_proc_write(struct file *file, const char __user *buffer,
 322                         unsigned long count, void *data)
 323 {
 324         unsigned char cmd[80];
 325         int len;
 326 #ifdef CONFIG_CISS_SCSI_TAPE
 327         ctlr_info_t *h = (ctlr_info_t *) data;
 328         int rc;
 329 #endif
 330
 331         if (count > sizeof(cmd)-1) return -EINVAL;
 332         if (copy_from_user(cmd, buffer, count)) return -EFAULT;
 333         cmd[count] = '\0';
 334         len = strlen(cmd);      // above 3 lines ensure safety
 335         if (len && cmd[len-1] == '\n')
 336                 cmd[--len] = '\0';
 337 #       ifdef CONFIG_CISS_SCSI_TAPE
 338                 if (strcmp("engage scsi", cmd)==0) {
 339                         rc = cciss_engage_scsi(h->ctlr);
 340                         if (rc != 0) return -rc;
 341                         return count;
 342                 }
 343                 /* might be nice to have "disengage" too, but it's not
 344                    safely possible. (only 1 module use count, lock issues.) */
 345 #       endif
 346         return -EINVAL;
 347 }
 348
 349 /*
 350  * Get us a file in /proc/cciss that says something about each controller.
 351  * Create /proc/cciss if it doesn't exist yet.
 352  */
 353 static void __devinit cciss_procinit(int i)
 354 {
 355         struct proc_dir_entry *pde;
 356
 357         if (proc_cciss == NULL) {
 358                 proc_cciss = proc_mkdir("cciss", proc_root_driver);
 359                 if (!proc_cciss)
 360                         return;
 361         }
 362
 363         pde = create_proc_read_entry(hba[i]->devname,
 364                 S_IWUSR | S_IRUSR | S_IRGRP | S_IROTH,
 365                 proc_cciss, cciss_proc_get_info, hba[i]);
 366         pde->write_proc = cciss_proc_write;
 367 }
 368 #endif /* CONFIG_PROC_FS */
 369
 370 /*
 371  * For operations that cannot sleep, a command block is allocated at init,
 372  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 373  * which ones are free or in use.  For operations that can wait for kmalloc
 374  * to possible sleep, this routine can be called with get_from_pool set to 0.
 375  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 376  */
 377 static CommandList_struct * cmd_alloc(ctlr_info_t *h, int get_from_pool)
 378 {
 379         CommandList_struct *c;
 380         int i;
 381         u64bit temp64;
 382         dma_addr_t cmd_dma_handle, err_dma_handle;
 383
 384         if (!get_from_pool)
 385         {
 386                 c = (CommandList_struct *) pci_alloc_consistent(
 387                         h->pdev, sizeof(CommandList_struct), &cmd_dma_handle);
 388                 if(c==NULL)
 389                         return NULL;
 390                 memset(c, 0, sizeof(CommandList_struct));
 391
 392                 c->cmdindex = -1;
 393
 394                 c->err_info = (ErrorInfo_struct *)pci_alloc_consistent(
 395                                         h->pdev, sizeof(ErrorInfo_struct),
 396                                         &err_dma_handle);
 397
 398                 if (c->err_info == NULL)
 399                 {
 400                         pci_free_consistent(h->pdev,
 401                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 402                         return NULL;
 403                 }
 404                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 405         } else /* get it out of the controllers pool */
 406         {
 407                 do {
 408                         i = find_first_zero_bit(h->cmd_pool_bits, NR_CMDS);
 409                         if (i == NR_CMDS)
 410                                 return NULL;
 411                 } while(test_and_set_bit(i & (BITS_PER_LONG - 1), h->cmd_pool_bits+(i/BITS_PER_LONG)) != 0);
 412 #ifdef CCISS_DEBUG
 413                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 414 #endif
 415                 c = h->cmd_pool + i;
 416                 memset(c, 0, sizeof(CommandList_struct));
 417                 cmd_dma_handle = h->cmd_pool_dhandle
 418                                         + i*sizeof(CommandList_struct);
 419                 c->err_info = h->errinfo_pool + i;
 420                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 421                 err_dma_handle = h->errinfo_pool_dhandle
 422                                         + i*sizeof(ErrorInfo_struct);
 423                 h->nr_allocs++;
 424
 425                 c->cmdindex = i;
 426         }
 427
 428         c->busaddr = (__u32) cmd_dma_handle;
 429         temp64.val = (__u64) err_dma_handle;
 430         c->ErrDesc.Addr.lower = temp64.val32.lower;
 431         c->ErrDesc.Addr.upper = temp64.val32.upper;
 432         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 433
 434         c->ctlr = h->ctlr;
 435         return c;
 436
 437
 438 }
 439
 440 /*
 441  * Frees a command block that was previously allocated with cmd_alloc().
 442  */
 443 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 444 {
 445         int i;
 446         u64bit temp64;
 447
 448         if( !got_from_pool)
 449         {
 450                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 451                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 452                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 453                         c->err_info, (dma_addr_t) temp64.val);
 454                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 455                         c, (dma_addr_t) c->busaddr);
 456         } else
 457         {
 458                 i = c - h->cmd_pool;
 459                 clear_bit(i&(BITS_PER_LONG-1), h->cmd_pool_bits+(i/BITS_PER_LONG));
 460                 h->nr_frees++;
 461         }
 462 }
 463
 464 static inline ctlr_info_t *get_host(struct gendisk *disk)
 465 {
 466         return disk->queue->queuedata;
 467 }
 468
 469 static inline drive_info_struct *get_drv(struct gendisk *disk)
 470 {
 471         return disk->private_data;
 472 }
 473
 474 /*
 475  * Open.  Make sure the device is really there.
 476  */
 477 static int cciss_open(struct inode *inode, struct file *filep)
 478 {
 479         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
 480         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
 481
 482 #ifdef CCISS_DEBUG
 483         printk(KERN_DEBUG "cciss_open %s\n", inode->i_bdev->bd_disk->disk_name);
 484 #endif /* CCISS_DEBUG */
 485
 486         if (host->busy_initializing || drv->busy_configuring)
 487                 return -EBUSY;
 488         /*
 489          * Root is allowed to open raw volume zero even if it's not configured
 490          * so array config can still work. Root is also allowed to open any
 491          * volume that has a LUN ID, so it can issue IOCTL to reread the
 492          * disk information.  I don't think I really like this
 493          * but I'm already using way to many device nodes to claim another one
 494          * for "raw controller".
 495          */
 496         if (drv->nr_blocks == 0) {
 497                 if (iminor(inode) != 0) {       /* not node 0? */
 498                         /* if not node 0 make sure it is a partition = 0 */
 499                         if (iminor(inode) & 0x0f) {
 500                         return -ENXIO;
 501                                 /* if it is, make sure we have a LUN ID */
 502                         } else if (drv->LunID == 0) {
 503                                 return -ENXIO;
 504                         }
 505                 }
 506                 if (!capable(CAP_SYS_ADMIN))
 507                         return -EPERM;
 508         }
 509         drv->usage_count++;
 510         host->usage_count++;
 511         return 0;
 512 }
 513 /*
 514  * Close.  Sync first.
 515  */
 516 static int cciss_release(struct inode *inode, struct file *filep)
 517 {
 518         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
 519         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
 520
 521 #ifdef CCISS_DEBUG
 522         printk(KERN_DEBUG "cciss_release %s\n", inode->i_bdev->bd_disk->disk_name);
 523 #endif /* CCISS_DEBUG */
 524
 525         drv->usage_count--;
 526         host->usage_count--;
 527         return 0;
 528 }
 529
 530 #ifdef CONFIG_COMPAT
 531
 532 static int do_ioctl(struct file *f, unsigned cmd, unsigned long arg)
 533 {
 534         int ret;
 535         lock_kernel();
 536         ret = cciss_ioctl(f->f_dentry->d_inode, f, cmd, arg);
 537         unlock_kernel();
 538         return ret;
 539 }
 540
 541 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd, unsigned long arg);
 542 static int cciss_ioctl32_big_passthru(struct file *f, unsigned cmd, unsigned long arg);
 543
 544 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg)
 545 {
 546         switch (cmd) {
 547         case CCISS_GETPCIINFO:
 548         case CCISS_GETINTINFO:
 549         case CCISS_SETINTINFO:
 550         case CCISS_GETNODENAME:
 551         case CCISS_SETNODENAME:
 552         case CCISS_GETHEARTBEAT:
 553         case CCISS_GETBUSTYPES:
 554         case CCISS_GETFIRMVER:
 555         case CCISS_GETDRIVVER:
 556         case CCISS_REVALIDVOLS:
 557         case CCISS_DEREGDISK:
 558         case CCISS_REGNEWDISK:
 559         case CCISS_REGNEWD:
 560         case CCISS_RESCANDISK:
 561         case CCISS_GETLUNINFO:
 562                 return do_ioctl(f, cmd, arg);
 563
 564         case CCISS_PASSTHRU32:
 565                 return cciss_ioctl32_passthru(f, cmd, arg);
 566         case CCISS_BIG_PASSTHRU32:
 567                 return cciss_ioctl32_big_passthru(f, cmd, arg);
 568
 569         default:
 570                 return -ENOIOCTLCMD;
 571         }
 572 }
 573
 574 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd, unsigned long arg)
 575 {
 576         IOCTL32_Command_struct __user *arg32 =
 577                 (IOCTL32_Command_struct __user *) arg;
 578         IOCTL_Command_struct arg64;
 579         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
 580         int err;
 581         u32 cp;
 582
 583         err = 0;
 584         err |= copy_from_user(&arg64.LUN_info, &arg32->LUN_info, sizeof(arg64.LUN_info));
 585         err |= copy_from_user(&arg64.Request, &arg32->Request, sizeof(arg64.Request));
 586         err |= copy_from_user(&arg64.error_info, &arg32->error_info, sizeof(arg64.error_info));
 587         err |= get_user(arg64.buf_size, &arg32->buf_size);
 588         err |= get_user(cp, &arg32->buf);
 589         arg64.buf = compat_ptr(cp);
 590         err |= copy_to_user(p, &arg64, sizeof(arg64));
 591
 592         if (err)
 593                 return -EFAULT;
 594
 595         err = do_ioctl(f, CCISS_PASSTHRU, (unsigned long) p);
 596         if (err)
 597                 return err;
 598         err |= copy_in_user(&arg32->error_info, &p->error_info, sizeof(arg32->error_info));
 599         if (err)
 600                 return -EFAULT;
 601         return err;
 602 }
 603
 604 static int cciss_ioctl32_big_passthru(struct file *file, unsigned cmd, unsigned long arg)
 605 {
 606         BIG_IOCTL32_Command_struct __user *arg32 =
 607                 (BIG_IOCTL32_Command_struct __user *) arg;
 608         BIG_IOCTL_Command_struct arg64;
 609         BIG_IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
 610         int err;
 611         u32 cp;
 612
 613         err = 0;
 614         err |= copy_from_user(&arg64.LUN_info, &arg32->LUN_info, sizeof(arg64.LUN_info));
 615         err |= copy_from_user(&arg64.Request, &arg32->Request, sizeof(arg64.Request));
 616         err |= copy_from_user(&arg64.error_info, &arg32->error_info, sizeof(arg64.error_info));
 617         err |= get_user(arg64.buf_size, &arg32->buf_size);
 618         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
 619         err |= get_user(cp, &arg32->buf);
 620         arg64.buf = compat_ptr(cp);
 621         err |= copy_to_user(p, &arg64, sizeof(arg64));
 622
 623         if (err)
 624                  return -EFAULT;
 625
 626         err = do_ioctl(file, CCISS_BIG_PASSTHRU, (unsigned long) p);
 627         if (err)
 628                 return err;
 629         err |= copy_in_user(&arg32->error_info, &p->error_info, sizeof(arg32->error_info));
 630         if (err)
 631                 return -EFAULT;
 632         return err;
 633 }
 634 #endif
 635 /*
 636  * ioctl
 637  */
 638 static int cciss_ioctl(struct inode *inode, struct file *filep,
 639                 unsigned int cmd, unsigned long arg)
 640 {
 641         struct block_device *bdev = inode->i_bdev;
 642         struct gendisk *disk = bdev->bd_disk;
 643         ctlr_info_t *host = get_host(disk);
 644         drive_info_struct *drv = get_drv(disk);
 645         int ctlr = host->ctlr;
 646         void __user *argp = (void __user *)arg;
 647
 648 #ifdef CCISS_DEBUG
 649         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
 650 #endif /* CCISS_DEBUG */
 651
 652         switch(cmd) {
 653         case HDIO_GETGEO:
 654         {
 655                 struct hd_geometry driver_geo;
 656                 if (drv->cylinders) {
 657                         driver_geo.heads = drv->heads;
 658                         driver_geo.sectors = drv->sectors;
 659                         driver_geo.cylinders = drv->cylinders;
 660                 } else
 661                         return -ENXIO;
 662                 driver_geo.start= get_start_sect(inode->i_bdev);
 663                 if (copy_to_user(argp, &driver_geo, sizeof(struct hd_geometry)))
 664                         return  -EFAULT;
 665                 return(0);
 666         }
 667
 668         case CCISS_GETPCIINFO:
 669         {
 670                 cciss_pci_info_struct pciinfo;
 671
 672                 if (!arg) return -EINVAL;
 673                 pciinfo.domain = pci_domain_nr(host->pdev->bus);
 674                 pciinfo.bus = host->pdev->bus->number;
 675                 pciinfo.dev_fn = host->pdev->devfn;
 676                 pciinfo.board_id = host->board_id;
 677                 if (copy_to_user(argp, &pciinfo,  sizeof( cciss_pci_info_struct )))
 678                         return  -EFAULT;
 679                 return(0);
 680         }
 681         case CCISS_GETINTINFO:
 682         {
 683                 cciss_coalint_struct intinfo;
 684                 if (!arg) return -EINVAL;
 685                 intinfo.delay = readl(&host->cfgtable->HostWrite.CoalIntDelay);
 686                 intinfo.count = readl(&host->cfgtable->HostWrite.CoalIntCount);
 687                 if (copy_to_user(argp, &intinfo, sizeof( cciss_coalint_struct )))
 688                         return -EFAULT;
 689                 return(0);
 690         }
 691         case CCISS_SETINTINFO:
 692         {
 693                 cciss_coalint_struct intinfo;
 694                 unsigned long flags;
 695                 int i;
 696
 697                 if (!arg) return -EINVAL;
 698                 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
 699                 if (copy_from_user(&intinfo, argp, sizeof( cciss_coalint_struct)))
 700                         return -EFAULT;
 701                 if ( (intinfo.delay == 0 ) && (intinfo.count == 0))
 702
 703                 {
 704 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
 705                         return( -EINVAL);
 706                 }
 707                 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 708                 /* Update the field, and then ring the doorbell */
 709                 writel( intinfo.delay,
 710                         &(host->cfgtable->HostWrite.CoalIntDelay));
 711                 writel( intinfo.count,
 712                         &(host->cfgtable->HostWrite.CoalIntCount));
 713                 writel( CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
 714
 715                 for(i=0;i<MAX_IOCTL_CONFIG_WAIT;i++) {
 716                         if (!(readl(host->vaddr + SA5_DOORBELL)
 717                                         & CFGTBL_ChangeReq))
 718                                 break;
 719                         /* delay and try again */
 720                         udelay(1000);
 721                 }
 722                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 723                 if (i >= MAX_IOCTL_CONFIG_WAIT)
 724                         return -EAGAIN;
 725                 return(0);
 726         }
 727         case CCISS_GETNODENAME:
 728         {
 729                 NodeName_type NodeName;
 730                 int i;
 731
 732                 if (!arg) return -EINVAL;
 733                 for(i=0;i<16;i++)
 734                         NodeName[i] = readb(&host->cfgtable->ServerName[i]);
 735                 if (copy_to_user(argp, NodeName, sizeof( NodeName_type)))
 736                         return  -EFAULT;
 737                 return(0);
 738         }
 739         case CCISS_SETNODENAME:
 740         {
 741                 NodeName_type NodeName;
 742                 unsigned long flags;
 743                 int i;
 744
 745                 if (!arg) return -EINVAL;
 746                 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
 747
 748                 if (copy_from_user(NodeName, argp, sizeof( NodeName_type)))
 749                         return -EFAULT;
 750
 751                 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 752
 753                         /* Update the field, and then ring the doorbell */
 754                 for(i=0;i<16;i++)
 755                         writeb( NodeName[i], &host->cfgtable->ServerName[i]);
 756
 757                 writel( CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
 758
 759                 for(i=0;i<MAX_IOCTL_CONFIG_WAIT;i++) {
 760                         if (!(readl(host->vaddr + SA5_DOORBELL)
 761                                         & CFGTBL_ChangeReq))
 762                                 break;
 763                         /* delay and try again */
 764                         udelay(1000);
 765                 }
 766                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 767                 if (i >= MAX_IOCTL_CONFIG_WAIT)
 768                         return -EAGAIN;
 769                 return(0);
 770         }
 771
 772         case CCISS_GETHEARTBEAT:
 773         {
 774                 Heartbeat_type heartbeat;
 775
 776                 if (!arg) return -EINVAL;
 777                 heartbeat = readl(&host->cfgtable->HeartBeat);
 778                 if (copy_to_user(argp, &heartbeat, sizeof( Heartbeat_type)))
 779                         return -EFAULT;
 780                 return(0);
 781         }
 782         case CCISS_GETBUSTYPES:
 783         {
 784                 BusTypes_type BusTypes;
 785
 786                 if (!arg) return -EINVAL;
 787                 BusTypes = readl(&host->cfgtable->BusTypes);
 788                 if (copy_to_user(argp, &BusTypes, sizeof( BusTypes_type) ))
 789                         return  -EFAULT;
 790                 return(0);
 791         }
 792         case CCISS_GETFIRMVER:
 793         {
 794                 FirmwareVer_type firmware;
 795
 796                 if (!arg) return -EINVAL;
 797                 memcpy(firmware, host->firm_ver, 4);
 798
 799                 if (copy_to_user(argp, firmware, sizeof( FirmwareVer_type)))
 800                         return -EFAULT;
 801                 return(0);
 802         }
 803         case CCISS_GETDRIVVER:
 804         {
 805                 DriverVer_type DriverVer = DRIVER_VERSION;
 806
 807                 if (!arg) return -EINVAL;
 808
 809                 if (copy_to_user(argp, &DriverVer, sizeof( DriverVer_type) ))
 810                         return -EFAULT;
 811                 return(0);
 812         }
 813
 814         case CCISS_REVALIDVOLS:
 815                 if (bdev != bdev->bd_contains || drv != host->drv)
 816                         return -ENXIO;
 817                 return revalidate_allvol(host);
 818
 819         case CCISS_GETLUNINFO: {
 820                 LogvolInfo_struct luninfo;
 821
 822                 luninfo.LunID = drv->LunID;
 823                 luninfo.num_opens = drv->usage_count;
 824                 luninfo.num_parts = 0;
 825                 if (copy_to_user(argp, &luninfo,
 826                                 sizeof(LogvolInfo_struct)))
 827                         return -EFAULT;
 828                 return(0);
 829         }
 830         case CCISS_DEREGDISK:
 831                 return rebuild_lun_table(host, disk);
 832
 833         case CCISS_REGNEWD:
 834                 return rebuild_lun_table(host, NULL);
 835
 836         case CCISS_PASSTHRU:
 837         {
 838                 IOCTL_Command_struct iocommand;
 839                 CommandList_struct *c;
 840                 char    *buff = NULL;
 841                 u64bit  temp64;
 842                 unsigned long flags;
 843                 DECLARE_COMPLETION(wait);
 844
 845                 if (!arg) return -EINVAL;
 846
 847                 if (!capable(CAP_SYS_RAWIO)) return -EPERM;
 848
 849                 if (copy_from_user(&iocommand, argp, sizeof( IOCTL_Command_struct) ))
 850                         return -EFAULT;
 851                 if((iocommand.buf_size < 1) &&
 852                                 (iocommand.Request.Type.Direction != XFER_NONE))
 853                 {
 854                         return -EINVAL;
 855                 }
 856 #if 0 /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
 857                 /* Check kmalloc limits */
 858                 if(iocommand.buf_size > 128000)
 859                         return -EINVAL;
 860 #endif
 861                 if(iocommand.buf_size > 0)
 862                 {
 863                         buff =  kmalloc(iocommand.buf_size, GFP_KERNEL);
 864                         if( buff == NULL)
 865                                 return -EFAULT;
 866                 }
 867                 if (iocommand.Request.Type.Direction == XFER_WRITE)
 868                 {
 869                         /* Copy the data into the buffer we created */
 870                         if (copy_from_user(buff, iocommand.buf, iocommand.buf_size))
 871                         {
 872                                 kfree(buff);
 873                                 return -EFAULT;
 874                         }
 875                 } else {
 876                         memset(buff, 0, iocommand.buf_size);
 877                 }
 878                 if ((c = cmd_alloc(host , 0)) == NULL)
 879                 {
 880                         kfree(buff);
 881                         return -ENOMEM;
 882                 }
 883                         // Fill in the command type
 884                 c->cmd_type = CMD_IOCTL_PEND;
 885                         // Fill in Command Header
 886                 c->Header.ReplyQueue = 0;  // unused in simple mode
 887                 if( iocommand.buf_size > 0)     // buffer to fill
 888                 {
 889                         c->Header.SGList = 1;
 890                         c->Header.SGTotal= 1;
 891                 } else  // no buffers to fill
 892                 {
 893                         c->Header.SGList = 0;
 894                         c->Header.SGTotal= 0;
 895                 }
 896                 c->Header.LUN = iocommand.LUN_info;
 897                 c->Header.Tag.lower = c->busaddr;  // use the kernel address the cmd block for tag
 898
 899                 // Fill in Request block
 900                 c->Request = iocommand.Request;
 901
 902                 // Fill in the scatter gather information
 903                 if (iocommand.buf_size > 0 )
 904                 {
 905                         temp64.val = pci_map_single( host->pdev, buff,
 906                                         iocommand.buf_size,
 907                                 PCI_DMA_BIDIRECTIONAL);
 908                         c->SG[0].Addr.lower = temp64.val32.lower;
 909                         c->SG[0].Addr.upper = temp64.val32.upper;
 910                         c->SG[0].Len = iocommand.buf_size;
 911                         c->SG[0].Ext = 0;  // we are not chaining
 912                 }
 913                 c->waiting = &wait;
 914
 915                 /* Put the request on the tail of the request queue */
 916                 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 917                 addQ(&host->reqQ, c);
 918                 host->Qdepth++;
 919                 start_io(host);
 920                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 921
 922                 wait_for_completion(&wait);
 923
 924                 /* unlock the buffers from DMA */
 925                 temp64.val32.lower = c->SG[0].Addr.lower;
 926                 temp64.val32.upper = c->SG[0].Addr.upper;
 927                 pci_unmap_single( host->pdev, (dma_addr_t) temp64.val,
 928                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
 929
 930                 /* Copy the error information out */
 931                 iocommand.error_info = *(c->err_info);
 932                 if ( copy_to_user(argp, &iocommand, sizeof( IOCTL_Command_struct) ) )
 933                 {
 934                         kfree(buff);
 935                         cmd_free(host, c, 0);
 936                         return( -EFAULT);
 937                 }
 938
 939                 if (iocommand.Request.Type.Direction == XFER_READ)
 940                 {
 941                         /* Copy the data out of the buffer we created */
 942                         if (copy_to_user(iocommand.buf, buff, iocommand.buf_size))
 943                         {
 944                                 kfree(buff);
 945                                 cmd_free(host, c, 0);
 946                                 return -EFAULT;
 947                         }
 948                 }
 949                 kfree(buff);
 950                 cmd_free(host, c, 0);
 951                 return(0);
 952         }
 953         case CCISS_BIG_PASSTHRU: {
 954                 BIG_IOCTL_Command_struct *ioc;
 955                 CommandList_struct *c;
 956                 unsigned char **buff = NULL;
 957                 int     *buff_size = NULL;
 958                 u64bit  temp64;
 959                 unsigned long flags;
 960                 BYTE sg_used = 0;
 961                 int status = 0;
 962                 int i;
 963                 DECLARE_COMPLETION(wait);
 964                 __u32   left;
 965                 __u32   sz;
 966                 BYTE    __user *data_ptr;
 967
 968                 if (!arg)
 969                         return -EINVAL;
 970                 if (!capable(CAP_SYS_RAWIO))
 971                         return -EPERM;
 972                 ioc = (BIG_IOCTL_Command_struct *)
 973                         kmalloc(sizeof(*ioc), GFP_KERNEL);
 974                 if (!ioc) {
 975                         status = -ENOMEM;
 976                         goto cleanup1;
 977                 }
 978                 if (copy_from_user(ioc, argp, sizeof(*ioc))) {
 979                         status = -EFAULT;
 980                         goto cleanup1;
 981                 }
 982                 if ((ioc->buf_size < 1) &&
 983                         (ioc->Request.Type.Direction != XFER_NONE)) {
 984                                 status = -EINVAL;
 985                                 goto cleanup1;
 986                 }
 987                 /* Check kmalloc limits  using all SGs */
 988                 if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
 989                         status = -EINVAL;
 990                         goto cleanup1;
 991                 }
 992                 if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
 993                         status = -EINVAL;
 994                         goto cleanup1;
 995                 }
 996                 buff = (unsigned char **) kmalloc(MAXSGENTRIES *
 997                                 sizeof(char *), GFP_KERNEL);
 998                 if (!buff) {
 999                         status = -ENOMEM;
1000                         goto cleanup1;
1001                 }
1002                 memset(buff, 0, MAXSGENTRIES);
1003                 buff_size = (int *) kmalloc(MAXSGENTRIES * sizeof(int),
1004                                         GFP_KERNEL);
1005                 if (!buff_size) {
1006                         status = -ENOMEM;
1007                         goto cleanup1;
1008                 }
1009                 left = ioc->buf_size;
1010                 data_ptr = ioc->buf;
1011                 while (left) {
1012                         sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1013                         buff_size[sg_used] = sz;
1014                         buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1015                         if (buff[sg_used] == NULL) {
1016                                 status = -ENOMEM;
1017                                 goto cleanup1;
1018                         }
1019                         if (ioc->Request.Type.Direction == XFER_WRITE &&
1020                                 copy_from_user(buff[sg_used], data_ptr, sz)) {
1021                                         status = -ENOMEM;
1022                                         goto cleanup1;
1023                         } else {
1024                                 memset(buff[sg_used], 0, sz);
1025                         }
1026                         left -= sz;
1027                         data_ptr += sz;
1028                         sg_used++;
1029                 }
1030                 if ((c = cmd_alloc(host , 0)) == NULL) {
1031                         status = -ENOMEM;
1032                         goto cleanup1;
1033                 }
1034                 c->cmd_type = CMD_IOCTL_PEND;
1035                 c->Header.ReplyQueue = 0;
1036
1037                 if( ioc->buf_size > 0) {
1038                         c->Header.SGList = sg_used;
1039                         c->Header.SGTotal= sg_used;
1040                 } else {
1041                         c->Header.SGList = 0;
1042                         c->Header.SGTotal= 0;
1043                 }
1044                 c->Header.LUN = ioc->LUN_info;
1045                 c->Header.Tag.lower = c->busaddr;
1046
1047                 c->Request = ioc->Request;
1048                 if (ioc->buf_size > 0 ) {
1049                         int i;
1050                         for(i=0; i<sg_used; i++) {
1051                                 temp64.val = pci_map_single( host->pdev, buff[i],
1052                                         buff_size[i],
1053                                         PCI_DMA_BIDIRECTIONAL);
1054                                 c->SG[i].Addr.lower = temp64.val32.lower;
1055                                 c->SG[i].Addr.upper = temp64.val32.upper;
1056                                 c->SG[i].Len = buff_size[i];
1057                                 c->SG[i].Ext = 0;  /* we are not chaining */
1058                         }
1059                 }
1060                 c->waiting = &wait;
1061                 /* Put the request on the tail of the request queue */
1062                 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1063                 addQ(&host->reqQ, c);
1064                 host->Qdepth++;
1065                 start_io(host);
1066                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1067                 wait_for_completion(&wait);
1068                 /* unlock the buffers from DMA */
1069                 for(i=0; i<sg_used; i++) {
1070                         temp64.val32.lower = c->SG[i].Addr.lower;
1071                         temp64.val32.upper = c->SG[i].Addr.upper;
1072                         pci_unmap_single( host->pdev, (dma_addr_t) temp64.val,
1073                                 buff_size[i], PCI_DMA_BIDIRECTIONAL);
1074                 }
1075                 /* Copy the error information out */
1076                 ioc->error_info = *(c->err_info);
1077                 if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1078                         cmd_free(host, c, 0);
1079                         status = -EFAULT;
1080                         goto cleanup1;
1081                 }
1082                 if (ioc->Request.Type.Direction == XFER_READ) {
1083                         /* Copy the data out of the buffer we created */
1084                         BYTE __user *ptr = ioc->buf;
1085                         for(i=0; i< sg_used; i++) {
1086                                 if (copy_to_user(ptr, buff[i], buff_size[i])) {
1087                                         cmd_free(host, c, 0);
1088                                         status = -EFAULT;
1089                                         goto cleanup1;
1090                                 }
1091                                 ptr += buff_size[i];
1092                         }
1093                 }
1094                 cmd_free(host, c, 0);
1095                 status = 0;
1096 cleanup1:
1097                 if (buff) {
1098                         for(i=0; i<sg_used; i++)
1099                                 if(buff[i] != NULL)
1100                                         kfree(buff[i]);
1101                         kfree(buff);
1102                 }
1103                 if (buff_size)
1104                         kfree(buff_size);
1105                 if (ioc)
1106                         kfree(ioc);
1107                 return(status);
1108         }
1109         default:
1110                 return -ENOTTY;
1111         }
1112
1113 }
1114
1115 /*
1116  * revalidate_allvol is for online array config utilities.  After a
1117  * utility reconfigures the drives in the array, it can use this function
1118  * (through an ioctl) to make the driver zap any previous disk structs for
1119  * that controller and get new ones.
1120  *
1121  * Right now I'm using the getgeometry() function to do this, but this
1122  * function should probably be finer grained and allow you to revalidate one
1123  * particualar logical volume (instead of all of them on a particular
1124  * controller).
1125  */
1126 static int revalidate_allvol(ctlr_info_t *host)
1127 {
1128         int ctlr = host->ctlr, i;
1129         unsigned long flags;
1130
1131         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1132         if (host->usage_count > 1) {
1133                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1134                 printk(KERN_WARNING "cciss: Device busy for volume"
1135                         " revalidation (usage=%d)\n", host->usage_count);
1136                 return -EBUSY;
1137         }
1138         host->usage_count++;
1139         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1140
1141         for(i=0; i< NWD; i++) {
1142                 struct gendisk *disk = host->gendisk[i];
1143                 if (disk->flags & GENHD_FL_UP)
1144                         del_gendisk(disk);
1145         }
1146
1147         /*
1148          * Set the partition and block size structures for all volumes
1149          * on this controller to zero.  We will reread all of this data
1150          */
1151         memset(host->drv,        0, sizeof(drive_info_struct)
1152                                                 * CISS_MAX_LUN);
1153         /*
1154          * Tell the array controller not to give us any interrupts while
1155          * we check the new geometry.  Then turn interrupts back on when
1156          * we're done.
1157          */
1158         host->access.set_intr_mask(host, CCISS_INTR_OFF);
1159         cciss_getgeometry(ctlr);
1160         host->access.set_intr_mask(host, CCISS_INTR_ON);
1161
1162         /* Loop through each real device */
1163         for (i = 0; i < NWD; i++) {
1164                 struct gendisk *disk = host->gendisk[i];
1165                 drive_info_struct *drv = &(host->drv[i]);
1166                 /* we must register the controller even if no disks exist */
1167                 /* this is for the online array utilities */
1168                 if (!drv->heads && i)
1169                         continue;
1170                 blk_queue_hardsect_size(drv->queue, drv->block_size);
1171                 set_capacity(disk, drv->nr_blocks);
1172                 add_disk(disk);
1173         }
1174         host->usage_count--;
1175         return 0;
1176 }
1177
1178 /* This function will check the usage_count of the drive to be updated/added.
1179  * If the usage_count is zero then the drive information will be updated and
1180  * the disk will be re-registered with the kernel.  If not then it will be
1181  * left alone for the next reboot.  The exception to this is disk 0 which
1182  * will always be left registered with the kernel since it is also the
1183  * controller node.  Any changes to disk 0 will show up on the next
1184  * reboot.
1185 */
1186 static void cciss_update_drive_info(int ctlr, int drv_index)
1187   {
1188         ctlr_info_t *h = hba[ctlr];
1189         struct gendisk *disk;
1190         ReadCapdata_struct *size_buff = NULL;
1191         InquiryData_struct *inq_buff = NULL;
1192         unsigned int block_size;
1193         unsigned int total_size;
1194         unsigned long flags = 0;
1195         int ret = 0;
1196
1197         /* if the disk already exists then deregister it before proceeding*/
1198         if (h->drv[drv_index].raid_level != -1){
1199                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1200                 h->drv[drv_index].busy_configuring = 1;
1201                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1202                 ret = deregister_disk(h->gendisk[drv_index],
1203                         &h->drv[drv_index], 0);
1204                 h->drv[drv_index].busy_configuring = 0;
1205         }
1206
1207         /* If the disk is in use return */
1208         if (ret)
1209                 return;
1210
1211
1212         /* Get information about the disk and modify the driver sturcture */
1213         size_buff = kmalloc(sizeof( ReadCapdata_struct), GFP_KERNEL);
1214         if (size_buff == NULL)
1215                 goto mem_msg;
1216         inq_buff = kmalloc(sizeof( InquiryData_struct), GFP_KERNEL);
1217         if (inq_buff == NULL)
1218                 goto mem_msg;
1219
1220         cciss_read_capacity(ctlr, drv_index, size_buff, 1,
1221                 &total_size, &block_size);
1222         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1223                 inq_buff, &h->drv[drv_index]);
1224
1225         ++h->num_luns;
1226         disk = h->gendisk[drv_index];
1227         set_capacity(disk, h->drv[drv_index].nr_blocks);
1228
1229
1230         /* if it's the controller it's already added */
1231         if (drv_index){
1232                 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1233
1234                 /* Set up queue information */
1235                 disk->queue->backing_dev_info.ra_pages = READ_AHEAD;
1236                 blk_queue_bounce_limit(disk->queue, hba[ctlr]->pdev->dma_mask);
1237
1238                 /* This is a hardware imposed limit. */
1239                 blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1240
1241                 /* This is a limit in the driver and could be eliminated. */
1242                 blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1243
1244                 blk_queue_max_sectors(disk->queue, 512);
1245
1246                 disk->queue->queuedata = hba[ctlr];
1247
1248                 blk_queue_hardsect_size(disk->queue,
1249                         hba[ctlr]->drv[drv_index].block_size);
1250
1251                 h->drv[drv_index].queue = disk->queue;
1252                 add_disk(disk);
1253         }
1254
1255 freeret:
1256         kfree(size_buff);
1257         kfree(inq_buff);
1258         return;
1259 mem_msg:
1260         printk(KERN_ERR "cciss: out of memory\n");
1261         goto freeret;
1262 }
1263
1264 /* This function will find the first index of the controllers drive array
1265  * that has a -1 for the raid_level and will return that index.  This is
1266  * where new drives will be added.  If the index to be returned is greater
1267  * than the highest_lun index for the controller then highest_lun is set
1268  * to this new index.  If there are no available indexes then -1 is returned.
1269 */
1270 static int cciss_find_free_drive_index(int ctlr)
1271 {
1272         int i;
1273
1274         for (i=0; i < CISS_MAX_LUN; i++){
1275                 if (hba[ctlr]->drv[i].raid_level == -1){
1276                         if (i > hba[ctlr]->highest_lun)
1277                                 hba[ctlr]->highest_lun = i;
1278                         return i;
1279                 }
1280         }
1281         return -1;
1282 }
1283
1284 /* This function will add and remove logical drives from the Logical
1285  * drive array of the controller and maintain persistancy of ordering
1286  * so that mount points are preserved until the next reboot.  This allows
1287  * for the removal of logical drives in the middle of the drive array
1288  * without a re-ordering of those drives.
1289  * INPUT
1290  * h            = The controller to perform the operations on
1291  * del_disk     = The disk to remove if specified.  If the value given
1292  *                is NULL then no disk is removed.
1293 */
1294 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk)
1295 {
1296         int ctlr = h->ctlr;
1297         int num_luns;
1298         ReportLunData_struct *ld_buff = NULL;
1299         drive_info_struct *drv = NULL;
1300         int return_code;
1301         int listlength = 0;
1302         int i;
1303         int drv_found;
1304         int drv_index = 0;
1305         __u32 lunid = 0;
1306         unsigned long flags;
1307
1308         /* Set busy_configuring flag for this operation */
1309         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1310         if (h->num_luns >= CISS_MAX_LUN){
1311                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1312                 return -EINVAL;
1313         }
1314
1315         if (h->busy_configuring){
1316                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1317                 return -EBUSY;
1318         }
1319         h->busy_configuring = 1;
1320
1321         /* if del_disk is NULL then we are being called to add a new disk
1322          * and update the logical drive table.  If it is not NULL then
1323          * we will check if the disk is in use or not.
1324          */
1325         if (del_disk != NULL){
1326                 drv = get_drv(del_disk);
1327                 drv->busy_configuring = 1;
1328                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1329                 return_code = deregister_disk(del_disk, drv, 1);
1330                 drv->busy_configuring = 0;
1331                 h->busy_configuring = 0;
1332                 return return_code;
1333         } else {
1334                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1335                 if (!capable(CAP_SYS_RAWIO))
1336                         return -EPERM;
1337
1338                 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1339                 if (ld_buff == NULL)
1340                         goto mem_msg;
1341
1342                 return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
1343                                 sizeof(ReportLunData_struct), 0, 0, 0,
1344                                 TYPE_CMD);
1345
1346                 if (return_code == IO_OK){
1347                         listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[0])) << 24;
1348                         listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[1])) << 16;
1349                         listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[2])) << 8;
1350                         listlength |= 0xff & (unsigned int)(ld_buff->LUNListLength[3]);
1351                 } else{ /* reading number of logical volumes failed */
1352                         printk(KERN_WARNING "cciss: report logical volume"
1353                                 " command failed\n");
1354                         listlength = 0;
1355                         goto freeret;
1356                 }
1357
1358                 num_luns = listlength / 8;      /* 8 bytes per entry */
1359                 if (num_luns > CISS_MAX_LUN){
1360                         num_luns = CISS_MAX_LUN;
1361                         printk(KERN_WARNING "cciss: more luns configured"
1362                                 " on controller than can be handled by"
1363                                 " this driver.\n");
1364                 }
1365
1366                 /* Compare controller drive array to drivers drive array.
1367                 * Check for updates in the drive information and any new drives
1368                 * on the controller.
1369                 */
1370                 for (i=0; i < num_luns; i++){
1371                         int j;
1372
1373                         drv_found = 0;
1374
1375                         lunid = (0xff &
1376                                 (unsigned int)(ld_buff->LUN[i][3])) << 24;
1377                         lunid |= (0xff &
1378                                 (unsigned int)(ld_buff->LUN[i][2])) << 16;
1379                         lunid |= (0xff &
1380                                 (unsigned int)(ld_buff->LUN[i][1])) << 8;
1381                         lunid |= 0xff &
1382                                 (unsigned int)(ld_buff->LUN[i][0]);
1383
1384                         /* Find if the LUN is already in the drive array
1385                          * of the controller.  If so then update its info
1386                          * if not is use.  If it does not exist then find
1387                          * the first free index and add it.
1388                         */
1389                         for (j=0; j <= h->highest_lun; j++){
1390                                 if (h->drv[j].LunID == lunid){
1391                                         drv_index = j;
1392                                         drv_found = 1;
1393                                 }
1394                         }
1395
1396                         /* check if the drive was found already in the array */
1397                         if (!drv_found){
1398                                 drv_index = cciss_find_free_drive_index(ctlr);
1399                                 if (drv_index == -1)
1400                                         goto freeret;
1401
1402                         }
1403                         h->drv[drv_index].LunID = lunid;
1404                         cciss_update_drive_info(ctlr, drv_index);
1405                 } /* end for */
1406         } /* end else */
1407
1408 freeret:
1409         kfree(ld_buff);
1410         h->busy_configuring = 0;
1411         /* We return -1 here to tell the ACU that we have registered/updated
1412          * all of the drives that we can and to keep it from calling us
1413          * additional times.
1414         */
1415         return -1;
1416 mem_msg:
1417         printk(KERN_ERR "cciss: out of memory\n");
1418         goto freeret;
1419 }
1420
1421 /* This function will deregister the disk and it's queue from the
1422  * kernel.  It must be called with the controller lock held and the
1423  * drv structures busy_configuring flag set.  It's parameters are:
1424  *
1425  * disk = This is the disk to be deregistered
1426  * drv  = This is the drive_info_struct associated with the disk to be
1427  *        deregistered.  It contains information about the disk used
1428  *        by the driver.
1429  * clear_all = This flag determines whether or not the disk information
1430  *             is going to be completely cleared out and the highest_lun
1431  *             reset.  Sometimes we want to clear out information about
1432  *             the disk in preperation for re-adding it.  In this case
1433  *             the highest_lun should be left unchanged and the LunID
1434  *             should not be cleared.
1435 */
1436 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
1437                            int clear_all)
1438 {
1439         ctlr_info_t *h = get_host(disk);
1440
1441         if (!capable(CAP_SYS_RAWIO))
1442                 return -EPERM;
1443
1444         /* make sure logical volume is NOT is use */
1445         if(clear_all || (h->gendisk[0] == disk)) {
1446         if (drv->usage_count > 1)
1447                 return -EBUSY;
1448         }
1449         else
1450                 if( drv->usage_count > 0 )
1451                         return -EBUSY;
1452
1453         /* invalidate the devices and deregister the disk.  If it is disk
1454          * zero do not deregister it but just zero out it's values.  This
1455          * allows us to delete disk zero but keep the controller registered.
1456         */
1457         if (h->gendisk[0] != disk){
1458                 if (disk->flags & GENHD_FL_UP){
1459                         blk_cleanup_queue(disk->queue);
1460                 del_gendisk(disk);
1461                         drv->queue = NULL;
1462                 }
1463         }
1464
1465         --h->num_luns;
1466         /* zero out the disk size info */
1467         drv->nr_blocks = 0;
1468         drv->block_size = 0;
1469         drv->heads = 0;
1470         drv->sectors = 0;
1471         drv->cylinders = 0;
1472         drv->raid_level = -1;   /* This can be used as a flag variable to
1473                                  * indicate that this element of the drive
1474                                  * array is free.
1475                                 */
1476
1477         if (clear_all){
1478         /* check to see if it was the last disk */
1479         if (drv == h->drv + h->highest_lun) {
1480                 /* if so, find the new hightest lun */
1481                 int i, newhighest =-1;
1482                 for(i=0; i<h->highest_lun; i++) {
1483                         /* if the disk has size > 0, it is available */
1484                                 if (h->drv[i].heads)
1485                                 newhighest = i;
1486                 }
1487                 h->highest_lun = newhighest;
1488         }
1489
1490         drv->LunID = 0;
1491         }
1492         return(0);
1493 }
1494
1495 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
1496         size_t size,
1497         unsigned int use_unit_num, /* 0: address the controller,
1498                                       1: address logical volume log_unit,
1499                                       2: periph device address is scsi3addr */
1500         unsigned int log_unit, __u8 page_code, unsigned char *scsi3addr,
1501         int cmd_type)
1502 {
1503         ctlr_info_t *h= hba[ctlr];
1504         u64bit buff_dma_handle;
1505         int status = IO_OK;
1506
1507         c->cmd_type = CMD_IOCTL_PEND;
1508         c->Header.ReplyQueue = 0;
1509         if( buff != NULL) {
1510                 c->Header.SGList = 1;
1511                 c->Header.SGTotal= 1;
1512         } else {
1513                 c->Header.SGList = 0;
1514                 c->Header.SGTotal= 0;
1515         }
1516         c->Header.Tag.lower = c->busaddr;
1517
1518         c->Request.Type.Type = cmd_type;
1519         if (cmd_type == TYPE_CMD) {
1520                 switch(cmd) {
1521                 case  CISS_INQUIRY:
1522                         /* If the logical unit number is 0 then, this is going
1523                         to controller so It's a physical command
1524                         mode = 0 target = 0.  So we have nothing to write.
1525                         otherwise, if use_unit_num == 1,
1526                         mode = 1(volume set addressing) target = LUNID
1527                         otherwise, if use_unit_num == 2,
1528                         mode = 0(periph dev addr) target = scsi3addr */
1529                         if (use_unit_num == 1) {
1530                                 c->Header.LUN.LogDev.VolId=
1531                                         h->drv[log_unit].LunID;
1532                                 c->Header.LUN.LogDev.Mode = 1;
1533                         } else if (use_unit_num == 2) {
1534                                 memcpy(c->Header.LUN.LunAddrBytes,scsi3addr,8);
1535                                 c->Header.LUN.LogDev.Mode = 0;
1536                         }
1537                         /* are we trying to read a vital product page */
1538                         if(page_code != 0) {
1539                                 c->Request.CDB[1] = 0x01;
1540                                 c->Request.CDB[2] = page_code;
1541                         }
1542                         c->Request.CDBLen = 6;
1543                         c->Request.Type.Attribute = ATTR_SIMPLE;
1544                         c->Request.Type.Direction = XFER_READ;
1545                         c->Request.Timeout = 0;
1546                         c->Request.CDB[0] =  CISS_INQUIRY;
1547                         c->Request.CDB[4] = size  & 0xFF;
1548                 break;
1549                 case CISS_REPORT_LOG:
1550                 case CISS_REPORT_PHYS:
1551                         /* Talking to controller so It's a physical command
1552                            mode = 00 target = 0.  Nothing to write.
1553                         */
1554                         c->Request.CDBLen = 12;
1555                         c->Request.Type.Attribute = ATTR_SIMPLE;
1556                         c->Request.Type.Direction = XFER_READ;
1557                         c->Request.Timeout = 0;
1558                         c->Request.CDB[0] = cmd;
1559                         c->Request.CDB[6] = (size >> 24) & 0xFF;  //MSB
1560                         c->Request.CDB[7] = (size >> 16) & 0xFF;
1561                         c->Request.CDB[8] = (size >> 8) & 0xFF;
1562                         c->Request.CDB[9] = size & 0xFF;
1563                         break;
1564
1565                 case CCISS_READ_CAPACITY:
1566                         c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1567                         c->Header.LUN.LogDev.Mode = 1;
1568                         c->Request.CDBLen = 10;
1569                         c->Request.Type.Attribute = ATTR_SIMPLE;
1570                         c->Request.Type.Direction = XFER_READ;
1571                         c->Request.Timeout = 0;
1572                         c->Request.CDB[0] = cmd;
1573                 break;
1574                 case CCISS_CACHE_FLUSH:
1575                         c->Request.CDBLen = 12;
1576                         c->Request.Type.Attribute = ATTR_SIMPLE;
1577                         c->Request.Type.Direction = XFER_WRITE;
1578                         c->Request.Timeout = 0;
1579                         c->Request.CDB[0] = BMIC_WRITE;
1580                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
1581                 break;
1582                 default:
1583                         printk(KERN_WARNING
1584                                 "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
1585                         return(IO_ERROR);
1586                 }
1587         } else if (cmd_type == TYPE_MSG) {
1588                 switch (cmd) {
1589                 case 3: /* No-Op message */
1590                         c->Request.CDBLen = 1;
1591                         c->Request.Type.Attribute = ATTR_SIMPLE;
1592                         c->Request.Type.Direction = XFER_WRITE;
1593                         c->Request.Timeout = 0;
1594                         c->Request.CDB[0] = cmd;
1595                         break;
1596                 default:
1597                         printk(KERN_WARNING
1598                                 "cciss%d: unknown message type %d\n",
1599                                 ctlr, cmd);
1600                         return IO_ERROR;
1601                 }
1602         } else {
1603                 printk(KERN_WARNING
1604                         "cciss%d: unknown command type %d\n", ctlr, cmd_type);
1605                 return IO_ERROR;
1606         }
1607         /* Fill in the scatter gather information */
1608         if (size > 0) {
1609                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
1610                         buff, size, PCI_DMA_BIDIRECTIONAL);
1611                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
1612                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
1613                 c->SG[0].Len = size;
1614                 c->SG[0].Ext = 0;  /* we are not chaining */
1615         }
1616         return status;
1617 }
1618 static int sendcmd_withirq(__u8 cmd,
1619         int     ctlr,
1620         void    *buff,
1621         size_t  size,
1622         unsigned int use_unit_num,
1623         unsigned int log_unit,
1624         __u8    page_code,
1625         int cmd_type)
1626 {
1627         ctlr_info_t *h = hba[ctlr];
1628         CommandList_struct *c;
1629         u64bit  buff_dma_handle;
1630         unsigned long flags;
1631         int return_status;
1632         DECLARE_COMPLETION(wait);
1633
1634         if ((c = cmd_alloc(h , 0)) == NULL)
1635                 return -ENOMEM;
1636         return_status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
1637                 log_unit, page_code, NULL, cmd_type);
1638         if (return_status != IO_OK) {
1639                 cmd_free(h, c, 0);
1640                 return return_status;
1641         }
1642 resend_cmd2:
1643         c->waiting = &wait;
1644
1645         /* Put the request on the tail of the queue and send it */
1646         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1647         addQ(&h->reqQ, c);
1648         h->Qdepth++;
1649         start_io(h);
1650         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1651
1652         wait_for_completion(&wait);
1653
1654         if(c->err_info->CommandStatus != 0)
1655         { /* an error has occurred */
1656                 switch(c->err_info->CommandStatus)
1657                 {
1658                         case CMD_TARGET_STATUS:
1659                                 printk(KERN_WARNING "cciss: cmd %p has "
1660                                         " completed with errors\n", c);
1661                                 if( c->err_info->ScsiStatus)
1662                                 {
1663                                         printk(KERN_WARNING "cciss: cmd %p "
1664                                         "has SCSI Status = %x\n",
1665                                                 c,
1666                                                 c->err_info->ScsiStatus);
1667                                 }
1668
1669                         break;
1670                         case CMD_DATA_UNDERRUN:
1671                         case CMD_DATA_OVERRUN:
1672                         /* expected for inquire and report lun commands */
1673                         break;
1674                         case CMD_INVALID:
1675                                 printk(KERN_WARNING "cciss: Cmd %p is "
1676                                         "reported invalid\n", c);
1677                                 return_status = IO_ERROR;
1678                         break;
1679                         case CMD_PROTOCOL_ERR:
1680                                 printk(KERN_WARNING "cciss: cmd %p has "
1681                                         "protocol error \n", c);
1682                                 return_status = IO_ERROR;
1683                         break;
1684 case CMD_HARDWARE_ERR:
1685                                 printk(KERN_WARNING "cciss: cmd %p had "
1686                                         " hardware error\n", c);
1687                                 return_status = IO_ERROR;
1688                         break;
1689                         case CMD_CONNECTION_LOST:
1690                                 printk(KERN_WARNING "cciss: cmd %p had "
1691                                         "connection lost\n", c);
1692                                 return_status = IO_ERROR;
1693                         break;
1694                         case CMD_ABORTED:
1695                                 printk(KERN_WARNING "cciss: cmd %p was "
1696                                         "aborted\n", c);
1697                                 return_status = IO_ERROR;
1698                         break;
1699                         case CMD_ABORT_FAILED:
1700                                 printk(KERN_WARNING "cciss: cmd %p reports "
1701                                         "abort failed\n", c);
1702                                 return_status = IO_ERROR;
1703                         break;
1704                         case CMD_UNSOLICITED_ABORT:
1705                                 printk(KERN_WARNING
1706                                         "cciss%d: unsolicited abort %p\n",
1707                                         ctlr, c);
1708                                 if (c->retry_count < MAX_CMD_RETRIES) {
1709                                         printk(KERN_WARNING
1710                                                 "cciss%d: retrying %p\n",
1711                                                 ctlr, c);
1712                                         c->retry_count++;
1713                                         /* erase the old error information */
1714                                         memset(c->err_info, 0,
1715                                                 sizeof(ErrorInfo_struct));
1716                                         return_status = IO_OK;
1717                                         INIT_COMPLETION(wait);
1718                                         goto resend_cmd2;
1719                                 }
1720                                 return_status = IO_ERROR;
1721                         break;
1722                         default:
1723                                 printk(KERN_WARNING "cciss: cmd %p returned "
1724                                         "unknown status %x\n", c,
1725                                                 c->err_info->CommandStatus);
1726                                 return_status = IO_ERROR;
1727                 }
1728         }
1729         /* unlock the buffers from DMA */
1730         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
1731         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
1732         pci_unmap_single( h->pdev, (dma_addr_t) buff_dma_handle.val,
1733                         c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
1734         cmd_free(h, c, 0);
1735         return(return_status);
1736
1737 }
1738 static void cciss_geometry_inquiry(int ctlr, int logvol,
1739                         int withirq, unsigned int total_size,
1740                         unsigned int block_size, InquiryData_struct *inq_buff,
1741                         drive_info_struct *drv)
1742 {
1743         int return_code;
1744         memset(inq_buff, 0, sizeof(InquiryData_struct));
1745         if (withirq)
1746                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
1747                         inq_buff, sizeof(*inq_buff), 1, logvol ,0xC1, TYPE_CMD);
1748         else
1749                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
1750                         sizeof(*inq_buff), 1, logvol ,0xC1, NULL, TYPE_CMD);
1751         if (return_code == IO_OK) {
1752                 if(inq_buff->data_byte[8] == 0xFF) {
1753                         printk(KERN_WARNING
1754                                 "cciss: reading geometry failed, volume "
1755                                 "does not support reading geometry\n");
1756                         drv->block_size = block_size;
1757                         drv->nr_blocks = total_size;
1758                         drv->heads = 255;
1759                         drv->sectors = 32; // Sectors per track
1760                         drv->cylinders = total_size / 255 / 32;
1761                 } else {
1762                         unsigned int t;
1763
1764                         drv->block_size = block_size;
1765                         drv->nr_blocks = total_size;
1766                         drv->heads = inq_buff->data_byte[6];
1767                         drv->sectors = inq_buff->data_byte[7];
1768                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
1769                         drv->cylinders += inq_buff->data_byte[5];
1770                         drv->raid_level = inq_buff->data_byte[8];
1771                         t = drv->heads * drv->sectors;
1772                         if (t > 1) {
1773                                 drv->cylinders = total_size/t;
1774                         }
1775                 }
1776         } else { /* Get geometry failed */
1777                 printk(KERN_WARNING "cciss: reading geometry failed\n");
1778         }
1779         printk(KERN_INFO "      heads= %d, sectors= %d, cylinders= %d\n\n",
1780                 drv->heads, drv->sectors, drv->cylinders);
1781 }
1782 static void
1783 cciss_read_capacity(int ctlr, int logvol, ReadCapdata_struct *buf,
1784                 int withirq, unsigned int *total_size, unsigned int *block_size)
1785 {
1786         int return_code;
1787         memset(buf, 0, sizeof(*buf));
1788         if (withirq)
1789                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
1790                         ctlr, buf, sizeof(*buf), 1, logvol, 0, TYPE_CMD);
1791         else
1792                 return_code = sendcmd(CCISS_READ_CAPACITY,
1793                         ctlr, buf, sizeof(*buf), 1, logvol, 0, NULL, TYPE_CMD);
1794         if (return_code == IO_OK) {
1795                 *total_size = be32_to_cpu(*((__be32 *) &buf->total_size[0]))+1;
1796                 *block_size = be32_to_cpu(*((__be32 *) &buf->block_size[0]));
1797         } else { /* read capacity command failed */
1798                 printk(KERN_WARNING "cciss: read capacity failed\n");
1799                 *total_size = 0;
1800                 *block_size = BLOCK_SIZE;
1801         }
1802         printk(KERN_INFO "      blocks= %u block_size= %d\n",
1803                 *total_size, *block_size);
1804         return;
1805 }
1806
1807 static int cciss_revalidate(struct gendisk *disk)
1808 {
1809         ctlr_info_t *h = get_host(disk);
1810         drive_info_struct *drv = get_drv(disk);
1811         int logvol;
1812         int FOUND=0;
1813         unsigned int block_size;
1814         unsigned int total_size;
1815         ReadCapdata_struct *size_buff = NULL;
1816         InquiryData_struct *inq_buff = NULL;
1817
1818         for(logvol=0; logvol < CISS_MAX_LUN; logvol++)
1819         {
1820                 if(h->drv[logvol].LunID == drv->LunID) {
1821                         FOUND=1;
1822                         break;
1823                 }
1824         }
1825
1826         if (!FOUND) return 1;
1827
1828         size_buff = kmalloc(sizeof( ReadCapdata_struct), GFP_KERNEL);
1829         if (size_buff == NULL)
1830         {
1831                 printk(KERN_WARNING "cciss: out of memory\n");
1832                 return 1;
1833         }
1834         inq_buff = kmalloc(sizeof( InquiryData_struct), GFP_KERNEL);
1835         if (inq_buff == NULL)
1836         {
1837                 printk(KERN_WARNING "cciss: out of memory\n");
1838                 kfree(size_buff);
1839                 return 1;
1840         }
1841
1842         cciss_read_capacity(h->ctlr, logvol, size_buff, 1, &total_size, &block_size);
1843         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size, inq_buff, drv);
1844
1845         blk_queue_hardsect_size(drv->queue, drv->block_size);
1846         set_capacity(disk, drv->nr_blocks);
1847
1848         kfree(size_buff);
1849         kfree(inq_buff);
1850         return 0;
1851 }
1852
1853 /*
1854  *   Wait polling for a command to complete.
1855  *   The memory mapped FIFO is polled for the completion.
1856  *   Used only at init time, interrupts from the HBA are disabled.
1857  */
1858 static unsigned long pollcomplete(int ctlr)
1859 {
1860         unsigned long done;
1861         int i;
1862
1863         /* Wait (up to 20 seconds) for a command to complete */
1864
1865         for (i = 20 * HZ; i > 0; i--) {
1866                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
1867                 if (done == FIFO_EMPTY)
1868                         schedule_timeout_uninterruptible(1);
1869                 else
1870                         return (done);
1871         }
1872         /* Invalid address to tell caller we ran out of time */
1873         return 1;
1874 }
1875 /*
1876  * Send a command to the controller, and wait for it to complete.
1877  * Only used at init time.
1878  */
1879 static int sendcmd(
1880         __u8    cmd,
1881         int     ctlr,
1882         void    *buff,
1883         size_t  size,
1884         unsigned int use_unit_num, /* 0: address the controller,
1885                                       1: address logical volume log_unit,
1886                                       2: periph device address is scsi3addr */
1887         unsigned int log_unit,
1888         __u8    page_code,
1889         unsigned char *scsi3addr,
1890         int cmd_type)
1891 {
1892         CommandList_struct *c;
1893         int i;
1894         unsigned long complete;
1895         ctlr_info_t *info_p= hba[ctlr];
1896         u64bit buff_dma_handle;
1897         int status;
1898
1899         if ((c = cmd_alloc(info_p, 1)) == NULL) {
1900                 printk(KERN_WARNING "cciss: unable to get memory");
1901                 return(IO_ERROR);
1902         }
1903         status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
1904                 log_unit, page_code, scsi3addr, cmd_type);
1905         if (status != IO_OK) {
1906                 cmd_free(info_p, c, 1);
1907                 return status;
1908         }
1909 resend_cmd1:
1910         /*
1911          * Disable interrupt
1912          */
1913 #ifdef CCISS_DEBUG
1914         printk(KERN_DEBUG "cciss: turning intr off\n");
1915 #endif /* CCISS_DEBUG */
1916         info_p->access.set_intr_mask(info_p, CCISS_INTR_OFF);
1917
1918         /* Make sure there is room in the command FIFO */
1919         /* Actually it should be completely empty at this time. */
1920         for (i = 200000; i > 0; i--)
1921         {
1922                 /* if fifo isn't full go */
1923                 if (!(info_p->access.fifo_full(info_p)))
1924                 {
1925
1926                         break;
1927                 }
1928                 udelay(10);
1929                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
1930                         " waiting!\n", ctlr);
1931         }
1932         /*
1933          * Send the cmd
1934          */
1935         info_p->access.submit_command(info_p, c);
1936         complete = pollcomplete(ctlr);
1937
1938 #ifdef CCISS_DEBUG
1939         printk(KERN_DEBUG "cciss: command completed\n");
1940 #endif /* CCISS_DEBUG */
1941
1942         if (complete != 1) {
1943                 if ( (complete & CISS_ERROR_BIT)
1944                      && (complete & ~CISS_ERROR_BIT) == c->busaddr)
1945                      {
1946                         /* if data overrun or underun on Report command
1947                                 ignore it
1948                         */
1949                         if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
1950                              (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
1951                              (c->Request.CDB[0] == CISS_INQUIRY)) &&
1952                                 ((c->err_info->CommandStatus ==
1953                                         CMD_DATA_OVERRUN) ||
1954                                  (c->err_info->CommandStatus ==
1955                                         CMD_DATA_UNDERRUN)
1956                                 ))
1957                         {
1958                                 complete = c->busaddr;
1959                         } else {
1960                                 if (c->err_info->CommandStatus ==
1961                                                 CMD_UNSOLICITED_ABORT) {
1962                                         printk(KERN_WARNING "cciss%d: "
1963                                                 "unsolicited abort %p\n",
1964                                                 ctlr, c);
1965                                         if (c->retry_count < MAX_CMD_RETRIES) {
1966                                                 printk(KERN_WARNING
1967                                                    "cciss%d: retrying %p\n",
1968                                                    ctlr, c);
1969                                                 c->retry_count++;
1970                                                 /* erase the old error */
1971                                                 /* information */
1972                                                 memset(c->err_info, 0,
1973                                                    sizeof(ErrorInfo_struct));
1974                                                 goto resend_cmd1;
1975                                         } else {
1976                                                 printk(KERN_WARNING
1977                                                    "cciss%d: retried %p too "
1978                                                    "many times\n", ctlr, c);
1979                                                 status = IO_ERROR;
1980                                                 goto cleanup1;
1981                                         }
1982                                 }
1983                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
1984                                 " Error %x \n", ctlr,
1985                                         c->err_info->CommandStatus);
1986                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
1987                                 " offensive info\n"
1988                                 "  size %x\n   num %x   value %x\n", ctlr,
1989                                   c->err_info->MoreErrInfo.Invalid_Cmd.offense_size,
1990                                   c->err_info->MoreErrInfo.Invalid_Cmd.offense_num,
1991                                   c->err_info->MoreErrInfo.Invalid_Cmd.offense_value);
1992                                 status = IO_ERROR;
1993                                 goto cleanup1;
1994                         }
1995                 }
1996                 if (complete != c->busaddr) {
1997                         printk( KERN_WARNING "cciss cciss%d: SendCmd "
1998                       "Invalid command list address returned! (%lx)\n",
1999                                 ctlr, complete);
2000                         status = IO_ERROR;
2001                         goto cleanup1;
2002                 }
2003         } else {
2004                 printk( KERN_WARNING
2005                         "cciss cciss%d: SendCmd Timeout out, "
2006                         "No command list address returned!\n",
2007                         ctlr);
2008                 status = IO_ERROR;
2009         }
2010
2011 cleanup1:
2012         /* unlock the data buffer from DMA */
2013         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2014         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2015         pci_unmap_single(info_p->pdev, (dma_addr_t) buff_dma_handle.val,
2016                                 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2017         cmd_free(info_p, c, 1);
2018         return (status);
2019 }
2020 /*
2021  * Map (physical) PCI mem into (virtual) kernel space
2022  */
2023 static void __iomem *remap_pci_mem(ulong base, ulong size)
2024 {
2025         ulong page_base        = ((ulong) base) & PAGE_MASK;
2026         ulong page_offs        = ((ulong) base) - page_base;
2027         void __iomem *page_remapped = ioremap(page_base, page_offs+size);
2028
2029         return page_remapped ? (page_remapped + page_offs) : NULL;
2030 }
2031
2032 /*
2033  * Takes jobs of the Q and sends them to the hardware, then puts it on
2034  * the Q to wait for completion.
2035  */
2036 static void start_io( ctlr_info_t *h)
2037 {
2038         CommandList_struct *c;
2039
2040         while(( c = h->reqQ) != NULL )
2041         {
2042                 /* can't do anything if fifo is full */
2043                 if ((h->access.fifo_full(h))) {
2044                         printk(KERN_WARNING "cciss: fifo full\n");
2045                         break;
2046                 }
2047
2048                 /* Get the frist entry from the Request Q */
2049                 removeQ(&(h->reqQ), c);
2050                 h->Qdepth--;
2051
2052                 /* Tell the controller execute command */
2053                 h->access.submit_command(h, c);
2054
2055                 /* Put job onto the completed Q */
2056                 addQ (&(h->cmpQ), c);
2057         }
2058 }
2059
2060 static inline void complete_buffers(struct bio *bio, int status)
2061 {
2062         while (bio) {
2063                 struct bio *xbh = bio->bi_next;
2064                 int nr_sectors = bio_sectors(bio);
2065
2066                 bio->bi_next = NULL;
2067                 blk_finished_io(len);
2068                 bio_endio(bio, nr_sectors << 9, status ? 0 : -EIO);
2069                 bio = xbh;
2070         }
2071
2072 }
2073 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2074 /* Zeros out the error record and then resends the command back */
2075 /* to the controller */
2076 static inline void resend_cciss_cmd( ctlr_info_t *h, CommandList_struct *c)
2077 {
2078         /* erase the old error information */
2079         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2080
2081         /* add it to software queue and then send it to the controller */
2082         addQ(&(h->reqQ),c);
2083         h->Qdepth++;
2084         if(h->Qdepth > h->maxQsinceinit)
2085                 h->maxQsinceinit = h->Qdepth;
2086
2087         start_io(h);
2088 }
2089 /* checks the status of the job and calls complete buffers to mark all
2090  * buffers for the completed job.
2091  */
2092 static inline void complete_command( ctlr_info_t *h, CommandList_struct *cmd,
2093                 int timeout)
2094 {
2095         int status = 1;
2096         int i;
2097         int retry_cmd = 0;
2098         u64bit temp64;
2099
2100         if (timeout)
2101                 status = 0;
2102
2103         if(cmd->err_info->CommandStatus != 0)
2104         { /* an error has occurred */
2105                 switch(cmd->err_info->CommandStatus)
2106                 {
2107                         unsigned char sense_key;
2108                         case CMD_TARGET_STATUS:
2109                                 status = 0;
2110
2111                                 if( cmd->err_info->ScsiStatus == 0x02)
2112                                 {
2113                                         printk(KERN_WARNING "cciss: cmd %p "
2114                                                 "has CHECK CONDITION "
2115                                                 " byte 2 = 0x%x\n", cmd,
2116                                                 cmd->err_info->SenseInfo[2]
2117                                         );
2118                                         /* check the sense key */
2119                                         sense_key = 0xf &
2120                                                 cmd->err_info->SenseInfo[2];
2121                                         /* no status or recovered error */
2122                                         if((sense_key == 0x0) ||
2123                                             (sense_key == 0x1))
2124                                         {
2125                                                         status = 1;
2126                                         }
2127                                 } else
2128                                 {
2129                                         printk(KERN_WARNING "cciss: cmd %p "
2130                                                 "has SCSI Status 0x%x\n",
2131                                                 cmd, cmd->err_info->ScsiStatus);
2132                                 }
2133                         break;
2134                         case CMD_DATA_UNDERRUN:
2135                                 printk(KERN_WARNING "cciss: cmd %p has"
2136                                         " completed with data underrun "
2137                                         "reported\n", cmd);
2138                         break;
2139                         case CMD_DATA_OVERRUN:
2140                                 printk(KERN_WARNING "cciss: cmd %p has"
2141                                         " completed with data overrun "
2142                                         "reported\n", cmd);
2143                         break;
2144                         case CMD_INVALID:
2145                                 printk(KERN_WARNING "cciss: cmd %p is "
2146                                         "reported invalid\n", cmd);
2147                                 status = 0;
2148                         break;
2149                         case CMD_PROTOCOL_ERR:
2150                                 printk(KERN_WARNING "cciss: cmd %p has "
2151                                         "protocol error \n", cmd);
2152                                 status = 0;
2153                         break;
2154                         case CMD_HARDWARE_ERR:
2155                                 printk(KERN_WARNING "cciss: cmd %p had "
2156                                         " hardware error\n", cmd);
2157                                 status = 0;
2158                         break;
2159                         case CMD_CONNECTION_LOST:
2160                                 printk(KERN_WARNING "cciss: cmd %p had "
2161                                         "connection lost\n", cmd);
2162                                 status=0;
2163                         break;
2164                         case CMD_ABORTED:
2165                                 printk(KERN_WARNING "cciss: cmd %p was "
2166                                         "aborted\n", cmd);
2167                                 status=0;
2168                         break;
2169                         case CMD_ABORT_FAILED:
2170                                 printk(KERN_WARNING "cciss: cmd %p reports "
2171                                         "abort failed\n", cmd);
2172                                 status=0;
2173                         break;
2174                         case CMD_UNSOLICITED_ABORT:
2175                                 printk(KERN_WARNING "cciss%d: unsolicited "
2176                                         "abort %p\n", h->ctlr, cmd);
2177                                 if (cmd->retry_count < MAX_CMD_RETRIES) {
2178                                         retry_cmd=1;
2179                                         printk(KERN_WARNING
2180                                                 "cciss%d: retrying %p\n",
2181                                                 h->ctlr, cmd);
2182                                         cmd->retry_count++;
2183                                 } else
2184                                         printk(KERN_WARNING
2185                                                 "cciss%d: %p retried too "
2186                                                 "many times\n", h->ctlr, cmd);
2187                                 status=0;
2188                         break;
2189                         case CMD_TIMEOUT:
2190                                 printk(KERN_WARNING "cciss: cmd %p timedout\n",
2191                                         cmd);
2192                                 status=0;
2193                         break;
2194                         default:
2195                                 printk(KERN_WARNING "cciss: cmd %p returned "
2196                                         "unknown status %x\n", cmd,
2197                                                 cmd->err_info->CommandStatus);
2198                                 status=0;
2199                 }
2200         }
2201         /* We need to return this command */
2202         if(retry_cmd) {
2203                 resend_cciss_cmd(h,cmd);
2204                 return;
2205         }
2206         /* command did not need to be retried */
2207         /* unmap the DMA mapping for all the scatter gather elements */
2208         for(i=0; i<cmd->Header.SGList; i++) {
2209                 temp64.val32.lower = cmd->SG[i].Addr.lower;
2210                 temp64.val32.upper = cmd->SG[i].Addr.upper;
2211                 pci_unmap_page(hba[cmd->ctlr]->pdev,
2212                         temp64.val, cmd->SG[i].Len,
2213                         (cmd->Request.Type.Direction == XFER_READ) ?
2214                                 PCI_DMA_FROMDEVICE : PCI_DMA_TODEVICE);
2215         }
2216         complete_buffers(cmd->rq->bio, status);
2217
2218 #ifdef CCISS_DEBUG
2219         printk("Done with %p\n", cmd->rq);
2220 #endif /* CCISS_DEBUG */
2221
2222         end_that_request_last(cmd->rq);
2223         cmd_free(h,cmd,1);
2224 }
2225
2226 /*
2227  * Get a request and submit it to the controller.
2228  */
2229 static void do_cciss_request(request_queue_t *q)
2230 {
2231         ctlr_info_t *h= q->queuedata;
2232         CommandList_struct *c;
2233         int start_blk, seg;
2234         struct request *creq;
2235         u64bit temp64;
2236         struct scatterlist tmp_sg[MAXSGENTRIES];
2237         drive_info_struct *drv;
2238         int i, dir;
2239
2240         /* We call start_io here in case there is a command waiting on the
2241          * queue that has not been sent.
2242         */
2243         if (blk_queue_plugged(q))
2244                 goto startio;
2245
2246 queue:
2247         creq = elv_next_request(q);
2248         if (!creq)
2249                 goto startio;
2250
2251         if (creq->nr_phys_segments > MAXSGENTRIES)
2252                 BUG();
2253
2254         if (( c = cmd_alloc(h, 1)) == NULL)
2255                 goto full;
2256
2257         blkdev_dequeue_request(creq);
2258
2259         spin_unlock_irq(q->queue_lock);
2260
2261         c->cmd_type = CMD_RWREQ;
2262         c->rq = creq;
2263
2264         /* fill in the request */
2265         drv = creq->rq_disk->private_data;
2266         c->Header.ReplyQueue = 0;  // unused in simple mode
2267         /* got command from pool, so use the command block index instead */
2268         /* for direct lookups. */
2269         /* The first 2 bits are reserved for controller error reporting. */
2270         c->Header.Tag.lower = (c->cmdindex << 3);
2271         c->Header.Tag.lower |= 0x04; /* flag for direct lookup. */
2272         c->Header.LUN.LogDev.VolId= drv->LunID;
2273         c->Header.LUN.LogDev.Mode = 1;
2274         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
2275         c->Request.Type.Type =  TYPE_CMD; // It is a command.
2276         c->Request.Type.Attribute = ATTR_SIMPLE;
2277         c->Request.Type.Direction =
2278                 (rq_data_dir(creq) == READ) ? XFER_READ: XFER_WRITE;
2279         c->Request.Timeout = 0; // Don't time out
2280         c->Request.CDB[0] = (rq_data_dir(creq) == READ) ? CCISS_READ : CCISS_WRITE;
2281         start_blk = creq->sector;
2282 #ifdef CCISS_DEBUG
2283         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",(int) creq->sector,
2284                 (int) creq->nr_sectors);
2285 #endif /* CCISS_DEBUG */
2286
2287         seg = blk_rq_map_sg(q, creq, tmp_sg);
2288
2289         /* get the DMA records for the setup */
2290         if (c->Request.Type.Direction == XFER_READ)
2291                 dir = PCI_DMA_FROMDEVICE;
2292         else
2293                 dir = PCI_DMA_TODEVICE;
2294
2295         for (i=0; i<seg; i++)
2296         {
2297                 c->SG[i].Len = tmp_sg[i].length;
2298                 temp64.val = (__u64) pci_map_page(h->pdev, tmp_sg[i].page,
2299                                           tmp_sg[i].offset, tmp_sg[i].length,
2300                                           dir);
2301                 c->SG[i].Addr.lower = temp64.val32.lower;
2302                 c->SG[i].Addr.upper = temp64.val32.upper;
2303                 c->SG[i].Ext = 0;  // we are not chaining
2304         }
2305         /* track how many SG entries we are using */
2306         if( seg > h->maxSG)
2307                 h->maxSG = seg;
2308
2309 #ifdef CCISS_DEBUG
2310         printk(KERN_DEBUG "cciss: Submitting %d sectors in %d segments\n", creq->nr_sectors, seg);
2311 #endif /* CCISS_DEBUG */
2312
2313         c->Header.SGList = c->Header.SGTotal = seg;
2314         c->Request.CDB[1]= 0;
2315         c->Request.CDB[2]= (start_blk >> 24) & 0xff;    //MSB
2316         c->Request.CDB[3]= (start_blk >> 16) & 0xff;
2317         c->Request.CDB[4]= (start_blk >>  8) & 0xff;
2318         c->Request.CDB[5]= start_blk & 0xff;
2319         c->Request.CDB[6]= 0; // (sect >> 24) & 0xff; MSB
2320         c->Request.CDB[7]= (creq->nr_sectors >>  8) & 0xff;
2321         c->Request.CDB[8]= creq->nr_sectors & 0xff;
2322         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
2323
2324         spin_lock_irq(q->queue_lock);
2325
2326         addQ(&(h->reqQ),c);
2327         h->Qdepth++;
2328         if(h->Qdepth > h->maxQsinceinit)
2329                 h->maxQsinceinit = h->Qdepth;
2330
2331         goto queue;
2332 full:
2333         blk_stop_queue(q);
2334 startio:
2335         /* We will already have the driver lock here so not need
2336          * to lock it.
2337         */
2338         start_io(h);
2339 }
2340
2341 static irqreturn_t do_cciss_intr(int irq, void *dev_id, struct pt_regs *regs)
2342 {
2343         ctlr_info_t *h = dev_id;
2344         CommandList_struct *c;
2345         unsigned long flags;
2346         __u32 a, a1, a2;
2347         int j;
2348         int start_queue = h->next_to_run;
2349
2350         /* Is this interrupt for us? */
2351         if (( h->access.intr_pending(h) == 0) || (h->interrupts_enabled == 0))
2352                 return IRQ_NONE;
2353
2354         /*
2355          * If there are completed commands in the completion queue,
2356          * we had better do something about it.
2357          */
2358         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2359         while( h->access.intr_pending(h))
2360         {
2361                 while((a = h->access.command_completed(h)) != FIFO_EMPTY)
2362                 {
2363                         a1 = a;
2364                         if ((a & 0x04)) {
2365                                 a2 = (a >> 3);
2366                                 if (a2 >= NR_CMDS) {
2367                                         printk(KERN_WARNING "cciss: controller cciss%d failed, stopping.\n", h->ctlr);
2368                                         fail_all_cmds(h->ctlr);
2369                                         return IRQ_HANDLED;
2370                                 }
2371
2372                                 c = h->cmd_pool + a2;
2373                                 a = c->busaddr;
2374
2375                         } else {
2376                         a &= ~3;
2377                                 if ((c = h->cmpQ) == NULL) {
2378                                         printk(KERN_WARNING "cciss: Completion of %08x ignored\n", a1);
2379                                 continue;
2380                         }
2381                         while(c->busaddr != a) {
2382                                 c = c->next;
2383                                 if (c == h->cmpQ)
2384                                         break;
2385                         }
2386                         }
2387                         /*
2388                          * If we've found the command, take it off the
2389                          * completion Q and free it
2390                          */
2391                          if (c->busaddr == a) {
2392                                 removeQ(&h->cmpQ, c);
2393                                 if (c->cmd_type == CMD_RWREQ) {
2394                                         complete_command(h, c, 0);
2395                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
2396                                         complete(c->waiting);
2397                                 }
2398 #                               ifdef CONFIG_CISS_SCSI_TAPE
2399                                 else if (c->cmd_type == CMD_SCSI)
2400                                         complete_scsi_command(c, 0, a1);
2401 #                               endif
2402                                 continue;
2403                         }
2404                 }
2405         }
2406
2407         /* check to see if we have maxed out the number of commands that can
2408          * be placed on the queue.  If so then exit.  We do this check here
2409          * in case the interrupt we serviced was from an ioctl and did not
2410          * free any new commands.
2411          */
2412         if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS)
2413                 goto cleanup;
2414
2415         /* We have room on the queue for more commands.  Now we need to queue
2416          * them up.  We will also keep track of the next queue to run so
2417          * that every queue gets a chance to be started first.
2418         */
2419         for (j=0; j < h->highest_lun + 1; j++){
2420                 int curr_queue = (start_queue + j) % (h->highest_lun + 1);
2421                 /* make sure the disk has been added and the drive is real
2422                  * because this can be called from the middle of init_one.
2423                 */
2424                 if(!(h->drv[curr_queue].queue) ||
2425                                    !(h->drv[curr_queue].heads))
2426                         continue;
2427                 blk_start_queue(h->gendisk[curr_queue]->queue);
2428
2429                 /* check to see if we have maxed out the number of commands
2430                  * that can be placed on the queue.
2431                 */
2432                 if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS)
2433                 {
2434                         if (curr_queue == start_queue){
2435                                 h->next_to_run = (start_queue + 1) % (h->highest_lun + 1);
2436                                 goto cleanup;
2437                         } else {
2438                                 h->next_to_run = curr_queue;
2439                                 goto cleanup;
2440         }
2441                 } else {
2442                         curr_queue = (curr_queue + 1) % (h->highest_lun + 1);
2443                 }
2444         }
2445
2446 cleanup:
2447         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2448         return IRQ_HANDLED;
2449 }
2450 /*
2451  *  We cannot read the structure directly, for portablity we must use
2452  *   the io functions.
2453  *   This is for debug only.
2454  */
2455 #ifdef CCISS_DEBUG
2456 static void print_cfg_table( CfgTable_struct *tb)
2457 {
2458         int i;
2459         char temp_name[17];
2460
2461         printk("Controller Configuration information\n");
2462         printk("------------------------------------\n");
2463         for(i=0;i<4;i++)
2464                 temp_name[i] = readb(&(tb->Signature[i]));
2465         temp_name[4]='\0';
2466         printk("   Signature = %s\n", temp_name);
2467         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
2468         printk("   Transport methods supported = 0x%x\n",
2469                                 readl(&(tb-> TransportSupport)));
2470         printk("   Transport methods active = 0x%x\n",
2471                                 readl(&(tb->TransportActive)));
2472         printk("   Requested transport Method = 0x%x\n",
2473                         readl(&(tb->HostWrite.TransportRequest)));
2474         printk("   Coalese Interrupt Delay = 0x%x\n",
2475                         readl(&(tb->HostWrite.CoalIntDelay)));
2476         printk("   Coalese Interrupt Count = 0x%x\n",
2477                         readl(&(tb->HostWrite.CoalIntCount)));
2478         printk("   Max outstanding commands = 0x%d\n",
2479                         readl(&(tb->CmdsOutMax)));
2480         printk("   Bus Types = 0x%x\n", readl(&(tb-> BusTypes)));
2481         for(i=0;i<16;i++)
2482                 temp_name[i] = readb(&(tb->ServerName[i]));
2483         temp_name[16] = '\0';
2484         printk("   Server Name = %s\n", temp_name);
2485         printk("   Heartbeat Counter = 0x%x\n\n\n",
2486                         readl(&(tb->HeartBeat)));
2487 }
2488 #endif /* CCISS_DEBUG */
2489
2490 static void release_io_mem(ctlr_info_t *c)
2491 {
2492         /* if IO mem was not protected do nothing */
2493         if( c->io_mem_addr == 0)
2494                 return;
2495         release_region(c->io_mem_addr, c->io_mem_length);
2496         c->io_mem_addr = 0;
2497         c->io_mem_length = 0;
2498 }
2499
2500 static int find_PCI_BAR_index(struct pci_dev *pdev,
2501                                 unsigned long pci_bar_addr)
2502 {
2503         int i, offset, mem_type, bar_type;
2504         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
2505                 return 0;
2506         offset = 0;
2507         for (i=0; i<DEVICE_COUNT_RESOURCE; i++) {
2508                 bar_type = pci_resource_flags(pdev, i) &
2509                         PCI_BASE_ADDRESS_SPACE;
2510                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
2511                         offset += 4;
2512                 else {
2513                         mem_type = pci_resource_flags(pdev, i) &
2514                                 PCI_BASE_ADDRESS_MEM_TYPE_MASK;
2515                         switch (mem_type) {
2516                                 case PCI_BASE_ADDRESS_MEM_TYPE_32:
2517                                 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
2518                                         offset += 4; /* 32 bit */
2519                                         break;
2520                                 case PCI_BASE_ADDRESS_MEM_TYPE_64:
2521                                         offset += 8;
2522                                         break;
2523                                 default: /* reserved in PCI 2.2 */
2524                                         printk(KERN_WARNING "Base address is invalid\n");
2525                                         return -1;
2526                                 break;
2527                         }
2528                 }
2529                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
2530                         return i+1;
2531         }
2532         return -1;
2533 }
2534
2535 static int cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
2536 {
2537         ushort subsystem_vendor_id, subsystem_device_id, command;
2538         __u32 board_id, scratchpad = 0;
2539         __u64 cfg_offset;
2540         __u32 cfg_base_addr;
2541         __u64 cfg_base_addr_index;
2542         int i;
2543
2544         /* check to see if controller has been disabled */
2545         /* BEFORE trying to enable it */
2546         (void) pci_read_config_word(pdev, PCI_COMMAND,&command);
2547         if(!(command & 0x02))
2548         {
2549                 printk(KERN_WARNING "cciss: controller appears to be disabled\n");
2550                 return(-1);
2551         }
2552
2553         if (pci_enable_device(pdev))
2554         {
2555                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
2556                 return( -1);
2557         }
2558
2559         subsystem_vendor_id = pdev->subsystem_vendor;
2560         subsystem_device_id = pdev->subsystem_device;
2561         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
2562                                         subsystem_vendor_id);
2563
2564         /* search for our IO range so we can protect it */
2565         for(i=0; i<DEVICE_COUNT_RESOURCE; i++)
2566         {
2567                 /* is this an IO range */
2568                 if( pci_resource_flags(pdev, i) & 0x01 ) {
2569                         c->io_mem_addr = pci_resource_start(pdev, i);
2570                         c->io_mem_length = pci_resource_end(pdev, i) -
2571                                 pci_resource_start(pdev, i) +1;
2572 #ifdef CCISS_DEBUG
2573                         printk("IO value found base_addr[%d] %lx %lx\n", i,
2574                                 c->io_mem_addr, c->io_mem_length);
2575 #endif /* CCISS_DEBUG */
2576                         /* register the IO range */
2577                         if(!request_region( c->io_mem_addr,
2578                                         c->io_mem_length, "cciss"))
2579                         {
2580                                 printk(KERN_WARNING "cciss I/O memory range already in use addr=%lx length=%ld\n",
2581                                 c->io_mem_addr, c->io_mem_length);
2582                                 c->io_mem_addr= 0;
2583                                 c->io_mem_length = 0;
2584                         }
2585                         break;
2586                 }
2587         }
2588
2589 #ifdef CCISS_DEBUG
2590         printk("command = %x\n", command);
2591         printk("irq = %x\n", pdev->irq);
2592         printk("board_id = %x\n", board_id);
2593 #endif /* CCISS_DEBUG */
2594
2595         c->intr = pdev->irq;
2596
2597         /*
2598          * Memory base addr is first addr , the second points to the config
2599          *   table
2600          */
2601
2602         c->paddr = pci_resource_start(pdev, 0); /* addressing mode bits already removed */
2603 #ifdef CCISS_DEBUG
2604         printk("address 0 = %x\n", c->paddr);
2605 #endif /* CCISS_DEBUG */
2606         c->vaddr = remap_pci_mem(c->paddr, 200);
2607
2608         /* Wait for the board to become ready.  (PCI hotplug needs this.)
2609          * We poll for up to 120 secs, once per 100ms. */
2610         for (i=0; i < 1200; i++) {
2611                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
2612                 if (scratchpad == CCISS_FIRMWARE_READY)
2613                         break;
2614                 set_current_state(TASK_INTERRUPTIBLE);
2615                 schedule_timeout(HZ / 10); /* wait 100ms */
2616         }
2617         if (scratchpad != CCISS_FIRMWARE_READY) {
2618                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
2619                 return -1;
2620         }
2621
2622         /* get the address index number */
2623         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
2624         cfg_base_addr &= (__u32) 0x0000ffff;
2625 #ifdef CCISS_DEBUG
2626         printk("cfg base address = %x\n", cfg_base_addr);
2627 #endif /* CCISS_DEBUG */
2628         cfg_base_addr_index =
2629                 find_PCI_BAR_index(pdev, cfg_base_addr);
2630 #ifdef CCISS_DEBUG
2631         printk("cfg base address index = %x\n", cfg_base_addr_index);
2632 #endif /* CCISS_DEBUG */
2633         if (cfg_base_addr_index == -1) {
2634                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
2635                 release_io_mem(c);
2636                 return -1;
2637         }
2638
2639         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
2640 #ifdef CCISS_DEBUG
2641         printk("cfg offset = %x\n", cfg_offset);
2642 #endif /* CCISS_DEBUG */
2643         c->cfgtable =  remap_pci_mem(pci_resource_start(pdev,
2644                                 cfg_base_addr_index) + cfg_offset,
2645                                 sizeof(CfgTable_struct));
2646         c->board_id = board_id;
2647
2648 #ifdef CCISS_DEBUG
2649         print_cfg_table(c->cfgtable);
2650 #endif /* CCISS_DEBUG */
2651
2652         for(i=0; i<NR_PRODUCTS; i++) {
2653                 if (board_id == products[i].board_id) {
2654                         c->product_name = products[i].product_name;
2655                         c->access = *(products[i].access);
2656                         break;
2657                 }
2658         }
2659         if (i == NR_PRODUCTS) {
2660                 printk(KERN_WARNING "cciss: Sorry, I don't know how"
2661                         " to access the Smart Array controller %08lx\n",
2662                                 (unsigned long)board_id);
2663                 return -1;
2664         }
2665         if (  (readb(&c->cfgtable->Signature[0]) != 'C') ||
2666               (readb(&c->cfgtable->Signature[1]) != 'I') ||
2667               (readb(&c->cfgtable->Signature[2]) != 'S') ||
2668               (readb(&c->cfgtable->Signature[3]) != 'S') )
2669         {
2670                 printk("Does not appear to be a valid CISS config table\n");
2671                 return -1;
2672         }
2673
2674 #ifdef CONFIG_X86
2675 {
2676         /* Need to enable prefetch in the SCSI core for 6400 in x86 */
2677         __u32 prefetch;
2678         prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
2679         prefetch |= 0x100;
2680         writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
2681 }
2682 #endif
2683
2684 #ifdef CCISS_DEBUG
2685         printk("Trying to put board into Simple mode\n");
2686 #endif /* CCISS_DEBUG */
2687         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
2688         /* Update the field, and then ring the doorbell */
2689         writel( CFGTBL_Trans_Simple,
2690                 &(c->cfgtable->HostWrite.TransportRequest));
2691         writel( CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
2692
2693         /* under certain very rare conditions, this can take awhile.
2694          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
2695          * as we enter this code.) */
2696         for(i=0;i<MAX_CONFIG_WAIT;i++) {
2697                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
2698                         break;
2699                 /* delay and try again */
2700                 set_current_state(TASK_INTERRUPTIBLE);
2701                 schedule_timeout(10);
2702         }
2703
2704 #ifdef CCISS_DEBUG
2705         printk(KERN_DEBUG "I counter got to %d %x\n", i, readl(c->vaddr + SA5_DOORBELL));
2706 #endif /* CCISS_DEBUG */
2707 #ifdef CCISS_DEBUG
2708         print_cfg_table(c->cfgtable);
2709 #endif /* CCISS_DEBUG */
2710
2711         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple))
2712         {
2713                 printk(KERN_WARNING "cciss: unable to get board into"
2714                                         " simple mode\n");
2715                 return -1;
2716         }
2717         return 0;
2718
2719 }
2720
2721 /*
2722  * Gets information about the local volumes attached to the controller.
2723  */
2724 static void cciss_getgeometry(int cntl_num)
2725 {
2726         ReportLunData_struct *ld_buff;
2727         ReadCapdata_struct *size_buff;
2728         InquiryData_struct *inq_buff;
2729         int return_code;
2730         int i;
2731         int listlength = 0;
2732         __u32 lunid = 0;
2733         int block_size;
2734         int total_size;
2735
2736         ld_buff = kmalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2737         if (ld_buff == NULL)
2738         {
2739                 printk(KERN_ERR "cciss: out of memory\n");
2740                 return;
2741         }
2742         memset(ld_buff, 0, sizeof(ReportLunData_struct));
2743         size_buff = kmalloc(sizeof( ReadCapdata_struct), GFP_KERNEL);
2744         if (size_buff == NULL)
2745         {
2746                 printk(KERN_ERR "cciss: out of memory\n");
2747                 kfree(ld_buff);
2748                 return;
2749         }
2750         inq_buff = kmalloc(sizeof( InquiryData_struct), GFP_KERNEL);
2751         if (inq_buff == NULL)
2752         {
2753                 printk(KERN_ERR "cciss: out of memory\n");
2754                 kfree(ld_buff);
2755                 kfree(size_buff);
2756                 return;
2757         }
2758         /* Get the firmware version */
2759         return_code = sendcmd(CISS_INQUIRY, cntl_num, inq_buff,
2760                 sizeof(InquiryData_struct), 0, 0 ,0, NULL, TYPE_CMD);
2761         if (return_code == IO_OK)
2762         {
2763                 hba[cntl_num]->firm_ver[0] = inq_buff->data_byte[32];
2764                 hba[cntl_num]->firm_ver[1] = inq_buff->data_byte[33];
2765                 hba[cntl_num]->firm_ver[2] = inq_buff->data_byte[34];
2766                 hba[cntl_num]->firm_ver[3] = inq_buff->data_byte[35];
2767         } else /* send command failed */
2768         {
2769                 printk(KERN_WARNING "cciss: unable to determine firmware"
2770                         " version of controller\n");
2771         }
2772         /* Get the number of logical volumes */
2773         return_code = sendcmd(CISS_REPORT_LOG, cntl_num, ld_buff,
2774                         sizeof(ReportLunData_struct), 0, 0, 0, NULL, TYPE_CMD);
2775
2776         if( return_code == IO_OK)
2777         {
2778 #ifdef CCISS_DEBUG
2779                 printk("LUN Data\n--------------------------\n");
2780 #endif /* CCISS_DEBUG */
2781
2782                 listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[0])) << 24;
2783                 listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[1])) << 16;
2784                 listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[2])) << 8;
2785                 listlength |= 0xff & (unsigned int)(ld_buff->LUNListLength[3]);
2786         } else /* reading number of logical volumes failed */
2787         {
2788                 printk(KERN_WARNING "cciss: report logical volume"
2789                         " command failed\n");
2790                 listlength = 0;
2791         }
2792         hba[cntl_num]->num_luns = listlength / 8; // 8 bytes pre entry
2793         if (hba[cntl_num]->num_luns > CISS_MAX_LUN)
2794         {
2795                 printk(KERN_ERR "ciss:  only %d number of logical volumes supported\n",
2796                         CISS_MAX_LUN);
2797                 hba[cntl_num]->num_luns = CISS_MAX_LUN;
2798         }
2799 #ifdef CCISS_DEBUG
2800         printk(KERN_DEBUG "Length = %x %x %x %x = %d\n", ld_buff->LUNListLength[0],
2801                 ld_buff->LUNListLength[1], ld_buff->LUNListLength[2],
2802                 ld_buff->LUNListLength[3],  hba[cntl_num]->num_luns);
2803 #endif /* CCISS_DEBUG */
2804
2805         hba[cntl_num]->highest_lun = hba[cntl_num]->num_luns-1;
2806 //      for(i=0; i<  hba[cntl_num]->num_luns; i++)
2807         for(i=0; i < CISS_MAX_LUN; i++)
2808         {
2809                 if (i < hba[cntl_num]->num_luns){
2810                         lunid = (0xff & (unsigned int)(ld_buff->LUN[i][3]))
2811                                  << 24;
2812                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][2]))
2813                                  << 16;
2814                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][1]))
2815                                  << 8;
2816                 lunid |= 0xff & (unsigned int)(ld_buff->LUN[i][0]);
2817
2818                 hba[cntl_num]->drv[i].LunID = lunid;
2819
2820
2821 #ifdef CCISS_DEBUG
2822                 printk(KERN_DEBUG "LUN[%d]:  %x %x %x %x = %x\n", i,
2823                         ld_buff->LUN[i][0], ld_buff->LUN[i][1],
2824                         ld_buff->LUN[i][2], ld_buff->LUN[i][3],
2825                         hba[cntl_num]->drv[i].LunID);
2826 #endif /* CCISS_DEBUG */
2827                 cciss_read_capacity(cntl_num, i, size_buff, 0,
2828                         &total_size, &block_size);
2829                         cciss_geometry_inquiry(cntl_num, i, 0, total_size,
2830                                 block_size, inq_buff, &hba[cntl_num]->drv[i]);
2831                 } else {
2832                         /* initialize raid_level to indicate a free space */
2833                         hba[cntl_num]->drv[i].raid_level = -1;
2834                 }
2835         }
2836         kfree(ld_buff);
2837         kfree(size_buff);
2838         kfree(inq_buff);
2839 }
2840
2841 /* Function to find the first free pointer into our hba[] array */
2842 /* Returns -1 if no free entries are left.  */
2843 static int alloc_cciss_hba(void)
2844 {
2845         struct gendisk *disk[NWD];
2846         int i, n;
2847         for (n = 0; n < NWD; n++) {
2848                 disk[n] = alloc_disk(1 << NWD_SHIFT);
2849                 if (!disk[n])
2850                         goto out;
2851         }
2852
2853         for(i=0; i< MAX_CTLR; i++) {
2854                 if (!hba[i]) {
2855                         ctlr_info_t *p;
2856                         p = kmalloc(sizeof(ctlr_info_t), GFP_KERNEL);
2857                         if (!p)
2858                                 goto Enomem;
2859                         memset(p, 0, sizeof(ctlr_info_t));
2860                         for (n = 0; n < NWD; n++)
2861                                 p->gendisk[n] = disk[n];
2862                         hba[i] = p;
2863                         return i;
2864                 }
2865         }
2866         printk(KERN_WARNING "cciss: This driver supports a maximum"
2867                 " of %d controllers.\n", MAX_CTLR);
2868         goto out;
2869 Enomem:
2870         printk(KERN_ERR "cciss: out of memory.\n");
2871 out:
2872         while (n--)
2873                 put_disk(disk[n]);
2874         return -1;
2875 }
2876
2877 static void free_hba(int i)
2878 {
2879         ctlr_info_t *p = hba[i];
2880         int n;
2881
2882         hba[i] = NULL;
2883         for (n = 0; n < NWD; n++)
2884                 put_disk(p->gendisk[n]);
2885         kfree(p);
2886 }
2887
2888 /*
2889  *  This is it.  Find all the controllers and register them.  I really hate
2890  *  stealing all these major device numbers.
2891  *  returns the number of block devices registered.
2892  */
2893 static int __devinit cciss_init_one(struct pci_dev *pdev,
2894         const struct pci_device_id *ent)
2895 {
2896         request_queue_t *q;
2897         int i;
2898         int j;
2899         int rc;
2900
2901         printk(KERN_DEBUG "cciss: Device 0x%x has been found at"
2902                         " bus %d dev %d func %d\n",
2903                 pdev->device, pdev->bus->number, PCI_SLOT(pdev->devfn),
2904                         PCI_FUNC(pdev->devfn));
2905         i = alloc_cciss_hba();
2906         if(i < 0)
2907                 return (-1);
2908
2909         hba[i]->busy_initializing = 1;
2910
2911         if (cciss_pci_init(hba[i], pdev) != 0)
2912                 goto clean1;
2913
2914         sprintf(hba[i]->devname, "cciss%d", i);
2915         hba[i]->ctlr = i;
2916         hba[i]->pdev = pdev;
2917
2918         /* configure PCI DMA stuff */
2919         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK))
2920                 printk("cciss: using DAC cycles\n");
2921         else if (!pci_set_dma_mask(pdev, DMA_32BIT_MASK))
2922                 printk("cciss: not using DAC cycles\n");
2923         else {
2924                 printk("cciss: no suitable DMA available\n");
2925                 goto clean1;
2926         }
2927
2928         /*
2929          * register with the major number, or get a dynamic major number
2930          * by passing 0 as argument.  This is done for greater than
2931          * 8 controller support.
2932          */
2933         if (i < MAX_CTLR_ORIG)
2934                 hba[i]->major = MAJOR_NR + i;
2935         rc = register_blkdev(hba[i]->major, hba[i]->devname);
2936         if(rc == -EBUSY || rc == -EINVAL) {
2937                 printk(KERN_ERR
2938                         "cciss:  Unable to get major number %d for %s "
2939                         "on hba %d\n", hba[i]->major, hba[i]->devname, i);
2940                 goto clean1;
2941         }
2942         else {
2943                 if (i >= MAX_CTLR_ORIG)
2944                         hba[i]->major = rc;
2945         }
2946
2947         /* make sure the board interrupts are off */
2948         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
2949         if( request_irq(hba[i]->intr, do_cciss_intr,
2950                 SA_INTERRUPT | SA_SHIRQ | SA_SAMPLE_RANDOM,
2951                         hba[i]->devname, hba[i])) {
2952                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
2953                         hba[i]->intr, hba[i]->devname);
2954                 goto clean2;
2955         }
2956         hba[i]->cmd_pool_bits = kmalloc(((NR_CMDS+BITS_PER_LONG-1)/BITS_PER_LONG)*sizeof(unsigned long), GFP_KERNEL);
2957         hba[i]->cmd_pool = (CommandList_struct *)pci_alloc_consistent(
2958                 hba[i]->pdev, NR_CMDS * sizeof(CommandList_struct),
2959                 &(hba[i]->cmd_pool_dhandle));
2960         hba[i]->errinfo_pool = (ErrorInfo_struct *)pci_alloc_consistent(
2961                 hba[i]->pdev, NR_CMDS * sizeof( ErrorInfo_struct),
2962                 &(hba[i]->errinfo_pool_dhandle));
2963         if((hba[i]->cmd_pool_bits == NULL)
2964                 || (hba[i]->cmd_pool == NULL)
2965                 || (hba[i]->errinfo_pool == NULL)) {
2966                 printk( KERN_ERR "cciss: out of memory");
2967                 goto clean4;
2968         }
2969
2970         spin_lock_init(&hba[i]->lock);
2971
2972         /* Initialize the pdev driver private data.
2973                 have it point to hba[i].  */
2974         pci_set_drvdata(pdev, hba[i]);
2975         /* command and error info recs zeroed out before
2976                         they are used */
2977         memset(hba[i]->cmd_pool_bits, 0, ((NR_CMDS+BITS_PER_LONG-1)/BITS_PER_LONG)*sizeof(unsigned long));
2978
2979 #ifdef CCISS_DEBUG
2980         printk(KERN_DEBUG "Scanning for drives on controller cciss%d\n",i);
2981 #endif /* CCISS_DEBUG */
2982
2983         cciss_getgeometry(i);
2984
2985         cciss_scsi_setup(i);
2986
2987         /* Turn the interrupts on so we can service requests */
2988         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
2989
2990         cciss_procinit(i);
2991         hba[i]->busy_initializing = 0;
2992
2993         for(j=0; j < NWD; j++) { /* mfm */
2994                 drive_info_struct *drv = &(hba[i]->drv[j]);
2995                 struct gendisk *disk = hba[i]->gendisk[j];
2996
2997                 q = blk_init_queue(do_cciss_request, &hba[i]->lock);
2998                 if (!q) {
2999                         printk(KERN_ERR
3000                            "cciss:  unable to allocate queue for disk %d\n",
3001                            j);
3002                         break;
3003                 }
3004                 drv->queue = q;
3005
3006                 q->backing_dev_info.ra_pages = READ_AHEAD;
3007         blk_queue_bounce_limit(q, hba[i]->pdev->dma_mask);
3008
3009         /* This is a hardware imposed limit. */
3010         blk_queue_max_hw_segments(q, MAXSGENTRIES);
3011
3012         /* This is a limit in the driver and could be eliminated. */
3013         blk_queue_max_phys_segments(q, MAXSGENTRIES);
3014
3015         blk_queue_max_sectors(q, 512);
3016
3017                 q->queuedata = hba[i];
3018                 sprintf(disk->disk_name, "cciss/c%dd%d", i, j);
3019                 sprintf(disk->devfs_name, "cciss/host%d/target%d", i, j);
3020                 disk->major = hba[i]->major;
3021                 disk->first_minor = j << NWD_SHIFT;
3022                 disk->fops = &cciss_fops;
3023                 disk->queue = q;
3024                 disk->private_data = drv;
3025                 /* we must register the controller even if no disks exist */
3026                 /* this is for the online array utilities */
3027                 if(!drv->heads && j)
3028                         continue;
3029                 blk_queue_hardsect_size(q, drv->block_size);
3030                 set_capacity(disk, drv->nr_blocks);
3031                 add_disk(disk);
3032         }
3033
3034         return(1);
3035
3036 clean4:
3037         if(hba[i]->cmd_pool_bits)
3038                 kfree(hba[i]->cmd_pool_bits);
3039         if(hba[i]->cmd_pool)
3040                 pci_free_consistent(hba[i]->pdev,
3041                         NR_CMDS * sizeof(CommandList_struct),
3042                         hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3043         if(hba[i]->errinfo_pool)
3044                 pci_free_consistent(hba[i]->pdev,
3045                         NR_CMDS * sizeof( ErrorInfo_struct),
3046                         hba[i]->errinfo_pool,
3047                         hba[i]->errinfo_pool_dhandle);
3048         free_irq(hba[i]->intr, hba[i]);
3049 clean2:
3050         unregister_blkdev(hba[i]->major, hba[i]->devname);
3051 clean1:
3052         release_io_mem(hba[i]);
3053         free_hba(i);
3054         hba[i]->busy_initializing = 0;
3055         return(-1);
3056 }
3057
3058 static void __devexit cciss_remove_one (struct pci_dev *pdev)
3059 {
3060         ctlr_info_t *tmp_ptr;
3061         int i, j;
3062         char flush_buf[4];
3063         int return_code;
3064
3065         if (pci_get_drvdata(pdev) == NULL)
3066         {
3067                 printk( KERN_ERR "cciss: Unable to remove device \n");
3068                 return;
3069         }
3070         tmp_ptr = pci_get_drvdata(pdev);
3071         i = tmp_ptr->ctlr;
3072         if (hba[i] == NULL)
3073         {
3074                 printk(KERN_ERR "cciss: device appears to "
3075                         "already be removed \n");
3076                 return;
3077         }
3078         /* Turn board interrupts off  and send the flush cache command */
3079         /* sendcmd will turn off interrupt, and send the flush...
3080         * To write all data in the battery backed cache to disks */
3081         memset(flush_buf, 0, 4);
3082         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0, 0, 0, NULL,
3083                                 TYPE_CMD);
3084         if(return_code != IO_OK)
3085         {
3086                 printk(KERN_WARNING "Error Flushing cache on controller %d\n",
3087                         i);
3088         }
3089         free_irq(hba[i]->intr, hba[i]);
3090         pci_set_drvdata(pdev, NULL);
3091         iounmap(hba[i]->vaddr);
3092         cciss_unregister_scsi(i);  /* unhook from SCSI subsystem */
3093         unregister_blkdev(hba[i]->major, hba[i]->devname);
3094         remove_proc_entry(hba[i]->devname, proc_cciss);
3095
3096         /* remove it from the disk list */
3097         for (j = 0; j < NWD; j++) {
3098                 struct gendisk *disk = hba[i]->gendisk[j];
3099                 if (disk->flags & GENHD_FL_UP) {
3100                         del_gendisk(disk);
3101                         blk_cleanup_queue(disk->queue);
3102                 }
3103         }
3104
3105         pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof(CommandList_struct),
3106                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3107         pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof( ErrorInfo_struct),
3108                 hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
3109         kfree(hba[i]->cmd_pool_bits);
3110         release_io_mem(hba[i]);
3111         free_hba(i);
3112 }
3113
3114 static struct pci_driver cciss_pci_driver = {
3115         .name =         "cciss",
3116         .probe =        cciss_init_one,
3117         .remove =       __devexit_p(cciss_remove_one),
3118         .id_table =     cciss_pci_device_id, /* id_table */
3119 };
3120
3121 /*
3122  *  This is it.  Register the PCI driver information for the cards we control
3123  *  the OS will call our registered routines when it finds one of our cards.
3124  */
3125 static int __init cciss_init(void)
3126 {
3127         printk(KERN_INFO DRIVER_NAME "\n");
3128
3129         /* Register for our PCI devices */
3130         return pci_module_init(&cciss_pci_driver);
3131 }
3132
3133 static void __exit cciss_cleanup(void)
3134 {
3135         int i;
3136
3137         pci_unregister_driver(&cciss_pci_driver);
3138         /* double check that all controller entrys have been removed */
3139         for (i=0; i< MAX_CTLR; i++)
3140         {
3141                 if (hba[i] != NULL)
3142                 {
3143                         printk(KERN_WARNING "cciss: had to remove"
3144                                         " controller %d\n", i);
3145                         cciss_remove_one(hba[i]->pdev);
3146                 }
3147         }
3148         remove_proc_entry("cciss", proc_root_driver);
3149 }
3150
3151 static void fail_all_cmds(unsigned long ctlr)
3152 {
3153         /* If we get here, the board is apparently dead. */
3154         ctlr_info_t *h = hba[ctlr];
3155         CommandList_struct *c;
3156         unsigned long flags;
3157
3158         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
3159         h->alive = 0;   /* the controller apparently died... */
3160
3161         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
3162
3163         pci_disable_device(h->pdev); /* Make sure it is really dead. */
3164
3165         /* move everything off the request queue onto the completed queue */
3166         while( (c = h->reqQ) != NULL ) {
3167                 removeQ(&(h->reqQ), c);
3168                 h->Qdepth--;
3169                 addQ (&(h->cmpQ), c);
3170         }
3171
3172         /* Now, fail everything on the completed queue with a HW error */
3173         while( (c = h->cmpQ) != NULL ) {
3174                 removeQ(&h->cmpQ, c);
3175                 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
3176                 if (c->cmd_type == CMD_RWREQ) {
3177                         complete_command(h, c, 0);
3178                 } else if (c->cmd_type == CMD_IOCTL_PEND)
3179                         complete(c->waiting);
3180 #ifdef CONFIG_CISS_SCSI_TAPE
3181                         else if (c->cmd_type == CMD_SCSI)
3182                                 complete_scsi_command(c, 0, 0);
3183 #endif
3184         }
3185         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
3186         return;
3187 }
3188
3189 module_init(cciss_init);
3190 module_exit(cciss_cleanup);