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)
 487                 return -EBUSY;
 488
 489         if (host->busy_initializing || drv->busy_configuring)
 490                 return -EBUSY;
 491         /*
 492          * Root is allowed to open raw volume zero even if it's not configured
 493          * so array config can still work. Root is also allowed to open any
 494          * volume that has a LUN ID, so it can issue IOCTL to reread the
 495          * disk information.  I don't think I really like this
 496          * but I'm already using way to many device nodes to claim another one
 497          * for "raw controller".
 498          */
 499         if (drv->nr_blocks == 0) {
 500                 if (iminor(inode) != 0) {       /* not node 0? */
 501                         /* if not node 0 make sure it is a partition = 0 */
 502                         if (iminor(inode) & 0x0f) {
 503                         return -ENXIO;
 504                                 /* if it is, make sure we have a LUN ID */
 505                         } else if (drv->LunID == 0) {
 506                                 return -ENXIO;
 507                         }
 508                 }
 509                 if (!capable(CAP_SYS_ADMIN))
 510                         return -EPERM;
 511         }
 512         drv->usage_count++;
 513         host->usage_count++;
 514         return 0;
 515 }
 516 /*
 517  * Close.  Sync first.
 518  */
 519 static int cciss_release(struct inode *inode, struct file *filep)
 520 {
 521         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
 522         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
 523
 524 #ifdef CCISS_DEBUG
 525         printk(KERN_DEBUG "cciss_release %s\n", inode->i_bdev->bd_disk->disk_name);
 526 #endif /* CCISS_DEBUG */
 527
 528         drv->usage_count--;
 529         host->usage_count--;
 530         return 0;
 531 }
 532
 533 #ifdef CONFIG_COMPAT
 534
 535 static int do_ioctl(struct file *f, unsigned cmd, unsigned long arg)
 536 {
 537         int ret;
 538         lock_kernel();
 539         ret = cciss_ioctl(f->f_dentry->d_inode, f, cmd, arg);
 540         unlock_kernel();
 541         return ret;
 542 }
 543
 544 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd, unsigned long arg);
 545 static int cciss_ioctl32_big_passthru(struct file *f, unsigned cmd, unsigned long arg);
 546
 547 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg)
 548 {
 549         switch (cmd) {
 550         case CCISS_GETPCIINFO:
 551         case CCISS_GETINTINFO:
 552         case CCISS_SETINTINFO:
 553         case CCISS_GETNODENAME:
 554         case CCISS_SETNODENAME:
 555         case CCISS_GETHEARTBEAT:
 556         case CCISS_GETBUSTYPES:
 557         case CCISS_GETFIRMVER:
 558         case CCISS_GETDRIVVER:
 559         case CCISS_REVALIDVOLS:
 560         case CCISS_DEREGDISK:
 561         case CCISS_REGNEWDISK:
 562         case CCISS_REGNEWD:
 563         case CCISS_RESCANDISK:
 564         case CCISS_GETLUNINFO:
 565                 return do_ioctl(f, cmd, arg);
 566
 567         case CCISS_PASSTHRU32:
 568                 return cciss_ioctl32_passthru(f, cmd, arg);
 569         case CCISS_BIG_PASSTHRU32:
 570                 return cciss_ioctl32_big_passthru(f, cmd, arg);
 571
 572         default:
 573                 return -ENOIOCTLCMD;
 574         }
 575 }
 576
 577 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd, unsigned long arg)
 578 {
 579         IOCTL32_Command_struct __user *arg32 =
 580                 (IOCTL32_Command_struct __user *) arg;
 581         IOCTL_Command_struct arg64;
 582         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
 583         int err;
 584         u32 cp;
 585
 586         err = 0;
 587         err |= copy_from_user(&arg64.LUN_info, &arg32->LUN_info, sizeof(arg64.LUN_info));
 588         err |= copy_from_user(&arg64.Request, &arg32->Request, sizeof(arg64.Request));
 589         err |= copy_from_user(&arg64.error_info, &arg32->error_info, sizeof(arg64.error_info));
 590         err |= get_user(arg64.buf_size, &arg32->buf_size);
 591         err |= get_user(cp, &arg32->buf);
 592         arg64.buf = compat_ptr(cp);
 593         err |= copy_to_user(p, &arg64, sizeof(arg64));
 594
 595         if (err)
 596                 return -EFAULT;
 597
 598         err = do_ioctl(f, CCISS_PASSTHRU, (unsigned long) p);
 599         if (err)
 600                 return err;
 601         err |= copy_in_user(&arg32->error_info, &p->error_info, sizeof(arg32->error_info));
 602         if (err)
 603                 return -EFAULT;
 604         return err;
 605 }
 606
 607 static int cciss_ioctl32_big_passthru(struct file *file, unsigned cmd, unsigned long arg)
 608 {
 609         BIG_IOCTL32_Command_struct __user *arg32 =
 610                 (BIG_IOCTL32_Command_struct __user *) arg;
 611         BIG_IOCTL_Command_struct arg64;
 612         BIG_IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
 613         int err;
 614         u32 cp;
 615
 616         err = 0;
 617         err |= copy_from_user(&arg64.LUN_info, &arg32->LUN_info, sizeof(arg64.LUN_info));
 618         err |= copy_from_user(&arg64.Request, &arg32->Request, sizeof(arg64.Request));
 619         err |= copy_from_user(&arg64.error_info, &arg32->error_info, sizeof(arg64.error_info));
 620         err |= get_user(arg64.buf_size, &arg32->buf_size);
 621         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
 622         err |= get_user(cp, &arg32->buf);
 623         arg64.buf = compat_ptr(cp);
 624         err |= copy_to_user(p, &arg64, sizeof(arg64));
 625
 626         if (err)
 627                  return -EFAULT;
 628
 629         err = do_ioctl(file, CCISS_BIG_PASSTHRU, (unsigned long) p);
 630         if (err)
 631                 return err;
 632         err |= copy_in_user(&arg32->error_info, &p->error_info, sizeof(arg32->error_info));
 633         if (err)
 634                 return -EFAULT;
 635         return err;
 636 }
 637 #endif
 638 /*
 639  * ioctl
 640  */
 641 static int cciss_ioctl(struct inode *inode, struct file *filep,
 642                 unsigned int cmd, unsigned long arg)
 643 {
 644         struct block_device *bdev = inode->i_bdev;
 645         struct gendisk *disk = bdev->bd_disk;
 646         ctlr_info_t *host = get_host(disk);
 647         drive_info_struct *drv = get_drv(disk);
 648         int ctlr = host->ctlr;
 649         void __user *argp = (void __user *)arg;
 650
 651 #ifdef CCISS_DEBUG
 652         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
 653 #endif /* CCISS_DEBUG */
 654
 655         switch(cmd) {
 656         case HDIO_GETGEO:
 657         {
 658                 struct hd_geometry driver_geo;
 659                 if (drv->cylinders) {
 660                         driver_geo.heads = drv->heads;
 661                         driver_geo.sectors = drv->sectors;
 662                         driver_geo.cylinders = drv->cylinders;
 663                 } else
 664                         return -ENXIO;
 665                 driver_geo.start= get_start_sect(inode->i_bdev);
 666                 if (copy_to_user(argp, &driver_geo, sizeof(struct hd_geometry)))
 667                         return  -EFAULT;
 668                 return(0);
 669         }
 670
 671         case CCISS_GETPCIINFO:
 672         {
 673                 cciss_pci_info_struct pciinfo;
 674
 675                 if (!arg) return -EINVAL;
 676                 pciinfo.domain = pci_domain_nr(host->pdev->bus);
 677                 pciinfo.bus = host->pdev->bus->number;
 678                 pciinfo.dev_fn = host->pdev->devfn;
 679                 pciinfo.board_id = host->board_id;
 680                 if (copy_to_user(argp, &pciinfo,  sizeof( cciss_pci_info_struct )))
 681                         return  -EFAULT;
 682                 return(0);
 683         }
 684         case CCISS_GETINTINFO:
 685         {
 686                 cciss_coalint_struct intinfo;
 687                 if (!arg) return -EINVAL;
 688                 intinfo.delay = readl(&host->cfgtable->HostWrite.CoalIntDelay);
 689                 intinfo.count = readl(&host->cfgtable->HostWrite.CoalIntCount);
 690                 if (copy_to_user(argp, &intinfo, sizeof( cciss_coalint_struct )))
 691                         return -EFAULT;
 692                 return(0);
 693         }
 694         case CCISS_SETINTINFO:
 695         {
 696                 cciss_coalint_struct intinfo;
 697                 unsigned long flags;
 698                 int i;
 699
 700                 if (!arg) return -EINVAL;
 701                 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
 702                 if (copy_from_user(&intinfo, argp, sizeof( cciss_coalint_struct)))
 703                         return -EFAULT;
 704                 if ( (intinfo.delay == 0 ) && (intinfo.count == 0))
 705
 706                 {
 707 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
 708                         return( -EINVAL);
 709                 }
 710                 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 711                 /* Update the field, and then ring the doorbell */
 712                 writel( intinfo.delay,
 713                         &(host->cfgtable->HostWrite.CoalIntDelay));
 714                 writel( intinfo.count,
 715                         &(host->cfgtable->HostWrite.CoalIntCount));
 716                 writel( CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
 717
 718                 for(i=0;i<MAX_IOCTL_CONFIG_WAIT;i++) {
 719                         if (!(readl(host->vaddr + SA5_DOORBELL)
 720                                         & CFGTBL_ChangeReq))
 721                                 break;
 722                         /* delay and try again */
 723                         udelay(1000);
 724                 }
 725                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 726                 if (i >= MAX_IOCTL_CONFIG_WAIT)
 727                         return -EAGAIN;
 728                 return(0);
 729         }
 730         case CCISS_GETNODENAME:
 731         {
 732                 NodeName_type NodeName;
 733                 int i;
 734
 735                 if (!arg) return -EINVAL;
 736                 for(i=0;i<16;i++)
 737                         NodeName[i] = readb(&host->cfgtable->ServerName[i]);
 738                 if (copy_to_user(argp, NodeName, sizeof( NodeName_type)))
 739                         return  -EFAULT;
 740                 return(0);
 741         }
 742         case CCISS_SETNODENAME:
 743         {
 744                 NodeName_type NodeName;
 745                 unsigned long flags;
 746                 int i;
 747
 748                 if (!arg) return -EINVAL;
 749                 if (!capable(CAP_SYS_ADMIN)) return -EPERM;
 750
 751                 if (copy_from_user(NodeName, argp, sizeof( NodeName_type)))
 752                         return -EFAULT;
 753
 754                 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 755
 756                         /* Update the field, and then ring the doorbell */
 757                 for(i=0;i<16;i++)
 758                         writeb( NodeName[i], &host->cfgtable->ServerName[i]);
 759
 760                 writel( CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
 761
 762                 for(i=0;i<MAX_IOCTL_CONFIG_WAIT;i++) {
 763                         if (!(readl(host->vaddr + SA5_DOORBELL)
 764                                         & CFGTBL_ChangeReq))
 765                                 break;
 766                         /* delay and try again */
 767                         udelay(1000);
 768                 }
 769                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 770                 if (i >= MAX_IOCTL_CONFIG_WAIT)
 771                         return -EAGAIN;
 772                 return(0);
 773         }
 774
 775         case CCISS_GETHEARTBEAT:
 776         {
 777                 Heartbeat_type heartbeat;
 778
 779                 if (!arg) return -EINVAL;
 780                 heartbeat = readl(&host->cfgtable->HeartBeat);
 781                 if (copy_to_user(argp, &heartbeat, sizeof( Heartbeat_type)))
 782                         return -EFAULT;
 783                 return(0);
 784         }
 785         case CCISS_GETBUSTYPES:
 786         {
 787                 BusTypes_type BusTypes;
 788
 789                 if (!arg) return -EINVAL;
 790                 BusTypes = readl(&host->cfgtable->BusTypes);
 791                 if (copy_to_user(argp, &BusTypes, sizeof( BusTypes_type) ))
 792                         return  -EFAULT;
 793                 return(0);
 794         }
 795         case CCISS_GETFIRMVER:
 796         {
 797                 FirmwareVer_type firmware;
 798
 799                 if (!arg) return -EINVAL;
 800                 memcpy(firmware, host->firm_ver, 4);
 801
 802                 if (copy_to_user(argp, firmware, sizeof( FirmwareVer_type)))
 803                         return -EFAULT;
 804                 return(0);
 805         }
 806         case CCISS_GETDRIVVER:
 807         {
 808                 DriverVer_type DriverVer = DRIVER_VERSION;
 809
 810                 if (!arg) return -EINVAL;
 811
 812                 if (copy_to_user(argp, &DriverVer, sizeof( DriverVer_type) ))
 813                         return -EFAULT;
 814                 return(0);
 815         }
 816
 817         case CCISS_REVALIDVOLS:
 818                 if (bdev != bdev->bd_contains || drv != host->drv)
 819                         return -ENXIO;
 820                 return revalidate_allvol(host);
 821
 822         case CCISS_GETLUNINFO: {
 823                 LogvolInfo_struct luninfo;
 824
 825                 luninfo.LunID = drv->LunID;
 826                 luninfo.num_opens = drv->usage_count;
 827                 luninfo.num_parts = 0;
 828                 if (copy_to_user(argp, &luninfo,
 829                                 sizeof(LogvolInfo_struct)))
 830                         return -EFAULT;
 831                 return(0);
 832         }
 833         case CCISS_DEREGDISK:
 834                 return rebuild_lun_table(host, disk);
 835
 836         case CCISS_REGNEWD:
 837                 return rebuild_lun_table(host, NULL);
 838
 839         case CCISS_PASSTHRU:
 840         {
 841                 IOCTL_Command_struct iocommand;
 842                 CommandList_struct *c;
 843                 char    *buff = NULL;
 844                 u64bit  temp64;
 845                 unsigned long flags;
 846                 DECLARE_COMPLETION(wait);
 847
 848                 if (!arg) return -EINVAL;
 849
 850                 if (!capable(CAP_SYS_RAWIO)) return -EPERM;
 851
 852                 if (copy_from_user(&iocommand, argp, sizeof( IOCTL_Command_struct) ))
 853                         return -EFAULT;
 854                 if((iocommand.buf_size < 1) &&
 855                                 (iocommand.Request.Type.Direction != XFER_NONE))
 856                 {
 857                         return -EINVAL;
 858                 }
 859 #if 0 /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
 860                 /* Check kmalloc limits */
 861                 if(iocommand.buf_size > 128000)
 862                         return -EINVAL;
 863 #endif
 864                 if(iocommand.buf_size > 0)
 865                 {
 866                         buff =  kmalloc(iocommand.buf_size, GFP_KERNEL);
 867                         if( buff == NULL)
 868                                 return -EFAULT;
 869                 }
 870                 if (iocommand.Request.Type.Direction == XFER_WRITE)
 871                 {
 872                         /* Copy the data into the buffer we created */
 873                         if (copy_from_user(buff, iocommand.buf, iocommand.buf_size))
 874                         {
 875                                 kfree(buff);
 876                                 return -EFAULT;
 877                         }
 878                 } else {
 879                         memset(buff, 0, iocommand.buf_size);
 880                 }
 881                 if ((c = cmd_alloc(host , 0)) == NULL)
 882                 {
 883                         kfree(buff);
 884                         return -ENOMEM;
 885                 }
 886                         // Fill in the command type
 887                 c->cmd_type = CMD_IOCTL_PEND;
 888                         // Fill in Command Header
 889                 c->Header.ReplyQueue = 0;  // unused in simple mode
 890                 if( iocommand.buf_size > 0)     // buffer to fill
 891                 {
 892                         c->Header.SGList = 1;
 893                         c->Header.SGTotal= 1;
 894                 } else  // no buffers to fill
 895                 {
 896                         c->Header.SGList = 0;
 897                         c->Header.SGTotal= 0;
 898                 }
 899                 c->Header.LUN = iocommand.LUN_info;
 900                 c->Header.Tag.lower = c->busaddr;  // use the kernel address the cmd block for tag
 901
 902                 // Fill in Request block
 903                 c->Request = iocommand.Request;
 904
 905                 // Fill in the scatter gather information
 906                 if (iocommand.buf_size > 0 )
 907                 {
 908                         temp64.val = pci_map_single( host->pdev, buff,
 909                                         iocommand.buf_size,
 910                                 PCI_DMA_BIDIRECTIONAL);
 911                         c->SG[0].Addr.lower = temp64.val32.lower;
 912                         c->SG[0].Addr.upper = temp64.val32.upper;
 913                         c->SG[0].Len = iocommand.buf_size;
 914                         c->SG[0].Ext = 0;  // we are not chaining
 915                 }
 916                 c->waiting = &wait;
 917
 918                 /* Put the request on the tail of the request queue */
 919                 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 920                 addQ(&host->reqQ, c);
 921                 host->Qdepth++;
 922                 start_io(host);
 923                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 924
 925                 wait_for_completion(&wait);
 926
 927                 /* unlock the buffers from DMA */
 928                 temp64.val32.lower = c->SG[0].Addr.lower;
 929                 temp64.val32.upper = c->SG[0].Addr.upper;
 930                 pci_unmap_single( host->pdev, (dma_addr_t) temp64.val,
 931                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
 932
 933                 /* Copy the error information out */
 934                 iocommand.error_info = *(c->err_info);
 935                 if ( copy_to_user(argp, &iocommand, sizeof( IOCTL_Command_struct) ) )
 936                 {
 937                         kfree(buff);
 938                         cmd_free(host, c, 0);
 939                         return( -EFAULT);
 940                 }
 941
 942                 if (iocommand.Request.Type.Direction == XFER_READ)
 943                 {
 944                         /* Copy the data out of the buffer we created */
 945                         if (copy_to_user(iocommand.buf, buff, iocommand.buf_size))
 946                         {
 947                                 kfree(buff);
 948                                 cmd_free(host, c, 0);
 949                                 return -EFAULT;
 950                         }
 951                 }
 952                 kfree(buff);
 953                 cmd_free(host, c, 0);
 954                 return(0);
 955         }
 956         case CCISS_BIG_PASSTHRU: {
 957                 BIG_IOCTL_Command_struct *ioc;
 958                 CommandList_struct *c;
 959                 unsigned char **buff = NULL;
 960                 int     *buff_size = NULL;
 961                 u64bit  temp64;
 962                 unsigned long flags;
 963                 BYTE sg_used = 0;
 964                 int status = 0;
 965                 int i;
 966                 DECLARE_COMPLETION(wait);
 967                 __u32   left;
 968                 __u32   sz;
 969                 BYTE    __user *data_ptr;
 970
 971                 if (!arg)
 972                         return -EINVAL;
 973                 if (!capable(CAP_SYS_RAWIO))
 974                         return -EPERM;
 975                 ioc = (BIG_IOCTL_Command_struct *)
 976                         kmalloc(sizeof(*ioc), GFP_KERNEL);
 977                 if (!ioc) {
 978                         status = -ENOMEM;
 979                         goto cleanup1;
 980                 }
 981                 if (copy_from_user(ioc, argp, sizeof(*ioc))) {
 982                         status = -EFAULT;
 983                         goto cleanup1;
 984                 }
 985                 if ((ioc->buf_size < 1) &&
 986                         (ioc->Request.Type.Direction != XFER_NONE)) {
 987                                 status = -EINVAL;
 988                                 goto cleanup1;
 989                 }
 990                 /* Check kmalloc limits  using all SGs */
 991                 if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
 992                         status = -EINVAL;
 993                         goto cleanup1;
 994                 }
 995                 if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
 996                         status = -EINVAL;
 997                         goto cleanup1;
 998                 }
 999                 buff = (unsigned char **) kmalloc(MAXSGENTRIES *
1000                                 sizeof(char *), GFP_KERNEL);
1001                 if (!buff) {
1002                         status = -ENOMEM;
1003                         goto cleanup1;
1004                 }
1005                 memset(buff, 0, MAXSGENTRIES);
1006                 buff_size = (int *) kmalloc(MAXSGENTRIES * sizeof(int),
1007                                         GFP_KERNEL);
1008                 if (!buff_size) {
1009                         status = -ENOMEM;
1010                         goto cleanup1;
1011                 }
1012                 left = ioc->buf_size;
1013                 data_ptr = ioc->buf;
1014                 while (left) {
1015                         sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1016                         buff_size[sg_used] = sz;
1017                         buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1018                         if (buff[sg_used] == NULL) {
1019                                 status = -ENOMEM;
1020                                 goto cleanup1;
1021                         }
1022                         if (ioc->Request.Type.Direction == XFER_WRITE &&
1023                                 copy_from_user(buff[sg_used], data_ptr, sz)) {
1024                                         status = -ENOMEM;
1025                                         goto cleanup1;
1026                         } else {
1027                                 memset(buff[sg_used], 0, sz);
1028                         }
1029                         left -= sz;
1030                         data_ptr += sz;
1031                         sg_used++;
1032                 }
1033                 if ((c = cmd_alloc(host , 0)) == NULL) {
1034                         status = -ENOMEM;
1035                         goto cleanup1;
1036                 }
1037                 c->cmd_type = CMD_IOCTL_PEND;
1038                 c->Header.ReplyQueue = 0;
1039
1040                 if( ioc->buf_size > 0) {
1041                         c->Header.SGList = sg_used;
1042                         c->Header.SGTotal= sg_used;
1043                 } else {
1044                         c->Header.SGList = 0;
1045                         c->Header.SGTotal= 0;
1046                 }
1047                 c->Header.LUN = ioc->LUN_info;
1048                 c->Header.Tag.lower = c->busaddr;
1049
1050                 c->Request = ioc->Request;
1051                 if (ioc->buf_size > 0 ) {
1052                         int i;
1053                         for(i=0; i<sg_used; i++) {
1054                                 temp64.val = pci_map_single( host->pdev, buff[i],
1055                                         buff_size[i],
1056                                         PCI_DMA_BIDIRECTIONAL);
1057                                 c->SG[i].Addr.lower = temp64.val32.lower;
1058                                 c->SG[i].Addr.upper = temp64.val32.upper;
1059                                 c->SG[i].Len = buff_size[i];
1060                                 c->SG[i].Ext = 0;  /* we are not chaining */
1061                         }
1062                 }
1063                 c->waiting = &wait;
1064                 /* Put the request on the tail of the request queue */
1065                 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1066                 addQ(&host->reqQ, c);
1067                 host->Qdepth++;
1068                 start_io(host);
1069                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1070                 wait_for_completion(&wait);
1071                 /* unlock the buffers from DMA */
1072                 for(i=0; i<sg_used; i++) {
1073                         temp64.val32.lower = c->SG[i].Addr.lower;
1074                         temp64.val32.upper = c->SG[i].Addr.upper;
1075                         pci_unmap_single( host->pdev, (dma_addr_t) temp64.val,
1076                                 buff_size[i], PCI_DMA_BIDIRECTIONAL);
1077                 }
1078                 /* Copy the error information out */
1079                 ioc->error_info = *(c->err_info);
1080                 if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1081                         cmd_free(host, c, 0);
1082                         status = -EFAULT;
1083                         goto cleanup1;
1084                 }
1085                 if (ioc->Request.Type.Direction == XFER_READ) {
1086                         /* Copy the data out of the buffer we created */
1087                         BYTE __user *ptr = ioc->buf;
1088                         for(i=0; i< sg_used; i++) {
1089                                 if (copy_to_user(ptr, buff[i], buff_size[i])) {
1090                                         cmd_free(host, c, 0);
1091                                         status = -EFAULT;
1092                                         goto cleanup1;
1093                                 }
1094                                 ptr += buff_size[i];
1095                         }
1096                 }
1097                 cmd_free(host, c, 0);
1098                 status = 0;
1099 cleanup1:
1100                 if (buff) {
1101                         for(i=0; i<sg_used; i++)
1102                                 if(buff[i] != NULL)
1103                                         kfree(buff[i]);
1104                         kfree(buff);
1105                 }
1106                 if (buff_size)
1107                         kfree(buff_size);
1108                 if (ioc)
1109                         kfree(ioc);
1110                 return(status);
1111         }
1112         default:
1113                 return -ENOTTY;
1114         }
1115
1116 }
1117
1118 /*
1119  * revalidate_allvol is for online array config utilities.  After a
1120  * utility reconfigures the drives in the array, it can use this function
1121  * (through an ioctl) to make the driver zap any previous disk structs for
1122  * that controller and get new ones.
1123  *
1124  * Right now I'm using the getgeometry() function to do this, but this
1125  * function should probably be finer grained and allow you to revalidate one
1126  * particualar logical volume (instead of all of them on a particular
1127  * controller).
1128  */
1129 static int revalidate_allvol(ctlr_info_t *host)
1130 {
1131         int ctlr = host->ctlr, i;
1132         unsigned long flags;
1133
1134         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1135         if (host->usage_count > 1) {
1136                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1137                 printk(KERN_WARNING "cciss: Device busy for volume"
1138                         " revalidation (usage=%d)\n", host->usage_count);
1139                 return -EBUSY;
1140         }
1141         host->usage_count++;
1142         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1143
1144         for(i=0; i< NWD; i++) {
1145                 struct gendisk *disk = host->gendisk[i];
1146                 if (disk->flags & GENHD_FL_UP)
1147                         del_gendisk(disk);
1148         }
1149
1150         /*
1151          * Set the partition and block size structures for all volumes
1152          * on this controller to zero.  We will reread all of this data
1153          */
1154         memset(host->drv,        0, sizeof(drive_info_struct)
1155                                                 * CISS_MAX_LUN);
1156         /*
1157          * Tell the array controller not to give us any interrupts while
1158          * we check the new geometry.  Then turn interrupts back on when
1159          * we're done.
1160          */
1161         host->access.set_intr_mask(host, CCISS_INTR_OFF);
1162         cciss_getgeometry(ctlr);
1163         host->access.set_intr_mask(host, CCISS_INTR_ON);
1164
1165         /* Loop through each real device */
1166         for (i = 0; i < NWD; i++) {
1167                 struct gendisk *disk = host->gendisk[i];
1168                 drive_info_struct *drv = &(host->drv[i]);
1169                 /* we must register the controller even if no disks exist */
1170                 /* this is for the online array utilities */
1171                 if (!drv->heads && i)
1172                         continue;
1173                 blk_queue_hardsect_size(drv->queue, drv->block_size);
1174                 set_capacity(disk, drv->nr_blocks);
1175                 add_disk(disk);
1176         }
1177         host->usage_count--;
1178         return 0;
1179 }
1180
1181 /* This function will check the usage_count of the drive to be updated/added.
1182  * If the usage_count is zero then the drive information will be updated and
1183  * the disk will be re-registered with the kernel.  If not then it will be
1184  * left alone for the next reboot.  The exception to this is disk 0 which
1185  * will always be left registered with the kernel since it is also the
1186  * controller node.  Any changes to disk 0 will show up on the next
1187  * reboot.
1188 */
1189 static void cciss_update_drive_info(int ctlr, int drv_index)
1190   {
1191         ctlr_info_t *h = hba[ctlr];
1192         struct gendisk *disk;
1193         ReadCapdata_struct *size_buff = NULL;
1194         InquiryData_struct *inq_buff = NULL;
1195         unsigned int block_size;
1196         unsigned int total_size;
1197         unsigned long flags = 0;
1198         int ret = 0;
1199
1200         /* if the disk already exists then deregister it before proceeding*/
1201         if (h->drv[drv_index].raid_level != -1){
1202                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1203                 h->drv[drv_index].busy_configuring = 1;
1204                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1205                 ret = deregister_disk(h->gendisk[drv_index],
1206                         &h->drv[drv_index], 0);
1207                 h->drv[drv_index].busy_configuring = 0;
1208         }
1209
1210         /* If the disk is in use return */
1211         if (ret)
1212                 return;
1213
1214
1215         /* Get information about the disk and modify the driver sturcture */
1216         size_buff = kmalloc(sizeof( ReadCapdata_struct), GFP_KERNEL);
1217         if (size_buff == NULL)
1218                 goto mem_msg;
1219         inq_buff = kmalloc(sizeof( InquiryData_struct), GFP_KERNEL);
1220         if (inq_buff == NULL)
1221                 goto mem_msg;
1222
1223         cciss_read_capacity(ctlr, drv_index, size_buff, 1,
1224                 &total_size, &block_size);
1225         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1226                 inq_buff, &h->drv[drv_index]);
1227
1228         ++h->num_luns;
1229         disk = h->gendisk[drv_index];
1230         set_capacity(disk, h->drv[drv_index].nr_blocks);
1231
1232
1233         /* if it's the controller it's already added */
1234         if (drv_index){
1235                 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1236
1237                 /* Set up queue information */
1238                 disk->queue->backing_dev_info.ra_pages = READ_AHEAD;
1239                 blk_queue_bounce_limit(disk->queue, hba[ctlr]->pdev->dma_mask);
1240
1241                 /* This is a hardware imposed limit. */
1242                 blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1243
1244                 /* This is a limit in the driver and could be eliminated. */
1245                 blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1246
1247                 blk_queue_max_sectors(disk->queue, 512);
1248
1249                 disk->queue->queuedata = hba[ctlr];
1250
1251                 blk_queue_hardsect_size(disk->queue,
1252                         hba[ctlr]->drv[drv_index].block_size);
1253
1254                 h->drv[drv_index].queue = disk->queue;
1255                 add_disk(disk);
1256         }
1257
1258 freeret:
1259         kfree(size_buff);
1260         kfree(inq_buff);
1261         return;
1262 mem_msg:
1263         printk(KERN_ERR "cciss: out of memory\n");
1264         goto freeret;
1265 }
1266
1267 /* This function will find the first index of the controllers drive array
1268  * that has a -1 for the raid_level and will return that index.  This is
1269  * where new drives will be added.  If the index to be returned is greater
1270  * than the highest_lun index for the controller then highest_lun is set
1271  * to this new index.  If there are no available indexes then -1 is returned.
1272 */
1273 static int cciss_find_free_drive_index(int ctlr)
1274 {
1275         int i;
1276
1277         for (i=0; i < CISS_MAX_LUN; i++){
1278                 if (hba[ctlr]->drv[i].raid_level == -1){
1279                         if (i > hba[ctlr]->highest_lun)
1280                                 hba[ctlr]->highest_lun = i;
1281                         return i;
1282                 }
1283         }
1284         return -1;
1285 }
1286
1287 /* This function will add and remove logical drives from the Logical
1288  * drive array of the controller and maintain persistancy of ordering
1289  * so that mount points are preserved until the next reboot.  This allows
1290  * for the removal of logical drives in the middle of the drive array
1291  * without a re-ordering of those drives.
1292  * INPUT
1293  * h            = The controller to perform the operations on
1294  * del_disk     = The disk to remove if specified.  If the value given
1295  *                is NULL then no disk is removed.
1296 */
1297 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk)
1298 {
1299         int ctlr = h->ctlr;
1300         int num_luns;
1301         ReportLunData_struct *ld_buff = NULL;
1302         drive_info_struct *drv = NULL;
1303         int return_code;
1304         int listlength = 0;
1305         int i;
1306         int drv_found;
1307         int drv_index = 0;
1308         __u32 lunid = 0;
1309         unsigned long flags;
1310
1311         /* Set busy_configuring flag for this operation */
1312         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1313         if (h->num_luns >= CISS_MAX_LUN){
1314                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1315                 return -EINVAL;
1316         }
1317
1318         if (h->busy_configuring){
1319                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1320                 return -EBUSY;
1321         }
1322         h->busy_configuring = 1;
1323
1324         /* if del_disk is NULL then we are being called to add a new disk
1325          * and update the logical drive table.  If it is not NULL then
1326          * we will check if the disk is in use or not.
1327          */
1328         if (del_disk != NULL){
1329                 drv = get_drv(del_disk);
1330                 drv->busy_configuring = 1;
1331                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1332                 return_code = deregister_disk(del_disk, drv, 1);
1333                 drv->busy_configuring = 0;
1334                 h->busy_configuring = 0;
1335                 return return_code;
1336         } else {
1337                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1338                 if (!capable(CAP_SYS_RAWIO))
1339                         return -EPERM;
1340
1341                 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1342                 if (ld_buff == NULL)
1343                         goto mem_msg;
1344
1345                 return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
1346                                 sizeof(ReportLunData_struct), 0, 0, 0,
1347                                 TYPE_CMD);
1348
1349                 if (return_code == IO_OK){
1350                         listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[0])) << 24;
1351                         listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[1])) << 16;
1352                         listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[2])) << 8;
1353                         listlength |= 0xff & (unsigned int)(ld_buff->LUNListLength[3]);
1354                 } else{ /* reading number of logical volumes failed */
1355                         printk(KERN_WARNING "cciss: report logical volume"
1356                                 " command failed\n");
1357                         listlength = 0;
1358                         goto freeret;
1359                 }
1360
1361                 num_luns = listlength / 8;      /* 8 bytes per entry */
1362                 if (num_luns > CISS_MAX_LUN){
1363                         num_luns = CISS_MAX_LUN;
1364                         printk(KERN_WARNING "cciss: more luns configured"
1365                                 " on controller than can be handled by"
1366                                 " this driver.\n");
1367                 }
1368
1369                 /* Compare controller drive array to drivers drive array.
1370                 * Check for updates in the drive information and any new drives
1371                 * on the controller.
1372                 */
1373                 for (i=0; i < num_luns; i++){
1374                         int j;
1375
1376                         drv_found = 0;
1377
1378                         lunid = (0xff &
1379                                 (unsigned int)(ld_buff->LUN[i][3])) << 24;
1380                         lunid |= (0xff &
1381                                 (unsigned int)(ld_buff->LUN[i][2])) << 16;
1382                         lunid |= (0xff &
1383                                 (unsigned int)(ld_buff->LUN[i][1])) << 8;
1384                         lunid |= 0xff &
1385                                 (unsigned int)(ld_buff->LUN[i][0]);
1386
1387                         /* Find if the LUN is already in the drive array
1388                          * of the controller.  If so then update its info
1389                          * if not is use.  If it does not exist then find
1390                          * the first free index and add it.
1391                         */
1392                         for (j=0; j <= h->highest_lun; j++){
1393                                 if (h->drv[j].LunID == lunid){
1394                                         drv_index = j;
1395                                         drv_found = 1;
1396                                 }
1397                         }
1398
1399                         /* check if the drive was found already in the array */
1400                         if (!drv_found){
1401                                 drv_index = cciss_find_free_drive_index(ctlr);
1402                                 if (drv_index == -1)
1403                                         goto freeret;
1404
1405                         }
1406                         h->drv[drv_index].LunID = lunid;
1407                         cciss_update_drive_info(ctlr, drv_index);
1408                 } /* end for */
1409         } /* end else */
1410
1411 freeret:
1412         kfree(ld_buff);
1413         h->busy_configuring = 0;
1414         /* We return -1 here to tell the ACU that we have registered/updated
1415          * all of the drives that we can and to keep it from calling us
1416          * additional times.
1417         */
1418         return -1;
1419 mem_msg:
1420         printk(KERN_ERR "cciss: out of memory\n");
1421         goto freeret;
1422 }
1423
1424 /* This function will deregister the disk and it's queue from the
1425  * kernel.  It must be called with the controller lock held and the
1426  * drv structures busy_configuring flag set.  It's parameters are:
1427  *
1428  * disk = This is the disk to be deregistered
1429  * drv  = This is the drive_info_struct associated with the disk to be
1430  *        deregistered.  It contains information about the disk used
1431  *        by the driver.
1432  * clear_all = This flag determines whether or not the disk information
1433  *             is going to be completely cleared out and the highest_lun
1434  *             reset.  Sometimes we want to clear out information about
1435  *             the disk in preperation for re-adding it.  In this case
1436  *             the highest_lun should be left unchanged and the LunID
1437  *             should not be cleared.
1438 */
1439 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
1440                            int clear_all)
1441 {
1442         ctlr_info_t *h = get_host(disk);
1443
1444         if (!capable(CAP_SYS_RAWIO))
1445                 return -EPERM;
1446
1447         /* make sure logical volume is NOT is use */
1448         if(clear_all || (h->gendisk[0] == disk)) {
1449         if (drv->usage_count > 1)
1450                 return -EBUSY;
1451         }
1452         else
1453                 if( drv->usage_count > 0 )
1454                         return -EBUSY;
1455
1456         /* invalidate the devices and deregister the disk.  If it is disk
1457          * zero do not deregister it but just zero out it's values.  This
1458          * allows us to delete disk zero but keep the controller registered.
1459         */
1460         if (h->gendisk[0] != disk){
1461                 if (disk->flags & GENHD_FL_UP){
1462                         blk_cleanup_queue(disk->queue);
1463                 del_gendisk(disk);
1464                         drv->queue = NULL;
1465                 }
1466         }
1467
1468         --h->num_luns;
1469         /* zero out the disk size info */
1470         drv->nr_blocks = 0;
1471         drv->block_size = 0;
1472         drv->heads = 0;
1473         drv->sectors = 0;
1474         drv->cylinders = 0;
1475         drv->raid_level = -1;   /* This can be used as a flag variable to
1476                                  * indicate that this element of the drive
1477                                  * array is free.
1478                                 */
1479
1480         if (clear_all){
1481         /* check to see if it was the last disk */
1482         if (drv == h->drv + h->highest_lun) {
1483                 /* if so, find the new hightest lun */
1484                 int i, newhighest =-1;
1485                 for(i=0; i<h->highest_lun; i++) {
1486                         /* if the disk has size > 0, it is available */
1487                                 if (h->drv[i].heads)
1488                                 newhighest = i;
1489                 }
1490                 h->highest_lun = newhighest;
1491         }
1492
1493         drv->LunID = 0;
1494         }
1495         return(0);
1496 }
1497
1498 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
1499         size_t size,
1500         unsigned int use_unit_num, /* 0: address the controller,
1501                                       1: address logical volume log_unit,
1502                                       2: periph device address is scsi3addr */
1503         unsigned int log_unit, __u8 page_code, unsigned char *scsi3addr,
1504         int cmd_type)
1505 {
1506         ctlr_info_t *h= hba[ctlr];
1507         u64bit buff_dma_handle;
1508         int status = IO_OK;
1509
1510         c->cmd_type = CMD_IOCTL_PEND;
1511         c->Header.ReplyQueue = 0;
1512         if( buff != NULL) {
1513                 c->Header.SGList = 1;
1514                 c->Header.SGTotal= 1;
1515         } else {
1516                 c->Header.SGList = 0;
1517                 c->Header.SGTotal= 0;
1518         }
1519         c->Header.Tag.lower = c->busaddr;
1520
1521         c->Request.Type.Type = cmd_type;
1522         if (cmd_type == TYPE_CMD) {
1523                 switch(cmd) {
1524                 case  CISS_INQUIRY:
1525                         /* If the logical unit number is 0 then, this is going
1526                         to controller so It's a physical command
1527                         mode = 0 target = 0.  So we have nothing to write.
1528                         otherwise, if use_unit_num == 1,
1529                         mode = 1(volume set addressing) target = LUNID
1530                         otherwise, if use_unit_num == 2,
1531                         mode = 0(periph dev addr) target = scsi3addr */
1532                         if (use_unit_num == 1) {
1533                                 c->Header.LUN.LogDev.VolId=
1534                                         h->drv[log_unit].LunID;
1535                                 c->Header.LUN.LogDev.Mode = 1;
1536                         } else if (use_unit_num == 2) {
1537                                 memcpy(c->Header.LUN.LunAddrBytes,scsi3addr,8);
1538                                 c->Header.LUN.LogDev.Mode = 0;
1539                         }
1540                         /* are we trying to read a vital product page */
1541                         if(page_code != 0) {
1542                                 c->Request.CDB[1] = 0x01;
1543                                 c->Request.CDB[2] = page_code;
1544                         }
1545                         c->Request.CDBLen = 6;
1546                         c->Request.Type.Attribute = ATTR_SIMPLE;
1547                         c->Request.Type.Direction = XFER_READ;
1548                         c->Request.Timeout = 0;
1549                         c->Request.CDB[0] =  CISS_INQUIRY;
1550                         c->Request.CDB[4] = size  & 0xFF;
1551                 break;
1552                 case CISS_REPORT_LOG:
1553                 case CISS_REPORT_PHYS:
1554                         /* Talking to controller so It's a physical command
1555                            mode = 00 target = 0.  Nothing to write.
1556                         */
1557                         c->Request.CDBLen = 12;
1558                         c->Request.Type.Attribute = ATTR_SIMPLE;
1559                         c->Request.Type.Direction = XFER_READ;
1560                         c->Request.Timeout = 0;
1561                         c->Request.CDB[0] = cmd;
1562                         c->Request.CDB[6] = (size >> 24) & 0xFF;  //MSB
1563                         c->Request.CDB[7] = (size >> 16) & 0xFF;
1564                         c->Request.CDB[8] = (size >> 8) & 0xFF;
1565                         c->Request.CDB[9] = size & 0xFF;
1566                         break;
1567
1568                 case CCISS_READ_CAPACITY:
1569                         c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1570                         c->Header.LUN.LogDev.Mode = 1;
1571                         c->Request.CDBLen = 10;
1572                         c->Request.Type.Attribute = ATTR_SIMPLE;
1573                         c->Request.Type.Direction = XFER_READ;
1574                         c->Request.Timeout = 0;
1575                         c->Request.CDB[0] = cmd;
1576                 break;
1577                 case CCISS_CACHE_FLUSH:
1578                         c->Request.CDBLen = 12;
1579                         c->Request.Type.Attribute = ATTR_SIMPLE;
1580                         c->Request.Type.Direction = XFER_WRITE;
1581                         c->Request.Timeout = 0;
1582                         c->Request.CDB[0] = BMIC_WRITE;
1583                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
1584                 break;
1585                 default:
1586                         printk(KERN_WARNING
1587                                 "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
1588                         return(IO_ERROR);
1589                 }
1590         } else if (cmd_type == TYPE_MSG) {
1591                 switch (cmd) {
1592                 case 3: /* No-Op message */
1593                         c->Request.CDBLen = 1;
1594                         c->Request.Type.Attribute = ATTR_SIMPLE;
1595                         c->Request.Type.Direction = XFER_WRITE;
1596                         c->Request.Timeout = 0;
1597                         c->Request.CDB[0] = cmd;
1598                         break;
1599                 default:
1600                         printk(KERN_WARNING
1601                                 "cciss%d: unknown message type %d\n",
1602                                 ctlr, cmd);
1603                         return IO_ERROR;
1604                 }
1605         } else {
1606                 printk(KERN_WARNING
1607                         "cciss%d: unknown command type %d\n", ctlr, cmd_type);
1608                 return IO_ERROR;
1609         }
1610         /* Fill in the scatter gather information */
1611         if (size > 0) {
1612                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
1613                         buff, size, PCI_DMA_BIDIRECTIONAL);
1614                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
1615                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
1616                 c->SG[0].Len = size;
1617                 c->SG[0].Ext = 0;  /* we are not chaining */
1618         }
1619         return status;
1620 }
1621 static int sendcmd_withirq(__u8 cmd,
1622         int     ctlr,
1623         void    *buff,
1624         size_t  size,
1625         unsigned int use_unit_num,
1626         unsigned int log_unit,
1627         __u8    page_code,
1628         int cmd_type)
1629 {
1630         ctlr_info_t *h = hba[ctlr];
1631         CommandList_struct *c;
1632         u64bit  buff_dma_handle;
1633         unsigned long flags;
1634         int return_status;
1635         DECLARE_COMPLETION(wait);
1636
1637         if ((c = cmd_alloc(h , 0)) == NULL)
1638                 return -ENOMEM;
1639         return_status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
1640                 log_unit, page_code, NULL, cmd_type);
1641         if (return_status != IO_OK) {
1642                 cmd_free(h, c, 0);
1643                 return return_status;
1644         }
1645 resend_cmd2:
1646         c->waiting = &wait;
1647
1648         /* Put the request on the tail of the queue and send it */
1649         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1650         addQ(&h->reqQ, c);
1651         h->Qdepth++;
1652         start_io(h);
1653         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1654
1655         wait_for_completion(&wait);
1656
1657         if(c->err_info->CommandStatus != 0)
1658         { /* an error has occurred */
1659                 switch(c->err_info->CommandStatus)
1660                 {
1661                         case CMD_TARGET_STATUS:
1662                                 printk(KERN_WARNING "cciss: cmd %p has "
1663                                         " completed with errors\n", c);
1664                                 if( c->err_info->ScsiStatus)
1665                                 {
1666                                         printk(KERN_WARNING "cciss: cmd %p "
1667                                         "has SCSI Status = %x\n",
1668                                                 c,
1669                                                 c->err_info->ScsiStatus);
1670                                 }
1671
1672                         break;
1673                         case CMD_DATA_UNDERRUN:
1674                         case CMD_DATA_OVERRUN:
1675                         /* expected for inquire and report lun commands */
1676                         break;
1677                         case CMD_INVALID:
1678                                 printk(KERN_WARNING "cciss: Cmd %p is "
1679                                         "reported invalid\n", c);
1680                                 return_status = IO_ERROR;
1681                         break;
1682                         case CMD_PROTOCOL_ERR:
1683                                 printk(KERN_WARNING "cciss: cmd %p has "
1684                                         "protocol error \n", c);
1685                                 return_status = IO_ERROR;
1686                         break;
1687 case CMD_HARDWARE_ERR:
1688                                 printk(KERN_WARNING "cciss: cmd %p had "
1689                                         " hardware error\n", c);
1690                                 return_status = IO_ERROR;
1691                         break;
1692                         case CMD_CONNECTION_LOST:
1693                                 printk(KERN_WARNING "cciss: cmd %p had "
1694                                         "connection lost\n", c);
1695                                 return_status = IO_ERROR;
1696                         break;
1697                         case CMD_ABORTED:
1698                                 printk(KERN_WARNING "cciss: cmd %p was "
1699                                         "aborted\n", c);
1700                                 return_status = IO_ERROR;
1701                         break;
1702                         case CMD_ABORT_FAILED:
1703                                 printk(KERN_WARNING "cciss: cmd %p reports "
1704                                         "abort failed\n", c);
1705                                 return_status = IO_ERROR;
1706                         break;
1707                         case CMD_UNSOLICITED_ABORT:
1708                                 printk(KERN_WARNING
1709                                         "cciss%d: unsolicited abort %p\n",
1710                                         ctlr, c);
1711                                 if (c->retry_count < MAX_CMD_RETRIES) {
1712                                         printk(KERN_WARNING
1713                                                 "cciss%d: retrying %p\n",
1714                                                 ctlr, c);
1715                                         c->retry_count++;
1716                                         /* erase the old error information */
1717                                         memset(c->err_info, 0,
1718                                                 sizeof(ErrorInfo_struct));
1719                                         return_status = IO_OK;
1720                                         INIT_COMPLETION(wait);
1721                                         goto resend_cmd2;
1722                                 }
1723                                 return_status = IO_ERROR;
1724                         break;
1725                         default:
1726                                 printk(KERN_WARNING "cciss: cmd %p returned "
1727                                         "unknown status %x\n", c,
1728                                                 c->err_info->CommandStatus);
1729                                 return_status = IO_ERROR;
1730                 }
1731         }
1732         /* unlock the buffers from DMA */
1733         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
1734         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
1735         pci_unmap_single( h->pdev, (dma_addr_t) buff_dma_handle.val,
1736                         c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
1737         cmd_free(h, c, 0);
1738         return(return_status);
1739
1740 }
1741 static void cciss_geometry_inquiry(int ctlr, int logvol,
1742                         int withirq, unsigned int total_size,
1743                         unsigned int block_size, InquiryData_struct *inq_buff,
1744                         drive_info_struct *drv)
1745 {
1746         int return_code;
1747         memset(inq_buff, 0, sizeof(InquiryData_struct));
1748         if (withirq)
1749                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
1750                         inq_buff, sizeof(*inq_buff), 1, logvol ,0xC1, TYPE_CMD);
1751         else
1752                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
1753                         sizeof(*inq_buff), 1, logvol ,0xC1, NULL, TYPE_CMD);
1754         if (return_code == IO_OK) {
1755                 if(inq_buff->data_byte[8] == 0xFF) {
1756                         printk(KERN_WARNING
1757                                 "cciss: reading geometry failed, volume "
1758                                 "does not support reading geometry\n");
1759                         drv->block_size = block_size;
1760                         drv->nr_blocks = total_size;
1761                         drv->heads = 255;
1762                         drv->sectors = 32; // Sectors per track
1763                         drv->cylinders = total_size / 255 / 32;
1764                 } else {
1765                         unsigned int t;
1766
1767                         drv->block_size = block_size;
1768                         drv->nr_blocks = total_size;
1769                         drv->heads = inq_buff->data_byte[6];
1770                         drv->sectors = inq_buff->data_byte[7];
1771                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
1772                         drv->cylinders += inq_buff->data_byte[5];
1773                         drv->raid_level = inq_buff->data_byte[8];
1774                         t = drv->heads * drv->sectors;
1775                         if (t > 1) {
1776                                 drv->cylinders = total_size/t;
1777                         }
1778                 }
1779         } else { /* Get geometry failed */
1780                 printk(KERN_WARNING "cciss: reading geometry failed\n");
1781         }
1782         printk(KERN_INFO "      heads= %d, sectors= %d, cylinders= %d\n\n",
1783                 drv->heads, drv->sectors, drv->cylinders);
1784 }
1785 static void
1786 cciss_read_capacity(int ctlr, int logvol, ReadCapdata_struct *buf,
1787                 int withirq, unsigned int *total_size, unsigned int *block_size)
1788 {
1789         int return_code;
1790         memset(buf, 0, sizeof(*buf));
1791         if (withirq)
1792                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
1793                         ctlr, buf, sizeof(*buf), 1, logvol, 0, TYPE_CMD);
1794         else
1795                 return_code = sendcmd(CCISS_READ_CAPACITY,
1796                         ctlr, buf, sizeof(*buf), 1, logvol, 0, NULL, TYPE_CMD);
1797         if (return_code == IO_OK) {
1798                 *total_size = be32_to_cpu(*((__be32 *) &buf->total_size[0]))+1;
1799                 *block_size = be32_to_cpu(*((__be32 *) &buf->block_size[0]));
1800         } else { /* read capacity command failed */
1801                 printk(KERN_WARNING "cciss: read capacity failed\n");
1802                 *total_size = 0;
1803                 *block_size = BLOCK_SIZE;
1804         }
1805         printk(KERN_INFO "      blocks= %u block_size= %d\n",
1806                 *total_size, *block_size);
1807         return;
1808 }
1809
1810 static int cciss_revalidate(struct gendisk *disk)
1811 {
1812         ctlr_info_t *h = get_host(disk);
1813         drive_info_struct *drv = get_drv(disk);
1814         int logvol;
1815         int FOUND=0;
1816         unsigned int block_size;
1817         unsigned int total_size;
1818         ReadCapdata_struct *size_buff = NULL;
1819         InquiryData_struct *inq_buff = NULL;
1820
1821         for(logvol=0; logvol < CISS_MAX_LUN; logvol++)
1822         {
1823                 if(h->drv[logvol].LunID == drv->LunID) {
1824                         FOUND=1;
1825                         break;
1826                 }
1827         }
1828
1829         if (!FOUND) return 1;
1830
1831         size_buff = kmalloc(sizeof( ReadCapdata_struct), GFP_KERNEL);
1832         if (size_buff == NULL)
1833         {
1834                 printk(KERN_WARNING "cciss: out of memory\n");
1835                 return 1;
1836         }
1837         inq_buff = kmalloc(sizeof( InquiryData_struct), GFP_KERNEL);
1838         if (inq_buff == NULL)
1839         {
1840                 printk(KERN_WARNING "cciss: out of memory\n");
1841                 kfree(size_buff);
1842                 return 1;
1843         }
1844
1845         cciss_read_capacity(h->ctlr, logvol, size_buff, 1, &total_size, &block_size);
1846         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size, inq_buff, drv);
1847
1848         blk_queue_hardsect_size(drv->queue, drv->block_size);
1849         set_capacity(disk, drv->nr_blocks);
1850
1851         kfree(size_buff);
1852         kfree(inq_buff);
1853         return 0;
1854 }
1855
1856 /*
1857  *   Wait polling for a command to complete.
1858  *   The memory mapped FIFO is polled for the completion.
1859  *   Used only at init time, interrupts from the HBA are disabled.
1860  */
1861 static unsigned long pollcomplete(int ctlr)
1862 {
1863         unsigned long done;
1864         int i;
1865
1866         /* Wait (up to 20 seconds) for a command to complete */
1867
1868         for (i = 20 * HZ; i > 0; i--) {
1869                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
1870                 if (done == FIFO_EMPTY)
1871                         schedule_timeout_uninterruptible(1);
1872                 else
1873                         return (done);
1874         }
1875         /* Invalid address to tell caller we ran out of time */
1876         return 1;
1877 }
1878 /*
1879  * Send a command to the controller, and wait for it to complete.
1880  * Only used at init time.
1881  */
1882 static int sendcmd(
1883         __u8    cmd,
1884         int     ctlr,
1885         void    *buff,
1886         size_t  size,
1887         unsigned int use_unit_num, /* 0: address the controller,
1888                                       1: address logical volume log_unit,
1889                                       2: periph device address is scsi3addr */
1890         unsigned int log_unit,
1891         __u8    page_code,
1892         unsigned char *scsi3addr,
1893         int cmd_type)
1894 {
1895         CommandList_struct *c;
1896         int i;
1897         unsigned long complete;
1898         ctlr_info_t *info_p= hba[ctlr];
1899         u64bit buff_dma_handle;
1900         int status;
1901
1902         if ((c = cmd_alloc(info_p, 1)) == NULL) {
1903                 printk(KERN_WARNING "cciss: unable to get memory");
1904                 return(IO_ERROR);
1905         }
1906         status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
1907                 log_unit, page_code, scsi3addr, cmd_type);
1908         if (status != IO_OK) {
1909                 cmd_free(info_p, c, 1);
1910                 return status;
1911         }
1912 resend_cmd1:
1913         /*
1914          * Disable interrupt
1915          */
1916 #ifdef CCISS_DEBUG
1917         printk(KERN_DEBUG "cciss: turning intr off\n");
1918 #endif /* CCISS_DEBUG */
1919         info_p->access.set_intr_mask(info_p, CCISS_INTR_OFF);
1920
1921         /* Make sure there is room in the command FIFO */
1922         /* Actually it should be completely empty at this time. */
1923         for (i = 200000; i > 0; i--)
1924         {
1925                 /* if fifo isn't full go */
1926                 if (!(info_p->access.fifo_full(info_p)))
1927                 {
1928
1929                         break;
1930                 }
1931                 udelay(10);
1932                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
1933                         " waiting!\n", ctlr);
1934         }
1935         /*
1936          * Send the cmd
1937          */
1938         info_p->access.submit_command(info_p, c);
1939         complete = pollcomplete(ctlr);
1940
1941 #ifdef CCISS_DEBUG
1942         printk(KERN_DEBUG "cciss: command completed\n");
1943 #endif /* CCISS_DEBUG */
1944
1945         if (complete != 1) {
1946                 if ( (complete & CISS_ERROR_BIT)
1947                      && (complete & ~CISS_ERROR_BIT) == c->busaddr)
1948                      {
1949                         /* if data overrun or underun on Report command
1950                                 ignore it
1951                         */
1952                         if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
1953                              (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
1954                              (c->Request.CDB[0] == CISS_INQUIRY)) &&
1955                                 ((c->err_info->CommandStatus ==
1956                                         CMD_DATA_OVERRUN) ||
1957                                  (c->err_info->CommandStatus ==
1958                                         CMD_DATA_UNDERRUN)
1959                                 ))
1960                         {
1961                                 complete = c->busaddr;
1962                         } else {
1963                                 if (c->err_info->CommandStatus ==
1964                                                 CMD_UNSOLICITED_ABORT) {
1965                                         printk(KERN_WARNING "cciss%d: "
1966                                                 "unsolicited abort %p\n",
1967                                                 ctlr, c);
1968                                         if (c->retry_count < MAX_CMD_RETRIES) {
1969                                                 printk(KERN_WARNING
1970                                                    "cciss%d: retrying %p\n",
1971                                                    ctlr, c);
1972                                                 c->retry_count++;
1973                                                 /* erase the old error */
1974                                                 /* information */
1975                                                 memset(c->err_info, 0,
1976                                                    sizeof(ErrorInfo_struct));
1977                                                 goto resend_cmd1;
1978                                         } else {
1979                                                 printk(KERN_WARNING
1980                                                    "cciss%d: retried %p too "
1981                                                    "many times\n", ctlr, c);
1982                                                 status = IO_ERROR;
1983                                                 goto cleanup1;
1984                                         }
1985                                 }
1986                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
1987                                 " Error %x \n", ctlr,
1988                                         c->err_info->CommandStatus);
1989                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
1990                                 " offensive info\n"
1991                                 "  size %x\n   num %x   value %x\n", ctlr,
1992                                   c->err_info->MoreErrInfo.Invalid_Cmd.offense_size,
1993                                   c->err_info->MoreErrInfo.Invalid_Cmd.offense_num,
1994                                   c->err_info->MoreErrInfo.Invalid_Cmd.offense_value);
1995                                 status = IO_ERROR;
1996                                 goto cleanup1;
1997                         }
1998                 }
1999                 if (complete != c->busaddr) {
2000                         printk( KERN_WARNING "cciss cciss%d: SendCmd "
2001                       "Invalid command list address returned! (%lx)\n",
2002                                 ctlr, complete);
2003                         status = IO_ERROR;
2004                         goto cleanup1;
2005                 }
2006         } else {
2007                 printk( KERN_WARNING
2008                         "cciss cciss%d: SendCmd Timeout out, "
2009                         "No command list address returned!\n",
2010                         ctlr);
2011                 status = IO_ERROR;
2012         }
2013
2014 cleanup1:
2015         /* unlock the data buffer from DMA */
2016         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2017         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2018         pci_unmap_single(info_p->pdev, (dma_addr_t) buff_dma_handle.val,
2019                                 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2020         cmd_free(info_p, c, 1);
2021         return (status);
2022 }
2023 /*
2024  * Map (physical) PCI mem into (virtual) kernel space
2025  */
2026 static void __iomem *remap_pci_mem(ulong base, ulong size)
2027 {
2028         ulong page_base        = ((ulong) base) & PAGE_MASK;
2029         ulong page_offs        = ((ulong) base) - page_base;
2030         void __iomem *page_remapped = ioremap(page_base, page_offs+size);
2031
2032         return page_remapped ? (page_remapped + page_offs) : NULL;
2033 }
2034
2035 /*
2036  * Takes jobs of the Q and sends them to the hardware, then puts it on
2037  * the Q to wait for completion.
2038  */
2039 static void start_io( ctlr_info_t *h)
2040 {
2041         CommandList_struct *c;
2042
2043         while(( c = h->reqQ) != NULL )
2044         {
2045                 /* can't do anything if fifo is full */
2046                 if ((h->access.fifo_full(h))) {
2047                         printk(KERN_WARNING "cciss: fifo full\n");
2048                         break;
2049                 }
2050
2051                 /* Get the frist entry from the Request Q */
2052                 removeQ(&(h->reqQ), c);
2053                 h->Qdepth--;
2054
2055                 /* Tell the controller execute command */
2056                 h->access.submit_command(h, c);
2057
2058                 /* Put job onto the completed Q */
2059                 addQ (&(h->cmpQ), c);
2060         }
2061 }
2062
2063 static inline void complete_buffers(struct bio *bio, int status)
2064 {
2065         while (bio) {
2066                 struct bio *xbh = bio->bi_next;
2067                 int nr_sectors = bio_sectors(bio);
2068
2069                 bio->bi_next = NULL;
2070                 blk_finished_io(len);
2071                 bio_endio(bio, nr_sectors << 9, status ? 0 : -EIO);
2072                 bio = xbh;
2073         }
2074
2075 }
2076 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2077 /* Zeros out the error record and then resends the command back */
2078 /* to the controller */
2079 static inline void resend_cciss_cmd( ctlr_info_t *h, CommandList_struct *c)
2080 {
2081         /* erase the old error information */
2082         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2083
2084         /* add it to software queue and then send it to the controller */
2085         addQ(&(h->reqQ),c);
2086         h->Qdepth++;
2087         if(h->Qdepth > h->maxQsinceinit)
2088                 h->maxQsinceinit = h->Qdepth;
2089
2090         start_io(h);
2091 }
2092 /* checks the status of the job and calls complete buffers to mark all
2093  * buffers for the completed job.
2094  */
2095 static inline void complete_command( ctlr_info_t *h, CommandList_struct *cmd,
2096                 int timeout)
2097 {
2098         int status = 1;
2099         int i;
2100         int retry_cmd = 0;
2101         u64bit temp64;
2102
2103         if (timeout)
2104                 status = 0;
2105
2106         if(cmd->err_info->CommandStatus != 0)
2107         { /* an error has occurred */
2108                 switch(cmd->err_info->CommandStatus)
2109                 {
2110                         unsigned char sense_key;
2111                         case CMD_TARGET_STATUS:
2112                                 status = 0;
2113
2114                                 if( cmd->err_info->ScsiStatus == 0x02)
2115                                 {
2116                                         printk(KERN_WARNING "cciss: cmd %p "
2117                                                 "has CHECK CONDITION "
2118                                                 " byte 2 = 0x%x\n", cmd,
2119                                                 cmd->err_info->SenseInfo[2]
2120                                         );
2121                                         /* check the sense key */
2122                                         sense_key = 0xf &
2123                                                 cmd->err_info->SenseInfo[2];
2124                                         /* no status or recovered error */
2125                                         if((sense_key == 0x0) ||
2126                                             (sense_key == 0x1))
2127                                         {
2128                                                         status = 1;
2129                                         }
2130                                 } else
2131                                 {
2132                                         printk(KERN_WARNING "cciss: cmd %p "
2133                                                 "has SCSI Status 0x%x\n",
2134                                                 cmd, cmd->err_info->ScsiStatus);
2135                                 }
2136                         break;
2137                         case CMD_DATA_UNDERRUN:
2138                                 printk(KERN_WARNING "cciss: cmd %p has"
2139                                         " completed with data underrun "
2140                                         "reported\n", cmd);
2141                         break;
2142                         case CMD_DATA_OVERRUN:
2143                                 printk(KERN_WARNING "cciss: cmd %p has"
2144                                         " completed with data overrun "
2145                                         "reported\n", cmd);
2146                         break;
2147                         case CMD_INVALID:
2148                                 printk(KERN_WARNING "cciss: cmd %p is "
2149                                         "reported invalid\n", cmd);
2150                                 status = 0;
2151                         break;
2152                         case CMD_PROTOCOL_ERR:
2153                                 printk(KERN_WARNING "cciss: cmd %p has "
2154                                         "protocol error \n", cmd);
2155                                 status = 0;
2156                         break;
2157                         case CMD_HARDWARE_ERR:
2158                                 printk(KERN_WARNING "cciss: cmd %p had "
2159                                         " hardware error\n", cmd);
2160                                 status = 0;
2161                         break;
2162                         case CMD_CONNECTION_LOST:
2163                                 printk(KERN_WARNING "cciss: cmd %p had "
2164                                         "connection lost\n", cmd);
2165                                 status=0;
2166                         break;
2167                         case CMD_ABORTED:
2168                                 printk(KERN_WARNING "cciss: cmd %p was "
2169                                         "aborted\n", cmd);
2170                                 status=0;
2171                         break;
2172                         case CMD_ABORT_FAILED:
2173                                 printk(KERN_WARNING "cciss: cmd %p reports "
2174                                         "abort failed\n", cmd);
2175                                 status=0;
2176                         break;
2177                         case CMD_UNSOLICITED_ABORT:
2178                                 printk(KERN_WARNING "cciss%d: unsolicited "
2179                                         "abort %p\n", h->ctlr, cmd);
2180                                 if (cmd->retry_count < MAX_CMD_RETRIES) {
2181                                         retry_cmd=1;
2182                                         printk(KERN_WARNING
2183                                                 "cciss%d: retrying %p\n",
2184                                                 h->ctlr, cmd);
2185                                         cmd->retry_count++;
2186                                 } else
2187                                         printk(KERN_WARNING
2188                                                 "cciss%d: %p retried too "
2189                                                 "many times\n", h->ctlr, cmd);
2190                                 status=0;
2191                         break;
2192                         case CMD_TIMEOUT:
2193                                 printk(KERN_WARNING "cciss: cmd %p timedout\n",
2194                                         cmd);
2195                                 status=0;
2196                         break;
2197                         default:
2198                                 printk(KERN_WARNING "cciss: cmd %p returned "
2199                                         "unknown status %x\n", cmd,
2200                                                 cmd->err_info->CommandStatus);
2201                                 status=0;
2202                 }
2203         }
2204         /* We need to return this command */
2205         if(retry_cmd) {
2206                 resend_cciss_cmd(h,cmd);
2207                 return;
2208         }
2209         /* command did not need to be retried */
2210         /* unmap the DMA mapping for all the scatter gather elements */
2211         for(i=0; i<cmd->Header.SGList; i++) {
2212                 temp64.val32.lower = cmd->SG[i].Addr.lower;
2213                 temp64.val32.upper = cmd->SG[i].Addr.upper;
2214                 pci_unmap_page(hba[cmd->ctlr]->pdev,
2215                         temp64.val, cmd->SG[i].Len,
2216                         (cmd->Request.Type.Direction == XFER_READ) ?
2217                                 PCI_DMA_FROMDEVICE : PCI_DMA_TODEVICE);
2218         }
2219         complete_buffers(cmd->rq->bio, status);
2220
2221 #ifdef CCISS_DEBUG
2222         printk("Done with %p\n", cmd->rq);
2223 #endif /* CCISS_DEBUG */
2224
2225         end_that_request_last(cmd->rq);
2226         cmd_free(h,cmd,1);
2227 }
2228
2229 /*
2230  * Get a request and submit it to the controller.
2231  */
2232 static void do_cciss_request(request_queue_t *q)
2233 {
2234         ctlr_info_t *h= q->queuedata;
2235         CommandList_struct *c;
2236         int start_blk, seg;
2237         struct request *creq;
2238         u64bit temp64;
2239         struct scatterlist tmp_sg[MAXSGENTRIES];
2240         drive_info_struct *drv;
2241         int i, dir;
2242
2243         /* We call start_io here in case there is a command waiting on the
2244          * queue that has not been sent.
2245         */
2246         if (blk_queue_plugged(q))
2247                 goto startio;
2248
2249 queue:
2250         creq = elv_next_request(q);
2251         if (!creq)
2252                 goto startio;
2253
2254         if (creq->nr_phys_segments > MAXSGENTRIES)
2255                 BUG();
2256
2257         if (( c = cmd_alloc(h, 1)) == NULL)
2258                 goto full;
2259
2260         blkdev_dequeue_request(creq);
2261
2262         spin_unlock_irq(q->queue_lock);
2263
2264         c->cmd_type = CMD_RWREQ;
2265         c->rq = creq;
2266
2267         /* fill in the request */
2268         drv = creq->rq_disk->private_data;
2269         c->Header.ReplyQueue = 0;  // unused in simple mode
2270         /* got command from pool, so use the command block index instead */
2271         /* for direct lookups. */
2272         /* The first 2 bits are reserved for controller error reporting. */
2273         c->Header.Tag.lower = (c->cmdindex << 3);
2274         c->Header.Tag.lower |= 0x04; /* flag for direct lookup. */
2275         c->Header.LUN.LogDev.VolId= drv->LunID;
2276         c->Header.LUN.LogDev.Mode = 1;
2277         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
2278         c->Request.Type.Type =  TYPE_CMD; // It is a command.
2279         c->Request.Type.Attribute = ATTR_SIMPLE;
2280         c->Request.Type.Direction =
2281                 (rq_data_dir(creq) == READ) ? XFER_READ: XFER_WRITE;
2282         c->Request.Timeout = 0; // Don't time out
2283         c->Request.CDB[0] = (rq_data_dir(creq) == READ) ? CCISS_READ : CCISS_WRITE;
2284         start_blk = creq->sector;
2285 #ifdef CCISS_DEBUG
2286         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",(int) creq->sector,
2287                 (int) creq->nr_sectors);
2288 #endif /* CCISS_DEBUG */
2289
2290         seg = blk_rq_map_sg(q, creq, tmp_sg);
2291
2292         /* get the DMA records for the setup */
2293         if (c->Request.Type.Direction == XFER_READ)
2294                 dir = PCI_DMA_FROMDEVICE;
2295         else
2296                 dir = PCI_DMA_TODEVICE;
2297
2298         for (i=0; i<seg; i++)
2299         {
2300                 c->SG[i].Len = tmp_sg[i].length;
2301                 temp64.val = (__u64) pci_map_page(h->pdev, tmp_sg[i].page,
2302                                           tmp_sg[i].offset, tmp_sg[i].length,
2303                                           dir);
2304                 c->SG[i].Addr.lower = temp64.val32.lower;
2305                 c->SG[i].Addr.upper = temp64.val32.upper;
2306                 c->SG[i].Ext = 0;  // we are not chaining
2307         }
2308         /* track how many SG entries we are using */
2309         if( seg > h->maxSG)
2310                 h->maxSG = seg;
2311
2312 #ifdef CCISS_DEBUG
2313         printk(KERN_DEBUG "cciss: Submitting %d sectors in %d segments\n", creq->nr_sectors, seg);
2314 #endif /* CCISS_DEBUG */
2315
2316         c->Header.SGList = c->Header.SGTotal = seg;
2317         c->Request.CDB[1]= 0;
2318         c->Request.CDB[2]= (start_blk >> 24) & 0xff;    //MSB
2319         c->Request.CDB[3]= (start_blk >> 16) & 0xff;
2320         c->Request.CDB[4]= (start_blk >>  8) & 0xff;
2321         c->Request.CDB[5]= start_blk & 0xff;
2322         c->Request.CDB[6]= 0; // (sect >> 24) & 0xff; MSB
2323         c->Request.CDB[7]= (creq->nr_sectors >>  8) & 0xff;
2324         c->Request.CDB[8]= creq->nr_sectors & 0xff;
2325         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
2326
2327         spin_lock_irq(q->queue_lock);
2328
2329         addQ(&(h->reqQ),c);
2330         h->Qdepth++;
2331         if(h->Qdepth > h->maxQsinceinit)
2332                 h->maxQsinceinit = h->Qdepth;
2333
2334         goto queue;
2335 full:
2336         blk_stop_queue(q);
2337 startio:
2338         /* We will already have the driver lock here so not need
2339          * to lock it.
2340         */
2341         start_io(h);
2342 }
2343
2344 static irqreturn_t do_cciss_intr(int irq, void *dev_id, struct pt_regs *regs)
2345 {
2346         ctlr_info_t *h = dev_id;
2347         CommandList_struct *c;
2348         unsigned long flags;
2349         __u32 a, a1, a2;
2350         int j;
2351         int start_queue = h->next_to_run;
2352
2353         /* Is this interrupt for us? */
2354         if (( h->access.intr_pending(h) == 0) || (h->interrupts_enabled == 0))
2355                 return IRQ_NONE;
2356
2357         /*
2358          * If there are completed commands in the completion queue,
2359          * we had better do something about it.
2360          */
2361         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2362         while( h->access.intr_pending(h))
2363         {
2364                 while((a = h->access.command_completed(h)) != FIFO_EMPTY)
2365                 {
2366                         a1 = a;
2367                         if ((a & 0x04)) {
2368                                 a2 = (a >> 3);
2369                                 if (a2 >= NR_CMDS) {
2370                                         printk(KERN_WARNING "cciss: controller cciss%d failed, stopping.\n", h->ctlr);
2371                                         fail_all_cmds(h->ctlr);
2372                                         return IRQ_HANDLED;
2373                                 }
2374
2375                                 c = h->cmd_pool + a2;
2376                                 a = c->busaddr;
2377
2378                         } else {
2379                         a &= ~3;
2380                                 if ((c = h->cmpQ) == NULL) {
2381                                         printk(KERN_WARNING "cciss: Completion of %08x ignored\n", a1);
2382                                 continue;
2383                         }
2384                         while(c->busaddr != a) {
2385                                 c = c->next;
2386                                 if (c == h->cmpQ)
2387                                         break;
2388                         }
2389                         }
2390                         /*
2391                          * If we've found the command, take it off the
2392                          * completion Q and free it
2393                          */
2394                          if (c->busaddr == a) {
2395                                 removeQ(&h->cmpQ, c);
2396                                 if (c->cmd_type == CMD_RWREQ) {
2397                                         complete_command(h, c, 0);
2398                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
2399                                         complete(c->waiting);
2400                                 }
2401 #                               ifdef CONFIG_CISS_SCSI_TAPE
2402                                 else if (c->cmd_type == CMD_SCSI)
2403                                         complete_scsi_command(c, 0, a1);
2404 #                               endif
2405                                 continue;
2406                         }
2407                 }
2408         }
2409
2410         /* check to see if we have maxed out the number of commands that can
2411          * be placed on the queue.  If so then exit.  We do this check here
2412          * in case the interrupt we serviced was from an ioctl and did not
2413          * free any new commands.
2414          */
2415         if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS)
2416                 goto cleanup;
2417
2418         /* We have room on the queue for more commands.  Now we need to queue
2419          * them up.  We will also keep track of the next queue to run so
2420          * that every queue gets a chance to be started first.
2421         */
2422         for (j=0; j < h->highest_lun + 1; j++){
2423                 int curr_queue = (start_queue + j) % (h->highest_lun + 1);
2424                 /* make sure the disk has been added and the drive is real
2425                  * because this can be called from the middle of init_one.
2426                 */
2427                 if(!(h->drv[curr_queue].queue) ||
2428                                    !(h->drv[curr_queue].heads))
2429                         continue;
2430                 blk_start_queue(h->gendisk[curr_queue]->queue);
2431
2432                 /* check to see if we have maxed out the number of commands
2433                  * that can be placed on the queue.
2434                 */
2435                 if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS)
2436                 {
2437                         if (curr_queue == start_queue){
2438                                 h->next_to_run = (start_queue + 1) % (h->highest_lun + 1);
2439                                 goto cleanup;
2440                         } else {
2441                                 h->next_to_run = curr_queue;
2442                                 goto cleanup;
2443         }
2444                 } else {
2445                         curr_queue = (curr_queue + 1) % (h->highest_lun + 1);
2446                 }
2447         }
2448
2449 cleanup:
2450         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2451         return IRQ_HANDLED;
2452 }
2453 /*
2454  *  We cannot read the structure directly, for portablity we must use
2455  *   the io functions.
2456  *   This is for debug only.
2457  */
2458 #ifdef CCISS_DEBUG
2459 static void print_cfg_table( CfgTable_struct *tb)
2460 {
2461         int i;
2462         char temp_name[17];
2463
2464         printk("Controller Configuration information\n");
2465         printk("------------------------------------\n");
2466         for(i=0;i<4;i++)
2467                 temp_name[i] = readb(&(tb->Signature[i]));
2468         temp_name[4]='\0';
2469         printk("   Signature = %s\n", temp_name);
2470         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
2471         printk("   Transport methods supported = 0x%x\n",
2472                                 readl(&(tb-> TransportSupport)));
2473         printk("   Transport methods active = 0x%x\n",
2474                                 readl(&(tb->TransportActive)));
2475         printk("   Requested transport Method = 0x%x\n",
2476                         readl(&(tb->HostWrite.TransportRequest)));
2477         printk("   Coalese Interrupt Delay = 0x%x\n",
2478                         readl(&(tb->HostWrite.CoalIntDelay)));
2479         printk("   Coalese Interrupt Count = 0x%x\n",
2480                         readl(&(tb->HostWrite.CoalIntCount)));
2481         printk("   Max outstanding commands = 0x%d\n",
2482                         readl(&(tb->CmdsOutMax)));
2483         printk("   Bus Types = 0x%x\n", readl(&(tb-> BusTypes)));
2484         for(i=0;i<16;i++)
2485                 temp_name[i] = readb(&(tb->ServerName[i]));
2486         temp_name[16] = '\0';
2487         printk("   Server Name = %s\n", temp_name);
2488         printk("   Heartbeat Counter = 0x%x\n\n\n",
2489                         readl(&(tb->HeartBeat)));
2490 }
2491 #endif /* CCISS_DEBUG */
2492
2493 static void release_io_mem(ctlr_info_t *c)
2494 {
2495         /* if IO mem was not protected do nothing */
2496         if( c->io_mem_addr == 0)
2497                 return;
2498         release_region(c->io_mem_addr, c->io_mem_length);
2499         c->io_mem_addr = 0;
2500         c->io_mem_length = 0;
2501 }
2502
2503 static int find_PCI_BAR_index(struct pci_dev *pdev,
2504                                 unsigned long pci_bar_addr)
2505 {
2506         int i, offset, mem_type, bar_type;
2507         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
2508                 return 0;
2509         offset = 0;
2510         for (i=0; i<DEVICE_COUNT_RESOURCE; i++) {
2511                 bar_type = pci_resource_flags(pdev, i) &
2512                         PCI_BASE_ADDRESS_SPACE;
2513                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
2514                         offset += 4;
2515                 else {
2516                         mem_type = pci_resource_flags(pdev, i) &
2517                                 PCI_BASE_ADDRESS_MEM_TYPE_MASK;
2518                         switch (mem_type) {
2519                                 case PCI_BASE_ADDRESS_MEM_TYPE_32:
2520                                 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
2521                                         offset += 4; /* 32 bit */
2522                                         break;
2523                                 case PCI_BASE_ADDRESS_MEM_TYPE_64:
2524                                         offset += 8;
2525                                         break;
2526                                 default: /* reserved in PCI 2.2 */
2527                                         printk(KERN_WARNING "Base address is invalid\n");
2528                                         return -1;
2529                                 break;
2530                         }
2531                 }
2532                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
2533                         return i+1;
2534         }
2535         return -1;
2536 }
2537
2538 static int cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
2539 {
2540         ushort subsystem_vendor_id, subsystem_device_id, command;
2541         __u32 board_id, scratchpad = 0;
2542         __u64 cfg_offset;
2543         __u32 cfg_base_addr;
2544         __u64 cfg_base_addr_index;
2545         int i;
2546
2547         /* check to see if controller has been disabled */
2548         /* BEFORE trying to enable it */
2549         (void) pci_read_config_word(pdev, PCI_COMMAND,&command);
2550         if(!(command & 0x02))
2551         {
2552                 printk(KERN_WARNING "cciss: controller appears to be disabled\n");
2553                 return(-1);
2554         }
2555
2556         if (pci_enable_device(pdev))
2557         {
2558                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
2559                 return( -1);
2560         }
2561
2562         subsystem_vendor_id = pdev->subsystem_vendor;
2563         subsystem_device_id = pdev->subsystem_device;
2564         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
2565                                         subsystem_vendor_id);
2566
2567         /* search for our IO range so we can protect it */
2568         for(i=0; i<DEVICE_COUNT_RESOURCE; i++)
2569         {
2570                 /* is this an IO range */
2571                 if( pci_resource_flags(pdev, i) & 0x01 ) {
2572                         c->io_mem_addr = pci_resource_start(pdev, i);
2573                         c->io_mem_length = pci_resource_end(pdev, i) -
2574                                 pci_resource_start(pdev, i) +1;
2575 #ifdef CCISS_DEBUG
2576                         printk("IO value found base_addr[%d] %lx %lx\n", i,
2577                                 c->io_mem_addr, c->io_mem_length);
2578 #endif /* CCISS_DEBUG */
2579                         /* register the IO range */
2580                         if(!request_region( c->io_mem_addr,
2581                                         c->io_mem_length, "cciss"))
2582                         {
2583                                 printk(KERN_WARNING "cciss I/O memory range already in use addr=%lx length=%ld\n",
2584                                 c->io_mem_addr, c->io_mem_length);
2585                                 c->io_mem_addr= 0;
2586                                 c->io_mem_length = 0;
2587                         }
2588                         break;
2589                 }
2590         }
2591
2592 #ifdef CCISS_DEBUG
2593         printk("command = %x\n", command);
2594         printk("irq = %x\n", pdev->irq);
2595         printk("board_id = %x\n", board_id);
2596 #endif /* CCISS_DEBUG */
2597
2598         c->intr = pdev->irq;
2599
2600         /*
2601          * Memory base addr is first addr , the second points to the config
2602          *   table
2603          */
2604
2605         c->paddr = pci_resource_start(pdev, 0); /* addressing mode bits already removed */
2606 #ifdef CCISS_DEBUG
2607         printk("address 0 = %x\n", c->paddr);
2608 #endif /* CCISS_DEBUG */
2609         c->vaddr = remap_pci_mem(c->paddr, 200);
2610
2611         /* Wait for the board to become ready.  (PCI hotplug needs this.)
2612          * We poll for up to 120 secs, once per 100ms. */
2613         for (i=0; i < 1200; i++) {
2614                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
2615                 if (scratchpad == CCISS_FIRMWARE_READY)
2616                         break;
2617                 set_current_state(TASK_INTERRUPTIBLE);
2618                 schedule_timeout(HZ / 10); /* wait 100ms */
2619         }
2620         if (scratchpad != CCISS_FIRMWARE_READY) {
2621                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
2622                 return -1;
2623         }
2624
2625         /* get the address index number */
2626         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
2627         cfg_base_addr &= (__u32) 0x0000ffff;
2628 #ifdef CCISS_DEBUG
2629         printk("cfg base address = %x\n", cfg_base_addr);
2630 #endif /* CCISS_DEBUG */
2631         cfg_base_addr_index =
2632                 find_PCI_BAR_index(pdev, cfg_base_addr);
2633 #ifdef CCISS_DEBUG
2634         printk("cfg base address index = %x\n", cfg_base_addr_index);
2635 #endif /* CCISS_DEBUG */
2636         if (cfg_base_addr_index == -1) {
2637                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
2638                 release_io_mem(c);
2639                 return -1;
2640         }
2641
2642         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
2643 #ifdef CCISS_DEBUG
2644         printk("cfg offset = %x\n", cfg_offset);
2645 #endif /* CCISS_DEBUG */
2646         c->cfgtable =  remap_pci_mem(pci_resource_start(pdev,
2647                                 cfg_base_addr_index) + cfg_offset,
2648                                 sizeof(CfgTable_struct));
2649         c->board_id = board_id;
2650
2651 #ifdef CCISS_DEBUG
2652         print_cfg_table(c->cfgtable);
2653 #endif /* CCISS_DEBUG */
2654
2655         for(i=0; i<NR_PRODUCTS; i++) {
2656                 if (board_id == products[i].board_id) {
2657                         c->product_name = products[i].product_name;
2658                         c->access = *(products[i].access);
2659                         break;
2660                 }
2661         }
2662         if (i == NR_PRODUCTS) {
2663                 printk(KERN_WARNING "cciss: Sorry, I don't know how"
2664                         " to access the Smart Array controller %08lx\n",
2665                                 (unsigned long)board_id);
2666                 return -1;
2667         }
2668         if (  (readb(&c->cfgtable->Signature[0]) != 'C') ||
2669               (readb(&c->cfgtable->Signature[1]) != 'I') ||
2670               (readb(&c->cfgtable->Signature[2]) != 'S') ||
2671               (readb(&c->cfgtable->Signature[3]) != 'S') )
2672         {
2673                 printk("Does not appear to be a valid CISS config table\n");
2674                 return -1;
2675         }
2676
2677 #ifdef CONFIG_X86
2678 {
2679         /* Need to enable prefetch in the SCSI core for 6400 in x86 */
2680         __u32 prefetch;
2681         prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
2682         prefetch |= 0x100;
2683         writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
2684 }
2685 #endif
2686
2687 #ifdef CCISS_DEBUG
2688         printk("Trying to put board into Simple mode\n");
2689 #endif /* CCISS_DEBUG */
2690         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
2691         /* Update the field, and then ring the doorbell */
2692         writel( CFGTBL_Trans_Simple,
2693                 &(c->cfgtable->HostWrite.TransportRequest));
2694         writel( CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
2695
2696         /* under certain very rare conditions, this can take awhile.
2697          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
2698          * as we enter this code.) */
2699         for(i=0;i<MAX_CONFIG_WAIT;i++) {
2700                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
2701                         break;
2702                 /* delay and try again */
2703                 set_current_state(TASK_INTERRUPTIBLE);
2704                 schedule_timeout(10);
2705         }
2706
2707 #ifdef CCISS_DEBUG
2708         printk(KERN_DEBUG "I counter got to %d %x\n", i, readl(c->vaddr + SA5_DOORBELL));
2709 #endif /* CCISS_DEBUG */
2710 #ifdef CCISS_DEBUG
2711         print_cfg_table(c->cfgtable);
2712 #endif /* CCISS_DEBUG */
2713
2714         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple))
2715         {
2716                 printk(KERN_WARNING "cciss: unable to get board into"
2717                                         " simple mode\n");
2718                 return -1;
2719         }
2720         return 0;
2721
2722 }
2723
2724 /*
2725  * Gets information about the local volumes attached to the controller.
2726  */
2727 static void cciss_getgeometry(int cntl_num)
2728 {
2729         ReportLunData_struct *ld_buff;
2730         ReadCapdata_struct *size_buff;
2731         InquiryData_struct *inq_buff;
2732         int return_code;
2733         int i;
2734         int listlength = 0;
2735         __u32 lunid = 0;
2736         int block_size;
2737         int total_size;
2738
2739         ld_buff = kmalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2740         if (ld_buff == NULL)
2741         {
2742                 printk(KERN_ERR "cciss: out of memory\n");
2743                 return;
2744         }
2745         memset(ld_buff, 0, sizeof(ReportLunData_struct));
2746         size_buff = kmalloc(sizeof( ReadCapdata_struct), GFP_KERNEL);
2747         if (size_buff == NULL)
2748         {
2749                 printk(KERN_ERR "cciss: out of memory\n");
2750                 kfree(ld_buff);
2751                 return;
2752         }
2753         inq_buff = kmalloc(sizeof( InquiryData_struct), GFP_KERNEL);
2754         if (inq_buff == NULL)
2755         {
2756                 printk(KERN_ERR "cciss: out of memory\n");
2757                 kfree(ld_buff);
2758                 kfree(size_buff);
2759                 return;
2760         }
2761         /* Get the firmware version */
2762         return_code = sendcmd(CISS_INQUIRY, cntl_num, inq_buff,
2763                 sizeof(InquiryData_struct), 0, 0 ,0, NULL, TYPE_CMD);
2764         if (return_code == IO_OK)
2765         {
2766                 hba[cntl_num]->firm_ver[0] = inq_buff->data_byte[32];
2767                 hba[cntl_num]->firm_ver[1] = inq_buff->data_byte[33];
2768                 hba[cntl_num]->firm_ver[2] = inq_buff->data_byte[34];
2769                 hba[cntl_num]->firm_ver[3] = inq_buff->data_byte[35];
2770         } else /* send command failed */
2771         {
2772                 printk(KERN_WARNING "cciss: unable to determine firmware"
2773                         " version of controller\n");
2774         }
2775         /* Get the number of logical volumes */
2776         return_code = sendcmd(CISS_REPORT_LOG, cntl_num, ld_buff,
2777                         sizeof(ReportLunData_struct), 0, 0, 0, NULL, TYPE_CMD);
2778
2779         if( return_code == IO_OK)
2780         {
2781 #ifdef CCISS_DEBUG
2782                 printk("LUN Data\n--------------------------\n");
2783 #endif /* CCISS_DEBUG */
2784
2785                 listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[0])) << 24;
2786                 listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[1])) << 16;
2787                 listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[2])) << 8;
2788                 listlength |= 0xff & (unsigned int)(ld_buff->LUNListLength[3]);
2789         } else /* reading number of logical volumes failed */
2790         {
2791                 printk(KERN_WARNING "cciss: report logical volume"
2792                         " command failed\n");
2793                 listlength = 0;
2794         }
2795         hba[cntl_num]->num_luns = listlength / 8; // 8 bytes pre entry
2796         if (hba[cntl_num]->num_luns > CISS_MAX_LUN)
2797         {
2798                 printk(KERN_ERR "ciss:  only %d number of logical volumes supported\n",
2799                         CISS_MAX_LUN);
2800                 hba[cntl_num]->num_luns = CISS_MAX_LUN;
2801         }
2802 #ifdef CCISS_DEBUG
2803         printk(KERN_DEBUG "Length = %x %x %x %x = %d\n", ld_buff->LUNListLength[0],
2804                 ld_buff->LUNListLength[1], ld_buff->LUNListLength[2],
2805                 ld_buff->LUNListLength[3],  hba[cntl_num]->num_luns);
2806 #endif /* CCISS_DEBUG */
2807
2808         hba[cntl_num]->highest_lun = hba[cntl_num]->num_luns-1;
2809 //      for(i=0; i<  hba[cntl_num]->num_luns; i++)
2810         for(i=0; i < CISS_MAX_LUN; i++)
2811         {
2812                 if (i < hba[cntl_num]->num_luns){
2813                         lunid = (0xff & (unsigned int)(ld_buff->LUN[i][3]))
2814                                  << 24;
2815                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][2]))
2816                                  << 16;
2817                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][1]))
2818                                  << 8;
2819                 lunid |= 0xff & (unsigned int)(ld_buff->LUN[i][0]);
2820
2821                 hba[cntl_num]->drv[i].LunID = lunid;
2822
2823
2824 #ifdef CCISS_DEBUG
2825                 printk(KERN_DEBUG "LUN[%d]:  %x %x %x %x = %x\n", i,
2826                         ld_buff->LUN[i][0], ld_buff->LUN[i][1],
2827                         ld_buff->LUN[i][2], ld_buff->LUN[i][3],
2828                         hba[cntl_num]->drv[i].LunID);
2829 #endif /* CCISS_DEBUG */
2830                 cciss_read_capacity(cntl_num, i, size_buff, 0,
2831                         &total_size, &block_size);
2832                         cciss_geometry_inquiry(cntl_num, i, 0, total_size,
2833                                 block_size, inq_buff, &hba[cntl_num]->drv[i]);
2834                 } else {
2835                         /* initialize raid_level to indicate a free space */
2836                         hba[cntl_num]->drv[i].raid_level = -1;
2837                 }
2838         }
2839         kfree(ld_buff);
2840         kfree(size_buff);
2841         kfree(inq_buff);
2842 }
2843
2844 /* Function to find the first free pointer into our hba[] array */
2845 /* Returns -1 if no free entries are left.  */
2846 static int alloc_cciss_hba(void)
2847 {
2848         struct gendisk *disk[NWD];
2849         int i, n;
2850         for (n = 0; n < NWD; n++) {
2851                 disk[n] = alloc_disk(1 << NWD_SHIFT);
2852                 if (!disk[n])
2853                         goto out;
2854         }
2855
2856         for(i=0; i< MAX_CTLR; i++) {
2857                 if (!hba[i]) {
2858                         ctlr_info_t *p;
2859                         p = kmalloc(sizeof(ctlr_info_t), GFP_KERNEL);
2860                         if (!p)
2861                                 goto Enomem;
2862                         memset(p, 0, sizeof(ctlr_info_t));
2863                         for (n = 0; n < NWD; n++)
2864                                 p->gendisk[n] = disk[n];
2865                         hba[i] = p;
2866                         return i;
2867                 }
2868         }
2869         printk(KERN_WARNING "cciss: This driver supports a maximum"
2870                 " of %d controllers.\n", MAX_CTLR);
2871         goto out;
2872 Enomem:
2873         printk(KERN_ERR "cciss: out of memory.\n");
2874 out:
2875         while (n--)
2876                 put_disk(disk[n]);
2877         return -1;
2878 }
2879
2880 static void free_hba(int i)
2881 {
2882         ctlr_info_t *p = hba[i];
2883         int n;
2884
2885         hba[i] = NULL;
2886         for (n = 0; n < NWD; n++)
2887                 put_disk(p->gendisk[n]);
2888         kfree(p);
2889 }
2890
2891 /*
2892  *  This is it.  Find all the controllers and register them.  I really hate
2893  *  stealing all these major device numbers.
2894  *  returns the number of block devices registered.
2895  */
2896 static int __devinit cciss_init_one(struct pci_dev *pdev,
2897         const struct pci_device_id *ent)
2898 {
2899         request_queue_t *q;
2900         int i;
2901         int j;
2902         int rc;
2903
2904         printk(KERN_DEBUG "cciss: Device 0x%x has been found at"
2905                         " bus %d dev %d func %d\n",
2906                 pdev->device, pdev->bus->number, PCI_SLOT(pdev->devfn),
2907                         PCI_FUNC(pdev->devfn));
2908         i = alloc_cciss_hba();
2909         if(i < 0)
2910                 return (-1);
2911
2912         hba[i]->busy_initializing = 1;
2913
2914         if (cciss_pci_init(hba[i], pdev) != 0)
2915                 goto clean1;
2916
2917         sprintf(hba[i]->devname, "cciss%d", i);
2918         hba[i]->ctlr = i;
2919         hba[i]->pdev = pdev;
2920
2921         /* configure PCI DMA stuff */
2922         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK))
2923                 printk("cciss: using DAC cycles\n");
2924         else if (!pci_set_dma_mask(pdev, DMA_32BIT_MASK))
2925                 printk("cciss: not using DAC cycles\n");
2926         else {
2927                 printk("cciss: no suitable DMA available\n");
2928                 goto clean1;
2929         }
2930
2931         /*
2932          * register with the major number, or get a dynamic major number
2933          * by passing 0 as argument.  This is done for greater than
2934          * 8 controller support.
2935          */
2936         if (i < MAX_CTLR_ORIG)
2937                 hba[i]->major = MAJOR_NR + i;
2938         rc = register_blkdev(hba[i]->major, hba[i]->devname);
2939         if(rc == -EBUSY || rc == -EINVAL) {
2940                 printk(KERN_ERR
2941                         "cciss:  Unable to get major number %d for %s "
2942                         "on hba %d\n", hba[i]->major, hba[i]->devname, i);
2943                 goto clean1;
2944         }
2945         else {
2946                 if (i >= MAX_CTLR_ORIG)
2947                         hba[i]->major = rc;
2948         }
2949
2950         /* make sure the board interrupts are off */
2951         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
2952         if( request_irq(hba[i]->intr, do_cciss_intr,
2953                 SA_INTERRUPT | SA_SHIRQ | SA_SAMPLE_RANDOM,
2954                         hba[i]->devname, hba[i])) {
2955                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
2956                         hba[i]->intr, hba[i]->devname);
2957                 goto clean2;
2958         }
2959         hba[i]->cmd_pool_bits = kmalloc(((NR_CMDS+BITS_PER_LONG-1)/BITS_PER_LONG)*sizeof(unsigned long), GFP_KERNEL);
2960         hba[i]->cmd_pool = (CommandList_struct *)pci_alloc_consistent(
2961                 hba[i]->pdev, NR_CMDS * sizeof(CommandList_struct),
2962                 &(hba[i]->cmd_pool_dhandle));
2963         hba[i]->errinfo_pool = (ErrorInfo_struct *)pci_alloc_consistent(
2964                 hba[i]->pdev, NR_CMDS * sizeof( ErrorInfo_struct),
2965                 &(hba[i]->errinfo_pool_dhandle));
2966         if((hba[i]->cmd_pool_bits == NULL)
2967                 || (hba[i]->cmd_pool == NULL)
2968                 || (hba[i]->errinfo_pool == NULL)) {
2969                 printk( KERN_ERR "cciss: out of memory");
2970                 goto clean4;
2971         }
2972
2973         spin_lock_init(&hba[i]->lock);
2974
2975         /* Initialize the pdev driver private data.
2976                 have it point to hba[i].  */
2977         pci_set_drvdata(pdev, hba[i]);
2978         /* command and error info recs zeroed out before
2979                         they are used */
2980         memset(hba[i]->cmd_pool_bits, 0, ((NR_CMDS+BITS_PER_LONG-1)/BITS_PER_LONG)*sizeof(unsigned long));
2981
2982 #ifdef CCISS_DEBUG
2983         printk(KERN_DEBUG "Scanning for drives on controller cciss%d\n",i);
2984 #endif /* CCISS_DEBUG */
2985
2986         cciss_getgeometry(i);
2987
2988         cciss_scsi_setup(i);
2989
2990         /* Turn the interrupts on so we can service requests */
2991         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
2992
2993         cciss_procinit(i);
2994
2995         for(j=0; j < NWD; j++) { /* mfm */
2996                 drive_info_struct *drv = &(hba[i]->drv[j]);
2997                 struct gendisk *disk = hba[i]->gendisk[j];
2998
2999                 q = blk_init_queue(do_cciss_request, &hba[i]->lock);
3000                 if (!q) {
3001                         printk(KERN_ERR
3002                            "cciss:  unable to allocate queue for disk %d\n",
3003                            j);
3004                         break;
3005                 }
3006                 drv->queue = q;
3007
3008                 q->backing_dev_info.ra_pages = READ_AHEAD;
3009         blk_queue_bounce_limit(q, hba[i]->pdev->dma_mask);
3010
3011         /* This is a hardware imposed limit. */
3012         blk_queue_max_hw_segments(q, MAXSGENTRIES);
3013
3014         /* This is a limit in the driver and could be eliminated. */
3015         blk_queue_max_phys_segments(q, MAXSGENTRIES);
3016
3017         blk_queue_max_sectors(q, 512);
3018
3019                 q->queuedata = hba[i];
3020                 sprintf(disk->disk_name, "cciss/c%dd%d", i, j);
3021                 sprintf(disk->devfs_name, "cciss/host%d/target%d", i, j);
3022                 disk->major = hba[i]->major;
3023                 disk->first_minor = j << NWD_SHIFT;
3024                 disk->fops = &cciss_fops;
3025                 disk->queue = q;
3026                 disk->private_data = drv;
3027                 /* we must register the controller even if no disks exist */
3028                 /* this is for the online array utilities */
3029                 if(!drv->heads && j)
3030                         continue;
3031                 blk_queue_hardsect_size(q, drv->block_size);
3032                 set_capacity(disk, drv->nr_blocks);
3033                 add_disk(disk);
3034         }
3035
3036         hba[i]->busy_initializing = 0;
3037         return(1);
3038
3039 clean4:
3040         if(hba[i]->cmd_pool_bits)
3041                 kfree(hba[i]->cmd_pool_bits);
3042         if(hba[i]->cmd_pool)
3043                 pci_free_consistent(hba[i]->pdev,
3044                         NR_CMDS * sizeof(CommandList_struct),
3045                         hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3046         if(hba[i]->errinfo_pool)
3047                 pci_free_consistent(hba[i]->pdev,
3048                         NR_CMDS * sizeof( ErrorInfo_struct),
3049                         hba[i]->errinfo_pool,
3050                         hba[i]->errinfo_pool_dhandle);
3051         free_irq(hba[i]->intr, hba[i]);
3052 clean2:
3053         unregister_blkdev(hba[i]->major, hba[i]->devname);
3054 clean1:
3055         release_io_mem(hba[i]);
3056         free_hba(i);
3057         hba[i]->busy_initializing = 0;
3058         return(-1);
3059 }
3060
3061 static void __devexit cciss_remove_one (struct pci_dev *pdev)
3062 {
3063         ctlr_info_t *tmp_ptr;
3064         int i, j;
3065         char flush_buf[4];
3066         int return_code;
3067
3068         if (pci_get_drvdata(pdev) == NULL)
3069         {
3070                 printk( KERN_ERR "cciss: Unable to remove device \n");
3071                 return;
3072         }
3073         tmp_ptr = pci_get_drvdata(pdev);
3074         i = tmp_ptr->ctlr;
3075         if (hba[i] == NULL)
3076         {
3077                 printk(KERN_ERR "cciss: device appears to "
3078                         "already be removed \n");
3079                 return;
3080         }
3081         /* Turn board interrupts off  and send the flush cache command */
3082         /* sendcmd will turn off interrupt, and send the flush...
3083         * To write all data in the battery backed cache to disks */
3084         memset(flush_buf, 0, 4);
3085         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0, 0, 0, NULL,
3086                                 TYPE_CMD);
3087         if(return_code != IO_OK)
3088         {
3089                 printk(KERN_WARNING "Error Flushing cache on controller %d\n",
3090                         i);
3091         }
3092         free_irq(hba[i]->intr, hba[i]);
3093         pci_set_drvdata(pdev, NULL);
3094         iounmap(hba[i]->vaddr);
3095         cciss_unregister_scsi(i);  /* unhook from SCSI subsystem */
3096         unregister_blkdev(hba[i]->major, hba[i]->devname);
3097         remove_proc_entry(hba[i]->devname, proc_cciss);
3098
3099         /* remove it from the disk list */
3100         for (j = 0; j < NWD; j++) {
3101                 struct gendisk *disk = hba[i]->gendisk[j];
3102                 if (disk->flags & GENHD_FL_UP) {
3103                         del_gendisk(disk);
3104                         blk_cleanup_queue(disk->queue);
3105                 }
3106         }
3107
3108         pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof(CommandList_struct),
3109                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3110         pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof( ErrorInfo_struct),
3111                 hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
3112         kfree(hba[i]->cmd_pool_bits);
3113         release_io_mem(hba[i]);
3114         free_hba(i);
3115 }
3116
3117 static struct pci_driver cciss_pci_driver = {
3118         .name =         "cciss",
3119         .probe =        cciss_init_one,
3120         .remove =       __devexit_p(cciss_remove_one),
3121         .id_table =     cciss_pci_device_id, /* id_table */
3122 };
3123
3124 /*
3125  *  This is it.  Register the PCI driver information for the cards we control
3126  *  the OS will call our registered routines when it finds one of our cards.
3127  */
3128 static int __init cciss_init(void)
3129 {
3130         printk(KERN_INFO DRIVER_NAME "\n");
3131
3132         /* Register for our PCI devices */
3133         return pci_module_init(&cciss_pci_driver);
3134 }
3135
3136 static void __exit cciss_cleanup(void)
3137 {
3138         int i;
3139
3140         pci_unregister_driver(&cciss_pci_driver);
3141         /* double check that all controller entrys have been removed */
3142         for (i=0; i< MAX_CTLR; i++)
3143         {
3144                 if (hba[i] != NULL)
3145                 {
3146                         printk(KERN_WARNING "cciss: had to remove"
3147                                         " controller %d\n", i);
3148                         cciss_remove_one(hba[i]->pdev);
3149                 }
3150         }
3151         remove_proc_entry("cciss", proc_root_driver);
3152 }
3153
3154 static void fail_all_cmds(unsigned long ctlr)
3155 {
3156         /* If we get here, the board is apparently dead. */
3157         ctlr_info_t *h = hba[ctlr];
3158         CommandList_struct *c;
3159         unsigned long flags;
3160
3161         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
3162         h->alive = 0;   /* the controller apparently died... */
3163
3164         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
3165
3166         pci_disable_device(h->pdev); /* Make sure it is really dead. */
3167
3168         /* move everything off the request queue onto the completed queue */
3169         while( (c = h->reqQ) != NULL ) {
3170                 removeQ(&(h->reqQ), c);
3171                 h->Qdepth--;
3172                 addQ (&(h->cmpQ), c);
3173         }
3174
3175         /* Now, fail everything on the completed queue with a HW error */
3176         while( (c = h->cmpQ) != NULL ) {
3177                 removeQ(&h->cmpQ, c);
3178                 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
3179                 if (c->cmd_type == CMD_RWREQ) {
3180                         complete_command(h, c, 0);
3181                 } else if (c->cmd_type == CMD_IOCTL_PEND)
3182                         complete(c->waiting);
3183 #ifdef CONFIG_CISS_SCSI_TAPE
3184                         else if (c->cmd_type == CMD_SCSI)
3185                                 complete_scsi_command(c, 0, 0);
3186 #endif
3187         }
3188         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
3189         return;
3190 }
3191
3192 module_init(cciss_init);
3193 module_exit(cciss_cleanup);