git.oblomov.eu Git - linux-2.6/blob - drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP SA 5xxx and 6xxx Controllers
   3  *    Copyright 2000, 2006 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 <linux/blktrace_api.h>
  42 #include <asm/uaccess.h>
  43 #include <asm/io.h>
  44
  45 #include <linux/dma-mapping.h>
  46 #include <linux/blkdev.h>
  47 #include <linux/genhd.h>
  48 #include <linux/completion.h>
  49
  50 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  51 #define DRIVER_NAME "HP CISS Driver (v 2.6.10)"
  52 #define DRIVER_VERSION CCISS_DRIVER_VERSION(2,6,10)
  53
  54 /* Embedded module documentation macros - see modules.h */
  55 MODULE_AUTHOR("Hewlett-Packard Company");
  56 MODULE_DESCRIPTION("Driver for HP Controller SA5xxx SA6xxx version 2.6.10");
  57 MODULE_SUPPORTED_DEVICE("HP SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
  58                         " SA6i P600 P800 P400 P400i E200 E200i");
  59 MODULE_LICENSE("GPL");
  60
  61 #include "cciss_cmd.h"
  62 #include "cciss.h"
  63 #include <linux/cciss_ioctl.h>
  64
  65 /* define the PCI info for the cards we can control */
  66 static const struct pci_device_id cciss_pci_device_id[] = {
  67         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,
  68                         0x0E11, 0x4070, 0, 0, 0},
  69         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB,
  70                         0x0E11, 0x4080, 0, 0, 0},
  71         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB,
  72                         0x0E11, 0x4082, 0, 0, 0},
  73         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB,
  74                         0x0E11, 0x4083, 0, 0, 0},
  75         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC,
  76                 0x0E11, 0x409A, 0, 0, 0},
  77         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC,
  78                 0x0E11, 0x409B, 0, 0, 0},
  79         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC,
  80                 0x0E11, 0x409C, 0, 0, 0},
  81         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC,
  82                 0x0E11, 0x409D, 0, 0, 0},
  83         { PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC,
  84                 0x0E11, 0x4091, 0, 0, 0},
  85         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSA,
  86                 0x103C, 0x3225, 0, 0, 0},
  87         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC,
  88                 0x103c, 0x3223, 0, 0, 0},
  89         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC,
  90                 0x103c, 0x3234, 0, 0, 0},
  91         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSC,
  92                 0x103c, 0x3235, 0, 0, 0},
  93         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD,
  94                 0x103c, 0x3211, 0, 0, 0},
  95         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD,
  96                 0x103c, 0x3212, 0, 0, 0},
  97         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD,
  98                 0x103c, 0x3213, 0, 0, 0},
  99         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD,
 100                 0x103c, 0x3214, 0, 0, 0},
 101         { PCI_VENDOR_ID_HP, PCI_DEVICE_ID_HP_CISSD,
 102                 0x103c, 0x3215, 0, 0, 0},
 103         {0,}
 104 };
 105 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 106
 107 #define NR_PRODUCTS ARRAY_SIZE(products)
 108
 109 /*  board_id = Subsystem Device ID & Vendor ID
 110  *  product = Marketing Name for the board
 111  *  access = Address of the struct of function pointers
 112  */
 113 static struct board_type products[] = {
 114         { 0x40700E11, "Smart Array 5300", &SA5_access },
 115         { 0x40800E11, "Smart Array 5i", &SA5B_access},
 116         { 0x40820E11, "Smart Array 532", &SA5B_access},
 117         { 0x40830E11, "Smart Array 5312", &SA5B_access},
 118         { 0x409A0E11, "Smart Array 641", &SA5_access},
 119         { 0x409B0E11, "Smart Array 642", &SA5_access},
 120         { 0x409C0E11, "Smart Array 6400", &SA5_access},
 121         { 0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 122         { 0x40910E11, "Smart Array 6i", &SA5_access},
 123         { 0x3225103C, "Smart Array P600", &SA5_access},
 124         { 0x3223103C, "Smart Array P800", &SA5_access},
 125         { 0x3234103C, "Smart Array P400", &SA5_access},
 126         { 0x3235103C, "Smart Array P400i", &SA5_access},
 127         { 0x3211103C, "Smart Array E200i", &SA5_access},
 128         { 0x3212103C, "Smart Array E200", &SA5_access},
 129         { 0x3213103C, "Smart Array E200i", &SA5_access},
 130         { 0x3214103C, "Smart Array E200i", &SA5_access},
 131         { 0x3215103C, "Smart Array E200i", &SA5_access},
 132 };
 133
 134 /* How long to wait (in millesconds) for board to go into simple mode */
 135 #define MAX_CONFIG_WAIT 30000
 136 #define MAX_IOCTL_CONFIG_WAIT 1000
 137
 138 /*define how many times we will try a command because of bus resets */
 139 #define MAX_CMD_RETRIES 3
 140
 141 #define READ_AHEAD       1024
 142 #define NR_CMDS          384 /* #commands that can be outstanding */
 143 #define MAX_CTLR        32
 144
 145 /* Originally cciss driver only supports 8 major numbers */
 146 #define MAX_CTLR_ORIG   8
 147
 148
 149 static ctlr_info_t *hba[MAX_CTLR];
 150
 151 static void do_cciss_request(request_queue_t *q);
 152 static irqreturn_t do_cciss_intr(int irq, void *dev_id, struct pt_regs *regs);
 153 static int cciss_open(struct inode *inode, struct file *filep);
 154 static int cciss_release(struct inode *inode, struct file *filep);
 155 static int cciss_ioctl(struct inode *inode, struct file *filep,
 156                 unsigned int cmd, unsigned long arg);
 157 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 158
 159 static int revalidate_allvol(ctlr_info_t *host);
 160 static int cciss_revalidate(struct gendisk *disk);
 161 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk);
 162 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv, int clear_all);
 163
 164 static void cciss_read_capacity(int ctlr, int logvol, ReadCapdata_struct *buf,
 165         int withirq, unsigned int *total_size, unsigned int *block_size);
 166 static void cciss_geometry_inquiry(int ctlr, int logvol,
 167                         int withirq, unsigned int total_size,
 168                         unsigned int block_size, InquiryData_struct *inq_buff,
 169                         drive_info_struct *drv);
 170 static void cciss_getgeometry(int cntl_num);
 171 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *, __u32);
 172 static void start_io( ctlr_info_t *h);
 173 static int sendcmd( __u8 cmd, int ctlr, void *buff, size_t size,
 174         unsigned int use_unit_num, unsigned int log_unit, __u8 page_code,
 175         unsigned char *scsi3addr, int cmd_type);
 176 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 177         unsigned int use_unit_num, unsigned int log_unit, __u8  page_code,
 178         int cmd_type);
 179
 180 static void fail_all_cmds(unsigned long ctlr);
 181
 182 #ifdef CONFIG_PROC_FS
 183 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
 184                 int length, int *eof, void *data);
 185 static void cciss_procinit(int i);
 186 #else
 187 static void cciss_procinit(int i) {}
 188 #endif /* CONFIG_PROC_FS */
 189
 190 #ifdef CONFIG_COMPAT
 191 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg);
 192 #endif
 193
 194 static struct block_device_operations cciss_fops  = {
 195         .owner          = THIS_MODULE,
 196         .open           = cciss_open,
 197         .release        = cciss_release,
 198         .ioctl          = cciss_ioctl,
 199         .getgeo         = cciss_getgeo,
 200 #ifdef CONFIG_COMPAT
 201         .compat_ioctl   = cciss_compat_ioctl,
 202 #endif
 203         .revalidate_disk= cciss_revalidate,
 204 };
 205
 206 /*
 207  * Enqueuing and dequeuing functions for cmdlists.
 208  */
 209 static inline void addQ(CommandList_struct **Qptr, CommandList_struct *c)
 210 {
 211         if (*Qptr == NULL) {
 212                 *Qptr = c;
 213                 c->next = c->prev = c;
 214         } else {
 215                 c->prev = (*Qptr)->prev;
 216                 c->next = (*Qptr);
 217                 (*Qptr)->prev->next = c;
 218                 (*Qptr)->prev = c;
 219         }
 220 }
 221
 222 static inline CommandList_struct *removeQ(CommandList_struct **Qptr,
 223                                                 CommandList_struct *c)
 224 {
 225         if (c && c->next != c) {
 226                 if (*Qptr == c) *Qptr = c->next;
 227                 c->prev->next = c->next;
 228                 c->next->prev = c->prev;
 229         } else {
 230                 *Qptr = NULL;
 231         }
 232         return c;
 233 }
 234
 235 #include "cciss_scsi.c"         /* For SCSI tape support */
 236
 237 #ifdef CONFIG_PROC_FS
 238
 239 /*
 240  * Report information about this controller.
 241  */
 242 #define ENG_GIG 1000000000
 243 #define ENG_GIG_FACTOR (ENG_GIG/512)
 244 #define RAID_UNKNOWN 6
 245 static const char *raid_label[] = {"0","4","1(1+0)","5","5+1","ADG",
 246                                            "UNKNOWN"};
 247
 248 static struct proc_dir_entry *proc_cciss;
 249
 250 static int cciss_proc_get_info(char *buffer, char **start, off_t offset,
 251                 int length, int *eof, void *data)
 252 {
 253         off_t pos = 0;
 254         off_t len = 0;
 255         int size, i, ctlr;
 256         ctlr_info_t *h = (ctlr_info_t*)data;
 257         drive_info_struct *drv;
 258         unsigned long flags;
 259         sector_t vol_sz, vol_sz_frac;
 260
 261         ctlr = h->ctlr;
 262
 263         /* prevent displaying bogus info during configuration
 264          * or deconfiguration of a logical volume
 265          */
 266         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 267         if (h->busy_configuring) {
 268                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 269         return -EBUSY;
 270         }
 271         h->busy_configuring = 1;
 272         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 273
 274         size = sprintf(buffer, "%s: HP %s Controller\n"
 275                 "Board ID: 0x%08lx\n"
 276                 "Firmware Version: %c%c%c%c\n"
 277                 "IRQ: %d\n"
 278                 "Logical drives: %d\n"
 279                 "Current Q depth: %d\n"
 280                 "Current # commands on controller: %d\n"
 281                 "Max Q depth since init: %d\n"
 282                 "Max # commands on controller since init: %d\n"
 283                 "Max SG entries since init: %d\n\n",
 284                 h->devname,
 285                 h->product_name,
 286                 (unsigned long)h->board_id,
 287                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2], h->firm_ver[3],
 288                 (unsigned int)h->intr[SIMPLE_MODE_INT],
 289                 h->num_luns,
 290                 h->Qdepth, h->commands_outstanding,
 291                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 292
 293         pos += size; len += size;
 294         cciss_proc_tape_report(ctlr, buffer, &pos, &len);
 295         for(i=0; i<=h->highest_lun; i++) {
 296
 297                 drv = &h->drv[i];
 298                 if (drv->heads == 0)
 299                         continue;
 300
 301                 vol_sz = drv->nr_blocks;
 302                 vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 303                 vol_sz_frac *= 100;
 304                 sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 305
 306                 if (drv->raid_level > 5)
 307                         drv->raid_level = RAID_UNKNOWN;
 308                 size = sprintf(buffer+len, "cciss/c%dd%d:"
 309                                 "\t%4u.%02uGB\tRAID %s\n",
 310                                 ctlr, i, (int)vol_sz, (int)vol_sz_frac,
 311                                 raid_label[drv->raid_level]);
 312                 pos += size; len += size;
 313         }
 314
 315         *eof = 1;
 316         *start = buffer+offset;
 317         len -= offset;
 318         if (len>length)
 319                 len = length;
 320         h->busy_configuring = 0;
 321         return len;
 322 }
 323
 324 static int
 325 cciss_proc_write(struct file *file, const char __user *buffer,
 326                         unsigned long count, void *data)
 327 {
 328         unsigned char cmd[80];
 329         int len;
 330 #ifdef CONFIG_CISS_SCSI_TAPE
 331         ctlr_info_t *h = (ctlr_info_t *) data;
 332         int rc;
 333 #endif
 334
 335         if (count > sizeof(cmd)-1) return -EINVAL;
 336         if (copy_from_user(cmd, buffer, count)) return -EFAULT;
 337         cmd[count] = '\0';
 338         len = strlen(cmd);      // above 3 lines ensure safety
 339         if (len && cmd[len-1] == '\n')
 340                 cmd[--len] = '\0';
 341 #       ifdef CONFIG_CISS_SCSI_TAPE
 342                 if (strcmp("engage scsi", cmd)==0) {
 343                         rc = cciss_engage_scsi(h->ctlr);
 344                         if (rc != 0) return -rc;
 345                         return count;
 346                 }
 347                 /* might be nice to have "disengage" too, but it's not
 348                    safely possible. (only 1 module use count, lock issues.) */
 349 #       endif
 350         return -EINVAL;
 351 }
 352
 353 /*
 354  * Get us a file in /proc/cciss that says something about each controller.
 355  * Create /proc/cciss if it doesn't exist yet.
 356  */
 357 static void __devinit cciss_procinit(int i)
 358 {
 359         struct proc_dir_entry *pde;
 360
 361         if (proc_cciss == NULL) {
 362                 proc_cciss = proc_mkdir("cciss", proc_root_driver);
 363                 if (!proc_cciss)
 364                         return;
 365         }
 366
 367         pde = create_proc_read_entry(hba[i]->devname,
 368                 S_IWUSR | S_IRUSR | S_IRGRP | S_IROTH,
 369                 proc_cciss, cciss_proc_get_info, hba[i]);
 370         pde->write_proc = cciss_proc_write;
 371 }
 372 #endif /* CONFIG_PROC_FS */
 373
 374 /*
 375  * For operations that cannot sleep, a command block is allocated at init,
 376  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 377  * which ones are free or in use.  For operations that can wait for kmalloc
 378  * to possible sleep, this routine can be called with get_from_pool set to 0.
 379  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 380  */
 381 static CommandList_struct * cmd_alloc(ctlr_info_t *h, int get_from_pool)
 382 {
 383         CommandList_struct *c;
 384         int i;
 385         u64bit temp64;
 386         dma_addr_t cmd_dma_handle, err_dma_handle;
 387
 388         if (!get_from_pool)
 389         {
 390                 c = (CommandList_struct *) pci_alloc_consistent(
 391                         h->pdev, sizeof(CommandList_struct), &cmd_dma_handle);
 392                 if(c==NULL)
 393                         return NULL;
 394                 memset(c, 0, sizeof(CommandList_struct));
 395
 396                 c->cmdindex = -1;
 397
 398                 c->err_info = (ErrorInfo_struct *)pci_alloc_consistent(
 399                                         h->pdev, sizeof(ErrorInfo_struct),
 400                                         &err_dma_handle);
 401
 402                 if (c->err_info == NULL)
 403                 {
 404                         pci_free_consistent(h->pdev,
 405                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 406                         return NULL;
 407                 }
 408                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 409         } else /* get it out of the controllers pool */
 410         {
 411                 do {
 412                         i = find_first_zero_bit(h->cmd_pool_bits, NR_CMDS);
 413                         if (i == NR_CMDS)
 414                                 return NULL;
 415                 } while(test_and_set_bit(i & (BITS_PER_LONG - 1), h->cmd_pool_bits+(i/BITS_PER_LONG)) != 0);
 416 #ifdef CCISS_DEBUG
 417                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 418 #endif
 419                 c = h->cmd_pool + i;
 420                 memset(c, 0, sizeof(CommandList_struct));
 421                 cmd_dma_handle = h->cmd_pool_dhandle
 422                                         + i*sizeof(CommandList_struct);
 423                 c->err_info = h->errinfo_pool + i;
 424                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 425                 err_dma_handle = h->errinfo_pool_dhandle
 426                                         + i*sizeof(ErrorInfo_struct);
 427                 h->nr_allocs++;
 428
 429                 c->cmdindex = i;
 430         }
 431
 432         c->busaddr = (__u32) cmd_dma_handle;
 433         temp64.val = (__u64) err_dma_handle;
 434         c->ErrDesc.Addr.lower = temp64.val32.lower;
 435         c->ErrDesc.Addr.upper = temp64.val32.upper;
 436         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 437
 438         c->ctlr = h->ctlr;
 439         return c;
 440
 441
 442 }
 443
 444 /*
 445  * Frees a command block that was previously allocated with cmd_alloc().
 446  */
 447 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 448 {
 449         int i;
 450         u64bit temp64;
 451
 452         if( !got_from_pool)
 453         {
 454                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 455                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 456                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 457                         c->err_info, (dma_addr_t) temp64.val);
 458                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 459                         c, (dma_addr_t) c->busaddr);
 460         } else
 461         {
 462                 i = c - h->cmd_pool;
 463                 clear_bit(i&(BITS_PER_LONG-1), h->cmd_pool_bits+(i/BITS_PER_LONG));
 464                 h->nr_frees++;
 465         }
 466 }
 467
 468 static inline ctlr_info_t *get_host(struct gendisk *disk)
 469 {
 470         return disk->queue->queuedata;
 471 }
 472
 473 static inline drive_info_struct *get_drv(struct gendisk *disk)
 474 {
 475         return disk->private_data;
 476 }
 477
 478 /*
 479  * Open.  Make sure the device is really there.
 480  */
 481 static int cciss_open(struct inode *inode, struct file *filep)
 482 {
 483         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
 484         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
 485
 486 #ifdef CCISS_DEBUG
 487         printk(KERN_DEBUG "cciss_open %s\n", inode->i_bdev->bd_disk->disk_name);
 488 #endif /* CCISS_DEBUG */
 489
 490         if (host->busy_initializing || drv->busy_configuring)
 491                 return -EBUSY;
 492         /*
 493          * Root is allowed to open raw volume zero even if it's not configured
 494          * so array config can still work. Root is also allowed to open any
 495          * volume that has a LUN ID, so it can issue IOCTL to reread the
 496          * disk information.  I don't think I really like this
 497          * but I'm already using way to many device nodes to claim another one
 498          * for "raw controller".
 499          */
 500         if (drv->nr_blocks == 0) {
 501                 if (iminor(inode) != 0) {       /* not node 0? */
 502                         /* if not node 0 make sure it is a partition = 0 */
 503                         if (iminor(inode) & 0x0f) {
 504                         return -ENXIO;
 505                                 /* if it is, make sure we have a LUN ID */
 506                         } else if (drv->LunID == 0) {
 507                                 return -ENXIO;
 508                         }
 509                 }
 510                 if (!capable(CAP_SYS_ADMIN))
 511                         return -EPERM;
 512         }
 513         drv->usage_count++;
 514         host->usage_count++;
 515         return 0;
 516 }
 517 /*
 518  * Close.  Sync first.
 519  */
 520 static int cciss_release(struct inode *inode, struct file *filep)
 521 {
 522         ctlr_info_t *host = get_host(inode->i_bdev->bd_disk);
 523         drive_info_struct *drv = get_drv(inode->i_bdev->bd_disk);
 524
 525 #ifdef CCISS_DEBUG
 526         printk(KERN_DEBUG "cciss_release %s\n", inode->i_bdev->bd_disk->disk_name);
 527 #endif /* CCISS_DEBUG */
 528
 529         drv->usage_count--;
 530         host->usage_count--;
 531         return 0;
 532 }
 533
 534 #ifdef CONFIG_COMPAT
 535
 536 static int do_ioctl(struct file *f, unsigned cmd, unsigned long arg)
 537 {
 538         int ret;
 539         lock_kernel();
 540         ret = cciss_ioctl(f->f_dentry->d_inode, f, cmd, arg);
 541         unlock_kernel();
 542         return ret;
 543 }
 544
 545 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd, unsigned long arg);
 546 static int cciss_ioctl32_big_passthru(struct file *f, unsigned cmd, unsigned long arg);
 547
 548 static long cciss_compat_ioctl(struct file *f, unsigned cmd, unsigned long arg)
 549 {
 550         switch (cmd) {
 551         case CCISS_GETPCIINFO:
 552         case CCISS_GETINTINFO:
 553         case CCISS_SETINTINFO:
 554         case CCISS_GETNODENAME:
 555         case CCISS_SETNODENAME:
 556         case CCISS_GETHEARTBEAT:
 557         case CCISS_GETBUSTYPES:
 558         case CCISS_GETFIRMVER:
 559         case CCISS_GETDRIVVER:
 560         case CCISS_REVALIDVOLS:
 561         case CCISS_DEREGDISK:
 562         case CCISS_REGNEWDISK:
 563         case CCISS_REGNEWD:
 564         case CCISS_RESCANDISK:
 565         case CCISS_GETLUNINFO:
 566                 return do_ioctl(f, cmd, arg);
 567
 568         case CCISS_PASSTHRU32:
 569                 return cciss_ioctl32_passthru(f, cmd, arg);
 570         case CCISS_BIG_PASSTHRU32:
 571                 return cciss_ioctl32_big_passthru(f, cmd, arg);
 572
 573         default:
 574                 return -ENOIOCTLCMD;
 575         }
 576 }
 577
 578 static int cciss_ioctl32_passthru(struct file *f, unsigned cmd, unsigned long arg)
 579 {
 580         IOCTL32_Command_struct __user *arg32 =
 581                 (IOCTL32_Command_struct __user *) arg;
 582         IOCTL_Command_struct arg64;
 583         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
 584         int err;
 585         u32 cp;
 586
 587         err = 0;
 588         err |= copy_from_user(&arg64.LUN_info, &arg32->LUN_info, sizeof(arg64.LUN_info));
 589         err |= copy_from_user(&arg64.Request, &arg32->Request, sizeof(arg64.Request));
 590         err |= copy_from_user(&arg64.error_info, &arg32->error_info, sizeof(arg64.error_info));
 591         err |= get_user(arg64.buf_size, &arg32->buf_size);
 592         err |= get_user(cp, &arg32->buf);
 593         arg64.buf = compat_ptr(cp);
 594         err |= copy_to_user(p, &arg64, sizeof(arg64));
 595
 596         if (err)
 597                 return -EFAULT;
 598
 599         err = do_ioctl(f, CCISS_PASSTHRU, (unsigned long) p);
 600         if (err)
 601                 return err;
 602         err |= copy_in_user(&arg32->error_info, &p->error_info, sizeof(arg32->error_info));
 603         if (err)
 604                 return -EFAULT;
 605         return err;
 606 }
 607
 608 static int cciss_ioctl32_big_passthru(struct file *file, unsigned cmd, unsigned long arg)
 609 {
 610         BIG_IOCTL32_Command_struct __user *arg32 =
 611                 (BIG_IOCTL32_Command_struct __user *) arg;
 612         BIG_IOCTL_Command_struct arg64;
 613         BIG_IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
 614         int err;
 615         u32 cp;
 616
 617         err = 0;
 618         err |= copy_from_user(&arg64.LUN_info, &arg32->LUN_info, sizeof(arg64.LUN_info));
 619         err |= copy_from_user(&arg64.Request, &arg32->Request, sizeof(arg64.Request));
 620         err |= copy_from_user(&arg64.error_info, &arg32->error_info, sizeof(arg64.error_info));
 621         err |= get_user(arg64.buf_size, &arg32->buf_size);
 622         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
 623         err |= get_user(cp, &arg32->buf);
 624         arg64.buf = compat_ptr(cp);
 625         err |= copy_to_user(p, &arg64, sizeof(arg64));
 626
 627         if (err)
 628                  return -EFAULT;
 629
 630         err = do_ioctl(file, CCISS_BIG_PASSTHRU, (unsigned long) p);
 631         if (err)
 632                 return err;
 633         err |= copy_in_user(&arg32->error_info, &p->error_info, sizeof(arg32->error_info));
 634         if (err)
 635                 return -EFAULT;
 636         return err;
 637 }
 638 #endif
 639
 640 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 641 {
 642         drive_info_struct *drv = get_drv(bdev->bd_disk);
 643
 644         if (!drv->cylinders)
 645                 return -ENXIO;
 646
 647         geo->heads = drv->heads;
 648         geo->sectors = drv->sectors;
 649         geo->cylinders = drv->cylinders;
 650         return 0;
 651 }
 652
 653 /*
 654  * ioctl
 655  */
 656 static int cciss_ioctl(struct inode *inode, struct file *filep,
 657                 unsigned int cmd, unsigned long arg)
 658 {
 659         struct block_device *bdev = inode->i_bdev;
 660         struct gendisk *disk = bdev->bd_disk;
 661         ctlr_info_t *host = get_host(disk);
 662         drive_info_struct *drv = get_drv(disk);
 663         int ctlr = host->ctlr;
 664         void __user *argp = (void __user *)arg;
 665
 666 #ifdef CCISS_DEBUG
 667         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
 668 #endif /* CCISS_DEBUG */
 669
 670         switch(cmd) {
 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 = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1000                 if (!buff) {
1001                         status = -ENOMEM;
1002                         goto cleanup1;
1003                 }
1004                 buff_size = (int *) kmalloc(MAXSGENTRIES * sizeof(int),
1005                                         GFP_KERNEL);
1006                 if (!buff_size) {
1007                         status = -ENOMEM;
1008                         goto cleanup1;
1009                 }
1010                 left = ioc->buf_size;
1011                 data_ptr = ioc->buf;
1012                 while (left) {
1013                         sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1014                         buff_size[sg_used] = sz;
1015                         buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1016                         if (buff[sg_used] == NULL) {
1017                                 status = -ENOMEM;
1018                                 goto cleanup1;
1019                         }
1020                         if (ioc->Request.Type.Direction == XFER_WRITE) {
1021                                 if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1022                                         status = -ENOMEM;
1023                                         goto cleanup1;
1024                                 }
1025                         } else {
1026                                 memset(buff[sg_used], 0, sz);
1027                         }
1028                         left -= sz;
1029                         data_ptr += sz;
1030                         sg_used++;
1031                 }
1032                 if ((c = cmd_alloc(host , 0)) == NULL) {
1033                         status = -ENOMEM;
1034                         goto cleanup1;
1035                 }
1036                 c->cmd_type = CMD_IOCTL_PEND;
1037                 c->Header.ReplyQueue = 0;
1038
1039                 if( ioc->buf_size > 0) {
1040                         c->Header.SGList = sg_used;
1041                         c->Header.SGTotal= sg_used;
1042                 } else {
1043                         c->Header.SGList = 0;
1044                         c->Header.SGTotal= 0;
1045                 }
1046                 c->Header.LUN = ioc->LUN_info;
1047                 c->Header.Tag.lower = c->busaddr;
1048
1049                 c->Request = ioc->Request;
1050                 if (ioc->buf_size > 0 ) {
1051                         int i;
1052                         for(i=0; i<sg_used; i++) {
1053                                 temp64.val = pci_map_single( host->pdev, buff[i],
1054                                         buff_size[i],
1055                                         PCI_DMA_BIDIRECTIONAL);
1056                                 c->SG[i].Addr.lower = temp64.val32.lower;
1057                                 c->SG[i].Addr.upper = temp64.val32.upper;
1058                                 c->SG[i].Len = buff_size[i];
1059                                 c->SG[i].Ext = 0;  /* we are not chaining */
1060                         }
1061                 }
1062                 c->waiting = &wait;
1063                 /* Put the request on the tail of the request queue */
1064                 spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1065                 addQ(&host->reqQ, c);
1066                 host->Qdepth++;
1067                 start_io(host);
1068                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1069                 wait_for_completion(&wait);
1070                 /* unlock the buffers from DMA */
1071                 for(i=0; i<sg_used; i++) {
1072                         temp64.val32.lower = c->SG[i].Addr.lower;
1073                         temp64.val32.upper = c->SG[i].Addr.upper;
1074                         pci_unmap_single( host->pdev, (dma_addr_t) temp64.val,
1075                                 buff_size[i], PCI_DMA_BIDIRECTIONAL);
1076                 }
1077                 /* Copy the error information out */
1078                 ioc->error_info = *(c->err_info);
1079                 if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1080                         cmd_free(host, c, 0);
1081                         status = -EFAULT;
1082                         goto cleanup1;
1083                 }
1084                 if (ioc->Request.Type.Direction == XFER_READ) {
1085                         /* Copy the data out of the buffer we created */
1086                         BYTE __user *ptr = ioc->buf;
1087                         for(i=0; i< sg_used; i++) {
1088                                 if (copy_to_user(ptr, buff[i], buff_size[i])) {
1089                                         cmd_free(host, c, 0);
1090                                         status = -EFAULT;
1091                                         goto cleanup1;
1092                                 }
1093                                 ptr += buff_size[i];
1094                         }
1095                 }
1096                 cmd_free(host, c, 0);
1097                 status = 0;
1098 cleanup1:
1099                 if (buff) {
1100                         for(i=0; i<sg_used; i++)
1101                                 kfree(buff[i]);
1102                         kfree(buff);
1103                 }
1104                 kfree(buff_size);
1105                 kfree(ioc);
1106                 return(status);
1107         }
1108         default:
1109                 return -ENOTTY;
1110         }
1111
1112 }
1113
1114 /*
1115  * revalidate_allvol is for online array config utilities.  After a
1116  * utility reconfigures the drives in the array, it can use this function
1117  * (through an ioctl) to make the driver zap any previous disk structs for
1118  * that controller and get new ones.
1119  *
1120  * Right now I'm using the getgeometry() function to do this, but this
1121  * function should probably be finer grained and allow you to revalidate one
1122  * particualar logical volume (instead of all of them on a particular
1123  * controller).
1124  */
1125 static int revalidate_allvol(ctlr_info_t *host)
1126 {
1127         int ctlr = host->ctlr, i;
1128         unsigned long flags;
1129
1130         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1131         if (host->usage_count > 1) {
1132                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1133                 printk(KERN_WARNING "cciss: Device busy for volume"
1134                         " revalidation (usage=%d)\n", host->usage_count);
1135                 return -EBUSY;
1136         }
1137         host->usage_count++;
1138         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1139
1140         for(i=0; i< NWD; i++) {
1141                 struct gendisk *disk = host->gendisk[i];
1142                 if (disk) {
1143                         request_queue_t *q = disk->queue;
1144
1145                         if (disk->flags & GENHD_FL_UP)
1146                                 del_gendisk(disk);
1147                         if (q)
1148                                 blk_cleanup_queue(q);
1149                 }
1150         }
1151
1152         /*
1153          * Set the partition and block size structures for all volumes
1154          * on this controller to zero.  We will reread all of this data
1155          */
1156         memset(host->drv,        0, sizeof(drive_info_struct)
1157                                                 * CISS_MAX_LUN);
1158         /*
1159          * Tell the array controller not to give us any interrupts while
1160          * we check the new geometry.  Then turn interrupts back on when
1161          * we're done.
1162          */
1163         host->access.set_intr_mask(host, CCISS_INTR_OFF);
1164         cciss_getgeometry(ctlr);
1165         host->access.set_intr_mask(host, CCISS_INTR_ON);
1166
1167         /* Loop through each real device */
1168         for (i = 0; i < NWD; i++) {
1169                 struct gendisk *disk = host->gendisk[i];
1170                 drive_info_struct *drv = &(host->drv[i]);
1171                 /* we must register the controller even if no disks exist */
1172                 /* this is for the online array utilities */
1173                 if (!drv->heads && i)
1174                         continue;
1175                 blk_queue_hardsect_size(drv->queue, drv->block_size);
1176                 set_capacity(disk, drv->nr_blocks);
1177                 add_disk(disk);
1178         }
1179         host->usage_count--;
1180         return 0;
1181 }
1182
1183 static inline void complete_buffers(struct bio *bio, int status)
1184 {
1185         while (bio) {
1186                 struct bio *xbh = bio->bi_next;
1187                 int nr_sectors = bio_sectors(bio);
1188
1189                 bio->bi_next = NULL;
1190                 blk_finished_io(len);
1191                 bio_endio(bio, nr_sectors << 9, status ? 0 : -EIO);
1192                 bio = xbh;
1193         }
1194
1195 }
1196
1197 static void cciss_softirq_done(struct request *rq)
1198 {
1199         CommandList_struct *cmd = rq->completion_data;
1200         ctlr_info_t *h = hba[cmd->ctlr];
1201         unsigned long flags;
1202         u64bit temp64;
1203         int i, ddir;
1204
1205         if (cmd->Request.Type.Direction == XFER_READ)
1206                 ddir = PCI_DMA_FROMDEVICE;
1207         else
1208                 ddir = PCI_DMA_TODEVICE;
1209
1210         /* command did not need to be retried */
1211         /* unmap the DMA mapping for all the scatter gather elements */
1212         for(i=0; i<cmd->Header.SGList; i++) {
1213                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1214                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1215                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1216         }
1217
1218         complete_buffers(rq->bio, rq->errors);
1219
1220 #ifdef CCISS_DEBUG
1221         printk("Done with %p\n", rq);
1222 #endif /* CCISS_DEBUG */
1223
1224         spin_lock_irqsave(&h->lock, flags);
1225         end_that_request_last(rq, rq->errors);
1226         cmd_free(h, cmd,1);
1227         spin_unlock_irqrestore(&h->lock, flags);
1228 }
1229
1230 /* This function will check the usage_count of the drive to be updated/added.
1231  * If the usage_count is zero then the drive information will be updated and
1232  * the disk will be re-registered with the kernel.  If not then it will be
1233  * left alone for the next reboot.  The exception to this is disk 0 which
1234  * will always be left registered with the kernel since it is also the
1235  * controller node.  Any changes to disk 0 will show up on the next
1236  * reboot.
1237 */
1238 static void cciss_update_drive_info(int ctlr, int drv_index)
1239   {
1240         ctlr_info_t *h = hba[ctlr];
1241         struct gendisk *disk;
1242         ReadCapdata_struct *size_buff = NULL;
1243         InquiryData_struct *inq_buff = NULL;
1244         unsigned int block_size;
1245         unsigned int total_size;
1246         unsigned long flags = 0;
1247         int ret = 0;
1248
1249         /* if the disk already exists then deregister it before proceeding*/
1250         if (h->drv[drv_index].raid_level != -1){
1251                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1252                 h->drv[drv_index].busy_configuring = 1;
1253                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1254                 ret = deregister_disk(h->gendisk[drv_index],
1255                         &h->drv[drv_index], 0);
1256                 h->drv[drv_index].busy_configuring = 0;
1257         }
1258
1259         /* If the disk is in use return */
1260         if (ret)
1261                 return;
1262
1263
1264         /* Get information about the disk and modify the driver sturcture */
1265         size_buff = kmalloc(sizeof( ReadCapdata_struct), GFP_KERNEL);
1266         if (size_buff == NULL)
1267                 goto mem_msg;
1268         inq_buff = kmalloc(sizeof( InquiryData_struct), GFP_KERNEL);
1269         if (inq_buff == NULL)
1270                 goto mem_msg;
1271
1272         cciss_read_capacity(ctlr, drv_index, size_buff, 1,
1273                 &total_size, &block_size);
1274         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1275                 inq_buff, &h->drv[drv_index]);
1276
1277         ++h->num_luns;
1278         disk = h->gendisk[drv_index];
1279         set_capacity(disk, h->drv[drv_index].nr_blocks);
1280
1281
1282         /* if it's the controller it's already added */
1283         if (drv_index){
1284                 disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1285
1286                 /* Set up queue information */
1287                 disk->queue->backing_dev_info.ra_pages = READ_AHEAD;
1288                 blk_queue_bounce_limit(disk->queue, hba[ctlr]->pdev->dma_mask);
1289
1290                 /* This is a hardware imposed limit. */
1291                 blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1292
1293                 /* This is a limit in the driver and could be eliminated. */
1294                 blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1295
1296                 blk_queue_max_sectors(disk->queue, 512);
1297
1298                 blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1299
1300                 disk->queue->queuedata = hba[ctlr];
1301
1302                 blk_queue_hardsect_size(disk->queue,
1303                         hba[ctlr]->drv[drv_index].block_size);
1304
1305                 h->drv[drv_index].queue = disk->queue;
1306                 add_disk(disk);
1307         }
1308
1309 freeret:
1310         kfree(size_buff);
1311         kfree(inq_buff);
1312         return;
1313 mem_msg:
1314         printk(KERN_ERR "cciss: out of memory\n");
1315         goto freeret;
1316 }
1317
1318 /* This function will find the first index of the controllers drive array
1319  * that has a -1 for the raid_level and will return that index.  This is
1320  * where new drives will be added.  If the index to be returned is greater
1321  * than the highest_lun index for the controller then highest_lun is set
1322  * to this new index.  If there are no available indexes then -1 is returned.
1323 */
1324 static int cciss_find_free_drive_index(int ctlr)
1325 {
1326         int i;
1327
1328         for (i=0; i < CISS_MAX_LUN; i++){
1329                 if (hba[ctlr]->drv[i].raid_level == -1){
1330                         if (i > hba[ctlr]->highest_lun)
1331                                 hba[ctlr]->highest_lun = i;
1332                         return i;
1333                 }
1334         }
1335         return -1;
1336 }
1337
1338 /* This function will add and remove logical drives from the Logical
1339  * drive array of the controller and maintain persistancy of ordering
1340  * so that mount points are preserved until the next reboot.  This allows
1341  * for the removal of logical drives in the middle of the drive array
1342  * without a re-ordering of those drives.
1343  * INPUT
1344  * h            = The controller to perform the operations on
1345  * del_disk     = The disk to remove if specified.  If the value given
1346  *                is NULL then no disk is removed.
1347 */
1348 static int rebuild_lun_table(ctlr_info_t *h, struct gendisk *del_disk)
1349 {
1350         int ctlr = h->ctlr;
1351         int num_luns;
1352         ReportLunData_struct *ld_buff = NULL;
1353         drive_info_struct *drv = NULL;
1354         int return_code;
1355         int listlength = 0;
1356         int i;
1357         int drv_found;
1358         int drv_index = 0;
1359         __u32 lunid = 0;
1360         unsigned long flags;
1361
1362         /* Set busy_configuring flag for this operation */
1363         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1364         if (h->num_luns >= CISS_MAX_LUN){
1365                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1366                 return -EINVAL;
1367         }
1368
1369         if (h->busy_configuring){
1370                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1371                 return -EBUSY;
1372         }
1373         h->busy_configuring = 1;
1374
1375         /* if del_disk is NULL then we are being called to add a new disk
1376          * and update the logical drive table.  If it is not NULL then
1377          * we will check if the disk is in use or not.
1378          */
1379         if (del_disk != NULL){
1380                 drv = get_drv(del_disk);
1381                 drv->busy_configuring = 1;
1382                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1383                 return_code = deregister_disk(del_disk, drv, 1);
1384                 drv->busy_configuring = 0;
1385                 h->busy_configuring = 0;
1386                 return return_code;
1387         } else {
1388                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1389                 if (!capable(CAP_SYS_RAWIO))
1390                         return -EPERM;
1391
1392                 ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
1393                 if (ld_buff == NULL)
1394                         goto mem_msg;
1395
1396                 return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
1397                                 sizeof(ReportLunData_struct), 0, 0, 0,
1398                                 TYPE_CMD);
1399
1400                 if (return_code == IO_OK){
1401                         listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[0])) << 24;
1402                         listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[1])) << 16;
1403                         listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[2])) << 8;
1404                         listlength |= 0xff & (unsigned int)(ld_buff->LUNListLength[3]);
1405                 } else{ /* reading number of logical volumes failed */
1406                         printk(KERN_WARNING "cciss: report logical volume"
1407                                 " command failed\n");
1408                         listlength = 0;
1409                         goto freeret;
1410                 }
1411
1412                 num_luns = listlength / 8;      /* 8 bytes per entry */
1413                 if (num_luns > CISS_MAX_LUN){
1414                         num_luns = CISS_MAX_LUN;
1415                         printk(KERN_WARNING "cciss: more luns configured"
1416                                 " on controller than can be handled by"
1417                                 " this driver.\n");
1418                 }
1419
1420                 /* Compare controller drive array to drivers drive array.
1421                 * Check for updates in the drive information and any new drives
1422                 * on the controller.
1423                 */
1424                 for (i=0; i < num_luns; i++){
1425                         int j;
1426
1427                         drv_found = 0;
1428
1429                         lunid = (0xff &
1430                                 (unsigned int)(ld_buff->LUN[i][3])) << 24;
1431                         lunid |= (0xff &
1432                                 (unsigned int)(ld_buff->LUN[i][2])) << 16;
1433                         lunid |= (0xff &
1434                                 (unsigned int)(ld_buff->LUN[i][1])) << 8;
1435                         lunid |= 0xff &
1436                                 (unsigned int)(ld_buff->LUN[i][0]);
1437
1438                         /* Find if the LUN is already in the drive array
1439                          * of the controller.  If so then update its info
1440                          * if not is use.  If it does not exist then find
1441                          * the first free index and add it.
1442                         */
1443                         for (j=0; j <= h->highest_lun; j++){
1444                                 if (h->drv[j].LunID == lunid){
1445                                         drv_index = j;
1446                                         drv_found = 1;
1447                                 }
1448                         }
1449
1450                         /* check if the drive was found already in the array */
1451                         if (!drv_found){
1452                                 drv_index = cciss_find_free_drive_index(ctlr);
1453                                 if (drv_index == -1)
1454                                         goto freeret;
1455
1456                         }
1457                         h->drv[drv_index].LunID = lunid;
1458                         cciss_update_drive_info(ctlr, drv_index);
1459                 } /* end for */
1460         } /* end else */
1461
1462 freeret:
1463         kfree(ld_buff);
1464         h->busy_configuring = 0;
1465         /* We return -1 here to tell the ACU that we have registered/updated
1466          * all of the drives that we can and to keep it from calling us
1467          * additional times.
1468         */
1469         return -1;
1470 mem_msg:
1471         printk(KERN_ERR "cciss: out of memory\n");
1472         goto freeret;
1473 }
1474
1475 /* This function will deregister the disk and it's queue from the
1476  * kernel.  It must be called with the controller lock held and the
1477  * drv structures busy_configuring flag set.  It's parameters are:
1478  *
1479  * disk = This is the disk to be deregistered
1480  * drv  = This is the drive_info_struct associated with the disk to be
1481  *        deregistered.  It contains information about the disk used
1482  *        by the driver.
1483  * clear_all = This flag determines whether or not the disk information
1484  *             is going to be completely cleared out and the highest_lun
1485  *             reset.  Sometimes we want to clear out information about
1486  *             the disk in preperation for re-adding it.  In this case
1487  *             the highest_lun should be left unchanged and the LunID
1488  *             should not be cleared.
1489 */
1490 static int deregister_disk(struct gendisk *disk, drive_info_struct *drv,
1491                            int clear_all)
1492 {
1493         ctlr_info_t *h = get_host(disk);
1494
1495         if (!capable(CAP_SYS_RAWIO))
1496                 return -EPERM;
1497
1498         /* make sure logical volume is NOT is use */
1499         if(clear_all || (h->gendisk[0] == disk)) {
1500         if (drv->usage_count > 1)
1501                 return -EBUSY;
1502         }
1503         else
1504                 if( drv->usage_count > 0 )
1505                         return -EBUSY;
1506
1507         /* invalidate the devices and deregister the disk.  If it is disk
1508          * zero do not deregister it but just zero out it's values.  This
1509          * allows us to delete disk zero but keep the controller registered.
1510         */
1511         if (h->gendisk[0] != disk){
1512                 if (disk) {
1513                         request_queue_t *q = disk->queue;
1514                         if (disk->flags & GENHD_FL_UP)
1515                                 del_gendisk(disk);
1516                         if (q) {
1517                                 blk_cleanup_queue(q);
1518                                 drv->queue = NULL;
1519                         }
1520                 }
1521         }
1522
1523         --h->num_luns;
1524         /* zero out the disk size info */
1525         drv->nr_blocks = 0;
1526         drv->block_size = 0;
1527         drv->heads = 0;
1528         drv->sectors = 0;
1529         drv->cylinders = 0;
1530         drv->raid_level = -1;   /* This can be used as a flag variable to
1531                                  * indicate that this element of the drive
1532                                  * array is free.
1533                                 */
1534
1535         if (clear_all){
1536         /* check to see if it was the last disk */
1537         if (drv == h->drv + h->highest_lun) {
1538                 /* if so, find the new hightest lun */
1539                 int i, newhighest =-1;
1540                 for(i=0; i<h->highest_lun; i++) {
1541                         /* if the disk has size > 0, it is available */
1542                                 if (h->drv[i].heads)
1543                                 newhighest = i;
1544                 }
1545                 h->highest_lun = newhighest;
1546         }
1547
1548         drv->LunID = 0;
1549         }
1550         return(0);
1551 }
1552
1553 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
1554         size_t size,
1555         unsigned int use_unit_num, /* 0: address the controller,
1556                                       1: address logical volume log_unit,
1557                                       2: periph device address is scsi3addr */
1558         unsigned int log_unit, __u8 page_code, unsigned char *scsi3addr,
1559         int cmd_type)
1560 {
1561         ctlr_info_t *h= hba[ctlr];
1562         u64bit buff_dma_handle;
1563         int status = IO_OK;
1564
1565         c->cmd_type = CMD_IOCTL_PEND;
1566         c->Header.ReplyQueue = 0;
1567         if( buff != NULL) {
1568                 c->Header.SGList = 1;
1569                 c->Header.SGTotal= 1;
1570         } else {
1571                 c->Header.SGList = 0;
1572                 c->Header.SGTotal= 0;
1573         }
1574         c->Header.Tag.lower = c->busaddr;
1575
1576         c->Request.Type.Type = cmd_type;
1577         if (cmd_type == TYPE_CMD) {
1578                 switch(cmd) {
1579                 case  CISS_INQUIRY:
1580                         /* If the logical unit number is 0 then, this is going
1581                         to controller so It's a physical command
1582                         mode = 0 target = 0.  So we have nothing to write.
1583                         otherwise, if use_unit_num == 1,
1584                         mode = 1(volume set addressing) target = LUNID
1585                         otherwise, if use_unit_num == 2,
1586                         mode = 0(periph dev addr) target = scsi3addr */
1587                         if (use_unit_num == 1) {
1588                                 c->Header.LUN.LogDev.VolId=
1589                                         h->drv[log_unit].LunID;
1590                                 c->Header.LUN.LogDev.Mode = 1;
1591                         } else if (use_unit_num == 2) {
1592                                 memcpy(c->Header.LUN.LunAddrBytes,scsi3addr,8);
1593                                 c->Header.LUN.LogDev.Mode = 0;
1594                         }
1595                         /* are we trying to read a vital product page */
1596                         if(page_code != 0) {
1597                                 c->Request.CDB[1] = 0x01;
1598                                 c->Request.CDB[2] = page_code;
1599                         }
1600                         c->Request.CDBLen = 6;
1601                         c->Request.Type.Attribute = ATTR_SIMPLE;
1602                         c->Request.Type.Direction = XFER_READ;
1603                         c->Request.Timeout = 0;
1604                         c->Request.CDB[0] =  CISS_INQUIRY;
1605                         c->Request.CDB[4] = size  & 0xFF;
1606                 break;
1607                 case CISS_REPORT_LOG:
1608                 case CISS_REPORT_PHYS:
1609                         /* Talking to controller so It's a physical command
1610                            mode = 00 target = 0.  Nothing to write.
1611                         */
1612                         c->Request.CDBLen = 12;
1613                         c->Request.Type.Attribute = ATTR_SIMPLE;
1614                         c->Request.Type.Direction = XFER_READ;
1615                         c->Request.Timeout = 0;
1616                         c->Request.CDB[0] = cmd;
1617                         c->Request.CDB[6] = (size >> 24) & 0xFF;  //MSB
1618                         c->Request.CDB[7] = (size >> 16) & 0xFF;
1619                         c->Request.CDB[8] = (size >> 8) & 0xFF;
1620                         c->Request.CDB[9] = size & 0xFF;
1621                         break;
1622
1623                 case CCISS_READ_CAPACITY:
1624                         c->Header.LUN.LogDev.VolId = h->drv[log_unit].LunID;
1625                         c->Header.LUN.LogDev.Mode = 1;
1626                         c->Request.CDBLen = 10;
1627                         c->Request.Type.Attribute = ATTR_SIMPLE;
1628                         c->Request.Type.Direction = XFER_READ;
1629                         c->Request.Timeout = 0;
1630                         c->Request.CDB[0] = cmd;
1631                 break;
1632                 case CCISS_CACHE_FLUSH:
1633                         c->Request.CDBLen = 12;
1634                         c->Request.Type.Attribute = ATTR_SIMPLE;
1635                         c->Request.Type.Direction = XFER_WRITE;
1636                         c->Request.Timeout = 0;
1637                         c->Request.CDB[0] = BMIC_WRITE;
1638                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
1639                 break;
1640                 default:
1641                         printk(KERN_WARNING
1642                                 "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
1643                         return(IO_ERROR);
1644                 }
1645         } else if (cmd_type == TYPE_MSG) {
1646                 switch (cmd) {
1647                 case 0: /* ABORT message */
1648                         c->Request.CDBLen = 12;
1649                         c->Request.Type.Attribute = ATTR_SIMPLE;
1650                         c->Request.Type.Direction = XFER_WRITE;
1651                         c->Request.Timeout = 0;
1652                         c->Request.CDB[0] = cmd; /* abort */
1653                         c->Request.CDB[1] = 0;   /* abort a command */
1654                         /* buff contains the tag of the command to abort */
1655                         memcpy(&c->Request.CDB[4], buff, 8);
1656                         break;
1657                 case 1: /* RESET message */
1658                         c->Request.CDBLen = 12;
1659                         c->Request.Type.Attribute = ATTR_SIMPLE;
1660                         c->Request.Type.Direction = XFER_WRITE;
1661                         c->Request.Timeout = 0;
1662                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
1663                         c->Request.CDB[0] = cmd;  /* reset */
1664                         c->Request.CDB[1] = 0x04; /* reset a LUN */
1665                 case 3: /* No-Op message */
1666                         c->Request.CDBLen = 1;
1667                         c->Request.Type.Attribute = ATTR_SIMPLE;
1668                         c->Request.Type.Direction = XFER_WRITE;
1669                         c->Request.Timeout = 0;
1670                         c->Request.CDB[0] = cmd;
1671                         break;
1672                 default:
1673                         printk(KERN_WARNING
1674                                 "cciss%d: unknown message type %d\n",
1675                                 ctlr, cmd);
1676                         return IO_ERROR;
1677                 }
1678         } else {
1679                 printk(KERN_WARNING
1680                         "cciss%d: unknown command type %d\n", ctlr, cmd_type);
1681                 return IO_ERROR;
1682         }
1683         /* Fill in the scatter gather information */
1684         if (size > 0) {
1685                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
1686                         buff, size, PCI_DMA_BIDIRECTIONAL);
1687                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
1688                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
1689                 c->SG[0].Len = size;
1690                 c->SG[0].Ext = 0;  /* we are not chaining */
1691         }
1692         return status;
1693 }
1694 static int sendcmd_withirq(__u8 cmd,
1695         int     ctlr,
1696         void    *buff,
1697         size_t  size,
1698         unsigned int use_unit_num,
1699         unsigned int log_unit,
1700         __u8    page_code,
1701         int cmd_type)
1702 {
1703         ctlr_info_t *h = hba[ctlr];
1704         CommandList_struct *c;
1705         u64bit  buff_dma_handle;
1706         unsigned long flags;
1707         int return_status;
1708         DECLARE_COMPLETION(wait);
1709
1710         if ((c = cmd_alloc(h , 0)) == NULL)
1711                 return -ENOMEM;
1712         return_status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
1713                 log_unit, page_code, NULL, cmd_type);
1714         if (return_status != IO_OK) {
1715                 cmd_free(h, c, 0);
1716                 return return_status;
1717         }
1718 resend_cmd2:
1719         c->waiting = &wait;
1720
1721         /* Put the request on the tail of the queue and send it */
1722         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1723         addQ(&h->reqQ, c);
1724         h->Qdepth++;
1725         start_io(h);
1726         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1727
1728         wait_for_completion(&wait);
1729
1730         if(c->err_info->CommandStatus != 0)
1731         { /* an error has occurred */
1732                 switch(c->err_info->CommandStatus)
1733                 {
1734                         case CMD_TARGET_STATUS:
1735                                 printk(KERN_WARNING "cciss: cmd %p has "
1736                                         " completed with errors\n", c);
1737                                 if( c->err_info->ScsiStatus)
1738                                 {
1739                                         printk(KERN_WARNING "cciss: cmd %p "
1740                                         "has SCSI Status = %x\n",
1741                                                 c,
1742                                                 c->err_info->ScsiStatus);
1743                                 }
1744
1745                         break;
1746                         case CMD_DATA_UNDERRUN:
1747                         case CMD_DATA_OVERRUN:
1748                         /* expected for inquire and report lun commands */
1749                         break;
1750                         case CMD_INVALID:
1751                                 printk(KERN_WARNING "cciss: Cmd %p is "
1752                                         "reported invalid\n", c);
1753                                 return_status = IO_ERROR;
1754                         break;
1755                         case CMD_PROTOCOL_ERR:
1756                                 printk(KERN_WARNING "cciss: cmd %p has "
1757                                         "protocol error \n", c);
1758                                 return_status = IO_ERROR;
1759                         break;
1760 case CMD_HARDWARE_ERR:
1761                                 printk(KERN_WARNING "cciss: cmd %p had "
1762                                         " hardware error\n", c);
1763                                 return_status = IO_ERROR;
1764                         break;
1765                         case CMD_CONNECTION_LOST:
1766                                 printk(KERN_WARNING "cciss: cmd %p had "
1767                                         "connection lost\n", c);
1768                                 return_status = IO_ERROR;
1769                         break;
1770                         case CMD_ABORTED:
1771                                 printk(KERN_WARNING "cciss: cmd %p was "
1772                                         "aborted\n", c);
1773                                 return_status = IO_ERROR;
1774                         break;
1775                         case CMD_ABORT_FAILED:
1776                                 printk(KERN_WARNING "cciss: cmd %p reports "
1777                                         "abort failed\n", c);
1778                                 return_status = IO_ERROR;
1779                         break;
1780                         case CMD_UNSOLICITED_ABORT:
1781                                 printk(KERN_WARNING
1782                                         "cciss%d: unsolicited abort %p\n",
1783                                         ctlr, c);
1784                                 if (c->retry_count < MAX_CMD_RETRIES) {
1785                                         printk(KERN_WARNING
1786                                                 "cciss%d: retrying %p\n",
1787                                                 ctlr, c);
1788                                         c->retry_count++;
1789                                         /* erase the old error information */
1790                                         memset(c->err_info, 0,
1791                                                 sizeof(ErrorInfo_struct));
1792                                         return_status = IO_OK;
1793                                         INIT_COMPLETION(wait);
1794                                         goto resend_cmd2;
1795                                 }
1796                                 return_status = IO_ERROR;
1797                         break;
1798                         default:
1799                                 printk(KERN_WARNING "cciss: cmd %p returned "
1800                                         "unknown status %x\n", c,
1801                                                 c->err_info->CommandStatus);
1802                                 return_status = IO_ERROR;
1803                 }
1804         }
1805         /* unlock the buffers from DMA */
1806         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
1807         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
1808         pci_unmap_single( h->pdev, (dma_addr_t) buff_dma_handle.val,
1809                         c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
1810         cmd_free(h, c, 0);
1811         return(return_status);
1812
1813 }
1814 static void cciss_geometry_inquiry(int ctlr, int logvol,
1815                         int withirq, unsigned int total_size,
1816                         unsigned int block_size, InquiryData_struct *inq_buff,
1817                         drive_info_struct *drv)
1818 {
1819         int return_code;
1820         memset(inq_buff, 0, sizeof(InquiryData_struct));
1821         if (withirq)
1822                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
1823                         inq_buff, sizeof(*inq_buff), 1, logvol ,0xC1, TYPE_CMD);
1824         else
1825                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
1826                         sizeof(*inq_buff), 1, logvol ,0xC1, NULL, TYPE_CMD);
1827         if (return_code == IO_OK) {
1828                 if(inq_buff->data_byte[8] == 0xFF) {
1829                         printk(KERN_WARNING
1830                                 "cciss: reading geometry failed, volume "
1831                                 "does not support reading geometry\n");
1832                         drv->block_size = block_size;
1833                         drv->nr_blocks = total_size;
1834                         drv->heads = 255;
1835                         drv->sectors = 32; // Sectors per track
1836                         drv->cylinders = total_size / 255 / 32;
1837                 } else {
1838                         unsigned int t;
1839
1840                         drv->block_size = block_size;
1841                         drv->nr_blocks = total_size;
1842                         drv->heads = inq_buff->data_byte[6];
1843                         drv->sectors = inq_buff->data_byte[7];
1844                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
1845                         drv->cylinders += inq_buff->data_byte[5];
1846                         drv->raid_level = inq_buff->data_byte[8];
1847                         t = drv->heads * drv->sectors;
1848                         if (t > 1) {
1849                                 drv->cylinders = total_size/t;
1850                         }
1851                 }
1852         } else { /* Get geometry failed */
1853                 printk(KERN_WARNING "cciss: reading geometry failed\n");
1854         }
1855         printk(KERN_INFO "      heads= %d, sectors= %d, cylinders= %d\n\n",
1856                 drv->heads, drv->sectors, drv->cylinders);
1857 }
1858 static void
1859 cciss_read_capacity(int ctlr, int logvol, ReadCapdata_struct *buf,
1860                 int withirq, unsigned int *total_size, unsigned int *block_size)
1861 {
1862         int return_code;
1863         memset(buf, 0, sizeof(*buf));
1864         if (withirq)
1865                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
1866                         ctlr, buf, sizeof(*buf), 1, logvol, 0, TYPE_CMD);
1867         else
1868                 return_code = sendcmd(CCISS_READ_CAPACITY,
1869                         ctlr, buf, sizeof(*buf), 1, logvol, 0, NULL, TYPE_CMD);
1870         if (return_code == IO_OK) {
1871                 *total_size = be32_to_cpu(*((__be32 *) &buf->total_size[0]))+1;
1872                 *block_size = be32_to_cpu(*((__be32 *) &buf->block_size[0]));
1873         } else { /* read capacity command failed */
1874                 printk(KERN_WARNING "cciss: read capacity failed\n");
1875                 *total_size = 0;
1876                 *block_size = BLOCK_SIZE;
1877         }
1878         printk(KERN_INFO "      blocks= %u block_size= %d\n",
1879                 *total_size, *block_size);
1880         return;
1881 }
1882
1883 static int cciss_revalidate(struct gendisk *disk)
1884 {
1885         ctlr_info_t *h = get_host(disk);
1886         drive_info_struct *drv = get_drv(disk);
1887         int logvol;
1888         int FOUND=0;
1889         unsigned int block_size;
1890         unsigned int total_size;
1891         ReadCapdata_struct *size_buff = NULL;
1892         InquiryData_struct *inq_buff = NULL;
1893
1894         for(logvol=0; logvol < CISS_MAX_LUN; logvol++)
1895         {
1896                 if(h->drv[logvol].LunID == drv->LunID) {
1897                         FOUND=1;
1898                         break;
1899                 }
1900         }
1901
1902         if (!FOUND) return 1;
1903
1904         size_buff = kmalloc(sizeof( ReadCapdata_struct), GFP_KERNEL);
1905         if (size_buff == NULL)
1906         {
1907                 printk(KERN_WARNING "cciss: out of memory\n");
1908                 return 1;
1909         }
1910         inq_buff = kmalloc(sizeof( InquiryData_struct), GFP_KERNEL);
1911         if (inq_buff == NULL)
1912         {
1913                 printk(KERN_WARNING "cciss: out of memory\n");
1914                 kfree(size_buff);
1915                 return 1;
1916         }
1917
1918         cciss_read_capacity(h->ctlr, logvol, size_buff, 1, &total_size, &block_size);
1919         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size, inq_buff, drv);
1920
1921         blk_queue_hardsect_size(drv->queue, drv->block_size);
1922         set_capacity(disk, drv->nr_blocks);
1923
1924         kfree(size_buff);
1925         kfree(inq_buff);
1926         return 0;
1927 }
1928
1929 /*
1930  *   Wait polling for a command to complete.
1931  *   The memory mapped FIFO is polled for the completion.
1932  *   Used only at init time, interrupts from the HBA are disabled.
1933  */
1934 static unsigned long pollcomplete(int ctlr)
1935 {
1936         unsigned long done;
1937         int i;
1938
1939         /* Wait (up to 20 seconds) for a command to complete */
1940
1941         for (i = 20 * HZ; i > 0; i--) {
1942                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
1943                 if (done == FIFO_EMPTY)
1944                         schedule_timeout_uninterruptible(1);
1945                 else
1946                         return (done);
1947         }
1948         /* Invalid address to tell caller we ran out of time */
1949         return 1;
1950 }
1951
1952 static int add_sendcmd_reject(__u8 cmd, int ctlr, unsigned long complete)
1953 {
1954         /* We get in here if sendcmd() is polling for completions
1955            and gets some command back that it wasn't expecting --
1956            something other than that which it just sent down.
1957            Ordinarily, that shouldn't happen, but it can happen when
1958            the scsi tape stuff gets into error handling mode, and
1959            starts using sendcmd() to try to abort commands and
1960            reset tape drives.  In that case, sendcmd may pick up
1961            completions of commands that were sent to logical drives
1962            through the block i/o system, or cciss ioctls completing, etc.
1963            In that case, we need to save those completions for later
1964            processing by the interrupt handler.
1965         */
1966
1967 #ifdef CONFIG_CISS_SCSI_TAPE
1968         struct sendcmd_reject_list *srl = &hba[ctlr]->scsi_rejects;
1969
1970         /* If it's not the scsi tape stuff doing error handling, (abort */
1971         /* or reset) then we don't expect anything weird. */
1972         if (cmd != CCISS_RESET_MSG && cmd != CCISS_ABORT_MSG) {
1973 #endif
1974                 printk( KERN_WARNING "cciss cciss%d: SendCmd "
1975                       "Invalid command list address returned! (%lx)\n",
1976                         ctlr, complete);
1977                 /* not much we can do. */
1978 #ifdef CONFIG_CISS_SCSI_TAPE
1979                 return 1;
1980         }
1981
1982         /* We've sent down an abort or reset, but something else
1983            has completed */
1984         if (srl->ncompletions >= (NR_CMDS + 2)) {
1985                 /* Uh oh.  No room to save it for later... */
1986                 printk(KERN_WARNING "cciss%d: Sendcmd: Invalid command addr, "
1987                         "reject list overflow, command lost!\n", ctlr);
1988                 return 1;
1989         }
1990         /* Save it for later */
1991         srl->complete[srl->ncompletions] = complete;
1992         srl->ncompletions++;
1993 #endif
1994         return 0;
1995 }
1996
1997 /*
1998  * Send a command to the controller, and wait for it to complete.
1999  * Only used at init time.
2000  */
2001 static int sendcmd(
2002         __u8    cmd,
2003         int     ctlr,
2004         void    *buff,
2005         size_t  size,
2006         unsigned int use_unit_num, /* 0: address the controller,
2007                                       1: address logical volume log_unit,
2008                                       2: periph device address is scsi3addr */
2009         unsigned int log_unit,
2010         __u8    page_code,
2011         unsigned char *scsi3addr,
2012         int cmd_type)
2013 {
2014         CommandList_struct *c;
2015         int i;
2016         unsigned long complete;
2017         ctlr_info_t *info_p= hba[ctlr];
2018         u64bit buff_dma_handle;
2019         int status, done = 0;
2020
2021         if ((c = cmd_alloc(info_p, 1)) == NULL) {
2022                 printk(KERN_WARNING "cciss: unable to get memory");
2023                 return(IO_ERROR);
2024         }
2025         status = fill_cmd(c, cmd, ctlr, buff, size, use_unit_num,
2026                 log_unit, page_code, scsi3addr, cmd_type);
2027         if (status != IO_OK) {
2028                 cmd_free(info_p, c, 1);
2029                 return status;
2030         }
2031 resend_cmd1:
2032         /*
2033          * Disable interrupt
2034          */
2035 #ifdef CCISS_DEBUG
2036         printk(KERN_DEBUG "cciss: turning intr off\n");
2037 #endif /* CCISS_DEBUG */
2038         info_p->access.set_intr_mask(info_p, CCISS_INTR_OFF);
2039
2040         /* Make sure there is room in the command FIFO */
2041         /* Actually it should be completely empty at this time */
2042         /* unless we are in here doing error handling for the scsi */
2043         /* tape side of the driver. */
2044         for (i = 200000; i > 0; i--)
2045         {
2046                 /* if fifo isn't full go */
2047                 if (!(info_p->access.fifo_full(info_p)))
2048                 {
2049
2050                         break;
2051                 }
2052                 udelay(10);
2053                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2054                         " waiting!\n", ctlr);
2055         }
2056         /*
2057          * Send the cmd
2058          */
2059         info_p->access.submit_command(info_p, c);
2060         done = 0;
2061         do {
2062                 complete = pollcomplete(ctlr);
2063
2064 #ifdef CCISS_DEBUG
2065                 printk(KERN_DEBUG "cciss: command completed\n");
2066 #endif /* CCISS_DEBUG */
2067
2068                 if (complete == 1) {
2069                         printk( KERN_WARNING
2070                                 "cciss cciss%d: SendCmd Timeout out, "
2071                                 "No command list address returned!\n",
2072                                 ctlr);
2073                         status = IO_ERROR;
2074                         done = 1;
2075                         break;
2076                 }
2077
2078                 /* This will need to change for direct lookup completions */
2079                 if ( (complete & CISS_ERROR_BIT)
2080                      && (complete & ~CISS_ERROR_BIT) == c->busaddr)
2081                      {
2082                         /* if data overrun or underun on Report command
2083                                 ignore it
2084                         */
2085                         if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2086                              (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2087                              (c->Request.CDB[0] == CISS_INQUIRY)) &&
2088                                 ((c->err_info->CommandStatus ==
2089                                         CMD_DATA_OVERRUN) ||
2090                                  (c->err_info->CommandStatus ==
2091                                         CMD_DATA_UNDERRUN)
2092                                 ))
2093                         {
2094                                 complete = c->busaddr;
2095                         } else {
2096                                 if (c->err_info->CommandStatus ==
2097                                                 CMD_UNSOLICITED_ABORT) {
2098                                         printk(KERN_WARNING "cciss%d: "
2099                                                 "unsolicited abort %p\n",
2100                                                 ctlr, c);
2101                                         if (c->retry_count < MAX_CMD_RETRIES) {
2102                                                 printk(KERN_WARNING
2103                                                    "cciss%d: retrying %p\n",
2104                                                    ctlr, c);
2105                                                 c->retry_count++;
2106                                                 /* erase the old error */
2107                                                 /* information */
2108                                                 memset(c->err_info, 0,
2109                                                    sizeof(ErrorInfo_struct));
2110                                                 goto resend_cmd1;
2111                                         } else {
2112                                                 printk(KERN_WARNING
2113                                                    "cciss%d: retried %p too "
2114                                                    "many times\n", ctlr, c);
2115                                                 status = IO_ERROR;
2116                                                 goto cleanup1;
2117                                         }
2118                                 } else if (c->err_info->CommandStatus == CMD_UNABORTABLE) {
2119                                         printk(KERN_WARNING "cciss%d: command could not be aborted.\n", ctlr);
2120                                         status = IO_ERROR;
2121                                         goto cleanup1;
2122                                 }
2123                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2124                                 " Error %x \n", ctlr,
2125                                         c->err_info->CommandStatus);
2126                                 printk(KERN_WARNING "ciss ciss%d: sendcmd"
2127                                 " offensive info\n"
2128                                 "  size %x\n   num %x   value %x\n", ctlr,
2129                                   c->err_info->MoreErrInfo.Invalid_Cmd.offense_size,
2130                                   c->err_info->MoreErrInfo.Invalid_Cmd.offense_num,
2131                                   c->err_info->MoreErrInfo.Invalid_Cmd.offense_value);
2132                                 status = IO_ERROR;
2133                                 goto cleanup1;
2134                         }
2135                 }
2136                 /* This will need changing for direct lookup completions */
2137                 if (complete != c->busaddr) {
2138                         if (add_sendcmd_reject(cmd, ctlr, complete) != 0) {
2139                                 BUG(); /* we are pretty much hosed if we get here. */
2140                         }
2141                         continue;
2142                 } else
2143                         done = 1;
2144         } while (!done);
2145
2146 cleanup1:
2147         /* unlock the data buffer from DMA */
2148         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2149         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2150         pci_unmap_single(info_p->pdev, (dma_addr_t) buff_dma_handle.val,
2151                                 c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2152 #ifdef CONFIG_CISS_SCSI_TAPE
2153         /* if we saved some commands for later, process them now. */
2154         if (info_p->scsi_rejects.ncompletions > 0)
2155                 do_cciss_intr(0, info_p, NULL);
2156 #endif
2157         cmd_free(info_p, c, 1);
2158         return (status);
2159 }
2160 /*
2161  * Map (physical) PCI mem into (virtual) kernel space
2162  */
2163 static void __iomem *remap_pci_mem(ulong base, ulong size)
2164 {
2165         ulong page_base        = ((ulong) base) & PAGE_MASK;
2166         ulong page_offs        = ((ulong) base) - page_base;
2167         void __iomem *page_remapped = ioremap(page_base, page_offs+size);
2168
2169         return page_remapped ? (page_remapped + page_offs) : NULL;
2170 }
2171
2172 /*
2173  * Takes jobs of the Q and sends them to the hardware, then puts it on
2174  * the Q to wait for completion.
2175  */
2176 static void start_io( ctlr_info_t *h)
2177 {
2178         CommandList_struct *c;
2179
2180         while(( c = h->reqQ) != NULL )
2181         {
2182                 /* can't do anything if fifo is full */
2183                 if ((h->access.fifo_full(h))) {
2184                         printk(KERN_WARNING "cciss: fifo full\n");
2185                         break;
2186                 }
2187
2188                 /* Get the first entry from the Request Q */
2189                 removeQ(&(h->reqQ), c);
2190                 h->Qdepth--;
2191
2192                 /* Tell the controller execute command */
2193                 h->access.submit_command(h, c);
2194
2195                 /* Put job onto the completed Q */
2196                 addQ (&(h->cmpQ), c);
2197         }
2198 }
2199 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2200 /* Zeros out the error record and then resends the command back */
2201 /* to the controller */
2202 static inline void resend_cciss_cmd( ctlr_info_t *h, CommandList_struct *c)
2203 {
2204         /* erase the old error information */
2205         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2206
2207         /* add it to software queue and then send it to the controller */
2208         addQ(&(h->reqQ),c);
2209         h->Qdepth++;
2210         if(h->Qdepth > h->maxQsinceinit)
2211                 h->maxQsinceinit = h->Qdepth;
2212
2213         start_io(h);
2214 }
2215
2216 /* checks the status of the job and calls complete buffers to mark all
2217  * buffers for the completed job. Note that this function does not need
2218  * to hold the hba/queue lock.
2219  */
2220 static inline void complete_command( ctlr_info_t *h, CommandList_struct *cmd,
2221                 int timeout)
2222 {
2223         int status = 1;
2224         int retry_cmd = 0;
2225
2226         if (timeout)
2227                 status = 0;
2228
2229         if(cmd->err_info->CommandStatus != 0)
2230         { /* an error has occurred */
2231                 switch(cmd->err_info->CommandStatus)
2232                 {
2233                         unsigned char sense_key;
2234                         case CMD_TARGET_STATUS:
2235                                 status = 0;
2236
2237                                 if( cmd->err_info->ScsiStatus == 0x02)
2238                                 {
2239                                         printk(KERN_WARNING "cciss: cmd %p "
2240                                                 "has CHECK CONDITION "
2241                                                 " byte 2 = 0x%x\n", cmd,
2242                                                 cmd->err_info->SenseInfo[2]
2243                                         );
2244                                         /* check the sense key */
2245                                         sense_key = 0xf &
2246                                                 cmd->err_info->SenseInfo[2];
2247                                         /* no status or recovered error */
2248                                         if((sense_key == 0x0) ||
2249                                             (sense_key == 0x1))
2250                                         {
2251                                                         status = 1;
2252                                         }
2253                                 } else
2254                                 {
2255                                         printk(KERN_WARNING "cciss: cmd %p "
2256                                                 "has SCSI Status 0x%x\n",
2257                                                 cmd, cmd->err_info->ScsiStatus);
2258                                 }
2259                         break;
2260                         case CMD_DATA_UNDERRUN:
2261                                 printk(KERN_WARNING "cciss: cmd %p has"
2262                                         " completed with data underrun "
2263                                         "reported\n", cmd);
2264                         break;
2265                         case CMD_DATA_OVERRUN:
2266                                 printk(KERN_WARNING "cciss: cmd %p has"
2267                                         " completed with data overrun "
2268                                         "reported\n", cmd);
2269                         break;
2270                         case CMD_INVALID:
2271                                 printk(KERN_WARNING "cciss: cmd %p is "
2272                                         "reported invalid\n", cmd);
2273                                 status = 0;
2274                         break;
2275                         case CMD_PROTOCOL_ERR:
2276                                 printk(KERN_WARNING "cciss: cmd %p has "
2277                                         "protocol error \n", cmd);
2278                                 status = 0;
2279                         break;
2280                         case CMD_HARDWARE_ERR:
2281                                 printk(KERN_WARNING "cciss: cmd %p had "
2282                                         " hardware error\n", cmd);
2283                                 status = 0;
2284                         break;
2285                         case CMD_CONNECTION_LOST:
2286                                 printk(KERN_WARNING "cciss: cmd %p had "
2287                                         "connection lost\n", cmd);
2288                                 status=0;
2289                         break;
2290                         case CMD_ABORTED:
2291                                 printk(KERN_WARNING "cciss: cmd %p was "
2292                                         "aborted\n", cmd);
2293                                 status=0;
2294                         break;
2295                         case CMD_ABORT_FAILED:
2296                                 printk(KERN_WARNING "cciss: cmd %p reports "
2297                                         "abort failed\n", cmd);
2298                                 status=0;
2299                         break;
2300                         case CMD_UNSOLICITED_ABORT:
2301                                 printk(KERN_WARNING "cciss%d: unsolicited "
2302                                         "abort %p\n", h->ctlr, cmd);
2303                                 if (cmd->retry_count < MAX_CMD_RETRIES) {
2304                                         retry_cmd=1;
2305                                         printk(KERN_WARNING
2306                                                 "cciss%d: retrying %p\n",
2307                                                 h->ctlr, cmd);
2308                                         cmd->retry_count++;
2309                                 } else
2310                                         printk(KERN_WARNING
2311                                                 "cciss%d: %p retried too "
2312                                                 "many times\n", h->ctlr, cmd);
2313                                 status=0;
2314                         break;
2315                         case CMD_TIMEOUT:
2316                                 printk(KERN_WARNING "cciss: cmd %p timedout\n",
2317                                         cmd);
2318                                 status=0;
2319                         break;
2320                         default:
2321                                 printk(KERN_WARNING "cciss: cmd %p returned "
2322                                         "unknown status %x\n", cmd,
2323                                                 cmd->err_info->CommandStatus);
2324                                 status=0;
2325                 }
2326         }
2327         /* We need to return this command */
2328         if(retry_cmd) {
2329                 resend_cciss_cmd(h,cmd);
2330                 return;
2331         }
2332
2333         cmd->rq->completion_data = cmd;
2334         cmd->rq->errors = status;
2335         blk_add_trace_rq(cmd->rq->q, cmd->rq, BLK_TA_COMPLETE);
2336         blk_complete_request(cmd->rq);
2337 }
2338
2339 /*
2340  * Get a request and submit it to the controller.
2341  */
2342 static void do_cciss_request(request_queue_t *q)
2343 {
2344         ctlr_info_t *h= q->queuedata;
2345         CommandList_struct *c;
2346         int start_blk, seg;
2347         struct request *creq;
2348         u64bit temp64;
2349         struct scatterlist tmp_sg[MAXSGENTRIES];
2350         drive_info_struct *drv;
2351         int i, dir;
2352
2353         /* We call start_io here in case there is a command waiting on the
2354          * queue that has not been sent.
2355         */
2356         if (blk_queue_plugged(q))
2357                 goto startio;
2358
2359 queue:
2360         creq = elv_next_request(q);
2361         if (!creq)
2362                 goto startio;
2363
2364         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
2365
2366         if (( c = cmd_alloc(h, 1)) == NULL)
2367                 goto full;
2368
2369         blkdev_dequeue_request(creq);
2370
2371         spin_unlock_irq(q->queue_lock);
2372
2373         c->cmd_type = CMD_RWREQ;
2374         c->rq = creq;
2375
2376         /* fill in the request */
2377         drv = creq->rq_disk->private_data;
2378         c->Header.ReplyQueue = 0;  // unused in simple mode
2379         /* got command from pool, so use the command block index instead */
2380         /* for direct lookups. */
2381         /* The first 2 bits are reserved for controller error reporting. */
2382         c->Header.Tag.lower = (c->cmdindex << 3);
2383         c->Header.Tag.lower |= 0x04; /* flag for direct lookup. */
2384         c->Header.LUN.LogDev.VolId= drv->LunID;
2385         c->Header.LUN.LogDev.Mode = 1;
2386         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
2387         c->Request.Type.Type =  TYPE_CMD; // It is a command.
2388         c->Request.Type.Attribute = ATTR_SIMPLE;
2389         c->Request.Type.Direction =
2390                 (rq_data_dir(creq) == READ) ? XFER_READ: XFER_WRITE;
2391         c->Request.Timeout = 0; // Don't time out
2392         c->Request.CDB[0] = (rq_data_dir(creq) == READ) ? CCISS_READ : CCISS_WRITE;
2393         start_blk = creq->sector;
2394 #ifdef CCISS_DEBUG
2395         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",(int) creq->sector,
2396                 (int) creq->nr_sectors);
2397 #endif /* CCISS_DEBUG */
2398
2399         seg = blk_rq_map_sg(q, creq, tmp_sg);
2400
2401         /* get the DMA records for the setup */
2402         if (c->Request.Type.Direction == XFER_READ)
2403                 dir = PCI_DMA_FROMDEVICE;
2404         else
2405                 dir = PCI_DMA_TODEVICE;
2406
2407         for (i=0; i<seg; i++)
2408         {
2409                 c->SG[i].Len = tmp_sg[i].length;
2410                 temp64.val = (__u64) pci_map_page(h->pdev, tmp_sg[i].page,
2411                                           tmp_sg[i].offset, tmp_sg[i].length,
2412                                           dir);
2413                 c->SG[i].Addr.lower = temp64.val32.lower;
2414                 c->SG[i].Addr.upper = temp64.val32.upper;
2415                 c->SG[i].Ext = 0;  // we are not chaining
2416         }
2417         /* track how many SG entries we are using */
2418         if( seg > h->maxSG)
2419                 h->maxSG = seg;
2420
2421 #ifdef CCISS_DEBUG
2422         printk(KERN_DEBUG "cciss: Submitting %d sectors in %d segments\n", creq->nr_sectors, seg);
2423 #endif /* CCISS_DEBUG */
2424
2425         c->Header.SGList = c->Header.SGTotal = seg;
2426         c->Request.CDB[1]= 0;
2427         c->Request.CDB[2]= (start_blk >> 24) & 0xff;    //MSB
2428         c->Request.CDB[3]= (start_blk >> 16) & 0xff;
2429         c->Request.CDB[4]= (start_blk >>  8) & 0xff;
2430         c->Request.CDB[5]= start_blk & 0xff;
2431         c->Request.CDB[6]= 0; // (sect >> 24) & 0xff; MSB
2432         c->Request.CDB[7]= (creq->nr_sectors >>  8) & 0xff;
2433         c->Request.CDB[8]= creq->nr_sectors & 0xff;
2434         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
2435
2436         spin_lock_irq(q->queue_lock);
2437
2438         addQ(&(h->reqQ),c);
2439         h->Qdepth++;
2440         if(h->Qdepth > h->maxQsinceinit)
2441                 h->maxQsinceinit = h->Qdepth;
2442
2443         goto queue;
2444 full:
2445         blk_stop_queue(q);
2446 startio:
2447         /* We will already have the driver lock here so not need
2448          * to lock it.
2449         */
2450         start_io(h);
2451 }
2452
2453 static inline unsigned long get_next_completion(ctlr_info_t *h)
2454 {
2455 #ifdef CONFIG_CISS_SCSI_TAPE
2456         /* Any rejects from sendcmd() lying around? Process them first */
2457         if (h->scsi_rejects.ncompletions == 0)
2458                 return h->access.command_completed(h);
2459         else {
2460                 struct sendcmd_reject_list *srl;
2461                 int n;
2462                 srl = &h->scsi_rejects;
2463                 n = --srl->ncompletions;
2464                 /* printk("cciss%d: processing saved reject\n", h->ctlr); */
2465                 printk("p");
2466                 return srl->complete[n];
2467         }
2468 #else
2469         return h->access.command_completed(h);
2470 #endif
2471 }
2472
2473 static inline int interrupt_pending(ctlr_info_t *h)
2474 {
2475 #ifdef CONFIG_CISS_SCSI_TAPE
2476         return ( h->access.intr_pending(h)
2477                 || (h->scsi_rejects.ncompletions > 0));
2478 #else
2479         return h->access.intr_pending(h);
2480 #endif
2481 }
2482
2483 static inline long interrupt_not_for_us(ctlr_info_t *h)
2484 {
2485 #ifdef CONFIG_CISS_SCSI_TAPE
2486         return (((h->access.intr_pending(h) == 0) ||
2487                  (h->interrupts_enabled == 0))
2488               && (h->scsi_rejects.ncompletions == 0));
2489 #else
2490         return (((h->access.intr_pending(h) == 0) ||
2491                  (h->interrupts_enabled == 0)));
2492 #endif
2493 }
2494
2495 static irqreturn_t do_cciss_intr(int irq, void *dev_id, struct pt_regs *regs)
2496 {
2497         ctlr_info_t *h = dev_id;
2498         CommandList_struct *c;
2499         unsigned long flags;
2500         __u32 a, a1, a2;
2501         int j;
2502         int start_queue = h->next_to_run;
2503
2504         if (interrupt_not_for_us(h))
2505                 return IRQ_NONE;
2506         /*
2507          * If there are completed commands in the completion queue,
2508          * we had better do something about it.
2509          */
2510         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2511         while (interrupt_pending(h)) {
2512                 while((a = get_next_completion(h)) != FIFO_EMPTY) {
2513                         a1 = a;
2514                         if ((a & 0x04)) {
2515                                 a2 = (a >> 3);
2516                                 if (a2 >= NR_CMDS) {
2517                                         printk(KERN_WARNING "cciss: controller cciss%d failed, stopping.\n", h->ctlr);
2518                                         fail_all_cmds(h->ctlr);
2519                                         return IRQ_HANDLED;
2520                                 }
2521
2522                                 c = h->cmd_pool + a2;
2523                                 a = c->busaddr;
2524
2525                         } else {
2526                         a &= ~3;
2527                                 if ((c = h->cmpQ) == NULL) {
2528                                         printk(KERN_WARNING "cciss: Completion of %08x ignored\n", a1);
2529                                 continue;
2530                         }
2531                         while(c->busaddr != a) {
2532                                 c = c->next;
2533                                 if (c == h->cmpQ)
2534                                         break;
2535                         }
2536                         }
2537                         /*
2538                          * If we've found the command, take it off the
2539                          * completion Q and free it
2540                          */
2541                          if (c->busaddr == a) {
2542                                 removeQ(&h->cmpQ, c);
2543                                 if (c->cmd_type == CMD_RWREQ) {
2544                                         complete_command(h, c, 0);
2545                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
2546                                         complete(c->waiting);
2547                                 }
2548 #                               ifdef CONFIG_CISS_SCSI_TAPE
2549                                 else if (c->cmd_type == CMD_SCSI)
2550                                         complete_scsi_command(c, 0, a1);
2551 #                               endif
2552                                 continue;
2553                         }
2554                 }
2555         }
2556
2557         /* check to see if we have maxed out the number of commands that can
2558          * be placed on the queue.  If so then exit.  We do this check here
2559          * in case the interrupt we serviced was from an ioctl and did not
2560          * free any new commands.
2561          */
2562         if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS)
2563                 goto cleanup;
2564
2565         /* We have room on the queue for more commands.  Now we need to queue
2566          * them up.  We will also keep track of the next queue to run so
2567          * that every queue gets a chance to be started first.
2568         */
2569         for (j=0; j < h->highest_lun + 1; j++){
2570                 int curr_queue = (start_queue + j) % (h->highest_lun + 1);
2571                 /* make sure the disk has been added and the drive is real
2572                  * because this can be called from the middle of init_one.
2573                 */
2574                 if(!(h->drv[curr_queue].queue) ||
2575                                    !(h->drv[curr_queue].heads))
2576                         continue;
2577                 blk_start_queue(h->gendisk[curr_queue]->queue);
2578
2579                 /* check to see if we have maxed out the number of commands
2580                  * that can be placed on the queue.
2581                 */
2582                 if ((find_first_zero_bit(h->cmd_pool_bits, NR_CMDS)) == NR_CMDS)
2583                 {
2584                         if (curr_queue == start_queue){
2585                                 h->next_to_run = (start_queue + 1) % (h->highest_lun + 1);
2586                                 goto cleanup;
2587                         } else {
2588                                 h->next_to_run = curr_queue;
2589                                 goto cleanup;
2590         }
2591                 } else {
2592                         curr_queue = (curr_queue + 1) % (h->highest_lun + 1);
2593                 }
2594         }
2595
2596 cleanup:
2597         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2598         return IRQ_HANDLED;
2599 }
2600 /*
2601  *  We cannot read the structure directly, for portablity we must use
2602  *   the io functions.
2603  *   This is for debug only.
2604  */
2605 #ifdef CCISS_DEBUG
2606 static void print_cfg_table( CfgTable_struct *tb)
2607 {
2608         int i;
2609         char temp_name[17];
2610
2611         printk("Controller Configuration information\n");
2612         printk("------------------------------------\n");
2613         for(i=0;i<4;i++)
2614                 temp_name[i] = readb(&(tb->Signature[i]));
2615         temp_name[4]='\0';
2616         printk("   Signature = %s\n", temp_name);
2617         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
2618         printk("   Transport methods supported = 0x%x\n",
2619                                 readl(&(tb-> TransportSupport)));
2620         printk("   Transport methods active = 0x%x\n",
2621                                 readl(&(tb->TransportActive)));
2622         printk("   Requested transport Method = 0x%x\n",
2623                         readl(&(tb->HostWrite.TransportRequest)));
2624         printk("   Coalese Interrupt Delay = 0x%x\n",
2625                         readl(&(tb->HostWrite.CoalIntDelay)));
2626         printk("   Coalese Interrupt Count = 0x%x\n",
2627                         readl(&(tb->HostWrite.CoalIntCount)));
2628         printk("   Max outstanding commands = 0x%d\n",
2629                         readl(&(tb->CmdsOutMax)));
2630         printk("   Bus Types = 0x%x\n", readl(&(tb-> BusTypes)));
2631         for(i=0;i<16;i++)
2632                 temp_name[i] = readb(&(tb->ServerName[i]));
2633         temp_name[16] = '\0';
2634         printk("   Server Name = %s\n", temp_name);
2635         printk("   Heartbeat Counter = 0x%x\n\n\n",
2636                         readl(&(tb->HeartBeat)));
2637 }
2638 #endif /* CCISS_DEBUG */
2639
2640 static void release_io_mem(ctlr_info_t *c)
2641 {
2642         /* if IO mem was not protected do nothing */
2643         if( c->io_mem_addr == 0)
2644                 return;
2645         release_region(c->io_mem_addr, c->io_mem_length);
2646         c->io_mem_addr = 0;
2647         c->io_mem_length = 0;
2648 }
2649
2650 static int find_PCI_BAR_index(struct pci_dev *pdev,
2651                                 unsigned long pci_bar_addr)
2652 {
2653         int i, offset, mem_type, bar_type;
2654         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
2655                 return 0;
2656         offset = 0;
2657         for (i=0; i<DEVICE_COUNT_RESOURCE; i++) {
2658                 bar_type = pci_resource_flags(pdev, i) &
2659                         PCI_BASE_ADDRESS_SPACE;
2660                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
2661                         offset += 4;
2662                 else {
2663                         mem_type = pci_resource_flags(pdev, i) &
2664                                 PCI_BASE_ADDRESS_MEM_TYPE_MASK;
2665                         switch (mem_type) {
2666                                 case PCI_BASE_ADDRESS_MEM_TYPE_32:
2667                                 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
2668                                         offset += 4; /* 32 bit */
2669                                         break;
2670                                 case PCI_BASE_ADDRESS_MEM_TYPE_64:
2671                                         offset += 8;
2672                                         break;
2673                                 default: /* reserved in PCI 2.2 */
2674                                         printk(KERN_WARNING "Base address is invalid\n");
2675                                         return -1;
2676                                 break;
2677                         }
2678                 }
2679                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
2680                         return i+1;
2681         }
2682         return -1;
2683 }
2684
2685 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
2686  * controllers that are capable. If not, we use IO-APIC mode.
2687  */
2688
2689 static void __devinit cciss_interrupt_mode(ctlr_info_t *c, struct pci_dev *pdev, __u32 board_id)
2690 {
2691 #ifdef CONFIG_PCI_MSI
2692         int err;
2693         struct msix_entry cciss_msix_entries[4] = {{0,0}, {0,1},
2694                                                    {0,2}, {0,3}};
2695
2696         /* Some boards advertise MSI but don't really support it */
2697         if ((board_id == 0x40700E11) ||
2698                 (board_id == 0x40800E11) ||
2699                 (board_id == 0x40820E11) ||
2700                 (board_id == 0x40830E11))
2701                 goto default_int_mode;
2702
2703         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
2704                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
2705                 if (!err) {
2706                         c->intr[0] = cciss_msix_entries[0].vector;
2707                         c->intr[1] = cciss_msix_entries[1].vector;
2708                         c->intr[2] = cciss_msix_entries[2].vector;
2709                         c->intr[3] = cciss_msix_entries[3].vector;
2710                         c->msix_vector = 1;
2711                         return;
2712                 }
2713                 if (err > 0) {
2714                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
2715                                         "available\n", err);
2716                 } else {
2717                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
2718                                                 err);
2719                 }
2720         }
2721         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
2722                 if (!pci_enable_msi(pdev)) {
2723                         c->intr[SIMPLE_MODE_INT] = pdev->irq;
2724                         c->msi_vector = 1;
2725                         return;
2726                 } else {
2727                         printk(KERN_WARNING "cciss: MSI init failed\n");
2728                         c->intr[SIMPLE_MODE_INT] = pdev->irq;
2729                         return;
2730                 }
2731         }
2732 default_int_mode:
2733 #endif /* CONFIG_PCI_MSI */
2734         /* if we get here we're going to use the default interrupt mode */
2735         c->intr[SIMPLE_MODE_INT] = pdev->irq;
2736         return;
2737 }
2738
2739 static int cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
2740 {
2741         ushort subsystem_vendor_id, subsystem_device_id, command;
2742         __u32 board_id, scratchpad = 0;
2743         __u64 cfg_offset;
2744         __u32 cfg_base_addr;
2745         __u64 cfg_base_addr_index;
2746         int i;
2747
2748         /* check to see if controller has been disabled */
2749         /* BEFORE trying to enable it */
2750         (void) pci_read_config_word(pdev, PCI_COMMAND,&command);
2751         if(!(command & 0x02))
2752         {
2753                 printk(KERN_WARNING "cciss: controller appears to be disabled\n");
2754                 return(-1);
2755         }
2756
2757         if (pci_enable_device(pdev))
2758         {
2759                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
2760                 return( -1);
2761         }
2762
2763         subsystem_vendor_id = pdev->subsystem_vendor;
2764         subsystem_device_id = pdev->subsystem_device;
2765         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
2766                                         subsystem_vendor_id);
2767
2768         /* search for our IO range so we can protect it */
2769         for(i=0; i<DEVICE_COUNT_RESOURCE; i++)
2770         {
2771                 /* is this an IO range */
2772                 if( pci_resource_flags(pdev, i) & 0x01 ) {
2773                         c->io_mem_addr = pci_resource_start(pdev, i);
2774                         c->io_mem_length = pci_resource_end(pdev, i) -
2775                                 pci_resource_start(pdev, i) +1;
2776 #ifdef CCISS_DEBUG
2777                         printk("IO value found base_addr[%d] %lx %lx\n", i,
2778                                 c->io_mem_addr, c->io_mem_length);
2779 #endif /* CCISS_DEBUG */
2780                         /* register the IO range */
2781                         if(!request_region( c->io_mem_addr,
2782                                         c->io_mem_length, "cciss"))
2783                         {
2784                                 printk(KERN_WARNING "cciss I/O memory range already in use addr=%lx length=%ld\n",
2785                                 c->io_mem_addr, c->io_mem_length);
2786                                 c->io_mem_addr= 0;
2787                                 c->io_mem_length = 0;
2788                         }
2789                         break;
2790                 }
2791         }
2792
2793 #ifdef CCISS_DEBUG
2794         printk("command = %x\n", command);
2795         printk("irq = %x\n", pdev->irq);
2796         printk("board_id = %x\n", board_id);
2797 #endif /* CCISS_DEBUG */
2798
2799 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
2800  * else we use the IO-APIC interrupt assigned to us by system ROM.
2801  */
2802         cciss_interrupt_mode(c, pdev, board_id);
2803
2804         /*
2805          * Memory base addr is first addr , the second points to the config
2806          *   table
2807          */
2808
2809         c->paddr = pci_resource_start(pdev, 0); /* addressing mode bits already removed */
2810 #ifdef CCISS_DEBUG
2811         printk("address 0 = %x\n", c->paddr);
2812 #endif /* CCISS_DEBUG */
2813         c->vaddr = remap_pci_mem(c->paddr, 200);
2814
2815         /* Wait for the board to become ready.  (PCI hotplug needs this.)
2816          * We poll for up to 120 secs, once per 100ms. */
2817         for (i=0; i < 1200; i++) {
2818                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
2819                 if (scratchpad == CCISS_FIRMWARE_READY)
2820                         break;
2821                 set_current_state(TASK_INTERRUPTIBLE);
2822                 schedule_timeout(HZ / 10); /* wait 100ms */
2823         }
2824         if (scratchpad != CCISS_FIRMWARE_READY) {
2825                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
2826                 return -1;
2827         }
2828
2829         /* get the address index number */
2830         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
2831         cfg_base_addr &= (__u32) 0x0000ffff;
2832 #ifdef CCISS_DEBUG
2833         printk("cfg base address = %x\n", cfg_base_addr);
2834 #endif /* CCISS_DEBUG */
2835         cfg_base_addr_index =
2836                 find_PCI_BAR_index(pdev, cfg_base_addr);
2837 #ifdef CCISS_DEBUG
2838         printk("cfg base address index = %x\n", cfg_base_addr_index);
2839 #endif /* CCISS_DEBUG */
2840         if (cfg_base_addr_index == -1) {
2841                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
2842                 release_io_mem(c);
2843                 return -1;
2844         }
2845
2846         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
2847 #ifdef CCISS_DEBUG
2848         printk("cfg offset = %x\n", cfg_offset);
2849 #endif /* CCISS_DEBUG */
2850         c->cfgtable =  remap_pci_mem(pci_resource_start(pdev,
2851                                 cfg_base_addr_index) + cfg_offset,
2852                                 sizeof(CfgTable_struct));
2853         c->board_id = board_id;
2854
2855 #ifdef CCISS_DEBUG
2856         print_cfg_table(c->cfgtable);
2857 #endif /* CCISS_DEBUG */
2858
2859         for(i=0; i<NR_PRODUCTS; i++) {
2860                 if (board_id == products[i].board_id) {
2861                         c->product_name = products[i].product_name;
2862                         c->access = *(products[i].access);
2863                         break;
2864                 }
2865         }
2866         if (i == NR_PRODUCTS) {
2867                 printk(KERN_WARNING "cciss: Sorry, I don't know how"
2868                         " to access the Smart Array controller %08lx\n",
2869                                 (unsigned long)board_id);
2870                 return -1;
2871         }
2872         if (  (readb(&c->cfgtable->Signature[0]) != 'C') ||
2873               (readb(&c->cfgtable->Signature[1]) != 'I') ||
2874               (readb(&c->cfgtable->Signature[2]) != 'S') ||
2875               (readb(&c->cfgtable->Signature[3]) != 'S') )
2876         {
2877                 printk("Does not appear to be a valid CISS config table\n");
2878                 return -1;
2879         }
2880
2881 #ifdef CONFIG_X86
2882 {
2883         /* Need to enable prefetch in the SCSI core for 6400 in x86 */
2884         __u32 prefetch;
2885         prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
2886         prefetch |= 0x100;
2887         writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
2888 }
2889 #endif
2890
2891 #ifdef CCISS_DEBUG
2892         printk("Trying to put board into Simple mode\n");
2893 #endif /* CCISS_DEBUG */
2894         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
2895         /* Update the field, and then ring the doorbell */
2896         writel( CFGTBL_Trans_Simple,
2897                 &(c->cfgtable->HostWrite.TransportRequest));
2898         writel( CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
2899
2900         /* under certain very rare conditions, this can take awhile.
2901          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
2902          * as we enter this code.) */
2903         for(i=0;i<MAX_CONFIG_WAIT;i++) {
2904                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
2905                         break;
2906                 /* delay and try again */
2907                 set_current_state(TASK_INTERRUPTIBLE);
2908                 schedule_timeout(10);
2909         }
2910
2911 #ifdef CCISS_DEBUG
2912         printk(KERN_DEBUG "I counter got to %d %x\n", i, readl(c->vaddr + SA5_DOORBELL));
2913 #endif /* CCISS_DEBUG */
2914 #ifdef CCISS_DEBUG
2915         print_cfg_table(c->cfgtable);
2916 #endif /* CCISS_DEBUG */
2917
2918         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple))
2919         {
2920                 printk(KERN_WARNING "cciss: unable to get board into"
2921                                         " simple mode\n");
2922                 return -1;
2923         }
2924         return 0;
2925
2926 }
2927
2928 /*
2929  * Gets information about the local volumes attached to the controller.
2930  */
2931 static void cciss_getgeometry(int cntl_num)
2932 {
2933         ReportLunData_struct *ld_buff;
2934         ReadCapdata_struct *size_buff;
2935         InquiryData_struct *inq_buff;
2936         int return_code;
2937         int i;
2938         int listlength = 0;
2939         __u32 lunid = 0;
2940         int block_size;
2941         int total_size;
2942
2943         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2944         if (ld_buff == NULL)
2945         {
2946                 printk(KERN_ERR "cciss: out of memory\n");
2947                 return;
2948         }
2949         size_buff = kmalloc(sizeof( ReadCapdata_struct), GFP_KERNEL);
2950         if (size_buff == NULL)
2951         {
2952                 printk(KERN_ERR "cciss: out of memory\n");
2953                 kfree(ld_buff);
2954                 return;
2955         }
2956         inq_buff = kmalloc(sizeof( InquiryData_struct), GFP_KERNEL);
2957         if (inq_buff == NULL)
2958         {
2959                 printk(KERN_ERR "cciss: out of memory\n");
2960                 kfree(ld_buff);
2961                 kfree(size_buff);
2962                 return;
2963         }
2964         /* Get the firmware version */
2965         return_code = sendcmd(CISS_INQUIRY, cntl_num, inq_buff,
2966                 sizeof(InquiryData_struct), 0, 0 ,0, NULL, TYPE_CMD);
2967         if (return_code == IO_OK)
2968         {
2969                 hba[cntl_num]->firm_ver[0] = inq_buff->data_byte[32];
2970                 hba[cntl_num]->firm_ver[1] = inq_buff->data_byte[33];
2971                 hba[cntl_num]->firm_ver[2] = inq_buff->data_byte[34];
2972                 hba[cntl_num]->firm_ver[3] = inq_buff->data_byte[35];
2973         } else /* send command failed */
2974         {
2975                 printk(KERN_WARNING "cciss: unable to determine firmware"
2976                         " version of controller\n");
2977         }
2978         /* Get the number of logical volumes */
2979         return_code = sendcmd(CISS_REPORT_LOG, cntl_num, ld_buff,
2980                         sizeof(ReportLunData_struct), 0, 0, 0, NULL, TYPE_CMD);
2981
2982         if( return_code == IO_OK)
2983         {
2984 #ifdef CCISS_DEBUG
2985                 printk("LUN Data\n--------------------------\n");
2986 #endif /* CCISS_DEBUG */
2987
2988                 listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[0])) << 24;
2989                 listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[1])) << 16;
2990                 listlength |= (0xff & (unsigned int)(ld_buff->LUNListLength[2])) << 8;
2991                 listlength |= 0xff & (unsigned int)(ld_buff->LUNListLength[3]);
2992         } else /* reading number of logical volumes failed */
2993         {
2994                 printk(KERN_WARNING "cciss: report logical volume"
2995                         " command failed\n");
2996                 listlength = 0;
2997         }
2998         hba[cntl_num]->num_luns = listlength / 8; // 8 bytes pre entry
2999         if (hba[cntl_num]->num_luns > CISS_MAX_LUN)
3000         {
3001                 printk(KERN_ERR "ciss:  only %d number of logical volumes supported\n",
3002                         CISS_MAX_LUN);
3003                 hba[cntl_num]->num_luns = CISS_MAX_LUN;
3004         }
3005 #ifdef CCISS_DEBUG
3006         printk(KERN_DEBUG "Length = %x %x %x %x = %d\n", ld_buff->LUNListLength[0],
3007                 ld_buff->LUNListLength[1], ld_buff->LUNListLength[2],
3008                 ld_buff->LUNListLength[3],  hba[cntl_num]->num_luns);
3009 #endif /* CCISS_DEBUG */
3010
3011         hba[cntl_num]->highest_lun = hba[cntl_num]->num_luns-1;
3012 //      for(i=0; i<  hba[cntl_num]->num_luns; i++)
3013         for(i=0; i < CISS_MAX_LUN; i++)
3014         {
3015                 if (i < hba[cntl_num]->num_luns){
3016                         lunid = (0xff & (unsigned int)(ld_buff->LUN[i][3]))
3017                                  << 24;
3018                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][2]))
3019                                  << 16;
3020                         lunid |= (0xff & (unsigned int)(ld_buff->LUN[i][1]))
3021                                  << 8;
3022                 lunid |= 0xff & (unsigned int)(ld_buff->LUN[i][0]);
3023
3024                 hba[cntl_num]->drv[i].LunID = lunid;
3025
3026
3027 #ifdef CCISS_DEBUG
3028                 printk(KERN_DEBUG "LUN[%d]:  %x %x %x %x = %x\n", i,
3029                         ld_buff->LUN[i][0], ld_buff->LUN[i][1],
3030                         ld_buff->LUN[i][2], ld_buff->LUN[i][3],
3031                         hba[cntl_num]->drv[i].LunID);
3032 #endif /* CCISS_DEBUG */
3033                 cciss_read_capacity(cntl_num, i, size_buff, 0,
3034                         &total_size, &block_size);
3035                         cciss_geometry_inquiry(cntl_num, i, 0, total_size,
3036                                 block_size, inq_buff, &hba[cntl_num]->drv[i]);
3037                 } else {
3038                         /* initialize raid_level to indicate a free space */
3039                         hba[cntl_num]->drv[i].raid_level = -1;
3040                 }
3041         }
3042         kfree(ld_buff);
3043         kfree(size_buff);
3044         kfree(inq_buff);
3045 }
3046
3047 /* Function to find the first free pointer into our hba[] array */
3048 /* Returns -1 if no free entries are left.  */
3049 static int alloc_cciss_hba(void)
3050 {
3051         struct gendisk *disk[NWD];
3052         int i, n;
3053         for (n = 0; n < NWD; n++) {
3054                 disk[n] = alloc_disk(1 << NWD_SHIFT);
3055                 if (!disk[n])
3056                         goto out;
3057         }
3058
3059         for(i=0; i< MAX_CTLR; i++) {
3060                 if (!hba[i]) {
3061                         ctlr_info_t *p;
3062                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3063                         if (!p)
3064                                 goto Enomem;
3065                         for (n = 0; n < NWD; n++)
3066                                 p->gendisk[n] = disk[n];
3067                         hba[i] = p;
3068                         return i;
3069                 }
3070         }
3071         printk(KERN_WARNING "cciss: This driver supports a maximum"
3072                 " of %d controllers.\n", MAX_CTLR);
3073         goto out;
3074 Enomem:
3075         printk(KERN_ERR "cciss: out of memory.\n");
3076 out:
3077         while (n--)
3078                 put_disk(disk[n]);
3079         return -1;
3080 }
3081
3082 static void free_hba(int i)
3083 {
3084         ctlr_info_t *p = hba[i];
3085         int n;
3086
3087         hba[i] = NULL;
3088         for (n = 0; n < NWD; n++)
3089                 put_disk(p->gendisk[n]);
3090         kfree(p);
3091 }
3092
3093 /*
3094  *  This is it.  Find all the controllers and register them.  I really hate
3095  *  stealing all these major device numbers.
3096  *  returns the number of block devices registered.
3097  */
3098 static int __devinit cciss_init_one(struct pci_dev *pdev,
3099         const struct pci_device_id *ent)
3100 {
3101         request_queue_t *q;
3102         int i;
3103         int j;
3104         int rc;
3105
3106         printk(KERN_DEBUG "cciss: Device 0x%x has been found at"
3107                         " bus %d dev %d func %d\n",
3108                 pdev->device, pdev->bus->number, PCI_SLOT(pdev->devfn),
3109                         PCI_FUNC(pdev->devfn));
3110         i = alloc_cciss_hba();
3111         if(i < 0)
3112                 return (-1);
3113
3114         hba[i]->busy_initializing = 1;
3115
3116         if (cciss_pci_init(hba[i], pdev) != 0)
3117                 goto clean1;
3118
3119         sprintf(hba[i]->devname, "cciss%d", i);
3120         hba[i]->ctlr = i;
3121         hba[i]->pdev = pdev;
3122
3123         /* configure PCI DMA stuff */
3124         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK))
3125                 printk("cciss: using DAC cycles\n");
3126         else if (!pci_set_dma_mask(pdev, DMA_32BIT_MASK))
3127                 printk("cciss: not using DAC cycles\n");
3128         else {
3129                 printk("cciss: no suitable DMA available\n");
3130                 goto clean1;
3131         }
3132
3133         /*
3134          * register with the major number, or get a dynamic major number
3135          * by passing 0 as argument.  This is done for greater than
3136          * 8 controller support.
3137          */
3138         if (i < MAX_CTLR_ORIG)
3139                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
3140         rc = register_blkdev(hba[i]->major, hba[i]->devname);
3141         if(rc == -EBUSY || rc == -EINVAL) {
3142                 printk(KERN_ERR
3143                         "cciss:  Unable to get major number %d for %s "
3144                         "on hba %d\n", hba[i]->major, hba[i]->devname, i);
3145                 goto clean1;
3146         }
3147         else {
3148                 if (i >= MAX_CTLR_ORIG)
3149                         hba[i]->major = rc;
3150         }
3151
3152         /* make sure the board interrupts are off */
3153         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
3154         if( request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
3155                 SA_INTERRUPT | SA_SHIRQ | SA_SAMPLE_RANDOM,
3156                         hba[i]->devname, hba[i])) {
3157                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
3158                         hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
3159                 goto clean2;
3160         }
3161         hba[i]->cmd_pool_bits = kmalloc(((NR_CMDS+BITS_PER_LONG-1)/BITS_PER_LONG)*sizeof(unsigned long), GFP_KERNEL);
3162         hba[i]->cmd_pool = (CommandList_struct *)pci_alloc_consistent(
3163                 hba[i]->pdev, NR_CMDS * sizeof(CommandList_struct),
3164                 &(hba[i]->cmd_pool_dhandle));
3165         hba[i]->errinfo_pool = (ErrorInfo_struct *)pci_alloc_consistent(
3166                 hba[i]->pdev, NR_CMDS * sizeof( ErrorInfo_struct),
3167                 &(hba[i]->errinfo_pool_dhandle));
3168         if((hba[i]->cmd_pool_bits == NULL)
3169                 || (hba[i]->cmd_pool == NULL)
3170                 || (hba[i]->errinfo_pool == NULL)) {
3171                 printk( KERN_ERR "cciss: out of memory");
3172                 goto clean4;
3173         }
3174 #ifdef CONFIG_CISS_SCSI_TAPE
3175         hba[i]->scsi_rejects.complete =
3176                 kmalloc(sizeof(hba[i]->scsi_rejects.complete[0]) *
3177                         (NR_CMDS + 5), GFP_KERNEL);
3178         if (hba[i]->scsi_rejects.complete == NULL) {
3179                 printk( KERN_ERR "cciss: out of memory");
3180                 goto clean4;
3181         }
3182 #endif
3183         spin_lock_init(&hba[i]->lock);
3184
3185         /* Initialize the pdev driver private data.
3186                 have it point to hba[i].  */
3187         pci_set_drvdata(pdev, hba[i]);
3188         /* command and error info recs zeroed out before
3189                         they are used */
3190         memset(hba[i]->cmd_pool_bits, 0, ((NR_CMDS+BITS_PER_LONG-1)/BITS_PER_LONG)*sizeof(unsigned long));
3191
3192 #ifdef CCISS_DEBUG
3193         printk(KERN_DEBUG "Scanning for drives on controller cciss%d\n",i);
3194 #endif /* CCISS_DEBUG */
3195
3196         cciss_getgeometry(i);
3197
3198         cciss_scsi_setup(i);
3199
3200         /* Turn the interrupts on so we can service requests */
3201         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
3202
3203         cciss_procinit(i);
3204         hba[i]->busy_initializing = 0;
3205
3206         for(j=0; j < NWD; j++) { /* mfm */
3207                 drive_info_struct *drv = &(hba[i]->drv[j]);
3208                 struct gendisk *disk = hba[i]->gendisk[j];
3209
3210                 q = blk_init_queue(do_cciss_request, &hba[i]->lock);
3211                 if (!q) {
3212                         printk(KERN_ERR
3213                            "cciss:  unable to allocate queue for disk %d\n",
3214                            j);
3215                         break;
3216                 }
3217                 drv->queue = q;
3218
3219                 q->backing_dev_info.ra_pages = READ_AHEAD;
3220                 blk_queue_bounce_limit(q, hba[i]->pdev->dma_mask);
3221
3222                 /* This is a hardware imposed limit. */
3223                 blk_queue_max_hw_segments(q, MAXSGENTRIES);
3224
3225                 /* This is a limit in the driver and could be eliminated. */
3226                 blk_queue_max_phys_segments(q, MAXSGENTRIES);
3227
3228                 blk_queue_max_sectors(q, 512);
3229
3230                 blk_queue_softirq_done(q, cciss_softirq_done);
3231
3232                 q->queuedata = hba[i];
3233                 sprintf(disk->disk_name, "cciss/c%dd%d", i, j);
3234                 sprintf(disk->devfs_name, "cciss/host%d/target%d", i, j);
3235                 disk->major = hba[i]->major;
3236                 disk->first_minor = j << NWD_SHIFT;
3237                 disk->fops = &cciss_fops;
3238                 disk->queue = q;
3239                 disk->private_data = drv;
3240                 /* we must register the controller even if no disks exist */
3241                 /* this is for the online array utilities */
3242                 if(!drv->heads && j)
3243                         continue;
3244                 blk_queue_hardsect_size(q, drv->block_size);
3245                 set_capacity(disk, drv->nr_blocks);
3246                 add_disk(disk);
3247         }
3248
3249         return(1);
3250
3251 clean4:
3252 #ifdef CONFIG_CISS_SCSI_TAPE
3253         kfree(hba[i]->scsi_rejects.complete);
3254 #endif
3255         kfree(hba[i]->cmd_pool_bits);
3256         if(hba[i]->cmd_pool)
3257                 pci_free_consistent(hba[i]->pdev,
3258                         NR_CMDS * sizeof(CommandList_struct),
3259                         hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3260         if(hba[i]->errinfo_pool)
3261                 pci_free_consistent(hba[i]->pdev,
3262                         NR_CMDS * sizeof( ErrorInfo_struct),
3263                         hba[i]->errinfo_pool,
3264                         hba[i]->errinfo_pool_dhandle);
3265         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
3266 clean2:
3267         unregister_blkdev(hba[i]->major, hba[i]->devname);
3268 clean1:
3269         release_io_mem(hba[i]);
3270         hba[i]->busy_initializing = 0;
3271         free_hba(i);
3272         return(-1);
3273 }
3274
3275 static void __devexit cciss_remove_one (struct pci_dev *pdev)
3276 {
3277         ctlr_info_t *tmp_ptr;
3278         int i, j;
3279         char flush_buf[4];
3280         int return_code;
3281
3282         if (pci_get_drvdata(pdev) == NULL)
3283         {
3284                 printk( KERN_ERR "cciss: Unable to remove device \n");
3285                 return;
3286         }
3287         tmp_ptr = pci_get_drvdata(pdev);
3288         i = tmp_ptr->ctlr;
3289         if (hba[i] == NULL)
3290         {
3291                 printk(KERN_ERR "cciss: device appears to "
3292                         "already be removed \n");
3293                 return;
3294         }
3295         /* Turn board interrupts off  and send the flush cache command */
3296         /* sendcmd will turn off interrupt, and send the flush...
3297         * To write all data in the battery backed cache to disks */
3298         memset(flush_buf, 0, 4);
3299         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0, 0, 0, NULL,
3300                                 TYPE_CMD);
3301         if(return_code != IO_OK)
3302         {
3303                 printk(KERN_WARNING "Error Flushing cache on controller %d\n",
3304                         i);
3305         }
3306         free_irq(hba[i]->intr[2], hba[i]);
3307
3308 #ifdef CONFIG_PCI_MSI
3309         if (hba[i]->msix_vector)
3310                 pci_disable_msix(hba[i]->pdev);
3311         else if (hba[i]->msi_vector)
3312                 pci_disable_msi(hba[i]->pdev);
3313 #endif /* CONFIG_PCI_MSI */
3314
3315         pci_set_drvdata(pdev, NULL);
3316         iounmap(hba[i]->vaddr);
3317         cciss_unregister_scsi(i);  /* unhook from SCSI subsystem */
3318         unregister_blkdev(hba[i]->major, hba[i]->devname);
3319         remove_proc_entry(hba[i]->devname, proc_cciss);
3320
3321         /* remove it from the disk list */
3322         for (j = 0; j < NWD; j++) {
3323                 struct gendisk *disk = hba[i]->gendisk[j];
3324                 if (disk) {
3325                         request_queue_t *q = disk->queue;
3326
3327                         if (disk->flags & GENHD_FL_UP)
3328                                 del_gendisk(disk);
3329                         if (q)
3330                                 blk_cleanup_queue(q);
3331                 }
3332         }
3333
3334         pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof(CommandList_struct),
3335                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
3336         pci_free_consistent(hba[i]->pdev, NR_CMDS * sizeof( ErrorInfo_struct),
3337                 hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
3338         kfree(hba[i]->cmd_pool_bits);
3339 #ifdef CONFIG_CISS_SCSI_TAPE
3340         kfree(hba[i]->scsi_rejects.complete);
3341 #endif
3342         release_io_mem(hba[i]);
3343         free_hba(i);
3344 }
3345
3346 static struct pci_driver cciss_pci_driver = {
3347         .name =         "cciss",
3348         .probe =        cciss_init_one,
3349         .remove =       __devexit_p(cciss_remove_one),
3350         .id_table =     cciss_pci_device_id, /* id_table */
3351 };
3352
3353 /*
3354  *  This is it.  Register the PCI driver information for the cards we control
3355  *  the OS will call our registered routines when it finds one of our cards.
3356  */
3357 static int __init cciss_init(void)
3358 {
3359         printk(KERN_INFO DRIVER_NAME "\n");
3360
3361         /* Register for our PCI devices */
3362         return pci_register_driver(&cciss_pci_driver);
3363 }
3364
3365 static void __exit cciss_cleanup(void)
3366 {
3367         int i;
3368
3369         pci_unregister_driver(&cciss_pci_driver);
3370         /* double check that all controller entrys have been removed */
3371         for (i=0; i< MAX_CTLR; i++)
3372         {
3373                 if (hba[i] != NULL)
3374                 {
3375                         printk(KERN_WARNING "cciss: had to remove"
3376                                         " controller %d\n", i);
3377                         cciss_remove_one(hba[i]->pdev);
3378                 }
3379         }
3380         remove_proc_entry("cciss", proc_root_driver);
3381 }
3382
3383 static void fail_all_cmds(unsigned long ctlr)
3384 {
3385         /* If we get here, the board is apparently dead. */
3386         ctlr_info_t *h = hba[ctlr];
3387         CommandList_struct *c;
3388         unsigned long flags;
3389
3390         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
3391         h->alive = 0;   /* the controller apparently died... */
3392
3393         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
3394
3395         pci_disable_device(h->pdev); /* Make sure it is really dead. */
3396
3397         /* move everything off the request queue onto the completed queue */
3398         while( (c = h->reqQ) != NULL ) {
3399                 removeQ(&(h->reqQ), c);
3400                 h->Qdepth--;
3401                 addQ (&(h->cmpQ), c);
3402         }
3403
3404         /* Now, fail everything on the completed queue with a HW error */
3405         while( (c = h->cmpQ) != NULL ) {
3406                 removeQ(&h->cmpQ, c);
3407                 c->err_info->CommandStatus = CMD_HARDWARE_ERR;
3408                 if (c->cmd_type == CMD_RWREQ) {
3409                         complete_command(h, c, 0);
3410                 } else if (c->cmd_type == CMD_IOCTL_PEND)
3411                         complete(c->waiting);
3412 #ifdef CONFIG_CISS_SCSI_TAPE
3413                         else if (c->cmd_type == CMD_SCSI)
3414                                 complete_scsi_command(c, 0, 0);
3415 #endif
3416         }
3417         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
3418         return;
3419 }
3420
3421 module_init(cciss_init);
3422 module_exit(cciss_cleanup);