drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/compat.h>
  43 #include <linux/mutex.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/io.h>
  46
  47 #include <linux/dma-mapping.h>
  48 #include <linux/blkdev.h>
  49 #include <linux/genhd.h>
  50 #include <linux/completion.h>
  51 #include <scsi/scsi.h>
  52 #include <scsi/sg.h>
  53 #include <scsi/scsi_ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/scatterlist.h>
  56 #include <linux/kthread.h>
  57
  58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  59 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  61
  62 /* Embedded module documentation macros - see modules.h */
  63 MODULE_AUTHOR("Hewlett-Packard Company");
  64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  65 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  66 MODULE_VERSION("3.6.26");
  67 MODULE_LICENSE("GPL");
  68
  69 static int cciss_allow_hpsa;
  70 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  71 MODULE_PARM_DESC(cciss_allow_hpsa,
  72         "Prevent cciss driver from accessing hardware known to be "
  73         " supported by the hpsa driver");
  74
  75 #include "cciss_cmd.h"
  76 #include "cciss.h"
  77 #include <linux/cciss_ioctl.h>
  78
  79 /* define the PCI info for the cards we can control */
  80 static const struct pci_device_id cciss_pci_device_id[] = {
  81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3250},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3251},
 110         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3252},
 111         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3253},
 112         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3254},
 113         {0,}
 114 };
 115
 116 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 117
 118 /*  board_id = Subsystem Device ID & Vendor ID
 119  *  product = Marketing Name for the board
 120  *  access = Address of the struct of function pointers
 121  */
 122 static struct board_type products[] = {
 123         {0x40700E11, "Smart Array 5300", &SA5_access},
 124         {0x40800E11, "Smart Array 5i", &SA5B_access},
 125         {0x40820E11, "Smart Array 532", &SA5B_access},
 126         {0x40830E11, "Smart Array 5312", &SA5B_access},
 127         {0x409A0E11, "Smart Array 641", &SA5_access},
 128         {0x409B0E11, "Smart Array 642", &SA5_access},
 129         {0x409C0E11, "Smart Array 6400", &SA5_access},
 130         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 131         {0x40910E11, "Smart Array 6i", &SA5_access},
 132         {0x3225103C, "Smart Array P600", &SA5_access},
 133         {0x3235103C, "Smart Array P400i", &SA5_access},
 134         {0x3211103C, "Smart Array E200i", &SA5_access},
 135         {0x3212103C, "Smart Array E200", &SA5_access},
 136         {0x3213103C, "Smart Array E200i", &SA5_access},
 137         {0x3214103C, "Smart Array E200i", &SA5_access},
 138         {0x3215103C, "Smart Array E200i", &SA5_access},
 139         {0x3237103C, "Smart Array E500", &SA5_access},
 140 /* controllers below this line are also supported by the hpsa driver. */
 141 #define HPSA_BOUNDARY 0x3223103C
 142         {0x3223103C, "Smart Array P800", &SA5_access},
 143         {0x3234103C, "Smart Array P400", &SA5_access},
 144         {0x323D103C, "Smart Array P700m", &SA5_access},
 145         {0x3241103C, "Smart Array P212", &SA5_access},
 146         {0x3243103C, "Smart Array P410", &SA5_access},
 147         {0x3245103C, "Smart Array P410i", &SA5_access},
 148         {0x3247103C, "Smart Array P411", &SA5_access},
 149         {0x3249103C, "Smart Array P812", &SA5_access},
 150         {0x324A103C, "Smart Array P712m", &SA5_access},
 151         {0x324B103C, "Smart Array P711m", &SA5_access},
 152         {0x3250103C, "Smart Array", &SA5_access},
 153         {0x3251103C, "Smart Array", &SA5_access},
 154         {0x3252103C, "Smart Array", &SA5_access},
 155         {0x3253103C, "Smart Array", &SA5_access},
 156         {0x3254103C, "Smart Array", &SA5_access},
 157 };
 158
 159 /* How long to wait (in milliseconds) for board to go into simple mode */
 160 #define MAX_CONFIG_WAIT 30000
 161 #define MAX_IOCTL_CONFIG_WAIT 1000
 162
 163 /*define how many times we will try a command because of bus resets */
 164 #define MAX_CMD_RETRIES 3
 165
 166 #define MAX_CTLR        32
 167
 168 /* Originally cciss driver only supports 8 major numbers */
 169 #define MAX_CTLR_ORIG   8
 170
 171 static ctlr_info_t *hba[MAX_CTLR];
 172
 173 static struct task_struct *cciss_scan_thread;
 174 static DEFINE_MUTEX(scan_mutex);
 175 static LIST_HEAD(scan_q);
 176
 177 static void do_cciss_request(struct request_queue *q);
 178 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 179 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 180 static int cciss_open(struct block_device *bdev, fmode_t mode);
 181 static int cciss_release(struct gendisk *disk, fmode_t mode);
 182 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 183                        unsigned int cmd, unsigned long arg);
 184 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 185
 186 static int cciss_revalidate(struct gendisk *disk);
 187 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 188 static int deregister_disk(ctlr_info_t *h, int drv_index,
 189                            int clear_all, int via_ioctl);
 190
 191 static void cciss_read_capacity(int ctlr, int logvol,
 192                         sector_t *total_size, unsigned int *block_size);
 193 static void cciss_read_capacity_16(int ctlr, int logvol,
 194                         sector_t *total_size, unsigned int *block_size);
 195 static void cciss_geometry_inquiry(int ctlr, int logvol,
 196                         sector_t total_size,
 197                         unsigned int block_size, InquiryData_struct *inq_buff,
 198                                    drive_info_struct *drv);
 199 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 200                                            __u32);
 201 static void start_io(ctlr_info_t *h);
 202 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 203                         __u8 page_code, unsigned char scsi3addr[],
 204                         int cmd_type);
 205 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 206         int attempt_retry);
 207 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 208
 209 static int add_to_scan_list(struct ctlr_info *h);
 210 static int scan_thread(void *data);
 211 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 212 static void cciss_hba_release(struct device *dev);
 213 static void cciss_device_release(struct device *dev);
 214 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 215 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 216
 217 /* performant mode helper functions */
 218 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 219                                 int *bucket_map);
 220 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 221
 222 #ifdef CONFIG_PROC_FS
 223 static void cciss_procinit(int i);
 224 #else
 225 static void cciss_procinit(int i)
 226 {
 227 }
 228 #endif                          /* CONFIG_PROC_FS */
 229
 230 #ifdef CONFIG_COMPAT
 231 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 232                               unsigned, unsigned long);
 233 #endif
 234
 235 static const struct block_device_operations cciss_fops = {
 236         .owner = THIS_MODULE,
 237         .open = cciss_open,
 238         .release = cciss_release,
 239         .locked_ioctl = cciss_ioctl,
 240         .getgeo = cciss_getgeo,
 241 #ifdef CONFIG_COMPAT
 242         .compat_ioctl = cciss_compat_ioctl,
 243 #endif
 244         .revalidate_disk = cciss_revalidate,
 245 };
 246
 247 /* set_performant_mode: Modify the tag for cciss performant
 248  * set bit 0 for pull model, bits 3-1 for block fetch
 249  * register number
 250  */
 251 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 252 {
 253         if (likely(h->transMethod == CFGTBL_Trans_Performant))
 254                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 255 }
 256
 257 /*
 258  * Enqueuing and dequeuing functions for cmdlists.
 259  */
 260 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 261 {
 262         hlist_add_head(&c->list, list);
 263 }
 264
 265 static inline void removeQ(CommandList_struct *c)
 266 {
 267         /*
 268          * After kexec/dump some commands might still
 269          * be in flight, which the firmware will try
 270          * to complete. Resetting the firmware doesn't work
 271          * with old fw revisions, so we have to mark
 272          * them off as 'stale' to prevent the driver from
 273          * falling over.
 274          */
 275         if (WARN_ON(hlist_unhashed(&c->list))) {
 276                 c->cmd_type = CMD_MSG_STALE;
 277                 return;
 278         }
 279
 280         hlist_del_init(&c->list);
 281 }
 282
 283 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 284         CommandList_struct *c)
 285 {
 286         unsigned long flags;
 287         set_performant_mode(h, c);
 288         spin_lock_irqsave(&h->lock, flags);
 289         addQ(&h->reqQ, c);
 290         h->Qdepth++;
 291         start_io(h);
 292         spin_unlock_irqrestore(&h->lock, flags);
 293 }
 294
 295 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 296         int nr_cmds)
 297 {
 298         int i;
 299
 300         if (!cmd_sg_list)
 301                 return;
 302         for (i = 0; i < nr_cmds; i++) {
 303                 kfree(cmd_sg_list[i]);
 304                 cmd_sg_list[i] = NULL;
 305         }
 306         kfree(cmd_sg_list);
 307 }
 308
 309 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 310         ctlr_info_t *h, int chainsize, int nr_cmds)
 311 {
 312         int j;
 313         SGDescriptor_struct **cmd_sg_list;
 314
 315         if (chainsize <= 0)
 316                 return NULL;
 317
 318         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 319         if (!cmd_sg_list)
 320                 return NULL;
 321
 322         /* Build up chain blocks for each command */
 323         for (j = 0; j < nr_cmds; j++) {
 324                 /* Need a block of chainsized s/g elements. */
 325                 cmd_sg_list[j] = kmalloc((chainsize *
 326                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 327                 if (!cmd_sg_list[j]) {
 328                         dev_err(&h->pdev->dev, "Cannot get memory "
 329                                 "for s/g chains.\n");
 330                         goto clean;
 331                 }
 332         }
 333         return cmd_sg_list;
 334 clean:
 335         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 336         return NULL;
 337 }
 338
 339 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 340 {
 341         SGDescriptor_struct *chain_sg;
 342         u64bit temp64;
 343
 344         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 345                 return;
 346
 347         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 348         temp64.val32.lower = chain_sg->Addr.lower;
 349         temp64.val32.upper = chain_sg->Addr.upper;
 350         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 351 }
 352
 353 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 354         SGDescriptor_struct *chain_block, int len)
 355 {
 356         SGDescriptor_struct *chain_sg;
 357         u64bit temp64;
 358
 359         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 360         chain_sg->Ext = CCISS_SG_CHAIN;
 361         chain_sg->Len = len;
 362         temp64.val = pci_map_single(h->pdev, chain_block, len,
 363                                 PCI_DMA_TODEVICE);
 364         chain_sg->Addr.lower = temp64.val32.lower;
 365         chain_sg->Addr.upper = temp64.val32.upper;
 366 }
 367
 368 #include "cciss_scsi.c"         /* For SCSI tape support */
 369
 370 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 371         "UNKNOWN"
 372 };
 373 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
 374
 375 #ifdef CONFIG_PROC_FS
 376
 377 static inline u32 next_command(ctlr_info_t *h)
 378 {
 379         u32 a;
 380
 381         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
 382                 return h->access.command_completed(h);
 383
 384         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
 385                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
 386                 (h->reply_pool_head)++;
 387                 h->commands_outstanding--;
 388         } else {
 389                 a = FIFO_EMPTY;
 390         }
 391         /* Check for wraparound */
 392         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
 393                 h->reply_pool_head = h->reply_pool;
 394                 h->reply_pool_wraparound ^= 1;
 395         }
 396         return a;
 397 }
 398
 399 /*
 400  * Report information about this controller.
 401  */
 402 #define ENG_GIG 1000000000
 403 #define ENG_GIG_FACTOR (ENG_GIG/512)
 404 #define ENGAGE_SCSI     "engage scsi"
 405
 406 static struct proc_dir_entry *proc_cciss;
 407
 408 static void cciss_seq_show_header(struct seq_file *seq)
 409 {
 410         ctlr_info_t *h = seq->private;
 411
 412         seq_printf(seq, "%s: HP %s Controller\n"
 413                 "Board ID: 0x%08lx\n"
 414                 "Firmware Version: %c%c%c%c\n"
 415                 "IRQ: %d\n"
 416                 "Logical drives: %d\n"
 417                 "Current Q depth: %d\n"
 418                 "Current # commands on controller: %d\n"
 419                 "Max Q depth since init: %d\n"
 420                 "Max # commands on controller since init: %d\n"
 421                 "Max SG entries since init: %d\n",
 422                 h->devname,
 423                 h->product_name,
 424                 (unsigned long)h->board_id,
 425                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 426                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 427                 h->num_luns,
 428                 h->Qdepth, h->commands_outstanding,
 429                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 430
 431 #ifdef CONFIG_CISS_SCSI_TAPE
 432         cciss_seq_tape_report(seq, h->ctlr);
 433 #endif /* CONFIG_CISS_SCSI_TAPE */
 434 }
 435
 436 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 437 {
 438         ctlr_info_t *h = seq->private;
 439         unsigned ctlr = h->ctlr;
 440         unsigned long flags;
 441
 442         /* prevent displaying bogus info during configuration
 443          * or deconfiguration of a logical volume
 444          */
 445         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 446         if (h->busy_configuring) {
 447                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 448                 return ERR_PTR(-EBUSY);
 449         }
 450         h->busy_configuring = 1;
 451         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 452
 453         if (*pos == 0)
 454                 cciss_seq_show_header(seq);
 455
 456         return pos;
 457 }
 458
 459 static int cciss_seq_show(struct seq_file *seq, void *v)
 460 {
 461         sector_t vol_sz, vol_sz_frac;
 462         ctlr_info_t *h = seq->private;
 463         unsigned ctlr = h->ctlr;
 464         loff_t *pos = v;
 465         drive_info_struct *drv = h->drv[*pos];
 466
 467         if (*pos > h->highest_lun)
 468                 return 0;
 469
 470         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 471                 return 0;
 472
 473         if (drv->heads == 0)
 474                 return 0;
 475
 476         vol_sz = drv->nr_blocks;
 477         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 478         vol_sz_frac *= 100;
 479         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 480
 481         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 482                 drv->raid_level = RAID_UNKNOWN;
 483         seq_printf(seq, "cciss/c%dd%d:"
 484                         "\t%4u.%02uGB\tRAID %s\n",
 485                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 486                         raid_label[drv->raid_level]);
 487         return 0;
 488 }
 489
 490 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 491 {
 492         ctlr_info_t *h = seq->private;
 493
 494         if (*pos > h->highest_lun)
 495                 return NULL;
 496         *pos += 1;
 497
 498         return pos;
 499 }
 500
 501 static void cciss_seq_stop(struct seq_file *seq, void *v)
 502 {
 503         ctlr_info_t *h = seq->private;
 504
 505         /* Only reset h->busy_configuring if we succeeded in setting
 506          * it during cciss_seq_start. */
 507         if (v == ERR_PTR(-EBUSY))
 508                 return;
 509
 510         h->busy_configuring = 0;
 511 }
 512
 513 static const struct seq_operations cciss_seq_ops = {
 514         .start = cciss_seq_start,
 515         .show  = cciss_seq_show,
 516         .next  = cciss_seq_next,
 517         .stop  = cciss_seq_stop,
 518 };
 519
 520 static int cciss_seq_open(struct inode *inode, struct file *file)
 521 {
 522         int ret = seq_open(file, &cciss_seq_ops);
 523         struct seq_file *seq = file->private_data;
 524
 525         if (!ret)
 526                 seq->private = PDE(inode)->data;
 527
 528         return ret;
 529 }
 530
 531 static ssize_t
 532 cciss_proc_write(struct file *file, const char __user *buf,
 533                  size_t length, loff_t *ppos)
 534 {
 535         int err;
 536         char *buffer;
 537
 538 #ifndef CONFIG_CISS_SCSI_TAPE
 539         return -EINVAL;
 540 #endif
 541
 542         if (!buf || length > PAGE_SIZE - 1)
 543                 return -EINVAL;
 544
 545         buffer = (char *)__get_free_page(GFP_KERNEL);
 546         if (!buffer)
 547                 return -ENOMEM;
 548
 549         err = -EFAULT;
 550         if (copy_from_user(buffer, buf, length))
 551                 goto out;
 552         buffer[length] = '\0';
 553
 554 #ifdef CONFIG_CISS_SCSI_TAPE
 555         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 556                 struct seq_file *seq = file->private_data;
 557                 ctlr_info_t *h = seq->private;
 558
 559                 err = cciss_engage_scsi(h->ctlr);
 560                 if (err == 0)
 561                         err = length;
 562         } else
 563 #endif /* CONFIG_CISS_SCSI_TAPE */
 564                 err = -EINVAL;
 565         /* might be nice to have "disengage" too, but it's not
 566            safely possible. (only 1 module use count, lock issues.) */
 567
 568 out:
 569         free_page((unsigned long)buffer);
 570         return err;
 571 }
 572
 573 static const struct file_operations cciss_proc_fops = {
 574         .owner   = THIS_MODULE,
 575         .open    = cciss_seq_open,
 576         .read    = seq_read,
 577         .llseek  = seq_lseek,
 578         .release = seq_release,
 579         .write   = cciss_proc_write,
 580 };
 581
 582 static void __devinit cciss_procinit(int i)
 583 {
 584         struct proc_dir_entry *pde;
 585
 586         if (proc_cciss == NULL)
 587                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 588         if (!proc_cciss)
 589                 return;
 590         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 591                                         S_IROTH, proc_cciss,
 592                                         &cciss_proc_fops, hba[i]);
 593 }
 594 #endif                          /* CONFIG_PROC_FS */
 595
 596 #define MAX_PRODUCT_NAME_LEN 19
 597
 598 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 599 #define to_drv(n) container_of(n, drive_info_struct, dev)
 600
 601 static ssize_t host_store_rescan(struct device *dev,
 602                                  struct device_attribute *attr,
 603                                  const char *buf, size_t count)
 604 {
 605         struct ctlr_info *h = to_hba(dev);
 606
 607         add_to_scan_list(h);
 608         wake_up_process(cciss_scan_thread);
 609         wait_for_completion_interruptible(&h->scan_wait);
 610
 611         return count;
 612 }
 613 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 614
 615 static ssize_t dev_show_unique_id(struct device *dev,
 616                                  struct device_attribute *attr,
 617                                  char *buf)
 618 {
 619         drive_info_struct *drv = to_drv(dev);
 620         struct ctlr_info *h = to_hba(drv->dev.parent);
 621         __u8 sn[16];
 622         unsigned long flags;
 623         int ret = 0;
 624
 625         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 626         if (h->busy_configuring)
 627                 ret = -EBUSY;
 628         else
 629                 memcpy(sn, drv->serial_no, sizeof(sn));
 630         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 631
 632         if (ret)
 633                 return ret;
 634         else
 635                 return snprintf(buf, 16 * 2 + 2,
 636                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 637                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 638                                 sn[0], sn[1], sn[2], sn[3],
 639                                 sn[4], sn[5], sn[6], sn[7],
 640                                 sn[8], sn[9], sn[10], sn[11],
 641                                 sn[12], sn[13], sn[14], sn[15]);
 642 }
 643 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 644
 645 static ssize_t dev_show_vendor(struct device *dev,
 646                                struct device_attribute *attr,
 647                                char *buf)
 648 {
 649         drive_info_struct *drv = to_drv(dev);
 650         struct ctlr_info *h = to_hba(drv->dev.parent);
 651         char vendor[VENDOR_LEN + 1];
 652         unsigned long flags;
 653         int ret = 0;
 654
 655         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 656         if (h->busy_configuring)
 657                 ret = -EBUSY;
 658         else
 659                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 660         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 661
 662         if (ret)
 663                 return ret;
 664         else
 665                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 666 }
 667 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 668
 669 static ssize_t dev_show_model(struct device *dev,
 670                               struct device_attribute *attr,
 671                               char *buf)
 672 {
 673         drive_info_struct *drv = to_drv(dev);
 674         struct ctlr_info *h = to_hba(drv->dev.parent);
 675         char model[MODEL_LEN + 1];
 676         unsigned long flags;
 677         int ret = 0;
 678
 679         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 680         if (h->busy_configuring)
 681                 ret = -EBUSY;
 682         else
 683                 memcpy(model, drv->model, MODEL_LEN + 1);
 684         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 685
 686         if (ret)
 687                 return ret;
 688         else
 689                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 690 }
 691 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 692
 693 static ssize_t dev_show_rev(struct device *dev,
 694                             struct device_attribute *attr,
 695                             char *buf)
 696 {
 697         drive_info_struct *drv = to_drv(dev);
 698         struct ctlr_info *h = to_hba(drv->dev.parent);
 699         char rev[REV_LEN + 1];
 700         unsigned long flags;
 701         int ret = 0;
 702
 703         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 704         if (h->busy_configuring)
 705                 ret = -EBUSY;
 706         else
 707                 memcpy(rev, drv->rev, REV_LEN + 1);
 708         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 709
 710         if (ret)
 711                 return ret;
 712         else
 713                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 714 }
 715 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 716
 717 static ssize_t cciss_show_lunid(struct device *dev,
 718                                 struct device_attribute *attr, char *buf)
 719 {
 720         drive_info_struct *drv = to_drv(dev);
 721         struct ctlr_info *h = to_hba(drv->dev.parent);
 722         unsigned long flags;
 723         unsigned char lunid[8];
 724
 725         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 726         if (h->busy_configuring) {
 727                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 728                 return -EBUSY;
 729         }
 730         if (!drv->heads) {
 731                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 732                 return -ENOTTY;
 733         }
 734         memcpy(lunid, drv->LunID, sizeof(lunid));
 735         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 736         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 737                 lunid[0], lunid[1], lunid[2], lunid[3],
 738                 lunid[4], lunid[5], lunid[6], lunid[7]);
 739 }
 740 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 741
 742 static ssize_t cciss_show_raid_level(struct device *dev,
 743                                      struct device_attribute *attr, char *buf)
 744 {
 745         drive_info_struct *drv = to_drv(dev);
 746         struct ctlr_info *h = to_hba(drv->dev.parent);
 747         int raid;
 748         unsigned long flags;
 749
 750         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 751         if (h->busy_configuring) {
 752                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 753                 return -EBUSY;
 754         }
 755         raid = drv->raid_level;
 756         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 757         if (raid < 0 || raid > RAID_UNKNOWN)
 758                 raid = RAID_UNKNOWN;
 759
 760         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 761                         raid_label[raid]);
 762 }
 763 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 764
 765 static ssize_t cciss_show_usage_count(struct device *dev,
 766                                       struct device_attribute *attr, char *buf)
 767 {
 768         drive_info_struct *drv = to_drv(dev);
 769         struct ctlr_info *h = to_hba(drv->dev.parent);
 770         unsigned long flags;
 771         int count;
 772
 773         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 774         if (h->busy_configuring) {
 775                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 776                 return -EBUSY;
 777         }
 778         count = drv->usage_count;
 779         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 780         return snprintf(buf, 20, "%d\n", count);
 781 }
 782 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 783
 784 static struct attribute *cciss_host_attrs[] = {
 785         &dev_attr_rescan.attr,
 786         NULL
 787 };
 788
 789 static struct attribute_group cciss_host_attr_group = {
 790         .attrs = cciss_host_attrs,
 791 };
 792
 793 static const struct attribute_group *cciss_host_attr_groups[] = {
 794         &cciss_host_attr_group,
 795         NULL
 796 };
 797
 798 static struct device_type cciss_host_type = {
 799         .name           = "cciss_host",
 800         .groups         = cciss_host_attr_groups,
 801         .release        = cciss_hba_release,
 802 };
 803
 804 static struct attribute *cciss_dev_attrs[] = {
 805         &dev_attr_unique_id.attr,
 806         &dev_attr_model.attr,
 807         &dev_attr_vendor.attr,
 808         &dev_attr_rev.attr,
 809         &dev_attr_lunid.attr,
 810         &dev_attr_raid_level.attr,
 811         &dev_attr_usage_count.attr,
 812         NULL
 813 };
 814
 815 static struct attribute_group cciss_dev_attr_group = {
 816         .attrs = cciss_dev_attrs,
 817 };
 818
 819 static const struct attribute_group *cciss_dev_attr_groups[] = {
 820         &cciss_dev_attr_group,
 821         NULL
 822 };
 823
 824 static struct device_type cciss_dev_type = {
 825         .name           = "cciss_device",
 826         .groups         = cciss_dev_attr_groups,
 827         .release        = cciss_device_release,
 828 };
 829
 830 static struct bus_type cciss_bus_type = {
 831         .name           = "cciss",
 832 };
 833
 834 /*
 835  * cciss_hba_release is called when the reference count
 836  * of h->dev goes to zero.
 837  */
 838 static void cciss_hba_release(struct device *dev)
 839 {
 840         /*
 841          * nothing to do, but need this to avoid a warning
 842          * about not having a release handler from lib/kref.c.
 843          */
 844 }
 845
 846 /*
 847  * Initialize sysfs entry for each controller.  This sets up and registers
 848  * the 'cciss#' directory for each individual controller under
 849  * /sys/bus/pci/devices/<dev>/.
 850  */
 851 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 852 {
 853         device_initialize(&h->dev);
 854         h->dev.type = &cciss_host_type;
 855         h->dev.bus = &cciss_bus_type;
 856         dev_set_name(&h->dev, "%s", h->devname);
 857         h->dev.parent = &h->pdev->dev;
 858
 859         return device_add(&h->dev);
 860 }
 861
 862 /*
 863  * Remove sysfs entries for an hba.
 864  */
 865 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 866 {
 867         device_del(&h->dev);
 868         put_device(&h->dev); /* final put. */
 869 }
 870
 871 /* cciss_device_release is called when the reference count
 872  * of h->drv[x]dev goes to zero.
 873  */
 874 static void cciss_device_release(struct device *dev)
 875 {
 876         drive_info_struct *drv = to_drv(dev);
 877         kfree(drv);
 878 }
 879
 880 /*
 881  * Initialize sysfs for each logical drive.  This sets up and registers
 882  * the 'c#d#' directory for each individual logical drive under
 883  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 884  * /sys/block/cciss!c#d# to this entry.
 885  */
 886 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 887                                        int drv_index)
 888 {
 889         struct device *dev;
 890
 891         if (h->drv[drv_index]->device_initialized)
 892                 return 0;
 893
 894         dev = &h->drv[drv_index]->dev;
 895         device_initialize(dev);
 896         dev->type = &cciss_dev_type;
 897         dev->bus = &cciss_bus_type;
 898         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 899         dev->parent = &h->dev;
 900         h->drv[drv_index]->device_initialized = 1;
 901         return device_add(dev);
 902 }
 903
 904 /*
 905  * Remove sysfs entries for a logical drive.
 906  */
 907 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 908         int ctlr_exiting)
 909 {
 910         struct device *dev = &h->drv[drv_index]->dev;
 911
 912         /* special case for c*d0, we only destroy it on controller exit */
 913         if (drv_index == 0 && !ctlr_exiting)
 914                 return;
 915
 916         device_del(dev);
 917         put_device(dev); /* the "final" put. */
 918         h->drv[drv_index] = NULL;
 919 }
 920
 921 /*
 922  * For operations that cannot sleep, a command block is allocated at init,
 923  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 924  * which ones are free or in use.  For operations that can wait for kmalloc
 925  * to possible sleep, this routine can be called with get_from_pool set to 0.
 926  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 927  */
 928 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 929 {
 930         CommandList_struct *c;
 931         int i;
 932         u64bit temp64;
 933         dma_addr_t cmd_dma_handle, err_dma_handle;
 934
 935         if (!get_from_pool) {
 936                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 937                         sizeof(CommandList_struct), &cmd_dma_handle);
 938                 if (c == NULL)
 939                         return NULL;
 940                 memset(c, 0, sizeof(CommandList_struct));
 941
 942                 c->cmdindex = -1;
 943
 944                 c->err_info = (ErrorInfo_struct *)
 945                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 946                             &err_dma_handle);
 947
 948                 if (c->err_info == NULL) {
 949                         pci_free_consistent(h->pdev,
 950                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 951                         return NULL;
 952                 }
 953                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 954         } else {                /* get it out of the controllers pool */
 955
 956                 do {
 957                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 958                         if (i == h->nr_cmds)
 959                                 return NULL;
 960                 } while (test_and_set_bit
 961                          (i & (BITS_PER_LONG - 1),
 962                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 963 #ifdef CCISS_DEBUG
 964                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 965 #endif
 966                 c = h->cmd_pool + i;
 967                 memset(c, 0, sizeof(CommandList_struct));
 968                 cmd_dma_handle = h->cmd_pool_dhandle
 969                     + i * sizeof(CommandList_struct);
 970                 c->err_info = h->errinfo_pool + i;
 971                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 972                 err_dma_handle = h->errinfo_pool_dhandle
 973                     + i * sizeof(ErrorInfo_struct);
 974                 h->nr_allocs++;
 975
 976                 c->cmdindex = i;
 977         }
 978
 979         INIT_HLIST_NODE(&c->list);
 980         c->busaddr = (__u32) cmd_dma_handle;
 981         temp64.val = (__u64) err_dma_handle;
 982         c->ErrDesc.Addr.lower = temp64.val32.lower;
 983         c->ErrDesc.Addr.upper = temp64.val32.upper;
 984         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 985
 986         c->ctlr = h->ctlr;
 987         return c;
 988 }
 989
 990 /*
 991  * Frees a command block that was previously allocated with cmd_alloc().
 992  */
 993 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 994 {
 995         int i;
 996         u64bit temp64;
 997
 998         if (!got_from_pool) {
 999                 temp64.val32.lower = c->ErrDesc.Addr.lower;
1000                 temp64.val32.upper = c->ErrDesc.Addr.upper;
1001                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1002                                     c->err_info, (dma_addr_t) temp64.val);
1003                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
1004                                     c, (dma_addr_t) c->busaddr);
1005         } else {
1006                 i = c - h->cmd_pool;
1007                 clear_bit(i & (BITS_PER_LONG - 1),
1008                           h->cmd_pool_bits + (i / BITS_PER_LONG));
1009                 h->nr_frees++;
1010         }
1011 }
1012
1013 static inline ctlr_info_t *get_host(struct gendisk *disk)
1014 {
1015         return disk->queue->queuedata;
1016 }
1017
1018 static inline drive_info_struct *get_drv(struct gendisk *disk)
1019 {
1020         return disk->private_data;
1021 }
1022
1023 /*
1024  * Open.  Make sure the device is really there.
1025  */
1026 static int cciss_open(struct block_device *bdev, fmode_t mode)
1027 {
1028         ctlr_info_t *host = get_host(bdev->bd_disk);
1029         drive_info_struct *drv = get_drv(bdev->bd_disk);
1030
1031 #ifdef CCISS_DEBUG
1032         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
1033 #endif                          /* CCISS_DEBUG */
1034
1035         if (drv->busy_configuring)
1036                 return -EBUSY;
1037         /*
1038          * Root is allowed to open raw volume zero even if it's not configured
1039          * so array config can still work. Root is also allowed to open any
1040          * volume that has a LUN ID, so it can issue IOCTL to reread the
1041          * disk information.  I don't think I really like this
1042          * but I'm already using way to many device nodes to claim another one
1043          * for "raw controller".
1044          */
1045         if (drv->heads == 0) {
1046                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1047                         /* if not node 0 make sure it is a partition = 0 */
1048                         if (MINOR(bdev->bd_dev) & 0x0f) {
1049                                 return -ENXIO;
1050                                 /* if it is, make sure we have a LUN ID */
1051                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1052                                 sizeof(drv->LunID))) {
1053                                 return -ENXIO;
1054                         }
1055                 }
1056                 if (!capable(CAP_SYS_ADMIN))
1057                         return -EPERM;
1058         }
1059         drv->usage_count++;
1060         host->usage_count++;
1061         return 0;
1062 }
1063
1064 /*
1065  * Close.  Sync first.
1066  */
1067 static int cciss_release(struct gendisk *disk, fmode_t mode)
1068 {
1069         ctlr_info_t *host = get_host(disk);
1070         drive_info_struct *drv = get_drv(disk);
1071
1072 #ifdef CCISS_DEBUG
1073         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
1074 #endif                          /* CCISS_DEBUG */
1075
1076         drv->usage_count--;
1077         host->usage_count--;
1078         return 0;
1079 }
1080
1081 #ifdef CONFIG_COMPAT
1082
1083 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1084                     unsigned cmd, unsigned long arg)
1085 {
1086         int ret;
1087         lock_kernel();
1088         ret = cciss_ioctl(bdev, mode, cmd, arg);
1089         unlock_kernel();
1090         return ret;
1091 }
1092
1093 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1094                                   unsigned cmd, unsigned long arg);
1095 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1096                                       unsigned cmd, unsigned long arg);
1097
1098 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1099                               unsigned cmd, unsigned long arg)
1100 {
1101         switch (cmd) {
1102         case CCISS_GETPCIINFO:
1103         case CCISS_GETINTINFO:
1104         case CCISS_SETINTINFO:
1105         case CCISS_GETNODENAME:
1106         case CCISS_SETNODENAME:
1107         case CCISS_GETHEARTBEAT:
1108         case CCISS_GETBUSTYPES:
1109         case CCISS_GETFIRMVER:
1110         case CCISS_GETDRIVVER:
1111         case CCISS_REVALIDVOLS:
1112         case CCISS_DEREGDISK:
1113         case CCISS_REGNEWDISK:
1114         case CCISS_REGNEWD:
1115         case CCISS_RESCANDISK:
1116         case CCISS_GETLUNINFO:
1117                 return do_ioctl(bdev, mode, cmd, arg);
1118
1119         case CCISS_PASSTHRU32:
1120                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1121         case CCISS_BIG_PASSTHRU32:
1122                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1123
1124         default:
1125                 return -ENOIOCTLCMD;
1126         }
1127 }
1128
1129 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1130                                   unsigned cmd, unsigned long arg)
1131 {
1132         IOCTL32_Command_struct __user *arg32 =
1133             (IOCTL32_Command_struct __user *) arg;
1134         IOCTL_Command_struct arg64;
1135         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1136         int err;
1137         u32 cp;
1138
1139         err = 0;
1140         err |=
1141             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1142                            sizeof(arg64.LUN_info));
1143         err |=
1144             copy_from_user(&arg64.Request, &arg32->Request,
1145                            sizeof(arg64.Request));
1146         err |=
1147             copy_from_user(&arg64.error_info, &arg32->error_info,
1148                            sizeof(arg64.error_info));
1149         err |= get_user(arg64.buf_size, &arg32->buf_size);
1150         err |= get_user(cp, &arg32->buf);
1151         arg64.buf = compat_ptr(cp);
1152         err |= copy_to_user(p, &arg64, sizeof(arg64));
1153
1154         if (err)
1155                 return -EFAULT;
1156
1157         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1158         if (err)
1159                 return err;
1160         err |=
1161             copy_in_user(&arg32->error_info, &p->error_info,
1162                          sizeof(arg32->error_info));
1163         if (err)
1164                 return -EFAULT;
1165         return err;
1166 }
1167
1168 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1169                                       unsigned cmd, unsigned long arg)
1170 {
1171         BIG_IOCTL32_Command_struct __user *arg32 =
1172             (BIG_IOCTL32_Command_struct __user *) arg;
1173         BIG_IOCTL_Command_struct arg64;
1174         BIG_IOCTL_Command_struct __user *p =
1175             compat_alloc_user_space(sizeof(arg64));
1176         int err;
1177         u32 cp;
1178
1179         err = 0;
1180         err |=
1181             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1182                            sizeof(arg64.LUN_info));
1183         err |=
1184             copy_from_user(&arg64.Request, &arg32->Request,
1185                            sizeof(arg64.Request));
1186         err |=
1187             copy_from_user(&arg64.error_info, &arg32->error_info,
1188                            sizeof(arg64.error_info));
1189         err |= get_user(arg64.buf_size, &arg32->buf_size);
1190         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1191         err |= get_user(cp, &arg32->buf);
1192         arg64.buf = compat_ptr(cp);
1193         err |= copy_to_user(p, &arg64, sizeof(arg64));
1194
1195         if (err)
1196                 return -EFAULT;
1197
1198         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1199         if (err)
1200                 return err;
1201         err |=
1202             copy_in_user(&arg32->error_info, &p->error_info,
1203                          sizeof(arg32->error_info));
1204         if (err)
1205                 return -EFAULT;
1206         return err;
1207 }
1208 #endif
1209
1210 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1211 {
1212         drive_info_struct *drv = get_drv(bdev->bd_disk);
1213
1214         if (!drv->cylinders)
1215                 return -ENXIO;
1216
1217         geo->heads = drv->heads;
1218         geo->sectors = drv->sectors;
1219         geo->cylinders = drv->cylinders;
1220         return 0;
1221 }
1222
1223 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1224 {
1225         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1226                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1227                 (void)check_for_unit_attention(host, c);
1228 }
1229 /*
1230  * ioctl
1231  */
1232 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1233                        unsigned int cmd, unsigned long arg)
1234 {
1235         struct gendisk *disk = bdev->bd_disk;
1236         ctlr_info_t *host = get_host(disk);
1237         drive_info_struct *drv = get_drv(disk);
1238         int ctlr = host->ctlr;
1239         void __user *argp = (void __user *)arg;
1240
1241 #ifdef CCISS_DEBUG
1242         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1243 #endif                          /* CCISS_DEBUG */
1244
1245         switch (cmd) {
1246         case CCISS_GETPCIINFO:
1247                 {
1248                         cciss_pci_info_struct pciinfo;
1249
1250                         if (!arg)
1251                                 return -EINVAL;
1252                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1253                         pciinfo.bus = host->pdev->bus->number;
1254                         pciinfo.dev_fn = host->pdev->devfn;
1255                         pciinfo.board_id = host->board_id;
1256                         if (copy_to_user
1257                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1258                                 return -EFAULT;
1259                         return 0;
1260                 }
1261         case CCISS_GETINTINFO:
1262                 {
1263                         cciss_coalint_struct intinfo;
1264                         if (!arg)
1265                                 return -EINVAL;
1266                         intinfo.delay =
1267                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1268                         intinfo.count =
1269                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1270                         if (copy_to_user
1271                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1272                                 return -EFAULT;
1273                         return 0;
1274                 }
1275         case CCISS_SETINTINFO:
1276                 {
1277                         cciss_coalint_struct intinfo;
1278                         unsigned long flags;
1279                         int i;
1280
1281                         if (!arg)
1282                                 return -EINVAL;
1283                         if (!capable(CAP_SYS_ADMIN))
1284                                 return -EPERM;
1285                         if (copy_from_user
1286                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1287                                 return -EFAULT;
1288                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1289                         {
1290 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1291                                 return -EINVAL;
1292                         }
1293                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1294                         /* Update the field, and then ring the doorbell */
1295                         writel(intinfo.delay,
1296                                &(host->cfgtable->HostWrite.CoalIntDelay));
1297                         writel(intinfo.count,
1298                                &(host->cfgtable->HostWrite.CoalIntCount));
1299                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1300
1301                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1302                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1303                                       & CFGTBL_ChangeReq))
1304                                         break;
1305                                 /* delay and try again */
1306                                 udelay(1000);
1307                         }
1308                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1309                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1310                                 return -EAGAIN;
1311                         return 0;
1312                 }
1313         case CCISS_GETNODENAME:
1314                 {
1315                         NodeName_type NodeName;
1316                         int i;
1317
1318                         if (!arg)
1319                                 return -EINVAL;
1320                         for (i = 0; i < 16; i++)
1321                                 NodeName[i] =
1322                                     readb(&host->cfgtable->ServerName[i]);
1323                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1324                                 return -EFAULT;
1325                         return 0;
1326                 }
1327         case CCISS_SETNODENAME:
1328                 {
1329                         NodeName_type NodeName;
1330                         unsigned long flags;
1331                         int i;
1332
1333                         if (!arg)
1334                                 return -EINVAL;
1335                         if (!capable(CAP_SYS_ADMIN))
1336                                 return -EPERM;
1337
1338                         if (copy_from_user
1339                             (NodeName, argp, sizeof(NodeName_type)))
1340                                 return -EFAULT;
1341
1342                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1343
1344                         /* Update the field, and then ring the doorbell */
1345                         for (i = 0; i < 16; i++)
1346                                 writeb(NodeName[i],
1347                                        &host->cfgtable->ServerName[i]);
1348
1349                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1350
1351                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1352                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1353                                       & CFGTBL_ChangeReq))
1354                                         break;
1355                                 /* delay and try again */
1356                                 udelay(1000);
1357                         }
1358                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1359                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1360                                 return -EAGAIN;
1361                         return 0;
1362                 }
1363
1364         case CCISS_GETHEARTBEAT:
1365                 {
1366                         Heartbeat_type heartbeat;
1367
1368                         if (!arg)
1369                                 return -EINVAL;
1370                         heartbeat = readl(&host->cfgtable->HeartBeat);
1371                         if (copy_to_user
1372                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1373                                 return -EFAULT;
1374                         return 0;
1375                 }
1376         case CCISS_GETBUSTYPES:
1377                 {
1378                         BusTypes_type BusTypes;
1379
1380                         if (!arg)
1381                                 return -EINVAL;
1382                         BusTypes = readl(&host->cfgtable->BusTypes);
1383                         if (copy_to_user
1384                             (argp, &BusTypes, sizeof(BusTypes_type)))
1385                                 return -EFAULT;
1386                         return 0;
1387                 }
1388         case CCISS_GETFIRMVER:
1389                 {
1390                         FirmwareVer_type firmware;
1391
1392                         if (!arg)
1393                                 return -EINVAL;
1394                         memcpy(firmware, host->firm_ver, 4);
1395
1396                         if (copy_to_user
1397                             (argp, firmware, sizeof(FirmwareVer_type)))
1398                                 return -EFAULT;
1399                         return 0;
1400                 }
1401         case CCISS_GETDRIVVER:
1402                 {
1403                         DriverVer_type DriverVer = DRIVER_VERSION;
1404
1405                         if (!arg)
1406                                 return -EINVAL;
1407
1408                         if (copy_to_user
1409                             (argp, &DriverVer, sizeof(DriverVer_type)))
1410                                 return -EFAULT;
1411                         return 0;
1412                 }
1413
1414         case CCISS_DEREGDISK:
1415         case CCISS_REGNEWD:
1416         case CCISS_REVALIDVOLS:
1417                 return rebuild_lun_table(host, 0, 1);
1418
1419         case CCISS_GETLUNINFO:{
1420                         LogvolInfo_struct luninfo;
1421
1422                         memcpy(&luninfo.LunID, drv->LunID,
1423                                 sizeof(luninfo.LunID));
1424                         luninfo.num_opens = drv->usage_count;
1425                         luninfo.num_parts = 0;
1426                         if (copy_to_user(argp, &luninfo,
1427                                          sizeof(LogvolInfo_struct)))
1428                                 return -EFAULT;
1429                         return 0;
1430                 }
1431         case CCISS_PASSTHRU:
1432                 {
1433                         IOCTL_Command_struct iocommand;
1434                         CommandList_struct *c;
1435                         char *buff = NULL;
1436                         u64bit temp64;
1437                         DECLARE_COMPLETION_ONSTACK(wait);
1438
1439                         if (!arg)
1440                                 return -EINVAL;
1441
1442                         if (!capable(CAP_SYS_RAWIO))
1443                                 return -EPERM;
1444
1445                         if (copy_from_user
1446                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1447                                 return -EFAULT;
1448                         if ((iocommand.buf_size < 1) &&
1449                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1450                                 return -EINVAL;
1451                         }
1452 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1453                         /* Check kmalloc limits */
1454                         if (iocommand.buf_size > 128000)
1455                                 return -EINVAL;
1456 #endif
1457                         if (iocommand.buf_size > 0) {
1458                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1459                                 if (buff == NULL)
1460                                         return -EFAULT;
1461                         }
1462                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1463                                 /* Copy the data into the buffer we created */
1464                                 if (copy_from_user
1465                                     (buff, iocommand.buf, iocommand.buf_size)) {
1466                                         kfree(buff);
1467                                         return -EFAULT;
1468                                 }
1469                         } else {
1470                                 memset(buff, 0, iocommand.buf_size);
1471                         }
1472                         if ((c = cmd_alloc(host, 0)) == NULL) {
1473                                 kfree(buff);
1474                                 return -ENOMEM;
1475                         }
1476                         /* Fill in the command type */
1477                         c->cmd_type = CMD_IOCTL_PEND;
1478                         /* Fill in Command Header */
1479                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1480                         if (iocommand.buf_size > 0) /* buffer to fill */
1481                         {
1482                                 c->Header.SGList = 1;
1483                                 c->Header.SGTotal = 1;
1484                         } else /* no buffers to fill */
1485                         {
1486                                 c->Header.SGList = 0;
1487                                 c->Header.SGTotal = 0;
1488                         }
1489                         c->Header.LUN = iocommand.LUN_info;
1490                         /* use the kernel address the cmd block for tag */
1491                         c->Header.Tag.lower = c->busaddr;
1492
1493                         /* Fill in Request block */
1494                         c->Request = iocommand.Request;
1495
1496                         /* Fill in the scatter gather information */
1497                         if (iocommand.buf_size > 0) {
1498                                 temp64.val = pci_map_single(host->pdev, buff,
1499                                         iocommand.buf_size,
1500                                         PCI_DMA_BIDIRECTIONAL);
1501                                 c->SG[0].Addr.lower = temp64.val32.lower;
1502                                 c->SG[0].Addr.upper = temp64.val32.upper;
1503                                 c->SG[0].Len = iocommand.buf_size;
1504                                 c->SG[0].Ext = 0;  /* we are not chaining */
1505                         }
1506                         c->waiting = &wait;
1507
1508                         enqueue_cmd_and_start_io(host, c);
1509                         wait_for_completion(&wait);
1510
1511                         /* unlock the buffers from DMA */
1512                         temp64.val32.lower = c->SG[0].Addr.lower;
1513                         temp64.val32.upper = c->SG[0].Addr.upper;
1514                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1515                                          iocommand.buf_size,
1516                                          PCI_DMA_BIDIRECTIONAL);
1517
1518                         check_ioctl_unit_attention(host, c);
1519
1520                         /* Copy the error information out */
1521                         iocommand.error_info = *(c->err_info);
1522                         if (copy_to_user
1523                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1524                                 kfree(buff);
1525                                 cmd_free(host, c, 0);
1526                                 return -EFAULT;
1527                         }
1528
1529                         if (iocommand.Request.Type.Direction == XFER_READ) {
1530                                 /* Copy the data out of the buffer we created */
1531                                 if (copy_to_user
1532                                     (iocommand.buf, buff, iocommand.buf_size)) {
1533                                         kfree(buff);
1534                                         cmd_free(host, c, 0);
1535                                         return -EFAULT;
1536                                 }
1537                         }
1538                         kfree(buff);
1539                         cmd_free(host, c, 0);
1540                         return 0;
1541                 }
1542         case CCISS_BIG_PASSTHRU:{
1543                         BIG_IOCTL_Command_struct *ioc;
1544                         CommandList_struct *c;
1545                         unsigned char **buff = NULL;
1546                         int *buff_size = NULL;
1547                         u64bit temp64;
1548                         BYTE sg_used = 0;
1549                         int status = 0;
1550                         int i;
1551                         DECLARE_COMPLETION_ONSTACK(wait);
1552                         __u32 left;
1553                         __u32 sz;
1554                         BYTE __user *data_ptr;
1555
1556                         if (!arg)
1557                                 return -EINVAL;
1558                         if (!capable(CAP_SYS_RAWIO))
1559                                 return -EPERM;
1560                         ioc = (BIG_IOCTL_Command_struct *)
1561                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1562                         if (!ioc) {
1563                                 status = -ENOMEM;
1564                                 goto cleanup1;
1565                         }
1566                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1567                                 status = -EFAULT;
1568                                 goto cleanup1;
1569                         }
1570                         if ((ioc->buf_size < 1) &&
1571                             (ioc->Request.Type.Direction != XFER_NONE)) {
1572                                 status = -EINVAL;
1573                                 goto cleanup1;
1574                         }
1575                         /* Check kmalloc limits  using all SGs */
1576                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1577                                 status = -EINVAL;
1578                                 goto cleanup1;
1579                         }
1580                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1581                                 status = -EINVAL;
1582                                 goto cleanup1;
1583                         }
1584                         buff =
1585                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1586                         if (!buff) {
1587                                 status = -ENOMEM;
1588                                 goto cleanup1;
1589                         }
1590                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1591                                                    GFP_KERNEL);
1592                         if (!buff_size) {
1593                                 status = -ENOMEM;
1594                                 goto cleanup1;
1595                         }
1596                         left = ioc->buf_size;
1597                         data_ptr = ioc->buf;
1598                         while (left) {
1599                                 sz = (left >
1600                                       ioc->malloc_size) ? ioc->
1601                                     malloc_size : left;
1602                                 buff_size[sg_used] = sz;
1603                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1604                                 if (buff[sg_used] == NULL) {
1605                                         status = -ENOMEM;
1606                                         goto cleanup1;
1607                                 }
1608                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1609                                         if (copy_from_user
1610                                             (buff[sg_used], data_ptr, sz)) {
1611                                                 status = -EFAULT;
1612                                                 goto cleanup1;
1613                                         }
1614                                 } else {
1615                                         memset(buff[sg_used], 0, sz);
1616                                 }
1617                                 left -= sz;
1618                                 data_ptr += sz;
1619                                 sg_used++;
1620                         }
1621                         if ((c = cmd_alloc(host, 0)) == NULL) {
1622                                 status = -ENOMEM;
1623                                 goto cleanup1;
1624                         }
1625                         c->cmd_type = CMD_IOCTL_PEND;
1626                         c->Header.ReplyQueue = 0;
1627
1628                         if (ioc->buf_size > 0) {
1629                                 c->Header.SGList = sg_used;
1630                                 c->Header.SGTotal = sg_used;
1631                         } else {
1632                                 c->Header.SGList = 0;
1633                                 c->Header.SGTotal = 0;
1634                         }
1635                         c->Header.LUN = ioc->LUN_info;
1636                         c->Header.Tag.lower = c->busaddr;
1637
1638                         c->Request = ioc->Request;
1639                         if (ioc->buf_size > 0) {
1640                                 for (i = 0; i < sg_used; i++) {
1641                                         temp64.val =
1642                                             pci_map_single(host->pdev, buff[i],
1643                                                     buff_size[i],
1644                                                     PCI_DMA_BIDIRECTIONAL);
1645                                         c->SG[i].Addr.lower =
1646                                             temp64.val32.lower;
1647                                         c->SG[i].Addr.upper =
1648                                             temp64.val32.upper;
1649                                         c->SG[i].Len = buff_size[i];
1650                                         c->SG[i].Ext = 0;       /* we are not chaining */
1651                                 }
1652                         }
1653                         c->waiting = &wait;
1654                         enqueue_cmd_and_start_io(host, c);
1655                         wait_for_completion(&wait);
1656                         /* unlock the buffers from DMA */
1657                         for (i = 0; i < sg_used; i++) {
1658                                 temp64.val32.lower = c->SG[i].Addr.lower;
1659                                 temp64.val32.upper = c->SG[i].Addr.upper;
1660                                 pci_unmap_single(host->pdev,
1661                                         (dma_addr_t) temp64.val, buff_size[i],
1662                                         PCI_DMA_BIDIRECTIONAL);
1663                         }
1664                         check_ioctl_unit_attention(host, c);
1665                         /* Copy the error information out */
1666                         ioc->error_info = *(c->err_info);
1667                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1668                                 cmd_free(host, c, 0);
1669                                 status = -EFAULT;
1670                                 goto cleanup1;
1671                         }
1672                         if (ioc->Request.Type.Direction == XFER_READ) {
1673                                 /* Copy the data out of the buffer we created */
1674                                 BYTE __user *ptr = ioc->buf;
1675                                 for (i = 0; i < sg_used; i++) {
1676                                         if (copy_to_user
1677                                             (ptr, buff[i], buff_size[i])) {
1678                                                 cmd_free(host, c, 0);
1679                                                 status = -EFAULT;
1680                                                 goto cleanup1;
1681                                         }
1682                                         ptr += buff_size[i];
1683                                 }
1684                         }
1685                         cmd_free(host, c, 0);
1686                         status = 0;
1687                       cleanup1:
1688                         if (buff) {
1689                                 for (i = 0; i < sg_used; i++)
1690                                         kfree(buff[i]);
1691                                 kfree(buff);
1692                         }
1693                         kfree(buff_size);
1694                         kfree(ioc);
1695                         return status;
1696                 }
1697
1698         /* scsi_cmd_ioctl handles these, below, though some are not */
1699         /* very meaningful for cciss.  SG_IO is the main one people want. */
1700
1701         case SG_GET_VERSION_NUM:
1702         case SG_SET_TIMEOUT:
1703         case SG_GET_TIMEOUT:
1704         case SG_GET_RESERVED_SIZE:
1705         case SG_SET_RESERVED_SIZE:
1706         case SG_EMULATED_HOST:
1707         case SG_IO:
1708         case SCSI_IOCTL_SEND_COMMAND:
1709                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1710
1711         /* scsi_cmd_ioctl would normally handle these, below, but */
1712         /* they aren't a good fit for cciss, as CD-ROMs are */
1713         /* not supported, and we don't have any bus/target/lun */
1714         /* which we present to the kernel. */
1715
1716         case CDROM_SEND_PACKET:
1717         case CDROMCLOSETRAY:
1718         case CDROMEJECT:
1719         case SCSI_IOCTL_GET_IDLUN:
1720         case SCSI_IOCTL_GET_BUS_NUMBER:
1721         default:
1722                 return -ENOTTY;
1723         }
1724 }
1725
1726 static void cciss_check_queues(ctlr_info_t *h)
1727 {
1728         int start_queue = h->next_to_run;
1729         int i;
1730
1731         /* check to see if we have maxed out the number of commands that can
1732          * be placed on the queue.  If so then exit.  We do this check here
1733          * in case the interrupt we serviced was from an ioctl and did not
1734          * free any new commands.
1735          */
1736         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1737                 return;
1738
1739         /* We have room on the queue for more commands.  Now we need to queue
1740          * them up.  We will also keep track of the next queue to run so
1741          * that every queue gets a chance to be started first.
1742          */
1743         for (i = 0; i < h->highest_lun + 1; i++) {
1744                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1745                 /* make sure the disk has been added and the drive is real
1746                  * because this can be called from the middle of init_one.
1747                  */
1748                 if (!h->drv[curr_queue])
1749                         continue;
1750                 if (!(h->drv[curr_queue]->queue) ||
1751                         !(h->drv[curr_queue]->heads))
1752                         continue;
1753                 blk_start_queue(h->gendisk[curr_queue]->queue);
1754
1755                 /* check to see if we have maxed out the number of commands
1756                  * that can be placed on the queue.
1757                  */
1758                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1759                         if (curr_queue == start_queue) {
1760                                 h->next_to_run =
1761                                     (start_queue + 1) % (h->highest_lun + 1);
1762                                 break;
1763                         } else {
1764                                 h->next_to_run = curr_queue;
1765                                 break;
1766                         }
1767                 }
1768         }
1769 }
1770
1771 static void cciss_softirq_done(struct request *rq)
1772 {
1773         CommandList_struct *cmd = rq->completion_data;
1774         ctlr_info_t *h = hba[cmd->ctlr];
1775         SGDescriptor_struct *curr_sg = cmd->SG;
1776         u64bit temp64;
1777         unsigned long flags;
1778         int i, ddir;
1779         int sg_index = 0;
1780
1781         if (cmd->Request.Type.Direction == XFER_READ)
1782                 ddir = PCI_DMA_FROMDEVICE;
1783         else
1784                 ddir = PCI_DMA_TODEVICE;
1785
1786         /* command did not need to be retried */
1787         /* unmap the DMA mapping for all the scatter gather elements */
1788         for (i = 0; i < cmd->Header.SGList; i++) {
1789                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1790                         cciss_unmap_sg_chain_block(h, cmd);
1791                         /* Point to the next block */
1792                         curr_sg = h->cmd_sg_list[cmd->cmdindex];
1793                         sg_index = 0;
1794                 }
1795                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1796                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1797                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1798                                 ddir);
1799                 ++sg_index;
1800         }
1801
1802 #ifdef CCISS_DEBUG
1803         printk("Done with %p\n", rq);
1804 #endif                          /* CCISS_DEBUG */
1805
1806         /* set the residual count for pc requests */
1807         if (blk_pc_request(rq))
1808                 rq->resid_len = cmd->err_info->ResidualCnt;
1809
1810         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1811
1812         spin_lock_irqsave(&h->lock, flags);
1813         cmd_free(h, cmd, 1);
1814         cciss_check_queues(h);
1815         spin_unlock_irqrestore(&h->lock, flags);
1816 }
1817
1818 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1819         unsigned char scsi3addr[], uint32_t log_unit)
1820 {
1821         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1822                 sizeof(h->drv[log_unit]->LunID));
1823 }
1824
1825 /* This function gets the SCSI vendor, model, and revision of a logical drive
1826  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1827  * they cannot be read.
1828  */
1829 static void cciss_get_device_descr(int ctlr, int logvol,
1830                                    char *vendor, char *model, char *rev)
1831 {
1832         int rc;
1833         InquiryData_struct *inq_buf;
1834         unsigned char scsi3addr[8];
1835
1836         *vendor = '\0';
1837         *model = '\0';
1838         *rev = '\0';
1839
1840         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1841         if (!inq_buf)
1842                 return;
1843
1844         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1845         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1846                         scsi3addr, TYPE_CMD);
1847         if (rc == IO_OK) {
1848                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1849                 vendor[VENDOR_LEN] = '\0';
1850                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1851                 model[MODEL_LEN] = '\0';
1852                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1853                 rev[REV_LEN] = '\0';
1854         }
1855
1856         kfree(inq_buf);
1857         return;
1858 }
1859
1860 /* This function gets the serial number of a logical drive via
1861  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1862  * number cannot be had, for whatever reason, 16 bytes of 0xff
1863  * are returned instead.
1864  */
1865 static void cciss_get_serial_no(int ctlr, int logvol,
1866                                 unsigned char *serial_no, int buflen)
1867 {
1868 #define PAGE_83_INQ_BYTES 64
1869         int rc;
1870         unsigned char *buf;
1871         unsigned char scsi3addr[8];
1872
1873         if (buflen > 16)
1874                 buflen = 16;
1875         memset(serial_no, 0xff, buflen);
1876         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1877         if (!buf)
1878                 return;
1879         memset(serial_no, 0, buflen);
1880         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1881         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1882                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1883         if (rc == IO_OK)
1884                 memcpy(serial_no, &buf[8], buflen);
1885         kfree(buf);
1886         return;
1887 }
1888
1889 /*
1890  * cciss_add_disk sets up the block device queue for a logical drive
1891  */
1892 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1893                                 int drv_index)
1894 {
1895         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1896         if (!disk->queue)
1897                 goto init_queue_failure;
1898         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1899         disk->major = h->major;
1900         disk->first_minor = drv_index << NWD_SHIFT;
1901         disk->fops = &cciss_fops;
1902         if (cciss_create_ld_sysfs_entry(h, drv_index))
1903                 goto cleanup_queue;
1904         disk->private_data = h->drv[drv_index];
1905         disk->driverfs_dev = &h->drv[drv_index]->dev;
1906
1907         /* Set up queue information */
1908         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1909
1910         /* This is a hardware imposed limit. */
1911         blk_queue_max_segments(disk->queue, h->maxsgentries);
1912
1913         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1914
1915         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1916
1917         disk->queue->queuedata = h;
1918
1919         blk_queue_logical_block_size(disk->queue,
1920                                      h->drv[drv_index]->block_size);
1921
1922         /* Make sure all queue data is written out before */
1923         /* setting h->drv[drv_index]->queue, as setting this */
1924         /* allows the interrupt handler to start the queue */
1925         wmb();
1926         h->drv[drv_index]->queue = disk->queue;
1927         add_disk(disk);
1928         return 0;
1929
1930 cleanup_queue:
1931         blk_cleanup_queue(disk->queue);
1932         disk->queue = NULL;
1933 init_queue_failure:
1934         return -1;
1935 }
1936
1937 /* This function will check the usage_count of the drive to be updated/added.
1938  * If the usage_count is zero and it is a heretofore unknown drive, or,
1939  * the drive's capacity, geometry, or serial number has changed,
1940  * then the drive information will be updated and the disk will be
1941  * re-registered with the kernel.  If these conditions don't hold,
1942  * then it will be left alone for the next reboot.  The exception to this
1943  * is disk 0 which will always be left registered with the kernel since it
1944  * is also the controller node.  Any changes to disk 0 will show up on
1945  * the next reboot.
1946  */
1947 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1948         int via_ioctl)
1949 {
1950         ctlr_info_t *h = hba[ctlr];
1951         struct gendisk *disk;
1952         InquiryData_struct *inq_buff = NULL;
1953         unsigned int block_size;
1954         sector_t total_size;
1955         unsigned long flags = 0;
1956         int ret = 0;
1957         drive_info_struct *drvinfo;
1958
1959         /* Get information about the disk and modify the driver structure */
1960         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1961         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1962         if (inq_buff == NULL || drvinfo == NULL)
1963                 goto mem_msg;
1964
1965         /* testing to see if 16-byte CDBs are already being used */
1966         if (h->cciss_read == CCISS_READ_16) {
1967                 cciss_read_capacity_16(h->ctlr, drv_index,
1968                         &total_size, &block_size);
1969
1970         } else {
1971                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1972                 /* if read_capacity returns all F's this volume is >2TB */
1973                 /* in size so we switch to 16-byte CDB's for all */
1974                 /* read/write ops */
1975                 if (total_size == 0xFFFFFFFFULL) {
1976                         cciss_read_capacity_16(ctlr, drv_index,
1977                         &total_size, &block_size);
1978                         h->cciss_read = CCISS_READ_16;
1979                         h->cciss_write = CCISS_WRITE_16;
1980                 } else {
1981                         h->cciss_read = CCISS_READ_10;
1982                         h->cciss_write = CCISS_WRITE_10;
1983                 }
1984         }
1985
1986         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1987                                inq_buff, drvinfo);
1988         drvinfo->block_size = block_size;
1989         drvinfo->nr_blocks = total_size + 1;
1990
1991         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1992                                 drvinfo->model, drvinfo->rev);
1993         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1994                         sizeof(drvinfo->serial_no));
1995         /* Save the lunid in case we deregister the disk, below. */
1996         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1997                 sizeof(drvinfo->LunID));
1998
1999         /* Is it the same disk we already know, and nothing's changed? */
2000         if (h->drv[drv_index]->raid_level != -1 &&
2001                 ((memcmp(drvinfo->serial_no,
2002                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2003                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2004                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2005                 drvinfo->heads == h->drv[drv_index]->heads &&
2006                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2007                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2008                         /* The disk is unchanged, nothing to update */
2009                         goto freeret;
2010
2011         /* If we get here it's not the same disk, or something's changed,
2012          * so we need to * deregister it, and re-register it, if it's not
2013          * in use.
2014          * If the disk already exists then deregister it before proceeding
2015          * (unless it's the first disk (for the controller node).
2016          */
2017         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2018                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
2019                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2020                 h->drv[drv_index]->busy_configuring = 1;
2021                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2022
2023                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2024                  * which keeps the interrupt handler from starting
2025                  * the queue.
2026                  */
2027                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2028         }
2029
2030         /* If the disk is in use return */
2031         if (ret)
2032                 goto freeret;
2033
2034         /* Save the new information from cciss_geometry_inquiry
2035          * and serial number inquiry.  If the disk was deregistered
2036          * above, then h->drv[drv_index] will be NULL.
2037          */
2038         if (h->drv[drv_index] == NULL) {
2039                 drvinfo->device_initialized = 0;
2040                 h->drv[drv_index] = drvinfo;
2041                 drvinfo = NULL; /* so it won't be freed below. */
2042         } else {
2043                 /* special case for cxd0 */
2044                 h->drv[drv_index]->block_size = drvinfo->block_size;
2045                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2046                 h->drv[drv_index]->heads = drvinfo->heads;
2047                 h->drv[drv_index]->sectors = drvinfo->sectors;
2048                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2049                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2050                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2051                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2052                         VENDOR_LEN + 1);
2053                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2054                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2055         }
2056
2057         ++h->num_luns;
2058         disk = h->gendisk[drv_index];
2059         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2060
2061         /* If it's not disk 0 (drv_index != 0)
2062          * or if it was disk 0, but there was previously
2063          * no actual corresponding configured logical drive
2064          * (raid_leve == -1) then we want to update the
2065          * logical drive's information.
2066          */
2067         if (drv_index || first_time) {
2068                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2069                         cciss_free_gendisk(h, drv_index);
2070                         cciss_free_drive_info(h, drv_index);
2071                         printk(KERN_WARNING "cciss:%d could not update "
2072                                 "disk %d\n", h->ctlr, drv_index);
2073                         --h->num_luns;
2074                 }
2075         }
2076
2077 freeret:
2078         kfree(inq_buff);
2079         kfree(drvinfo);
2080         return;
2081 mem_msg:
2082         printk(KERN_ERR "cciss: out of memory\n");
2083         goto freeret;
2084 }
2085
2086 /* This function will find the first index of the controllers drive array
2087  * that has a null drv pointer and allocate the drive info struct and
2088  * will return that index   This is where new drives will be added.
2089  * If the index to be returned is greater than the highest_lun index for
2090  * the controller then highest_lun is set * to this new index.
2091  * If there are no available indexes or if tha allocation fails, then -1
2092  * is returned.  * "controller_node" is used to know if this is a real
2093  * logical drive, or just the controller node, which determines if this
2094  * counts towards highest_lun.
2095  */
2096 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2097 {
2098         int i;
2099         drive_info_struct *drv;
2100
2101         /* Search for an empty slot for our drive info */
2102         for (i = 0; i < CISS_MAX_LUN; i++) {
2103
2104                 /* if not cxd0 case, and it's occupied, skip it. */
2105                 if (h->drv[i] && i != 0)
2106                         continue;
2107                 /*
2108                  * If it's cxd0 case, and drv is alloc'ed already, and a
2109                  * disk is configured there, skip it.
2110                  */
2111                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2112                         continue;
2113
2114                 /*
2115                  * We've found an empty slot.  Update highest_lun
2116                  * provided this isn't just the fake cxd0 controller node.
2117                  */
2118                 if (i > h->highest_lun && !controller_node)
2119                         h->highest_lun = i;
2120
2121                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2122                 if (i == 0 && h->drv[i] != NULL)
2123                         return i;
2124
2125                 /*
2126                  * Found an empty slot, not already alloc'ed.  Allocate it.
2127                  * Mark it with raid_level == -1, so we know it's new later on.
2128                  */
2129                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2130                 if (!drv)
2131                         return -1;
2132                 drv->raid_level = -1; /* so we know it's new */
2133                 h->drv[i] = drv;
2134                 return i;
2135         }
2136         return -1;
2137 }
2138
2139 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2140 {
2141         kfree(h->drv[drv_index]);
2142         h->drv[drv_index] = NULL;
2143 }
2144
2145 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2146 {
2147         put_disk(h->gendisk[drv_index]);
2148         h->gendisk[drv_index] = NULL;
2149 }
2150
2151 /* cciss_add_gendisk finds a free hba[]->drv structure
2152  * and allocates a gendisk if needed, and sets the lunid
2153  * in the drvinfo structure.   It returns the index into
2154  * the ->drv[] array, or -1 if none are free.
2155  * is_controller_node indicates whether highest_lun should
2156  * count this disk, or if it's only being added to provide
2157  * a means to talk to the controller in case no logical
2158  * drives have yet been configured.
2159  */
2160 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2161         int controller_node)
2162 {
2163         int drv_index;
2164
2165         drv_index = cciss_alloc_drive_info(h, controller_node);
2166         if (drv_index == -1)
2167                 return -1;
2168
2169         /*Check if the gendisk needs to be allocated */
2170         if (!h->gendisk[drv_index]) {
2171                 h->gendisk[drv_index] =
2172                         alloc_disk(1 << NWD_SHIFT);
2173                 if (!h->gendisk[drv_index]) {
2174                         printk(KERN_ERR "cciss%d: could not "
2175                                 "allocate a new disk %d\n",
2176                                 h->ctlr, drv_index);
2177                         goto err_free_drive_info;
2178                 }
2179         }
2180         memcpy(h->drv[drv_index]->LunID, lunid,
2181                 sizeof(h->drv[drv_index]->LunID));
2182         if (cciss_create_ld_sysfs_entry(h, drv_index))
2183                 goto err_free_disk;
2184         /* Don't need to mark this busy because nobody */
2185         /* else knows about this disk yet to contend */
2186         /* for access to it. */
2187         h->drv[drv_index]->busy_configuring = 0;
2188         wmb();
2189         return drv_index;
2190
2191 err_free_disk:
2192         cciss_free_gendisk(h, drv_index);
2193 err_free_drive_info:
2194         cciss_free_drive_info(h, drv_index);
2195         return -1;
2196 }
2197
2198 /* This is for the special case of a controller which
2199  * has no logical drives.  In this case, we still need
2200  * to register a disk so the controller can be accessed
2201  * by the Array Config Utility.
2202  */
2203 static void cciss_add_controller_node(ctlr_info_t *h)
2204 {
2205         struct gendisk *disk;
2206         int drv_index;
2207
2208         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2209                 return;
2210
2211         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2212         if (drv_index == -1)
2213                 goto error;
2214         h->drv[drv_index]->block_size = 512;
2215         h->drv[drv_index]->nr_blocks = 0;
2216         h->drv[drv_index]->heads = 0;
2217         h->drv[drv_index]->sectors = 0;
2218         h->drv[drv_index]->cylinders = 0;
2219         h->drv[drv_index]->raid_level = -1;
2220         memset(h->drv[drv_index]->serial_no, 0, 16);
2221         disk = h->gendisk[drv_index];
2222         if (cciss_add_disk(h, disk, drv_index) == 0)
2223                 return;
2224         cciss_free_gendisk(h, drv_index);
2225         cciss_free_drive_info(h, drv_index);
2226 error:
2227         printk(KERN_WARNING "cciss%d: could not "
2228                 "add disk 0.\n", h->ctlr);
2229         return;
2230 }
2231
2232 /* This function will add and remove logical drives from the Logical
2233  * drive array of the controller and maintain persistency of ordering
2234  * so that mount points are preserved until the next reboot.  This allows
2235  * for the removal of logical drives in the middle of the drive array
2236  * without a re-ordering of those drives.
2237  * INPUT
2238  * h            = The controller to perform the operations on
2239  */
2240 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2241         int via_ioctl)
2242 {
2243         int ctlr = h->ctlr;
2244         int num_luns;
2245         ReportLunData_struct *ld_buff = NULL;
2246         int return_code;
2247         int listlength = 0;
2248         int i;
2249         int drv_found;
2250         int drv_index = 0;
2251         unsigned char lunid[8] = CTLR_LUNID;
2252         unsigned long flags;
2253
2254         if (!capable(CAP_SYS_RAWIO))
2255                 return -EPERM;
2256
2257         /* Set busy_configuring flag for this operation */
2258         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2259         if (h->busy_configuring) {
2260                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2261                 return -EBUSY;
2262         }
2263         h->busy_configuring = 1;
2264         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2265
2266         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2267         if (ld_buff == NULL)
2268                 goto mem_msg;
2269
2270         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2271                                       sizeof(ReportLunData_struct),
2272                                       0, CTLR_LUNID, TYPE_CMD);
2273
2274         if (return_code == IO_OK)
2275                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2276         else {  /* reading number of logical volumes failed */
2277                 printk(KERN_WARNING "cciss: report logical volume"
2278                        " command failed\n");
2279                 listlength = 0;
2280                 goto freeret;
2281         }
2282
2283         num_luns = listlength / 8;      /* 8 bytes per entry */
2284         if (num_luns > CISS_MAX_LUN) {
2285                 num_luns = CISS_MAX_LUN;
2286                 printk(KERN_WARNING "cciss: more luns configured"
2287                        " on controller than can be handled by"
2288                        " this driver.\n");
2289         }
2290
2291         if (num_luns == 0)
2292                 cciss_add_controller_node(h);
2293
2294         /* Compare controller drive array to driver's drive array
2295          * to see if any drives are missing on the controller due
2296          * to action of Array Config Utility (user deletes drive)
2297          * and deregister logical drives which have disappeared.
2298          */
2299         for (i = 0; i <= h->highest_lun; i++) {
2300                 int j;
2301                 drv_found = 0;
2302
2303                 /* skip holes in the array from already deleted drives */
2304                 if (h->drv[i] == NULL)
2305                         continue;
2306
2307                 for (j = 0; j < num_luns; j++) {
2308                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2309                         if (memcmp(h->drv[i]->LunID, lunid,
2310                                 sizeof(lunid)) == 0) {
2311                                 drv_found = 1;
2312                                 break;
2313                         }
2314                 }
2315                 if (!drv_found) {
2316                         /* Deregister it from the OS, it's gone. */
2317                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2318                         h->drv[i]->busy_configuring = 1;
2319                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2320                         return_code = deregister_disk(h, i, 1, via_ioctl);
2321                         if (h->drv[i] != NULL)
2322                                 h->drv[i]->busy_configuring = 0;
2323                 }
2324         }
2325
2326         /* Compare controller drive array to driver's drive array.
2327          * Check for updates in the drive information and any new drives
2328          * on the controller due to ACU adding logical drives, or changing
2329          * a logical drive's size, etc.  Reregister any new/changed drives
2330          */
2331         for (i = 0; i < num_luns; i++) {
2332                 int j;
2333
2334                 drv_found = 0;
2335
2336                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2337                 /* Find if the LUN is already in the drive array
2338                  * of the driver.  If so then update its info
2339                  * if not in use.  If it does not exist then find
2340                  * the first free index and add it.
2341                  */
2342                 for (j = 0; j <= h->highest_lun; j++) {
2343                         if (h->drv[j] != NULL &&
2344                                 memcmp(h->drv[j]->LunID, lunid,
2345                                         sizeof(h->drv[j]->LunID)) == 0) {
2346                                 drv_index = j;
2347                                 drv_found = 1;
2348                                 break;
2349                         }
2350                 }
2351
2352                 /* check if the drive was found already in the array */
2353                 if (!drv_found) {
2354                         drv_index = cciss_add_gendisk(h, lunid, 0);
2355                         if (drv_index == -1)
2356                                 goto freeret;
2357                 }
2358                 cciss_update_drive_info(ctlr, drv_index, first_time,
2359                         via_ioctl);
2360         }               /* end for */
2361
2362 freeret:
2363         kfree(ld_buff);
2364         h->busy_configuring = 0;
2365         /* We return -1 here to tell the ACU that we have registered/updated
2366          * all of the drives that we can and to keep it from calling us
2367          * additional times.
2368          */
2369         return -1;
2370 mem_msg:
2371         printk(KERN_ERR "cciss: out of memory\n");
2372         h->busy_configuring = 0;
2373         goto freeret;
2374 }
2375
2376 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2377 {
2378         /* zero out the disk size info */
2379         drive_info->nr_blocks = 0;
2380         drive_info->block_size = 0;
2381         drive_info->heads = 0;
2382         drive_info->sectors = 0;
2383         drive_info->cylinders = 0;
2384         drive_info->raid_level = -1;
2385         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2386         memset(drive_info->model, 0, sizeof(drive_info->model));
2387         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2388         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2389         /*
2390          * don't clear the LUNID though, we need to remember which
2391          * one this one is.
2392          */
2393 }
2394
2395 /* This function will deregister the disk and it's queue from the
2396  * kernel.  It must be called with the controller lock held and the
2397  * drv structures busy_configuring flag set.  It's parameters are:
2398  *
2399  * disk = This is the disk to be deregistered
2400  * drv  = This is the drive_info_struct associated with the disk to be
2401  *        deregistered.  It contains information about the disk used
2402  *        by the driver.
2403  * clear_all = This flag determines whether or not the disk information
2404  *             is going to be completely cleared out and the highest_lun
2405  *             reset.  Sometimes we want to clear out information about
2406  *             the disk in preparation for re-adding it.  In this case
2407  *             the highest_lun should be left unchanged and the LunID
2408  *             should not be cleared.
2409  * via_ioctl
2410  *    This indicates whether we've reached this path via ioctl.
2411  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2412  *    If this path is reached via ioctl(), then the max_usage_count will
2413  *    be 1, as the process calling ioctl() has got to have the device open.
2414  *    If we get here via sysfs, then the max usage count will be zero.
2415 */
2416 static int deregister_disk(ctlr_info_t *h, int drv_index,
2417                            int clear_all, int via_ioctl)
2418 {
2419         int i;
2420         struct gendisk *disk;
2421         drive_info_struct *drv;
2422         int recalculate_highest_lun;
2423
2424         if (!capable(CAP_SYS_RAWIO))
2425                 return -EPERM;
2426
2427         drv = h->drv[drv_index];
2428         disk = h->gendisk[drv_index];
2429
2430         /* make sure logical volume is NOT is use */
2431         if (clear_all || (h->gendisk[0] == disk)) {
2432                 if (drv->usage_count > via_ioctl)
2433                         return -EBUSY;
2434         } else if (drv->usage_count > 0)
2435                 return -EBUSY;
2436
2437         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2438
2439         /* invalidate the devices and deregister the disk.  If it is disk
2440          * zero do not deregister it but just zero out it's values.  This
2441          * allows us to delete disk zero but keep the controller registered.
2442          */
2443         if (h->gendisk[0] != disk) {
2444                 struct request_queue *q = disk->queue;
2445                 if (disk->flags & GENHD_FL_UP) {
2446                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2447                         del_gendisk(disk);
2448                 }
2449                 if (q)
2450                         blk_cleanup_queue(q);
2451                 /* If clear_all is set then we are deleting the logical
2452                  * drive, not just refreshing its info.  For drives
2453                  * other than disk 0 we will call put_disk.  We do not
2454                  * do this for disk 0 as we need it to be able to
2455                  * configure the controller.
2456                  */
2457                 if (clear_all){
2458                         /* This isn't pretty, but we need to find the
2459                          * disk in our array and NULL our the pointer.
2460                          * This is so that we will call alloc_disk if
2461                          * this index is used again later.
2462                          */
2463                         for (i=0; i < CISS_MAX_LUN; i++){
2464                                 if (h->gendisk[i] == disk) {
2465                                         h->gendisk[i] = NULL;
2466                                         break;
2467                                 }
2468                         }
2469                         put_disk(disk);
2470                 }
2471         } else {
2472                 set_capacity(disk, 0);
2473                 cciss_clear_drive_info(drv);
2474         }
2475
2476         --h->num_luns;
2477
2478         /* if it was the last disk, find the new hightest lun */
2479         if (clear_all && recalculate_highest_lun) {
2480                 int newhighest = -1;
2481                 for (i = 0; i <= h->highest_lun; i++) {
2482                         /* if the disk has size > 0, it is available */
2483                         if (h->drv[i] && h->drv[i]->heads)
2484                                 newhighest = i;
2485                 }
2486                 h->highest_lun = newhighest;
2487         }
2488         return 0;
2489 }
2490
2491 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2492                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2493                 int cmd_type)
2494 {
2495         ctlr_info_t *h = hba[ctlr];
2496         u64bit buff_dma_handle;
2497         int status = IO_OK;
2498
2499         c->cmd_type = CMD_IOCTL_PEND;
2500         c->Header.ReplyQueue = 0;
2501         if (buff != NULL) {
2502                 c->Header.SGList = 1;
2503                 c->Header.SGTotal = 1;
2504         } else {
2505                 c->Header.SGList = 0;
2506                 c->Header.SGTotal = 0;
2507         }
2508         c->Header.Tag.lower = c->busaddr;
2509         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2510
2511         c->Request.Type.Type = cmd_type;
2512         if (cmd_type == TYPE_CMD) {
2513                 switch (cmd) {
2514                 case CISS_INQUIRY:
2515                         /* are we trying to read a vital product page */
2516                         if (page_code != 0) {
2517                                 c->Request.CDB[1] = 0x01;
2518                                 c->Request.CDB[2] = page_code;
2519                         }
2520                         c->Request.CDBLen = 6;
2521                         c->Request.Type.Attribute = ATTR_SIMPLE;
2522                         c->Request.Type.Direction = XFER_READ;
2523                         c->Request.Timeout = 0;
2524                         c->Request.CDB[0] = CISS_INQUIRY;
2525                         c->Request.CDB[4] = size & 0xFF;
2526                         break;
2527                 case CISS_REPORT_LOG:
2528                 case CISS_REPORT_PHYS:
2529                         /* Talking to controller so It's a physical command
2530                            mode = 00 target = 0.  Nothing to write.
2531                          */
2532                         c->Request.CDBLen = 12;
2533                         c->Request.Type.Attribute = ATTR_SIMPLE;
2534                         c->Request.Type.Direction = XFER_READ;
2535                         c->Request.Timeout = 0;
2536                         c->Request.CDB[0] = cmd;
2537                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2538                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2539                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2540                         c->Request.CDB[9] = size & 0xFF;
2541                         break;
2542
2543                 case CCISS_READ_CAPACITY:
2544                         c->Request.CDBLen = 10;
2545                         c->Request.Type.Attribute = ATTR_SIMPLE;
2546                         c->Request.Type.Direction = XFER_READ;
2547                         c->Request.Timeout = 0;
2548                         c->Request.CDB[0] = cmd;
2549                         break;
2550                 case CCISS_READ_CAPACITY_16:
2551                         c->Request.CDBLen = 16;
2552                         c->Request.Type.Attribute = ATTR_SIMPLE;
2553                         c->Request.Type.Direction = XFER_READ;
2554                         c->Request.Timeout = 0;
2555                         c->Request.CDB[0] = cmd;
2556                         c->Request.CDB[1] = 0x10;
2557                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2558                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2559                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2560                         c->Request.CDB[13] = size & 0xFF;
2561                         c->Request.Timeout = 0;
2562                         c->Request.CDB[0] = cmd;
2563                         break;
2564                 case CCISS_CACHE_FLUSH:
2565                         c->Request.CDBLen = 12;
2566                         c->Request.Type.Attribute = ATTR_SIMPLE;
2567                         c->Request.Type.Direction = XFER_WRITE;
2568                         c->Request.Timeout = 0;
2569                         c->Request.CDB[0] = BMIC_WRITE;
2570                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2571                         break;
2572                 case TEST_UNIT_READY:
2573                         c->Request.CDBLen = 6;
2574                         c->Request.Type.Attribute = ATTR_SIMPLE;
2575                         c->Request.Type.Direction = XFER_NONE;
2576                         c->Request.Timeout = 0;
2577                         break;
2578                 default:
2579                         printk(KERN_WARNING
2580                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2581                         return IO_ERROR;
2582                 }
2583         } else if (cmd_type == TYPE_MSG) {
2584                 switch (cmd) {
2585                 case 0: /* ABORT message */
2586                         c->Request.CDBLen = 12;
2587                         c->Request.Type.Attribute = ATTR_SIMPLE;
2588                         c->Request.Type.Direction = XFER_WRITE;
2589                         c->Request.Timeout = 0;
2590                         c->Request.CDB[0] = cmd;        /* abort */
2591                         c->Request.CDB[1] = 0;  /* abort a command */
2592                         /* buff contains the tag of the command to abort */
2593                         memcpy(&c->Request.CDB[4], buff, 8);
2594                         break;
2595                 case 1: /* RESET message */
2596                         c->Request.CDBLen = 16;
2597                         c->Request.Type.Attribute = ATTR_SIMPLE;
2598                         c->Request.Type.Direction = XFER_NONE;
2599                         c->Request.Timeout = 0;
2600                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2601                         c->Request.CDB[0] = cmd;        /* reset */
2602                         c->Request.CDB[1] = 0x03;       /* reset a target */
2603                         break;
2604                 case 3: /* No-Op message */
2605                         c->Request.CDBLen = 1;
2606                         c->Request.Type.Attribute = ATTR_SIMPLE;
2607                         c->Request.Type.Direction = XFER_WRITE;
2608                         c->Request.Timeout = 0;
2609                         c->Request.CDB[0] = cmd;
2610                         break;
2611                 default:
2612                         printk(KERN_WARNING
2613                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2614                         return IO_ERROR;
2615                 }
2616         } else {
2617                 printk(KERN_WARNING
2618                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2619                 return IO_ERROR;
2620         }
2621         /* Fill in the scatter gather information */
2622         if (size > 0) {
2623                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2624                                                              buff, size,
2625                                                              PCI_DMA_BIDIRECTIONAL);
2626                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2627                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2628                 c->SG[0].Len = size;
2629                 c->SG[0].Ext = 0;       /* we are not chaining */
2630         }
2631         return status;
2632 }
2633
2634 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2635 {
2636         switch (c->err_info->ScsiStatus) {
2637         case SAM_STAT_GOOD:
2638                 return IO_OK;
2639         case SAM_STAT_CHECK_CONDITION:
2640                 switch (0xf & c->err_info->SenseInfo[2]) {
2641                 case 0: return IO_OK; /* no sense */
2642                 case 1: return IO_OK; /* recovered error */
2643                 default:
2644                         if (check_for_unit_attention(h, c))
2645                                 return IO_NEEDS_RETRY;
2646                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2647                                 "check condition, sense key = 0x%02x\n",
2648                                 h->ctlr, c->Request.CDB[0],
2649                                 c->err_info->SenseInfo[2]);
2650                 }
2651                 break;
2652         default:
2653                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2654                         "scsi status = 0x%02x\n", h->ctlr,
2655                         c->Request.CDB[0], c->err_info->ScsiStatus);
2656                 break;
2657         }
2658         return IO_ERROR;
2659 }
2660
2661 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2662 {
2663         int return_status = IO_OK;
2664
2665         if (c->err_info->CommandStatus == CMD_SUCCESS)
2666                 return IO_OK;
2667
2668         switch (c->err_info->CommandStatus) {
2669         case CMD_TARGET_STATUS:
2670                 return_status = check_target_status(h, c);
2671                 break;
2672         case CMD_DATA_UNDERRUN:
2673         case CMD_DATA_OVERRUN:
2674                 /* expected for inquiry and report lun commands */
2675                 break;
2676         case CMD_INVALID:
2677                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2678                        "reported invalid\n", c->Request.CDB[0]);
2679                 return_status = IO_ERROR;
2680                 break;
2681         case CMD_PROTOCOL_ERR:
2682                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2683                        "protocol error \n", c->Request.CDB[0]);
2684                 return_status = IO_ERROR;
2685                 break;
2686         case CMD_HARDWARE_ERR:
2687                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2688                        " hardware error\n", c->Request.CDB[0]);
2689                 return_status = IO_ERROR;
2690                 break;
2691         case CMD_CONNECTION_LOST:
2692                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2693                        "connection lost\n", c->Request.CDB[0]);
2694                 return_status = IO_ERROR;
2695                 break;
2696         case CMD_ABORTED:
2697                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2698                        "aborted\n", c->Request.CDB[0]);
2699                 return_status = IO_ERROR;
2700                 break;
2701         case CMD_ABORT_FAILED:
2702                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2703                        "abort failed\n", c->Request.CDB[0]);
2704                 return_status = IO_ERROR;
2705                 break;
2706         case CMD_UNSOLICITED_ABORT:
2707                 printk(KERN_WARNING
2708                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2709                         c->Request.CDB[0]);
2710                 return_status = IO_NEEDS_RETRY;
2711                 break;
2712         default:
2713                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2714                        "unknown status %x\n", c->Request.CDB[0],
2715                        c->err_info->CommandStatus);
2716                 return_status = IO_ERROR;
2717         }
2718         return return_status;
2719 }
2720
2721 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2722         int attempt_retry)
2723 {
2724         DECLARE_COMPLETION_ONSTACK(wait);
2725         u64bit buff_dma_handle;
2726         int return_status = IO_OK;
2727
2728 resend_cmd2:
2729         c->waiting = &wait;
2730         enqueue_cmd_and_start_io(h, c);
2731
2732         wait_for_completion(&wait);
2733
2734         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2735                 goto command_done;
2736
2737         return_status = process_sendcmd_error(h, c);
2738
2739         if (return_status == IO_NEEDS_RETRY &&
2740                 c->retry_count < MAX_CMD_RETRIES) {
2741                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2742                         c->Request.CDB[0]);
2743                 c->retry_count++;
2744                 /* erase the old error information */
2745                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2746                 return_status = IO_OK;
2747                 INIT_COMPLETION(wait);
2748                 goto resend_cmd2;
2749         }
2750
2751 command_done:
2752         /* unlock the buffers from DMA */
2753         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2754         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2755         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2756                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2757         return return_status;
2758 }
2759
2760 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2761                            __u8 page_code, unsigned char scsi3addr[],
2762                         int cmd_type)
2763 {
2764         ctlr_info_t *h = hba[ctlr];
2765         CommandList_struct *c;
2766         int return_status;
2767
2768         c = cmd_alloc(h, 0);
2769         if (!c)
2770                 return -ENOMEM;
2771         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2772                 scsi3addr, cmd_type);
2773         if (return_status == IO_OK)
2774                 return_status = sendcmd_withirq_core(h, c, 1);
2775
2776         cmd_free(h, c, 0);
2777         return return_status;
2778 }
2779
2780 static void cciss_geometry_inquiry(int ctlr, int logvol,
2781                                    sector_t total_size,
2782                                    unsigned int block_size,
2783                                    InquiryData_struct *inq_buff,
2784                                    drive_info_struct *drv)
2785 {
2786         int return_code;
2787         unsigned long t;
2788         unsigned char scsi3addr[8];
2789
2790         memset(inq_buff, 0, sizeof(InquiryData_struct));
2791         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2792         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2793                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2794         if (return_code == IO_OK) {
2795                 if (inq_buff->data_byte[8] == 0xFF) {
2796                         printk(KERN_WARNING
2797                                "cciss: reading geometry failed, volume "
2798                                "does not support reading geometry\n");
2799                         drv->heads = 255;
2800                         drv->sectors = 32;      /* Sectors per track */
2801                         drv->cylinders = total_size + 1;
2802                         drv->raid_level = RAID_UNKNOWN;
2803                 } else {
2804                         drv->heads = inq_buff->data_byte[6];
2805                         drv->sectors = inq_buff->data_byte[7];
2806                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2807                         drv->cylinders += inq_buff->data_byte[5];
2808                         drv->raid_level = inq_buff->data_byte[8];
2809                 }
2810                 drv->block_size = block_size;
2811                 drv->nr_blocks = total_size + 1;
2812                 t = drv->heads * drv->sectors;
2813                 if (t > 1) {
2814                         sector_t real_size = total_size + 1;
2815                         unsigned long rem = sector_div(real_size, t);
2816                         if (rem)
2817                                 real_size++;
2818                         drv->cylinders = real_size;
2819                 }
2820         } else {                /* Get geometry failed */
2821                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2822         }
2823 }
2824
2825 static void
2826 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2827                     unsigned int *block_size)
2828 {
2829         ReadCapdata_struct *buf;
2830         int return_code;
2831         unsigned char scsi3addr[8];
2832
2833         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2834         if (!buf) {
2835                 printk(KERN_WARNING "cciss: out of memory\n");
2836                 return;
2837         }
2838
2839         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2840         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2841                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2842         if (return_code == IO_OK) {
2843                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2844                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2845         } else {                /* read capacity command failed */
2846                 printk(KERN_WARNING "cciss: read capacity failed\n");
2847                 *total_size = 0;
2848                 *block_size = BLOCK_SIZE;
2849         }
2850         kfree(buf);
2851 }
2852
2853 static void cciss_read_capacity_16(int ctlr, int logvol,
2854         sector_t *total_size, unsigned int *block_size)
2855 {
2856         ReadCapdata_struct_16 *buf;
2857         int return_code;
2858         unsigned char scsi3addr[8];
2859
2860         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2861         if (!buf) {
2862                 printk(KERN_WARNING "cciss: out of memory\n");
2863                 return;
2864         }
2865
2866         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2867         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2868                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2869                         0, scsi3addr, TYPE_CMD);
2870         if (return_code == IO_OK) {
2871                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2872                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2873         } else {                /* read capacity command failed */
2874                 printk(KERN_WARNING "cciss: read capacity failed\n");
2875                 *total_size = 0;
2876                 *block_size = BLOCK_SIZE;
2877         }
2878         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2879                (unsigned long long)*total_size+1, *block_size);
2880         kfree(buf);
2881 }
2882
2883 static int cciss_revalidate(struct gendisk *disk)
2884 {
2885         ctlr_info_t *h = get_host(disk);
2886         drive_info_struct *drv = get_drv(disk);
2887         int logvol;
2888         int FOUND = 0;
2889         unsigned int block_size;
2890         sector_t total_size;
2891         InquiryData_struct *inq_buff = NULL;
2892
2893         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2894                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2895                         sizeof(drv->LunID)) == 0) {
2896                         FOUND = 1;
2897                         break;
2898                 }
2899         }
2900
2901         if (!FOUND)
2902                 return 1;
2903
2904         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2905         if (inq_buff == NULL) {
2906                 printk(KERN_WARNING "cciss: out of memory\n");
2907                 return 1;
2908         }
2909         if (h->cciss_read == CCISS_READ_10) {
2910                 cciss_read_capacity(h->ctlr, logvol,
2911                                         &total_size, &block_size);
2912         } else {
2913                 cciss_read_capacity_16(h->ctlr, logvol,
2914                                         &total_size, &block_size);
2915         }
2916         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2917                                inq_buff, drv);
2918
2919         blk_queue_logical_block_size(drv->queue, drv->block_size);
2920         set_capacity(disk, drv->nr_blocks);
2921
2922         kfree(inq_buff);
2923         return 0;
2924 }
2925
2926 /*
2927  * Map (physical) PCI mem into (virtual) kernel space
2928  */
2929 static void __iomem *remap_pci_mem(ulong base, ulong size)
2930 {
2931         ulong page_base = ((ulong) base) & PAGE_MASK;
2932         ulong page_offs = ((ulong) base) - page_base;
2933         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2934
2935         return page_remapped ? (page_remapped + page_offs) : NULL;
2936 }
2937
2938 /*
2939  * Takes jobs of the Q and sends them to the hardware, then puts it on
2940  * the Q to wait for completion.
2941  */
2942 static void start_io(ctlr_info_t *h)
2943 {
2944         CommandList_struct *c;
2945
2946         while (!hlist_empty(&h->reqQ)) {
2947                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2948                 /* can't do anything if fifo is full */
2949                 if ((h->access.fifo_full(h))) {
2950                         printk(KERN_WARNING "cciss: fifo full\n");
2951                         break;
2952                 }
2953
2954                 /* Get the first entry from the Request Q */
2955                 removeQ(c);
2956                 h->Qdepth--;
2957
2958                 /* Tell the controller execute command */
2959                 h->access.submit_command(h, c);
2960
2961                 /* Put job onto the completed Q */
2962                 addQ(&h->cmpQ, c);
2963         }
2964 }
2965
2966 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2967 /* Zeros out the error record and then resends the command back */
2968 /* to the controller */
2969 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2970 {
2971         /* erase the old error information */
2972         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2973
2974         /* add it to software queue and then send it to the controller */
2975         addQ(&h->reqQ, c);
2976         h->Qdepth++;
2977         if (h->Qdepth > h->maxQsinceinit)
2978                 h->maxQsinceinit = h->Qdepth;
2979
2980         start_io(h);
2981 }
2982
2983 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2984         unsigned int msg_byte, unsigned int host_byte,
2985         unsigned int driver_byte)
2986 {
2987         /* inverse of macros in scsi.h */
2988         return (scsi_status_byte & 0xff) |
2989                 ((msg_byte & 0xff) << 8) |
2990                 ((host_byte & 0xff) << 16) |
2991                 ((driver_byte & 0xff) << 24);
2992 }
2993
2994 static inline int evaluate_target_status(ctlr_info_t *h,
2995                         CommandList_struct *cmd, int *retry_cmd)
2996 {
2997         unsigned char sense_key;
2998         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2999         int error_value;
3000
3001         *retry_cmd = 0;
3002         /* If we get in here, it means we got "target status", that is, scsi status */
3003         status_byte = cmd->err_info->ScsiStatus;
3004         driver_byte = DRIVER_OK;
3005         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3006
3007         if (blk_pc_request(cmd->rq))
3008                 host_byte = DID_PASSTHROUGH;
3009         else
3010                 host_byte = DID_OK;
3011
3012         error_value = make_status_bytes(status_byte, msg_byte,
3013                 host_byte, driver_byte);
3014
3015         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3016                 if (!blk_pc_request(cmd->rq))
3017                         printk(KERN_WARNING "cciss: cmd %p "
3018                                "has SCSI Status 0x%x\n",
3019                                cmd, cmd->err_info->ScsiStatus);
3020                 return error_value;
3021         }
3022
3023         /* check the sense key */
3024         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3025         /* no status or recovered error */
3026         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
3027                 error_value = 0;
3028
3029         if (check_for_unit_attention(h, cmd)) {
3030                 *retry_cmd = !blk_pc_request(cmd->rq);
3031                 return 0;
3032         }
3033
3034         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
3035                 if (error_value != 0)
3036                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3037                                " sense key = 0x%x\n", cmd, sense_key);
3038                 return error_value;
3039         }
3040
3041         /* SG_IO or similar, copy sense data back */
3042         if (cmd->rq->sense) {
3043                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3044                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3045                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3046                         cmd->rq->sense_len);
3047         } else
3048                 cmd->rq->sense_len = 0;
3049
3050         return error_value;
3051 }
3052
3053 /* checks the status of the job and calls complete buffers to mark all
3054  * buffers for the completed job. Note that this function does not need
3055  * to hold the hba/queue lock.
3056  */
3057 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3058                                     int timeout)
3059 {
3060         int retry_cmd = 0;
3061         struct request *rq = cmd->rq;
3062
3063         rq->errors = 0;
3064
3065         if (timeout)
3066                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3067
3068         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3069                 goto after_error_processing;
3070
3071         switch (cmd->err_info->CommandStatus) {
3072         case CMD_TARGET_STATUS:
3073                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3074                 break;
3075         case CMD_DATA_UNDERRUN:
3076                 if (blk_fs_request(cmd->rq)) {
3077                         printk(KERN_WARNING "cciss: cmd %p has"
3078                                " completed with data underrun "
3079                                "reported\n", cmd);
3080                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3081                 }
3082                 break;
3083         case CMD_DATA_OVERRUN:
3084                 if (blk_fs_request(cmd->rq))
3085                         printk(KERN_WARNING "cciss: cmd %p has"
3086                                " completed with data overrun "
3087                                "reported\n", cmd);
3088                 break;
3089         case CMD_INVALID:
3090                 printk(KERN_WARNING "cciss: cmd %p is "
3091                        "reported invalid\n", cmd);
3092                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3093                         cmd->err_info->CommandStatus, DRIVER_OK,
3094                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3095                 break;
3096         case CMD_PROTOCOL_ERR:
3097                 printk(KERN_WARNING "cciss: cmd %p has "
3098                        "protocol error \n", cmd);
3099                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3100                         cmd->err_info->CommandStatus, DRIVER_OK,
3101                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3102                 break;
3103         case CMD_HARDWARE_ERR:
3104                 printk(KERN_WARNING "cciss: cmd %p had "
3105                        " hardware error\n", cmd);
3106                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3107                         cmd->err_info->CommandStatus, DRIVER_OK,
3108                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3109                 break;
3110         case CMD_CONNECTION_LOST:
3111                 printk(KERN_WARNING "cciss: cmd %p had "
3112                        "connection lost\n", cmd);
3113                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3114                         cmd->err_info->CommandStatus, DRIVER_OK,
3115                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3116                 break;
3117         case CMD_ABORTED:
3118                 printk(KERN_WARNING "cciss: cmd %p was "
3119                        "aborted\n", cmd);
3120                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3121                         cmd->err_info->CommandStatus, DRIVER_OK,
3122                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3123                 break;
3124         case CMD_ABORT_FAILED:
3125                 printk(KERN_WARNING "cciss: cmd %p reports "
3126                        "abort failed\n", cmd);
3127                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3128                         cmd->err_info->CommandStatus, DRIVER_OK,
3129                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3130                 break;
3131         case CMD_UNSOLICITED_ABORT:
3132                 printk(KERN_WARNING "cciss%d: unsolicited "
3133                        "abort %p\n", h->ctlr, cmd);
3134                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3135                         retry_cmd = 1;
3136                         printk(KERN_WARNING
3137                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3138                         cmd->retry_count++;
3139                 } else
3140                         printk(KERN_WARNING
3141                                "cciss%d: %p retried too "
3142                                "many times\n", h->ctlr, cmd);
3143                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3144                         cmd->err_info->CommandStatus, DRIVER_OK,
3145                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3146                 break;
3147         case CMD_TIMEOUT:
3148                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3149                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3150                         cmd->err_info->CommandStatus, DRIVER_OK,
3151                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3152                 break;
3153         default:
3154                 printk(KERN_WARNING "cciss: cmd %p returned "
3155                        "unknown status %x\n", cmd,
3156                        cmd->err_info->CommandStatus);
3157                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3158                         cmd->err_info->CommandStatus, DRIVER_OK,
3159                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3160         }
3161
3162 after_error_processing:
3163
3164         /* We need to return this command */
3165         if (retry_cmd) {
3166                 resend_cciss_cmd(h, cmd);
3167                 return;
3168         }
3169         cmd->rq->completion_data = cmd;
3170         blk_complete_request(cmd->rq);
3171 }
3172
3173 static inline u32 cciss_tag_contains_index(u32 tag)
3174 {
3175 #define DIRECT_LOOKUP_BIT 0x10
3176         return tag & DIRECT_LOOKUP_BIT;
3177 }
3178
3179 static inline u32 cciss_tag_to_index(u32 tag)
3180 {
3181 #define DIRECT_LOOKUP_SHIFT 5
3182         return tag >> DIRECT_LOOKUP_SHIFT;
3183 }
3184
3185 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3186 {
3187 #define CCISS_ERROR_BITS 0x03
3188         return tag & ~CCISS_ERROR_BITS;
3189 }
3190
3191 static inline void cciss_mark_tag_indexed(u32 *tag)
3192 {
3193         *tag |= DIRECT_LOOKUP_BIT;
3194 }
3195
3196 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3197 {
3198         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3199 }
3200
3201 /*
3202  * Get a request and submit it to the controller.
3203  */
3204 static void do_cciss_request(struct request_queue *q)
3205 {
3206         ctlr_info_t *h = q->queuedata;
3207         CommandList_struct *c;
3208         sector_t start_blk;
3209         int seg;
3210         struct request *creq;
3211         u64bit temp64;
3212         struct scatterlist *tmp_sg;
3213         SGDescriptor_struct *curr_sg;
3214         drive_info_struct *drv;
3215         int i, dir;
3216         int sg_index = 0;
3217         int chained = 0;
3218
3219         /* We call start_io here in case there is a command waiting on the
3220          * queue that has not been sent.
3221          */
3222         if (blk_queue_plugged(q))
3223                 goto startio;
3224
3225       queue:
3226         creq = blk_peek_request(q);
3227         if (!creq)
3228                 goto startio;
3229
3230         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3231
3232         if ((c = cmd_alloc(h, 1)) == NULL)
3233                 goto full;
3234
3235         blk_start_request(creq);
3236
3237         tmp_sg = h->scatter_list[c->cmdindex];
3238         spin_unlock_irq(q->queue_lock);
3239
3240         c->cmd_type = CMD_RWREQ;
3241         c->rq = creq;
3242
3243         /* fill in the request */
3244         drv = creq->rq_disk->private_data;
3245         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3246         /* got command from pool, so use the command block index instead */
3247         /* for direct lookups. */
3248         /* The first 2 bits are reserved for controller error reporting. */
3249         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3250         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3251         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3252         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3253         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3254         c->Request.Type.Attribute = ATTR_SIMPLE;
3255         c->Request.Type.Direction =
3256             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3257         c->Request.Timeout = 0; /* Don't time out */
3258         c->Request.CDB[0] =
3259             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3260         start_blk = blk_rq_pos(creq);
3261 #ifdef CCISS_DEBUG
3262         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3263                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3264 #endif                          /* CCISS_DEBUG */
3265
3266         sg_init_table(tmp_sg, h->maxsgentries);
3267         seg = blk_rq_map_sg(q, creq, tmp_sg);
3268
3269         /* get the DMA records for the setup */
3270         if (c->Request.Type.Direction == XFER_READ)
3271                 dir = PCI_DMA_FROMDEVICE;
3272         else
3273                 dir = PCI_DMA_TODEVICE;
3274
3275         curr_sg = c->SG;
3276         sg_index = 0;
3277         chained = 0;
3278
3279         for (i = 0; i < seg; i++) {
3280                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3281                         !chained && ((seg - i) > 1)) {
3282                         /* Point to next chain block. */
3283                         curr_sg = h->cmd_sg_list[c->cmdindex];
3284                         sg_index = 0;
3285                         chained = 1;
3286                 }
3287                 curr_sg[sg_index].Len = tmp_sg[i].length;
3288                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3289                                                 tmp_sg[i].offset,
3290                                                 tmp_sg[i].length, dir);
3291                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3292                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3293                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3294                 ++sg_index;
3295         }
3296         if (chained)
3297                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3298                         (seg - (h->max_cmd_sgentries - 1)) *
3299                                 sizeof(SGDescriptor_struct));
3300
3301         /* track how many SG entries we are using */
3302         if (seg > h->maxSG)
3303                 h->maxSG = seg;
3304
3305 #ifdef CCISS_DEBUG
3306         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3307                         "chained[%d]\n",
3308                         blk_rq_sectors(creq), seg, chained);
3309 #endif                          /* CCISS_DEBUG */
3310
3311         c->Header.SGTotal = seg + chained;
3312         if (seg <= h->max_cmd_sgentries)
3313                 c->Header.SGList = c->Header.SGTotal;
3314         else
3315                 c->Header.SGList = h->max_cmd_sgentries;
3316         set_performant_mode(h, c);
3317
3318         if (likely(blk_fs_request(creq))) {
3319                 if(h->cciss_read == CCISS_READ_10) {
3320                         c->Request.CDB[1] = 0;
3321                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3322                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3323                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3324                         c->Request.CDB[5] = start_blk & 0xff;
3325                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3326                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3327                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3328                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3329                 } else {
3330                         u32 upper32 = upper_32_bits(start_blk);
3331
3332                         c->Request.CDBLen = 16;
3333                         c->Request.CDB[1]= 0;
3334                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3335                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3336                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3337                         c->Request.CDB[5]= upper32 & 0xff;
3338                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3339                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3340                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3341                         c->Request.CDB[9]= start_blk & 0xff;
3342                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3343                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3344                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3345                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3346                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3347                 }
3348         } else if (blk_pc_request(creq)) {
3349                 c->Request.CDBLen = creq->cmd_len;
3350                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3351         } else {
3352                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3353                 BUG();
3354         }
3355
3356         spin_lock_irq(q->queue_lock);
3357
3358         addQ(&h->reqQ, c);
3359         h->Qdepth++;
3360         if (h->Qdepth > h->maxQsinceinit)
3361                 h->maxQsinceinit = h->Qdepth;
3362
3363         goto queue;
3364 full:
3365         blk_stop_queue(q);
3366 startio:
3367         /* We will already have the driver lock here so not need
3368          * to lock it.
3369          */
3370         start_io(h);
3371 }
3372
3373 static inline unsigned long get_next_completion(ctlr_info_t *h)
3374 {
3375         return h->access.command_completed(h);
3376 }
3377
3378 static inline int interrupt_pending(ctlr_info_t *h)
3379 {
3380         return h->access.intr_pending(h);
3381 }
3382
3383 static inline long interrupt_not_for_us(ctlr_info_t *h)
3384 {
3385         return !(h->msi_vector || h->msix_vector) &&
3386                 ((h->access.intr_pending(h) == 0) ||
3387                 (h->interrupts_enabled == 0));
3388 }
3389
3390 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3391                         u32 raw_tag)
3392 {
3393         if (unlikely(tag_index >= h->nr_cmds)) {
3394                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3395                 return 1;
3396         }
3397         return 0;
3398 }
3399
3400 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3401                                 u32 raw_tag)
3402 {
3403         removeQ(c);
3404         if (likely(c->cmd_type == CMD_RWREQ))
3405                 complete_command(h, c, 0);
3406         else if (c->cmd_type == CMD_IOCTL_PEND)
3407                 complete(c->waiting);
3408 #ifdef CONFIG_CISS_SCSI_TAPE
3409         else if (c->cmd_type == CMD_SCSI)
3410                 complete_scsi_command(c, 0, raw_tag);
3411 #endif
3412 }
3413
3414 /* process completion of an indexed ("direct lookup") command */
3415 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3416 {
3417         u32 tag_index;
3418         CommandList_struct *c;
3419
3420         tag_index = cciss_tag_to_index(raw_tag);
3421         if (bad_tag(h, tag_index, raw_tag))
3422                 return next_command(h);
3423         c = h->cmd_pool + tag_index;
3424         finish_cmd(h, c, raw_tag);
3425         return next_command(h);
3426 }
3427
3428 /* process completion of a non-indexed command */
3429 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3430 {
3431         u32 tag;
3432         CommandList_struct *c = NULL;
3433         struct hlist_node *tmp;
3434         __u32 busaddr_masked, tag_masked;
3435
3436         tag = cciss_tag_discard_error_bits(raw_tag);
3437         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3438                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3439                 tag_masked = cciss_tag_discard_error_bits(tag);
3440                 if (busaddr_masked == tag_masked) {
3441                         finish_cmd(h, c, raw_tag);
3442                         return next_command(h);
3443                 }
3444         }
3445         bad_tag(h, h->nr_cmds + 1, raw_tag);
3446         return next_command(h);
3447 }
3448
3449 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3450 {
3451         ctlr_info_t *h = dev_id;
3452         unsigned long flags;
3453         u32 raw_tag;
3454
3455         if (interrupt_not_for_us(h))
3456                 return IRQ_NONE;
3457         /*
3458          * If there are completed commands in the completion queue,
3459          * we had better do something about it.
3460          */
3461         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3462         while (interrupt_pending(h)) {
3463                 raw_tag = get_next_completion(h);
3464                 while (raw_tag != FIFO_EMPTY) {
3465                         if (cciss_tag_contains_index(raw_tag))
3466                                 raw_tag = process_indexed_cmd(h, raw_tag);
3467                         else
3468                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3469                 }
3470         }
3471
3472         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3473         return IRQ_HANDLED;
3474 }
3475
3476 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3477  * check the interrupt pending register because it is not set.
3478  */
3479 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3480 {
3481         ctlr_info_t *h = dev_id;
3482         unsigned long flags;
3483         u32 raw_tag;
3484
3485         if (interrupt_not_for_us(h))
3486                 return IRQ_NONE;
3487         /*
3488          * If there are completed commands in the completion queue,
3489          * we had better do something about it.
3490          */
3491         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3492         raw_tag = get_next_completion(h);
3493         while (raw_tag != FIFO_EMPTY) {
3494                 if (cciss_tag_contains_index(raw_tag))
3495                         raw_tag = process_indexed_cmd(h, raw_tag);
3496                 else
3497                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3498         }
3499
3500         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3501         return IRQ_HANDLED;
3502 }
3503
3504 /**
3505  * add_to_scan_list() - add controller to rescan queue
3506  * @h:                Pointer to the controller.
3507  *
3508  * Adds the controller to the rescan queue if not already on the queue.
3509  *
3510  * returns 1 if added to the queue, 0 if skipped (could be on the
3511  * queue already, or the controller could be initializing or shutting
3512  * down).
3513  **/
3514 static int add_to_scan_list(struct ctlr_info *h)
3515 {
3516         struct ctlr_info *test_h;
3517         int found = 0;
3518         int ret = 0;
3519
3520         if (h->busy_initializing)
3521                 return 0;
3522
3523         if (!mutex_trylock(&h->busy_shutting_down))
3524                 return 0;
3525
3526         mutex_lock(&scan_mutex);
3527         list_for_each_entry(test_h, &scan_q, scan_list) {
3528                 if (test_h == h) {
3529                         found = 1;
3530                         break;
3531                 }
3532         }
3533         if (!found && !h->busy_scanning) {
3534                 INIT_COMPLETION(h->scan_wait);
3535                 list_add_tail(&h->scan_list, &scan_q);
3536                 ret = 1;
3537         }
3538         mutex_unlock(&scan_mutex);
3539         mutex_unlock(&h->busy_shutting_down);
3540
3541         return ret;
3542 }
3543
3544 /**
3545  * remove_from_scan_list() - remove controller from rescan queue
3546  * @h:                     Pointer to the controller.
3547  *
3548  * Removes the controller from the rescan queue if present. Blocks if
3549  * the controller is currently conducting a rescan.  The controller
3550  * can be in one of three states:
3551  * 1. Doesn't need a scan
3552  * 2. On the scan list, but not scanning yet (we remove it)
3553  * 3. Busy scanning (and not on the list). In this case we want to wait for
3554  *    the scan to complete to make sure the scanning thread for this
3555  *    controller is completely idle.
3556  **/
3557 static void remove_from_scan_list(struct ctlr_info *h)
3558 {
3559         struct ctlr_info *test_h, *tmp_h;
3560
3561         mutex_lock(&scan_mutex);
3562         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3563                 if (test_h == h) { /* state 2. */
3564                         list_del(&h->scan_list);
3565                         complete_all(&h->scan_wait);
3566                         mutex_unlock(&scan_mutex);
3567                         return;
3568                 }
3569         }
3570         if (h->busy_scanning) { /* state 3. */
3571                 mutex_unlock(&scan_mutex);
3572                 wait_for_completion(&h->scan_wait);
3573         } else { /* state 1, nothing to do. */
3574                 mutex_unlock(&scan_mutex);
3575         }
3576 }
3577
3578 /**
3579  * scan_thread() - kernel thread used to rescan controllers
3580  * @data:        Ignored.
3581  *
3582  * A kernel thread used scan for drive topology changes on
3583  * controllers. The thread processes only one controller at a time
3584  * using a queue.  Controllers are added to the queue using
3585  * add_to_scan_list() and removed from the queue either after done
3586  * processing or using remove_from_scan_list().
3587  *
3588  * returns 0.
3589  **/
3590 static int scan_thread(void *data)
3591 {
3592         struct ctlr_info *h;
3593
3594         while (1) {
3595                 set_current_state(TASK_INTERRUPTIBLE);
3596                 schedule();
3597                 if (kthread_should_stop())
3598                         break;
3599
3600                 while (1) {
3601                         mutex_lock(&scan_mutex);
3602                         if (list_empty(&scan_q)) {
3603                                 mutex_unlock(&scan_mutex);
3604                                 break;
3605                         }
3606
3607                         h = list_entry(scan_q.next,
3608                                        struct ctlr_info,
3609                                        scan_list);
3610                         list_del(&h->scan_list);
3611                         h->busy_scanning = 1;
3612                         mutex_unlock(&scan_mutex);
3613
3614                         rebuild_lun_table(h, 0, 0);
3615                         complete_all(&h->scan_wait);
3616                         mutex_lock(&scan_mutex);
3617                         h->busy_scanning = 0;
3618                         mutex_unlock(&scan_mutex);
3619                 }
3620         }
3621
3622         return 0;
3623 }
3624
3625 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3626 {
3627         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3628                 return 0;
3629
3630         switch (c->err_info->SenseInfo[12]) {
3631         case STATE_CHANGED:
3632                 printk(KERN_WARNING "cciss%d: a state change "
3633                         "detected, command retried\n", h->ctlr);
3634                 return 1;
3635         break;
3636         case LUN_FAILED:
3637                 printk(KERN_WARNING "cciss%d: LUN failure "
3638                         "detected, action required\n", h->ctlr);
3639                 return 1;
3640         break;
3641         case REPORT_LUNS_CHANGED:
3642                 printk(KERN_WARNING "cciss%d: report LUN data "
3643                         "changed\n", h->ctlr);
3644         /*
3645          * Here, we could call add_to_scan_list and wake up the scan thread,
3646          * except that it's quite likely that we will get more than one
3647          * REPORT_LUNS_CHANGED condition in quick succession, which means
3648          * that those which occur after the first one will likely happen
3649          * *during* the scan_thread's rescan.  And the rescan code is not
3650          * robust enough to restart in the middle, undoing what it has already
3651          * done, and it's not clear that it's even possible to do this, since
3652          * part of what it does is notify the block layer, which starts
3653          * doing it's own i/o to read partition tables and so on, and the
3654          * driver doesn't have visibility to know what might need undoing.
3655          * In any event, if possible, it is horribly complicated to get right
3656          * so we just don't do it for now.
3657          *
3658          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3659          */
3660                 return 1;
3661         break;
3662         case POWER_OR_RESET:
3663                 printk(KERN_WARNING "cciss%d: a power on "
3664                         "or device reset detected\n", h->ctlr);
3665                 return 1;
3666         break;
3667         case UNIT_ATTENTION_CLEARED:
3668                 printk(KERN_WARNING "cciss%d: unit attention "
3669                     "cleared by another initiator\n", h->ctlr);
3670                 return 1;
3671         break;
3672         default:
3673                 printk(KERN_WARNING "cciss%d: unknown "
3674                         "unit attention detected\n", h->ctlr);
3675                                 return 1;
3676         }
3677 }
3678
3679 /*
3680  *  We cannot read the structure directly, for portability we must use
3681  *   the io functions.
3682  *   This is for debug only.
3683  */
3684 #ifdef CCISS_DEBUG
3685 static void print_cfg_table(CfgTable_struct *tb)
3686 {
3687         int i;
3688         char temp_name[17];
3689
3690         printk("Controller Configuration information\n");
3691         printk("------------------------------------\n");
3692         for (i = 0; i < 4; i++)
3693                 temp_name[i] = readb(&(tb->Signature[i]));
3694         temp_name[4] = '\0';
3695         printk("   Signature = %s\n", temp_name);
3696         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3697         printk("   Transport methods supported = 0x%x\n",
3698                readl(&(tb->TransportSupport)));
3699         printk("   Transport methods active = 0x%x\n",
3700                readl(&(tb->TransportActive)));
3701         printk("   Requested transport Method = 0x%x\n",
3702                readl(&(tb->HostWrite.TransportRequest)));
3703         printk("   Coalesce Interrupt Delay = 0x%x\n",
3704                readl(&(tb->HostWrite.CoalIntDelay)));
3705         printk("   Coalesce Interrupt Count = 0x%x\n",
3706                readl(&(tb->HostWrite.CoalIntCount)));
3707         printk("   Max outstanding commands = 0x%d\n",
3708                readl(&(tb->CmdsOutMax)));
3709         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3710         for (i = 0; i < 16; i++)
3711                 temp_name[i] = readb(&(tb->ServerName[i]));
3712         temp_name[16] = '\0';
3713         printk("   Server Name = %s\n", temp_name);
3714         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3715 }
3716 #endif                          /* CCISS_DEBUG */
3717
3718 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3719 {
3720         int i, offset, mem_type, bar_type;
3721         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3722                 return 0;
3723         offset = 0;
3724         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3725                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3726                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3727                         offset += 4;
3728                 else {
3729                         mem_type = pci_resource_flags(pdev, i) &
3730                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3731                         switch (mem_type) {
3732                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3733                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3734                                 offset += 4;    /* 32 bit */
3735                                 break;
3736                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3737                                 offset += 8;
3738                                 break;
3739                         default:        /* reserved in PCI 2.2 */
3740                                 printk(KERN_WARNING
3741                                        "Base address is invalid\n");
3742                                 return -1;
3743                                 break;
3744                         }
3745                 }
3746                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3747                         return i + 1;
3748         }
3749         return -1;
3750 }
3751
3752 /* Fill in bucket_map[], given nsgs (the max number of
3753  * scatter gather elements supported) and bucket[],
3754  * which is an array of 8 integers.  The bucket[] array
3755  * contains 8 different DMA transfer sizes (in 16
3756  * byte increments) which the controller uses to fetch
3757  * commands.  This function fills in bucket_map[], which
3758  * maps a given number of scatter gather elements to one of
3759  * the 8 DMA transfer sizes.  The point of it is to allow the
3760  * controller to only do as much DMA as needed to fetch the
3761  * command, with the DMA transfer size encoded in the lower
3762  * bits of the command address.
3763  */
3764 static void  calc_bucket_map(int bucket[], int num_buckets,
3765         int nsgs, int *bucket_map)
3766 {
3767         int i, j, b, size;
3768
3769         /* even a command with 0 SGs requires 4 blocks */
3770 #define MINIMUM_TRANSFER_BLOCKS 4
3771 #define NUM_BUCKETS 8
3772         /* Note, bucket_map must have nsgs+1 entries. */
3773         for (i = 0; i <= nsgs; i++) {
3774                 /* Compute size of a command with i SG entries */
3775                 size = i + MINIMUM_TRANSFER_BLOCKS;
3776                 b = num_buckets; /* Assume the biggest bucket */
3777                 /* Find the bucket that is just big enough */
3778                 for (j = 0; j < 8; j++) {
3779                         if (bucket[j] >= size) {
3780                                 b = j;
3781                                 break;
3782                         }
3783                 }
3784                 /* for a command with i SG entries, use bucket b. */
3785                 bucket_map[i] = b;
3786         }
3787 }
3788
3789 static void
3790 cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3791 {
3792         int l = 0;
3793         __u32 trans_support;
3794         __u32 trans_offset;
3795                         /*
3796                          *  5 = 1 s/g entry or 4k
3797                          *  6 = 2 s/g entry or 8k
3798                          *  8 = 4 s/g entry or 16k
3799                          * 10 = 6 s/g entry or 24k
3800                          */
3801         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3802         unsigned long register_value;
3803
3804         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3805
3806         /* Attempt to put controller into performant mode if supported */
3807         /* Does board support performant mode? */
3808         trans_support = readl(&(h->cfgtable->TransportSupport));
3809         if (!(trans_support & PERFORMANT_MODE))
3810                 return;
3811
3812         printk(KERN_WARNING "cciss%d: Placing controller into "
3813                                 "performant mode\n", h->ctlr);
3814         /* Performant mode demands commands on a 32 byte boundary
3815          * pci_alloc_consistent aligns on page boundarys already.
3816          * Just need to check if divisible by 32
3817          */
3818         if ((sizeof(CommandList_struct) % 32) != 0) {
3819                 printk(KERN_WARNING "%s %d %s\n",
3820                         "cciss info: command size[",
3821                         (int)sizeof(CommandList_struct),
3822                         "] not divisible by 32, no performant mode..\n");
3823                 return;
3824         }
3825
3826         /* Performant mode ring buffer and supporting data structures */
3827         h->reply_pool = (__u64 *)pci_alloc_consistent(
3828                 h->pdev, h->max_commands * sizeof(__u64),
3829                 &(h->reply_pool_dhandle));
3830
3831         /* Need a block fetch table for performant mode */
3832         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3833                 sizeof(__u32)), GFP_KERNEL);
3834
3835         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3836                 goto clean_up;
3837
3838         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3839
3840         /* Controller spec: zero out this buffer. */
3841         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3842         h->reply_pool_head = h->reply_pool;
3843
3844         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3845         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3846                                 h->blockFetchTable);
3847         writel(bft[0], &h->transtable->BlockFetch0);
3848         writel(bft[1], &h->transtable->BlockFetch1);
3849         writel(bft[2], &h->transtable->BlockFetch2);
3850         writel(bft[3], &h->transtable->BlockFetch3);
3851         writel(bft[4], &h->transtable->BlockFetch4);
3852         writel(bft[5], &h->transtable->BlockFetch5);
3853         writel(bft[6], &h->transtable->BlockFetch6);
3854         writel(bft[7], &h->transtable->BlockFetch7);
3855
3856         /* size of controller ring buffer */
3857         writel(h->max_commands, &h->transtable->RepQSize);
3858         writel(1, &h->transtable->RepQCount);
3859         writel(0, &h->transtable->RepQCtrAddrLow32);
3860         writel(0, &h->transtable->RepQCtrAddrHigh32);
3861         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3862         writel(0, &h->transtable->RepQAddr0High32);
3863         writel(CFGTBL_Trans_Performant,
3864                         &(h->cfgtable->HostWrite.TransportRequest));
3865
3866         h->transMethod = CFGTBL_Trans_Performant;
3867         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3868         /* under certain very rare conditions, this can take awhile.
3869          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3870          * as we enter this code.) */
3871         for (l = 0; l < MAX_CONFIG_WAIT; l++) {
3872                 register_value = readl(h->vaddr + SA5_DOORBELL);
3873                 if (!(register_value & CFGTBL_ChangeReq))
3874                         break;
3875                 /* delay and try again */
3876                 set_current_state(TASK_INTERRUPTIBLE);
3877                 schedule_timeout(10);
3878         }
3879         register_value = readl(&(h->cfgtable->TransportActive));
3880         if (!(register_value & CFGTBL_Trans_Performant)) {
3881                 printk(KERN_WARNING "cciss: unable to get board into"
3882                                         " performant mode\n");
3883                 return;
3884         }
3885
3886         /* Change the access methods to the performant access methods */
3887         h->access = SA5_performant_access;
3888
3889         return;
3890 clean_up:
3891         kfree(h->blockFetchTable);
3892         if (h->reply_pool)
3893                 pci_free_consistent(h->pdev,
3894                                 h->max_commands * sizeof(__u64),
3895                                 h->reply_pool,
3896                                 h->reply_pool_dhandle);
3897         return;
3898
3899 } /* cciss_put_controller_into_performant_mode */
3900
3901 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3902  * controllers that are capable. If not, we use IO-APIC mode.
3903  */
3904
3905 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3906                                            struct pci_dev *pdev, __u32 board_id)
3907 {
3908 #ifdef CONFIG_PCI_MSI
3909         int err;
3910         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3911         {0, 2}, {0, 3}
3912         };
3913
3914         /* Some boards advertise MSI but don't really support it */
3915         if ((board_id == 0x40700E11) ||
3916             (board_id == 0x40800E11) ||
3917             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3918                 goto default_int_mode;
3919
3920         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3921                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3922                 if (!err) {
3923                         c->intr[0] = cciss_msix_entries[0].vector;
3924                         c->intr[1] = cciss_msix_entries[1].vector;
3925                         c->intr[2] = cciss_msix_entries[2].vector;
3926                         c->intr[3] = cciss_msix_entries[3].vector;
3927                         c->msix_vector = 1;
3928                         return;
3929                 }
3930                 if (err > 0) {
3931                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3932                                "available\n", err);
3933                         goto default_int_mode;
3934                 } else {
3935                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3936                                err);
3937                         goto default_int_mode;
3938                 }
3939         }
3940         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3941                 if (!pci_enable_msi(pdev)) {
3942                         c->msi_vector = 1;
3943                 } else {
3944                         printk(KERN_WARNING "cciss: MSI init failed\n");
3945                 }
3946         }
3947 default_int_mode:
3948 #endif                          /* CONFIG_PCI_MSI */
3949         /* if we get here we're going to use the default interrupt mode */
3950         c->intr[PERF_MODE_INT] = pdev->irq;
3951         return;
3952 }
3953
3954 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3955 {
3956         ushort subsystem_vendor_id, subsystem_device_id, command;
3957         __u32 board_id, scratchpad = 0;
3958         __u64 cfg_offset;
3959         __u32 cfg_base_addr;
3960         __u64 cfg_base_addr_index;
3961         int i, prod_index, err;
3962         __u32 trans_offset;
3963
3964         subsystem_vendor_id = pdev->subsystem_vendor;
3965         subsystem_device_id = pdev->subsystem_device;
3966         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3967                     subsystem_vendor_id);
3968
3969         for (i = 0; i < ARRAY_SIZE(products); i++) {
3970                 /* Stand aside for hpsa driver on request */
3971                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3972                         return -ENODEV;
3973                 if (board_id == products[i].board_id)
3974                         break;
3975         }
3976         prod_index = i;
3977         if (prod_index == ARRAY_SIZE(products)) {
3978                 dev_warn(&pdev->dev,
3979                         "unrecognized board ID: 0x%08lx, ignoring.\n",
3980                         (unsigned long) board_id);
3981                 return -ENODEV;
3982         }
3983
3984         /* check to see if controller has been disabled */
3985         /* BEFORE trying to enable it */
3986         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3987         if (!(command & 0x02)) {
3988                 printk(KERN_WARNING
3989                        "cciss: controller appears to be disabled\n");
3990                 return -ENODEV;
3991         }
3992
3993         err = pci_enable_device(pdev);
3994         if (err) {
3995                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3996                 return err;
3997         }
3998
3999         err = pci_request_regions(pdev, "cciss");
4000         if (err) {
4001                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
4002                        "aborting\n");
4003                 return err;
4004         }
4005
4006 #ifdef CCISS_DEBUG
4007         printk("command = %x\n", command);
4008         printk("irq = %x\n", pdev->irq);
4009         printk("board_id = %x\n", board_id);
4010 #endif                          /* CCISS_DEBUG */
4011
4012 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4013  * else we use the IO-APIC interrupt assigned to us by system ROM.
4014  */
4015         cciss_interrupt_mode(c, pdev, board_id);
4016
4017         /* find the memory BAR */
4018         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
4019                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
4020                         break;
4021         }
4022         if (i == DEVICE_COUNT_RESOURCE) {
4023                 printk(KERN_WARNING "cciss: No memory BAR found\n");
4024                 err = -ENODEV;
4025                 goto err_out_free_res;
4026         }
4027
4028         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
4029                                                  * already removed
4030                                                  */
4031
4032 #ifdef CCISS_DEBUG
4033         printk("address 0 = %lx\n", c->paddr);
4034 #endif                          /* CCISS_DEBUG */
4035         c->vaddr = remap_pci_mem(c->paddr, 0x250);
4036
4037         /* Wait for the board to become ready.  (PCI hotplug needs this.)
4038          * We poll for up to 120 secs, once per 100ms. */
4039         for (i = 0; i < 1200; i++) {
4040                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
4041                 if (scratchpad == CCISS_FIRMWARE_READY)
4042                         break;
4043                 set_current_state(TASK_INTERRUPTIBLE);
4044                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
4045         }
4046         if (scratchpad != CCISS_FIRMWARE_READY) {
4047                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
4048                 err = -ENODEV;
4049                 goto err_out_free_res;
4050         }
4051
4052         /* get the address index number */
4053         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
4054         cfg_base_addr &= (__u32) 0x0000ffff;
4055 #ifdef CCISS_DEBUG
4056         printk("cfg base address = %x\n", cfg_base_addr);
4057 #endif                          /* CCISS_DEBUG */
4058         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
4059 #ifdef CCISS_DEBUG
4060         printk("cfg base address index = %llx\n",
4061                 (unsigned long long)cfg_base_addr_index);
4062 #endif                          /* CCISS_DEBUG */
4063         if (cfg_base_addr_index == -1) {
4064                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
4065                 err = -ENODEV;
4066                 goto err_out_free_res;
4067         }
4068
4069         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
4070 #ifdef CCISS_DEBUG
4071         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
4072 #endif                          /* CCISS_DEBUG */
4073         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
4074                                                        cfg_base_addr_index) +
4075                                     cfg_offset, sizeof(CfgTable_struct));
4076         /* Find performant mode table. */
4077         trans_offset = readl(&(c->cfgtable->TransMethodOffset));
4078         c->transtable = remap_pci_mem(pci_resource_start(pdev,
4079                 cfg_base_addr_index) + cfg_offset+trans_offset,
4080                 sizeof(*c->transtable));
4081         c->board_id = board_id;
4082
4083         #ifdef CCISS_DEBUG
4084                 print_cfg_table(c->cfgtable);
4085         #endif                          /* CCISS_DEBUG */
4086
4087         /* Some controllers support Zero Memory Raid (ZMR).
4088          * When configured in ZMR mode the number of supported
4089          * commands drops to 64. So instead of just setting an
4090          * arbitrary value we make the driver a little smarter.
4091          * We read the config table to tell us how many commands
4092          * are supported on the controller then subtract 4 to
4093          * leave a little room for ioctl calls.
4094          */
4095         c->max_commands = readl(&(c->cfgtable->MaxPerformantModeCommands));
4096         c->maxsgentries = readl(&(c->cfgtable->MaxSGElements));
4097
4098         /*
4099          * Limit native command to 32 s/g elements to save dma'able memory.
4100          * Howvever spec says if 0, use 31
4101          */
4102
4103         c->max_cmd_sgentries = 31;
4104         if (c->maxsgentries > 512) {
4105                 c->max_cmd_sgentries = 32;
4106                 c->chainsize = c->maxsgentries - c->max_cmd_sgentries + 1;
4107                 c->maxsgentries -= 1;   /* account for chain pointer */
4108         } else {
4109                 c->maxsgentries = 31;   /* Default to traditional value */
4110                 c->chainsize = 0;       /* traditional */
4111         }
4112
4113         c->product_name = products[prod_index].product_name;
4114         c->access = *(products[prod_index].access);
4115         c->nr_cmds = c->max_commands - 4;
4116         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
4117             (readb(&c->cfgtable->Signature[1]) != 'I') ||
4118             (readb(&c->cfgtable->Signature[2]) != 'S') ||
4119             (readb(&c->cfgtable->Signature[3]) != 'S')) {
4120                 printk("Does not appear to be a valid CISS config table\n");
4121                 err = -ENODEV;
4122                 goto err_out_free_res;
4123         }
4124 #ifdef CONFIG_X86
4125         {
4126                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4127                 __u32 prefetch;
4128                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
4129                 prefetch |= 0x100;
4130                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
4131         }
4132 #endif
4133
4134         /* Disabling DMA prefetch and refetch for the P600.
4135          * An ASIC bug may result in accesses to invalid memory addresses.
4136          * We've disabled prefetch for some time now. Testing with XEN
4137          * kernels revealed a bug in the refetch if dom0 resides on a P600.
4138          */
4139         if(board_id == 0x3225103C) {
4140                         __u32 dma_prefetch;
4141                 __u32 dma_refetch;
4142                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
4143                 dma_prefetch |= 0x8000;
4144                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
4145                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
4146                 dma_refetch |= 0x1;
4147                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
4148         }
4149
4150 #ifdef CCISS_DEBUG
4151         printk(KERN_WARNING "Trying to put board into Performant mode\n");
4152 #endif                          /* CCISS_DEBUG */
4153         return 0;
4154
4155 err_out_free_res:
4156         /*
4157          * Deliberately omit pci_disable_device(): it does something nasty to
4158          * Smart Array controllers that pci_enable_device does not undo
4159          */
4160         pci_release_regions(pdev);
4161         cciss_put_controller_into_performant_mode(c);
4162         return err;
4163 }
4164
4165 /* Function to find the first free pointer into our hba[] array
4166  * Returns -1 if no free entries are left.
4167  */
4168 static int alloc_cciss_hba(void)
4169 {
4170         int i;
4171
4172         for (i = 0; i < MAX_CTLR; i++) {
4173                 if (!hba[i]) {
4174                         ctlr_info_t *p;
4175
4176                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4177                         if (!p)
4178                                 goto Enomem;
4179                         hba[i] = p;
4180                         return i;
4181                 }
4182         }
4183         printk(KERN_WARNING "cciss: This driver supports a maximum"
4184                " of %d controllers.\n", MAX_CTLR);
4185         return -1;
4186 Enomem:
4187         printk(KERN_ERR "cciss: out of memory.\n");
4188         return -1;
4189 }
4190
4191 static void free_hba(int n)
4192 {
4193         ctlr_info_t *h = hba[n];
4194         int i;
4195
4196         hba[n] = NULL;
4197         for (i = 0; i < h->highest_lun + 1; i++)
4198                 if (h->gendisk[i] != NULL)
4199                         put_disk(h->gendisk[i]);
4200         kfree(h);
4201 }
4202
4203 /* Send a message CDB to the firmware. */
4204 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4205 {
4206         typedef struct {
4207                 CommandListHeader_struct CommandHeader;
4208                 RequestBlock_struct Request;
4209                 ErrDescriptor_struct ErrorDescriptor;
4210         } Command;
4211         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4212         Command *cmd;
4213         dma_addr_t paddr64;
4214         uint32_t paddr32, tag;
4215         void __iomem *vaddr;
4216         int i, err;
4217
4218         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4219         if (vaddr == NULL)
4220                 return -ENOMEM;
4221
4222         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4223            CCISS commands, so they must be allocated from the lower 4GiB of
4224            memory. */
4225         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4226         if (err) {
4227                 iounmap(vaddr);
4228                 return -ENOMEM;
4229         }
4230
4231         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4232         if (cmd == NULL) {
4233                 iounmap(vaddr);
4234                 return -ENOMEM;
4235         }
4236
4237         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4238            although there's no guarantee, we assume that the address is at
4239            least 4-byte aligned (most likely, it's page-aligned). */
4240         paddr32 = paddr64;
4241
4242         cmd->CommandHeader.ReplyQueue = 0;
4243         cmd->CommandHeader.SGList = 0;
4244         cmd->CommandHeader.SGTotal = 0;
4245         cmd->CommandHeader.Tag.lower = paddr32;
4246         cmd->CommandHeader.Tag.upper = 0;
4247         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4248
4249         cmd->Request.CDBLen = 16;
4250         cmd->Request.Type.Type = TYPE_MSG;
4251         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4252         cmd->Request.Type.Direction = XFER_NONE;
4253         cmd->Request.Timeout = 0; /* Don't time out */
4254         cmd->Request.CDB[0] = opcode;
4255         cmd->Request.CDB[1] = type;
4256         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4257
4258         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4259         cmd->ErrorDescriptor.Addr.upper = 0;
4260         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4261
4262         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4263
4264         for (i = 0; i < 10; i++) {
4265                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4266                 if ((tag & ~3) == paddr32)
4267                         break;
4268                 schedule_timeout_uninterruptible(HZ);
4269         }
4270
4271         iounmap(vaddr);
4272
4273         /* we leak the DMA buffer here ... no choice since the controller could
4274            still complete the command. */
4275         if (i == 10) {
4276                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4277                         opcode, type);
4278                 return -ETIMEDOUT;
4279         }
4280
4281         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4282
4283         if (tag & 2) {
4284                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4285                         opcode, type);
4286                 return -EIO;
4287         }
4288
4289         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4290                 opcode, type);
4291         return 0;
4292 }
4293
4294 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4295 #define cciss_noop(p) cciss_message(p, 3, 0)
4296
4297 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4298 {
4299 /* the #defines are stolen from drivers/pci/msi.h. */
4300 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4301 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4302
4303         int pos;
4304         u16 control = 0;
4305
4306         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4307         if (pos) {
4308                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4309                 if (control & PCI_MSI_FLAGS_ENABLE) {
4310                         printk(KERN_INFO "cciss: resetting MSI\n");
4311                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4312                 }
4313         }
4314
4315         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4316         if (pos) {
4317                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4318                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4319                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4320                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4321                 }
4322         }
4323
4324         return 0;
4325 }
4326
4327 /* This does a hard reset of the controller using PCI power management
4328  * states. */
4329 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4330 {
4331         u16 pmcsr, saved_config_space[32];
4332         int i, pos;
4333
4334         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4335
4336         /* This is very nearly the same thing as
4337
4338            pci_save_state(pci_dev);
4339            pci_set_power_state(pci_dev, PCI_D3hot);
4340            pci_set_power_state(pci_dev, PCI_D0);
4341            pci_restore_state(pci_dev);
4342
4343            but we can't use these nice canned kernel routines on
4344            kexec, because they also check the MSI/MSI-X state in PCI
4345            configuration space and do the wrong thing when it is
4346            set/cleared.  Also, the pci_save/restore_state functions
4347            violate the ordering requirements for restoring the
4348            configuration space from the CCISS document (see the
4349            comment below).  So we roll our own .... */
4350
4351         for (i = 0; i < 32; i++)
4352                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4353
4354         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4355         if (pos == 0) {
4356                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4357                 return -ENODEV;
4358         }
4359
4360         /* Quoting from the Open CISS Specification: "The Power
4361          * Management Control/Status Register (CSR) controls the power
4362          * state of the device.  The normal operating state is D0,
4363          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4364          * the controller, place the interface device in D3 then to
4365          * D0, this causes a secondary PCI reset which will reset the
4366          * controller." */
4367
4368         /* enter the D3hot power management state */
4369         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4370         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4371         pmcsr |= PCI_D3hot;
4372         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4373
4374         schedule_timeout_uninterruptible(HZ >> 1);
4375
4376         /* enter the D0 power management state */
4377         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4378         pmcsr |= PCI_D0;
4379         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4380
4381         schedule_timeout_uninterruptible(HZ >> 1);
4382
4383         /* Restore the PCI configuration space.  The Open CISS
4384          * Specification says, "Restore the PCI Configuration
4385          * Registers, offsets 00h through 60h. It is important to
4386          * restore the command register, 16-bits at offset 04h,
4387          * last. Do not restore the configuration status register,
4388          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4389         for (i = 0; i < 32; i++) {
4390                 if (i == 2 || i == 3)
4391                         continue;
4392                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4393         }
4394         wmb();
4395         pci_write_config_word(pdev, 4, saved_config_space[2]);
4396
4397         return 0;
4398 }
4399
4400 /*
4401  *  This is it.  Find all the controllers and register them.  I really hate
4402  *  stealing all these major device numbers.
4403  *  returns the number of block devices registered.
4404  */
4405 static int __devinit cciss_init_one(struct pci_dev *pdev,
4406                                     const struct pci_device_id *ent)
4407 {
4408         int i;
4409         int j = 0;
4410         int k = 0;
4411         int rc;
4412         int dac, return_code;
4413         InquiryData_struct *inq_buff;
4414
4415         if (reset_devices) {
4416                 /* Reset the controller with a PCI power-cycle */
4417                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4418                         return -ENODEV;
4419
4420                 /* Now try to get the controller to respond to a no-op. Some
4421                    devices (notably the HP Smart Array 5i Controller) need
4422                    up to 30 seconds to respond. */
4423                 for (i=0; i<30; i++) {
4424                         if (cciss_noop(pdev) == 0)
4425                                 break;
4426
4427                         schedule_timeout_uninterruptible(HZ);
4428                 }
4429                 if (i == 30) {
4430                         printk(KERN_ERR "cciss: controller seems dead\n");
4431                         return -EBUSY;
4432                 }
4433         }
4434
4435         i = alloc_cciss_hba();
4436         if (i < 0)
4437                 return -1;
4438         hba[i]->busy_initializing = 1;
4439         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4440         INIT_HLIST_HEAD(&hba[i]->reqQ);
4441         mutex_init(&hba[i]->busy_shutting_down);
4442
4443         if (cciss_pci_init(hba[i], pdev) != 0)
4444                 goto clean_no_release_regions;
4445
4446         sprintf(hba[i]->devname, "cciss%d", i);
4447         hba[i]->ctlr = i;
4448         hba[i]->pdev = pdev;
4449
4450         init_completion(&hba[i]->scan_wait);
4451
4452         if (cciss_create_hba_sysfs_entry(hba[i]))
4453                 goto clean0;
4454
4455         /* configure PCI DMA stuff */
4456         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4457                 dac = 1;
4458         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4459                 dac = 0;
4460         else {
4461                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4462                 goto clean1;
4463         }
4464
4465         /*
4466          * register with the major number, or get a dynamic major number
4467          * by passing 0 as argument.  This is done for greater than
4468          * 8 controller support.
4469          */
4470         if (i < MAX_CTLR_ORIG)
4471                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4472         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4473         if (rc == -EBUSY || rc == -EINVAL) {
4474                 printk(KERN_ERR
4475                        "cciss:  Unable to get major number %d for %s "
4476                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4477                 goto clean1;
4478         } else {
4479                 if (i >= MAX_CTLR_ORIG)
4480                         hba[i]->major = rc;
4481         }
4482
4483         /* make sure the board interrupts are off */
4484         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4485         if (hba[i]->msi_vector || hba[i]->msix_vector) {
4486                 if (request_irq(hba[i]->intr[SIMPLE_MODE_INT],
4487                                 do_cciss_msix_intr,
4488                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4489                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4490                                hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4491                         goto clean2;
4492                 }
4493         } else {
4494                 if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intx,
4495                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4496                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4497                                hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4498                         goto clean2;
4499                 }
4500         }
4501
4502         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4503                hba[i]->devname, pdev->device, pci_name(pdev),
4504                hba[i]->intr[PERF_MODE_INT], dac ? "" : " not");
4505
4506         hba[i]->cmd_pool_bits =
4507             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4508                         * sizeof(unsigned long), GFP_KERNEL);
4509         hba[i]->cmd_pool = (CommandList_struct *)
4510             pci_alloc_consistent(hba[i]->pdev,
4511                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4512                     &(hba[i]->cmd_pool_dhandle));
4513         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4514             pci_alloc_consistent(hba[i]->pdev,
4515                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4516                     &(hba[i]->errinfo_pool_dhandle));
4517         if ((hba[i]->cmd_pool_bits == NULL)
4518             || (hba[i]->cmd_pool == NULL)
4519             || (hba[i]->errinfo_pool == NULL)) {
4520                 printk(KERN_ERR "cciss: out of memory");
4521                 goto clean4;
4522         }
4523
4524         /* Need space for temp scatter list */
4525         hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4526                                                 sizeof(struct scatterlist *),
4527                                                 GFP_KERNEL);
4528         for (k = 0; k < hba[i]->nr_cmds; k++) {
4529                 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4530                                                         hba[i]->maxsgentries,
4531                                                         GFP_KERNEL);
4532                 if (hba[i]->scatter_list[k] == NULL) {
4533                         printk(KERN_ERR "cciss%d: could not allocate "
4534                                 "s/g lists\n", i);
4535                         goto clean4;
4536                 }
4537         }
4538         hba[i]->cmd_sg_list = cciss_allocate_sg_chain_blocks(hba[i],
4539                 hba[i]->chainsize, hba[i]->nr_cmds);
4540         if (!hba[i]->cmd_sg_list && hba[i]->chainsize > 0)
4541                 goto clean4;
4542
4543         spin_lock_init(&hba[i]->lock);
4544
4545         /* Initialize the pdev driver private data.
4546            have it point to hba[i].  */
4547         pci_set_drvdata(pdev, hba[i]);
4548         /* command and error info recs zeroed out before
4549            they are used */
4550         memset(hba[i]->cmd_pool_bits, 0,
4551                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4552                         * sizeof(unsigned long));
4553
4554         hba[i]->num_luns = 0;
4555         hba[i]->highest_lun = -1;
4556         for (j = 0; j < CISS_MAX_LUN; j++) {
4557                 hba[i]->drv[j] = NULL;
4558                 hba[i]->gendisk[j] = NULL;
4559         }
4560
4561         cciss_scsi_setup(i);
4562
4563         /* Turn the interrupts on so we can service requests */
4564         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4565
4566         /* Get the firmware version */
4567         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4568         if (inq_buff == NULL) {
4569                 printk(KERN_ERR "cciss: out of memory\n");
4570                 goto clean4;
4571         }
4572
4573         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4574                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4575         if (return_code == IO_OK) {
4576                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4577                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4578                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4579                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4580         } else {         /* send command failed */
4581                 printk(KERN_WARNING "cciss: unable to determine firmware"
4582                         " version of controller\n");
4583         }
4584         kfree(inq_buff);
4585
4586         cciss_procinit(i);
4587
4588         hba[i]->cciss_max_sectors = 8192;
4589
4590         rebuild_lun_table(hba[i], 1, 0);
4591         hba[i]->busy_initializing = 0;
4592         return 1;
4593
4594 clean4:
4595         kfree(hba[i]->cmd_pool_bits);
4596         /* Free up sg elements */
4597         for (k = 0; k < hba[i]->nr_cmds; k++)
4598                 kfree(hba[i]->scatter_list[k]);
4599         kfree(hba[i]->scatter_list);
4600         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4601         if (hba[i]->cmd_pool)
4602                 pci_free_consistent(hba[i]->pdev,
4603                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4604                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4605         if (hba[i]->errinfo_pool)
4606                 pci_free_consistent(hba[i]->pdev,
4607                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4608                                     hba[i]->errinfo_pool,
4609                                     hba[i]->errinfo_pool_dhandle);
4610         free_irq(hba[i]->intr[PERF_MODE_INT], hba[i]);
4611 clean2:
4612         unregister_blkdev(hba[i]->major, hba[i]->devname);
4613 clean1:
4614         cciss_destroy_hba_sysfs_entry(hba[i]);
4615 clean0:
4616         pci_release_regions(pdev);
4617 clean_no_release_regions:
4618         hba[i]->busy_initializing = 0;
4619
4620         /*
4621          * Deliberately omit pci_disable_device(): it does something nasty to
4622          * Smart Array controllers that pci_enable_device does not undo
4623          */
4624         pci_set_drvdata(pdev, NULL);
4625         free_hba(i);
4626         return -1;
4627 }
4628
4629 static void cciss_shutdown(struct pci_dev *pdev)
4630 {
4631         ctlr_info_t *h;
4632         char *flush_buf;
4633         int return_code;
4634
4635         h = pci_get_drvdata(pdev);
4636         flush_buf = kzalloc(4, GFP_KERNEL);
4637         if (!flush_buf) {
4638                 printk(KERN_WARNING
4639                         "cciss:%d cache not flushed, out of memory.\n",
4640                         h->ctlr);
4641                 return;
4642         }
4643         /* write all data in the battery backed cache to disk */
4644         memset(flush_buf, 0, 4);
4645         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4646                 4, 0, CTLR_LUNID, TYPE_CMD);
4647         kfree(flush_buf);
4648         if (return_code != IO_OK)
4649                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4650                         h->ctlr);
4651         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4652         free_irq(h->intr[PERF_MODE_INT], h);
4653 }
4654
4655 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4656 {
4657         ctlr_info_t *tmp_ptr;
4658         int i, j;
4659
4660         if (pci_get_drvdata(pdev) == NULL) {
4661                 printk(KERN_ERR "cciss: Unable to remove device \n");
4662                 return;
4663         }
4664
4665         tmp_ptr = pci_get_drvdata(pdev);
4666         i = tmp_ptr->ctlr;
4667         if (hba[i] == NULL) {
4668                 printk(KERN_ERR "cciss: device appears to "
4669                        "already be removed \n");
4670                 return;
4671         }
4672
4673         mutex_lock(&hba[i]->busy_shutting_down);
4674
4675         remove_from_scan_list(hba[i]);
4676         remove_proc_entry(hba[i]->devname, proc_cciss);
4677         unregister_blkdev(hba[i]->major, hba[i]->devname);
4678
4679         /* remove it from the disk list */
4680         for (j = 0; j < CISS_MAX_LUN; j++) {
4681                 struct gendisk *disk = hba[i]->gendisk[j];
4682                 if (disk) {
4683                         struct request_queue *q = disk->queue;
4684
4685                         if (disk->flags & GENHD_FL_UP) {
4686                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4687                                 del_gendisk(disk);
4688                         }
4689                         if (q)
4690                                 blk_cleanup_queue(q);
4691                 }
4692         }
4693
4694 #ifdef CONFIG_CISS_SCSI_TAPE
4695         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4696 #endif
4697
4698         cciss_shutdown(pdev);
4699
4700 #ifdef CONFIG_PCI_MSI
4701         if (hba[i]->msix_vector)
4702                 pci_disable_msix(hba[i]->pdev);
4703         else if (hba[i]->msi_vector)
4704                 pci_disable_msi(hba[i]->pdev);
4705 #endif                          /* CONFIG_PCI_MSI */
4706
4707         iounmap(hba[i]->vaddr);
4708
4709         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4710                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4711         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4712                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4713         kfree(hba[i]->cmd_pool_bits);
4714         /* Free up sg elements */
4715         for (j = 0; j < hba[i]->nr_cmds; j++)
4716                 kfree(hba[i]->scatter_list[j]);
4717         kfree(hba[i]->scatter_list);
4718         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4719         /*
4720          * Deliberately omit pci_disable_device(): it does something nasty to
4721          * Smart Array controllers that pci_enable_device does not undo
4722          */
4723         pci_release_regions(pdev);
4724         pci_set_drvdata(pdev, NULL);
4725         cciss_destroy_hba_sysfs_entry(hba[i]);
4726         mutex_unlock(&hba[i]->busy_shutting_down);
4727         free_hba(i);
4728 }
4729
4730 static struct pci_driver cciss_pci_driver = {
4731         .name = "cciss",
4732         .probe = cciss_init_one,
4733         .remove = __devexit_p(cciss_remove_one),
4734         .id_table = cciss_pci_device_id,        /* id_table */
4735         .shutdown = cciss_shutdown,
4736 };
4737
4738 /*
4739  *  This is it.  Register the PCI driver information for the cards we control
4740  *  the OS will call our registered routines when it finds one of our cards.
4741  */
4742 static int __init cciss_init(void)
4743 {
4744         int err;
4745
4746         /*
4747          * The hardware requires that commands are aligned on a 64-bit
4748          * boundary. Given that we use pci_alloc_consistent() to allocate an
4749          * array of them, the size must be a multiple of 8 bytes.
4750          */
4751         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4752         printk(KERN_INFO DRIVER_NAME "\n");
4753
4754         err = bus_register(&cciss_bus_type);
4755         if (err)
4756                 return err;
4757
4758         /* Start the scan thread */
4759         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4760         if (IS_ERR(cciss_scan_thread)) {
4761                 err = PTR_ERR(cciss_scan_thread);
4762                 goto err_bus_unregister;
4763         }
4764
4765         /* Register for our PCI devices */
4766         err = pci_register_driver(&cciss_pci_driver);
4767         if (err)
4768                 goto err_thread_stop;
4769
4770         return err;
4771
4772 err_thread_stop:
4773         kthread_stop(cciss_scan_thread);
4774 err_bus_unregister:
4775         bus_unregister(&cciss_bus_type);
4776
4777         return err;
4778 }
4779
4780 static void __exit cciss_cleanup(void)
4781 {
4782         int i;
4783
4784         pci_unregister_driver(&cciss_pci_driver);
4785         /* double check that all controller entrys have been removed */
4786         for (i = 0; i < MAX_CTLR; i++) {
4787                 if (hba[i] != NULL) {
4788                         printk(KERN_WARNING "cciss: had to remove"
4789                                " controller %d\n", i);
4790                         cciss_remove_one(hba[i]->pdev);
4791                 }
4792         }
4793         kthread_stop(cciss_scan_thread);
4794         remove_proc_entry("driver/cciss", NULL);
4795         bus_unregister(&cciss_bus_type);
4796 }
4797
4798 module_init(cciss_init);
4799 module_exit(cciss_cleanup);