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_unlocked_open(struct block_device *bdev, fmode_t mode);
 182 static int cciss_release(struct gendisk *disk, fmode_t mode);
 183 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 184                     unsigned int cmd, unsigned long arg);
 185 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 186                        unsigned int cmd, unsigned long arg);
 187 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 188
 189 static int cciss_revalidate(struct gendisk *disk);
 190 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 191 static int deregister_disk(ctlr_info_t *h, int drv_index,
 192                            int clear_all, int via_ioctl);
 193
 194 static void cciss_read_capacity(int ctlr, int logvol,
 195                         sector_t *total_size, unsigned int *block_size);
 196 static void cciss_read_capacity_16(int ctlr, int logvol,
 197                         sector_t *total_size, unsigned int *block_size);
 198 static void cciss_geometry_inquiry(int ctlr, int logvol,
 199                         sector_t total_size,
 200                         unsigned int block_size, InquiryData_struct *inq_buff,
 201                                    drive_info_struct *drv);
 202 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 203                                            __u32);
 204 static void start_io(ctlr_info_t *h);
 205 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 206                         __u8 page_code, unsigned char scsi3addr[],
 207                         int cmd_type);
 208 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 209         int attempt_retry);
 210 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 211
 212 static int add_to_scan_list(struct ctlr_info *h);
 213 static int scan_thread(void *data);
 214 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 215 static void cciss_hba_release(struct device *dev);
 216 static void cciss_device_release(struct device *dev);
 217 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 218 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 219 static inline u32 next_command(ctlr_info_t *h);
 220
 221 /* performant mode helper functions */
 222 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 223                                 int *bucket_map);
 224 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 225
 226 #ifdef CONFIG_PROC_FS
 227 static void cciss_procinit(int i);
 228 #else
 229 static void cciss_procinit(int i)
 230 {
 231 }
 232 #endif                          /* CONFIG_PROC_FS */
 233
 234 #ifdef CONFIG_COMPAT
 235 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 236                               unsigned, unsigned long);
 237 #endif
 238
 239 static const struct block_device_operations cciss_fops = {
 240         .owner = THIS_MODULE,
 241         .open = cciss_unlocked_open,
 242         .release = cciss_release,
 243         .ioctl = do_ioctl,
 244         .getgeo = cciss_getgeo,
 245 #ifdef CONFIG_COMPAT
 246         .compat_ioctl = cciss_compat_ioctl,
 247 #endif
 248         .revalidate_disk = cciss_revalidate,
 249 };
 250
 251 /* set_performant_mode: Modify the tag for cciss performant
 252  * set bit 0 for pull model, bits 3-1 for block fetch
 253  * register number
 254  */
 255 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 256 {
 257         if (likely(h->transMethod == CFGTBL_Trans_Performant))
 258                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 259 }
 260
 261 /*
 262  * Enqueuing and dequeuing functions for cmdlists.
 263  */
 264 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 265 {
 266         hlist_add_head(&c->list, list);
 267 }
 268
 269 static inline void removeQ(CommandList_struct *c)
 270 {
 271         /*
 272          * After kexec/dump some commands might still
 273          * be in flight, which the firmware will try
 274          * to complete. Resetting the firmware doesn't work
 275          * with old fw revisions, so we have to mark
 276          * them off as 'stale' to prevent the driver from
 277          * falling over.
 278          */
 279         if (WARN_ON(hlist_unhashed(&c->list))) {
 280                 c->cmd_type = CMD_MSG_STALE;
 281                 return;
 282         }
 283
 284         hlist_del_init(&c->list);
 285 }
 286
 287 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 288         CommandList_struct *c)
 289 {
 290         unsigned long flags;
 291         set_performant_mode(h, c);
 292         spin_lock_irqsave(&h->lock, flags);
 293         addQ(&h->reqQ, c);
 294         h->Qdepth++;
 295         start_io(h);
 296         spin_unlock_irqrestore(&h->lock, flags);
 297 }
 298
 299 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 300         int nr_cmds)
 301 {
 302         int i;
 303
 304         if (!cmd_sg_list)
 305                 return;
 306         for (i = 0; i < nr_cmds; i++) {
 307                 kfree(cmd_sg_list[i]);
 308                 cmd_sg_list[i] = NULL;
 309         }
 310         kfree(cmd_sg_list);
 311 }
 312
 313 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 314         ctlr_info_t *h, int chainsize, int nr_cmds)
 315 {
 316         int j;
 317         SGDescriptor_struct **cmd_sg_list;
 318
 319         if (chainsize <= 0)
 320                 return NULL;
 321
 322         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 323         if (!cmd_sg_list)
 324                 return NULL;
 325
 326         /* Build up chain blocks for each command */
 327         for (j = 0; j < nr_cmds; j++) {
 328                 /* Need a block of chainsized s/g elements. */
 329                 cmd_sg_list[j] = kmalloc((chainsize *
 330                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 331                 if (!cmd_sg_list[j]) {
 332                         dev_err(&h->pdev->dev, "Cannot get memory "
 333                                 "for s/g chains.\n");
 334                         goto clean;
 335                 }
 336         }
 337         return cmd_sg_list;
 338 clean:
 339         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 340         return NULL;
 341 }
 342
 343 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 344 {
 345         SGDescriptor_struct *chain_sg;
 346         u64bit temp64;
 347
 348         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 349                 return;
 350
 351         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 352         temp64.val32.lower = chain_sg->Addr.lower;
 353         temp64.val32.upper = chain_sg->Addr.upper;
 354         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 355 }
 356
 357 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 358         SGDescriptor_struct *chain_block, int len)
 359 {
 360         SGDescriptor_struct *chain_sg;
 361         u64bit temp64;
 362
 363         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 364         chain_sg->Ext = CCISS_SG_CHAIN;
 365         chain_sg->Len = len;
 366         temp64.val = pci_map_single(h->pdev, chain_block, len,
 367                                 PCI_DMA_TODEVICE);
 368         chain_sg->Addr.lower = temp64.val32.lower;
 369         chain_sg->Addr.upper = temp64.val32.upper;
 370 }
 371
 372 #include "cciss_scsi.c"         /* For SCSI tape support */
 373
 374 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 375         "UNKNOWN"
 376 };
 377 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
 378
 379 #ifdef CONFIG_PROC_FS
 380
 381 /*
 382  * Report information about this controller.
 383  */
 384 #define ENG_GIG 1000000000
 385 #define ENG_GIG_FACTOR (ENG_GIG/512)
 386 #define ENGAGE_SCSI     "engage scsi"
 387
 388 static struct proc_dir_entry *proc_cciss;
 389
 390 static void cciss_seq_show_header(struct seq_file *seq)
 391 {
 392         ctlr_info_t *h = seq->private;
 393
 394         seq_printf(seq, "%s: HP %s Controller\n"
 395                 "Board ID: 0x%08lx\n"
 396                 "Firmware Version: %c%c%c%c\n"
 397                 "IRQ: %d\n"
 398                 "Logical drives: %d\n"
 399                 "Current Q depth: %d\n"
 400                 "Current # commands on controller: %d\n"
 401                 "Max Q depth since init: %d\n"
 402                 "Max # commands on controller since init: %d\n"
 403                 "Max SG entries since init: %d\n",
 404                 h->devname,
 405                 h->product_name,
 406                 (unsigned long)h->board_id,
 407                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 408                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 409                 h->num_luns,
 410                 h->Qdepth, h->commands_outstanding,
 411                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 412
 413 #ifdef CONFIG_CISS_SCSI_TAPE
 414         cciss_seq_tape_report(seq, h->ctlr);
 415 #endif /* CONFIG_CISS_SCSI_TAPE */
 416 }
 417
 418 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 419 {
 420         ctlr_info_t *h = seq->private;
 421         unsigned ctlr = h->ctlr;
 422         unsigned long flags;
 423
 424         /* prevent displaying bogus info during configuration
 425          * or deconfiguration of a logical volume
 426          */
 427         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 428         if (h->busy_configuring) {
 429                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 430                 return ERR_PTR(-EBUSY);
 431         }
 432         h->busy_configuring = 1;
 433         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 434
 435         if (*pos == 0)
 436                 cciss_seq_show_header(seq);
 437
 438         return pos;
 439 }
 440
 441 static int cciss_seq_show(struct seq_file *seq, void *v)
 442 {
 443         sector_t vol_sz, vol_sz_frac;
 444         ctlr_info_t *h = seq->private;
 445         unsigned ctlr = h->ctlr;
 446         loff_t *pos = v;
 447         drive_info_struct *drv = h->drv[*pos];
 448
 449         if (*pos > h->highest_lun)
 450                 return 0;
 451
 452         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 453                 return 0;
 454
 455         if (drv->heads == 0)
 456                 return 0;
 457
 458         vol_sz = drv->nr_blocks;
 459         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 460         vol_sz_frac *= 100;
 461         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 462
 463         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 464                 drv->raid_level = RAID_UNKNOWN;
 465         seq_printf(seq, "cciss/c%dd%d:"
 466                         "\t%4u.%02uGB\tRAID %s\n",
 467                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 468                         raid_label[drv->raid_level]);
 469         return 0;
 470 }
 471
 472 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 473 {
 474         ctlr_info_t *h = seq->private;
 475
 476         if (*pos > h->highest_lun)
 477                 return NULL;
 478         *pos += 1;
 479
 480         return pos;
 481 }
 482
 483 static void cciss_seq_stop(struct seq_file *seq, void *v)
 484 {
 485         ctlr_info_t *h = seq->private;
 486
 487         /* Only reset h->busy_configuring if we succeeded in setting
 488          * it during cciss_seq_start. */
 489         if (v == ERR_PTR(-EBUSY))
 490                 return;
 491
 492         h->busy_configuring = 0;
 493 }
 494
 495 static const struct seq_operations cciss_seq_ops = {
 496         .start = cciss_seq_start,
 497         .show  = cciss_seq_show,
 498         .next  = cciss_seq_next,
 499         .stop  = cciss_seq_stop,
 500 };
 501
 502 static int cciss_seq_open(struct inode *inode, struct file *file)
 503 {
 504         int ret = seq_open(file, &cciss_seq_ops);
 505         struct seq_file *seq = file->private_data;
 506
 507         if (!ret)
 508                 seq->private = PDE(inode)->data;
 509
 510         return ret;
 511 }
 512
 513 static ssize_t
 514 cciss_proc_write(struct file *file, const char __user *buf,
 515                  size_t length, loff_t *ppos)
 516 {
 517         int err;
 518         char *buffer;
 519
 520 #ifndef CONFIG_CISS_SCSI_TAPE
 521         return -EINVAL;
 522 #endif
 523
 524         if (!buf || length > PAGE_SIZE - 1)
 525                 return -EINVAL;
 526
 527         buffer = (char *)__get_free_page(GFP_KERNEL);
 528         if (!buffer)
 529                 return -ENOMEM;
 530
 531         err = -EFAULT;
 532         if (copy_from_user(buffer, buf, length))
 533                 goto out;
 534         buffer[length] = '\0';
 535
 536 #ifdef CONFIG_CISS_SCSI_TAPE
 537         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 538                 struct seq_file *seq = file->private_data;
 539                 ctlr_info_t *h = seq->private;
 540
 541                 err = cciss_engage_scsi(h->ctlr);
 542                 if (err == 0)
 543                         err = length;
 544         } else
 545 #endif /* CONFIG_CISS_SCSI_TAPE */
 546                 err = -EINVAL;
 547         /* might be nice to have "disengage" too, but it's not
 548            safely possible. (only 1 module use count, lock issues.) */
 549
 550 out:
 551         free_page((unsigned long)buffer);
 552         return err;
 553 }
 554
 555 static const struct file_operations cciss_proc_fops = {
 556         .owner   = THIS_MODULE,
 557         .open    = cciss_seq_open,
 558         .read    = seq_read,
 559         .llseek  = seq_lseek,
 560         .release = seq_release,
 561         .write   = cciss_proc_write,
 562 };
 563
 564 static void __devinit cciss_procinit(int i)
 565 {
 566         struct proc_dir_entry *pde;
 567
 568         if (proc_cciss == NULL)
 569                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 570         if (!proc_cciss)
 571                 return;
 572         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 573                                         S_IROTH, proc_cciss,
 574                                         &cciss_proc_fops, hba[i]);
 575 }
 576 #endif                          /* CONFIG_PROC_FS */
 577
 578 #define MAX_PRODUCT_NAME_LEN 19
 579
 580 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 581 #define to_drv(n) container_of(n, drive_info_struct, dev)
 582
 583 static ssize_t host_store_rescan(struct device *dev,
 584                                  struct device_attribute *attr,
 585                                  const char *buf, size_t count)
 586 {
 587         struct ctlr_info *h = to_hba(dev);
 588
 589         add_to_scan_list(h);
 590         wake_up_process(cciss_scan_thread);
 591         wait_for_completion_interruptible(&h->scan_wait);
 592
 593         return count;
 594 }
 595 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 596
 597 static ssize_t dev_show_unique_id(struct device *dev,
 598                                  struct device_attribute *attr,
 599                                  char *buf)
 600 {
 601         drive_info_struct *drv = to_drv(dev);
 602         struct ctlr_info *h = to_hba(drv->dev.parent);
 603         __u8 sn[16];
 604         unsigned long flags;
 605         int ret = 0;
 606
 607         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 608         if (h->busy_configuring)
 609                 ret = -EBUSY;
 610         else
 611                 memcpy(sn, drv->serial_no, sizeof(sn));
 612         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 613
 614         if (ret)
 615                 return ret;
 616         else
 617                 return snprintf(buf, 16 * 2 + 2,
 618                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 619                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 620                                 sn[0], sn[1], sn[2], sn[3],
 621                                 sn[4], sn[5], sn[6], sn[7],
 622                                 sn[8], sn[9], sn[10], sn[11],
 623                                 sn[12], sn[13], sn[14], sn[15]);
 624 }
 625 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 626
 627 static ssize_t dev_show_vendor(struct device *dev,
 628                                struct device_attribute *attr,
 629                                char *buf)
 630 {
 631         drive_info_struct *drv = to_drv(dev);
 632         struct ctlr_info *h = to_hba(drv->dev.parent);
 633         char vendor[VENDOR_LEN + 1];
 634         unsigned long flags;
 635         int ret = 0;
 636
 637         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 638         if (h->busy_configuring)
 639                 ret = -EBUSY;
 640         else
 641                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 642         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 643
 644         if (ret)
 645                 return ret;
 646         else
 647                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 648 }
 649 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 650
 651 static ssize_t dev_show_model(struct device *dev,
 652                               struct device_attribute *attr,
 653                               char *buf)
 654 {
 655         drive_info_struct *drv = to_drv(dev);
 656         struct ctlr_info *h = to_hba(drv->dev.parent);
 657         char model[MODEL_LEN + 1];
 658         unsigned long flags;
 659         int ret = 0;
 660
 661         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 662         if (h->busy_configuring)
 663                 ret = -EBUSY;
 664         else
 665                 memcpy(model, drv->model, MODEL_LEN + 1);
 666         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 667
 668         if (ret)
 669                 return ret;
 670         else
 671                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 672 }
 673 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 674
 675 static ssize_t dev_show_rev(struct device *dev,
 676                             struct device_attribute *attr,
 677                             char *buf)
 678 {
 679         drive_info_struct *drv = to_drv(dev);
 680         struct ctlr_info *h = to_hba(drv->dev.parent);
 681         char rev[REV_LEN + 1];
 682         unsigned long flags;
 683         int ret = 0;
 684
 685         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 686         if (h->busy_configuring)
 687                 ret = -EBUSY;
 688         else
 689                 memcpy(rev, drv->rev, REV_LEN + 1);
 690         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 691
 692         if (ret)
 693                 return ret;
 694         else
 695                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 696 }
 697 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 698
 699 static ssize_t cciss_show_lunid(struct device *dev,
 700                                 struct device_attribute *attr, char *buf)
 701 {
 702         drive_info_struct *drv = to_drv(dev);
 703         struct ctlr_info *h = to_hba(drv->dev.parent);
 704         unsigned long flags;
 705         unsigned char lunid[8];
 706
 707         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 708         if (h->busy_configuring) {
 709                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 710                 return -EBUSY;
 711         }
 712         if (!drv->heads) {
 713                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 714                 return -ENOTTY;
 715         }
 716         memcpy(lunid, drv->LunID, sizeof(lunid));
 717         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 718         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 719                 lunid[0], lunid[1], lunid[2], lunid[3],
 720                 lunid[4], lunid[5], lunid[6], lunid[7]);
 721 }
 722 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 723
 724 static ssize_t cciss_show_raid_level(struct device *dev,
 725                                      struct device_attribute *attr, char *buf)
 726 {
 727         drive_info_struct *drv = to_drv(dev);
 728         struct ctlr_info *h = to_hba(drv->dev.parent);
 729         int raid;
 730         unsigned long flags;
 731
 732         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 733         if (h->busy_configuring) {
 734                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 735                 return -EBUSY;
 736         }
 737         raid = drv->raid_level;
 738         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 739         if (raid < 0 || raid > RAID_UNKNOWN)
 740                 raid = RAID_UNKNOWN;
 741
 742         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 743                         raid_label[raid]);
 744 }
 745 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 746
 747 static ssize_t cciss_show_usage_count(struct device *dev,
 748                                       struct device_attribute *attr, char *buf)
 749 {
 750         drive_info_struct *drv = to_drv(dev);
 751         struct ctlr_info *h = to_hba(drv->dev.parent);
 752         unsigned long flags;
 753         int count;
 754
 755         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 756         if (h->busy_configuring) {
 757                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 758                 return -EBUSY;
 759         }
 760         count = drv->usage_count;
 761         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 762         return snprintf(buf, 20, "%d\n", count);
 763 }
 764 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 765
 766 static struct attribute *cciss_host_attrs[] = {
 767         &dev_attr_rescan.attr,
 768         NULL
 769 };
 770
 771 static struct attribute_group cciss_host_attr_group = {
 772         .attrs = cciss_host_attrs,
 773 };
 774
 775 static const struct attribute_group *cciss_host_attr_groups[] = {
 776         &cciss_host_attr_group,
 777         NULL
 778 };
 779
 780 static struct device_type cciss_host_type = {
 781         .name           = "cciss_host",
 782         .groups         = cciss_host_attr_groups,
 783         .release        = cciss_hba_release,
 784 };
 785
 786 static struct attribute *cciss_dev_attrs[] = {
 787         &dev_attr_unique_id.attr,
 788         &dev_attr_model.attr,
 789         &dev_attr_vendor.attr,
 790         &dev_attr_rev.attr,
 791         &dev_attr_lunid.attr,
 792         &dev_attr_raid_level.attr,
 793         &dev_attr_usage_count.attr,
 794         NULL
 795 };
 796
 797 static struct attribute_group cciss_dev_attr_group = {
 798         .attrs = cciss_dev_attrs,
 799 };
 800
 801 static const struct attribute_group *cciss_dev_attr_groups[] = {
 802         &cciss_dev_attr_group,
 803         NULL
 804 };
 805
 806 static struct device_type cciss_dev_type = {
 807         .name           = "cciss_device",
 808         .groups         = cciss_dev_attr_groups,
 809         .release        = cciss_device_release,
 810 };
 811
 812 static struct bus_type cciss_bus_type = {
 813         .name           = "cciss",
 814 };
 815
 816 /*
 817  * cciss_hba_release is called when the reference count
 818  * of h->dev goes to zero.
 819  */
 820 static void cciss_hba_release(struct device *dev)
 821 {
 822         /*
 823          * nothing to do, but need this to avoid a warning
 824          * about not having a release handler from lib/kref.c.
 825          */
 826 }
 827
 828 /*
 829  * Initialize sysfs entry for each controller.  This sets up and registers
 830  * the 'cciss#' directory for each individual controller under
 831  * /sys/bus/pci/devices/<dev>/.
 832  */
 833 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 834 {
 835         device_initialize(&h->dev);
 836         h->dev.type = &cciss_host_type;
 837         h->dev.bus = &cciss_bus_type;
 838         dev_set_name(&h->dev, "%s", h->devname);
 839         h->dev.parent = &h->pdev->dev;
 840
 841         return device_add(&h->dev);
 842 }
 843
 844 /*
 845  * Remove sysfs entries for an hba.
 846  */
 847 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 848 {
 849         device_del(&h->dev);
 850         put_device(&h->dev); /* final put. */
 851 }
 852
 853 /* cciss_device_release is called when the reference count
 854  * of h->drv[x]dev goes to zero.
 855  */
 856 static void cciss_device_release(struct device *dev)
 857 {
 858         drive_info_struct *drv = to_drv(dev);
 859         kfree(drv);
 860 }
 861
 862 /*
 863  * Initialize sysfs for each logical drive.  This sets up and registers
 864  * the 'c#d#' directory for each individual logical drive under
 865  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 866  * /sys/block/cciss!c#d# to this entry.
 867  */
 868 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 869                                        int drv_index)
 870 {
 871         struct device *dev;
 872
 873         if (h->drv[drv_index]->device_initialized)
 874                 return 0;
 875
 876         dev = &h->drv[drv_index]->dev;
 877         device_initialize(dev);
 878         dev->type = &cciss_dev_type;
 879         dev->bus = &cciss_bus_type;
 880         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 881         dev->parent = &h->dev;
 882         h->drv[drv_index]->device_initialized = 1;
 883         return device_add(dev);
 884 }
 885
 886 /*
 887  * Remove sysfs entries for a logical drive.
 888  */
 889 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 890         int ctlr_exiting)
 891 {
 892         struct device *dev = &h->drv[drv_index]->dev;
 893
 894         /* special case for c*d0, we only destroy it on controller exit */
 895         if (drv_index == 0 && !ctlr_exiting)
 896                 return;
 897
 898         device_del(dev);
 899         put_device(dev); /* the "final" put. */
 900         h->drv[drv_index] = NULL;
 901 }
 902
 903 /*
 904  * For operations that cannot sleep, a command block is allocated at init,
 905  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 906  * which ones are free or in use.  For operations that can wait for kmalloc
 907  * to possible sleep, this routine can be called with get_from_pool set to 0.
 908  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 909  */
 910 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 911 {
 912         CommandList_struct *c;
 913         int i;
 914         u64bit temp64;
 915         dma_addr_t cmd_dma_handle, err_dma_handle;
 916
 917         if (!get_from_pool) {
 918                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 919                         sizeof(CommandList_struct), &cmd_dma_handle);
 920                 if (c == NULL)
 921                         return NULL;
 922                 memset(c, 0, sizeof(CommandList_struct));
 923
 924                 c->cmdindex = -1;
 925
 926                 c->err_info = (ErrorInfo_struct *)
 927                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 928                             &err_dma_handle);
 929
 930                 if (c->err_info == NULL) {
 931                         pci_free_consistent(h->pdev,
 932                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 933                         return NULL;
 934                 }
 935                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 936         } else {                /* get it out of the controllers pool */
 937
 938                 do {
 939                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 940                         if (i == h->nr_cmds)
 941                                 return NULL;
 942                 } while (test_and_set_bit
 943                          (i & (BITS_PER_LONG - 1),
 944                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 945 #ifdef CCISS_DEBUG
 946                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 947 #endif
 948                 c = h->cmd_pool + i;
 949                 memset(c, 0, sizeof(CommandList_struct));
 950                 cmd_dma_handle = h->cmd_pool_dhandle
 951                     + i * sizeof(CommandList_struct);
 952                 c->err_info = h->errinfo_pool + i;
 953                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 954                 err_dma_handle = h->errinfo_pool_dhandle
 955                     + i * sizeof(ErrorInfo_struct);
 956                 h->nr_allocs++;
 957
 958                 c->cmdindex = i;
 959         }
 960
 961         INIT_HLIST_NODE(&c->list);
 962         c->busaddr = (__u32) cmd_dma_handle;
 963         temp64.val = (__u64) err_dma_handle;
 964         c->ErrDesc.Addr.lower = temp64.val32.lower;
 965         c->ErrDesc.Addr.upper = temp64.val32.upper;
 966         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 967
 968         c->ctlr = h->ctlr;
 969         return c;
 970 }
 971
 972 /*
 973  * Frees a command block that was previously allocated with cmd_alloc().
 974  */
 975 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 976 {
 977         int i;
 978         u64bit temp64;
 979
 980         if (!got_from_pool) {
 981                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 982                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 983                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 984                                     c->err_info, (dma_addr_t) temp64.val);
 985                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 986                                     c, (dma_addr_t) c->busaddr);
 987         } else {
 988                 i = c - h->cmd_pool;
 989                 clear_bit(i & (BITS_PER_LONG - 1),
 990                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 991                 h->nr_frees++;
 992         }
 993 }
 994
 995 static inline ctlr_info_t *get_host(struct gendisk *disk)
 996 {
 997         return disk->queue->queuedata;
 998 }
 999
1000 static inline drive_info_struct *get_drv(struct gendisk *disk)
1001 {
1002         return disk->private_data;
1003 }
1004
1005 /*
1006  * Open.  Make sure the device is really there.
1007  */
1008 static int cciss_open(struct block_device *bdev, fmode_t mode)
1009 {
1010         ctlr_info_t *host = get_host(bdev->bd_disk);
1011         drive_info_struct *drv = get_drv(bdev->bd_disk);
1012
1013 #ifdef CCISS_DEBUG
1014         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
1015 #endif                          /* CCISS_DEBUG */
1016
1017         if (drv->busy_configuring)
1018                 return -EBUSY;
1019         /*
1020          * Root is allowed to open raw volume zero even if it's not configured
1021          * so array config can still work. Root is also allowed to open any
1022          * volume that has a LUN ID, so it can issue IOCTL to reread the
1023          * disk information.  I don't think I really like this
1024          * but I'm already using way to many device nodes to claim another one
1025          * for "raw controller".
1026          */
1027         if (drv->heads == 0) {
1028                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1029                         /* if not node 0 make sure it is a partition = 0 */
1030                         if (MINOR(bdev->bd_dev) & 0x0f) {
1031                                 return -ENXIO;
1032                                 /* if it is, make sure we have a LUN ID */
1033                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1034                                 sizeof(drv->LunID))) {
1035                                 return -ENXIO;
1036                         }
1037                 }
1038                 if (!capable(CAP_SYS_ADMIN))
1039                         return -EPERM;
1040         }
1041         drv->usage_count++;
1042         host->usage_count++;
1043         return 0;
1044 }
1045
1046 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1047 {
1048         int ret;
1049
1050         lock_kernel();
1051         ret = cciss_open(bdev, mode);
1052         unlock_kernel();
1053
1054         return ret;
1055 }
1056
1057 /*
1058  * Close.  Sync first.
1059  */
1060 static int cciss_release(struct gendisk *disk, fmode_t mode)
1061 {
1062         ctlr_info_t *host;
1063         drive_info_struct *drv;
1064
1065         lock_kernel();
1066         host = get_host(disk);
1067         drv = get_drv(disk);
1068
1069 #ifdef CCISS_DEBUG
1070         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
1071 #endif                          /* CCISS_DEBUG */
1072
1073         drv->usage_count--;
1074         host->usage_count--;
1075         unlock_kernel();
1076         return 0;
1077 }
1078
1079 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1080                     unsigned cmd, unsigned long arg)
1081 {
1082         int ret;
1083         lock_kernel();
1084         ret = cciss_ioctl(bdev, mode, cmd, arg);
1085         unlock_kernel();
1086         return ret;
1087 }
1088
1089 #ifdef CONFIG_COMPAT
1090
1091 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1092                                   unsigned cmd, unsigned long arg);
1093 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1094                                       unsigned cmd, unsigned long arg);
1095
1096 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1097                               unsigned cmd, unsigned long arg)
1098 {
1099         switch (cmd) {
1100         case CCISS_GETPCIINFO:
1101         case CCISS_GETINTINFO:
1102         case CCISS_SETINTINFO:
1103         case CCISS_GETNODENAME:
1104         case CCISS_SETNODENAME:
1105         case CCISS_GETHEARTBEAT:
1106         case CCISS_GETBUSTYPES:
1107         case CCISS_GETFIRMVER:
1108         case CCISS_GETDRIVVER:
1109         case CCISS_REVALIDVOLS:
1110         case CCISS_DEREGDISK:
1111         case CCISS_REGNEWDISK:
1112         case CCISS_REGNEWD:
1113         case CCISS_RESCANDISK:
1114         case CCISS_GETLUNINFO:
1115                 return do_ioctl(bdev, mode, cmd, arg);
1116
1117         case CCISS_PASSTHRU32:
1118                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1119         case CCISS_BIG_PASSTHRU32:
1120                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1121
1122         default:
1123                 return -ENOIOCTLCMD;
1124         }
1125 }
1126
1127 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1128                                   unsigned cmd, unsigned long arg)
1129 {
1130         IOCTL32_Command_struct __user *arg32 =
1131             (IOCTL32_Command_struct __user *) arg;
1132         IOCTL_Command_struct arg64;
1133         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1134         int err;
1135         u32 cp;
1136
1137         err = 0;
1138         err |=
1139             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1140                            sizeof(arg64.LUN_info));
1141         err |=
1142             copy_from_user(&arg64.Request, &arg32->Request,
1143                            sizeof(arg64.Request));
1144         err |=
1145             copy_from_user(&arg64.error_info, &arg32->error_info,
1146                            sizeof(arg64.error_info));
1147         err |= get_user(arg64.buf_size, &arg32->buf_size);
1148         err |= get_user(cp, &arg32->buf);
1149         arg64.buf = compat_ptr(cp);
1150         err |= copy_to_user(p, &arg64, sizeof(arg64));
1151
1152         if (err)
1153                 return -EFAULT;
1154
1155         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1156         if (err)
1157                 return err;
1158         err |=
1159             copy_in_user(&arg32->error_info, &p->error_info,
1160                          sizeof(arg32->error_info));
1161         if (err)
1162                 return -EFAULT;
1163         return err;
1164 }
1165
1166 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1167                                       unsigned cmd, unsigned long arg)
1168 {
1169         BIG_IOCTL32_Command_struct __user *arg32 =
1170             (BIG_IOCTL32_Command_struct __user *) arg;
1171         BIG_IOCTL_Command_struct arg64;
1172         BIG_IOCTL_Command_struct __user *p =
1173             compat_alloc_user_space(sizeof(arg64));
1174         int err;
1175         u32 cp;
1176
1177         err = 0;
1178         err |=
1179             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1180                            sizeof(arg64.LUN_info));
1181         err |=
1182             copy_from_user(&arg64.Request, &arg32->Request,
1183                            sizeof(arg64.Request));
1184         err |=
1185             copy_from_user(&arg64.error_info, &arg32->error_info,
1186                            sizeof(arg64.error_info));
1187         err |= get_user(arg64.buf_size, &arg32->buf_size);
1188         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1189         err |= get_user(cp, &arg32->buf);
1190         arg64.buf = compat_ptr(cp);
1191         err |= copy_to_user(p, &arg64, sizeof(arg64));
1192
1193         if (err)
1194                 return -EFAULT;
1195
1196         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1197         if (err)
1198                 return err;
1199         err |=
1200             copy_in_user(&arg32->error_info, &p->error_info,
1201                          sizeof(arg32->error_info));
1202         if (err)
1203                 return -EFAULT;
1204         return err;
1205 }
1206 #endif
1207
1208 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1209 {
1210         drive_info_struct *drv = get_drv(bdev->bd_disk);
1211
1212         if (!drv->cylinders)
1213                 return -ENXIO;
1214
1215         geo->heads = drv->heads;
1216         geo->sectors = drv->sectors;
1217         geo->cylinders = drv->cylinders;
1218         return 0;
1219 }
1220
1221 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1222 {
1223         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1224                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1225                 (void)check_for_unit_attention(host, c);
1226 }
1227 /*
1228  * ioctl
1229  */
1230 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1231                        unsigned int cmd, unsigned long arg)
1232 {
1233         struct gendisk *disk = bdev->bd_disk;
1234         ctlr_info_t *host = get_host(disk);
1235         drive_info_struct *drv = get_drv(disk);
1236         int ctlr = host->ctlr;
1237         void __user *argp = (void __user *)arg;
1238
1239 #ifdef CCISS_DEBUG
1240         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1241 #endif                          /* CCISS_DEBUG */
1242
1243         switch (cmd) {
1244         case CCISS_GETPCIINFO:
1245                 {
1246                         cciss_pci_info_struct pciinfo;
1247
1248                         if (!arg)
1249                                 return -EINVAL;
1250                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1251                         pciinfo.bus = host->pdev->bus->number;
1252                         pciinfo.dev_fn = host->pdev->devfn;
1253                         pciinfo.board_id = host->board_id;
1254                         if (copy_to_user
1255                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1256                                 return -EFAULT;
1257                         return 0;
1258                 }
1259         case CCISS_GETINTINFO:
1260                 {
1261                         cciss_coalint_struct intinfo;
1262                         if (!arg)
1263                                 return -EINVAL;
1264                         intinfo.delay =
1265                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1266                         intinfo.count =
1267                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1268                         if (copy_to_user
1269                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1270                                 return -EFAULT;
1271                         return 0;
1272                 }
1273         case CCISS_SETINTINFO:
1274                 {
1275                         cciss_coalint_struct intinfo;
1276                         unsigned long flags;
1277                         int i;
1278
1279                         if (!arg)
1280                                 return -EINVAL;
1281                         if (!capable(CAP_SYS_ADMIN))
1282                                 return -EPERM;
1283                         if (copy_from_user
1284                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1285                                 return -EFAULT;
1286                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1287                         {
1288 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1289                                 return -EINVAL;
1290                         }
1291                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1292                         /* Update the field, and then ring the doorbell */
1293                         writel(intinfo.delay,
1294                                &(host->cfgtable->HostWrite.CoalIntDelay));
1295                         writel(intinfo.count,
1296                                &(host->cfgtable->HostWrite.CoalIntCount));
1297                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1298
1299                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1300                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1301                                       & CFGTBL_ChangeReq))
1302                                         break;
1303                                 /* delay and try again */
1304                                 udelay(1000);
1305                         }
1306                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1307                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1308                                 return -EAGAIN;
1309                         return 0;
1310                 }
1311         case CCISS_GETNODENAME:
1312                 {
1313                         NodeName_type NodeName;
1314                         int i;
1315
1316                         if (!arg)
1317                                 return -EINVAL;
1318                         for (i = 0; i < 16; i++)
1319                                 NodeName[i] =
1320                                     readb(&host->cfgtable->ServerName[i]);
1321                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1322                                 return -EFAULT;
1323                         return 0;
1324                 }
1325         case CCISS_SETNODENAME:
1326                 {
1327                         NodeName_type NodeName;
1328                         unsigned long flags;
1329                         int i;
1330
1331                         if (!arg)
1332                                 return -EINVAL;
1333                         if (!capable(CAP_SYS_ADMIN))
1334                                 return -EPERM;
1335
1336                         if (copy_from_user
1337                             (NodeName, argp, sizeof(NodeName_type)))
1338                                 return -EFAULT;
1339
1340                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1341
1342                         /* Update the field, and then ring the doorbell */
1343                         for (i = 0; i < 16; i++)
1344                                 writeb(NodeName[i],
1345                                        &host->cfgtable->ServerName[i]);
1346
1347                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1348
1349                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1350                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1351                                       & CFGTBL_ChangeReq))
1352                                         break;
1353                                 /* delay and try again */
1354                                 udelay(1000);
1355                         }
1356                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1357                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1358                                 return -EAGAIN;
1359                         return 0;
1360                 }
1361
1362         case CCISS_GETHEARTBEAT:
1363                 {
1364                         Heartbeat_type heartbeat;
1365
1366                         if (!arg)
1367                                 return -EINVAL;
1368                         heartbeat = readl(&host->cfgtable->HeartBeat);
1369                         if (copy_to_user
1370                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1371                                 return -EFAULT;
1372                         return 0;
1373                 }
1374         case CCISS_GETBUSTYPES:
1375                 {
1376                         BusTypes_type BusTypes;
1377
1378                         if (!arg)
1379                                 return -EINVAL;
1380                         BusTypes = readl(&host->cfgtable->BusTypes);
1381                         if (copy_to_user
1382                             (argp, &BusTypes, sizeof(BusTypes_type)))
1383                                 return -EFAULT;
1384                         return 0;
1385                 }
1386         case CCISS_GETFIRMVER:
1387                 {
1388                         FirmwareVer_type firmware;
1389
1390                         if (!arg)
1391                                 return -EINVAL;
1392                         memcpy(firmware, host->firm_ver, 4);
1393
1394                         if (copy_to_user
1395                             (argp, firmware, sizeof(FirmwareVer_type)))
1396                                 return -EFAULT;
1397                         return 0;
1398                 }
1399         case CCISS_GETDRIVVER:
1400                 {
1401                         DriverVer_type DriverVer = DRIVER_VERSION;
1402
1403                         if (!arg)
1404                                 return -EINVAL;
1405
1406                         if (copy_to_user
1407                             (argp, &DriverVer, sizeof(DriverVer_type)))
1408                                 return -EFAULT;
1409                         return 0;
1410                 }
1411
1412         case CCISS_DEREGDISK:
1413         case CCISS_REGNEWD:
1414         case CCISS_REVALIDVOLS:
1415                 return rebuild_lun_table(host, 0, 1);
1416
1417         case CCISS_GETLUNINFO:{
1418                         LogvolInfo_struct luninfo;
1419
1420                         memcpy(&luninfo.LunID, drv->LunID,
1421                                 sizeof(luninfo.LunID));
1422                         luninfo.num_opens = drv->usage_count;
1423                         luninfo.num_parts = 0;
1424                         if (copy_to_user(argp, &luninfo,
1425                                          sizeof(LogvolInfo_struct)))
1426                                 return -EFAULT;
1427                         return 0;
1428                 }
1429         case CCISS_PASSTHRU:
1430                 {
1431                         IOCTL_Command_struct iocommand;
1432                         CommandList_struct *c;
1433                         char *buff = NULL;
1434                         u64bit temp64;
1435                         DECLARE_COMPLETION_ONSTACK(wait);
1436
1437                         if (!arg)
1438                                 return -EINVAL;
1439
1440                         if (!capable(CAP_SYS_RAWIO))
1441                                 return -EPERM;
1442
1443                         if (copy_from_user
1444                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1445                                 return -EFAULT;
1446                         if ((iocommand.buf_size < 1) &&
1447                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1448                                 return -EINVAL;
1449                         }
1450 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1451                         /* Check kmalloc limits */
1452                         if (iocommand.buf_size > 128000)
1453                                 return -EINVAL;
1454 #endif
1455                         if (iocommand.buf_size > 0) {
1456                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1457                                 if (buff == NULL)
1458                                         return -EFAULT;
1459                         }
1460                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1461                                 /* Copy the data into the buffer we created */
1462                                 if (copy_from_user
1463                                     (buff, iocommand.buf, iocommand.buf_size)) {
1464                                         kfree(buff);
1465                                         return -EFAULT;
1466                                 }
1467                         } else {
1468                                 memset(buff, 0, iocommand.buf_size);
1469                         }
1470                         if ((c = cmd_alloc(host, 0)) == NULL) {
1471                                 kfree(buff);
1472                                 return -ENOMEM;
1473                         }
1474                         /* Fill in the command type */
1475                         c->cmd_type = CMD_IOCTL_PEND;
1476                         /* Fill in Command Header */
1477                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1478                         if (iocommand.buf_size > 0) /* buffer to fill */
1479                         {
1480                                 c->Header.SGList = 1;
1481                                 c->Header.SGTotal = 1;
1482                         } else /* no buffers to fill */
1483                         {
1484                                 c->Header.SGList = 0;
1485                                 c->Header.SGTotal = 0;
1486                         }
1487                         c->Header.LUN = iocommand.LUN_info;
1488                         /* use the kernel address the cmd block for tag */
1489                         c->Header.Tag.lower = c->busaddr;
1490
1491                         /* Fill in Request block */
1492                         c->Request = iocommand.Request;
1493
1494                         /* Fill in the scatter gather information */
1495                         if (iocommand.buf_size > 0) {
1496                                 temp64.val = pci_map_single(host->pdev, buff,
1497                                         iocommand.buf_size,
1498                                         PCI_DMA_BIDIRECTIONAL);
1499                                 c->SG[0].Addr.lower = temp64.val32.lower;
1500                                 c->SG[0].Addr.upper = temp64.val32.upper;
1501                                 c->SG[0].Len = iocommand.buf_size;
1502                                 c->SG[0].Ext = 0;  /* we are not chaining */
1503                         }
1504                         c->waiting = &wait;
1505
1506                         enqueue_cmd_and_start_io(host, c);
1507                         wait_for_completion(&wait);
1508
1509                         /* unlock the buffers from DMA */
1510                         temp64.val32.lower = c->SG[0].Addr.lower;
1511                         temp64.val32.upper = c->SG[0].Addr.upper;
1512                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1513                                          iocommand.buf_size,
1514                                          PCI_DMA_BIDIRECTIONAL);
1515
1516                         check_ioctl_unit_attention(host, c);
1517
1518                         /* Copy the error information out */
1519                         iocommand.error_info = *(c->err_info);
1520                         if (copy_to_user
1521                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1522                                 kfree(buff);
1523                                 cmd_free(host, c, 0);
1524                                 return -EFAULT;
1525                         }
1526
1527                         if (iocommand.Request.Type.Direction == XFER_READ) {
1528                                 /* Copy the data out of the buffer we created */
1529                                 if (copy_to_user
1530                                     (iocommand.buf, buff, iocommand.buf_size)) {
1531                                         kfree(buff);
1532                                         cmd_free(host, c, 0);
1533                                         return -EFAULT;
1534                                 }
1535                         }
1536                         kfree(buff);
1537                         cmd_free(host, c, 0);
1538                         return 0;
1539                 }
1540         case CCISS_BIG_PASSTHRU:{
1541                         BIG_IOCTL_Command_struct *ioc;
1542                         CommandList_struct *c;
1543                         unsigned char **buff = NULL;
1544                         int *buff_size = NULL;
1545                         u64bit temp64;
1546                         BYTE sg_used = 0;
1547                         int status = 0;
1548                         int i;
1549                         DECLARE_COMPLETION_ONSTACK(wait);
1550                         __u32 left;
1551                         __u32 sz;
1552                         BYTE __user *data_ptr;
1553
1554                         if (!arg)
1555                                 return -EINVAL;
1556                         if (!capable(CAP_SYS_RAWIO))
1557                                 return -EPERM;
1558                         ioc = (BIG_IOCTL_Command_struct *)
1559                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1560                         if (!ioc) {
1561                                 status = -ENOMEM;
1562                                 goto cleanup1;
1563                         }
1564                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1565                                 status = -EFAULT;
1566                                 goto cleanup1;
1567                         }
1568                         if ((ioc->buf_size < 1) &&
1569                             (ioc->Request.Type.Direction != XFER_NONE)) {
1570                                 status = -EINVAL;
1571                                 goto cleanup1;
1572                         }
1573                         /* Check kmalloc limits  using all SGs */
1574                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1575                                 status = -EINVAL;
1576                                 goto cleanup1;
1577                         }
1578                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1579                                 status = -EINVAL;
1580                                 goto cleanup1;
1581                         }
1582                         buff =
1583                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1584                         if (!buff) {
1585                                 status = -ENOMEM;
1586                                 goto cleanup1;
1587                         }
1588                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1589                                                    GFP_KERNEL);
1590                         if (!buff_size) {
1591                                 status = -ENOMEM;
1592                                 goto cleanup1;
1593                         }
1594                         left = ioc->buf_size;
1595                         data_ptr = ioc->buf;
1596                         while (left) {
1597                                 sz = (left >
1598                                       ioc->malloc_size) ? ioc->
1599                                     malloc_size : left;
1600                                 buff_size[sg_used] = sz;
1601                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1602                                 if (buff[sg_used] == NULL) {
1603                                         status = -ENOMEM;
1604                                         goto cleanup1;
1605                                 }
1606                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1607                                         if (copy_from_user
1608                                             (buff[sg_used], data_ptr, sz)) {
1609                                                 status = -EFAULT;
1610                                                 goto cleanup1;
1611                                         }
1612                                 } else {
1613                                         memset(buff[sg_used], 0, sz);
1614                                 }
1615                                 left -= sz;
1616                                 data_ptr += sz;
1617                                 sg_used++;
1618                         }
1619                         if ((c = cmd_alloc(host, 0)) == NULL) {
1620                                 status = -ENOMEM;
1621                                 goto cleanup1;
1622                         }
1623                         c->cmd_type = CMD_IOCTL_PEND;
1624                         c->Header.ReplyQueue = 0;
1625
1626                         if (ioc->buf_size > 0) {
1627                                 c->Header.SGList = sg_used;
1628                                 c->Header.SGTotal = sg_used;
1629                         } else {
1630                                 c->Header.SGList = 0;
1631                                 c->Header.SGTotal = 0;
1632                         }
1633                         c->Header.LUN = ioc->LUN_info;
1634                         c->Header.Tag.lower = c->busaddr;
1635
1636                         c->Request = ioc->Request;
1637                         if (ioc->buf_size > 0) {
1638                                 for (i = 0; i < sg_used; i++) {
1639                                         temp64.val =
1640                                             pci_map_single(host->pdev, buff[i],
1641                                                     buff_size[i],
1642                                                     PCI_DMA_BIDIRECTIONAL);
1643                                         c->SG[i].Addr.lower =
1644                                             temp64.val32.lower;
1645                                         c->SG[i].Addr.upper =
1646                                             temp64.val32.upper;
1647                                         c->SG[i].Len = buff_size[i];
1648                                         c->SG[i].Ext = 0;       /* we are not chaining */
1649                                 }
1650                         }
1651                         c->waiting = &wait;
1652                         enqueue_cmd_and_start_io(host, c);
1653                         wait_for_completion(&wait);
1654                         /* unlock the buffers from DMA */
1655                         for (i = 0; i < sg_used; i++) {
1656                                 temp64.val32.lower = c->SG[i].Addr.lower;
1657                                 temp64.val32.upper = c->SG[i].Addr.upper;
1658                                 pci_unmap_single(host->pdev,
1659                                         (dma_addr_t) temp64.val, buff_size[i],
1660                                         PCI_DMA_BIDIRECTIONAL);
1661                         }
1662                         check_ioctl_unit_attention(host, c);
1663                         /* Copy the error information out */
1664                         ioc->error_info = *(c->err_info);
1665                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1666                                 cmd_free(host, c, 0);
1667                                 status = -EFAULT;
1668                                 goto cleanup1;
1669                         }
1670                         if (ioc->Request.Type.Direction == XFER_READ) {
1671                                 /* Copy the data out of the buffer we created */
1672                                 BYTE __user *ptr = ioc->buf;
1673                                 for (i = 0; i < sg_used; i++) {
1674                                         if (copy_to_user
1675                                             (ptr, buff[i], buff_size[i])) {
1676                                                 cmd_free(host, c, 0);
1677                                                 status = -EFAULT;
1678                                                 goto cleanup1;
1679                                         }
1680                                         ptr += buff_size[i];
1681                                 }
1682                         }
1683                         cmd_free(host, c, 0);
1684                         status = 0;
1685                       cleanup1:
1686                         if (buff) {
1687                                 for (i = 0; i < sg_used; i++)
1688                                         kfree(buff[i]);
1689                                 kfree(buff);
1690                         }
1691                         kfree(buff_size);
1692                         kfree(ioc);
1693                         return status;
1694                 }
1695
1696         /* scsi_cmd_ioctl handles these, below, though some are not */
1697         /* very meaningful for cciss.  SG_IO is the main one people want. */
1698
1699         case SG_GET_VERSION_NUM:
1700         case SG_SET_TIMEOUT:
1701         case SG_GET_TIMEOUT:
1702         case SG_GET_RESERVED_SIZE:
1703         case SG_SET_RESERVED_SIZE:
1704         case SG_EMULATED_HOST:
1705         case SG_IO:
1706         case SCSI_IOCTL_SEND_COMMAND:
1707                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1708
1709         /* scsi_cmd_ioctl would normally handle these, below, but */
1710         /* they aren't a good fit for cciss, as CD-ROMs are */
1711         /* not supported, and we don't have any bus/target/lun */
1712         /* which we present to the kernel. */
1713
1714         case CDROM_SEND_PACKET:
1715         case CDROMCLOSETRAY:
1716         case CDROMEJECT:
1717         case SCSI_IOCTL_GET_IDLUN:
1718         case SCSI_IOCTL_GET_BUS_NUMBER:
1719         default:
1720                 return -ENOTTY;
1721         }
1722 }
1723
1724 static void cciss_check_queues(ctlr_info_t *h)
1725 {
1726         int start_queue = h->next_to_run;
1727         int i;
1728
1729         /* check to see if we have maxed out the number of commands that can
1730          * be placed on the queue.  If so then exit.  We do this check here
1731          * in case the interrupt we serviced was from an ioctl and did not
1732          * free any new commands.
1733          */
1734         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1735                 return;
1736
1737         /* We have room on the queue for more commands.  Now we need to queue
1738          * them up.  We will also keep track of the next queue to run so
1739          * that every queue gets a chance to be started first.
1740          */
1741         for (i = 0; i < h->highest_lun + 1; i++) {
1742                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1743                 /* make sure the disk has been added and the drive is real
1744                  * because this can be called from the middle of init_one.
1745                  */
1746                 if (!h->drv[curr_queue])
1747                         continue;
1748                 if (!(h->drv[curr_queue]->queue) ||
1749                         !(h->drv[curr_queue]->heads))
1750                         continue;
1751                 blk_start_queue(h->gendisk[curr_queue]->queue);
1752
1753                 /* check to see if we have maxed out the number of commands
1754                  * that can be placed on the queue.
1755                  */
1756                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1757                         if (curr_queue == start_queue) {
1758                                 h->next_to_run =
1759                                     (start_queue + 1) % (h->highest_lun + 1);
1760                                 break;
1761                         } else {
1762                                 h->next_to_run = curr_queue;
1763                                 break;
1764                         }
1765                 }
1766         }
1767 }
1768
1769 static void cciss_softirq_done(struct request *rq)
1770 {
1771         CommandList_struct *cmd = rq->completion_data;
1772         ctlr_info_t *h = hba[cmd->ctlr];
1773         SGDescriptor_struct *curr_sg = cmd->SG;
1774         u64bit temp64;
1775         unsigned long flags;
1776         int i, ddir;
1777         int sg_index = 0;
1778
1779         if (cmd->Request.Type.Direction == XFER_READ)
1780                 ddir = PCI_DMA_FROMDEVICE;
1781         else
1782                 ddir = PCI_DMA_TODEVICE;
1783
1784         /* command did not need to be retried */
1785         /* unmap the DMA mapping for all the scatter gather elements */
1786         for (i = 0; i < cmd->Header.SGList; i++) {
1787                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1788                         cciss_unmap_sg_chain_block(h, cmd);
1789                         /* Point to the next block */
1790                         curr_sg = h->cmd_sg_list[cmd->cmdindex];
1791                         sg_index = 0;
1792                 }
1793                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1794                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1795                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1796                                 ddir);
1797                 ++sg_index;
1798         }
1799
1800 #ifdef CCISS_DEBUG
1801         printk("Done with %p\n", rq);
1802 #endif                          /* CCISS_DEBUG */
1803
1804         /* set the residual count for pc requests */
1805         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1806                 rq->resid_len = cmd->err_info->ResidualCnt;
1807
1808         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1809
1810         spin_lock_irqsave(&h->lock, flags);
1811         cmd_free(h, cmd, 1);
1812         cciss_check_queues(h);
1813         spin_unlock_irqrestore(&h->lock, flags);
1814 }
1815
1816 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1817         unsigned char scsi3addr[], uint32_t log_unit)
1818 {
1819         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1820                 sizeof(h->drv[log_unit]->LunID));
1821 }
1822
1823 /* This function gets the SCSI vendor, model, and revision of a logical drive
1824  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1825  * they cannot be read.
1826  */
1827 static void cciss_get_device_descr(int ctlr, int logvol,
1828                                    char *vendor, char *model, char *rev)
1829 {
1830         int rc;
1831         InquiryData_struct *inq_buf;
1832         unsigned char scsi3addr[8];
1833
1834         *vendor = '\0';
1835         *model = '\0';
1836         *rev = '\0';
1837
1838         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1839         if (!inq_buf)
1840                 return;
1841
1842         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1843         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf, sizeof(*inq_buf), 0,
1844                         scsi3addr, TYPE_CMD);
1845         if (rc == IO_OK) {
1846                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1847                 vendor[VENDOR_LEN] = '\0';
1848                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1849                 model[MODEL_LEN] = '\0';
1850                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1851                 rev[REV_LEN] = '\0';
1852         }
1853
1854         kfree(inq_buf);
1855         return;
1856 }
1857
1858 /* This function gets the serial number of a logical drive via
1859  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1860  * number cannot be had, for whatever reason, 16 bytes of 0xff
1861  * are returned instead.
1862  */
1863 static void cciss_get_serial_no(int ctlr, int logvol,
1864                                 unsigned char *serial_no, int buflen)
1865 {
1866 #define PAGE_83_INQ_BYTES 64
1867         int rc;
1868         unsigned char *buf;
1869         unsigned char scsi3addr[8];
1870
1871         if (buflen > 16)
1872                 buflen = 16;
1873         memset(serial_no, 0xff, buflen);
1874         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1875         if (!buf)
1876                 return;
1877         memset(serial_no, 0, buflen);
1878         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1879         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1880                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1881         if (rc == IO_OK)
1882                 memcpy(serial_no, &buf[8], buflen);
1883         kfree(buf);
1884         return;
1885 }
1886
1887 /*
1888  * cciss_add_disk sets up the block device queue for a logical drive
1889  */
1890 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1891                                 int drv_index)
1892 {
1893         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1894         if (!disk->queue)
1895                 goto init_queue_failure;
1896         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1897         disk->major = h->major;
1898         disk->first_minor = drv_index << NWD_SHIFT;
1899         disk->fops = &cciss_fops;
1900         if (cciss_create_ld_sysfs_entry(h, drv_index))
1901                 goto cleanup_queue;
1902         disk->private_data = h->drv[drv_index];
1903         disk->driverfs_dev = &h->drv[drv_index]->dev;
1904
1905         /* Set up queue information */
1906         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1907
1908         /* This is a hardware imposed limit. */
1909         blk_queue_max_segments(disk->queue, h->maxsgentries);
1910
1911         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1912
1913         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1914
1915         disk->queue->queuedata = h;
1916
1917         blk_queue_logical_block_size(disk->queue,
1918                                      h->drv[drv_index]->block_size);
1919
1920         /* Make sure all queue data is written out before */
1921         /* setting h->drv[drv_index]->queue, as setting this */
1922         /* allows the interrupt handler to start the queue */
1923         wmb();
1924         h->drv[drv_index]->queue = disk->queue;
1925         add_disk(disk);
1926         return 0;
1927
1928 cleanup_queue:
1929         blk_cleanup_queue(disk->queue);
1930         disk->queue = NULL;
1931 init_queue_failure:
1932         return -1;
1933 }
1934
1935 /* This function will check the usage_count of the drive to be updated/added.
1936  * If the usage_count is zero and it is a heretofore unknown drive, or,
1937  * the drive's capacity, geometry, or serial number has changed,
1938  * then the drive information will be updated and the disk will be
1939  * re-registered with the kernel.  If these conditions don't hold,
1940  * then it will be left alone for the next reboot.  The exception to this
1941  * is disk 0 which will always be left registered with the kernel since it
1942  * is also the controller node.  Any changes to disk 0 will show up on
1943  * the next reboot.
1944  */
1945 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1946         int via_ioctl)
1947 {
1948         ctlr_info_t *h = hba[ctlr];
1949         struct gendisk *disk;
1950         InquiryData_struct *inq_buff = NULL;
1951         unsigned int block_size;
1952         sector_t total_size;
1953         unsigned long flags = 0;
1954         int ret = 0;
1955         drive_info_struct *drvinfo;
1956
1957         /* Get information about the disk and modify the driver structure */
1958         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1959         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1960         if (inq_buff == NULL || drvinfo == NULL)
1961                 goto mem_msg;
1962
1963         /* testing to see if 16-byte CDBs are already being used */
1964         if (h->cciss_read == CCISS_READ_16) {
1965                 cciss_read_capacity_16(h->ctlr, drv_index,
1966                         &total_size, &block_size);
1967
1968         } else {
1969                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1970                 /* if read_capacity returns all F's this volume is >2TB */
1971                 /* in size so we switch to 16-byte CDB's for all */
1972                 /* read/write ops */
1973                 if (total_size == 0xFFFFFFFFULL) {
1974                         cciss_read_capacity_16(ctlr, drv_index,
1975                         &total_size, &block_size);
1976                         h->cciss_read = CCISS_READ_16;
1977                         h->cciss_write = CCISS_WRITE_16;
1978                 } else {
1979                         h->cciss_read = CCISS_READ_10;
1980                         h->cciss_write = CCISS_WRITE_10;
1981                 }
1982         }
1983
1984         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1985                                inq_buff, drvinfo);
1986         drvinfo->block_size = block_size;
1987         drvinfo->nr_blocks = total_size + 1;
1988
1989         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1990                                 drvinfo->model, drvinfo->rev);
1991         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1992                         sizeof(drvinfo->serial_no));
1993         /* Save the lunid in case we deregister the disk, below. */
1994         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1995                 sizeof(drvinfo->LunID));
1996
1997         /* Is it the same disk we already know, and nothing's changed? */
1998         if (h->drv[drv_index]->raid_level != -1 &&
1999                 ((memcmp(drvinfo->serial_no,
2000                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2001                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2002                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2003                 drvinfo->heads == h->drv[drv_index]->heads &&
2004                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2005                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2006                         /* The disk is unchanged, nothing to update */
2007                         goto freeret;
2008
2009         /* If we get here it's not the same disk, or something's changed,
2010          * so we need to * deregister it, and re-register it, if it's not
2011          * in use.
2012          * If the disk already exists then deregister it before proceeding
2013          * (unless it's the first disk (for the controller node).
2014          */
2015         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2016                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
2017                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2018                 h->drv[drv_index]->busy_configuring = 1;
2019                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2020
2021                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2022                  * which keeps the interrupt handler from starting
2023                  * the queue.
2024                  */
2025                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2026         }
2027
2028         /* If the disk is in use return */
2029         if (ret)
2030                 goto freeret;
2031
2032         /* Save the new information from cciss_geometry_inquiry
2033          * and serial number inquiry.  If the disk was deregistered
2034          * above, then h->drv[drv_index] will be NULL.
2035          */
2036         if (h->drv[drv_index] == NULL) {
2037                 drvinfo->device_initialized = 0;
2038                 h->drv[drv_index] = drvinfo;
2039                 drvinfo = NULL; /* so it won't be freed below. */
2040         } else {
2041                 /* special case for cxd0 */
2042                 h->drv[drv_index]->block_size = drvinfo->block_size;
2043                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2044                 h->drv[drv_index]->heads = drvinfo->heads;
2045                 h->drv[drv_index]->sectors = drvinfo->sectors;
2046                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2047                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2048                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2049                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2050                         VENDOR_LEN + 1);
2051                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2052                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2053         }
2054
2055         ++h->num_luns;
2056         disk = h->gendisk[drv_index];
2057         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2058
2059         /* If it's not disk 0 (drv_index != 0)
2060          * or if it was disk 0, but there was previously
2061          * no actual corresponding configured logical drive
2062          * (raid_leve == -1) then we want to update the
2063          * logical drive's information.
2064          */
2065         if (drv_index || first_time) {
2066                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2067                         cciss_free_gendisk(h, drv_index);
2068                         cciss_free_drive_info(h, drv_index);
2069                         printk(KERN_WARNING "cciss:%d could not update "
2070                                 "disk %d\n", h->ctlr, drv_index);
2071                         --h->num_luns;
2072                 }
2073         }
2074
2075 freeret:
2076         kfree(inq_buff);
2077         kfree(drvinfo);
2078         return;
2079 mem_msg:
2080         printk(KERN_ERR "cciss: out of memory\n");
2081         goto freeret;
2082 }
2083
2084 /* This function will find the first index of the controllers drive array
2085  * that has a null drv pointer and allocate the drive info struct and
2086  * will return that index   This is where new drives will be added.
2087  * If the index to be returned is greater than the highest_lun index for
2088  * the controller then highest_lun is set * to this new index.
2089  * If there are no available indexes or if tha allocation fails, then -1
2090  * is returned.  * "controller_node" is used to know if this is a real
2091  * logical drive, or just the controller node, which determines if this
2092  * counts towards highest_lun.
2093  */
2094 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2095 {
2096         int i;
2097         drive_info_struct *drv;
2098
2099         /* Search for an empty slot for our drive info */
2100         for (i = 0; i < CISS_MAX_LUN; i++) {
2101
2102                 /* if not cxd0 case, and it's occupied, skip it. */
2103                 if (h->drv[i] && i != 0)
2104                         continue;
2105                 /*
2106                  * If it's cxd0 case, and drv is alloc'ed already, and a
2107                  * disk is configured there, skip it.
2108                  */
2109                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2110                         continue;
2111
2112                 /*
2113                  * We've found an empty slot.  Update highest_lun
2114                  * provided this isn't just the fake cxd0 controller node.
2115                  */
2116                 if (i > h->highest_lun && !controller_node)
2117                         h->highest_lun = i;
2118
2119                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2120                 if (i == 0 && h->drv[i] != NULL)
2121                         return i;
2122
2123                 /*
2124                  * Found an empty slot, not already alloc'ed.  Allocate it.
2125                  * Mark it with raid_level == -1, so we know it's new later on.
2126                  */
2127                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2128                 if (!drv)
2129                         return -1;
2130                 drv->raid_level = -1; /* so we know it's new */
2131                 h->drv[i] = drv;
2132                 return i;
2133         }
2134         return -1;
2135 }
2136
2137 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2138 {
2139         kfree(h->drv[drv_index]);
2140         h->drv[drv_index] = NULL;
2141 }
2142
2143 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2144 {
2145         put_disk(h->gendisk[drv_index]);
2146         h->gendisk[drv_index] = NULL;
2147 }
2148
2149 /* cciss_add_gendisk finds a free hba[]->drv structure
2150  * and allocates a gendisk if needed, and sets the lunid
2151  * in the drvinfo structure.   It returns the index into
2152  * the ->drv[] array, or -1 if none are free.
2153  * is_controller_node indicates whether highest_lun should
2154  * count this disk, or if it's only being added to provide
2155  * a means to talk to the controller in case no logical
2156  * drives have yet been configured.
2157  */
2158 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2159         int controller_node)
2160 {
2161         int drv_index;
2162
2163         drv_index = cciss_alloc_drive_info(h, controller_node);
2164         if (drv_index == -1)
2165                 return -1;
2166
2167         /*Check if the gendisk needs to be allocated */
2168         if (!h->gendisk[drv_index]) {
2169                 h->gendisk[drv_index] =
2170                         alloc_disk(1 << NWD_SHIFT);
2171                 if (!h->gendisk[drv_index]) {
2172                         printk(KERN_ERR "cciss%d: could not "
2173                                 "allocate a new disk %d\n",
2174                                 h->ctlr, drv_index);
2175                         goto err_free_drive_info;
2176                 }
2177         }
2178         memcpy(h->drv[drv_index]->LunID, lunid,
2179                 sizeof(h->drv[drv_index]->LunID));
2180         if (cciss_create_ld_sysfs_entry(h, drv_index))
2181                 goto err_free_disk;
2182         /* Don't need to mark this busy because nobody */
2183         /* else knows about this disk yet to contend */
2184         /* for access to it. */
2185         h->drv[drv_index]->busy_configuring = 0;
2186         wmb();
2187         return drv_index;
2188
2189 err_free_disk:
2190         cciss_free_gendisk(h, drv_index);
2191 err_free_drive_info:
2192         cciss_free_drive_info(h, drv_index);
2193         return -1;
2194 }
2195
2196 /* This is for the special case of a controller which
2197  * has no logical drives.  In this case, we still need
2198  * to register a disk so the controller can be accessed
2199  * by the Array Config Utility.
2200  */
2201 static void cciss_add_controller_node(ctlr_info_t *h)
2202 {
2203         struct gendisk *disk;
2204         int drv_index;
2205
2206         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2207                 return;
2208
2209         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2210         if (drv_index == -1)
2211                 goto error;
2212         h->drv[drv_index]->block_size = 512;
2213         h->drv[drv_index]->nr_blocks = 0;
2214         h->drv[drv_index]->heads = 0;
2215         h->drv[drv_index]->sectors = 0;
2216         h->drv[drv_index]->cylinders = 0;
2217         h->drv[drv_index]->raid_level = -1;
2218         memset(h->drv[drv_index]->serial_no, 0, 16);
2219         disk = h->gendisk[drv_index];
2220         if (cciss_add_disk(h, disk, drv_index) == 0)
2221                 return;
2222         cciss_free_gendisk(h, drv_index);
2223         cciss_free_drive_info(h, drv_index);
2224 error:
2225         printk(KERN_WARNING "cciss%d: could not "
2226                 "add disk 0.\n", h->ctlr);
2227         return;
2228 }
2229
2230 /* This function will add and remove logical drives from the Logical
2231  * drive array of the controller and maintain persistency of ordering
2232  * so that mount points are preserved until the next reboot.  This allows
2233  * for the removal of logical drives in the middle of the drive array
2234  * without a re-ordering of those drives.
2235  * INPUT
2236  * h            = The controller to perform the operations on
2237  */
2238 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2239         int via_ioctl)
2240 {
2241         int ctlr = h->ctlr;
2242         int num_luns;
2243         ReportLunData_struct *ld_buff = NULL;
2244         int return_code;
2245         int listlength = 0;
2246         int i;
2247         int drv_found;
2248         int drv_index = 0;
2249         unsigned char lunid[8] = CTLR_LUNID;
2250         unsigned long flags;
2251
2252         if (!capable(CAP_SYS_RAWIO))
2253                 return -EPERM;
2254
2255         /* Set busy_configuring flag for this operation */
2256         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2257         if (h->busy_configuring) {
2258                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2259                 return -EBUSY;
2260         }
2261         h->busy_configuring = 1;
2262         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2263
2264         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2265         if (ld_buff == NULL)
2266                 goto mem_msg;
2267
2268         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2269                                       sizeof(ReportLunData_struct),
2270                                       0, CTLR_LUNID, TYPE_CMD);
2271
2272         if (return_code == IO_OK)
2273                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2274         else {  /* reading number of logical volumes failed */
2275                 printk(KERN_WARNING "cciss: report logical volume"
2276                        " command failed\n");
2277                 listlength = 0;
2278                 goto freeret;
2279         }
2280
2281         num_luns = listlength / 8;      /* 8 bytes per entry */
2282         if (num_luns > CISS_MAX_LUN) {
2283                 num_luns = CISS_MAX_LUN;
2284                 printk(KERN_WARNING "cciss: more luns configured"
2285                        " on controller than can be handled by"
2286                        " this driver.\n");
2287         }
2288
2289         if (num_luns == 0)
2290                 cciss_add_controller_node(h);
2291
2292         /* Compare controller drive array to driver's drive array
2293          * to see if any drives are missing on the controller due
2294          * to action of Array Config Utility (user deletes drive)
2295          * and deregister logical drives which have disappeared.
2296          */
2297         for (i = 0; i <= h->highest_lun; i++) {
2298                 int j;
2299                 drv_found = 0;
2300
2301                 /* skip holes in the array from already deleted drives */
2302                 if (h->drv[i] == NULL)
2303                         continue;
2304
2305                 for (j = 0; j < num_luns; j++) {
2306                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2307                         if (memcmp(h->drv[i]->LunID, lunid,
2308                                 sizeof(lunid)) == 0) {
2309                                 drv_found = 1;
2310                                 break;
2311                         }
2312                 }
2313                 if (!drv_found) {
2314                         /* Deregister it from the OS, it's gone. */
2315                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2316                         h->drv[i]->busy_configuring = 1;
2317                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2318                         return_code = deregister_disk(h, i, 1, via_ioctl);
2319                         if (h->drv[i] != NULL)
2320                                 h->drv[i]->busy_configuring = 0;
2321                 }
2322         }
2323
2324         /* Compare controller drive array to driver's drive array.
2325          * Check for updates in the drive information and any new drives
2326          * on the controller due to ACU adding logical drives, or changing
2327          * a logical drive's size, etc.  Reregister any new/changed drives
2328          */
2329         for (i = 0; i < num_luns; i++) {
2330                 int j;
2331
2332                 drv_found = 0;
2333
2334                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2335                 /* Find if the LUN is already in the drive array
2336                  * of the driver.  If so then update its info
2337                  * if not in use.  If it does not exist then find
2338                  * the first free index and add it.
2339                  */
2340                 for (j = 0; j <= h->highest_lun; j++) {
2341                         if (h->drv[j] != NULL &&
2342                                 memcmp(h->drv[j]->LunID, lunid,
2343                                         sizeof(h->drv[j]->LunID)) == 0) {
2344                                 drv_index = j;
2345                                 drv_found = 1;
2346                                 break;
2347                         }
2348                 }
2349
2350                 /* check if the drive was found already in the array */
2351                 if (!drv_found) {
2352                         drv_index = cciss_add_gendisk(h, lunid, 0);
2353                         if (drv_index == -1)
2354                                 goto freeret;
2355                 }
2356                 cciss_update_drive_info(ctlr, drv_index, first_time,
2357                         via_ioctl);
2358         }               /* end for */
2359
2360 freeret:
2361         kfree(ld_buff);
2362         h->busy_configuring = 0;
2363         /* We return -1 here to tell the ACU that we have registered/updated
2364          * all of the drives that we can and to keep it from calling us
2365          * additional times.
2366          */
2367         return -1;
2368 mem_msg:
2369         printk(KERN_ERR "cciss: out of memory\n");
2370         h->busy_configuring = 0;
2371         goto freeret;
2372 }
2373
2374 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2375 {
2376         /* zero out the disk size info */
2377         drive_info->nr_blocks = 0;
2378         drive_info->block_size = 0;
2379         drive_info->heads = 0;
2380         drive_info->sectors = 0;
2381         drive_info->cylinders = 0;
2382         drive_info->raid_level = -1;
2383         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2384         memset(drive_info->model, 0, sizeof(drive_info->model));
2385         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2386         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2387         /*
2388          * don't clear the LUNID though, we need to remember which
2389          * one this one is.
2390          */
2391 }
2392
2393 /* This function will deregister the disk and it's queue from the
2394  * kernel.  It must be called with the controller lock held and the
2395  * drv structures busy_configuring flag set.  It's parameters are:
2396  *
2397  * disk = This is the disk to be deregistered
2398  * drv  = This is the drive_info_struct associated with the disk to be
2399  *        deregistered.  It contains information about the disk used
2400  *        by the driver.
2401  * clear_all = This flag determines whether or not the disk information
2402  *             is going to be completely cleared out and the highest_lun
2403  *             reset.  Sometimes we want to clear out information about
2404  *             the disk in preparation for re-adding it.  In this case
2405  *             the highest_lun should be left unchanged and the LunID
2406  *             should not be cleared.
2407  * via_ioctl
2408  *    This indicates whether we've reached this path via ioctl.
2409  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2410  *    If this path is reached via ioctl(), then the max_usage_count will
2411  *    be 1, as the process calling ioctl() has got to have the device open.
2412  *    If we get here via sysfs, then the max usage count will be zero.
2413 */
2414 static int deregister_disk(ctlr_info_t *h, int drv_index,
2415                            int clear_all, int via_ioctl)
2416 {
2417         int i;
2418         struct gendisk *disk;
2419         drive_info_struct *drv;
2420         int recalculate_highest_lun;
2421
2422         if (!capable(CAP_SYS_RAWIO))
2423                 return -EPERM;
2424
2425         drv = h->drv[drv_index];
2426         disk = h->gendisk[drv_index];
2427
2428         /* make sure logical volume is NOT is use */
2429         if (clear_all || (h->gendisk[0] == disk)) {
2430                 if (drv->usage_count > via_ioctl)
2431                         return -EBUSY;
2432         } else if (drv->usage_count > 0)
2433                 return -EBUSY;
2434
2435         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2436
2437         /* invalidate the devices and deregister the disk.  If it is disk
2438          * zero do not deregister it but just zero out it's values.  This
2439          * allows us to delete disk zero but keep the controller registered.
2440          */
2441         if (h->gendisk[0] != disk) {
2442                 struct request_queue *q = disk->queue;
2443                 if (disk->flags & GENHD_FL_UP) {
2444                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2445                         del_gendisk(disk);
2446                 }
2447                 if (q)
2448                         blk_cleanup_queue(q);
2449                 /* If clear_all is set then we are deleting the logical
2450                  * drive, not just refreshing its info.  For drives
2451                  * other than disk 0 we will call put_disk.  We do not
2452                  * do this for disk 0 as we need it to be able to
2453                  * configure the controller.
2454                  */
2455                 if (clear_all){
2456                         /* This isn't pretty, but we need to find the
2457                          * disk in our array and NULL our the pointer.
2458                          * This is so that we will call alloc_disk if
2459                          * this index is used again later.
2460                          */
2461                         for (i=0; i < CISS_MAX_LUN; i++){
2462                                 if (h->gendisk[i] == disk) {
2463                                         h->gendisk[i] = NULL;
2464                                         break;
2465                                 }
2466                         }
2467                         put_disk(disk);
2468                 }
2469         } else {
2470                 set_capacity(disk, 0);
2471                 cciss_clear_drive_info(drv);
2472         }
2473
2474         --h->num_luns;
2475
2476         /* if it was the last disk, find the new hightest lun */
2477         if (clear_all && recalculate_highest_lun) {
2478                 int newhighest = -1;
2479                 for (i = 0; i <= h->highest_lun; i++) {
2480                         /* if the disk has size > 0, it is available */
2481                         if (h->drv[i] && h->drv[i]->heads)
2482                                 newhighest = i;
2483                 }
2484                 h->highest_lun = newhighest;
2485         }
2486         return 0;
2487 }
2488
2489 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2490                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2491                 int cmd_type)
2492 {
2493         ctlr_info_t *h = hba[ctlr];
2494         u64bit buff_dma_handle;
2495         int status = IO_OK;
2496
2497         c->cmd_type = CMD_IOCTL_PEND;
2498         c->Header.ReplyQueue = 0;
2499         if (buff != NULL) {
2500                 c->Header.SGList = 1;
2501                 c->Header.SGTotal = 1;
2502         } else {
2503                 c->Header.SGList = 0;
2504                 c->Header.SGTotal = 0;
2505         }
2506         c->Header.Tag.lower = c->busaddr;
2507         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2508
2509         c->Request.Type.Type = cmd_type;
2510         if (cmd_type == TYPE_CMD) {
2511                 switch (cmd) {
2512                 case CISS_INQUIRY:
2513                         /* are we trying to read a vital product page */
2514                         if (page_code != 0) {
2515                                 c->Request.CDB[1] = 0x01;
2516                                 c->Request.CDB[2] = page_code;
2517                         }
2518                         c->Request.CDBLen = 6;
2519                         c->Request.Type.Attribute = ATTR_SIMPLE;
2520                         c->Request.Type.Direction = XFER_READ;
2521                         c->Request.Timeout = 0;
2522                         c->Request.CDB[0] = CISS_INQUIRY;
2523                         c->Request.CDB[4] = size & 0xFF;
2524                         break;
2525                 case CISS_REPORT_LOG:
2526                 case CISS_REPORT_PHYS:
2527                         /* Talking to controller so It's a physical command
2528                            mode = 00 target = 0.  Nothing to write.
2529                          */
2530                         c->Request.CDBLen = 12;
2531                         c->Request.Type.Attribute = ATTR_SIMPLE;
2532                         c->Request.Type.Direction = XFER_READ;
2533                         c->Request.Timeout = 0;
2534                         c->Request.CDB[0] = cmd;
2535                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2536                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2537                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2538                         c->Request.CDB[9] = size & 0xFF;
2539                         break;
2540
2541                 case CCISS_READ_CAPACITY:
2542                         c->Request.CDBLen = 10;
2543                         c->Request.Type.Attribute = ATTR_SIMPLE;
2544                         c->Request.Type.Direction = XFER_READ;
2545                         c->Request.Timeout = 0;
2546                         c->Request.CDB[0] = cmd;
2547                         break;
2548                 case CCISS_READ_CAPACITY_16:
2549                         c->Request.CDBLen = 16;
2550                         c->Request.Type.Attribute = ATTR_SIMPLE;
2551                         c->Request.Type.Direction = XFER_READ;
2552                         c->Request.Timeout = 0;
2553                         c->Request.CDB[0] = cmd;
2554                         c->Request.CDB[1] = 0x10;
2555                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2556                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2557                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2558                         c->Request.CDB[13] = size & 0xFF;
2559                         c->Request.Timeout = 0;
2560                         c->Request.CDB[0] = cmd;
2561                         break;
2562                 case CCISS_CACHE_FLUSH:
2563                         c->Request.CDBLen = 12;
2564                         c->Request.Type.Attribute = ATTR_SIMPLE;
2565                         c->Request.Type.Direction = XFER_WRITE;
2566                         c->Request.Timeout = 0;
2567                         c->Request.CDB[0] = BMIC_WRITE;
2568                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2569                         break;
2570                 case TEST_UNIT_READY:
2571                         c->Request.CDBLen = 6;
2572                         c->Request.Type.Attribute = ATTR_SIMPLE;
2573                         c->Request.Type.Direction = XFER_NONE;
2574                         c->Request.Timeout = 0;
2575                         break;
2576                 default:
2577                         printk(KERN_WARNING
2578                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2579                         return IO_ERROR;
2580                 }
2581         } else if (cmd_type == TYPE_MSG) {
2582                 switch (cmd) {
2583                 case 0: /* ABORT message */
2584                         c->Request.CDBLen = 12;
2585                         c->Request.Type.Attribute = ATTR_SIMPLE;
2586                         c->Request.Type.Direction = XFER_WRITE;
2587                         c->Request.Timeout = 0;
2588                         c->Request.CDB[0] = cmd;        /* abort */
2589                         c->Request.CDB[1] = 0;  /* abort a command */
2590                         /* buff contains the tag of the command to abort */
2591                         memcpy(&c->Request.CDB[4], buff, 8);
2592                         break;
2593                 case 1: /* RESET message */
2594                         c->Request.CDBLen = 16;
2595                         c->Request.Type.Attribute = ATTR_SIMPLE;
2596                         c->Request.Type.Direction = XFER_NONE;
2597                         c->Request.Timeout = 0;
2598                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2599                         c->Request.CDB[0] = cmd;        /* reset */
2600                         c->Request.CDB[1] = 0x03;       /* reset a target */
2601                         break;
2602                 case 3: /* No-Op message */
2603                         c->Request.CDBLen = 1;
2604                         c->Request.Type.Attribute = ATTR_SIMPLE;
2605                         c->Request.Type.Direction = XFER_WRITE;
2606                         c->Request.Timeout = 0;
2607                         c->Request.CDB[0] = cmd;
2608                         break;
2609                 default:
2610                         printk(KERN_WARNING
2611                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2612                         return IO_ERROR;
2613                 }
2614         } else {
2615                 printk(KERN_WARNING
2616                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2617                 return IO_ERROR;
2618         }
2619         /* Fill in the scatter gather information */
2620         if (size > 0) {
2621                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2622                                                              buff, size,
2623                                                              PCI_DMA_BIDIRECTIONAL);
2624                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2625                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2626                 c->SG[0].Len = size;
2627                 c->SG[0].Ext = 0;       /* we are not chaining */
2628         }
2629         return status;
2630 }
2631
2632 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2633 {
2634         switch (c->err_info->ScsiStatus) {
2635         case SAM_STAT_GOOD:
2636                 return IO_OK;
2637         case SAM_STAT_CHECK_CONDITION:
2638                 switch (0xf & c->err_info->SenseInfo[2]) {
2639                 case 0: return IO_OK; /* no sense */
2640                 case 1: return IO_OK; /* recovered error */
2641                 default:
2642                         if (check_for_unit_attention(h, c))
2643                                 return IO_NEEDS_RETRY;
2644                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2645                                 "check condition, sense key = 0x%02x\n",
2646                                 h->ctlr, c->Request.CDB[0],
2647                                 c->err_info->SenseInfo[2]);
2648                 }
2649                 break;
2650         default:
2651                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2652                         "scsi status = 0x%02x\n", h->ctlr,
2653                         c->Request.CDB[0], c->err_info->ScsiStatus);
2654                 break;
2655         }
2656         return IO_ERROR;
2657 }
2658
2659 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2660 {
2661         int return_status = IO_OK;
2662
2663         if (c->err_info->CommandStatus == CMD_SUCCESS)
2664                 return IO_OK;
2665
2666         switch (c->err_info->CommandStatus) {
2667         case CMD_TARGET_STATUS:
2668                 return_status = check_target_status(h, c);
2669                 break;
2670         case CMD_DATA_UNDERRUN:
2671         case CMD_DATA_OVERRUN:
2672                 /* expected for inquiry and report lun commands */
2673                 break;
2674         case CMD_INVALID:
2675                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2676                        "reported invalid\n", c->Request.CDB[0]);
2677                 return_status = IO_ERROR;
2678                 break;
2679         case CMD_PROTOCOL_ERR:
2680                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2681                        "protocol error \n", c->Request.CDB[0]);
2682                 return_status = IO_ERROR;
2683                 break;
2684         case CMD_HARDWARE_ERR:
2685                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2686                        " hardware error\n", c->Request.CDB[0]);
2687                 return_status = IO_ERROR;
2688                 break;
2689         case CMD_CONNECTION_LOST:
2690                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2691                        "connection lost\n", c->Request.CDB[0]);
2692                 return_status = IO_ERROR;
2693                 break;
2694         case CMD_ABORTED:
2695                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2696                        "aborted\n", c->Request.CDB[0]);
2697                 return_status = IO_ERROR;
2698                 break;
2699         case CMD_ABORT_FAILED:
2700                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2701                        "abort failed\n", c->Request.CDB[0]);
2702                 return_status = IO_ERROR;
2703                 break;
2704         case CMD_UNSOLICITED_ABORT:
2705                 printk(KERN_WARNING
2706                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2707                         c->Request.CDB[0]);
2708                 return_status = IO_NEEDS_RETRY;
2709                 break;
2710         default:
2711                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2712                        "unknown status %x\n", c->Request.CDB[0],
2713                        c->err_info->CommandStatus);
2714                 return_status = IO_ERROR;
2715         }
2716         return return_status;
2717 }
2718
2719 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2720         int attempt_retry)
2721 {
2722         DECLARE_COMPLETION_ONSTACK(wait);
2723         u64bit buff_dma_handle;
2724         int return_status = IO_OK;
2725
2726 resend_cmd2:
2727         c->waiting = &wait;
2728         enqueue_cmd_and_start_io(h, c);
2729
2730         wait_for_completion(&wait);
2731
2732         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2733                 goto command_done;
2734
2735         return_status = process_sendcmd_error(h, c);
2736
2737         if (return_status == IO_NEEDS_RETRY &&
2738                 c->retry_count < MAX_CMD_RETRIES) {
2739                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2740                         c->Request.CDB[0]);
2741                 c->retry_count++;
2742                 /* erase the old error information */
2743                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2744                 return_status = IO_OK;
2745                 INIT_COMPLETION(wait);
2746                 goto resend_cmd2;
2747         }
2748
2749 command_done:
2750         /* unlock the buffers from DMA */
2751         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2752         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2753         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2754                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2755         return return_status;
2756 }
2757
2758 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2759                            __u8 page_code, unsigned char scsi3addr[],
2760                         int cmd_type)
2761 {
2762         ctlr_info_t *h = hba[ctlr];
2763         CommandList_struct *c;
2764         int return_status;
2765
2766         c = cmd_alloc(h, 0);
2767         if (!c)
2768                 return -ENOMEM;
2769         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2770                 scsi3addr, cmd_type);
2771         if (return_status == IO_OK)
2772                 return_status = sendcmd_withirq_core(h, c, 1);
2773
2774         cmd_free(h, c, 0);
2775         return return_status;
2776 }
2777
2778 static void cciss_geometry_inquiry(int ctlr, int logvol,
2779                                    sector_t total_size,
2780                                    unsigned int block_size,
2781                                    InquiryData_struct *inq_buff,
2782                                    drive_info_struct *drv)
2783 {
2784         int return_code;
2785         unsigned long t;
2786         unsigned char scsi3addr[8];
2787
2788         memset(inq_buff, 0, sizeof(InquiryData_struct));
2789         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2790         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2791                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2792         if (return_code == IO_OK) {
2793                 if (inq_buff->data_byte[8] == 0xFF) {
2794                         printk(KERN_WARNING
2795                                "cciss: reading geometry failed, volume "
2796                                "does not support reading geometry\n");
2797                         drv->heads = 255;
2798                         drv->sectors = 32;      /* Sectors per track */
2799                         drv->cylinders = total_size + 1;
2800                         drv->raid_level = RAID_UNKNOWN;
2801                 } else {
2802                         drv->heads = inq_buff->data_byte[6];
2803                         drv->sectors = inq_buff->data_byte[7];
2804                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2805                         drv->cylinders += inq_buff->data_byte[5];
2806                         drv->raid_level = inq_buff->data_byte[8];
2807                 }
2808                 drv->block_size = block_size;
2809                 drv->nr_blocks = total_size + 1;
2810                 t = drv->heads * drv->sectors;
2811                 if (t > 1) {
2812                         sector_t real_size = total_size + 1;
2813                         unsigned long rem = sector_div(real_size, t);
2814                         if (rem)
2815                                 real_size++;
2816                         drv->cylinders = real_size;
2817                 }
2818         } else {                /* Get geometry failed */
2819                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2820         }
2821 }
2822
2823 static void
2824 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2825                     unsigned int *block_size)
2826 {
2827         ReadCapdata_struct *buf;
2828         int return_code;
2829         unsigned char scsi3addr[8];
2830
2831         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2832         if (!buf) {
2833                 printk(KERN_WARNING "cciss: out of memory\n");
2834                 return;
2835         }
2836
2837         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2838         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2839                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2840         if (return_code == IO_OK) {
2841                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2842                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2843         } else {                /* read capacity command failed */
2844                 printk(KERN_WARNING "cciss: read capacity failed\n");
2845                 *total_size = 0;
2846                 *block_size = BLOCK_SIZE;
2847         }
2848         kfree(buf);
2849 }
2850
2851 static void cciss_read_capacity_16(int ctlr, int logvol,
2852         sector_t *total_size, unsigned int *block_size)
2853 {
2854         ReadCapdata_struct_16 *buf;
2855         int return_code;
2856         unsigned char scsi3addr[8];
2857
2858         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2859         if (!buf) {
2860                 printk(KERN_WARNING "cciss: out of memory\n");
2861                 return;
2862         }
2863
2864         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2865         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2866                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2867                         0, scsi3addr, TYPE_CMD);
2868         if (return_code == IO_OK) {
2869                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2870                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2871         } else {                /* read capacity command failed */
2872                 printk(KERN_WARNING "cciss: read capacity failed\n");
2873                 *total_size = 0;
2874                 *block_size = BLOCK_SIZE;
2875         }
2876         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2877                (unsigned long long)*total_size+1, *block_size);
2878         kfree(buf);
2879 }
2880
2881 static int cciss_revalidate(struct gendisk *disk)
2882 {
2883         ctlr_info_t *h = get_host(disk);
2884         drive_info_struct *drv = get_drv(disk);
2885         int logvol;
2886         int FOUND = 0;
2887         unsigned int block_size;
2888         sector_t total_size;
2889         InquiryData_struct *inq_buff = NULL;
2890
2891         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2892                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2893                         sizeof(drv->LunID)) == 0) {
2894                         FOUND = 1;
2895                         break;
2896                 }
2897         }
2898
2899         if (!FOUND)
2900                 return 1;
2901
2902         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2903         if (inq_buff == NULL) {
2904                 printk(KERN_WARNING "cciss: out of memory\n");
2905                 return 1;
2906         }
2907         if (h->cciss_read == CCISS_READ_10) {
2908                 cciss_read_capacity(h->ctlr, logvol,
2909                                         &total_size, &block_size);
2910         } else {
2911                 cciss_read_capacity_16(h->ctlr, logvol,
2912                                         &total_size, &block_size);
2913         }
2914         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2915                                inq_buff, drv);
2916
2917         blk_queue_logical_block_size(drv->queue, drv->block_size);
2918         set_capacity(disk, drv->nr_blocks);
2919
2920         kfree(inq_buff);
2921         return 0;
2922 }
2923
2924 /*
2925  * Map (physical) PCI mem into (virtual) kernel space
2926  */
2927 static void __iomem *remap_pci_mem(ulong base, ulong size)
2928 {
2929         ulong page_base = ((ulong) base) & PAGE_MASK;
2930         ulong page_offs = ((ulong) base) - page_base;
2931         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2932
2933         return page_remapped ? (page_remapped + page_offs) : NULL;
2934 }
2935
2936 /*
2937  * Takes jobs of the Q and sends them to the hardware, then puts it on
2938  * the Q to wait for completion.
2939  */
2940 static void start_io(ctlr_info_t *h)
2941 {
2942         CommandList_struct *c;
2943
2944         while (!hlist_empty(&h->reqQ)) {
2945                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2946                 /* can't do anything if fifo is full */
2947                 if ((h->access.fifo_full(h))) {
2948                         printk(KERN_WARNING "cciss: fifo full\n");
2949                         break;
2950                 }
2951
2952                 /* Get the first entry from the Request Q */
2953                 removeQ(c);
2954                 h->Qdepth--;
2955
2956                 /* Tell the controller execute command */
2957                 h->access.submit_command(h, c);
2958
2959                 /* Put job onto the completed Q */
2960                 addQ(&h->cmpQ, c);
2961         }
2962 }
2963
2964 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2965 /* Zeros out the error record and then resends the command back */
2966 /* to the controller */
2967 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2968 {
2969         /* erase the old error information */
2970         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2971
2972         /* add it to software queue and then send it to the controller */
2973         addQ(&h->reqQ, c);
2974         h->Qdepth++;
2975         if (h->Qdepth > h->maxQsinceinit)
2976                 h->maxQsinceinit = h->Qdepth;
2977
2978         start_io(h);
2979 }
2980
2981 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2982         unsigned int msg_byte, unsigned int host_byte,
2983         unsigned int driver_byte)
2984 {
2985         /* inverse of macros in scsi.h */
2986         return (scsi_status_byte & 0xff) |
2987                 ((msg_byte & 0xff) << 8) |
2988                 ((host_byte & 0xff) << 16) |
2989                 ((driver_byte & 0xff) << 24);
2990 }
2991
2992 static inline int evaluate_target_status(ctlr_info_t *h,
2993                         CommandList_struct *cmd, int *retry_cmd)
2994 {
2995         unsigned char sense_key;
2996         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2997         int error_value;
2998
2999         *retry_cmd = 0;
3000         /* If we get in here, it means we got "target status", that is, scsi status */
3001         status_byte = cmd->err_info->ScsiStatus;
3002         driver_byte = DRIVER_OK;
3003         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3004
3005         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3006                 host_byte = DID_PASSTHROUGH;
3007         else
3008                 host_byte = DID_OK;
3009
3010         error_value = make_status_bytes(status_byte, msg_byte,
3011                 host_byte, driver_byte);
3012
3013         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3014                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3015                         printk(KERN_WARNING "cciss: cmd %p "
3016                                "has SCSI Status 0x%x\n",
3017                                cmd, cmd->err_info->ScsiStatus);
3018                 return error_value;
3019         }
3020
3021         /* check the sense key */
3022         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3023         /* no status or recovered error */
3024         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3025             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3026                 error_value = 0;
3027
3028         if (check_for_unit_attention(h, cmd)) {
3029                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3030                 return 0;
3031         }
3032
3033         /* Not SG_IO or similar? */
3034         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
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 (cmd->rq->cmd_type == REQ_TYPE_FS) {
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 (cmd->rq->cmd_type == REQ_TYPE_FS)
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                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3095                                 DID_PASSTHROUGH : DID_ERROR);
3096                 break;
3097         case CMD_PROTOCOL_ERR:
3098                 printk(KERN_WARNING "cciss: cmd %p has "
3099                        "protocol error \n", cmd);
3100                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3101                         cmd->err_info->CommandStatus, DRIVER_OK,
3102                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3103                                 DID_PASSTHROUGH : DID_ERROR);
3104                 break;
3105         case CMD_HARDWARE_ERR:
3106                 printk(KERN_WARNING "cciss: cmd %p had "
3107                        " hardware error\n", cmd);
3108                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3109                         cmd->err_info->CommandStatus, DRIVER_OK,
3110                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3111                                 DID_PASSTHROUGH : DID_ERROR);
3112                 break;
3113         case CMD_CONNECTION_LOST:
3114                 printk(KERN_WARNING "cciss: cmd %p had "
3115                        "connection lost\n", cmd);
3116                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3117                         cmd->err_info->CommandStatus, DRIVER_OK,
3118                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3119                                 DID_PASSTHROUGH : DID_ERROR);
3120                 break;
3121         case CMD_ABORTED:
3122                 printk(KERN_WARNING "cciss: cmd %p was "
3123                        "aborted\n", cmd);
3124                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3125                         cmd->err_info->CommandStatus, DRIVER_OK,
3126                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3127                                 DID_PASSTHROUGH : DID_ABORT);
3128                 break;
3129         case CMD_ABORT_FAILED:
3130                 printk(KERN_WARNING "cciss: cmd %p reports "
3131                        "abort failed\n", cmd);
3132                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3133                         cmd->err_info->CommandStatus, DRIVER_OK,
3134                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3135                                 DID_PASSTHROUGH : DID_ERROR);
3136                 break;
3137         case CMD_UNSOLICITED_ABORT:
3138                 printk(KERN_WARNING "cciss%d: unsolicited "
3139                        "abort %p\n", h->ctlr, cmd);
3140                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3141                         retry_cmd = 1;
3142                         printk(KERN_WARNING
3143                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3144                         cmd->retry_count++;
3145                 } else
3146                         printk(KERN_WARNING
3147                                "cciss%d: %p retried too "
3148                                "many times\n", h->ctlr, cmd);
3149                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3150                         cmd->err_info->CommandStatus, DRIVER_OK,
3151                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3152                                 DID_PASSTHROUGH : DID_ABORT);
3153                 break;
3154         case CMD_TIMEOUT:
3155                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3156                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3157                         cmd->err_info->CommandStatus, DRIVER_OK,
3158                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3159                                 DID_PASSTHROUGH : DID_ERROR);
3160                 break;
3161         default:
3162                 printk(KERN_WARNING "cciss: cmd %p returned "
3163                        "unknown status %x\n", cmd,
3164                        cmd->err_info->CommandStatus);
3165                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3166                         cmd->err_info->CommandStatus, DRIVER_OK,
3167                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3168                                 DID_PASSTHROUGH : DID_ERROR);
3169         }
3170
3171 after_error_processing:
3172
3173         /* We need to return this command */
3174         if (retry_cmd) {
3175                 resend_cciss_cmd(h, cmd);
3176                 return;
3177         }
3178         cmd->rq->completion_data = cmd;
3179         blk_complete_request(cmd->rq);
3180 }
3181
3182 static inline u32 cciss_tag_contains_index(u32 tag)
3183 {
3184 #define DIRECT_LOOKUP_BIT 0x10
3185         return tag & DIRECT_LOOKUP_BIT;
3186 }
3187
3188 static inline u32 cciss_tag_to_index(u32 tag)
3189 {
3190 #define DIRECT_LOOKUP_SHIFT 5
3191         return tag >> DIRECT_LOOKUP_SHIFT;
3192 }
3193
3194 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3195 {
3196 #define CCISS_ERROR_BITS 0x03
3197         return tag & ~CCISS_ERROR_BITS;
3198 }
3199
3200 static inline void cciss_mark_tag_indexed(u32 *tag)
3201 {
3202         *tag |= DIRECT_LOOKUP_BIT;
3203 }
3204
3205 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3206 {
3207         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3208 }
3209
3210 /*
3211  * Get a request and submit it to the controller.
3212  */
3213 static void do_cciss_request(struct request_queue *q)
3214 {
3215         ctlr_info_t *h = q->queuedata;
3216         CommandList_struct *c;
3217         sector_t start_blk;
3218         int seg;
3219         struct request *creq;
3220         u64bit temp64;
3221         struct scatterlist *tmp_sg;
3222         SGDescriptor_struct *curr_sg;
3223         drive_info_struct *drv;
3224         int i, dir;
3225         int sg_index = 0;
3226         int chained = 0;
3227
3228         /* We call start_io here in case there is a command waiting on the
3229          * queue that has not been sent.
3230          */
3231         if (blk_queue_plugged(q))
3232                 goto startio;
3233
3234       queue:
3235         creq = blk_peek_request(q);
3236         if (!creq)
3237                 goto startio;
3238
3239         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3240
3241         if ((c = cmd_alloc(h, 1)) == NULL)
3242                 goto full;
3243
3244         blk_start_request(creq);
3245
3246         tmp_sg = h->scatter_list[c->cmdindex];
3247         spin_unlock_irq(q->queue_lock);
3248
3249         c->cmd_type = CMD_RWREQ;
3250         c->rq = creq;
3251
3252         /* fill in the request */
3253         drv = creq->rq_disk->private_data;
3254         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3255         /* got command from pool, so use the command block index instead */
3256         /* for direct lookups. */
3257         /* The first 2 bits are reserved for controller error reporting. */
3258         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3259         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3260         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3261         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3262         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3263         c->Request.Type.Attribute = ATTR_SIMPLE;
3264         c->Request.Type.Direction =
3265             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3266         c->Request.Timeout = 0; /* Don't time out */
3267         c->Request.CDB[0] =
3268             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3269         start_blk = blk_rq_pos(creq);
3270 #ifdef CCISS_DEBUG
3271         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3272                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3273 #endif                          /* CCISS_DEBUG */
3274
3275         sg_init_table(tmp_sg, h->maxsgentries);
3276         seg = blk_rq_map_sg(q, creq, tmp_sg);
3277
3278         /* get the DMA records for the setup */
3279         if (c->Request.Type.Direction == XFER_READ)
3280                 dir = PCI_DMA_FROMDEVICE;
3281         else
3282                 dir = PCI_DMA_TODEVICE;
3283
3284         curr_sg = c->SG;
3285         sg_index = 0;
3286         chained = 0;
3287
3288         for (i = 0; i < seg; i++) {
3289                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3290                         !chained && ((seg - i) > 1)) {
3291                         /* Point to next chain block. */
3292                         curr_sg = h->cmd_sg_list[c->cmdindex];
3293                         sg_index = 0;
3294                         chained = 1;
3295                 }
3296                 curr_sg[sg_index].Len = tmp_sg[i].length;
3297                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3298                                                 tmp_sg[i].offset,
3299                                                 tmp_sg[i].length, dir);
3300                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3301                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3302                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3303                 ++sg_index;
3304         }
3305         if (chained)
3306                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3307                         (seg - (h->max_cmd_sgentries - 1)) *
3308                                 sizeof(SGDescriptor_struct));
3309
3310         /* track how many SG entries we are using */
3311         if (seg > h->maxSG)
3312                 h->maxSG = seg;
3313
3314 #ifdef CCISS_DEBUG
3315         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3316                         "chained[%d]\n",
3317                         blk_rq_sectors(creq), seg, chained);
3318 #endif                          /* CCISS_DEBUG */
3319
3320         c->Header.SGTotal = seg + chained;
3321         if (seg <= h->max_cmd_sgentries)
3322                 c->Header.SGList = c->Header.SGTotal;
3323         else
3324                 c->Header.SGList = h->max_cmd_sgentries;
3325         set_performant_mode(h, c);
3326
3327         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3328                 if(h->cciss_read == CCISS_READ_10) {
3329                         c->Request.CDB[1] = 0;
3330                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3331                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3332                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3333                         c->Request.CDB[5] = start_blk & 0xff;
3334                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3335                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3336                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3337                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3338                 } else {
3339                         u32 upper32 = upper_32_bits(start_blk);
3340
3341                         c->Request.CDBLen = 16;
3342                         c->Request.CDB[1]= 0;
3343                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3344                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3345                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3346                         c->Request.CDB[5]= upper32 & 0xff;
3347                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3348                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3349                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3350                         c->Request.CDB[9]= start_blk & 0xff;
3351                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3352                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3353                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3354                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3355                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3356                 }
3357         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3358                 c->Request.CDBLen = creq->cmd_len;
3359                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3360         } else {
3361                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3362                 BUG();
3363         }
3364
3365         spin_lock_irq(q->queue_lock);
3366
3367         addQ(&h->reqQ, c);
3368         h->Qdepth++;
3369         if (h->Qdepth > h->maxQsinceinit)
3370                 h->maxQsinceinit = h->Qdepth;
3371
3372         goto queue;
3373 full:
3374         blk_stop_queue(q);
3375 startio:
3376         /* We will already have the driver lock here so not need
3377          * to lock it.
3378          */
3379         start_io(h);
3380 }
3381
3382 static inline unsigned long get_next_completion(ctlr_info_t *h)
3383 {
3384         return h->access.command_completed(h);
3385 }
3386
3387 static inline int interrupt_pending(ctlr_info_t *h)
3388 {
3389         return h->access.intr_pending(h);
3390 }
3391
3392 static inline long interrupt_not_for_us(ctlr_info_t *h)
3393 {
3394         return !(h->msi_vector || h->msix_vector) &&
3395                 ((h->access.intr_pending(h) == 0) ||
3396                 (h->interrupts_enabled == 0));
3397 }
3398
3399 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3400                         u32 raw_tag)
3401 {
3402         if (unlikely(tag_index >= h->nr_cmds)) {
3403                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3404                 return 1;
3405         }
3406         return 0;
3407 }
3408
3409 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3410                                 u32 raw_tag)
3411 {
3412         removeQ(c);
3413         if (likely(c->cmd_type == CMD_RWREQ))
3414                 complete_command(h, c, 0);
3415         else if (c->cmd_type == CMD_IOCTL_PEND)
3416                 complete(c->waiting);
3417 #ifdef CONFIG_CISS_SCSI_TAPE
3418         else if (c->cmd_type == CMD_SCSI)
3419                 complete_scsi_command(c, 0, raw_tag);
3420 #endif
3421 }
3422
3423 static inline u32 next_command(ctlr_info_t *h)
3424 {
3425         u32 a;
3426
3427         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3428                 return h->access.command_completed(h);
3429
3430         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3431                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3432                 (h->reply_pool_head)++;
3433                 h->commands_outstanding--;
3434         } else {
3435                 a = FIFO_EMPTY;
3436         }
3437         /* Check for wraparound */
3438         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3439                 h->reply_pool_head = h->reply_pool;
3440                 h->reply_pool_wraparound ^= 1;
3441         }
3442         return a;
3443 }
3444
3445 /* process completion of an indexed ("direct lookup") command */
3446 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3447 {
3448         u32 tag_index;
3449         CommandList_struct *c;
3450
3451         tag_index = cciss_tag_to_index(raw_tag);
3452         if (bad_tag(h, tag_index, raw_tag))
3453                 return next_command(h);
3454         c = h->cmd_pool + tag_index;
3455         finish_cmd(h, c, raw_tag);
3456         return next_command(h);
3457 }
3458
3459 /* process completion of a non-indexed command */
3460 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3461 {
3462         u32 tag;
3463         CommandList_struct *c = NULL;
3464         struct hlist_node *tmp;
3465         __u32 busaddr_masked, tag_masked;
3466
3467         tag = cciss_tag_discard_error_bits(raw_tag);
3468         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3469                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3470                 tag_masked = cciss_tag_discard_error_bits(tag);
3471                 if (busaddr_masked == tag_masked) {
3472                         finish_cmd(h, c, raw_tag);
3473                         return next_command(h);
3474                 }
3475         }
3476         bad_tag(h, h->nr_cmds + 1, raw_tag);
3477         return next_command(h);
3478 }
3479
3480 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3481 {
3482         ctlr_info_t *h = dev_id;
3483         unsigned long flags;
3484         u32 raw_tag;
3485
3486         if (interrupt_not_for_us(h))
3487                 return IRQ_NONE;
3488         /*
3489          * If there are completed commands in the completion queue,
3490          * we had better do something about it.
3491          */
3492         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3493         while (interrupt_pending(h)) {
3494                 raw_tag = get_next_completion(h);
3495                 while (raw_tag != FIFO_EMPTY) {
3496                         if (cciss_tag_contains_index(raw_tag))
3497                                 raw_tag = process_indexed_cmd(h, raw_tag);
3498                         else
3499                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3500                 }
3501         }
3502
3503         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3504         return IRQ_HANDLED;
3505 }
3506
3507 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3508  * check the interrupt pending register because it is not set.
3509  */
3510 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3511 {
3512         ctlr_info_t *h = dev_id;
3513         unsigned long flags;
3514         u32 raw_tag;
3515
3516         if (interrupt_not_for_us(h))
3517                 return IRQ_NONE;
3518         /*
3519          * If there are completed commands in the completion queue,
3520          * we had better do something about it.
3521          */
3522         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3523         raw_tag = get_next_completion(h);
3524         while (raw_tag != FIFO_EMPTY) {
3525                 if (cciss_tag_contains_index(raw_tag))
3526                         raw_tag = process_indexed_cmd(h, raw_tag);
3527                 else
3528                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3529         }
3530
3531         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3532         return IRQ_HANDLED;
3533 }
3534
3535 /**
3536  * add_to_scan_list() - add controller to rescan queue
3537  * @h:                Pointer to the controller.
3538  *
3539  * Adds the controller to the rescan queue if not already on the queue.
3540  *
3541  * returns 1 if added to the queue, 0 if skipped (could be on the
3542  * queue already, or the controller could be initializing or shutting
3543  * down).
3544  **/
3545 static int add_to_scan_list(struct ctlr_info *h)
3546 {
3547         struct ctlr_info *test_h;
3548         int found = 0;
3549         int ret = 0;
3550
3551         if (h->busy_initializing)
3552                 return 0;
3553
3554         if (!mutex_trylock(&h->busy_shutting_down))
3555                 return 0;
3556
3557         mutex_lock(&scan_mutex);
3558         list_for_each_entry(test_h, &scan_q, scan_list) {
3559                 if (test_h == h) {
3560                         found = 1;
3561                         break;
3562                 }
3563         }
3564         if (!found && !h->busy_scanning) {
3565                 INIT_COMPLETION(h->scan_wait);
3566                 list_add_tail(&h->scan_list, &scan_q);
3567                 ret = 1;
3568         }
3569         mutex_unlock(&scan_mutex);
3570         mutex_unlock(&h->busy_shutting_down);
3571
3572         return ret;
3573 }
3574
3575 /**
3576  * remove_from_scan_list() - remove controller from rescan queue
3577  * @h:                     Pointer to the controller.
3578  *
3579  * Removes the controller from the rescan queue if present. Blocks if
3580  * the controller is currently conducting a rescan.  The controller
3581  * can be in one of three states:
3582  * 1. Doesn't need a scan
3583  * 2. On the scan list, but not scanning yet (we remove it)
3584  * 3. Busy scanning (and not on the list). In this case we want to wait for
3585  *    the scan to complete to make sure the scanning thread for this
3586  *    controller is completely idle.
3587  **/
3588 static void remove_from_scan_list(struct ctlr_info *h)
3589 {
3590         struct ctlr_info *test_h, *tmp_h;
3591
3592         mutex_lock(&scan_mutex);
3593         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3594                 if (test_h == h) { /* state 2. */
3595                         list_del(&h->scan_list);
3596                         complete_all(&h->scan_wait);
3597                         mutex_unlock(&scan_mutex);
3598                         return;
3599                 }
3600         }
3601         if (h->busy_scanning) { /* state 3. */
3602                 mutex_unlock(&scan_mutex);
3603                 wait_for_completion(&h->scan_wait);
3604         } else { /* state 1, nothing to do. */
3605                 mutex_unlock(&scan_mutex);
3606         }
3607 }
3608
3609 /**
3610  * scan_thread() - kernel thread used to rescan controllers
3611  * @data:        Ignored.
3612  *
3613  * A kernel thread used scan for drive topology changes on
3614  * controllers. The thread processes only one controller at a time
3615  * using a queue.  Controllers are added to the queue using
3616  * add_to_scan_list() and removed from the queue either after done
3617  * processing or using remove_from_scan_list().
3618  *
3619  * returns 0.
3620  **/
3621 static int scan_thread(void *data)
3622 {
3623         struct ctlr_info *h;
3624
3625         while (1) {
3626                 set_current_state(TASK_INTERRUPTIBLE);
3627                 schedule();
3628                 if (kthread_should_stop())
3629                         break;
3630
3631                 while (1) {
3632                         mutex_lock(&scan_mutex);
3633                         if (list_empty(&scan_q)) {
3634                                 mutex_unlock(&scan_mutex);
3635                                 break;
3636                         }
3637
3638                         h = list_entry(scan_q.next,
3639                                        struct ctlr_info,
3640                                        scan_list);
3641                         list_del(&h->scan_list);
3642                         h->busy_scanning = 1;
3643                         mutex_unlock(&scan_mutex);
3644
3645                         rebuild_lun_table(h, 0, 0);
3646                         complete_all(&h->scan_wait);
3647                         mutex_lock(&scan_mutex);
3648                         h->busy_scanning = 0;
3649                         mutex_unlock(&scan_mutex);
3650                 }
3651         }
3652
3653         return 0;
3654 }
3655
3656 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3657 {
3658         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3659                 return 0;
3660
3661         switch (c->err_info->SenseInfo[12]) {
3662         case STATE_CHANGED:
3663                 printk(KERN_WARNING "cciss%d: a state change "
3664                         "detected, command retried\n", h->ctlr);
3665                 return 1;
3666         break;
3667         case LUN_FAILED:
3668                 printk(KERN_WARNING "cciss%d: LUN failure "
3669                         "detected, action required\n", h->ctlr);
3670                 return 1;
3671         break;
3672         case REPORT_LUNS_CHANGED:
3673                 printk(KERN_WARNING "cciss%d: report LUN data "
3674                         "changed\n", h->ctlr);
3675         /*
3676          * Here, we could call add_to_scan_list and wake up the scan thread,
3677          * except that it's quite likely that we will get more than one
3678          * REPORT_LUNS_CHANGED condition in quick succession, which means
3679          * that those which occur after the first one will likely happen
3680          * *during* the scan_thread's rescan.  And the rescan code is not
3681          * robust enough to restart in the middle, undoing what it has already
3682          * done, and it's not clear that it's even possible to do this, since
3683          * part of what it does is notify the block layer, which starts
3684          * doing it's own i/o to read partition tables and so on, and the
3685          * driver doesn't have visibility to know what might need undoing.
3686          * In any event, if possible, it is horribly complicated to get right
3687          * so we just don't do it for now.
3688          *
3689          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3690          */
3691                 return 1;
3692         break;
3693         case POWER_OR_RESET:
3694                 printk(KERN_WARNING "cciss%d: a power on "
3695                         "or device reset detected\n", h->ctlr);
3696                 return 1;
3697         break;
3698         case UNIT_ATTENTION_CLEARED:
3699                 printk(KERN_WARNING "cciss%d: unit attention "
3700                     "cleared by another initiator\n", h->ctlr);
3701                 return 1;
3702         break;
3703         default:
3704                 printk(KERN_WARNING "cciss%d: unknown "
3705                         "unit attention detected\n", h->ctlr);
3706                                 return 1;
3707         }
3708 }
3709
3710 /*
3711  *  We cannot read the structure directly, for portability we must use
3712  *   the io functions.
3713  *   This is for debug only.
3714  */
3715 #ifdef CCISS_DEBUG
3716 static void print_cfg_table(CfgTable_struct *tb)
3717 {
3718         int i;
3719         char temp_name[17];
3720
3721         printk("Controller Configuration information\n");
3722         printk("------------------------------------\n");
3723         for (i = 0; i < 4; i++)
3724                 temp_name[i] = readb(&(tb->Signature[i]));
3725         temp_name[4] = '\0';
3726         printk("   Signature = %s\n", temp_name);
3727         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3728         printk("   Transport methods supported = 0x%x\n",
3729                readl(&(tb->TransportSupport)));
3730         printk("   Transport methods active = 0x%x\n",
3731                readl(&(tb->TransportActive)));
3732         printk("   Requested transport Method = 0x%x\n",
3733                readl(&(tb->HostWrite.TransportRequest)));
3734         printk("   Coalesce Interrupt Delay = 0x%x\n",
3735                readl(&(tb->HostWrite.CoalIntDelay)));
3736         printk("   Coalesce Interrupt Count = 0x%x\n",
3737                readl(&(tb->HostWrite.CoalIntCount)));
3738         printk("   Max outstanding commands = 0x%d\n",
3739                readl(&(tb->CmdsOutMax)));
3740         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3741         for (i = 0; i < 16; i++)
3742                 temp_name[i] = readb(&(tb->ServerName[i]));
3743         temp_name[16] = '\0';
3744         printk("   Server Name = %s\n", temp_name);
3745         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3746 }
3747 #endif                          /* CCISS_DEBUG */
3748
3749 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3750 {
3751         int i, offset, mem_type, bar_type;
3752         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3753                 return 0;
3754         offset = 0;
3755         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3756                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3757                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3758                         offset += 4;
3759                 else {
3760                         mem_type = pci_resource_flags(pdev, i) &
3761                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3762                         switch (mem_type) {
3763                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3764                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3765                                 offset += 4;    /* 32 bit */
3766                                 break;
3767                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3768                                 offset += 8;
3769                                 break;
3770                         default:        /* reserved in PCI 2.2 */
3771                                 printk(KERN_WARNING
3772                                        "Base address is invalid\n");
3773                                 return -1;
3774                                 break;
3775                         }
3776                 }
3777                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3778                         return i + 1;
3779         }
3780         return -1;
3781 }
3782
3783 /* Fill in bucket_map[], given nsgs (the max number of
3784  * scatter gather elements supported) and bucket[],
3785  * which is an array of 8 integers.  The bucket[] array
3786  * contains 8 different DMA transfer sizes (in 16
3787  * byte increments) which the controller uses to fetch
3788  * commands.  This function fills in bucket_map[], which
3789  * maps a given number of scatter gather elements to one of
3790  * the 8 DMA transfer sizes.  The point of it is to allow the
3791  * controller to only do as much DMA as needed to fetch the
3792  * command, with the DMA transfer size encoded in the lower
3793  * bits of the command address.
3794  */
3795 static void  calc_bucket_map(int bucket[], int num_buckets,
3796         int nsgs, int *bucket_map)
3797 {
3798         int i, j, b, size;
3799
3800         /* even a command with 0 SGs requires 4 blocks */
3801 #define MINIMUM_TRANSFER_BLOCKS 4
3802 #define NUM_BUCKETS 8
3803         /* Note, bucket_map must have nsgs+1 entries. */
3804         for (i = 0; i <= nsgs; i++) {
3805                 /* Compute size of a command with i SG entries */
3806                 size = i + MINIMUM_TRANSFER_BLOCKS;
3807                 b = num_buckets; /* Assume the biggest bucket */
3808                 /* Find the bucket that is just big enough */
3809                 for (j = 0; j < 8; j++) {
3810                         if (bucket[j] >= size) {
3811                                 b = j;
3812                                 break;
3813                         }
3814                 }
3815                 /* for a command with i SG entries, use bucket b. */
3816                 bucket_map[i] = b;
3817         }
3818 }
3819
3820 static void
3821 cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3822 {
3823         int l = 0;
3824         __u32 trans_support;
3825         __u32 trans_offset;
3826                         /*
3827                          *  5 = 1 s/g entry or 4k
3828                          *  6 = 2 s/g entry or 8k
3829                          *  8 = 4 s/g entry or 16k
3830                          * 10 = 6 s/g entry or 24k
3831                          */
3832         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3833         unsigned long register_value;
3834
3835         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3836
3837         /* Attempt to put controller into performant mode if supported */
3838         /* Does board support performant mode? */
3839         trans_support = readl(&(h->cfgtable->TransportSupport));
3840         if (!(trans_support & PERFORMANT_MODE))
3841                 return;
3842
3843         printk(KERN_WARNING "cciss%d: Placing controller into "
3844                                 "performant mode\n", h->ctlr);
3845         /* Performant mode demands commands on a 32 byte boundary
3846          * pci_alloc_consistent aligns on page boundarys already.
3847          * Just need to check if divisible by 32
3848          */
3849         if ((sizeof(CommandList_struct) % 32) != 0) {
3850                 printk(KERN_WARNING "%s %d %s\n",
3851                         "cciss info: command size[",
3852                         (int)sizeof(CommandList_struct),
3853                         "] not divisible by 32, no performant mode..\n");
3854                 return;
3855         }
3856
3857         /* Performant mode ring buffer and supporting data structures */
3858         h->reply_pool = (__u64 *)pci_alloc_consistent(
3859                 h->pdev, h->max_commands * sizeof(__u64),
3860                 &(h->reply_pool_dhandle));
3861
3862         /* Need a block fetch table for performant mode */
3863         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3864                 sizeof(__u32)), GFP_KERNEL);
3865
3866         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3867                 goto clean_up;
3868
3869         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3870
3871         /* Controller spec: zero out this buffer. */
3872         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3873         h->reply_pool_head = h->reply_pool;
3874
3875         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3876         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3877                                 h->blockFetchTable);
3878         writel(bft[0], &h->transtable->BlockFetch0);
3879         writel(bft[1], &h->transtable->BlockFetch1);
3880         writel(bft[2], &h->transtable->BlockFetch2);
3881         writel(bft[3], &h->transtable->BlockFetch3);
3882         writel(bft[4], &h->transtable->BlockFetch4);
3883         writel(bft[5], &h->transtable->BlockFetch5);
3884         writel(bft[6], &h->transtable->BlockFetch6);
3885         writel(bft[7], &h->transtable->BlockFetch7);
3886
3887         /* size of controller ring buffer */
3888         writel(h->max_commands, &h->transtable->RepQSize);
3889         writel(1, &h->transtable->RepQCount);
3890         writel(0, &h->transtable->RepQCtrAddrLow32);
3891         writel(0, &h->transtable->RepQCtrAddrHigh32);
3892         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3893         writel(0, &h->transtable->RepQAddr0High32);
3894         writel(CFGTBL_Trans_Performant,
3895                         &(h->cfgtable->HostWrite.TransportRequest));
3896
3897         h->transMethod = CFGTBL_Trans_Performant;
3898         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3899         /* under certain very rare conditions, this can take awhile.
3900          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3901          * as we enter this code.) */
3902         for (l = 0; l < MAX_CONFIG_WAIT; l++) {
3903                 register_value = readl(h->vaddr + SA5_DOORBELL);
3904                 if (!(register_value & CFGTBL_ChangeReq))
3905                         break;
3906                 /* delay and try again */
3907                 set_current_state(TASK_INTERRUPTIBLE);
3908                 schedule_timeout(10);
3909         }
3910         register_value = readl(&(h->cfgtable->TransportActive));
3911         if (!(register_value & CFGTBL_Trans_Performant)) {
3912                 printk(KERN_WARNING "cciss: unable to get board into"
3913                                         " performant mode\n");
3914                 return;
3915         }
3916
3917         /* Change the access methods to the performant access methods */
3918         h->access = SA5_performant_access;
3919
3920         return;
3921 clean_up:
3922         kfree(h->blockFetchTable);
3923         if (h->reply_pool)
3924                 pci_free_consistent(h->pdev,
3925                                 h->max_commands * sizeof(__u64),
3926                                 h->reply_pool,
3927                                 h->reply_pool_dhandle);
3928         return;
3929
3930 } /* cciss_put_controller_into_performant_mode */
3931
3932 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3933  * controllers that are capable. If not, we use IO-APIC mode.
3934  */
3935
3936 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3937                                            struct pci_dev *pdev, __u32 board_id)
3938 {
3939 #ifdef CONFIG_PCI_MSI
3940         int err;
3941         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3942         {0, 2}, {0, 3}
3943         };
3944
3945         /* Some boards advertise MSI but don't really support it */
3946         if ((board_id == 0x40700E11) ||
3947             (board_id == 0x40800E11) ||
3948             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3949                 goto default_int_mode;
3950
3951         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3952                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3953                 if (!err) {
3954                         c->intr[0] = cciss_msix_entries[0].vector;
3955                         c->intr[1] = cciss_msix_entries[1].vector;
3956                         c->intr[2] = cciss_msix_entries[2].vector;
3957                         c->intr[3] = cciss_msix_entries[3].vector;
3958                         c->msix_vector = 1;
3959                         return;
3960                 }
3961                 if (err > 0) {
3962                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3963                                "available\n", err);
3964                         goto default_int_mode;
3965                 } else {
3966                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3967                                err);
3968                         goto default_int_mode;
3969                 }
3970         }
3971         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3972                 if (!pci_enable_msi(pdev)) {
3973                         c->msi_vector = 1;
3974                 } else {
3975                         printk(KERN_WARNING "cciss: MSI init failed\n");
3976                 }
3977         }
3978 default_int_mode:
3979 #endif                          /* CONFIG_PCI_MSI */
3980         /* if we get here we're going to use the default interrupt mode */
3981         c->intr[PERF_MODE_INT] = pdev->irq;
3982         return;
3983 }
3984
3985 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3986 {
3987         ushort subsystem_vendor_id, subsystem_device_id, command;
3988         __u32 board_id, scratchpad = 0;
3989         __u64 cfg_offset;
3990         __u32 cfg_base_addr;
3991         __u64 cfg_base_addr_index;
3992         int i, prod_index, err;
3993         __u32 trans_offset;
3994
3995         subsystem_vendor_id = pdev->subsystem_vendor;
3996         subsystem_device_id = pdev->subsystem_device;
3997         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3998                     subsystem_vendor_id);
3999
4000         for (i = 0; i < ARRAY_SIZE(products); i++) {
4001                 /* Stand aside for hpsa driver on request */
4002                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
4003                         return -ENODEV;
4004                 if (board_id == products[i].board_id)
4005                         break;
4006         }
4007         prod_index = i;
4008         if (prod_index == ARRAY_SIZE(products)) {
4009                 dev_warn(&pdev->dev,
4010                         "unrecognized board ID: 0x%08lx, ignoring.\n",
4011                         (unsigned long) board_id);
4012                 return -ENODEV;
4013         }
4014
4015         /* check to see if controller has been disabled */
4016         /* BEFORE trying to enable it */
4017         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
4018         if (!(command & 0x02)) {
4019                 printk(KERN_WARNING
4020                        "cciss: controller appears to be disabled\n");
4021                 return -ENODEV;
4022         }
4023
4024         err = pci_enable_device(pdev);
4025         if (err) {
4026                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
4027                 return err;
4028         }
4029
4030         err = pci_request_regions(pdev, "cciss");
4031         if (err) {
4032                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
4033                        "aborting\n");
4034                 return err;
4035         }
4036
4037 #ifdef CCISS_DEBUG
4038         printk("command = %x\n", command);
4039         printk("irq = %x\n", pdev->irq);
4040         printk("board_id = %x\n", board_id);
4041 #endif                          /* CCISS_DEBUG */
4042
4043 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4044  * else we use the IO-APIC interrupt assigned to us by system ROM.
4045  */
4046         cciss_interrupt_mode(c, pdev, board_id);
4047
4048         /* find the memory BAR */
4049         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
4050                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
4051                         break;
4052         }
4053         if (i == DEVICE_COUNT_RESOURCE) {
4054                 printk(KERN_WARNING "cciss: No memory BAR found\n");
4055                 err = -ENODEV;
4056                 goto err_out_free_res;
4057         }
4058
4059         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
4060                                                  * already removed
4061                                                  */
4062
4063 #ifdef CCISS_DEBUG
4064         printk("address 0 = %lx\n", c->paddr);
4065 #endif                          /* CCISS_DEBUG */
4066         c->vaddr = remap_pci_mem(c->paddr, 0x250);
4067
4068         /* Wait for the board to become ready.  (PCI hotplug needs this.)
4069          * We poll for up to 120 secs, once per 100ms. */
4070         for (i = 0; i < 1200; i++) {
4071                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
4072                 if (scratchpad == CCISS_FIRMWARE_READY)
4073                         break;
4074                 set_current_state(TASK_INTERRUPTIBLE);
4075                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
4076         }
4077         if (scratchpad != CCISS_FIRMWARE_READY) {
4078                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
4079                 err = -ENODEV;
4080                 goto err_out_free_res;
4081         }
4082
4083         /* get the address index number */
4084         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
4085         cfg_base_addr &= (__u32) 0x0000ffff;
4086 #ifdef CCISS_DEBUG
4087         printk("cfg base address = %x\n", cfg_base_addr);
4088 #endif                          /* CCISS_DEBUG */
4089         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
4090 #ifdef CCISS_DEBUG
4091         printk("cfg base address index = %llx\n",
4092                 (unsigned long long)cfg_base_addr_index);
4093 #endif                          /* CCISS_DEBUG */
4094         if (cfg_base_addr_index == -1) {
4095                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
4096                 err = -ENODEV;
4097                 goto err_out_free_res;
4098         }
4099
4100         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
4101 #ifdef CCISS_DEBUG
4102         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
4103 #endif                          /* CCISS_DEBUG */
4104         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
4105                                                        cfg_base_addr_index) +
4106                                     cfg_offset, sizeof(CfgTable_struct));
4107         /* Find performant mode table. */
4108         trans_offset = readl(&(c->cfgtable->TransMethodOffset));
4109         c->transtable = remap_pci_mem(pci_resource_start(pdev,
4110                 cfg_base_addr_index) + cfg_offset+trans_offset,
4111                 sizeof(*c->transtable));
4112         c->board_id = board_id;
4113
4114         #ifdef CCISS_DEBUG
4115                 print_cfg_table(c->cfgtable);
4116         #endif                          /* CCISS_DEBUG */
4117
4118         /* Some controllers support Zero Memory Raid (ZMR).
4119          * When configured in ZMR mode the number of supported
4120          * commands drops to 64. So instead of just setting an
4121          * arbitrary value we make the driver a little smarter.
4122          * We read the config table to tell us how many commands
4123          * are supported on the controller then subtract 4 to
4124          * leave a little room for ioctl calls.
4125          */
4126         c->max_commands = readl(&(c->cfgtable->MaxPerformantModeCommands));
4127         c->maxsgentries = readl(&(c->cfgtable->MaxSGElements));
4128
4129         /*
4130          * Limit native command to 32 s/g elements to save dma'able memory.
4131          * Howvever spec says if 0, use 31
4132          */
4133
4134         c->max_cmd_sgentries = 31;
4135         if (c->maxsgentries > 512) {
4136                 c->max_cmd_sgentries = 32;
4137                 c->chainsize = c->maxsgentries - c->max_cmd_sgentries + 1;
4138                 c->maxsgentries -= 1;   /* account for chain pointer */
4139         } else {
4140                 c->maxsgentries = 31;   /* Default to traditional value */
4141                 c->chainsize = 0;       /* traditional */
4142         }
4143
4144         c->product_name = products[prod_index].product_name;
4145         c->access = *(products[prod_index].access);
4146         c->nr_cmds = c->max_commands - 4;
4147         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
4148             (readb(&c->cfgtable->Signature[1]) != 'I') ||
4149             (readb(&c->cfgtable->Signature[2]) != 'S') ||
4150             (readb(&c->cfgtable->Signature[3]) != 'S')) {
4151                 printk("Does not appear to be a valid CISS config table\n");
4152                 err = -ENODEV;
4153                 goto err_out_free_res;
4154         }
4155 #ifdef CONFIG_X86
4156         {
4157                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4158                 __u32 prefetch;
4159                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
4160                 prefetch |= 0x100;
4161                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
4162         }
4163 #endif
4164
4165         /* Disabling DMA prefetch and refetch for the P600.
4166          * An ASIC bug may result in accesses to invalid memory addresses.
4167          * We've disabled prefetch for some time now. Testing with XEN
4168          * kernels revealed a bug in the refetch if dom0 resides on a P600.
4169          */
4170         if(board_id == 0x3225103C) {
4171                         __u32 dma_prefetch;
4172                 __u32 dma_refetch;
4173                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
4174                 dma_prefetch |= 0x8000;
4175                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
4176                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
4177                 dma_refetch |= 0x1;
4178                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
4179         }
4180
4181 #ifdef CCISS_DEBUG
4182         printk(KERN_WARNING "Trying to put board into Performant mode\n");
4183 #endif                          /* CCISS_DEBUG */
4184         cciss_put_controller_into_performant_mode(c);
4185         return 0;
4186
4187 err_out_free_res:
4188         /*
4189          * Deliberately omit pci_disable_device(): it does something nasty to
4190          * Smart Array controllers that pci_enable_device does not undo
4191          */
4192         pci_release_regions(pdev);
4193         return err;
4194 }
4195
4196 /* Function to find the first free pointer into our hba[] array
4197  * Returns -1 if no free entries are left.
4198  */
4199 static int alloc_cciss_hba(void)
4200 {
4201         int i;
4202
4203         for (i = 0; i < MAX_CTLR; i++) {
4204                 if (!hba[i]) {
4205                         ctlr_info_t *p;
4206
4207                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4208                         if (!p)
4209                                 goto Enomem;
4210                         hba[i] = p;
4211                         return i;
4212                 }
4213         }
4214         printk(KERN_WARNING "cciss: This driver supports a maximum"
4215                " of %d controllers.\n", MAX_CTLR);
4216         return -1;
4217 Enomem:
4218         printk(KERN_ERR "cciss: out of memory.\n");
4219         return -1;
4220 }
4221
4222 static void free_hba(int n)
4223 {
4224         ctlr_info_t *h = hba[n];
4225         int i;
4226
4227         hba[n] = NULL;
4228         for (i = 0; i < h->highest_lun + 1; i++)
4229                 if (h->gendisk[i] != NULL)
4230                         put_disk(h->gendisk[i]);
4231         kfree(h);
4232 }
4233
4234 /* Send a message CDB to the firmware. */
4235 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4236 {
4237         typedef struct {
4238                 CommandListHeader_struct CommandHeader;
4239                 RequestBlock_struct Request;
4240                 ErrDescriptor_struct ErrorDescriptor;
4241         } Command;
4242         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4243         Command *cmd;
4244         dma_addr_t paddr64;
4245         uint32_t paddr32, tag;
4246         void __iomem *vaddr;
4247         int i, err;
4248
4249         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4250         if (vaddr == NULL)
4251                 return -ENOMEM;
4252
4253         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4254            CCISS commands, so they must be allocated from the lower 4GiB of
4255            memory. */
4256         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4257         if (err) {
4258                 iounmap(vaddr);
4259                 return -ENOMEM;
4260         }
4261
4262         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4263         if (cmd == NULL) {
4264                 iounmap(vaddr);
4265                 return -ENOMEM;
4266         }
4267
4268         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4269            although there's no guarantee, we assume that the address is at
4270            least 4-byte aligned (most likely, it's page-aligned). */
4271         paddr32 = paddr64;
4272
4273         cmd->CommandHeader.ReplyQueue = 0;
4274         cmd->CommandHeader.SGList = 0;
4275         cmd->CommandHeader.SGTotal = 0;
4276         cmd->CommandHeader.Tag.lower = paddr32;
4277         cmd->CommandHeader.Tag.upper = 0;
4278         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4279
4280         cmd->Request.CDBLen = 16;
4281         cmd->Request.Type.Type = TYPE_MSG;
4282         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4283         cmd->Request.Type.Direction = XFER_NONE;
4284         cmd->Request.Timeout = 0; /* Don't time out */
4285         cmd->Request.CDB[0] = opcode;
4286         cmd->Request.CDB[1] = type;
4287         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4288
4289         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4290         cmd->ErrorDescriptor.Addr.upper = 0;
4291         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4292
4293         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4294
4295         for (i = 0; i < 10; i++) {
4296                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4297                 if ((tag & ~3) == paddr32)
4298                         break;
4299                 schedule_timeout_uninterruptible(HZ);
4300         }
4301
4302         iounmap(vaddr);
4303
4304         /* we leak the DMA buffer here ... no choice since the controller could
4305            still complete the command. */
4306         if (i == 10) {
4307                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4308                         opcode, type);
4309                 return -ETIMEDOUT;
4310         }
4311
4312         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4313
4314         if (tag & 2) {
4315                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4316                         opcode, type);
4317                 return -EIO;
4318         }
4319
4320         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4321                 opcode, type);
4322         return 0;
4323 }
4324
4325 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4326 #define cciss_noop(p) cciss_message(p, 3, 0)
4327
4328 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4329 {
4330 /* the #defines are stolen from drivers/pci/msi.h. */
4331 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4332 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4333
4334         int pos;
4335         u16 control = 0;
4336
4337         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4338         if (pos) {
4339                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4340                 if (control & PCI_MSI_FLAGS_ENABLE) {
4341                         printk(KERN_INFO "cciss: resetting MSI\n");
4342                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4343                 }
4344         }
4345
4346         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4347         if (pos) {
4348                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4349                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4350                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4351                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4352                 }
4353         }
4354
4355         return 0;
4356 }
4357
4358 /* This does a hard reset of the controller using PCI power management
4359  * states. */
4360 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4361 {
4362         u16 pmcsr, saved_config_space[32];
4363         int i, pos;
4364
4365         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4366
4367         /* This is very nearly the same thing as
4368
4369            pci_save_state(pci_dev);
4370            pci_set_power_state(pci_dev, PCI_D3hot);
4371            pci_set_power_state(pci_dev, PCI_D0);
4372            pci_restore_state(pci_dev);
4373
4374            but we can't use these nice canned kernel routines on
4375            kexec, because they also check the MSI/MSI-X state in PCI
4376            configuration space and do the wrong thing when it is
4377            set/cleared.  Also, the pci_save/restore_state functions
4378            violate the ordering requirements for restoring the
4379            configuration space from the CCISS document (see the
4380            comment below).  So we roll our own .... */
4381
4382         for (i = 0; i < 32; i++)
4383                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4384
4385         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4386         if (pos == 0) {
4387                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4388                 return -ENODEV;
4389         }
4390
4391         /* Quoting from the Open CISS Specification: "The Power
4392          * Management Control/Status Register (CSR) controls the power
4393          * state of the device.  The normal operating state is D0,
4394          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4395          * the controller, place the interface device in D3 then to
4396          * D0, this causes a secondary PCI reset which will reset the
4397          * controller." */
4398
4399         /* enter the D3hot power management state */
4400         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4401         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4402         pmcsr |= PCI_D3hot;
4403         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4404
4405         schedule_timeout_uninterruptible(HZ >> 1);
4406
4407         /* enter the D0 power management state */
4408         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4409         pmcsr |= PCI_D0;
4410         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4411
4412         schedule_timeout_uninterruptible(HZ >> 1);
4413
4414         /* Restore the PCI configuration space.  The Open CISS
4415          * Specification says, "Restore the PCI Configuration
4416          * Registers, offsets 00h through 60h. It is important to
4417          * restore the command register, 16-bits at offset 04h,
4418          * last. Do not restore the configuration status register,
4419          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4420         for (i = 0; i < 32; i++) {
4421                 if (i == 2 || i == 3)
4422                         continue;
4423                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4424         }
4425         wmb();
4426         pci_write_config_word(pdev, 4, saved_config_space[2]);
4427
4428         return 0;
4429 }
4430
4431 /*
4432  *  This is it.  Find all the controllers and register them.  I really hate
4433  *  stealing all these major device numbers.
4434  *  returns the number of block devices registered.
4435  */
4436 static int __devinit cciss_init_one(struct pci_dev *pdev,
4437                                     const struct pci_device_id *ent)
4438 {
4439         int i;
4440         int j = 0;
4441         int k = 0;
4442         int rc;
4443         int dac, return_code;
4444         InquiryData_struct *inq_buff;
4445
4446         if (reset_devices) {
4447                 /* Reset the controller with a PCI power-cycle */
4448                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4449                         return -ENODEV;
4450
4451                 /* Now try to get the controller to respond to a no-op. Some
4452                    devices (notably the HP Smart Array 5i Controller) need
4453                    up to 30 seconds to respond. */
4454                 for (i=0; i<30; i++) {
4455                         if (cciss_noop(pdev) == 0)
4456                                 break;
4457
4458                         schedule_timeout_uninterruptible(HZ);
4459                 }
4460                 if (i == 30) {
4461                         printk(KERN_ERR "cciss: controller seems dead\n");
4462                         return -EBUSY;
4463                 }
4464         }
4465
4466         i = alloc_cciss_hba();
4467         if (i < 0)
4468                 return -1;
4469         hba[i]->busy_initializing = 1;
4470         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4471         INIT_HLIST_HEAD(&hba[i]->reqQ);
4472         mutex_init(&hba[i]->busy_shutting_down);
4473
4474         if (cciss_pci_init(hba[i], pdev) != 0)
4475                 goto clean_no_release_regions;
4476
4477         sprintf(hba[i]->devname, "cciss%d", i);
4478         hba[i]->ctlr = i;
4479         hba[i]->pdev = pdev;
4480
4481         init_completion(&hba[i]->scan_wait);
4482
4483         if (cciss_create_hba_sysfs_entry(hba[i]))
4484                 goto clean0;
4485
4486         /* configure PCI DMA stuff */
4487         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4488                 dac = 1;
4489         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4490                 dac = 0;
4491         else {
4492                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4493                 goto clean1;
4494         }
4495
4496         /*
4497          * register with the major number, or get a dynamic major number
4498          * by passing 0 as argument.  This is done for greater than
4499          * 8 controller support.
4500          */
4501         if (i < MAX_CTLR_ORIG)
4502                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4503         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4504         if (rc == -EBUSY || rc == -EINVAL) {
4505                 printk(KERN_ERR
4506                        "cciss:  Unable to get major number %d for %s "
4507                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4508                 goto clean1;
4509         } else {
4510                 if (i >= MAX_CTLR_ORIG)
4511                         hba[i]->major = rc;
4512         }
4513
4514         /* make sure the board interrupts are off */
4515         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4516         if (hba[i]->msi_vector || hba[i]->msix_vector) {
4517                 if (request_irq(hba[i]->intr[PERF_MODE_INT],
4518                                 do_cciss_msix_intr,
4519                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4520                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4521                                hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4522                         goto clean2;
4523                 }
4524         } else {
4525                 if (request_irq(hba[i]->intr[PERF_MODE_INT], do_cciss_intx,
4526                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4527                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4528                                hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4529                         goto clean2;
4530                 }
4531         }
4532
4533         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4534                hba[i]->devname, pdev->device, pci_name(pdev),
4535                hba[i]->intr[PERF_MODE_INT], dac ? "" : " not");
4536
4537         hba[i]->cmd_pool_bits =
4538             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4539                         * sizeof(unsigned long), GFP_KERNEL);
4540         hba[i]->cmd_pool = (CommandList_struct *)
4541             pci_alloc_consistent(hba[i]->pdev,
4542                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4543                     &(hba[i]->cmd_pool_dhandle));
4544         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4545             pci_alloc_consistent(hba[i]->pdev,
4546                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4547                     &(hba[i]->errinfo_pool_dhandle));
4548         if ((hba[i]->cmd_pool_bits == NULL)
4549             || (hba[i]->cmd_pool == NULL)
4550             || (hba[i]->errinfo_pool == NULL)) {
4551                 printk(KERN_ERR "cciss: out of memory");
4552                 goto clean4;
4553         }
4554
4555         /* Need space for temp scatter list */
4556         hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4557                                                 sizeof(struct scatterlist *),
4558                                                 GFP_KERNEL);
4559         for (k = 0; k < hba[i]->nr_cmds; k++) {
4560                 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4561                                                         hba[i]->maxsgentries,
4562                                                         GFP_KERNEL);
4563                 if (hba[i]->scatter_list[k] == NULL) {
4564                         printk(KERN_ERR "cciss%d: could not allocate "
4565                                 "s/g lists\n", i);
4566                         goto clean4;
4567                 }
4568         }
4569         hba[i]->cmd_sg_list = cciss_allocate_sg_chain_blocks(hba[i],
4570                 hba[i]->chainsize, hba[i]->nr_cmds);
4571         if (!hba[i]->cmd_sg_list && hba[i]->chainsize > 0)
4572                 goto clean4;
4573
4574         spin_lock_init(&hba[i]->lock);
4575
4576         /* Initialize the pdev driver private data.
4577            have it point to hba[i].  */
4578         pci_set_drvdata(pdev, hba[i]);
4579         /* command and error info recs zeroed out before
4580            they are used */
4581         memset(hba[i]->cmd_pool_bits, 0,
4582                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4583                         * sizeof(unsigned long));
4584
4585         hba[i]->num_luns = 0;
4586         hba[i]->highest_lun = -1;
4587         for (j = 0; j < CISS_MAX_LUN; j++) {
4588                 hba[i]->drv[j] = NULL;
4589                 hba[i]->gendisk[j] = NULL;
4590         }
4591
4592         cciss_scsi_setup(i);
4593
4594         /* Turn the interrupts on so we can service requests */
4595         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4596
4597         /* Get the firmware version */
4598         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4599         if (inq_buff == NULL) {
4600                 printk(KERN_ERR "cciss: out of memory\n");
4601                 goto clean4;
4602         }
4603
4604         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4605                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4606         if (return_code == IO_OK) {
4607                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4608                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4609                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4610                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4611         } else {         /* send command failed */
4612                 printk(KERN_WARNING "cciss: unable to determine firmware"
4613                         " version of controller\n");
4614         }
4615         kfree(inq_buff);
4616
4617         cciss_procinit(i);
4618
4619         hba[i]->cciss_max_sectors = 8192;
4620
4621         rebuild_lun_table(hba[i], 1, 0);
4622         hba[i]->busy_initializing = 0;
4623         return 1;
4624
4625 clean4:
4626         kfree(hba[i]->cmd_pool_bits);
4627         /* Free up sg elements */
4628         for (k = 0; k < hba[i]->nr_cmds; k++)
4629                 kfree(hba[i]->scatter_list[k]);
4630         kfree(hba[i]->scatter_list);
4631         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4632         if (hba[i]->cmd_pool)
4633                 pci_free_consistent(hba[i]->pdev,
4634                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4635                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4636         if (hba[i]->errinfo_pool)
4637                 pci_free_consistent(hba[i]->pdev,
4638                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4639                                     hba[i]->errinfo_pool,
4640                                     hba[i]->errinfo_pool_dhandle);
4641         free_irq(hba[i]->intr[PERF_MODE_INT], hba[i]);
4642 clean2:
4643         unregister_blkdev(hba[i]->major, hba[i]->devname);
4644 clean1:
4645         cciss_destroy_hba_sysfs_entry(hba[i]);
4646 clean0:
4647         pci_release_regions(pdev);
4648 clean_no_release_regions:
4649         hba[i]->busy_initializing = 0;
4650
4651         /*
4652          * Deliberately omit pci_disable_device(): it does something nasty to
4653          * Smart Array controllers that pci_enable_device does not undo
4654          */
4655         pci_set_drvdata(pdev, NULL);
4656         free_hba(i);
4657         return -1;
4658 }
4659
4660 static void cciss_shutdown(struct pci_dev *pdev)
4661 {
4662         ctlr_info_t *h;
4663         char *flush_buf;
4664         int return_code;
4665
4666         h = pci_get_drvdata(pdev);
4667         flush_buf = kzalloc(4, GFP_KERNEL);
4668         if (!flush_buf) {
4669                 printk(KERN_WARNING
4670                         "cciss:%d cache not flushed, out of memory.\n",
4671                         h->ctlr);
4672                 return;
4673         }
4674         /* write all data in the battery backed cache to disk */
4675         memset(flush_buf, 0, 4);
4676         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4677                 4, 0, CTLR_LUNID, TYPE_CMD);
4678         kfree(flush_buf);
4679         if (return_code != IO_OK)
4680                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4681                         h->ctlr);
4682         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4683         free_irq(h->intr[PERF_MODE_INT], h);
4684 }
4685
4686 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4687 {
4688         ctlr_info_t *tmp_ptr;
4689         int i, j;
4690
4691         if (pci_get_drvdata(pdev) == NULL) {
4692                 printk(KERN_ERR "cciss: Unable to remove device \n");
4693                 return;
4694         }
4695
4696         tmp_ptr = pci_get_drvdata(pdev);
4697         i = tmp_ptr->ctlr;
4698         if (hba[i] == NULL) {
4699                 printk(KERN_ERR "cciss: device appears to "
4700                        "already be removed \n");
4701                 return;
4702         }
4703
4704         mutex_lock(&hba[i]->busy_shutting_down);
4705
4706         remove_from_scan_list(hba[i]);
4707         remove_proc_entry(hba[i]->devname, proc_cciss);
4708         unregister_blkdev(hba[i]->major, hba[i]->devname);
4709
4710         /* remove it from the disk list */
4711         for (j = 0; j < CISS_MAX_LUN; j++) {
4712                 struct gendisk *disk = hba[i]->gendisk[j];
4713                 if (disk) {
4714                         struct request_queue *q = disk->queue;
4715
4716                         if (disk->flags & GENHD_FL_UP) {
4717                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4718                                 del_gendisk(disk);
4719                         }
4720                         if (q)
4721                                 blk_cleanup_queue(q);
4722                 }
4723         }
4724
4725 #ifdef CONFIG_CISS_SCSI_TAPE
4726         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4727 #endif
4728
4729         cciss_shutdown(pdev);
4730
4731 #ifdef CONFIG_PCI_MSI
4732         if (hba[i]->msix_vector)
4733                 pci_disable_msix(hba[i]->pdev);
4734         else if (hba[i]->msi_vector)
4735                 pci_disable_msi(hba[i]->pdev);
4736 #endif                          /* CONFIG_PCI_MSI */
4737
4738         iounmap(hba[i]->vaddr);
4739
4740         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4741                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4742         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4743                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4744         kfree(hba[i]->cmd_pool_bits);
4745         /* Free up sg elements */
4746         for (j = 0; j < hba[i]->nr_cmds; j++)
4747                 kfree(hba[i]->scatter_list[j]);
4748         kfree(hba[i]->scatter_list);
4749         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4750         /*
4751          * Deliberately omit pci_disable_device(): it does something nasty to
4752          * Smart Array controllers that pci_enable_device does not undo
4753          */
4754         pci_release_regions(pdev);
4755         pci_set_drvdata(pdev, NULL);
4756         cciss_destroy_hba_sysfs_entry(hba[i]);
4757         mutex_unlock(&hba[i]->busy_shutting_down);
4758         free_hba(i);
4759 }
4760
4761 static struct pci_driver cciss_pci_driver = {
4762         .name = "cciss",
4763         .probe = cciss_init_one,
4764         .remove = __devexit_p(cciss_remove_one),
4765         .id_table = cciss_pci_device_id,        /* id_table */
4766         .shutdown = cciss_shutdown,
4767 };
4768
4769 /*
4770  *  This is it.  Register the PCI driver information for the cards we control
4771  *  the OS will call our registered routines when it finds one of our cards.
4772  */
4773 static int __init cciss_init(void)
4774 {
4775         int err;
4776
4777         /*
4778          * The hardware requires that commands are aligned on a 64-bit
4779          * boundary. Given that we use pci_alloc_consistent() to allocate an
4780          * array of them, the size must be a multiple of 8 bytes.
4781          */
4782         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4783         printk(KERN_INFO DRIVER_NAME "\n");
4784
4785         err = bus_register(&cciss_bus_type);
4786         if (err)
4787                 return err;
4788
4789         /* Start the scan thread */
4790         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4791         if (IS_ERR(cciss_scan_thread)) {
4792                 err = PTR_ERR(cciss_scan_thread);
4793                 goto err_bus_unregister;
4794         }
4795
4796         /* Register for our PCI devices */
4797         err = pci_register_driver(&cciss_pci_driver);
4798         if (err)
4799                 goto err_thread_stop;
4800
4801         return err;
4802
4803 err_thread_stop:
4804         kthread_stop(cciss_scan_thread);
4805 err_bus_unregister:
4806         bus_unregister(&cciss_bus_type);
4807
4808         return err;
4809 }
4810
4811 static void __exit cciss_cleanup(void)
4812 {
4813         int i;
4814
4815         pci_unregister_driver(&cciss_pci_driver);
4816         /* double check that all controller entrys have been removed */
4817         for (i = 0; i < MAX_CTLR; i++) {
4818                 if (hba[i] != NULL) {
4819                         printk(KERN_WARNING "cciss: had to remove"
4820                                " controller %d\n", i);
4821                         cciss_remove_one(hba[i]->pdev);
4822                 }
4823         }
4824         kthread_stop(cciss_scan_thread);
4825         remove_proc_entry("driver/cciss", NULL);
4826         bus_unregister(&cciss_bus_type);
4827 }
4828
4829 module_init(cciss_init);
4830 module_exit(cciss_cleanup);