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/pci-aspm.h>
  28 #include <linux/kernel.h>
  29 #include <linux/slab.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/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 <linux/bitmap.h>
  45 #include <linux/io.h>
  46 #include <asm/uaccess.h>
  47
  48 #include <linux/dma-mapping.h>
  49 #include <linux/blkdev.h>
  50 #include <linux/genhd.h>
  51 #include <linux/completion.h>
  52 #include <scsi/scsi.h>
  53 #include <scsi/sg.h>
  54 #include <scsi/scsi_ioctl.h>
  55 #include <linux/cdrom.h>
  56 #include <linux/scatterlist.h>
  57 #include <linux/kthread.h>
  58
  59 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  60 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  61 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  62
  63 /* Embedded module documentation macros - see modules.h */
  64 MODULE_AUTHOR("Hewlett-Packard Company");
  65 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  66 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  67 MODULE_VERSION("3.6.26");
  68 MODULE_LICENSE("GPL");
  69 static int cciss_tape_cmds = 6;
  70 module_param(cciss_tape_cmds, int, 0644);
  71 MODULE_PARM_DESC(cciss_tape_cmds,
  72         "number of commands to allocate for tape devices (default: 6)");
  73 static int cciss_simple_mode;
  74 module_param(cciss_simple_mode, int, S_IRUGO|S_IWUSR);
  75 MODULE_PARM_DESC(cciss_simple_mode,
  76         "Use 'simple mode' rather than 'performant mode'");
  77
  78 static DEFINE_MUTEX(cciss_mutex);
  79 static struct proc_dir_entry *proc_cciss;
  80
  81 #include "cciss_cmd.h"
  82 #include "cciss.h"
  83 #include <linux/cciss_ioctl.h>
  84
  85 /* define the PCI info for the cards we can control */
  86 static const struct pci_device_id cciss_pci_device_id[] = {
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  90         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  91         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  92         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  93         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  94         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  95         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 107         {0,}
 108 };
 109
 110 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 111
 112 /*  board_id = Subsystem Device ID & Vendor ID
 113  *  product = Marketing Name for the board
 114  *  access = Address of the struct of function pointers
 115  */
 116 static struct board_type products[] = {
 117         {0x40700E11, "Smart Array 5300", &SA5_access},
 118         {0x40800E11, "Smart Array 5i", &SA5B_access},
 119         {0x40820E11, "Smart Array 532", &SA5B_access},
 120         {0x40830E11, "Smart Array 5312", &SA5B_access},
 121         {0x409A0E11, "Smart Array 641", &SA5_access},
 122         {0x409B0E11, "Smart Array 642", &SA5_access},
 123         {0x409C0E11, "Smart Array 6400", &SA5_access},
 124         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 125         {0x40910E11, "Smart Array 6i", &SA5_access},
 126         {0x3225103C, "Smart Array P600", &SA5_access},
 127         {0x3223103C, "Smart Array P800", &SA5_access},
 128         {0x3234103C, "Smart Array P400", &SA5_access},
 129         {0x3235103C, "Smart Array P400i", &SA5_access},
 130         {0x3211103C, "Smart Array E200i", &SA5_access},
 131         {0x3212103C, "Smart Array E200", &SA5_access},
 132         {0x3213103C, "Smart Array E200i", &SA5_access},
 133         {0x3214103C, "Smart Array E200i", &SA5_access},
 134         {0x3215103C, "Smart Array E200i", &SA5_access},
 135         {0x3237103C, "Smart Array E500", &SA5_access},
 136         {0x3223103C, "Smart Array P800", &SA5_access},
 137         {0x3234103C, "Smart Array P400", &SA5_access},
 138         {0x323D103C, "Smart Array P700m", &SA5_access},
 139 };
 140
 141 /* How long to wait (in milliseconds) for board to go into simple mode */
 142 #define MAX_CONFIG_WAIT 30000
 143 #define MAX_IOCTL_CONFIG_WAIT 1000
 144
 145 /*define how many times we will try a command because of bus resets */
 146 #define MAX_CMD_RETRIES 3
 147
 148 #define MAX_CTLR        32
 149
 150 /* Originally cciss driver only supports 8 major numbers */
 151 #define MAX_CTLR_ORIG   8
 152
 153 static ctlr_info_t *hba[MAX_CTLR];
 154
 155 static struct task_struct *cciss_scan_thread;
 156 static DEFINE_MUTEX(scan_mutex);
 157 static LIST_HEAD(scan_q);
 158
 159 static void do_cciss_request(struct request_queue *q);
 160 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 161 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 162 static int cciss_open(struct block_device *bdev, fmode_t mode);
 163 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 164 static int cciss_release(struct gendisk *disk, fmode_t mode);
 165 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 166                     unsigned int cmd, unsigned long arg);
 167 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 168                        unsigned int cmd, unsigned long arg);
 169 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 170
 171 static int cciss_revalidate(struct gendisk *disk);
 172 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 173 static int deregister_disk(ctlr_info_t *h, int drv_index,
 174                            int clear_all, int via_ioctl);
 175
 176 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 177                         sector_t *total_size, unsigned int *block_size);
 178 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 179                         sector_t *total_size, unsigned int *block_size);
 180 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 181                         sector_t total_size,
 182                         unsigned int block_size, InquiryData_struct *inq_buff,
 183                                    drive_info_struct *drv);
 184 static void cciss_interrupt_mode(ctlr_info_t *);
 185 static int cciss_enter_simple_mode(struct ctlr_info *h);
 186 static void start_io(ctlr_info_t *h);
 187 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 188                         __u8 page_code, unsigned char scsi3addr[],
 189                         int cmd_type);
 190 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 191         int attempt_retry);
 192 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 193
 194 static int add_to_scan_list(struct ctlr_info *h);
 195 static int scan_thread(void *data);
 196 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 197 static void cciss_hba_release(struct device *dev);
 198 static void cciss_device_release(struct device *dev);
 199 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 200 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 201 static inline u32 next_command(ctlr_info_t *h);
 202 static int cciss_find_cfg_addrs(struct pci_dev *pdev, void __iomem *vaddr,
 203                                 u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 204                                 u64 *cfg_offset);
 205 static int cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 206                                      unsigned long *memory_bar);
 207 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag);
 208 static int write_driver_ver_to_cfgtable(CfgTable_struct __iomem *cfgtable);
 209
 210 /* performant mode helper functions */
 211 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 212                                 int *bucket_map);
 213 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 214
 215 #ifdef CONFIG_PROC_FS
 216 static void cciss_procinit(ctlr_info_t *h);
 217 #else
 218 static void cciss_procinit(ctlr_info_t *h)
 219 {
 220 }
 221 #endif                          /* CONFIG_PROC_FS */
 222
 223 #ifdef CONFIG_COMPAT
 224 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 225                               unsigned, unsigned long);
 226 #endif
 227
 228 static const struct block_device_operations cciss_fops = {
 229         .owner = THIS_MODULE,
 230         .open = cciss_unlocked_open,
 231         .release = cciss_release,
 232         .ioctl = do_ioctl,
 233         .getgeo = cciss_getgeo,
 234 #ifdef CONFIG_COMPAT
 235         .compat_ioctl = cciss_compat_ioctl,
 236 #endif
 237         .revalidate_disk = cciss_revalidate,
 238 };
 239
 240 /* set_performant_mode: Modify the tag for cciss performant
 241  * set bit 0 for pull model, bits 3-1 for block fetch
 242  * register number
 243  */
 244 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 245 {
 246         if (likely(h->transMethod & CFGTBL_Trans_Performant))
 247                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 248 }
 249
 250 /*
 251  * Enqueuing and dequeuing functions for cmdlists.
 252  */
 253 static inline void addQ(struct list_head *list, CommandList_struct *c)
 254 {
 255         list_add_tail(&c->list, list);
 256 }
 257
 258 static inline void removeQ(CommandList_struct *c)
 259 {
 260         /*
 261          * After kexec/dump some commands might still
 262          * be in flight, which the firmware will try
 263          * to complete. Resetting the firmware doesn't work
 264          * with old fw revisions, so we have to mark
 265          * them off as 'stale' to prevent the driver from
 266          * falling over.
 267          */
 268         if (WARN_ON(list_empty(&c->list))) {
 269                 c->cmd_type = CMD_MSG_STALE;
 270                 return;
 271         }
 272
 273         list_del_init(&c->list);
 274 }
 275
 276 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 277         CommandList_struct *c)
 278 {
 279         unsigned long flags;
 280         set_performant_mode(h, c);
 281         spin_lock_irqsave(&h->lock, flags);
 282         addQ(&h->reqQ, c);
 283         h->Qdepth++;
 284         if (h->Qdepth > h->maxQsinceinit)
 285                 h->maxQsinceinit = h->Qdepth;
 286         start_io(h);
 287         spin_unlock_irqrestore(&h->lock, flags);
 288 }
 289
 290 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 291         int nr_cmds)
 292 {
 293         int i;
 294
 295         if (!cmd_sg_list)
 296                 return;
 297         for (i = 0; i < nr_cmds; i++) {
 298                 kfree(cmd_sg_list[i]);
 299                 cmd_sg_list[i] = NULL;
 300         }
 301         kfree(cmd_sg_list);
 302 }
 303
 304 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 305         ctlr_info_t *h, int chainsize, int nr_cmds)
 306 {
 307         int j;
 308         SGDescriptor_struct **cmd_sg_list;
 309
 310         if (chainsize <= 0)
 311                 return NULL;
 312
 313         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 314         if (!cmd_sg_list)
 315                 return NULL;
 316
 317         /* Build up chain blocks for each command */
 318         for (j = 0; j < nr_cmds; j++) {
 319                 /* Need a block of chainsized s/g elements. */
 320                 cmd_sg_list[j] = kmalloc((chainsize *
 321                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 322                 if (!cmd_sg_list[j]) {
 323                         dev_err(&h->pdev->dev, "Cannot get memory "
 324                                 "for s/g chains.\n");
 325                         goto clean;
 326                 }
 327         }
 328         return cmd_sg_list;
 329 clean:
 330         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 331         return NULL;
 332 }
 333
 334 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 335 {
 336         SGDescriptor_struct *chain_sg;
 337         u64bit temp64;
 338
 339         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 340                 return;
 341
 342         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 343         temp64.val32.lower = chain_sg->Addr.lower;
 344         temp64.val32.upper = chain_sg->Addr.upper;
 345         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 346 }
 347
 348 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 349         SGDescriptor_struct *chain_block, int len)
 350 {
 351         SGDescriptor_struct *chain_sg;
 352         u64bit temp64;
 353
 354         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 355         chain_sg->Ext = CCISS_SG_CHAIN;
 356         chain_sg->Len = len;
 357         temp64.val = pci_map_single(h->pdev, chain_block, len,
 358                                 PCI_DMA_TODEVICE);
 359         chain_sg->Addr.lower = temp64.val32.lower;
 360         chain_sg->Addr.upper = temp64.val32.upper;
 361 }
 362
 363 #include "cciss_scsi.c"         /* For SCSI tape support */
 364
 365 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 366         "UNKNOWN"
 367 };
 368 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 369
 370 #ifdef CONFIG_PROC_FS
 371
 372 /*
 373  * Report information about this controller.
 374  */
 375 #define ENG_GIG 1000000000
 376 #define ENG_GIG_FACTOR (ENG_GIG/512)
 377 #define ENGAGE_SCSI     "engage scsi"
 378
 379 static void cciss_seq_show_header(struct seq_file *seq)
 380 {
 381         ctlr_info_t *h = seq->private;
 382
 383         seq_printf(seq, "%s: HP %s Controller\n"
 384                 "Board ID: 0x%08lx\n"
 385                 "Firmware Version: %c%c%c%c\n"
 386                 "IRQ: %d\n"
 387                 "Logical drives: %d\n"
 388                 "Current Q depth: %d\n"
 389                 "Current # commands on controller: %d\n"
 390                 "Max Q depth since init: %d\n"
 391                 "Max # commands on controller since init: %d\n"
 392                 "Max SG entries since init: %d\n",
 393                 h->devname,
 394                 h->product_name,
 395                 (unsigned long)h->board_id,
 396                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 397                 h->firm_ver[3], (unsigned int)h->intr[h->intr_mode],
 398                 h->num_luns,
 399                 h->Qdepth, h->commands_outstanding,
 400                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 401
 402 #ifdef CONFIG_CISS_SCSI_TAPE
 403         cciss_seq_tape_report(seq, h);
 404 #endif /* CONFIG_CISS_SCSI_TAPE */
 405 }
 406
 407 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 408 {
 409         ctlr_info_t *h = seq->private;
 410         unsigned long flags;
 411
 412         /* prevent displaying bogus info during configuration
 413          * or deconfiguration of a logical volume
 414          */
 415         spin_lock_irqsave(&h->lock, flags);
 416         if (h->busy_configuring) {
 417                 spin_unlock_irqrestore(&h->lock, flags);
 418                 return ERR_PTR(-EBUSY);
 419         }
 420         h->busy_configuring = 1;
 421         spin_unlock_irqrestore(&h->lock, flags);
 422
 423         if (*pos == 0)
 424                 cciss_seq_show_header(seq);
 425
 426         return pos;
 427 }
 428
 429 static int cciss_seq_show(struct seq_file *seq, void *v)
 430 {
 431         sector_t vol_sz, vol_sz_frac;
 432         ctlr_info_t *h = seq->private;
 433         unsigned ctlr = h->ctlr;
 434         loff_t *pos = v;
 435         drive_info_struct *drv = h->drv[*pos];
 436
 437         if (*pos > h->highest_lun)
 438                 return 0;
 439
 440         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 441                 return 0;
 442
 443         if (drv->heads == 0)
 444                 return 0;
 445
 446         vol_sz = drv->nr_blocks;
 447         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 448         vol_sz_frac *= 100;
 449         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 450
 451         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 452                 drv->raid_level = RAID_UNKNOWN;
 453         seq_printf(seq, "cciss/c%dd%d:"
 454                         "\t%4u.%02uGB\tRAID %s\n",
 455                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 456                         raid_label[drv->raid_level]);
 457         return 0;
 458 }
 459
 460 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 461 {
 462         ctlr_info_t *h = seq->private;
 463
 464         if (*pos > h->highest_lun)
 465                 return NULL;
 466         *pos += 1;
 467
 468         return pos;
 469 }
 470
 471 static void cciss_seq_stop(struct seq_file *seq, void *v)
 472 {
 473         ctlr_info_t *h = seq->private;
 474
 475         /* Only reset h->busy_configuring if we succeeded in setting
 476          * it during cciss_seq_start. */
 477         if (v == ERR_PTR(-EBUSY))
 478                 return;
 479
 480         h->busy_configuring = 0;
 481 }
 482
 483 static const struct seq_operations cciss_seq_ops = {
 484         .start = cciss_seq_start,
 485         .show  = cciss_seq_show,
 486         .next  = cciss_seq_next,
 487         .stop  = cciss_seq_stop,
 488 };
 489
 490 static int cciss_seq_open(struct inode *inode, struct file *file)
 491 {
 492         int ret = seq_open(file, &cciss_seq_ops);
 493         struct seq_file *seq = file->private_data;
 494
 495         if (!ret)
 496                 seq->private = PDE(inode)->data;
 497
 498         return ret;
 499 }
 500
 501 static ssize_t
 502 cciss_proc_write(struct file *file, const char __user *buf,
 503                  size_t length, loff_t *ppos)
 504 {
 505         int err;
 506         char *buffer;
 507
 508 #ifndef CONFIG_CISS_SCSI_TAPE
 509         return -EINVAL;
 510 #endif
 511
 512         if (!buf || length > PAGE_SIZE - 1)
 513                 return -EINVAL;
 514
 515         buffer = (char *)__get_free_page(GFP_KERNEL);
 516         if (!buffer)
 517                 return -ENOMEM;
 518
 519         err = -EFAULT;
 520         if (copy_from_user(buffer, buf, length))
 521                 goto out;
 522         buffer[length] = '\0';
 523
 524 #ifdef CONFIG_CISS_SCSI_TAPE
 525         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 526                 struct seq_file *seq = file->private_data;
 527                 ctlr_info_t *h = seq->private;
 528
 529                 err = cciss_engage_scsi(h);
 530                 if (err == 0)
 531                         err = length;
 532         } else
 533 #endif /* CONFIG_CISS_SCSI_TAPE */
 534                 err = -EINVAL;
 535         /* might be nice to have "disengage" too, but it's not
 536            safely possible. (only 1 module use count, lock issues.) */
 537
 538 out:
 539         free_page((unsigned long)buffer);
 540         return err;
 541 }
 542
 543 static const struct file_operations cciss_proc_fops = {
 544         .owner   = THIS_MODULE,
 545         .open    = cciss_seq_open,
 546         .read    = seq_read,
 547         .llseek  = seq_lseek,
 548         .release = seq_release,
 549         .write   = cciss_proc_write,
 550 };
 551
 552 static void cciss_procinit(ctlr_info_t *h)
 553 {
 554         struct proc_dir_entry *pde;
 555
 556         if (proc_cciss == NULL)
 557                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 558         if (!proc_cciss)
 559                 return;
 560         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 561                                         S_IROTH, proc_cciss,
 562                                         &cciss_proc_fops, h);
 563 }
 564 #endif                          /* CONFIG_PROC_FS */
 565
 566 #define MAX_PRODUCT_NAME_LEN 19
 567
 568 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 569 #define to_drv(n) container_of(n, drive_info_struct, dev)
 570
 571 /* List of controllers which cannot be hard reset on kexec with reset_devices */
 572 static u32 unresettable_controller[] = {
 573         0x324a103C, /* Smart Array P712m */
 574         0x324b103C, /* SmartArray P711m */
 575         0x3223103C, /* Smart Array P800 */
 576         0x3234103C, /* Smart Array P400 */
 577         0x3235103C, /* Smart Array P400i */
 578         0x3211103C, /* Smart Array E200i */
 579         0x3212103C, /* Smart Array E200 */
 580         0x3213103C, /* Smart Array E200i */
 581         0x3214103C, /* Smart Array E200i */
 582         0x3215103C, /* Smart Array E200i */
 583         0x3237103C, /* Smart Array E500 */
 584         0x323D103C, /* Smart Array P700m */
 585         0x409C0E11, /* Smart Array 6400 */
 586         0x409D0E11, /* Smart Array 6400 EM */
 587 };
 588
 589 /* List of controllers which cannot even be soft reset */
 590 static u32 soft_unresettable_controller[] = {
 591         0x409C0E11, /* Smart Array 6400 */
 592         0x409D0E11, /* Smart Array 6400 EM */
 593 };
 594
 595 static int ctlr_is_hard_resettable(u32 board_id)
 596 {
 597         int i;
 598
 599         for (i = 0; i < ARRAY_SIZE(unresettable_controller); i++)
 600                 if (unresettable_controller[i] == board_id)
 601                         return 0;
 602         return 1;
 603 }
 604
 605 static int ctlr_is_soft_resettable(u32 board_id)
 606 {
 607         int i;
 608
 609         for (i = 0; i < ARRAY_SIZE(soft_unresettable_controller); i++)
 610                 if (soft_unresettable_controller[i] == board_id)
 611                         return 0;
 612         return 1;
 613 }
 614
 615 static int ctlr_is_resettable(u32 board_id)
 616 {
 617         return ctlr_is_hard_resettable(board_id) ||
 618                 ctlr_is_soft_resettable(board_id);
 619 }
 620
 621 static ssize_t host_show_resettable(struct device *dev,
 622                                     struct device_attribute *attr,
 623                                     char *buf)
 624 {
 625         struct ctlr_info *h = to_hba(dev);
 626
 627         return snprintf(buf, 20, "%d\n", ctlr_is_resettable(h->board_id));
 628 }
 629 static DEVICE_ATTR(resettable, S_IRUGO, host_show_resettable, NULL);
 630
 631 static ssize_t host_store_rescan(struct device *dev,
 632                                  struct device_attribute *attr,
 633                                  const char *buf, size_t count)
 634 {
 635         struct ctlr_info *h = to_hba(dev);
 636
 637         add_to_scan_list(h);
 638         wake_up_process(cciss_scan_thread);
 639         wait_for_completion_interruptible(&h->scan_wait);
 640
 641         return count;
 642 }
 643 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 644
 645 static ssize_t host_show_transport_mode(struct device *dev,
 646                                  struct device_attribute *attr,
 647                                  char *buf)
 648 {
 649         struct ctlr_info *h = to_hba(dev);
 650
 651         return snprintf(buf, 20, "%s\n",
 652                 h->transMethod & CFGTBL_Trans_Performant ?
 653                         "performant" : "simple");
 654 }
 655 static DEVICE_ATTR(transport_mode, S_IRUGO, host_show_transport_mode, NULL);
 656
 657 static ssize_t dev_show_unique_id(struct device *dev,
 658                                  struct device_attribute *attr,
 659                                  char *buf)
 660 {
 661         drive_info_struct *drv = to_drv(dev);
 662         struct ctlr_info *h = to_hba(drv->dev.parent);
 663         __u8 sn[16];
 664         unsigned long flags;
 665         int ret = 0;
 666
 667         spin_lock_irqsave(&h->lock, flags);
 668         if (h->busy_configuring)
 669                 ret = -EBUSY;
 670         else
 671                 memcpy(sn, drv->serial_no, sizeof(sn));
 672         spin_unlock_irqrestore(&h->lock, flags);
 673
 674         if (ret)
 675                 return ret;
 676         else
 677                 return snprintf(buf, 16 * 2 + 2,
 678                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 679                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 680                                 sn[0], sn[1], sn[2], sn[3],
 681                                 sn[4], sn[5], sn[6], sn[7],
 682                                 sn[8], sn[9], sn[10], sn[11],
 683                                 sn[12], sn[13], sn[14], sn[15]);
 684 }
 685 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 686
 687 static ssize_t dev_show_vendor(struct device *dev,
 688                                struct device_attribute *attr,
 689                                char *buf)
 690 {
 691         drive_info_struct *drv = to_drv(dev);
 692         struct ctlr_info *h = to_hba(drv->dev.parent);
 693         char vendor[VENDOR_LEN + 1];
 694         unsigned long flags;
 695         int ret = 0;
 696
 697         spin_lock_irqsave(&h->lock, flags);
 698         if (h->busy_configuring)
 699                 ret = -EBUSY;
 700         else
 701                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 702         spin_unlock_irqrestore(&h->lock, flags);
 703
 704         if (ret)
 705                 return ret;
 706         else
 707                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 708 }
 709 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 710
 711 static ssize_t dev_show_model(struct device *dev,
 712                               struct device_attribute *attr,
 713                               char *buf)
 714 {
 715         drive_info_struct *drv = to_drv(dev);
 716         struct ctlr_info *h = to_hba(drv->dev.parent);
 717         char model[MODEL_LEN + 1];
 718         unsigned long flags;
 719         int ret = 0;
 720
 721         spin_lock_irqsave(&h->lock, flags);
 722         if (h->busy_configuring)
 723                 ret = -EBUSY;
 724         else
 725                 memcpy(model, drv->model, MODEL_LEN + 1);
 726         spin_unlock_irqrestore(&h->lock, flags);
 727
 728         if (ret)
 729                 return ret;
 730         else
 731                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 732 }
 733 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 734
 735 static ssize_t dev_show_rev(struct device *dev,
 736                             struct device_attribute *attr,
 737                             char *buf)
 738 {
 739         drive_info_struct *drv = to_drv(dev);
 740         struct ctlr_info *h = to_hba(drv->dev.parent);
 741         char rev[REV_LEN + 1];
 742         unsigned long flags;
 743         int ret = 0;
 744
 745         spin_lock_irqsave(&h->lock, flags);
 746         if (h->busy_configuring)
 747                 ret = -EBUSY;
 748         else
 749                 memcpy(rev, drv->rev, REV_LEN + 1);
 750         spin_unlock_irqrestore(&h->lock, flags);
 751
 752         if (ret)
 753                 return ret;
 754         else
 755                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 756 }
 757 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 758
 759 static ssize_t cciss_show_lunid(struct device *dev,
 760                                 struct device_attribute *attr, char *buf)
 761 {
 762         drive_info_struct *drv = to_drv(dev);
 763         struct ctlr_info *h = to_hba(drv->dev.parent);
 764         unsigned long flags;
 765         unsigned char lunid[8];
 766
 767         spin_lock_irqsave(&h->lock, flags);
 768         if (h->busy_configuring) {
 769                 spin_unlock_irqrestore(&h->lock, flags);
 770                 return -EBUSY;
 771         }
 772         if (!drv->heads) {
 773                 spin_unlock_irqrestore(&h->lock, flags);
 774                 return -ENOTTY;
 775         }
 776         memcpy(lunid, drv->LunID, sizeof(lunid));
 777         spin_unlock_irqrestore(&h->lock, flags);
 778         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 779                 lunid[0], lunid[1], lunid[2], lunid[3],
 780                 lunid[4], lunid[5], lunid[6], lunid[7]);
 781 }
 782 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 783
 784 static ssize_t cciss_show_raid_level(struct device *dev,
 785                                      struct device_attribute *attr, char *buf)
 786 {
 787         drive_info_struct *drv = to_drv(dev);
 788         struct ctlr_info *h = to_hba(drv->dev.parent);
 789         int raid;
 790         unsigned long flags;
 791
 792         spin_lock_irqsave(&h->lock, flags);
 793         if (h->busy_configuring) {
 794                 spin_unlock_irqrestore(&h->lock, flags);
 795                 return -EBUSY;
 796         }
 797         raid = drv->raid_level;
 798         spin_unlock_irqrestore(&h->lock, flags);
 799         if (raid < 0 || raid > RAID_UNKNOWN)
 800                 raid = RAID_UNKNOWN;
 801
 802         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 803                         raid_label[raid]);
 804 }
 805 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 806
 807 static ssize_t cciss_show_usage_count(struct device *dev,
 808                                       struct device_attribute *attr, char *buf)
 809 {
 810         drive_info_struct *drv = to_drv(dev);
 811         struct ctlr_info *h = to_hba(drv->dev.parent);
 812         unsigned long flags;
 813         int count;
 814
 815         spin_lock_irqsave(&h->lock, flags);
 816         if (h->busy_configuring) {
 817                 spin_unlock_irqrestore(&h->lock, flags);
 818                 return -EBUSY;
 819         }
 820         count = drv->usage_count;
 821         spin_unlock_irqrestore(&h->lock, flags);
 822         return snprintf(buf, 20, "%d\n", count);
 823 }
 824 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 825
 826 static struct attribute *cciss_host_attrs[] = {
 827         &dev_attr_rescan.attr,
 828         &dev_attr_resettable.attr,
 829         &dev_attr_transport_mode.attr,
 830         NULL
 831 };
 832
 833 static struct attribute_group cciss_host_attr_group = {
 834         .attrs = cciss_host_attrs,
 835 };
 836
 837 static const struct attribute_group *cciss_host_attr_groups[] = {
 838         &cciss_host_attr_group,
 839         NULL
 840 };
 841
 842 static struct device_type cciss_host_type = {
 843         .name           = "cciss_host",
 844         .groups         = cciss_host_attr_groups,
 845         .release        = cciss_hba_release,
 846 };
 847
 848 static struct attribute *cciss_dev_attrs[] = {
 849         &dev_attr_unique_id.attr,
 850         &dev_attr_model.attr,
 851         &dev_attr_vendor.attr,
 852         &dev_attr_rev.attr,
 853         &dev_attr_lunid.attr,
 854         &dev_attr_raid_level.attr,
 855         &dev_attr_usage_count.attr,
 856         NULL
 857 };
 858
 859 static struct attribute_group cciss_dev_attr_group = {
 860         .attrs = cciss_dev_attrs,
 861 };
 862
 863 static const struct attribute_group *cciss_dev_attr_groups[] = {
 864         &cciss_dev_attr_group,
 865         NULL
 866 };
 867
 868 static struct device_type cciss_dev_type = {
 869         .name           = "cciss_device",
 870         .groups         = cciss_dev_attr_groups,
 871         .release        = cciss_device_release,
 872 };
 873
 874 static struct bus_type cciss_bus_type = {
 875         .name           = "cciss",
 876 };
 877
 878 /*
 879  * cciss_hba_release is called when the reference count
 880  * of h->dev goes to zero.
 881  */
 882 static void cciss_hba_release(struct device *dev)
 883 {
 884         /*
 885          * nothing to do, but need this to avoid a warning
 886          * about not having a release handler from lib/kref.c.
 887          */
 888 }
 889
 890 /*
 891  * Initialize sysfs entry for each controller.  This sets up and registers
 892  * the 'cciss#' directory for each individual controller under
 893  * /sys/bus/pci/devices/<dev>/.
 894  */
 895 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 896 {
 897         device_initialize(&h->dev);
 898         h->dev.type = &cciss_host_type;
 899         h->dev.bus = &cciss_bus_type;
 900         dev_set_name(&h->dev, "%s", h->devname);
 901         h->dev.parent = &h->pdev->dev;
 902
 903         return device_add(&h->dev);
 904 }
 905
 906 /*
 907  * Remove sysfs entries for an hba.
 908  */
 909 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 910 {
 911         device_del(&h->dev);
 912         put_device(&h->dev); /* final put. */
 913 }
 914
 915 /* cciss_device_release is called when the reference count
 916  * of h->drv[x]dev goes to zero.
 917  */
 918 static void cciss_device_release(struct device *dev)
 919 {
 920         drive_info_struct *drv = to_drv(dev);
 921         kfree(drv);
 922 }
 923
 924 /*
 925  * Initialize sysfs for each logical drive.  This sets up and registers
 926  * the 'c#d#' directory for each individual logical drive under
 927  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 928  * /sys/block/cciss!c#d# to this entry.
 929  */
 930 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 931                                        int drv_index)
 932 {
 933         struct device *dev;
 934
 935         if (h->drv[drv_index]->device_initialized)
 936                 return 0;
 937
 938         dev = &h->drv[drv_index]->dev;
 939         device_initialize(dev);
 940         dev->type = &cciss_dev_type;
 941         dev->bus = &cciss_bus_type;
 942         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 943         dev->parent = &h->dev;
 944         h->drv[drv_index]->device_initialized = 1;
 945         return device_add(dev);
 946 }
 947
 948 /*
 949  * Remove sysfs entries for a logical drive.
 950  */
 951 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 952         int ctlr_exiting)
 953 {
 954         struct device *dev = &h->drv[drv_index]->dev;
 955
 956         /* special case for c*d0, we only destroy it on controller exit */
 957         if (drv_index == 0 && !ctlr_exiting)
 958                 return;
 959
 960         device_del(dev);
 961         put_device(dev); /* the "final" put. */
 962         h->drv[drv_index] = NULL;
 963 }
 964
 965 /*
 966  * For operations that cannot sleep, a command block is allocated at init,
 967  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 968  * which ones are free or in use.
 969  */
 970 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 971 {
 972         CommandList_struct *c;
 973         int i;
 974         u64bit temp64;
 975         dma_addr_t cmd_dma_handle, err_dma_handle;
 976
 977         do {
 978                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 979                 if (i == h->nr_cmds)
 980                         return NULL;
 981         } while (test_and_set_bit(i, h->cmd_pool_bits) != 0);
 982         c = h->cmd_pool + i;
 983         memset(c, 0, sizeof(CommandList_struct));
 984         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
 985         c->err_info = h->errinfo_pool + i;
 986         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 987         err_dma_handle = h->errinfo_pool_dhandle
 988             + i * sizeof(ErrorInfo_struct);
 989         h->nr_allocs++;
 990
 991         c->cmdindex = i;
 992
 993         INIT_LIST_HEAD(&c->list);
 994         c->busaddr = (__u32) cmd_dma_handle;
 995         temp64.val = (__u64) err_dma_handle;
 996         c->ErrDesc.Addr.lower = temp64.val32.lower;
 997         c->ErrDesc.Addr.upper = temp64.val32.upper;
 998         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 999
1000         c->ctlr = h->ctlr;
1001         return c;
1002 }
1003
1004 /* allocate a command using pci_alloc_consistent, used for ioctls,
1005  * etc., not for the main i/o path.
1006  */
1007 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
1008 {
1009         CommandList_struct *c;
1010         u64bit temp64;
1011         dma_addr_t cmd_dma_handle, err_dma_handle;
1012
1013         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
1014                 sizeof(CommandList_struct), &cmd_dma_handle);
1015         if (c == NULL)
1016                 return NULL;
1017         memset(c, 0, sizeof(CommandList_struct));
1018
1019         c->cmdindex = -1;
1020
1021         c->err_info = (ErrorInfo_struct *)
1022             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
1023                     &err_dma_handle);
1024
1025         if (c->err_info == NULL) {
1026                 pci_free_consistent(h->pdev,
1027                         sizeof(CommandList_struct), c, cmd_dma_handle);
1028                 return NULL;
1029         }
1030         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
1031
1032         INIT_LIST_HEAD(&c->list);
1033         c->busaddr = (__u32) cmd_dma_handle;
1034         temp64.val = (__u64) err_dma_handle;
1035         c->ErrDesc.Addr.lower = temp64.val32.lower;
1036         c->ErrDesc.Addr.upper = temp64.val32.upper;
1037         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
1038
1039         c->ctlr = h->ctlr;
1040         return c;
1041 }
1042
1043 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
1044 {
1045         int i;
1046
1047         i = c - h->cmd_pool;
1048         clear_bit(i, h->cmd_pool_bits);
1049         h->nr_frees++;
1050 }
1051
1052 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
1053 {
1054         u64bit temp64;
1055
1056         temp64.val32.lower = c->ErrDesc.Addr.lower;
1057         temp64.val32.upper = c->ErrDesc.Addr.upper;
1058         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1059                             c->err_info, (dma_addr_t) temp64.val);
1060         pci_free_consistent(h->pdev, sizeof(CommandList_struct), c,
1061                 (dma_addr_t) cciss_tag_discard_error_bits(h, (u32) c->busaddr));
1062 }
1063
1064 static inline ctlr_info_t *get_host(struct gendisk *disk)
1065 {
1066         return disk->queue->queuedata;
1067 }
1068
1069 static inline drive_info_struct *get_drv(struct gendisk *disk)
1070 {
1071         return disk->private_data;
1072 }
1073
1074 /*
1075  * Open.  Make sure the device is really there.
1076  */
1077 static int cciss_open(struct block_device *bdev, fmode_t mode)
1078 {
1079         ctlr_info_t *h = get_host(bdev->bd_disk);
1080         drive_info_struct *drv = get_drv(bdev->bd_disk);
1081
1082         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1083         if (drv->busy_configuring)
1084                 return -EBUSY;
1085         /*
1086          * Root is allowed to open raw volume zero even if it's not configured
1087          * so array config can still work. Root is also allowed to open any
1088          * volume that has a LUN ID, so it can issue IOCTL to reread the
1089          * disk information.  I don't think I really like this
1090          * but I'm already using way to many device nodes to claim another one
1091          * for "raw controller".
1092          */
1093         if (drv->heads == 0) {
1094                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1095                         /* if not node 0 make sure it is a partition = 0 */
1096                         if (MINOR(bdev->bd_dev) & 0x0f) {
1097                                 return -ENXIO;
1098                                 /* if it is, make sure we have a LUN ID */
1099                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1100                                 sizeof(drv->LunID))) {
1101                                 return -ENXIO;
1102                         }
1103                 }
1104                 if (!capable(CAP_SYS_ADMIN))
1105                         return -EPERM;
1106         }
1107         drv->usage_count++;
1108         h->usage_count++;
1109         return 0;
1110 }
1111
1112 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1113 {
1114         int ret;
1115
1116         mutex_lock(&cciss_mutex);
1117         ret = cciss_open(bdev, mode);
1118         mutex_unlock(&cciss_mutex);
1119
1120         return ret;
1121 }
1122
1123 /*
1124  * Close.  Sync first.
1125  */
1126 static int cciss_release(struct gendisk *disk, fmode_t mode)
1127 {
1128         ctlr_info_t *h;
1129         drive_info_struct *drv;
1130
1131         mutex_lock(&cciss_mutex);
1132         h = get_host(disk);
1133         drv = get_drv(disk);
1134         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1135         drv->usage_count--;
1136         h->usage_count--;
1137         mutex_unlock(&cciss_mutex);
1138         return 0;
1139 }
1140
1141 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1142                     unsigned cmd, unsigned long arg)
1143 {
1144         int ret;
1145         mutex_lock(&cciss_mutex);
1146         ret = cciss_ioctl(bdev, mode, cmd, arg);
1147         mutex_unlock(&cciss_mutex);
1148         return ret;
1149 }
1150
1151 #ifdef CONFIG_COMPAT
1152
1153 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1154                                   unsigned cmd, unsigned long arg);
1155 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1156                                       unsigned cmd, unsigned long arg);
1157
1158 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1159                               unsigned cmd, unsigned long arg)
1160 {
1161         switch (cmd) {
1162         case CCISS_GETPCIINFO:
1163         case CCISS_GETINTINFO:
1164         case CCISS_SETINTINFO:
1165         case CCISS_GETNODENAME:
1166         case CCISS_SETNODENAME:
1167         case CCISS_GETHEARTBEAT:
1168         case CCISS_GETBUSTYPES:
1169         case CCISS_GETFIRMVER:
1170         case CCISS_GETDRIVVER:
1171         case CCISS_REVALIDVOLS:
1172         case CCISS_DEREGDISK:
1173         case CCISS_REGNEWDISK:
1174         case CCISS_REGNEWD:
1175         case CCISS_RESCANDISK:
1176         case CCISS_GETLUNINFO:
1177                 return do_ioctl(bdev, mode, cmd, arg);
1178
1179         case CCISS_PASSTHRU32:
1180                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1181         case CCISS_BIG_PASSTHRU32:
1182                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1183
1184         default:
1185                 return -ENOIOCTLCMD;
1186         }
1187 }
1188
1189 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1190                                   unsigned cmd, unsigned long arg)
1191 {
1192         IOCTL32_Command_struct __user *arg32 =
1193             (IOCTL32_Command_struct __user *) arg;
1194         IOCTL_Command_struct arg64;
1195         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1196         int err;
1197         u32 cp;
1198
1199         err = 0;
1200         err |=
1201             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1202                            sizeof(arg64.LUN_info));
1203         err |=
1204             copy_from_user(&arg64.Request, &arg32->Request,
1205                            sizeof(arg64.Request));
1206         err |=
1207             copy_from_user(&arg64.error_info, &arg32->error_info,
1208                            sizeof(arg64.error_info));
1209         err |= get_user(arg64.buf_size, &arg32->buf_size);
1210         err |= get_user(cp, &arg32->buf);
1211         arg64.buf = compat_ptr(cp);
1212         err |= copy_to_user(p, &arg64, sizeof(arg64));
1213
1214         if (err)
1215                 return -EFAULT;
1216
1217         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1218         if (err)
1219                 return err;
1220         err |=
1221             copy_in_user(&arg32->error_info, &p->error_info,
1222                          sizeof(arg32->error_info));
1223         if (err)
1224                 return -EFAULT;
1225         return err;
1226 }
1227
1228 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1229                                       unsigned cmd, unsigned long arg)
1230 {
1231         BIG_IOCTL32_Command_struct __user *arg32 =
1232             (BIG_IOCTL32_Command_struct __user *) arg;
1233         BIG_IOCTL_Command_struct arg64;
1234         BIG_IOCTL_Command_struct __user *p =
1235             compat_alloc_user_space(sizeof(arg64));
1236         int err;
1237         u32 cp;
1238
1239         memset(&arg64, 0, sizeof(arg64));
1240         err = 0;
1241         err |=
1242             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1243                            sizeof(arg64.LUN_info));
1244         err |=
1245             copy_from_user(&arg64.Request, &arg32->Request,
1246                            sizeof(arg64.Request));
1247         err |=
1248             copy_from_user(&arg64.error_info, &arg32->error_info,
1249                            sizeof(arg64.error_info));
1250         err |= get_user(arg64.buf_size, &arg32->buf_size);
1251         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1252         err |= get_user(cp, &arg32->buf);
1253         arg64.buf = compat_ptr(cp);
1254         err |= copy_to_user(p, &arg64, sizeof(arg64));
1255
1256         if (err)
1257                 return -EFAULT;
1258
1259         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1260         if (err)
1261                 return err;
1262         err |=
1263             copy_in_user(&arg32->error_info, &p->error_info,
1264                          sizeof(arg32->error_info));
1265         if (err)
1266                 return -EFAULT;
1267         return err;
1268 }
1269 #endif
1270
1271 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1272 {
1273         drive_info_struct *drv = get_drv(bdev->bd_disk);
1274
1275         if (!drv->cylinders)
1276                 return -ENXIO;
1277
1278         geo->heads = drv->heads;
1279         geo->sectors = drv->sectors;
1280         geo->cylinders = drv->cylinders;
1281         return 0;
1282 }
1283
1284 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1285 {
1286         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1287                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1288                 (void)check_for_unit_attention(h, c);
1289 }
1290
1291 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1292 {
1293         cciss_pci_info_struct pciinfo;
1294
1295         if (!argp)
1296                 return -EINVAL;
1297         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1298         pciinfo.bus = h->pdev->bus->number;
1299         pciinfo.dev_fn = h->pdev->devfn;
1300         pciinfo.board_id = h->board_id;
1301         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1302                 return -EFAULT;
1303         return 0;
1304 }
1305
1306 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1307 {
1308         cciss_coalint_struct intinfo;
1309
1310         if (!argp)
1311                 return -EINVAL;
1312         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1313         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1314         if (copy_to_user
1315             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1316                 return -EFAULT;
1317         return 0;
1318 }
1319
1320 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1321 {
1322         cciss_coalint_struct intinfo;
1323         unsigned long flags;
1324         int i;
1325
1326         if (!argp)
1327                 return -EINVAL;
1328         if (!capable(CAP_SYS_ADMIN))
1329                 return -EPERM;
1330         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1331                 return -EFAULT;
1332         if ((intinfo.delay == 0) && (intinfo.count == 0))
1333                 return -EINVAL;
1334         spin_lock_irqsave(&h->lock, flags);
1335         /* Update the field, and then ring the doorbell */
1336         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1337         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1338         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1339
1340         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1341                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1342                         break;
1343                 udelay(1000); /* delay and try again */
1344         }
1345         spin_unlock_irqrestore(&h->lock, flags);
1346         if (i >= MAX_IOCTL_CONFIG_WAIT)
1347                 return -EAGAIN;
1348         return 0;
1349 }
1350
1351 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1352 {
1353         NodeName_type NodeName;
1354         int i;
1355
1356         if (!argp)
1357                 return -EINVAL;
1358         for (i = 0; i < 16; i++)
1359                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1360         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1361                 return -EFAULT;
1362         return 0;
1363 }
1364
1365 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1366 {
1367         NodeName_type NodeName;
1368         unsigned long flags;
1369         int i;
1370
1371         if (!argp)
1372                 return -EINVAL;
1373         if (!capable(CAP_SYS_ADMIN))
1374                 return -EPERM;
1375         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1376                 return -EFAULT;
1377         spin_lock_irqsave(&h->lock, flags);
1378         /* Update the field, and then ring the doorbell */
1379         for (i = 0; i < 16; i++)
1380                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1381         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1382         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1383                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1384                         break;
1385                 udelay(1000); /* delay and try again */
1386         }
1387         spin_unlock_irqrestore(&h->lock, flags);
1388         if (i >= MAX_IOCTL_CONFIG_WAIT)
1389                 return -EAGAIN;
1390         return 0;
1391 }
1392
1393 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1394 {
1395         Heartbeat_type heartbeat;
1396
1397         if (!argp)
1398                 return -EINVAL;
1399         heartbeat = readl(&h->cfgtable->HeartBeat);
1400         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1401                 return -EFAULT;
1402         return 0;
1403 }
1404
1405 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1406 {
1407         BusTypes_type BusTypes;
1408
1409         if (!argp)
1410                 return -EINVAL;
1411         BusTypes = readl(&h->cfgtable->BusTypes);
1412         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1413                 return -EFAULT;
1414         return 0;
1415 }
1416
1417 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1418 {
1419         FirmwareVer_type firmware;
1420
1421         if (!argp)
1422                 return -EINVAL;
1423         memcpy(firmware, h->firm_ver, 4);
1424
1425         if (copy_to_user
1426             (argp, firmware, sizeof(FirmwareVer_type)))
1427                 return -EFAULT;
1428         return 0;
1429 }
1430
1431 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1432 {
1433         DriverVer_type DriverVer = DRIVER_VERSION;
1434
1435         if (!argp)
1436                 return -EINVAL;
1437         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1438                 return -EFAULT;
1439         return 0;
1440 }
1441
1442 static int cciss_getluninfo(ctlr_info_t *h,
1443         struct gendisk *disk, void __user *argp)
1444 {
1445         LogvolInfo_struct luninfo;
1446         drive_info_struct *drv = get_drv(disk);
1447
1448         if (!argp)
1449                 return -EINVAL;
1450         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1451         luninfo.num_opens = drv->usage_count;
1452         luninfo.num_parts = 0;
1453         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1454                 return -EFAULT;
1455         return 0;
1456 }
1457
1458 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1459 {
1460         IOCTL_Command_struct iocommand;
1461         CommandList_struct *c;
1462         char *buff = NULL;
1463         u64bit temp64;
1464         DECLARE_COMPLETION_ONSTACK(wait);
1465
1466         if (!argp)
1467                 return -EINVAL;
1468
1469         if (!capable(CAP_SYS_RAWIO))
1470                 return -EPERM;
1471
1472         if (copy_from_user
1473             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1474                 return -EFAULT;
1475         if ((iocommand.buf_size < 1) &&
1476             (iocommand.Request.Type.Direction != XFER_NONE)) {
1477                 return -EINVAL;
1478         }
1479         if (iocommand.buf_size > 0) {
1480                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1481                 if (buff == NULL)
1482                         return -EFAULT;
1483         }
1484         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1485                 /* Copy the data into the buffer we created */
1486                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1487                         kfree(buff);
1488                         return -EFAULT;
1489                 }
1490         } else {
1491                 memset(buff, 0, iocommand.buf_size);
1492         }
1493         c = cmd_special_alloc(h);
1494         if (!c) {
1495                 kfree(buff);
1496                 return -ENOMEM;
1497         }
1498         /* Fill in the command type */
1499         c->cmd_type = CMD_IOCTL_PEND;
1500         /* Fill in Command Header */
1501         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1502         if (iocommand.buf_size > 0) { /* buffer to fill */
1503                 c->Header.SGList = 1;
1504                 c->Header.SGTotal = 1;
1505         } else { /* no buffers to fill */
1506                 c->Header.SGList = 0;
1507                 c->Header.SGTotal = 0;
1508         }
1509         c->Header.LUN = iocommand.LUN_info;
1510         /* use the kernel address the cmd block for tag */
1511         c->Header.Tag.lower = c->busaddr;
1512
1513         /* Fill in Request block */
1514         c->Request = iocommand.Request;
1515
1516         /* Fill in the scatter gather information */
1517         if (iocommand.buf_size > 0) {
1518                 temp64.val = pci_map_single(h->pdev, buff,
1519                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1520                 c->SG[0].Addr.lower = temp64.val32.lower;
1521                 c->SG[0].Addr.upper = temp64.val32.upper;
1522                 c->SG[0].Len = iocommand.buf_size;
1523                 c->SG[0].Ext = 0;  /* we are not chaining */
1524         }
1525         c->waiting = &wait;
1526
1527         enqueue_cmd_and_start_io(h, c);
1528         wait_for_completion(&wait);
1529
1530         /* unlock the buffers from DMA */
1531         temp64.val32.lower = c->SG[0].Addr.lower;
1532         temp64.val32.upper = c->SG[0].Addr.upper;
1533         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1534                          PCI_DMA_BIDIRECTIONAL);
1535         check_ioctl_unit_attention(h, c);
1536
1537         /* Copy the error information out */
1538         iocommand.error_info = *(c->err_info);
1539         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1540                 kfree(buff);
1541                 cmd_special_free(h, c);
1542                 return -EFAULT;
1543         }
1544
1545         if (iocommand.Request.Type.Direction == XFER_READ) {
1546                 /* Copy the data out of the buffer we created */
1547                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1548                         kfree(buff);
1549                         cmd_special_free(h, c);
1550                         return -EFAULT;
1551                 }
1552         }
1553         kfree(buff);
1554         cmd_special_free(h, c);
1555         return 0;
1556 }
1557
1558 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1559 {
1560         BIG_IOCTL_Command_struct *ioc;
1561         CommandList_struct *c;
1562         unsigned char **buff = NULL;
1563         int *buff_size = NULL;
1564         u64bit temp64;
1565         BYTE sg_used = 0;
1566         int status = 0;
1567         int i;
1568         DECLARE_COMPLETION_ONSTACK(wait);
1569         __u32 left;
1570         __u32 sz;
1571         BYTE __user *data_ptr;
1572
1573         if (!argp)
1574                 return -EINVAL;
1575         if (!capable(CAP_SYS_RAWIO))
1576                 return -EPERM;
1577         ioc = kmalloc(sizeof(*ioc), GFP_KERNEL);
1578         if (!ioc) {
1579                 status = -ENOMEM;
1580                 goto cleanup1;
1581         }
1582         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1583                 status = -EFAULT;
1584                 goto cleanup1;
1585         }
1586         if ((ioc->buf_size < 1) &&
1587             (ioc->Request.Type.Direction != XFER_NONE)) {
1588                 status = -EINVAL;
1589                 goto cleanup1;
1590         }
1591         /* Check kmalloc limits  using all SGs */
1592         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1593                 status = -EINVAL;
1594                 goto cleanup1;
1595         }
1596         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1597                 status = -EINVAL;
1598                 goto cleanup1;
1599         }
1600         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1601         if (!buff) {
1602                 status = -ENOMEM;
1603                 goto cleanup1;
1604         }
1605         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1606         if (!buff_size) {
1607                 status = -ENOMEM;
1608                 goto cleanup1;
1609         }
1610         left = ioc->buf_size;
1611         data_ptr = ioc->buf;
1612         while (left) {
1613                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1614                 buff_size[sg_used] = sz;
1615                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1616                 if (buff[sg_used] == NULL) {
1617                         status = -ENOMEM;
1618                         goto cleanup1;
1619                 }
1620                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1621                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1622                                 status = -EFAULT;
1623                                 goto cleanup1;
1624                         }
1625                 } else {
1626                         memset(buff[sg_used], 0, sz);
1627                 }
1628                 left -= sz;
1629                 data_ptr += sz;
1630                 sg_used++;
1631         }
1632         c = cmd_special_alloc(h);
1633         if (!c) {
1634                 status = -ENOMEM;
1635                 goto cleanup1;
1636         }
1637         c->cmd_type = CMD_IOCTL_PEND;
1638         c->Header.ReplyQueue = 0;
1639         c->Header.SGList = sg_used;
1640         c->Header.SGTotal = sg_used;
1641         c->Header.LUN = ioc->LUN_info;
1642         c->Header.Tag.lower = c->busaddr;
1643
1644         c->Request = ioc->Request;
1645         for (i = 0; i < sg_used; i++) {
1646                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1647                                     PCI_DMA_BIDIRECTIONAL);
1648                 c->SG[i].Addr.lower = temp64.val32.lower;
1649                 c->SG[i].Addr.upper = temp64.val32.upper;
1650                 c->SG[i].Len = buff_size[i];
1651                 c->SG[i].Ext = 0;       /* we are not chaining */
1652         }
1653         c->waiting = &wait;
1654         enqueue_cmd_and_start_io(h, c);
1655         wait_for_completion(&wait);
1656         /* unlock the buffers from DMA */
1657         for (i = 0; i < sg_used; i++) {
1658                 temp64.val32.lower = c->SG[i].Addr.lower;
1659                 temp64.val32.upper = c->SG[i].Addr.upper;
1660                 pci_unmap_single(h->pdev,
1661                         (dma_addr_t) temp64.val, buff_size[i],
1662                         PCI_DMA_BIDIRECTIONAL);
1663         }
1664         check_ioctl_unit_attention(h, c);
1665         /* Copy the error information out */
1666         ioc->error_info = *(c->err_info);
1667         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1668                 cmd_special_free(h, c);
1669                 status = -EFAULT;
1670                 goto cleanup1;
1671         }
1672         if (ioc->Request.Type.Direction == XFER_READ) {
1673                 /* Copy the data out of the buffer we created */
1674                 BYTE __user *ptr = ioc->buf;
1675                 for (i = 0; i < sg_used; i++) {
1676                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1677                                 cmd_special_free(h, c);
1678                                 status = -EFAULT;
1679                                 goto cleanup1;
1680                         }
1681                         ptr += buff_size[i];
1682                 }
1683         }
1684         cmd_special_free(h, c);
1685         status = 0;
1686 cleanup1:
1687         if (buff) {
1688                 for (i = 0; i < sg_used; i++)
1689                         kfree(buff[i]);
1690                 kfree(buff);
1691         }
1692         kfree(buff_size);
1693         kfree(ioc);
1694         return status;
1695 }
1696
1697 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1698         unsigned int cmd, unsigned long arg)
1699 {
1700         struct gendisk *disk = bdev->bd_disk;
1701         ctlr_info_t *h = get_host(disk);
1702         void __user *argp = (void __user *)arg;
1703
1704         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1705                 cmd, arg);
1706         switch (cmd) {
1707         case CCISS_GETPCIINFO:
1708                 return cciss_getpciinfo(h, argp);
1709         case CCISS_GETINTINFO:
1710                 return cciss_getintinfo(h, argp);
1711         case CCISS_SETINTINFO:
1712                 return cciss_setintinfo(h, argp);
1713         case CCISS_GETNODENAME:
1714                 return cciss_getnodename(h, argp);
1715         case CCISS_SETNODENAME:
1716                 return cciss_setnodename(h, argp);
1717         case CCISS_GETHEARTBEAT:
1718                 return cciss_getheartbeat(h, argp);
1719         case CCISS_GETBUSTYPES:
1720                 return cciss_getbustypes(h, argp);
1721         case CCISS_GETFIRMVER:
1722                 return cciss_getfirmver(h, argp);
1723         case CCISS_GETDRIVVER:
1724                 return cciss_getdrivver(h, argp);
1725         case CCISS_DEREGDISK:
1726         case CCISS_REGNEWD:
1727         case CCISS_REVALIDVOLS:
1728                 return rebuild_lun_table(h, 0, 1);
1729         case CCISS_GETLUNINFO:
1730                 return cciss_getluninfo(h, disk, argp);
1731         case CCISS_PASSTHRU:
1732                 return cciss_passthru(h, argp);
1733         case CCISS_BIG_PASSTHRU:
1734                 return cciss_bigpassthru(h, argp);
1735
1736         /* scsi_cmd_blk_ioctl handles these, below, though some are not */
1737         /* very meaningful for cciss.  SG_IO is the main one people want. */
1738
1739         case SG_GET_VERSION_NUM:
1740         case SG_SET_TIMEOUT:
1741         case SG_GET_TIMEOUT:
1742         case SG_GET_RESERVED_SIZE:
1743         case SG_SET_RESERVED_SIZE:
1744         case SG_EMULATED_HOST:
1745         case SG_IO:
1746         case SCSI_IOCTL_SEND_COMMAND:
1747                 return scsi_cmd_blk_ioctl(bdev, mode, cmd, argp);
1748
1749         /* scsi_cmd_blk_ioctl would normally handle these, below, but */
1750         /* they aren't a good fit for cciss, as CD-ROMs are */
1751         /* not supported, and we don't have any bus/target/lun */
1752         /* which we present to the kernel. */
1753
1754         case CDROM_SEND_PACKET:
1755         case CDROMCLOSETRAY:
1756         case CDROMEJECT:
1757         case SCSI_IOCTL_GET_IDLUN:
1758         case SCSI_IOCTL_GET_BUS_NUMBER:
1759         default:
1760                 return -ENOTTY;
1761         }
1762 }
1763
1764 static void cciss_check_queues(ctlr_info_t *h)
1765 {
1766         int start_queue = h->next_to_run;
1767         int i;
1768
1769         /* check to see if we have maxed out the number of commands that can
1770          * be placed on the queue.  If so then exit.  We do this check here
1771          * in case the interrupt we serviced was from an ioctl and did not
1772          * free any new commands.
1773          */
1774         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1775                 return;
1776
1777         /* We have room on the queue for more commands.  Now we need to queue
1778          * them up.  We will also keep track of the next queue to run so
1779          * that every queue gets a chance to be started first.
1780          */
1781         for (i = 0; i < h->highest_lun + 1; i++) {
1782                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1783                 /* make sure the disk has been added and the drive is real
1784                  * because this can be called from the middle of init_one.
1785                  */
1786                 if (!h->drv[curr_queue])
1787                         continue;
1788                 if (!(h->drv[curr_queue]->queue) ||
1789                         !(h->drv[curr_queue]->heads))
1790                         continue;
1791                 blk_start_queue(h->gendisk[curr_queue]->queue);
1792
1793                 /* check to see if we have maxed out the number of commands
1794                  * that can be placed on the queue.
1795                  */
1796                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1797                         if (curr_queue == start_queue) {
1798                                 h->next_to_run =
1799                                     (start_queue + 1) % (h->highest_lun + 1);
1800                                 break;
1801                         } else {
1802                                 h->next_to_run = curr_queue;
1803                                 break;
1804                         }
1805                 }
1806         }
1807 }
1808
1809 static void cciss_softirq_done(struct request *rq)
1810 {
1811         CommandList_struct *c = rq->completion_data;
1812         ctlr_info_t *h = hba[c->ctlr];
1813         SGDescriptor_struct *curr_sg = c->SG;
1814         u64bit temp64;
1815         unsigned long flags;
1816         int i, ddir;
1817         int sg_index = 0;
1818
1819         if (c->Request.Type.Direction == XFER_READ)
1820                 ddir = PCI_DMA_FROMDEVICE;
1821         else
1822                 ddir = PCI_DMA_TODEVICE;
1823
1824         /* command did not need to be retried */
1825         /* unmap the DMA mapping for all the scatter gather elements */
1826         for (i = 0; i < c->Header.SGList; i++) {
1827                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1828                         cciss_unmap_sg_chain_block(h, c);
1829                         /* Point to the next block */
1830                         curr_sg = h->cmd_sg_list[c->cmdindex];
1831                         sg_index = 0;
1832                 }
1833                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1834                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1835                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1836                                 ddir);
1837                 ++sg_index;
1838         }
1839
1840         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1841
1842         /* set the residual count for pc requests */
1843         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1844                 rq->resid_len = c->err_info->ResidualCnt;
1845
1846         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1847
1848         spin_lock_irqsave(&h->lock, flags);
1849         cmd_free(h, c);
1850         cciss_check_queues(h);
1851         spin_unlock_irqrestore(&h->lock, flags);
1852 }
1853
1854 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1855         unsigned char scsi3addr[], uint32_t log_unit)
1856 {
1857         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1858                 sizeof(h->drv[log_unit]->LunID));
1859 }
1860
1861 /* This function gets the SCSI vendor, model, and revision of a logical drive
1862  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1863  * they cannot be read.
1864  */
1865 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1866                                    char *vendor, char *model, char *rev)
1867 {
1868         int rc;
1869         InquiryData_struct *inq_buf;
1870         unsigned char scsi3addr[8];
1871
1872         *vendor = '\0';
1873         *model = '\0';
1874         *rev = '\0';
1875
1876         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1877         if (!inq_buf)
1878                 return;
1879
1880         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1881         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1882                         scsi3addr, TYPE_CMD);
1883         if (rc == IO_OK) {
1884                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1885                 vendor[VENDOR_LEN] = '\0';
1886                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1887                 model[MODEL_LEN] = '\0';
1888                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1889                 rev[REV_LEN] = '\0';
1890         }
1891
1892         kfree(inq_buf);
1893         return;
1894 }
1895
1896 /* This function gets the serial number of a logical drive via
1897  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1898  * number cannot be had, for whatever reason, 16 bytes of 0xff
1899  * are returned instead.
1900  */
1901 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1902                                 unsigned char *serial_no, int buflen)
1903 {
1904 #define PAGE_83_INQ_BYTES 64
1905         int rc;
1906         unsigned char *buf;
1907         unsigned char scsi3addr[8];
1908
1909         if (buflen > 16)
1910                 buflen = 16;
1911         memset(serial_no, 0xff, buflen);
1912         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1913         if (!buf)
1914                 return;
1915         memset(serial_no, 0, buflen);
1916         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1917         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1918                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1919         if (rc == IO_OK)
1920                 memcpy(serial_no, &buf[8], buflen);
1921         kfree(buf);
1922         return;
1923 }
1924
1925 /*
1926  * cciss_add_disk sets up the block device queue for a logical drive
1927  */
1928 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1929                                 int drv_index)
1930 {
1931         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1932         if (!disk->queue)
1933                 goto init_queue_failure;
1934         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1935         disk->major = h->major;
1936         disk->first_minor = drv_index << NWD_SHIFT;
1937         disk->fops = &cciss_fops;
1938         if (cciss_create_ld_sysfs_entry(h, drv_index))
1939                 goto cleanup_queue;
1940         disk->private_data = h->drv[drv_index];
1941         disk->driverfs_dev = &h->drv[drv_index]->dev;
1942
1943         /* Set up queue information */
1944         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1945
1946         /* This is a hardware imposed limit. */
1947         blk_queue_max_segments(disk->queue, h->maxsgentries);
1948
1949         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1950
1951         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1952
1953         disk->queue->queuedata = h;
1954
1955         blk_queue_logical_block_size(disk->queue,
1956                                      h->drv[drv_index]->block_size);
1957
1958         /* Make sure all queue data is written out before */
1959         /* setting h->drv[drv_index]->queue, as setting this */
1960         /* allows the interrupt handler to start the queue */
1961         wmb();
1962         h->drv[drv_index]->queue = disk->queue;
1963         add_disk(disk);
1964         return 0;
1965
1966 cleanup_queue:
1967         blk_cleanup_queue(disk->queue);
1968         disk->queue = NULL;
1969 init_queue_failure:
1970         return -1;
1971 }
1972
1973 /* This function will check the usage_count of the drive to be updated/added.
1974  * If the usage_count is zero and it is a heretofore unknown drive, or,
1975  * the drive's capacity, geometry, or serial number has changed,
1976  * then the drive information will be updated and the disk will be
1977  * re-registered with the kernel.  If these conditions don't hold,
1978  * then it will be left alone for the next reboot.  The exception to this
1979  * is disk 0 which will always be left registered with the kernel since it
1980  * is also the controller node.  Any changes to disk 0 will show up on
1981  * the next reboot.
1982  */
1983 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1984         int first_time, int via_ioctl)
1985 {
1986         struct gendisk *disk;
1987         InquiryData_struct *inq_buff = NULL;
1988         unsigned int block_size;
1989         sector_t total_size;
1990         unsigned long flags = 0;
1991         int ret = 0;
1992         drive_info_struct *drvinfo;
1993
1994         /* Get information about the disk and modify the driver structure */
1995         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1996         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1997         if (inq_buff == NULL || drvinfo == NULL)
1998                 goto mem_msg;
1999
2000         /* testing to see if 16-byte CDBs are already being used */
2001         if (h->cciss_read == CCISS_READ_16) {
2002                 cciss_read_capacity_16(h, drv_index,
2003                         &total_size, &block_size);
2004
2005         } else {
2006                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
2007                 /* if read_capacity returns all F's this volume is >2TB */
2008                 /* in size so we switch to 16-byte CDB's for all */
2009                 /* read/write ops */
2010                 if (total_size == 0xFFFFFFFFULL) {
2011                         cciss_read_capacity_16(h, drv_index,
2012                         &total_size, &block_size);
2013                         h->cciss_read = CCISS_READ_16;
2014                         h->cciss_write = CCISS_WRITE_16;
2015                 } else {
2016                         h->cciss_read = CCISS_READ_10;
2017                         h->cciss_write = CCISS_WRITE_10;
2018                 }
2019         }
2020
2021         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
2022                                inq_buff, drvinfo);
2023         drvinfo->block_size = block_size;
2024         drvinfo->nr_blocks = total_size + 1;
2025
2026         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
2027                                 drvinfo->model, drvinfo->rev);
2028         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
2029                         sizeof(drvinfo->serial_no));
2030         /* Save the lunid in case we deregister the disk, below. */
2031         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
2032                 sizeof(drvinfo->LunID));
2033
2034         /* Is it the same disk we already know, and nothing's changed? */
2035         if (h->drv[drv_index]->raid_level != -1 &&
2036                 ((memcmp(drvinfo->serial_no,
2037                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2038                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2039                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2040                 drvinfo->heads == h->drv[drv_index]->heads &&
2041                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2042                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2043                         /* The disk is unchanged, nothing to update */
2044                         goto freeret;
2045
2046         /* If we get here it's not the same disk, or something's changed,
2047          * so we need to * deregister it, and re-register it, if it's not
2048          * in use.
2049          * If the disk already exists then deregister it before proceeding
2050          * (unless it's the first disk (for the controller node).
2051          */
2052         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2053                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2054                 spin_lock_irqsave(&h->lock, flags);
2055                 h->drv[drv_index]->busy_configuring = 1;
2056                 spin_unlock_irqrestore(&h->lock, flags);
2057
2058                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2059                  * which keeps the interrupt handler from starting
2060                  * the queue.
2061                  */
2062                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2063         }
2064
2065         /* If the disk is in use return */
2066         if (ret)
2067                 goto freeret;
2068
2069         /* Save the new information from cciss_geometry_inquiry
2070          * and serial number inquiry.  If the disk was deregistered
2071          * above, then h->drv[drv_index] will be NULL.
2072          */
2073         if (h->drv[drv_index] == NULL) {
2074                 drvinfo->device_initialized = 0;
2075                 h->drv[drv_index] = drvinfo;
2076                 drvinfo = NULL; /* so it won't be freed below. */
2077         } else {
2078                 /* special case for cxd0 */
2079                 h->drv[drv_index]->block_size = drvinfo->block_size;
2080                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2081                 h->drv[drv_index]->heads = drvinfo->heads;
2082                 h->drv[drv_index]->sectors = drvinfo->sectors;
2083                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2084                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2085                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2086                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2087                         VENDOR_LEN + 1);
2088                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2089                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2090         }
2091
2092         ++h->num_luns;
2093         disk = h->gendisk[drv_index];
2094         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2095
2096         /* If it's not disk 0 (drv_index != 0)
2097          * or if it was disk 0, but there was previously
2098          * no actual corresponding configured logical drive
2099          * (raid_leve == -1) then we want to update the
2100          * logical drive's information.
2101          */
2102         if (drv_index || first_time) {
2103                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2104                         cciss_free_gendisk(h, drv_index);
2105                         cciss_free_drive_info(h, drv_index);
2106                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2107                                 drv_index);
2108                         --h->num_luns;
2109                 }
2110         }
2111
2112 freeret:
2113         kfree(inq_buff);
2114         kfree(drvinfo);
2115         return;
2116 mem_msg:
2117         dev_err(&h->pdev->dev, "out of memory\n");
2118         goto freeret;
2119 }
2120
2121 /* This function will find the first index of the controllers drive array
2122  * that has a null drv pointer and allocate the drive info struct and
2123  * will return that index   This is where new drives will be added.
2124  * If the index to be returned is greater than the highest_lun index for
2125  * the controller then highest_lun is set * to this new index.
2126  * If there are no available indexes or if tha allocation fails, then -1
2127  * is returned.  * "controller_node" is used to know if this is a real
2128  * logical drive, or just the controller node, which determines if this
2129  * counts towards highest_lun.
2130  */
2131 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2132 {
2133         int i;
2134         drive_info_struct *drv;
2135
2136         /* Search for an empty slot for our drive info */
2137         for (i = 0; i < CISS_MAX_LUN; i++) {
2138
2139                 /* if not cxd0 case, and it's occupied, skip it. */
2140                 if (h->drv[i] && i != 0)
2141                         continue;
2142                 /*
2143                  * If it's cxd0 case, and drv is alloc'ed already, and a
2144                  * disk is configured there, skip it.
2145                  */
2146                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2147                         continue;
2148
2149                 /*
2150                  * We've found an empty slot.  Update highest_lun
2151                  * provided this isn't just the fake cxd0 controller node.
2152                  */
2153                 if (i > h->highest_lun && !controller_node)
2154                         h->highest_lun = i;
2155
2156                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2157                 if (i == 0 && h->drv[i] != NULL)
2158                         return i;
2159
2160                 /*
2161                  * Found an empty slot, not already alloc'ed.  Allocate it.
2162                  * Mark it with raid_level == -1, so we know it's new later on.
2163                  */
2164                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2165                 if (!drv)
2166                         return -1;
2167                 drv->raid_level = -1; /* so we know it's new */
2168                 h->drv[i] = drv;
2169                 return i;
2170         }
2171         return -1;
2172 }
2173
2174 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2175 {
2176         kfree(h->drv[drv_index]);
2177         h->drv[drv_index] = NULL;
2178 }
2179
2180 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2181 {
2182         put_disk(h->gendisk[drv_index]);
2183         h->gendisk[drv_index] = NULL;
2184 }
2185
2186 /* cciss_add_gendisk finds a free hba[]->drv structure
2187  * and allocates a gendisk if needed, and sets the lunid
2188  * in the drvinfo structure.   It returns the index into
2189  * the ->drv[] array, or -1 if none are free.
2190  * is_controller_node indicates whether highest_lun should
2191  * count this disk, or if it's only being added to provide
2192  * a means to talk to the controller in case no logical
2193  * drives have yet been configured.
2194  */
2195 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2196         int controller_node)
2197 {
2198         int drv_index;
2199
2200         drv_index = cciss_alloc_drive_info(h, controller_node);
2201         if (drv_index == -1)
2202                 return -1;
2203
2204         /*Check if the gendisk needs to be allocated */
2205         if (!h->gendisk[drv_index]) {
2206                 h->gendisk[drv_index] =
2207                         alloc_disk(1 << NWD_SHIFT);
2208                 if (!h->gendisk[drv_index]) {
2209                         dev_err(&h->pdev->dev,
2210                                 "could not allocate a new disk %d\n",
2211                                 drv_index);
2212                         goto err_free_drive_info;
2213                 }
2214         }
2215         memcpy(h->drv[drv_index]->LunID, lunid,
2216                 sizeof(h->drv[drv_index]->LunID));
2217         if (cciss_create_ld_sysfs_entry(h, drv_index))
2218                 goto err_free_disk;
2219         /* Don't need to mark this busy because nobody */
2220         /* else knows about this disk yet to contend */
2221         /* for access to it. */
2222         h->drv[drv_index]->busy_configuring = 0;
2223         wmb();
2224         return drv_index;
2225
2226 err_free_disk:
2227         cciss_free_gendisk(h, drv_index);
2228 err_free_drive_info:
2229         cciss_free_drive_info(h, drv_index);
2230         return -1;
2231 }
2232
2233 /* This is for the special case of a controller which
2234  * has no logical drives.  In this case, we still need
2235  * to register a disk so the controller can be accessed
2236  * by the Array Config Utility.
2237  */
2238 static void cciss_add_controller_node(ctlr_info_t *h)
2239 {
2240         struct gendisk *disk;
2241         int drv_index;
2242
2243         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2244                 return;
2245
2246         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2247         if (drv_index == -1)
2248                 goto error;
2249         h->drv[drv_index]->block_size = 512;
2250         h->drv[drv_index]->nr_blocks = 0;
2251         h->drv[drv_index]->heads = 0;
2252         h->drv[drv_index]->sectors = 0;
2253         h->drv[drv_index]->cylinders = 0;
2254         h->drv[drv_index]->raid_level = -1;
2255         memset(h->drv[drv_index]->serial_no, 0, 16);
2256         disk = h->gendisk[drv_index];
2257         if (cciss_add_disk(h, disk, drv_index) == 0)
2258                 return;
2259         cciss_free_gendisk(h, drv_index);
2260         cciss_free_drive_info(h, drv_index);
2261 error:
2262         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2263         return;
2264 }
2265
2266 /* This function will add and remove logical drives from the Logical
2267  * drive array of the controller and maintain persistency of ordering
2268  * so that mount points are preserved until the next reboot.  This allows
2269  * for the removal of logical drives in the middle of the drive array
2270  * without a re-ordering of those drives.
2271  * INPUT
2272  * h            = The controller to perform the operations on
2273  */
2274 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2275         int via_ioctl)
2276 {
2277         int num_luns;
2278         ReportLunData_struct *ld_buff = NULL;
2279         int return_code;
2280         int listlength = 0;
2281         int i;
2282         int drv_found;
2283         int drv_index = 0;
2284         unsigned char lunid[8] = CTLR_LUNID;
2285         unsigned long flags;
2286
2287         if (!capable(CAP_SYS_RAWIO))
2288                 return -EPERM;
2289
2290         /* Set busy_configuring flag for this operation */
2291         spin_lock_irqsave(&h->lock, flags);
2292         if (h->busy_configuring) {
2293                 spin_unlock_irqrestore(&h->lock, flags);
2294                 return -EBUSY;
2295         }
2296         h->busy_configuring = 1;
2297         spin_unlock_irqrestore(&h->lock, flags);
2298
2299         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2300         if (ld_buff == NULL)
2301                 goto mem_msg;
2302
2303         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2304                                       sizeof(ReportLunData_struct),
2305                                       0, CTLR_LUNID, TYPE_CMD);
2306
2307         if (return_code == IO_OK)
2308                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2309         else {  /* reading number of logical volumes failed */
2310                 dev_warn(&h->pdev->dev,
2311                         "report logical volume command failed\n");
2312                 listlength = 0;
2313                 goto freeret;
2314         }
2315
2316         num_luns = listlength / 8;      /* 8 bytes per entry */
2317         if (num_luns > CISS_MAX_LUN) {
2318                 num_luns = CISS_MAX_LUN;
2319                 dev_warn(&h->pdev->dev, "more luns configured"
2320                        " on controller than can be handled by"
2321                        " this driver.\n");
2322         }
2323
2324         if (num_luns == 0)
2325                 cciss_add_controller_node(h);
2326
2327         /* Compare controller drive array to driver's drive array
2328          * to see if any drives are missing on the controller due
2329          * to action of Array Config Utility (user deletes drive)
2330          * and deregister logical drives which have disappeared.
2331          */
2332         for (i = 0; i <= h->highest_lun; i++) {
2333                 int j;
2334                 drv_found = 0;
2335
2336                 /* skip holes in the array from already deleted drives */
2337                 if (h->drv[i] == NULL)
2338                         continue;
2339
2340                 for (j = 0; j < num_luns; j++) {
2341                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2342                         if (memcmp(h->drv[i]->LunID, lunid,
2343                                 sizeof(lunid)) == 0) {
2344                                 drv_found = 1;
2345                                 break;
2346                         }
2347                 }
2348                 if (!drv_found) {
2349                         /* Deregister it from the OS, it's gone. */
2350                         spin_lock_irqsave(&h->lock, flags);
2351                         h->drv[i]->busy_configuring = 1;
2352                         spin_unlock_irqrestore(&h->lock, flags);
2353                         return_code = deregister_disk(h, i, 1, via_ioctl);
2354                         if (h->drv[i] != NULL)
2355                                 h->drv[i]->busy_configuring = 0;
2356                 }
2357         }
2358
2359         /* Compare controller drive array to driver's drive array.
2360          * Check for updates in the drive information and any new drives
2361          * on the controller due to ACU adding logical drives, or changing
2362          * a logical drive's size, etc.  Reregister any new/changed drives
2363          */
2364         for (i = 0; i < num_luns; i++) {
2365                 int j;
2366
2367                 drv_found = 0;
2368
2369                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2370                 /* Find if the LUN is already in the drive array
2371                  * of the driver.  If so then update its info
2372                  * if not in use.  If it does not exist then find
2373                  * the first free index and add it.
2374                  */
2375                 for (j = 0; j <= h->highest_lun; j++) {
2376                         if (h->drv[j] != NULL &&
2377                                 memcmp(h->drv[j]->LunID, lunid,
2378                                         sizeof(h->drv[j]->LunID)) == 0) {
2379                                 drv_index = j;
2380                                 drv_found = 1;
2381                                 break;
2382                         }
2383                 }
2384
2385                 /* check if the drive was found already in the array */
2386                 if (!drv_found) {
2387                         drv_index = cciss_add_gendisk(h, lunid, 0);
2388                         if (drv_index == -1)
2389                                 goto freeret;
2390                 }
2391                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2392         }               /* end for */
2393
2394 freeret:
2395         kfree(ld_buff);
2396         h->busy_configuring = 0;
2397         /* We return -1 here to tell the ACU that we have registered/updated
2398          * all of the drives that we can and to keep it from calling us
2399          * additional times.
2400          */
2401         return -1;
2402 mem_msg:
2403         dev_err(&h->pdev->dev, "out of memory\n");
2404         h->busy_configuring = 0;
2405         goto freeret;
2406 }
2407
2408 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2409 {
2410         /* zero out the disk size info */
2411         drive_info->nr_blocks = 0;
2412         drive_info->block_size = 0;
2413         drive_info->heads = 0;
2414         drive_info->sectors = 0;
2415         drive_info->cylinders = 0;
2416         drive_info->raid_level = -1;
2417         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2418         memset(drive_info->model, 0, sizeof(drive_info->model));
2419         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2420         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2421         /*
2422          * don't clear the LUNID though, we need to remember which
2423          * one this one is.
2424          */
2425 }
2426
2427 /* This function will deregister the disk and it's queue from the
2428  * kernel.  It must be called with the controller lock held and the
2429  * drv structures busy_configuring flag set.  It's parameters are:
2430  *
2431  * disk = This is the disk to be deregistered
2432  * drv  = This is the drive_info_struct associated with the disk to be
2433  *        deregistered.  It contains information about the disk used
2434  *        by the driver.
2435  * clear_all = This flag determines whether or not the disk information
2436  *             is going to be completely cleared out and the highest_lun
2437  *             reset.  Sometimes we want to clear out information about
2438  *             the disk in preparation for re-adding it.  In this case
2439  *             the highest_lun should be left unchanged and the LunID
2440  *             should not be cleared.
2441  * via_ioctl
2442  *    This indicates whether we've reached this path via ioctl.
2443  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2444  *    If this path is reached via ioctl(), then the max_usage_count will
2445  *    be 1, as the process calling ioctl() has got to have the device open.
2446  *    If we get here via sysfs, then the max usage count will be zero.
2447 */
2448 static int deregister_disk(ctlr_info_t *h, int drv_index,
2449                            int clear_all, int via_ioctl)
2450 {
2451         int i;
2452         struct gendisk *disk;
2453         drive_info_struct *drv;
2454         int recalculate_highest_lun;
2455
2456         if (!capable(CAP_SYS_RAWIO))
2457                 return -EPERM;
2458
2459         drv = h->drv[drv_index];
2460         disk = h->gendisk[drv_index];
2461
2462         /* make sure logical volume is NOT is use */
2463         if (clear_all || (h->gendisk[0] == disk)) {
2464                 if (drv->usage_count > via_ioctl)
2465                         return -EBUSY;
2466         } else if (drv->usage_count > 0)
2467                 return -EBUSY;
2468
2469         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2470
2471         /* invalidate the devices and deregister the disk.  If it is disk
2472          * zero do not deregister it but just zero out it's values.  This
2473          * allows us to delete disk zero but keep the controller registered.
2474          */
2475         if (h->gendisk[0] != disk) {
2476                 struct request_queue *q = disk->queue;
2477                 if (disk->flags & GENHD_FL_UP) {
2478                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2479                         del_gendisk(disk);
2480                 }
2481                 if (q)
2482                         blk_cleanup_queue(q);
2483                 /* If clear_all is set then we are deleting the logical
2484                  * drive, not just refreshing its info.  For drives
2485                  * other than disk 0 we will call put_disk.  We do not
2486                  * do this for disk 0 as we need it to be able to
2487                  * configure the controller.
2488                  */
2489                 if (clear_all){
2490                         /* This isn't pretty, but we need to find the
2491                          * disk in our array and NULL our the pointer.
2492                          * This is so that we will call alloc_disk if
2493                          * this index is used again later.
2494                          */
2495                         for (i=0; i < CISS_MAX_LUN; i++){
2496                                 if (h->gendisk[i] == disk) {
2497                                         h->gendisk[i] = NULL;
2498                                         break;
2499                                 }
2500                         }
2501                         put_disk(disk);
2502                 }
2503         } else {
2504                 set_capacity(disk, 0);
2505                 cciss_clear_drive_info(drv);
2506         }
2507
2508         --h->num_luns;
2509
2510         /* if it was the last disk, find the new hightest lun */
2511         if (clear_all && recalculate_highest_lun) {
2512                 int newhighest = -1;
2513                 for (i = 0; i <= h->highest_lun; i++) {
2514                         /* if the disk has size > 0, it is available */
2515                         if (h->drv[i] && h->drv[i]->heads)
2516                                 newhighest = i;
2517                 }
2518                 h->highest_lun = newhighest;
2519         }
2520         return 0;
2521 }
2522
2523 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2524                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2525                 int cmd_type)
2526 {
2527         u64bit buff_dma_handle;
2528         int status = IO_OK;
2529
2530         c->cmd_type = CMD_IOCTL_PEND;
2531         c->Header.ReplyQueue = 0;
2532         if (buff != NULL) {
2533                 c->Header.SGList = 1;
2534                 c->Header.SGTotal = 1;
2535         } else {
2536                 c->Header.SGList = 0;
2537                 c->Header.SGTotal = 0;
2538         }
2539         c->Header.Tag.lower = c->busaddr;
2540         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2541
2542         c->Request.Type.Type = cmd_type;
2543         if (cmd_type == TYPE_CMD) {
2544                 switch (cmd) {
2545                 case CISS_INQUIRY:
2546                         /* are we trying to read a vital product page */
2547                         if (page_code != 0) {
2548                                 c->Request.CDB[1] = 0x01;
2549                                 c->Request.CDB[2] = page_code;
2550                         }
2551                         c->Request.CDBLen = 6;
2552                         c->Request.Type.Attribute = ATTR_SIMPLE;
2553                         c->Request.Type.Direction = XFER_READ;
2554                         c->Request.Timeout = 0;
2555                         c->Request.CDB[0] = CISS_INQUIRY;
2556                         c->Request.CDB[4] = size & 0xFF;
2557                         break;
2558                 case CISS_REPORT_LOG:
2559                 case CISS_REPORT_PHYS:
2560                         /* Talking to controller so It's a physical command
2561                            mode = 00 target = 0.  Nothing to write.
2562                          */
2563                         c->Request.CDBLen = 12;
2564                         c->Request.Type.Attribute = ATTR_SIMPLE;
2565                         c->Request.Type.Direction = XFER_READ;
2566                         c->Request.Timeout = 0;
2567                         c->Request.CDB[0] = cmd;
2568                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2569                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2570                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2571                         c->Request.CDB[9] = size & 0xFF;
2572                         break;
2573
2574                 case CCISS_READ_CAPACITY:
2575                         c->Request.CDBLen = 10;
2576                         c->Request.Type.Attribute = ATTR_SIMPLE;
2577                         c->Request.Type.Direction = XFER_READ;
2578                         c->Request.Timeout = 0;
2579                         c->Request.CDB[0] = cmd;
2580                         break;
2581                 case CCISS_READ_CAPACITY_16:
2582                         c->Request.CDBLen = 16;
2583                         c->Request.Type.Attribute = ATTR_SIMPLE;
2584                         c->Request.Type.Direction = XFER_READ;
2585                         c->Request.Timeout = 0;
2586                         c->Request.CDB[0] = cmd;
2587                         c->Request.CDB[1] = 0x10;
2588                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2589                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2590                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2591                         c->Request.CDB[13] = size & 0xFF;
2592                         c->Request.Timeout = 0;
2593                         c->Request.CDB[0] = cmd;
2594                         break;
2595                 case CCISS_CACHE_FLUSH:
2596                         c->Request.CDBLen = 12;
2597                         c->Request.Type.Attribute = ATTR_SIMPLE;
2598                         c->Request.Type.Direction = XFER_WRITE;
2599                         c->Request.Timeout = 0;
2600                         c->Request.CDB[0] = BMIC_WRITE;
2601                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2602                         c->Request.CDB[7] = (size >> 8) & 0xFF;
2603                         c->Request.CDB[8] = size & 0xFF;
2604                         break;
2605                 case TEST_UNIT_READY:
2606                         c->Request.CDBLen = 6;
2607                         c->Request.Type.Attribute = ATTR_SIMPLE;
2608                         c->Request.Type.Direction = XFER_NONE;
2609                         c->Request.Timeout = 0;
2610                         break;
2611                 default:
2612                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2613                         return IO_ERROR;
2614                 }
2615         } else if (cmd_type == TYPE_MSG) {
2616                 switch (cmd) {
2617                 case CCISS_ABORT_MSG:
2618                         c->Request.CDBLen = 12;
2619                         c->Request.Type.Attribute = ATTR_SIMPLE;
2620                         c->Request.Type.Direction = XFER_WRITE;
2621                         c->Request.Timeout = 0;
2622                         c->Request.CDB[0] = cmd;        /* abort */
2623                         c->Request.CDB[1] = 0;  /* abort a command */
2624                         /* buff contains the tag of the command to abort */
2625                         memcpy(&c->Request.CDB[4], buff, 8);
2626                         break;
2627                 case CCISS_RESET_MSG:
2628                         c->Request.CDBLen = 16;
2629                         c->Request.Type.Attribute = ATTR_SIMPLE;
2630                         c->Request.Type.Direction = XFER_NONE;
2631                         c->Request.Timeout = 0;
2632                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2633                         c->Request.CDB[0] = cmd;        /* reset */
2634                         c->Request.CDB[1] = CCISS_RESET_TYPE_TARGET;
2635                         break;
2636                 case CCISS_NOOP_MSG:
2637                         c->Request.CDBLen = 1;
2638                         c->Request.Type.Attribute = ATTR_SIMPLE;
2639                         c->Request.Type.Direction = XFER_WRITE;
2640                         c->Request.Timeout = 0;
2641                         c->Request.CDB[0] = cmd;
2642                         break;
2643                 default:
2644                         dev_warn(&h->pdev->dev,
2645                                 "unknown message type %d\n", cmd);
2646                         return IO_ERROR;
2647                 }
2648         } else {
2649                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2650                 return IO_ERROR;
2651         }
2652         /* Fill in the scatter gather information */
2653         if (size > 0) {
2654                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2655                                                              buff, size,
2656                                                              PCI_DMA_BIDIRECTIONAL);
2657                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2658                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2659                 c->SG[0].Len = size;
2660                 c->SG[0].Ext = 0;       /* we are not chaining */
2661         }
2662         return status;
2663 }
2664
2665 static int cciss_send_reset(ctlr_info_t *h, unsigned char *scsi3addr,
2666                             u8 reset_type)
2667 {
2668         CommandList_struct *c;
2669         int return_status;
2670
2671         c = cmd_alloc(h);
2672         if (!c)
2673                 return -ENOMEM;
2674         return_status = fill_cmd(h, c, CCISS_RESET_MSG, NULL, 0, 0,
2675                 CTLR_LUNID, TYPE_MSG);
2676         c->Request.CDB[1] = reset_type; /* fill_cmd defaults to target reset */
2677         if (return_status != IO_OK) {
2678                 cmd_special_free(h, c);
2679                 return return_status;
2680         }
2681         c->waiting = NULL;
2682         enqueue_cmd_and_start_io(h, c);
2683         /* Don't wait for completion, the reset won't complete.  Don't free
2684          * the command either.  This is the last command we will send before
2685          * re-initializing everything, so it doesn't matter and won't leak.
2686          */
2687         return 0;
2688 }
2689
2690 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2691 {
2692         switch (c->err_info->ScsiStatus) {
2693         case SAM_STAT_GOOD:
2694                 return IO_OK;
2695         case SAM_STAT_CHECK_CONDITION:
2696                 switch (0xf & c->err_info->SenseInfo[2]) {
2697                 case 0: return IO_OK; /* no sense */
2698                 case 1: return IO_OK; /* recovered error */
2699                 default:
2700                         if (check_for_unit_attention(h, c))
2701                                 return IO_NEEDS_RETRY;
2702                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2703                                 "check condition, sense key = 0x%02x\n",
2704                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2705                 }
2706                 break;
2707         default:
2708                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2709                         "scsi status = 0x%02x\n",
2710                         c->Request.CDB[0], c->err_info->ScsiStatus);
2711                 break;
2712         }
2713         return IO_ERROR;
2714 }
2715
2716 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2717 {
2718         int return_status = IO_OK;
2719
2720         if (c->err_info->CommandStatus == CMD_SUCCESS)
2721                 return IO_OK;
2722
2723         switch (c->err_info->CommandStatus) {
2724         case CMD_TARGET_STATUS:
2725                 return_status = check_target_status(h, c);
2726                 break;
2727         case CMD_DATA_UNDERRUN:
2728         case CMD_DATA_OVERRUN:
2729                 /* expected for inquiry and report lun commands */
2730                 break;
2731         case CMD_INVALID:
2732                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2733                        "reported invalid\n", c->Request.CDB[0]);
2734                 return_status = IO_ERROR;
2735                 break;
2736         case CMD_PROTOCOL_ERR:
2737                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2738                        "protocol error\n", c->Request.CDB[0]);
2739                 return_status = IO_ERROR;
2740                 break;
2741         case CMD_HARDWARE_ERR:
2742                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2743                        " hardware error\n", c->Request.CDB[0]);
2744                 return_status = IO_ERROR;
2745                 break;
2746         case CMD_CONNECTION_LOST:
2747                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2748                        "connection lost\n", c->Request.CDB[0]);
2749                 return_status = IO_ERROR;
2750                 break;
2751         case CMD_ABORTED:
2752                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2753                        "aborted\n", c->Request.CDB[0]);
2754                 return_status = IO_ERROR;
2755                 break;
2756         case CMD_ABORT_FAILED:
2757                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2758                        "abort failed\n", c->Request.CDB[0]);
2759                 return_status = IO_ERROR;
2760                 break;
2761         case CMD_UNSOLICITED_ABORT:
2762                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2763                         c->Request.CDB[0]);
2764                 return_status = IO_NEEDS_RETRY;
2765                 break;
2766         case CMD_UNABORTABLE:
2767                 dev_warn(&h->pdev->dev, "cmd unabortable\n");
2768                 return_status = IO_ERROR;
2769                 break;
2770         default:
2771                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2772                        "unknown status %x\n", c->Request.CDB[0],
2773                        c->err_info->CommandStatus);
2774                 return_status = IO_ERROR;
2775         }
2776         return return_status;
2777 }
2778
2779 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2780         int attempt_retry)
2781 {
2782         DECLARE_COMPLETION_ONSTACK(wait);
2783         u64bit buff_dma_handle;
2784         int return_status = IO_OK;
2785
2786 resend_cmd2:
2787         c->waiting = &wait;
2788         enqueue_cmd_and_start_io(h, c);
2789
2790         wait_for_completion(&wait);
2791
2792         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2793                 goto command_done;
2794
2795         return_status = process_sendcmd_error(h, c);
2796
2797         if (return_status == IO_NEEDS_RETRY &&
2798                 c->retry_count < MAX_CMD_RETRIES) {
2799                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2800                         c->Request.CDB[0]);
2801                 c->retry_count++;
2802                 /* erase the old error information */
2803                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2804                 return_status = IO_OK;
2805                 INIT_COMPLETION(wait);
2806                 goto resend_cmd2;
2807         }
2808
2809 command_done:
2810         /* unlock the buffers from DMA */
2811         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2812         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2813         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2814                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2815         return return_status;
2816 }
2817
2818 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2819                            __u8 page_code, unsigned char scsi3addr[],
2820                         int cmd_type)
2821 {
2822         CommandList_struct *c;
2823         int return_status;
2824
2825         c = cmd_special_alloc(h);
2826         if (!c)
2827                 return -ENOMEM;
2828         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2829                 scsi3addr, cmd_type);
2830         if (return_status == IO_OK)
2831                 return_status = sendcmd_withirq_core(h, c, 1);
2832
2833         cmd_special_free(h, c);
2834         return return_status;
2835 }
2836
2837 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2838                                    sector_t total_size,
2839                                    unsigned int block_size,
2840                                    InquiryData_struct *inq_buff,
2841                                    drive_info_struct *drv)
2842 {
2843         int return_code;
2844         unsigned long t;
2845         unsigned char scsi3addr[8];
2846
2847         memset(inq_buff, 0, sizeof(InquiryData_struct));
2848         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2849         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2850                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2851         if (return_code == IO_OK) {
2852                 if (inq_buff->data_byte[8] == 0xFF) {
2853                         dev_warn(&h->pdev->dev,
2854                                "reading geometry failed, volume "
2855                                "does not support reading geometry\n");
2856                         drv->heads = 255;
2857                         drv->sectors = 32;      /* Sectors per track */
2858                         drv->cylinders = total_size + 1;
2859                         drv->raid_level = RAID_UNKNOWN;
2860                 } else {
2861                         drv->heads = inq_buff->data_byte[6];
2862                         drv->sectors = inq_buff->data_byte[7];
2863                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2864                         drv->cylinders += inq_buff->data_byte[5];
2865                         drv->raid_level = inq_buff->data_byte[8];
2866                 }
2867                 drv->block_size = block_size;
2868                 drv->nr_blocks = total_size + 1;
2869                 t = drv->heads * drv->sectors;
2870                 if (t > 1) {
2871                         sector_t real_size = total_size + 1;
2872                         unsigned long rem = sector_div(real_size, t);
2873                         if (rem)
2874                                 real_size++;
2875                         drv->cylinders = real_size;
2876                 }
2877         } else {                /* Get geometry failed */
2878                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2879         }
2880 }
2881
2882 static void
2883 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2884                     unsigned int *block_size)
2885 {
2886         ReadCapdata_struct *buf;
2887         int return_code;
2888         unsigned char scsi3addr[8];
2889
2890         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2891         if (!buf) {
2892                 dev_warn(&h->pdev->dev, "out of memory\n");
2893                 return;
2894         }
2895
2896         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2897         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2898                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2899         if (return_code == IO_OK) {
2900                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2901                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2902         } else {                /* read capacity command failed */
2903                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2904                 *total_size = 0;
2905                 *block_size = BLOCK_SIZE;
2906         }
2907         kfree(buf);
2908 }
2909
2910 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2911         sector_t *total_size, unsigned int *block_size)
2912 {
2913         ReadCapdata_struct_16 *buf;
2914         int return_code;
2915         unsigned char scsi3addr[8];
2916
2917         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2918         if (!buf) {
2919                 dev_warn(&h->pdev->dev, "out of memory\n");
2920                 return;
2921         }
2922
2923         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2924         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2925                 buf, sizeof(ReadCapdata_struct_16),
2926                         0, scsi3addr, TYPE_CMD);
2927         if (return_code == IO_OK) {
2928                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2929                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2930         } else {                /* read capacity command failed */
2931                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2932                 *total_size = 0;
2933                 *block_size = BLOCK_SIZE;
2934         }
2935         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2936                (unsigned long long)*total_size+1, *block_size);
2937         kfree(buf);
2938 }
2939
2940 static int cciss_revalidate(struct gendisk *disk)
2941 {
2942         ctlr_info_t *h = get_host(disk);
2943         drive_info_struct *drv = get_drv(disk);
2944         int logvol;
2945         int FOUND = 0;
2946         unsigned int block_size;
2947         sector_t total_size;
2948         InquiryData_struct *inq_buff = NULL;
2949
2950         for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2951                 if (!h->drv[logvol])
2952                         continue;
2953                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2954                         sizeof(drv->LunID)) == 0) {
2955                         FOUND = 1;
2956                         break;
2957                 }
2958         }
2959
2960         if (!FOUND)
2961                 return 1;
2962
2963         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2964         if (inq_buff == NULL) {
2965                 dev_warn(&h->pdev->dev, "out of memory\n");
2966                 return 1;
2967         }
2968         if (h->cciss_read == CCISS_READ_10) {
2969                 cciss_read_capacity(h, logvol,
2970                                         &total_size, &block_size);
2971         } else {
2972                 cciss_read_capacity_16(h, logvol,
2973                                         &total_size, &block_size);
2974         }
2975         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2976                                inq_buff, drv);
2977
2978         blk_queue_logical_block_size(drv->queue, drv->block_size);
2979         set_capacity(disk, drv->nr_blocks);
2980
2981         kfree(inq_buff);
2982         return 0;
2983 }
2984
2985 /*
2986  * Map (physical) PCI mem into (virtual) kernel space
2987  */
2988 static void __iomem *remap_pci_mem(ulong base, ulong size)
2989 {
2990         ulong page_base = ((ulong) base) & PAGE_MASK;
2991         ulong page_offs = ((ulong) base) - page_base;
2992         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2993
2994         return page_remapped ? (page_remapped + page_offs) : NULL;
2995 }
2996
2997 /*
2998  * Takes jobs of the Q and sends them to the hardware, then puts it on
2999  * the Q to wait for completion.
3000  */
3001 static void start_io(ctlr_info_t *h)
3002 {
3003         CommandList_struct *c;
3004
3005         while (!list_empty(&h->reqQ)) {
3006                 c = list_entry(h->reqQ.next, CommandList_struct, list);
3007                 /* can't do anything if fifo is full */
3008                 if ((h->access.fifo_full(h))) {
3009                         dev_warn(&h->pdev->dev, "fifo full\n");
3010                         break;
3011                 }
3012
3013                 /* Get the first entry from the Request Q */
3014                 removeQ(c);
3015                 h->Qdepth--;
3016
3017                 /* Tell the controller execute command */
3018                 h->access.submit_command(h, c);
3019
3020                 /* Put job onto the completed Q */
3021                 addQ(&h->cmpQ, c);
3022         }
3023 }
3024
3025 /* Assumes that h->lock is held. */
3026 /* Zeros out the error record and then resends the command back */
3027 /* to the controller */
3028 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
3029 {
3030         /* erase the old error information */
3031         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
3032
3033         /* add it to software queue and then send it to the controller */
3034         addQ(&h->reqQ, c);
3035         h->Qdepth++;
3036         if (h->Qdepth > h->maxQsinceinit)
3037                 h->maxQsinceinit = h->Qdepth;
3038
3039         start_io(h);
3040 }
3041
3042 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
3043         unsigned int msg_byte, unsigned int host_byte,
3044         unsigned int driver_byte)
3045 {
3046         /* inverse of macros in scsi.h */
3047         return (scsi_status_byte & 0xff) |
3048                 ((msg_byte & 0xff) << 8) |
3049                 ((host_byte & 0xff) << 16) |
3050                 ((driver_byte & 0xff) << 24);
3051 }
3052
3053 static inline int evaluate_target_status(ctlr_info_t *h,
3054                         CommandList_struct *cmd, int *retry_cmd)
3055 {
3056         unsigned char sense_key;
3057         unsigned char status_byte, msg_byte, host_byte, driver_byte;
3058         int error_value;
3059
3060         *retry_cmd = 0;
3061         /* If we get in here, it means we got "target status", that is, scsi status */
3062         status_byte = cmd->err_info->ScsiStatus;
3063         driver_byte = DRIVER_OK;
3064         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3065
3066         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3067                 host_byte = DID_PASSTHROUGH;
3068         else
3069                 host_byte = DID_OK;
3070
3071         error_value = make_status_bytes(status_byte, msg_byte,
3072                 host_byte, driver_byte);
3073
3074         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3075                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3076                         dev_warn(&h->pdev->dev, "cmd %p "
3077                                "has SCSI Status 0x%x\n",
3078                                cmd, cmd->err_info->ScsiStatus);
3079                 return error_value;
3080         }
3081
3082         /* check the sense key */
3083         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3084         /* no status or recovered error */
3085         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3086             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3087                 error_value = 0;
3088
3089         if (check_for_unit_attention(h, cmd)) {
3090                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3091                 return 0;
3092         }
3093
3094         /* Not SG_IO or similar? */
3095         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3096                 if (error_value != 0)
3097                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3098                                " sense key = 0x%x\n", cmd, sense_key);
3099                 return error_value;
3100         }
3101
3102         /* SG_IO or similar, copy sense data back */
3103         if (cmd->rq->sense) {
3104                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3105                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3106                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3107                         cmd->rq->sense_len);
3108         } else
3109                 cmd->rq->sense_len = 0;
3110
3111         return error_value;
3112 }
3113
3114 /* checks the status of the job and calls complete buffers to mark all
3115  * buffers for the completed job. Note that this function does not need
3116  * to hold the hba/queue lock.
3117  */
3118 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3119                                     int timeout)
3120 {
3121         int retry_cmd = 0;
3122         struct request *rq = cmd->rq;
3123
3124         rq->errors = 0;
3125
3126         if (timeout)
3127                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3128
3129         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3130                 goto after_error_processing;
3131
3132         switch (cmd->err_info->CommandStatus) {
3133         case CMD_TARGET_STATUS:
3134                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3135                 break;
3136         case CMD_DATA_UNDERRUN:
3137                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3138                         dev_warn(&h->pdev->dev, "cmd %p has"
3139                                " completed with data underrun "
3140                                "reported\n", cmd);
3141                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3142                 }
3143                 break;
3144         case CMD_DATA_OVERRUN:
3145                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3146                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3147                                " completed with data overrun "
3148                                "reported\n", cmd);
3149                 break;
3150         case CMD_INVALID:
3151                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3152                        "reported invalid\n", cmd);
3153                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3154                         cmd->err_info->CommandStatus, DRIVER_OK,
3155                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3156                                 DID_PASSTHROUGH : DID_ERROR);
3157                 break;
3158         case CMD_PROTOCOL_ERR:
3159                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3160                        "protocol error\n", cmd);
3161                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3162                         cmd->err_info->CommandStatus, DRIVER_OK,
3163                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3164                                 DID_PASSTHROUGH : DID_ERROR);
3165                 break;
3166         case CMD_HARDWARE_ERR:
3167                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3168                        " hardware error\n", cmd);
3169                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3170                         cmd->err_info->CommandStatus, DRIVER_OK,
3171                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3172                                 DID_PASSTHROUGH : DID_ERROR);
3173                 break;
3174         case CMD_CONNECTION_LOST:
3175                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3176                        "connection lost\n", cmd);
3177                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3178                         cmd->err_info->CommandStatus, DRIVER_OK,
3179                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3180                                 DID_PASSTHROUGH : DID_ERROR);
3181                 break;
3182         case CMD_ABORTED:
3183                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3184                        "aborted\n", cmd);
3185                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3186                         cmd->err_info->CommandStatus, DRIVER_OK,
3187                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3188                                 DID_PASSTHROUGH : DID_ABORT);
3189                 break;
3190         case CMD_ABORT_FAILED:
3191                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3192                        "abort failed\n", cmd);
3193                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3194                         cmd->err_info->CommandStatus, DRIVER_OK,
3195                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3196                                 DID_PASSTHROUGH : DID_ERROR);
3197                 break;
3198         case CMD_UNSOLICITED_ABORT:
3199                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3200                        "abort %p\n", h->ctlr, cmd);
3201                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3202                         retry_cmd = 1;
3203                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3204                         cmd->retry_count++;
3205                 } else
3206                         dev_warn(&h->pdev->dev,
3207                                 "%p retried too many times\n", cmd);
3208                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3209                         cmd->err_info->CommandStatus, DRIVER_OK,
3210                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3211                                 DID_PASSTHROUGH : DID_ABORT);
3212                 break;
3213         case CMD_TIMEOUT:
3214                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3215                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3216                         cmd->err_info->CommandStatus, DRIVER_OK,
3217                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3218                                 DID_PASSTHROUGH : DID_ERROR);
3219                 break;
3220         case CMD_UNABORTABLE:
3221                 dev_warn(&h->pdev->dev, "cmd %p unabortable\n", cmd);
3222                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3223                         cmd->err_info->CommandStatus, DRIVER_OK,
3224                         cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC ?
3225                                 DID_PASSTHROUGH : DID_ERROR);
3226                 break;
3227         default:
3228                 dev_warn(&h->pdev->dev, "cmd %p returned "
3229                        "unknown status %x\n", cmd,
3230                        cmd->err_info->CommandStatus);
3231                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3232                         cmd->err_info->CommandStatus, DRIVER_OK,
3233                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3234                                 DID_PASSTHROUGH : DID_ERROR);
3235         }
3236
3237 after_error_processing:
3238
3239         /* We need to return this command */
3240         if (retry_cmd) {
3241                 resend_cciss_cmd(h, cmd);
3242                 return;
3243         }
3244         cmd->rq->completion_data = cmd;
3245         blk_complete_request(cmd->rq);
3246 }
3247
3248 static inline u32 cciss_tag_contains_index(u32 tag)
3249 {
3250 #define DIRECT_LOOKUP_BIT 0x10
3251         return tag & DIRECT_LOOKUP_BIT;
3252 }
3253
3254 static inline u32 cciss_tag_to_index(u32 tag)
3255 {
3256 #define DIRECT_LOOKUP_SHIFT 5
3257         return tag >> DIRECT_LOOKUP_SHIFT;
3258 }
3259
3260 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag)
3261 {
3262 #define CCISS_PERF_ERROR_BITS ((1 << DIRECT_LOOKUP_SHIFT) - 1)
3263 #define CCISS_SIMPLE_ERROR_BITS 0x03
3264         if (likely(h->transMethod & CFGTBL_Trans_Performant))
3265                 return tag & ~CCISS_PERF_ERROR_BITS;
3266         return tag & ~CCISS_SIMPLE_ERROR_BITS;
3267 }
3268
3269 static inline void cciss_mark_tag_indexed(u32 *tag)
3270 {
3271         *tag |= DIRECT_LOOKUP_BIT;
3272 }
3273
3274 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3275 {
3276         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3277 }
3278
3279 /*
3280  * Get a request and submit it to the controller.
3281  */
3282 static void do_cciss_request(struct request_queue *q)
3283 {
3284         ctlr_info_t *h = q->queuedata;
3285         CommandList_struct *c;
3286         sector_t start_blk;
3287         int seg;
3288         struct request *creq;
3289         u64bit temp64;
3290         struct scatterlist *tmp_sg;
3291         SGDescriptor_struct *curr_sg;
3292         drive_info_struct *drv;
3293         int i, dir;
3294         int sg_index = 0;
3295         int chained = 0;
3296
3297       queue:
3298         creq = blk_peek_request(q);
3299         if (!creq)
3300                 goto startio;
3301
3302         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3303
3304         c = cmd_alloc(h);
3305         if (!c)
3306                 goto full;
3307
3308         blk_start_request(creq);
3309
3310         tmp_sg = h->scatter_list[c->cmdindex];
3311         spin_unlock_irq(q->queue_lock);
3312
3313         c->cmd_type = CMD_RWREQ;
3314         c->rq = creq;
3315
3316         /* fill in the request */
3317         drv = creq->rq_disk->private_data;
3318         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3319         /* got command from pool, so use the command block index instead */
3320         /* for direct lookups. */
3321         /* The first 2 bits are reserved for controller error reporting. */
3322         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3323         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3324         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3325         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3326         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3327         c->Request.Type.Attribute = ATTR_SIMPLE;
3328         c->Request.Type.Direction =
3329          &nbs