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 int cciss_allow_hpsa;
  79 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  80 MODULE_PARM_DESC(cciss_allow_hpsa,
  81         "Prevent cciss driver from accessing hardware known to be "
  82         " supported by the hpsa driver");
  83
  84 static DEFINE_MUTEX(cciss_mutex);
  85 static struct proc_dir_entry *proc_cciss;
  86
  87 #include "cciss_cmd.h"
  88 #include "cciss.h"
  89 #include <linux/cciss_ioctl.h>
  90
  91 /* define the PCI info for the cards we can control */
  92 static const struct pci_device_id cciss_pci_device_id[] = {
  93         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  94         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  95         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  96         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  97         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  98         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  99         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
 100         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
 101         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
 110         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
 111         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 112         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 113         {0,}
 114 };
 115
 116 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 117
 118 /*  board_id = Subsystem Device ID & Vendor ID
 119  *  product = Marketing Name for the board
 120  *  access = Address of the struct of function pointers
 121  */
 122 static struct board_type products[] = {
 123         {0x40700E11, "Smart Array 5300", &SA5_access},
 124         {0x40800E11, "Smart Array 5i", &SA5B_access},
 125         {0x40820E11, "Smart Array 532", &SA5B_access},
 126         {0x40830E11, "Smart Array 5312", &SA5B_access},
 127         {0x409A0E11, "Smart Array 641", &SA5_access},
 128         {0x409B0E11, "Smart Array 642", &SA5_access},
 129         {0x409C0E11, "Smart Array 6400", &SA5_access},
 130         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 131         {0x40910E11, "Smart Array 6i", &SA5_access},
 132         {0x3225103C, "Smart Array P600", &SA5_access},
 133         {0x3223103C, "Smart Array P800", &SA5_access},
 134         {0x3234103C, "Smart Array P400", &SA5_access},
 135         {0x3235103C, "Smart Array P400i", &SA5_access},
 136         {0x3211103C, "Smart Array E200i", &SA5_access},
 137         {0x3212103C, "Smart Array E200", &SA5_access},
 138         {0x3213103C, "Smart Array E200i", &SA5_access},
 139         {0x3214103C, "Smart Array E200i", &SA5_access},
 140         {0x3215103C, "Smart Array E200i", &SA5_access},
 141         {0x3237103C, "Smart Array E500", &SA5_access},
 142         {0x3223103C, "Smart Array P800", &SA5_access},
 143         {0x3234103C, "Smart Array P400", &SA5_access},
 144         {0x323D103C, "Smart Array P700m", &SA5_access},
 145 };
 146
 147 /* How long to wait (in milliseconds) for board to go into simple mode */
 148 #define MAX_CONFIG_WAIT 30000
 149 #define MAX_IOCTL_CONFIG_WAIT 1000
 150
 151 /*define how many times we will try a command because of bus resets */
 152 #define MAX_CMD_RETRIES 3
 153
 154 #define MAX_CTLR        32
 155
 156 /* Originally cciss driver only supports 8 major numbers */
 157 #define MAX_CTLR_ORIG   8
 158
 159 static ctlr_info_t *hba[MAX_CTLR];
 160
 161 static struct task_struct *cciss_scan_thread;
 162 static DEFINE_MUTEX(scan_mutex);
 163 static LIST_HEAD(scan_q);
 164
 165 static void do_cciss_request(struct request_queue *q);
 166 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 167 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 168 static int cciss_open(struct block_device *bdev, fmode_t mode);
 169 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 170 static void cciss_release(struct gendisk *disk, fmode_t mode);
 171 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 172                        unsigned int cmd, unsigned long arg);
 173 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 174
 175 static int cciss_revalidate(struct gendisk *disk);
 176 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 177 static int deregister_disk(ctlr_info_t *h, int drv_index,
 178                            int clear_all, int via_ioctl);
 179
 180 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 181                         sector_t *total_size, unsigned int *block_size);
 182 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 183                         sector_t *total_size, unsigned int *block_size);
 184 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 185                         sector_t total_size,
 186                         unsigned int block_size, InquiryData_struct *inq_buff,
 187                                    drive_info_struct *drv);
 188 static void cciss_interrupt_mode(ctlr_info_t *);
 189 static int cciss_enter_simple_mode(struct ctlr_info *h);
 190 static void start_io(ctlr_info_t *h);
 191 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 192                         __u8 page_code, unsigned char scsi3addr[],
 193                         int cmd_type);
 194 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 195         int attempt_retry);
 196 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 197
 198 static int add_to_scan_list(struct ctlr_info *h);
 199 static int scan_thread(void *data);
 200 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 201 static void cciss_hba_release(struct device *dev);
 202 static void cciss_device_release(struct device *dev);
 203 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 204 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 205 static inline u32 next_command(ctlr_info_t *h);
 206 static int cciss_find_cfg_addrs(struct pci_dev *pdev, void __iomem *vaddr,
 207                                 u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 208                                 u64 *cfg_offset);
 209 static int cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 210                                      unsigned long *memory_bar);
 211 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag);
 212 static int write_driver_ver_to_cfgtable(CfgTable_struct __iomem *cfgtable);
 213
 214 /* performant mode helper functions */
 215 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 216                                 int *bucket_map);
 217 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 218
 219 #ifdef CONFIG_PROC_FS
 220 static void cciss_procinit(ctlr_info_t *h);
 221 #else
 222 static void cciss_procinit(ctlr_info_t *h)
 223 {
 224 }
 225 #endif                          /* CONFIG_PROC_FS */
 226
 227 #ifdef CONFIG_COMPAT
 228 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 229                               unsigned, unsigned long);
 230 #endif
 231
 232 static const struct block_device_operations cciss_fops = {
 233         .owner = THIS_MODULE,
 234         .open = cciss_unlocked_open,
 235         .release = cciss_release,
 236         .ioctl = cciss_ioctl,
 237         .getgeo = cciss_getgeo,
 238 #ifdef CONFIG_COMPAT
 239         .compat_ioctl = cciss_compat_ioctl,
 240 #endif
 241         .revalidate_disk = cciss_revalidate,
 242 };
 243
 244 /* set_performant_mode: Modify the tag for cciss performant
 245  * set bit 0 for pull model, bits 3-1 for block fetch
 246  * register number
 247  */
 248 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 249 {
 250         if (likely(h->transMethod & CFGTBL_Trans_Performant))
 251                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 252 }
 253
 254 /*
 255  * Enqueuing and dequeuing functions for cmdlists.
 256  */
 257 static inline void addQ(struct list_head *list, CommandList_struct *c)
 258 {
 259         list_add_tail(&c->list, list);
 260 }
 261
 262 static inline void removeQ(CommandList_struct *c)
 263 {
 264         /*
 265          * After kexec/dump some commands might still
 266          * be in flight, which the firmware will try
 267          * to complete. Resetting the firmware doesn't work
 268          * with old fw revisions, so we have to mark
 269          * them off as 'stale' to prevent the driver from
 270          * falling over.
 271          */
 272         if (WARN_ON(list_empty(&c->list))) {
 273                 c->cmd_type = CMD_MSG_STALE;
 274                 return;
 275         }
 276
 277         list_del_init(&c->list);
 278 }
 279
 280 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 281         CommandList_struct *c)
 282 {
 283         unsigned long flags;
 284         set_performant_mode(h, c);
 285         spin_lock_irqsave(&h->lock, flags);
 286         addQ(&h->reqQ, c);
 287         h->Qdepth++;
 288         if (h->Qdepth > h->maxQsinceinit)
 289                 h->maxQsinceinit = h->Qdepth;
 290         start_io(h);
 291         spin_unlock_irqrestore(&h->lock, flags);
 292 }
 293
 294 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 295         int nr_cmds)
 296 {
 297         int i;
 298
 299         if (!cmd_sg_list)
 300                 return;
 301         for (i = 0; i < nr_cmds; i++) {
 302                 kfree(cmd_sg_list[i]);
 303                 cmd_sg_list[i] = NULL;
 304         }
 305         kfree(cmd_sg_list);
 306 }
 307
 308 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 309         ctlr_info_t *h, int chainsize, int nr_cmds)
 310 {
 311         int j;
 312         SGDescriptor_struct **cmd_sg_list;
 313
 314         if (chainsize <= 0)
 315                 return NULL;
 316
 317         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 318         if (!cmd_sg_list)
 319                 return NULL;
 320
 321         /* Build up chain blocks for each command */
 322         for (j = 0; j < nr_cmds; j++) {
 323                 /* Need a block of chainsized s/g elements. */
 324                 cmd_sg_list[j] = kmalloc((chainsize *
 325                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 326                 if (!cmd_sg_list[j]) {
 327                         dev_err(&h->pdev->dev, "Cannot get memory "
 328                                 "for s/g chains.\n");
 329                         goto clean;
 330                 }
 331         }
 332         return cmd_sg_list;
 333 clean:
 334         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 335         return NULL;
 336 }
 337
 338 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 339 {
 340         SGDescriptor_struct *chain_sg;
 341         u64bit temp64;
 342
 343         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 344                 return;
 345
 346         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 347         temp64.val32.lower = chain_sg->Addr.lower;
 348         temp64.val32.upper = chain_sg->Addr.upper;
 349         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 350 }
 351
 352 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 353         SGDescriptor_struct *chain_block, int len)
 354 {
 355         SGDescriptor_struct *chain_sg;
 356         u64bit temp64;
 357
 358         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 359         chain_sg->Ext = CCISS_SG_CHAIN;
 360         chain_sg->Len = len;
 361         temp64.val = pci_map_single(h->pdev, chain_block, len,
 362                                 PCI_DMA_TODEVICE);
 363         chain_sg->Addr.lower = temp64.val32.lower;
 364         chain_sg->Addr.upper = temp64.val32.upper;
 365 }
 366
 367 #include "cciss_scsi.c"         /* For SCSI tape support */
 368
 369 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 370         "UNKNOWN"
 371 };
 372 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 373
 374 #ifdef CONFIG_PROC_FS
 375
 376 /*
 377  * Report information about this controller.
 378  */
 379 #define ENG_GIG 1000000000
 380 #define ENG_GIG_FACTOR (ENG_GIG/512)
 381 #define ENGAGE_SCSI     "engage scsi"
 382
 383 static void cciss_seq_show_header(struct seq_file *seq)
 384 {
 385         ctlr_info_t *h = seq->private;
 386
 387         seq_printf(seq, "%s: HP %s Controller\n"
 388                 "Board ID: 0x%08lx\n"
 389                 "Firmware Version: %c%c%c%c\n"
 390                 "IRQ: %d\n"
 391                 "Logical drives: %d\n"
 392                 "Current Q depth: %d\n"
 393                 "Current # commands on controller: %d\n"
 394                 "Max Q depth since init: %d\n"
 395                 "Max # commands on controller since init: %d\n"
 396                 "Max SG entries since init: %d\n",
 397                 h->devname,
 398                 h->product_name,
 399                 (unsigned long)h->board_id,
 400                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 401                 h->firm_ver[3], (unsigned int)h->intr[h->intr_mode],
 402                 h->num_luns,
 403                 h->Qdepth, h->commands_outstanding,
 404                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 405
 406 #ifdef CONFIG_CISS_SCSI_TAPE
 407         cciss_seq_tape_report(seq, h);
 408 #endif /* CONFIG_CISS_SCSI_TAPE */
 409 }
 410
 411 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 412 {
 413         ctlr_info_t *h = seq->private;
 414         unsigned long flags;
 415
 416         /* prevent displaying bogus info during configuration
 417          * or deconfiguration of a logical volume
 418          */
 419         spin_lock_irqsave(&h->lock, flags);
 420         if (h->busy_configuring) {
 421                 spin_unlock_irqrestore(&h->lock, flags);
 422                 return ERR_PTR(-EBUSY);
 423         }
 424         h->busy_configuring = 1;
 425         spin_unlock_irqrestore(&h->lock, flags);
 426
 427         if (*pos == 0)
 428                 cciss_seq_show_header(seq);
 429
 430         return pos;
 431 }
 432
 433 static int cciss_seq_show(struct seq_file *seq, void *v)
 434 {
 435         sector_t vol_sz, vol_sz_frac;
 436         ctlr_info_t *h = seq->private;
 437         unsigned ctlr = h->ctlr;
 438         loff_t *pos = v;
 439         drive_info_struct *drv = h->drv[*pos];
 440
 441         if (*pos > h->highest_lun)
 442                 return 0;
 443
 444         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 445                 return 0;
 446
 447         if (drv->heads == 0)
 448                 return 0;
 449
 450         vol_sz = drv->nr_blocks;
 451         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 452         vol_sz_frac *= 100;
 453         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 454
 455         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 456                 drv->raid_level = RAID_UNKNOWN;
 457         seq_printf(seq, "cciss/c%dd%d:"
 458                         "\t%4u.%02uGB\tRAID %s\n",
 459                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 460                         raid_label[drv->raid_level]);
 461         return 0;
 462 }
 463
 464 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 465 {
 466         ctlr_info_t *h = seq->private;
 467
 468         if (*pos > h->highest_lun)
 469                 return NULL;
 470         *pos += 1;
 471
 472         return pos;
 473 }
 474
 475 static void cciss_seq_stop(struct seq_file *seq, void *v)
 476 {
 477         ctlr_info_t *h = seq->private;
 478
 479         /* Only reset h->busy_configuring if we succeeded in setting
 480          * it during cciss_seq_start. */
 481         if (v == ERR_PTR(-EBUSY))
 482                 return;
 483
 484         h->busy_configuring = 0;
 485 }
 486
 487 static const struct seq_operations cciss_seq_ops = {
 488         .start = cciss_seq_start,
 489         .show  = cciss_seq_show,
 490         .next  = cciss_seq_next,
 491         .stop  = cciss_seq_stop,
 492 };
 493
 494 static int cciss_seq_open(struct inode *inode, struct file *file)
 495 {
 496         int ret = seq_open(file, &cciss_seq_ops);
 497         struct seq_file *seq = file->private_data;
 498
 499         if (!ret)
 500                 seq->private = PDE_DATA(inode);
 501
 502         return ret;
 503 }
 504
 505 static ssize_t
 506 cciss_proc_write(struct file *file, const char __user *buf,
 507                  size_t length, loff_t *ppos)
 508 {
 509         int err;
 510         char *buffer;
 511
 512 #ifndef CONFIG_CISS_SCSI_TAPE
 513         return -EINVAL;
 514 #endif
 515
 516         if (!buf || length > PAGE_SIZE - 1)
 517                 return -EINVAL;
 518
 519         buffer = (char *)__get_free_page(GFP_KERNEL);
 520         if (!buffer)
 521                 return -ENOMEM;
 522
 523         err = -EFAULT;
 524         if (copy_from_user(buffer, buf, length))
 525                 goto out;
 526         buffer[length] = '\0';
 527
 528 #ifdef CONFIG_CISS_SCSI_TAPE
 529         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 530                 struct seq_file *seq = file->private_data;
 531                 ctlr_info_t *h = seq->private;
 532
 533                 err = cciss_engage_scsi(h);
 534                 if (err == 0)
 535                         err = length;
 536         } else
 537 #endif /* CONFIG_CISS_SCSI_TAPE */
 538                 err = -EINVAL;
 539         /* might be nice to have "disengage" too, but it's not
 540            safely possible. (only 1 module use count, lock issues.) */
 541
 542 out:
 543         free_page((unsigned long)buffer);
 544         return err;
 545 }
 546
 547 static const struct file_operations cciss_proc_fops = {
 548         .owner   = THIS_MODULE,
 549         .open    = cciss_seq_open,
 550         .read    = seq_read,
 551         .llseek  = seq_lseek,
 552         .release = seq_release,
 553         .write   = cciss_proc_write,
 554 };
 555
 556 static void cciss_procinit(ctlr_info_t *h)
 557 {
 558         struct proc_dir_entry *pde;
 559
 560         if (proc_cciss == NULL)
 561                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 562         if (!proc_cciss)
 563                 return;
 564         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 565                                         S_IROTH, proc_cciss,
 566                                         &cciss_proc_fops, h);
 567 }
 568 #endif                          /* CONFIG_PROC_FS */
 569
 570 #define MAX_PRODUCT_NAME_LEN 19
 571
 572 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 573 #define to_drv(n) container_of(n, drive_info_struct, dev)
 574
 575 /* List of controllers which cannot be hard reset on kexec with reset_devices */
 576 static u32 unresettable_controller[] = {
 577         0x324a103C, /* Smart Array P712m */
 578         0x324b103C, /* SmartArray P711m */
 579         0x3223103C, /* Smart Array P800 */
 580         0x3234103C, /* Smart Array P400 */
 581         0x3235103C, /* Smart Array P400i */
 582         0x3211103C, /* Smart Array E200i */
 583         0x3212103C, /* Smart Array E200 */
 584         0x3213103C, /* Smart Array E200i */
 585         0x3214103C, /* Smart Array E200i */
 586         0x3215103C, /* Smart Array E200i */
 587         0x3237103C, /* Smart Array E500 */
 588         0x323D103C, /* Smart Array P700m */
 589         0x409C0E11, /* Smart Array 6400 */
 590         0x409D0E11, /* Smart Array 6400 EM */
 591 };
 592
 593 /* List of controllers which cannot even be soft reset */
 594 static u32 soft_unresettable_controller[] = {
 595         0x409C0E11, /* Smart Array 6400 */
 596         0x409D0E11, /* Smart Array 6400 EM */
 597 };
 598
 599 static int ctlr_is_hard_resettable(u32 board_id)
 600 {
 601         int i;
 602
 603         for (i = 0; i < ARRAY_SIZE(unresettable_controller); i++)
 604                 if (unresettable_controller[i] == board_id)
 605                         return 0;
 606         return 1;
 607 }
 608
 609 static int ctlr_is_soft_resettable(u32 board_id)
 610 {
 611         int i;
 612
 613         for (i = 0; i < ARRAY_SIZE(soft_unresettable_controller); i++)
 614                 if (soft_unresettable_controller[i] == board_id)
 615                         return 0;
 616         return 1;
 617 }
 618
 619 static int ctlr_is_resettable(u32 board_id)
 620 {
 621         return ctlr_is_hard_resettable(board_id) ||
 622                 ctlr_is_soft_resettable(board_id);
 623 }
 624
 625 static ssize_t host_show_resettable(struct device *dev,
 626                                     struct device_attribute *attr,
 627                                     char *buf)
 628 {
 629         struct ctlr_info *h = to_hba(dev);
 630
 631         return snprintf(buf, 20, "%d\n", ctlr_is_resettable(h->board_id));
 632 }
 633 static DEVICE_ATTR(resettable, S_IRUGO, host_show_resettable, NULL);
 634
 635 static ssize_t host_store_rescan(struct device *dev,
 636                                  struct device_attribute *attr,
 637                                  const char *buf, size_t count)
 638 {
 639         struct ctlr_info *h = to_hba(dev);
 640
 641         add_to_scan_list(h);
 642         wake_up_process(cciss_scan_thread);
 643         wait_for_completion_interruptible(&h->scan_wait);
 644
 645         return count;
 646 }
 647 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 648
 649 static ssize_t host_show_transport_mode(struct device *dev,
 650                                  struct device_attribute *attr,
 651                                  char *buf)
 652 {
 653         struct ctlr_info *h = to_hba(dev);
 654
 655         return snprintf(buf, 20, "%s\n",
 656                 h->transMethod & CFGTBL_Trans_Performant ?
 657                         "performant" : "simple");
 658 }
 659 static DEVICE_ATTR(transport_mode, S_IRUGO, host_show_transport_mode, NULL);
 660
 661 static ssize_t dev_show_unique_id(struct device *dev,
 662                                  struct device_attribute *attr,
 663                                  char *buf)
 664 {
 665         drive_info_struct *drv = to_drv(dev);
 666         struct ctlr_info *h = to_hba(drv->dev.parent);
 667         __u8 sn[16];
 668         unsigned long flags;
 669         int ret = 0;
 670
 671         spin_lock_irqsave(&h->lock, flags);
 672         if (h->busy_configuring)
 673                 ret = -EBUSY;
 674         else
 675                 memcpy(sn, drv->serial_no, sizeof(sn));
 676         spin_unlock_irqrestore(&h->lock, flags);
 677
 678         if (ret)
 679                 return ret;
 680         else
 681                 return snprintf(buf, 16 * 2 + 2,
 682                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 683                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 684                                 sn[0], sn[1], sn[2], sn[3],
 685                                 sn[4], sn[5], sn[6], sn[7],
 686                                 sn[8], sn[9], sn[10], sn[11],
 687                                 sn[12], sn[13], sn[14], sn[15]);
 688 }
 689 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 690
 691 static ssize_t dev_show_vendor(struct device *dev,
 692                                struct device_attribute *attr,
 693                                char *buf)
 694 {
 695         drive_info_struct *drv = to_drv(dev);
 696         struct ctlr_info *h = to_hba(drv->dev.parent);
 697         char vendor[VENDOR_LEN + 1];
 698         unsigned long flags;
 699         int ret = 0;
 700
 701         spin_lock_irqsave(&h->lock, flags);
 702         if (h->busy_configuring)
 703                 ret = -EBUSY;
 704         else
 705                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 706         spin_unlock_irqrestore(&h->lock, flags);
 707
 708         if (ret)
 709                 return ret;
 710         else
 711                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 712 }
 713 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 714
 715 static ssize_t dev_show_model(struct device *dev,
 716                               struct device_attribute *attr,
 717                               char *buf)
 718 {
 719         drive_info_struct *drv = to_drv(dev);
 720         struct ctlr_info *h = to_hba(drv->dev.parent);
 721         char model[MODEL_LEN + 1];
 722         unsigned long flags;
 723         int ret = 0;
 724
 725         spin_lock_irqsave(&h->lock, flags);
 726         if (h->busy_configuring)
 727                 ret = -EBUSY;
 728         else
 729                 memcpy(model, drv->model, MODEL_LEN + 1);
 730         spin_unlock_irqrestore(&h->lock, flags);
 731
 732         if (ret)
 733                 return ret;
 734         else
 735                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 736 }
 737 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 738
 739 static ssize_t dev_show_rev(struct device *dev,
 740                             struct device_attribute *attr,
 741                             char *buf)
 742 {
 743         drive_info_struct *drv = to_drv(dev);
 744         struct ctlr_info *h = to_hba(drv->dev.parent);
 745         char rev[REV_LEN + 1];
 746         unsigned long flags;
 747         int ret = 0;
 748
 749         spin_lock_irqsave(&h->lock, flags);
 750         if (h->busy_configuring)
 751                 ret = -EBUSY;
 752         else
 753                 memcpy(rev, drv->rev, REV_LEN + 1);
 754         spin_unlock_irqrestore(&h->lock, flags);
 755
 756         if (ret)
 757                 return ret;
 758         else
 759                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 760 }
 761 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 762
 763 static ssize_t cciss_show_lunid(struct device *dev,
 764                                 struct device_attribute *attr, char *buf)
 765 {
 766         drive_info_struct *drv = to_drv(dev);
 767         struct ctlr_info *h = to_hba(drv->dev.parent);
 768         unsigned long flags;
 769         unsigned char lunid[8];
 770
 771         spin_lock_irqsave(&h->lock, flags);
 772         if (h->busy_configuring) {
 773                 spin_unlock_irqrestore(&h->lock, flags);
 774                 return -EBUSY;
 775         }
 776         if (!drv->heads) {
 777                 spin_unlock_irqrestore(&h->lock, flags);
 778                 return -ENOTTY;
 779         }
 780         memcpy(lunid, drv->LunID, sizeof(lunid));
 781         spin_unlock_irqrestore(&h->lock, flags);
 782         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 783                 lunid[0], lunid[1], lunid[2], lunid[3],
 784                 lunid[4], lunid[5], lunid[6], lunid[7]);
 785 }
 786 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 787
 788 static ssize_t cciss_show_raid_level(struct device *dev,
 789                                      struct device_attribute *attr, char *buf)
 790 {
 791         drive_info_struct *drv = to_drv(dev);
 792         struct ctlr_info *h = to_hba(drv->dev.parent);
 793         int raid;
 794         unsigned long flags;
 795
 796         spin_lock_irqsave(&h->lock, flags);
 797         if (h->busy_configuring) {
 798                 spin_unlock_irqrestore(&h->lock, flags);
 799                 return -EBUSY;
 800         }
 801         raid = drv->raid_level;
 802         spin_unlock_irqrestore(&h->lock, flags);
 803         if (raid < 0 || raid > RAID_UNKNOWN)
 804                 raid = RAID_UNKNOWN;
 805
 806         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 807                         raid_label[raid]);
 808 }
 809 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 810
 811 static ssize_t cciss_show_usage_count(struct device *dev,
 812                                       struct device_attribute *attr, char *buf)
 813 {
 814         drive_info_struct *drv = to_drv(dev);
 815         struct ctlr_info *h = to_hba(drv->dev.parent);
 816         unsigned long flags;
 817         int count;
 818
 819         spin_lock_irqsave(&h->lock, flags);
 820         if (h->busy_configuring) {
 821                 spin_unlock_irqrestore(&h->lock, flags);
 822                 return -EBUSY;
 823         }
 824         count = drv->usage_count;
 825         spin_unlock_irqrestore(&h->lock, flags);
 826         return snprintf(buf, 20, "%d\n", count);
 827 }
 828 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 829
 830 static struct attribute *cciss_host_attrs[] = {
 831         &dev_attr_rescan.attr,
 832         &dev_attr_resettable.attr,
 833         &dev_attr_transport_mode.attr,
 834         NULL
 835 };
 836
 837 static struct attribute_group cciss_host_attr_group = {
 838         .attrs = cciss_host_attrs,
 839 };
 840
 841 static const struct attribute_group *cciss_host_attr_groups[] = {
 842         &cciss_host_attr_group,
 843         NULL
 844 };
 845
 846 static struct device_type cciss_host_type = {
 847         .name           = "cciss_host",
 848         .groups         = cciss_host_attr_groups,
 849         .release        = cciss_hba_release,
 850 };
 851
 852 static struct attribute *cciss_dev_attrs[] = {
 853         &dev_attr_unique_id.attr,
 854         &dev_attr_model.attr,
 855         &dev_attr_vendor.attr,
 856         &dev_attr_rev.attr,
 857         &dev_attr_lunid.attr,
 858         &dev_attr_raid_level.attr,
 859         &dev_attr_usage_count.attr,
 860         NULL
 861 };
 862
 863 static struct attribute_group cciss_dev_attr_group = {
 864         .attrs = cciss_dev_attrs,
 865 };
 866
 867 static const struct attribute_group *cciss_dev_attr_groups[] = {
 868         &cciss_dev_attr_group,
 869         NULL
 870 };
 871
 872 static struct device_type cciss_dev_type = {
 873         .name           = "cciss_device",
 874         .groups         = cciss_dev_attr_groups,
 875         .release        = cciss_device_release,
 876 };
 877
 878 static struct bus_type cciss_bus_type = {
 879         .name           = "cciss",
 880 };
 881
 882 /*
 883  * cciss_hba_release is called when the reference count
 884  * of h->dev goes to zero.
 885  */
 886 static void cciss_hba_release(struct device *dev)
 887 {
 888         /*
 889          * nothing to do, but need this to avoid a warning
 890          * about not having a release handler from lib/kref.c.
 891          */
 892 }
 893
 894 /*
 895  * Initialize sysfs entry for each controller.  This sets up and registers
 896  * the 'cciss#' directory for each individual controller under
 897  * /sys/bus/pci/devices/<dev>/.
 898  */
 899 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 900 {
 901         device_initialize(&h->dev);
 902         h->dev.type = &cciss_host_type;
 903         h->dev.bus = &cciss_bus_type;
 904         dev_set_name(&h->dev, "%s", h->devname);
 905         h->dev.parent = &h->pdev->dev;
 906
 907         return device_add(&h->dev);
 908 }
 909
 910 /*
 911  * Remove sysfs entries for an hba.
 912  */
 913 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 914 {
 915         device_del(&h->dev);
 916         put_device(&h->dev); /* final put. */
 917 }
 918
 919 /* cciss_device_release is called when the reference count
 920  * of h->drv[x]dev goes to zero.
 921  */
 922 static void cciss_device_release(struct device *dev)
 923 {
 924         drive_info_struct *drv = to_drv(dev);
 925         kfree(drv);
 926 }
 927
 928 /*
 929  * Initialize sysfs for each logical drive.  This sets up and registers
 930  * the 'c#d#' directory for each individual logical drive under
 931  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 932  * /sys/block/cciss!c#d# to this entry.
 933  */
 934 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 935                                        int drv_index)
 936 {
 937         struct device *dev;
 938
 939         if (h->drv[drv_index]->device_initialized)
 940                 return 0;
 941
 942         dev = &h->drv[drv_index]->dev;
 943         device_initialize(dev);
 944         dev->type = &cciss_dev_type;
 945         dev->bus = &cciss_bus_type;
 946         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 947         dev->parent = &h->dev;
 948         h->drv[drv_index]->device_initialized = 1;
 949         return device_add(dev);
 950 }
 951
 952 /*
 953  * Remove sysfs entries for a logical drive.
 954  */
 955 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 956         int ctlr_exiting)
 957 {
 958         struct device *dev = &h->drv[drv_index]->dev;
 959
 960         /* special case for c*d0, we only destroy it on controller exit */
 961         if (drv_index == 0 && !ctlr_exiting)
 962                 return;
 963
 964         device_del(dev);
 965         put_device(dev); /* the "final" put. */
 966         h->drv[drv_index] = NULL;
 967 }
 968
 969 /*
 970  * For operations that cannot sleep, a command block is allocated at init,
 971  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 972  * which ones are free or in use.
 973  */
 974 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 975 {
 976         CommandList_struct *c;
 977         int i;
 978         u64bit temp64;
 979         dma_addr_t cmd_dma_handle, err_dma_handle;
 980
 981         do {
 982                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 983                 if (i == h->nr_cmds)
 984                         return NULL;
 985         } while (test_and_set_bit(i, h->cmd_pool_bits) != 0);
 986         c = h->cmd_pool + i;
 987         memset(c, 0, sizeof(CommandList_struct));
 988         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
 989         c->err_info = h->errinfo_pool + i;
 990         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 991         err_dma_handle = h->errinfo_pool_dhandle
 992             + i * sizeof(ErrorInfo_struct);
 993         h->nr_allocs++;
 994
 995         c->cmdindex = i;
 996
 997         INIT_LIST_HEAD(&c->list);
 998         c->busaddr = (__u32) cmd_dma_handle;
 999         temp64.val = (__u64) err_dma_handle;
1000         c->ErrDesc.Addr.lower = temp64.val32.lower;
1001         c->ErrDesc.Addr.upper = temp64.val32.upper;
1002         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
1003
1004         c->ctlr = h->ctlr;
1005         return c;
1006 }
1007
1008 /* allocate a command using pci_alloc_consistent, used for ioctls,
1009  * etc., not for the main i/o path.
1010  */
1011 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
1012 {
1013         CommandList_struct *c;
1014         u64bit temp64;
1015         dma_addr_t cmd_dma_handle, err_dma_handle;
1016
1017         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
1018                 sizeof(CommandList_struct), &cmd_dma_handle);
1019         if (c == NULL)
1020                 return NULL;
1021         memset(c, 0, sizeof(CommandList_struct));
1022
1023         c->cmdindex = -1;
1024
1025         c->err_info = (ErrorInfo_struct *)
1026             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
1027                     &err_dma_handle);
1028
1029         if (c->err_info == NULL) {
1030                 pci_free_consistent(h->pdev,
1031                         sizeof(CommandList_struct), c, cmd_dma_handle);
1032                 return NULL;
1033         }
1034         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
1035
1036         INIT_LIST_HEAD(&c->list);
1037         c->busaddr = (__u32) cmd_dma_handle;
1038         temp64.val = (__u64) err_dma_handle;
1039         c->ErrDesc.Addr.lower = temp64.val32.lower;
1040         c->ErrDesc.Addr.upper = temp64.val32.upper;
1041         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
1042
1043         c->ctlr = h->ctlr;
1044         return c;
1045 }
1046
1047 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
1048 {
1049         int i;
1050
1051         i = c - h->cmd_pool;
1052         clear_bit(i, h->cmd_pool_bits);
1053         h->nr_frees++;
1054 }
1055
1056 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
1057 {
1058         u64bit temp64;
1059
1060         temp64.val32.lower = c->ErrDesc.Addr.lower;
1061         temp64.val32.upper = c->ErrDesc.Addr.upper;
1062         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1063                             c->err_info, (dma_addr_t) temp64.val);
1064         pci_free_consistent(h->pdev, sizeof(CommandList_struct), c,
1065                 (dma_addr_t) cciss_tag_discard_error_bits(h, (u32) c->busaddr));
1066 }
1067
1068 static inline ctlr_info_t *get_host(struct gendisk *disk)
1069 {
1070         return disk->queue->queuedata;
1071 }
1072
1073 static inline drive_info_struct *get_drv(struct gendisk *disk)
1074 {
1075         return disk->private_data;
1076 }
1077
1078 /*
1079  * Open.  Make sure the device is really there.
1080  */
1081 static int cciss_open(struct block_device *bdev, fmode_t mode)
1082 {
1083         ctlr_info_t *h = get_host(bdev->bd_disk);
1084         drive_info_struct *drv = get_drv(bdev->bd_disk);
1085
1086         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1087         if (drv->busy_configuring)
1088                 return -EBUSY;
1089         /*
1090          * Root is allowed to open raw volume zero even if it's not configured
1091          * so array config can still work. Root is also allowed to open any
1092          * volume that has a LUN ID, so it can issue IOCTL to reread the
1093          * disk information.  I don't think I really like this
1094          * but I'm already using way to many device nodes to claim another one
1095          * for "raw controller".
1096          */
1097         if (drv->heads == 0) {
1098                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1099                         /* if not node 0 make sure it is a partition = 0 */
1100                         if (MINOR(bdev->bd_dev) & 0x0f) {
1101                                 return -ENXIO;
1102                                 /* if it is, make sure we have a LUN ID */
1103                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1104                                 sizeof(drv->LunID))) {
1105                                 return -ENXIO;
1106                         }
1107                 }
1108                 if (!capable(CAP_SYS_ADMIN))
1109                         return -EPERM;
1110         }
1111         drv->usage_count++;
1112         h->usage_count++;
1113         return 0;
1114 }
1115
1116 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1117 {
1118         int ret;
1119
1120         mutex_lock(&cciss_mutex);
1121         ret = cciss_open(bdev, mode);
1122         mutex_unlock(&cciss_mutex);
1123
1124         return ret;
1125 }
1126
1127 /*
1128  * Close.  Sync first.
1129  */
1130 static void cciss_release(struct gendisk *disk, fmode_t mode)
1131 {
1132         ctlr_info_t *h;
1133         drive_info_struct *drv;
1134
1135         mutex_lock(&cciss_mutex);
1136         h = get_host(disk);
1137         drv = get_drv(disk);
1138         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1139         drv->usage_count--;
1140         h->usage_count--;
1141         mutex_unlock(&cciss_mutex);
1142 }
1143
1144 #ifdef CONFIG_COMPAT
1145
1146 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1147                                   unsigned cmd, unsigned long arg);
1148 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1149                                       unsigned cmd, unsigned long arg);
1150
1151 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1152                               unsigned cmd, unsigned long arg)
1153 {
1154         switch (cmd) {
1155         case CCISS_GETPCIINFO:
1156         case CCISS_GETINTINFO:
1157         case CCISS_SETINTINFO:
1158         case CCISS_GETNODENAME:
1159         case CCISS_SETNODENAME:
1160         case CCISS_GETHEARTBEAT:
1161         case CCISS_GETBUSTYPES:
1162         case CCISS_GETFIRMVER:
1163         case CCISS_GETDRIVVER:
1164         case CCISS_REVALIDVOLS:
1165         case CCISS_DEREGDISK:
1166         case CCISS_REGNEWDISK:
1167         case CCISS_REGNEWD:
1168         case CCISS_RESCANDISK:
1169         case CCISS_GETLUNINFO:
1170                 return cciss_ioctl(bdev, mode, cmd, arg);
1171
1172         case CCISS_PASSTHRU32:
1173                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1174         case CCISS_BIG_PASSTHRU32:
1175                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1176
1177         default:
1178                 return -ENOIOCTLCMD;
1179         }
1180 }
1181
1182 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1183                                   unsigned cmd, unsigned long arg)
1184 {
1185         IOCTL32_Command_struct __user *arg32 =
1186             (IOCTL32_Command_struct __user *) arg;
1187         IOCTL_Command_struct arg64;
1188         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1189         int err;
1190         u32 cp;
1191
1192         memset(&arg64, 0, sizeof(arg64));
1193         err = 0;
1194         err |=
1195             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1196                            sizeof(arg64.LUN_info));
1197         err |=
1198             copy_from_user(&arg64.Request, &arg32->Request,
1199                            sizeof(arg64.Request));
1200         err |=
1201             copy_from_user(&arg64.error_info, &arg32->error_info,
1202                            sizeof(arg64.error_info));
1203         err |= get_user(arg64.buf_size, &arg32->buf_size);
1204         err |= get_user(cp, &arg32->buf);
1205         arg64.buf = compat_ptr(cp);
1206         err |= copy_to_user(p, &arg64, sizeof(arg64));
1207
1208         if (err)
1209                 return -EFAULT;
1210
1211         err = cciss_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1212         if (err)
1213                 return err;
1214         err |=
1215             copy_in_user(&arg32->error_info, &p->error_info,
1216                          sizeof(arg32->error_info));
1217         if (err)
1218                 return -EFAULT;
1219         return err;
1220 }
1221
1222 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1223                                       unsigned cmd, unsigned long arg)
1224 {
1225         BIG_IOCTL32_Command_struct __user *arg32 =
1226             (BIG_IOCTL32_Command_struct __user *) arg;
1227         BIG_IOCTL_Command_struct arg64;
1228         BIG_IOCTL_Command_struct __user *p =
1229             compat_alloc_user_space(sizeof(arg64));
1230         int err;
1231         u32 cp;
1232
1233         memset(&arg64, 0, sizeof(arg64));
1234         err = 0;
1235         err |=
1236             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1237                            sizeof(arg64.LUN_info));
1238         err |=
1239             copy_from_user(&arg64.Request, &arg32->Request,
1240                            sizeof(arg64.Request));
1241         err |=
1242             copy_from_user(&arg64.error_info, &arg32->error_info,
1243                            sizeof(arg64.error_info));
1244         err |= get_user(arg64.buf_size, &arg32->buf_size);
1245         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1246         err |= get_user(cp, &arg32->buf);
1247         arg64.buf = compat_ptr(cp);
1248         err |= copy_to_user(p, &arg64, sizeof(arg64));
1249
1250         if (err)
1251                 return -EFAULT;
1252
1253         err = cciss_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1254         if (err)
1255                 return err;
1256         err |=
1257             copy_in_user(&arg32->error_info, &p->error_info,
1258                          sizeof(arg32->error_info));
1259         if (err)
1260                 return -EFAULT;
1261         return err;
1262 }
1263 #endif
1264
1265 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1266 {
1267         drive_info_struct *drv = get_drv(bdev->bd_disk);
1268
1269         if (!drv->cylinders)
1270                 return -ENXIO;
1271
1272         geo->heads = drv->heads;
1273         geo->sectors = drv->sectors;
1274         geo->cylinders = drv->cylinders;
1275         return 0;
1276 }
1277
1278 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1279 {
1280         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1281                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1282                 (void)check_for_unit_attention(h, c);
1283 }
1284
1285 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1286 {
1287         cciss_pci_info_struct pciinfo;
1288
1289         if (!argp)
1290                 return -EINVAL;
1291         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1292         pciinfo.bus = h->pdev->bus->number;
1293         pciinfo.dev_fn = h->pdev->devfn;
1294         pciinfo.board_id = h->board_id;
1295         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1296                 return -EFAULT;
1297         return 0;
1298 }
1299
1300 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1301 {
1302         cciss_coalint_struct intinfo;
1303         unsigned long flags;
1304
1305         if (!argp)
1306                 return -EINVAL;
1307         spin_lock_irqsave(&h->lock, flags);
1308         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1309         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1310         spin_unlock_irqrestore(&h->lock, flags);
1311         if (copy_to_user
1312             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1313                 return -EFAULT;
1314         return 0;
1315 }
1316
1317 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1318 {
1319         cciss_coalint_struct intinfo;
1320         unsigned long flags;
1321         int i;
1322
1323         if (!argp)
1324                 return -EINVAL;
1325         if (!capable(CAP_SYS_ADMIN))
1326                 return -EPERM;
1327         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1328                 return -EFAULT;
1329         if ((intinfo.delay == 0) && (intinfo.count == 0))
1330                 return -EINVAL;
1331         spin_lock_irqsave(&h->lock, flags);
1332         /* Update the field, and then ring the doorbell */
1333         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1334         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1335         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1336
1337         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1338                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1339                         break;
1340                 udelay(1000); /* delay and try again */
1341         }
1342         spin_unlock_irqrestore(&h->lock, flags);
1343         if (i >= MAX_IOCTL_CONFIG_WAIT)
1344                 return -EAGAIN;
1345         return 0;
1346 }
1347
1348 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1349 {
1350         NodeName_type NodeName;
1351         unsigned long flags;
1352         int i;
1353
1354         if (!argp)
1355                 return -EINVAL;
1356         spin_lock_irqsave(&h->lock, flags);
1357         for (i = 0; i < 16; i++)
1358                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1359         spin_unlock_irqrestore(&h->lock, flags);
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         unsigned long flags;
1397
1398         if (!argp)
1399                 return -EINVAL;
1400         spin_lock_irqsave(&h->lock, flags);
1401         heartbeat = readl(&h->cfgtable->HeartBeat);
1402         spin_unlock_irqrestore(&h->lock, flags);
1403         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1404                 return -EFAULT;
1405         return 0;
1406 }
1407
1408 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1409 {
1410         BusTypes_type BusTypes;
1411         unsigned long flags;
1412
1413         if (!argp)
1414                 return -EINVAL;
1415         spin_lock_irqsave(&h->lock, flags);
1416         BusTypes = readl(&h->cfgtable->BusTypes);
1417         spin_unlock_irqrestore(&h->lock, flags);
1418         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1419                 return -EFAULT;
1420         return 0;
1421 }
1422
1423 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1424 {
1425         FirmwareVer_type firmware;
1426
1427         if (!argp)
1428                 return -EINVAL;
1429         memcpy(firmware, h->firm_ver, 4);
1430
1431         if (copy_to_user
1432             (argp, firmware, sizeof(FirmwareVer_type)))
1433                 return -EFAULT;
1434         return 0;
1435 }
1436
1437 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1438 {
1439         DriverVer_type DriverVer = DRIVER_VERSION;
1440
1441         if (!argp)
1442                 return -EINVAL;
1443         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1444                 return -EFAULT;
1445         return 0;
1446 }
1447
1448 static int cciss_getluninfo(ctlr_info_t *h,
1449         struct gendisk *disk, void __user *argp)
1450 {
1451         LogvolInfo_struct luninfo;
1452         drive_info_struct *drv = get_drv(disk);
1453
1454         if (!argp)
1455                 return -EINVAL;
1456         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1457         luninfo.num_opens = drv->usage_count;
1458         luninfo.num_parts = 0;
1459         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1460                 return -EFAULT;
1461         return 0;
1462 }
1463
1464 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1465 {
1466         IOCTL_Command_struct iocommand;
1467         CommandList_struct *c;
1468         char *buff = NULL;
1469         u64bit temp64;
1470         DECLARE_COMPLETION_ONSTACK(wait);
1471
1472         if (!argp)
1473                 return -EINVAL;
1474
1475         if (!capable(CAP_SYS_RAWIO))
1476                 return -EPERM;
1477
1478         if (copy_from_user
1479             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1480                 return -EFAULT;
1481         if ((iocommand.buf_size < 1) &&
1482             (iocommand.Request.Type.Direction != XFER_NONE)) {
1483                 return -EINVAL;
1484         }
1485         if (iocommand.buf_size > 0) {
1486                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1487                 if (buff == NULL)
1488                         return -EFAULT;
1489         }
1490         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1491                 /* Copy the data into the buffer we created */
1492                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1493                         kfree(buff);
1494                         return -EFAULT;
1495                 }
1496         } else {
1497                 memset(buff, 0, iocommand.buf_size);
1498         }
1499         c = cmd_special_alloc(h);
1500         if (!c) {
1501                 kfree(buff);
1502                 return -ENOMEM;
1503         }
1504         /* Fill in the command type */
1505         c->cmd_type = CMD_IOCTL_PEND;
1506         /* Fill in Command Header */
1507         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1508         if (iocommand.buf_size > 0) { /* buffer to fill */
1509                 c->Header.SGList = 1;
1510                 c->Header.SGTotal = 1;
1511         } else { /* no buffers to fill */
1512                 c->Header.SGList = 0;
1513                 c->Header.SGTotal = 0;
1514         }
1515         c->Header.LUN = iocommand.LUN_info;
1516         /* use the kernel address the cmd block for tag */
1517         c->Header.Tag.lower = c->busaddr;
1518
1519         /* Fill in Request block */
1520         c->Request = iocommand.Request;
1521
1522         /* Fill in the scatter gather information */
1523         if (iocommand.buf_size > 0) {
1524                 temp64.val = pci_map_single(h->pdev, buff,
1525                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1526                 c->SG[0].Addr.lower = temp64.val32.lower;
1527                 c->SG[0].Addr.upper = temp64.val32.upper;
1528                 c->SG[0].Len = iocommand.buf_size;
1529                 c->SG[0].Ext = 0;  /* we are not chaining */
1530         }
1531         c->waiting = &wait;
1532
1533         enqueue_cmd_and_start_io(h, c);
1534         wait_for_completion(&wait);
1535
1536         /* unlock the buffers from DMA */
1537         temp64.val32.lower = c->SG[0].Addr.lower;
1538         temp64.val32.upper = c->SG[0].Addr.upper;
1539         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1540                          PCI_DMA_BIDIRECTIONAL);
1541         check_ioctl_unit_attention(h, c);
1542
1543         /* Copy the error information out */
1544         iocommand.error_info = *(c->err_info);
1545         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1546                 kfree(buff);
1547                 cmd_special_free(h, c);
1548                 return -EFAULT;
1549         }
1550
1551         if (iocommand.Request.Type.Direction == XFER_READ) {
1552                 /* Copy the data out of the buffer we created */
1553                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1554                         kfree(buff);
1555                         cmd_special_free(h, c);
1556                         return -EFAULT;
1557                 }
1558         }
1559         kfree(buff);
1560         cmd_special_free(h, c);
1561         return 0;
1562 }
1563
1564 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1565 {
1566         BIG_IOCTL_Command_struct *ioc;
1567         CommandList_struct *c;
1568         unsigned char **buff = NULL;
1569         int *buff_size = NULL;
1570         u64bit temp64;
1571         BYTE sg_used = 0;
1572         int status = 0;
1573         int i;
1574         DECLARE_COMPLETION_ONSTACK(wait);
1575         __u32 left;
1576         __u32 sz;
1577         BYTE __user *data_ptr;
1578
1579         if (!argp)
1580                 return -EINVAL;
1581         if (!capable(CAP_SYS_RAWIO))
1582                 return -EPERM;
1583         ioc = kmalloc(sizeof(*ioc), GFP_KERNEL);
1584         if (!ioc) {
1585                 status = -ENOMEM;
1586                 goto cleanup1;
1587         }
1588         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1589                 status = -EFAULT;
1590                 goto cleanup1;
1591         }
1592         if ((ioc->buf_size < 1) &&
1593             (ioc->Request.Type.Direction != XFER_NONE)) {
1594                 status = -EINVAL;
1595                 goto cleanup1;
1596         }
1597         /* Check kmalloc limits  using all SGs */
1598         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1599                 status = -EINVAL;
1600                 goto cleanup1;
1601         }
1602         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1603                 status = -EINVAL;
1604                 goto cleanup1;
1605         }
1606         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1607         if (!buff) {
1608                 status = -ENOMEM;
1609                 goto cleanup1;
1610         }
1611         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1612         if (!buff_size) {
1613                 status = -ENOMEM;
1614                 goto cleanup1;
1615         }
1616         left = ioc->buf_size;
1617         data_ptr = ioc->buf;
1618         while (left) {
1619                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1620                 buff_size[sg_used] = sz;
1621                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1622                 if (buff[sg_used] == NULL) {
1623                         status = -ENOMEM;
1624                         goto cleanup1;
1625                 }
1626                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1627                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1628                                 status = -EFAULT;
1629                                 goto cleanup1;
1630                         }
1631                 } else {
1632                         memset(buff[sg_used], 0, sz);
1633                 }
1634                 left -= sz;
1635                 data_ptr += sz;
1636                 sg_used++;
1637         }
1638         c = cmd_special_alloc(h);
1639         if (!c) {
1640                 status = -ENOMEM;
1641                 goto cleanup1;
1642         }
1643         c->cmd_type = CMD_IOCTL_PEND;
1644         c->Header.ReplyQueue = 0;
1645         c->Header.SGList = sg_used;
1646         c->Header.SGTotal = sg_used;
1647         c->Header.LUN = ioc->LUN_info;
1648         c->Header.Tag.lower = c->busaddr;
1649
1650         c->Request = ioc->Request;
1651         for (i = 0; i < sg_used; i++) {
1652                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1653                                     PCI_DMA_BIDIRECTIONAL);
1654                 c->SG[i].Addr.lower = temp64.val32.lower;
1655                 c->SG[i].Addr.upper = temp64.val32.upper;
1656                 c->SG[i].Len = buff_size[i];
1657                 c->SG[i].Ext = 0;       /* we are not chaining */
1658         }
1659         c->waiting = &wait;
1660         enqueue_cmd_and_start_io(h, c);
1661         wait_for_completion(&wait);
1662         /* unlock the buffers from DMA */
1663         for (i = 0; i < sg_used; i++) {
1664                 temp64.val32.lower = c->SG[i].Addr.lower;
1665                 temp64.val32.upper = c->SG[i].Addr.upper;
1666                 pci_unmap_single(h->pdev,
1667                         (dma_addr_t) temp64.val, buff_size[i],
1668                         PCI_DMA_BIDIRECTIONAL);
1669         }
1670         check_ioctl_unit_attention(h, c);
1671         /* Copy the error information out */
1672         ioc->error_info = *(c->err_info);
1673         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1674                 cmd_special_free(h, c);
1675                 status = -EFAULT;
1676                 goto cleanup1;
1677         }
1678         if (ioc->Request.Type.Direction == XFER_READ) {
1679                 /* Copy the data out of the buffer we created */
1680                 BYTE __user *ptr = ioc->buf;
1681                 for (i = 0; i < sg_used; i++) {
1682                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1683                                 cmd_special_free(h, c);
1684                                 status = -EFAULT;
1685                                 goto cleanup1;
1686                         }
1687                         ptr += buff_size[i];
1688                 }
1689         }
1690         cmd_special_free(h, c);
1691         status = 0;
1692 cleanup1:
1693         if (buff) {
1694                 for (i = 0; i < sg_used; i++)
1695                         kfree(buff[i]);
1696                 kfree(buff);
1697         }
1698         kfree(buff_size);
1699         kfree(ioc);
1700         return status;
1701 }
1702
1703 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1704         unsigned int cmd, unsigned long arg)
1705 {
1706         struct gendisk *disk = bdev->bd_disk;
1707         ctlr_info_t *h = get_host(disk);
1708         void __user *argp = (void __user *)arg;
1709
1710         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1711                 cmd, arg);
1712         switch (cmd) {
1713         case CCISS_GETPCIINFO:
1714                 return cciss_getpciinfo(h, argp);
1715         case CCISS_GETINTINFO:
1716                 return cciss_getintinfo(h, argp);
1717         case CCISS_SETINTINFO:
1718                 return cciss_setintinfo(h, argp);
1719         case CCISS_GETNODENAME:
1720                 return cciss_getnodename(h, argp);
1721         case CCISS_SETNODENAME:
1722                 return cciss_setnodename(h, argp);
1723         case CCISS_GETHEARTBEAT:
1724                 return cciss_getheartbeat(h, argp);
1725         case CCISS_GETBUSTYPES:
1726                 return cciss_getbustypes(h, argp);
1727         case CCISS_GETFIRMVER:
1728                 return cciss_getfirmver(h, argp);
1729         case CCISS_GETDRIVVER:
1730                 return cciss_getdrivver(h, argp);
1731         case CCISS_DEREGDISK:
1732         case CCISS_REGNEWD:
1733         case CCISS_REVALIDVOLS:
1734                 return rebuild_lun_table(h, 0, 1);
1735         case CCISS_GETLUNINFO:
1736                 return cciss_getluninfo(h, disk, argp);
1737         case CCISS_PASSTHRU:
1738                 return cciss_passthru(h, argp);
1739         case CCISS_BIG_PASSTHRU:
1740                 return cciss_bigpassthru(h, argp);
1741
1742         /* scsi_cmd_blk_ioctl handles these, below, though some are not */
1743         /* very meaningful for cciss.  SG_IO is the main one people want. */
1744
1745         case SG_GET_VERSION_NUM:
1746         case SG_SET_TIMEOUT:
1747         case SG_GET_TIMEOUT:
1748         case SG_GET_RESERVED_SIZE:
1749         case SG_SET_RESERVED_SIZE:
1750         case SG_EMULATED_HOST:
1751         case SG_IO:
1752         case SCSI_IOCTL_SEND_COMMAND:
1753                 return scsi_cmd_blk_ioctl(bdev, mode, cmd, argp);
1754
1755         /* scsi_cmd_blk_ioctl would normally handle these, below, but */
1756         /* they aren't a good fit for cciss, as CD-ROMs are */
1757         /* not supported, and we don't have any bus/target/lun */
1758         /* which we present to the kernel. */
1759
1760         case CDROM_SEND_PACKET:
1761         case CDROMCLOSETRAY:
1762         case CDROMEJECT:
1763         case SCSI_IOCTL_GET_IDLUN:
1764         case SCSI_IOCTL_GET_BUS_NUMBER:
1765         default:
1766                 return -ENOTTY;
1767         }
1768 }
1769
1770 static void cciss_check_queues(ctlr_info_t *h)
1771 {
1772         int start_queue = h->next_to_run;
1773         int i;
1774
1775         /* check to see if we have maxed out the number of commands that can
1776          * be placed on the queue.  If so then exit.  We do this check here
1777          * in case the interrupt we serviced was from an ioctl and did not
1778          * free any new commands.
1779          */
1780         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1781                 return;
1782
1783         /* We have room on the queue for more commands.  Now we need to queue
1784          * them up.  We will also keep track of the next queue to run so
1785          * that every queue gets a chance to be started first.
1786          */
1787         for (i = 0; i < h->highest_lun + 1; i++) {
1788                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1789                 /* make sure the disk has been added and the drive is real
1790                  * because this can be called from the middle of init_one.
1791                  */
1792                 if (!h->drv[curr_queue])
1793                         continue;
1794                 if (!(h->drv[curr_queue]->queue) ||
1795                         !(h->drv[curr_queue]->heads))
1796                         continue;
1797                 blk_start_queue(h->gendisk[curr_queue]->queue);
1798
1799                 /* check to see if we have maxed out the number of commands
1800                  * that can be placed on the queue.
1801                  */
1802                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1803                         if (curr_queue == start_queue) {
1804                                 h->next_to_run =
1805                                     (start_queue + 1) % (h->highest_lun + 1);
1806                                 break;
1807                         } else {
1808                                 h->next_to_run = curr_queue;
1809                                 break;
1810                         }
1811                 }
1812         }
1813 }
1814
1815 static void cciss_softirq_done(struct request *rq)
1816 {
1817         CommandList_struct *c = rq->completion_data;
1818         ctlr_info_t *h = hba[c->ctlr];
1819         SGDescriptor_struct *curr_sg = c->SG;
1820         u64bit temp64;
1821         unsigned long flags;
1822         int i, ddir;
1823         int sg_index = 0;
1824
1825         if (c->Request.Type.Direction == XFER_READ)
1826                 ddir = PCI_DMA_FROMDEVICE;
1827         else
1828                 ddir = PCI_DMA_TODEVICE;
1829
1830         /* command did not need to be retried */
1831         /* unmap the DMA mapping for all the scatter gather elements */
1832         for (i = 0; i < c->Header.SGList; i++) {
1833                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1834                         cciss_unmap_sg_chain_block(h, c);
1835                         /* Point to the next block */
1836                         curr_sg = h->cmd_sg_list[c->cmdindex];
1837                         sg_index = 0;
1838                 }
1839                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1840                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1841                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1842                                 ddir);
1843                 ++sg_index;
1844         }
1845
1846         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1847
1848         /* set the residual count for pc requests */
1849         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1850                 rq->resid_len = c->err_info->ResidualCnt;
1851
1852         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1853
1854         spin_lock_irqsave(&h->lock, flags);
1855         cmd_free(h, c);
1856         cciss_check_queues(h);
1857         spin_unlock_irqrestore(&h->lock, flags);
1858 }
1859
1860 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1861         unsigned char scsi3addr[], uint32_t log_unit)
1862 {
1863         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1864                 sizeof(h->drv[log_unit]->LunID));
1865 }
1866
1867 /* This function gets the SCSI vendor, model, and revision of a logical drive
1868  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1869  * they cannot be read.
1870  */
1871 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1872                                    char *vendor, char *model, char *rev)
1873 {
1874         int rc;
1875         InquiryData_struct *inq_buf;
1876         unsigned char scsi3addr[8];
1877
1878         *vendor = '\0';
1879         *model = '\0';
1880         *rev = '\0';
1881
1882         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1883         if (!inq_buf)
1884                 return;
1885
1886         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1887         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1888                         scsi3addr, TYPE_CMD);
1889         if (rc == IO_OK) {
1890                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1891                 vendor[VENDOR_LEN] = '\0';
1892                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1893                 model[MODEL_LEN] = '\0';
1894                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1895                 rev[REV_LEN] = '\0';
1896         }
1897
1898         kfree(inq_buf);
1899         return;
1900 }
1901
1902 /* This function gets the serial number of a logical drive via
1903  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1904  * number cannot be had, for whatever reason, 16 bytes of 0xff
1905  * are returned instead.
1906  */
1907 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1908                                 unsigned char *serial_no, int buflen)
1909 {
1910 #define PAGE_83_INQ_BYTES 64
1911         int rc;
1912         unsigned char *buf;
1913         unsigned char scsi3addr[8];
1914
1915         if (buflen > 16)
1916                 buflen = 16;
1917         memset(serial_no, 0xff, buflen);
1918         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1919         if (!buf)
1920                 return;
1921         memset(serial_no, 0, buflen);
1922         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1923         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1924                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1925         if (rc == IO_OK)
1926                 memcpy(serial_no, &buf[8], buflen);
1927         kfree(buf);
1928         return;
1929 }
1930
1931 /*
1932  * cciss_add_disk sets up the block device queue for a logical drive
1933  */
1934 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1935                                 int drv_index)
1936 {
1937         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1938         if (!disk->queue)
1939                 goto init_queue_failure;
1940         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1941         disk->major = h->major;
1942         disk->first_minor = drv_index << NWD_SHIFT;
1943         disk->fops = &cciss_fops;
1944         if (cciss_create_ld_sysfs_entry(h, drv_index))
1945                 goto cleanup_queue;
1946         disk->private_data = h->drv[drv_index];
1947         disk->driverfs_dev = &h->drv[drv_index]->dev;
1948
1949         /* Set up queue information */
1950         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1951
1952         /* This is a hardware imposed limit. */
1953         blk_queue_max_segments(disk->queue, h->maxsgentries);
1954
1955         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1956
1957         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1958
1959         disk->queue->queuedata = h;
1960
1961         blk_queue_logical_block_size(disk->queue,
1962                                      h->drv[drv_index]->block_size);
1963
1964         /* Make sure all queue data is written out before */
1965         /* setting h->drv[drv_index]->queue, as setting this */
1966         /* allows the interrupt handler to start the queue */
1967         wmb();
1968         h->drv[drv_index]->queue = disk->queue;
1969         add_disk(disk);
1970         return 0;
1971
1972 cleanup_queue:
1973         blk_cleanup_queue(disk->queue);
1974         disk->queue = NULL;
1975 init_queue_failure:
1976         return -1;
1977 }
1978
1979 /* This function will check the usage_count of the drive to be updated/added.
1980  * If the usage_count is zero and it is a heretofore unknown drive, or,
1981  * the drive's capacity, geometry, or serial number has changed,
1982  * then the drive information will be updated and the disk will be
1983  * re-registered with the kernel.  If these conditions don't hold,
1984  * then it will be left alone for the next reboot.  The exception to this
1985  * is disk 0 which will always be left registered with the kernel since it
1986  * is also the controller node.  Any changes to disk 0 will show up on
1987  * the next reboot.
1988  */
1989 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1990         int first_time, int via_ioctl)
1991 {
1992         struct gendisk *disk;
1993         InquiryData_struct *inq_buff = NULL;
1994         unsigned int block_size;
1995         sector_t total_size;
1996         unsigned long flags = 0;
1997         int ret = 0;
1998         drive_info_struct *drvinfo;
1999
2000         /* Get information about the disk and modify the driver structure */
2001         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2002         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
2003         if (inq_buff == NULL || drvinfo == NULL)
2004                 goto mem_msg;
2005
2006         /* testing to see if 16-byte CDBs are already being used */
2007         if (h->cciss_read == CCISS_READ_16) {
2008                 cciss_read_capacity_16(h, drv_index,
2009                         &total_size, &block_size);
2010
2011         } else {
2012                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
2013                 /* if read_capacity returns all F's this volume is >2TB */
2014                 /* in size so we switch to 16-byte CDB's for all */
2015                 /* read/write ops */
2016                 if (total_size == 0xFFFFFFFFULL) {
2017                         cciss_read_capacity_16(h, drv_index,
2018                         &total_size, &block_size);
2019                         h->cciss_read = CCISS_READ_16;
2020                         h->cciss_write = CCISS_WRITE_16;
2021                 } else {
2022                         h->cciss_read = CCISS_READ_10;
2023                         h->cciss_write = CCISS_WRITE_10;
2024                 }
2025         }
2026
2027         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
2028                                inq_buff, drvinfo);
2029         drvinfo->block_size = block_size;
2030         drvinfo->nr_blocks = total_size + 1;
2031
2032         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
2033                                 drvinfo->model, drvinfo->rev);
2034         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
2035                         sizeof(drvinfo->serial_no));
2036         /* Save the lunid in case we deregister the disk, below. */
2037         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
2038                 sizeof(drvinfo->LunID));
2039
2040         /* Is it the same disk we already know, and nothing's changed? */
2041         if (h->drv[drv_index]->raid_level != -1 &&
2042                 ((memcmp(drvinfo->serial_no,
2043                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2044                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2045                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2046                 drvinfo->heads == h->drv[drv_index]->heads &&
2047                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2048                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2049                         /* The disk is unchanged, nothing to update */
2050                         goto freeret;
2051
2052         /* If we get here it's not the same disk, or something's changed,
2053          * so we need to * deregister it, and re-register it, if it's not
2054          * in use.
2055          * If the disk already exists then deregister it before proceeding
2056          * (unless it's the first disk (for the controller node).
2057          */
2058         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2059                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2060                 spin_lock_irqsave(&h->lock, flags);
2061                 h->drv[drv_index]->busy_configuring = 1;
2062                 spin_unlock_irqrestore(&h->lock, flags);
2063
2064                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2065                  * which keeps the interrupt handler from starting
2066                  * the queue.
2067                  */
2068                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2069         }
2070
2071         /* If the disk is in use return */
2072         if (ret)
2073                 goto freeret;
2074
2075         /* Save the new information from cciss_geometry_inquiry
2076          * and serial number inquiry.  If the disk was deregistered
2077          * above, then h->drv[drv_index] will be NULL.
2078          */
2079         if (h->drv[drv_index] == NULL) {
2080                 drvinfo->device_initialized = 0;
2081                 h->drv[drv_index] = drvinfo;
2082                 drvinfo = NULL; /* so it won't be freed below. */
2083         } else {
2084                 /* special case for cxd0 */
2085                 h->drv[drv_index]->block_size = drvinfo->block_size;
2086                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2087                 h->drv[drv_index]->heads = drvinfo->heads;
2088                 h->drv[drv_index]->sectors = drvinfo->sectors;
2089                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2090                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2091                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2092                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2093                         VENDOR_LEN + 1);
2094                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2095                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2096         }
2097
2098         ++h->num_luns;
2099         disk = h->gendisk[drv_index];
2100         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2101
2102         /* If it's not disk 0 (drv_index != 0)
2103          * or if it was disk 0, but there was previously
2104          * no actual corresponding configured logical drive
2105          * (raid_leve == -1) then we want to update the
2106          * logical drive's information.
2107          */
2108         if (drv_index || first_time) {
2109                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2110                         cciss_free_gendisk(h, drv_index);
2111                         cciss_free_drive_info(h, drv_index);
2112                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2113                                 drv_index);
2114                         --h->num_luns;
2115                 }
2116         }
2117
2118 freeret:
2119         kfree(inq_buff);
2120         kfree(drvinfo);
2121         return;
2122 mem_msg:
2123         dev_err(&h->pdev->dev, "out of memory\n");
2124         goto freeret;
2125 }
2126
2127 /* This function will find the first index of the controllers drive array
2128  * that has a null drv pointer and allocate the drive info struct and
2129  * will return that index   This is where new drives will be added.
2130  * If the index to be returned is greater than the highest_lun index for
2131  * the controller then highest_lun is set * to this new index.
2132  * If there are no available indexes or if tha allocation fails, then -1
2133  * is returned.  * "controller_node" is used to know if this is a real
2134  * logical drive, or just the controller node, which determines if this
2135  * counts towards highest_lun.
2136  */
2137 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2138 {
2139         int i;
2140         drive_info_struct *drv;
2141
2142         /* Search for an empty slot for our drive info */
2143         for (i = 0; i < CISS_MAX_LUN; i++) {
2144
2145                 /* if not cxd0 case, and it's occupied, skip it. */
2146                 if (h->drv[i] && i != 0)
2147                         continue;
2148                 /*
2149                  * If it's cxd0 case, and drv is alloc'ed already, and a
2150                  * disk is configured there, skip it.
2151                  */
2152                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2153                         continue;
2154
2155                 /*
2156                  * We've found an empty slot.  Update highest_lun
2157                  * provided this isn't just the fake cxd0 controller node.
2158                  */
2159                 if (i > h->highest_lun && !controller_node)
2160                         h->highest_lun = i;
2161
2162                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2163                 if (i == 0 && h->drv[i] != NULL)
2164                         return i;
2165
2166                 /*
2167                  * Found an empty slot, not already alloc'ed.  Allocate it.
2168                  * Mark it with raid_level == -1, so we know it's new later on.
2169                  */
2170                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2171                 if (!drv)
2172                         return -1;
2173                 drv->raid_level = -1; /* so we know it's new */
2174                 h->drv[i] = drv;
2175                 return i;
2176         }
2177         return -1;
2178 }
2179
2180 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2181 {
2182         kfree(h->drv[drv_index]);
2183         h->drv[drv_index] = NULL;
2184 }
2185
2186 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2187 {
2188         put_disk(h->gendisk[drv_index]);
2189         h->gendisk[drv_index] = NULL;
2190 }
2191
2192 /* cciss_add_gendisk finds a free hba[]->drv structure
2193  * and allocates a gendisk if needed, and sets the lunid
2194  * in the drvinfo structure.   It returns the index into
2195  * the ->drv[] array, or -1 if none are free.
2196  * is_controller_node indicates whether highest_lun should
2197  * count this disk, or if it's only being added to provide
2198  * a means to talk to the controller in case no logical
2199  * drives have yet been configured.
2200  */
2201 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2202         int controller_node)
2203 {
2204         int drv_index;
2205
2206         drv_index = cciss_alloc_drive_info(h, controller_node);
2207         if (drv_index == -1)
2208                 return -1;
2209
2210         /*Check if the gendisk needs to be allocated */
2211         if (!h->gendisk[drv_index]) {
2212                 h->gendisk[drv_index] =
2213                         alloc_disk(1 << NWD_SHIFT);
2214                 if (!h->gendisk[drv_index]) {
2215                         dev_err(&h->pdev->dev,
2216                                 "could not allocate a new disk %d\n",
2217                                 drv_index);
2218                         goto err_free_drive_info;
2219                 }
2220         }
2221         memcpy(h->drv[drv_index]->LunID, lunid,
2222                 sizeof(h->drv[drv_index]->LunID));
2223         if (cciss_create_ld_sysfs_entry(h, drv_index))
2224                 goto err_free_disk;
2225         /* Don't need to mark this busy because nobody */
2226         /* else knows about this disk yet to contend */
2227         /* for access to it. */
2228         h->drv[drv_index]->busy_configuring = 0;
2229         wmb();
2230         return drv_index;
2231
2232 err_free_disk:
2233         cciss_free_gendisk(h, drv_index);
2234 err_free_drive_info:
2235         cciss_free_drive_info(h, drv_index);
2236         return -1;
2237 }
2238
2239 /* This is for the special case of a controller which
2240  * has no logical drives.  In this case, we still need
2241  * to register a disk so the controller can be accessed
2242  * by the Array Config Utility.
2243  */
2244 static void cciss_add_controller_node(ctlr_info_t *h)
2245 {
2246         struct gendisk *disk;
2247         int drv_index;
2248
2249         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2250                 return;
2251
2252         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2253         if (drv_index == -1)
2254                 goto error;
2255         h->drv[drv_index]->block_size = 512;
2256         h->drv[drv_index]->nr_blocks = 0;
2257         h->drv[drv_index]->heads = 0;
2258         h->drv[drv_index]->sectors = 0;
2259         h->drv[drv_index]->cylinders = 0;
2260         h->drv[drv_index]->raid_level = -1;
2261         memset(h->drv[drv_index]->serial_no, 0, 16);
2262         disk = h->gendisk[drv_index];
2263         if (cciss_add_disk(h, disk, drv_index) == 0)
2264                 return;
2265         cciss_free_gendisk(h, drv_index);
2266         cciss_free_drive_info(h, drv_index);
2267 error:
2268         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2269         return;
2270 }
2271
2272 /* This function will add and remove logical drives from the Logical
2273  * drive array of the controller and maintain persistency of ordering
2274  * so that mount points are preserved until the next reboot.  This allows
2275  * for the removal of logical drives in the middle of the drive array
2276  * without a re-ordering of those drives.
2277  * INPUT
2278  * h            = The controller to perform the operations on
2279  */
2280 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2281         int via_ioctl)
2282 {
2283         int num_luns;
2284         ReportLunData_struct *ld_buff = NULL;
2285         int return_code;
2286         int listlength = 0;
2287         int i;
2288         int drv_found;
2289         int drv_index = 0;
2290         unsigned char lunid[8] = CTLR_LUNID;
2291         unsigned long flags;
2292
2293         if (!capable(CAP_SYS_RAWIO))
2294                 return -EPERM;
2295
2296         /* Set busy_configuring flag for this operation */
2297         spin_lock_irqsave(&h->lock, flags);
2298         if (h->busy_configuring) {
2299                 spin_unlock_irqrestore(&h->lock, flags);
2300                 return -EBUSY;
2301         }
2302         h->busy_configuring = 1;
2303         spin_unlock_irqrestore(&h->lock, flags);
2304
2305         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2306         if (ld_buff == NULL)
2307                 goto mem_msg;
2308
2309         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2310                                       sizeof(ReportLunData_struct),
2311                                       0, CTLR_LUNID, TYPE_CMD);
2312
2313         if (return_code == IO_OK)
2314                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2315         else {  /* reading number of logical volumes failed */
2316                 dev_warn(&h->pdev->dev,
2317                         "report logical volume command failed\n");
2318                 listlength = 0;
2319                 goto freeret;
2320         }
2321
2322         num_luns = listlength / 8;      /* 8 bytes per entry */
2323         if (num_luns > CISS_MAX_LUN) {
2324                 num_luns = CISS_MAX_LUN;
2325                 dev_warn(&h->pdev->dev, "more luns configured"
2326                        " on controller than can be handled by"
2327                        " this driver.\n");
2328         }
2329
2330         if (num_luns == 0)
2331                 cciss_add_controller_node(h);
2332
2333         /* Compare controller drive array to driver's drive array
2334          * to see if any drives are missing on the controller due
2335          * to action of Array Config Utility (user deletes drive)
2336          * and deregister logical drives which have disappeared.
2337          */
2338         for (i = 0; i <= h->highest_lun; i++) {
2339                 int j;
2340                 drv_found = 0;
2341
2342                 /* skip holes in the array from already deleted drives */
2343                 if (h->drv[i] == NULL)
2344                         continue;
2345
2346                 for (j = 0; j < num_luns; j++) {
2347                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2348                         if (memcmp(h->drv[i]->LunID, lunid,
2349                                 sizeof(lunid)) == 0) {
2350                                 drv_found = 1;
2351                                 break;
2352                         }
2353                 }
2354                 if (!drv_found) {
2355                         /* Deregister it from the OS, it's gone. */
2356                         spin_lock_irqsave(&h->lock, flags);
2357                         h->drv[i]->busy_configuring = 1;
2358                         spin_unlock_irqrestore(&h->lock, flags);
2359                         return_code = deregister_disk(h, i, 1, via_ioctl);
2360                         if (h->drv[i] != NULL)
2361                                 h->drv[i]->busy_configuring = 0;
2362                 }
2363         }
2364
2365         /* Compare controller drive array to driver's drive array.
2366          * Check for updates in the drive information and any new drives
2367          * on the controller due to ACU adding logical drives, or changing
2368          * a logical drive's size, etc.  Reregister any new/changed drives
2369          */
2370         for (i = 0; i < num_luns; i++) {
2371                 int j;
2372
2373                 drv_found = 0;
2374
2375                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2376                 /* Find if the LUN is already in the drive array
2377                  * of the driver.  If so then update its info
2378                  * if not in use.  If it does not exist then find
2379                  * the first free index and add it.
2380                  */
2381                 for (j = 0; j <= h->highest_lun; j++) {
2382                         if (h->drv[j] != NULL &&
2383                                 memcmp(h->drv[j]->LunID, lunid,
2384                                         sizeof(h->drv[j]->LunID)) == 0) {
2385                                 drv_index = j;
2386                                 drv_found = 1;
2387                                 break;
2388                         }
2389                 }
2390
2391                 /* check if the drive was found already in the array */
2392                 if (!drv_found) {
2393                         drv_index = cciss_add_gendisk(h, lunid, 0);
2394                         if (drv_index == -1)
2395                                 goto freeret;
2396                 }
2397                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2398         }               /* end for */
2399
2400 freeret:
2401         kfree(ld_buff);
2402         h->busy_configuring = 0;
2403         /* We return -1 here to tell the ACU that we have registered/updated
2404          * all of the drives that we can and to keep it from calling us
2405          * additional times.
2406          */
2407         return -1;
2408 mem_msg:
2409         dev_err(&h->pdev->dev, "out of memory\n");
2410         h->busy_configuring = 0;
2411         goto freeret;
2412 }
2413
2414 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2415 {
2416         /* zero out the disk size info */
2417         drive_info->nr_blocks = 0;
2418         drive_info->block_size = 0;
2419         drive_info->heads = 0;
2420         drive_info->sectors = 0;
2421         drive_info->cylinders = 0;
2422         drive_info->raid_level = -1;
2423         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2424         memset(drive_info->model, 0, sizeof(drive_info->model));
2425         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2426         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2427         /*
2428          * don't clear the LUNID though, we need to remember which
2429          * one this one is.
2430          */
2431 }
2432
2433 /* This function will deregister the disk and it's queue from the
2434  * kernel.  It must be called with the controller lock held and the
2435  * drv structures busy_configuring flag set.  It's parameters are:
2436  *
2437  * disk = This is the disk to be deregistered
2438  * drv  = This is the drive_info_struct associated with the disk to be
2439  *        deregistered.  It contains information about the disk used
2440  *        by the driver.
2441  * clear_all = This flag determines whether or not the disk information
2442  *             is going to be completely cleared out and the highest_lun
2443  *             reset.  Sometimes we want to clear out information about
2444  *             the disk in preparation for re-adding it.  In this case
2445  *             the highest_lun should be left unchanged and the LunID
2446  *             should not be cleared.
2447  * via_ioctl
2448  *    This indicates whether we've reached this path via ioctl.
2449  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2450  *    If this path is reached via ioctl(), then the max_usage_count will
2451  *    be 1, as the process calling ioctl() has got to have the device open.
2452  *    If we get here via sysfs, then the max usage count will be zero.
2453 */
2454 static int deregister_disk(ctlr_info_t *h, int drv_index,
2455                            int clear_all, int via_ioctl)
2456 {
2457         int i;
2458         struct gendisk *disk;
2459         drive_info_struct *drv;
2460         int recalculate_highest_lun;
2461
2462         if (!capable(CAP_SYS_RAWIO))
2463                 return -EPERM;
2464
2465         drv = h->drv[drv_index];
2466         disk = h->gendisk[drv_index];
2467
2468         /* make sure logical volume is NOT is use */
2469         if (clear_all || (h->gendisk[0] == disk)) {
2470                 if (drv->usage_count > via_ioctl)
2471                         return -EBUSY;
2472         } else if (drv->usage_count > 0)
2473                 return -EBUSY;
2474
2475         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2476
2477         /* invalidate the devices and deregister the disk.  If it is disk
2478          * zero do not deregister it but just zero out it's values.  This
2479          * allows us to delete disk zero but keep the controller registered.
2480          */
2481         if (h->gendisk[0] != disk) {
2482                 struct request_queue *q = disk->queue;
2483                 if (disk->flags & GENHD_FL_UP) {
2484                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2485                         del_gendisk(disk);
2486                 }
2487                 if (q)
2488                         blk_cleanup_queue(q);
2489                 /* If clear_all is set then we are deleting the logical
2490                  * drive, not just refreshing its info.  For drives
2491                  * other than disk 0 we will call put_disk.  We do not
2492                  * do this for disk 0 as we need it to be able to
2493                  * configure the controller.
2494                  */
2495                 if (clear_all){
2496                         /* This isn't pretty, but we need to find the
2497                          * disk in our array and NULL our the pointer.
2498                          * This is so that we will call alloc_disk if
2499                          * this index is used again later.
2500                          */
2501                         for (i=0; i < CISS_MAX_LUN; i++){
2502                                 if (h->gendisk[i] == disk) {
2503                                         h->gendisk[i] = NULL;
2504                                         break;
2505                                 }
2506                         }
2507                         put_disk(disk);
2508                 }
2509         } else {
2510                 set_capacity(disk, 0);
2511                 cciss_clear_drive_info(drv);
2512         }
2513
2514         --h->num_luns;
2515
2516         /* if it was the last disk, find the new hightest lun */
2517         if (clear_all && recalculate_highest_lun) {
2518                 int newhighest = -1;
2519                 for (i = 0; i <= h->highest_lun; i++) {
2520                         /* if the disk has size > 0, it is available */
2521                         if (h->drv[i] && h->drv[i]->heads)
2522                                 newhighest = i;
2523                 }
2524                 h->highest_lun = newhighest;
2525         }
2526         return 0;
2527 }
2528
2529 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2530                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2531                 int cmd_type)
2532 {
2533         u64bit buff_dma_handle;
2534         int status = IO_OK;
2535
2536         c->cmd_type = CMD_IOCTL_PEND;
2537         c->Header.ReplyQueue = 0;
2538         if (buff != NULL) {
2539                 c->Header.SGList = 1;
2540                 c->Header.SGTotal = 1;
2541         } else {
2542                 c->Header.SGList = 0;
2543                 c->Header.SGTotal = 0;
2544         }
2545         c->Header.Tag.lower = c->busaddr;
2546         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2547
2548         c->Request.Type.Type = cmd_type;
2549         if (cmd_type == TYPE_CMD) {
2550                 switch (cmd) {
2551                 case CISS_INQUIRY:
2552                         /* are we trying to read a vital product page */
2553                         if (page_code != 0) {
2554                                 c->Request.CDB[1] = 0x01;
2555                                 c->Request.CDB[2] = page_code;
2556                         }
2557                         c->Request.CDBLen = 6;
2558                         c->Request.Type.Attribute = ATTR_SIMPLE;
2559                         c->Request.Type.Direction = XFER_READ;
2560                         c->Request.Timeout = 0;
2561                         c->Request.CDB[0] = CISS_INQUIRY;
2562                         c->Request.CDB[4] = size & 0xFF;
2563                         break;
2564                 case CISS_REPORT_LOG:
2565                 case CISS_REPORT_PHYS:
2566                         /* Talking to controller so It's a physical command
2567                            mode = 00 target = 0.  Nothing to write.
2568                          */
2569                         c->Request.CDBLen = 12;
2570                         c->Request.Type.Attribute = ATTR_SIMPLE;
2571                         c->Request.Type.Direction = XFER_READ;
2572                         c->Request.Timeout = 0;
2573                         c->Request.CDB[0] = cmd;
2574                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2575                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2576                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2577                         c->Request.CDB[9] = size & 0xFF;
2578                         break;
2579
2580                 case CCISS_READ_CAPACITY:
2581                         c->Request.CDBLen = 10;
2582                         c->Request.Type.Attribute = ATTR_SIMPLE;
2583                         c->Request.Type.Direction = XFER_READ;
2584                         c->Request.Timeout = 0;
2585                         c->Request.CDB[0] = cmd;
2586                         break;
2587                 case CCISS_READ_CAPACITY_16:
2588                         c->Request.CDBLen = 16;
2589                         c->Request.Type.Attribute = ATTR_SIMPLE;
2590                         c->Request.Type.Direction = XFER_READ;
2591                         c->Request.Timeout = 0;
2592                         c->Request.CDB[0] = cmd;
2593                         c->Request.CDB[1] = 0x10;
2594                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2595                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2596                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2597                         c->Request.CDB[13] = size & 0xFF;
2598                         c->Request.Timeout = 0;
2599                         c->Request.CDB[0] = cmd;
2600                         break;
2601                 case CCISS_CACHE_FLUSH:
2602                         c->Request.CDBLen = 12;
2603                         c->Request.Type.Attribute = ATTR_SIMPLE;
2604                         c->Request.Type.Direction = XFER_WRITE;
2605                         c->Request.Timeout = 0;
2606                         c->Request.CDB[0] = BMIC_WRITE;
2607                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2608                         c->Request.CDB[7] = (size >> 8) & 0xFF;
2609                         c->Request.CDB[8] = size & 0xFF;
2610                         break;
2611                 case TEST_UNIT_READY:
2612                         c->Request.CDBLen = 6;
2613                         c->Request.Type.Attribute = ATTR_SIMPLE;
2614                         c->Request.Type.Direction = XFER_NONE;
2615                         c->Request.Timeout = 0;
2616                         break;
2617                 default:
2618                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2619                         return IO_ERROR;
2620                 }
2621         } else if (cmd_type == TYPE_MSG) {
2622                 switch (cmd) {
2623                 case CCISS_ABORT_MSG:
2624                         c->Request.CDBLen = 12;
2625                         c->Request.Type.Attribute = ATTR_SIMPLE;
2626                         c->Request.Type.Direction = XFER_WRITE;
2627                         c->Request.Timeout = 0;
2628                         c->Request.CDB[0] = cmd;        /* abort */
2629                         c->Request.CDB[1] = 0;  /* abort a command */
2630                         /* buff contains the tag of the command to abort */
2631                         memcpy(&c->Request.CDB[4], buff, 8);
2632                         break;
2633                 case CCISS_RESET_MSG:
2634                         c->Request.CDBLen = 16;
2635                         c->Request.Type.Attribute = ATTR_SIMPLE;
2636                         c->Request.Type.Direction = XFER_NONE;
2637                         c->Request.Timeout = 0;
2638                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2639                         c->Request.CDB[0] = cmd;        /* reset */
2640                         c->Request.CDB[1] = CCISS_RESET_TYPE_TARGET;
2641                         break;
2642                 case CCISS_NOOP_MSG:
2643                         c->Request.CDBLen = 1;
2644                         c->Request.Type.Attribute = ATTR_SIMPLE;
2645                         c->Request.Type.Direction = XFER_WRITE;
2646                         c->Request.Timeout = 0;
2647                         c->Request.CDB[0] = cmd;
2648                         break;
2649                 default:
2650                         dev_warn(&h->pdev->dev,
2651                                 "unknown message type %d\n", cmd);
2652                         return IO_ERROR;
2653                 }
2654         } else {
2655                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2656                 return IO_ERROR;
2657         }
2658         /* Fill in the scatter gather information */
2659         if (size > 0) {
2660                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2661                                                              buff, size,
2662                                                              PCI_DMA_BIDIRECTIONAL);
2663                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2664                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2665                 c->SG[0].Len = size;
2666                 c->SG[0].Ext = 0;       /* we are not chaining */
2667         }
2668         return status;
2669 }
2670
2671 static int cciss_send_reset(ctlr_info_t *h, unsigned char *scsi3addr,
2672                             u8 reset_type)
2673 {
2674         CommandList_struct *c;
2675         int return_status;
2676
2677         c = cmd_alloc(h);
2678         if (!c)
2679                 return -ENOMEM;
2680         return_status = fill_cmd(h, c, CCISS_RESET_MSG, NULL, 0, 0,
2681                 CTLR_LUNID, TYPE_MSG);
2682         c->Request.CDB[1] = reset_type; /* fill_cmd defaults to target reset */
2683         if (return_status != IO_OK) {
2684                 cmd_special_free(h, c);
2685                 return return_status;
2686         }
2687         c->waiting = NULL;
2688         enqueue_cmd_and_start_io(h, c);
2689         /* Don't wait for completion, the reset won't complete.  Don't free
2690          * the command either.  This is the last command we will send before
2691          * re-initializing everything, so it doesn't matter and won't leak.
2692          */
2693         return 0;
2694 }
2695
2696 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2697 {
2698         switch (c->err_info->ScsiStatus) {
2699         case SAM_STAT_GOOD:
2700                 return IO_OK;
2701         case SAM_STAT_CHECK_CONDITION:
2702                 switch (0xf & c->err_info->SenseInfo[2]) {
2703                 case 0: return IO_OK; /* no sense */
2704                 case 1: return IO_OK; /* recovered error */
2705                 default:
2706                         if (check_for_unit_attention(h, c))
2707                                 return IO_NEEDS_RETRY;
2708                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2709                                 "check condition, sense key = 0x%02x\n",
2710                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2711                 }
2712                 break;
2713         default:
2714                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2715                         "scsi status = 0x%02x\n",
2716                         c->Request.CDB[0], c->err_info->ScsiStatus);
2717                 break;
2718         }
2719         return IO_ERROR;
2720 }
2721
2722 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2723 {
2724         int return_status = IO_OK;
2725
2726         if (c->err_info->CommandStatus == CMD_SUCCESS)
2727                 return IO_OK;
2728
2729         switch (c->err_info->CommandStatus) {
2730         case CMD_TARGET_STATUS:
2731                 return_status = check_target_status(h, c);
2732                 break;
2733         case CMD_DATA_UNDERRUN:
2734         case CMD_DATA_OVERRUN:
2735                 /* expected for inquiry and report lun commands */
2736                 break;
2737         case CMD_INVALID:
2738                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2739                        "reported invalid\n", c->Request.CDB[0]);
2740                 return_status = IO_ERROR;
2741                 break;
2742         case CMD_PROTOCOL_ERR:
2743                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2744                        "protocol error\n", c->Request.CDB[0]);
2745                 return_status = IO_ERROR;
2746                 break;
2747         case CMD_HARDWARE_ERR:
2748                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2749                        " hardware error\n", c->Request.CDB[0]);
2750                 return_status = IO_ERROR;
2751                 break;
2752         case CMD_CONNECTION_LOST:
2753                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2754                        "connection lost\n", c->Request.CDB[0]);
2755                 return_status = IO_ERROR;
2756                 break;
2757         case CMD_ABORTED:
2758                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2759                        "aborted\n", c->Request.CDB[0]);
2760                 return_status = IO_ERROR;
2761                 break;
2762         case CMD_ABORT_FAILED:
2763                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2764                        "abort failed\n", c->Request.CDB[0]);
2765                 return_status = IO_ERROR;
2766                 break;
2767         case CMD_UNSOLICITED_ABORT:
2768                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2769                         c->Request.CDB[0]);
2770                 return_status = IO_NEEDS_RETRY;
2771                 break;
2772         case CMD_UNABORTABLE:
2773                 dev_warn(&h->pdev->dev, "cmd unabortable\n");
2774                 return_status = IO_ERROR;
2775                 break;
2776         default:
2777                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2778                        "unknown status %x\n", c->Request.CDB[0],
2779                        c->err_info->CommandStatus);
2780                 return_status = IO_ERROR;
2781         }
2782         return return_status;
2783 }
2784
2785 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2786         int attempt_retry)
2787 {
2788         DECLARE_COMPLETION_ONSTACK(wait);
2789         u64bit buff_dma_handle;
2790         int return_status = IO_OK;
2791
2792 resend_cmd2:
2793         c->waiting = &wait;
2794         enqueue_cmd_and_start_io(h, c);
2795
2796         wait_for_completion(&wait);
2797
2798         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2799                 goto command_done;
2800
2801         return_status = process_sendcmd_error(h, c);
2802
2803         if (return_status == IO_NEEDS_RETRY &&
2804                 c->retry_count < MAX_CMD_RETRIES) {
2805                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2806                         c->Request.CDB[0]);
2807                 c->retry_count++;
2808                 /* erase the old error information */
2809                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2810                 return_status = IO_OK;
2811                 INIT_COMPLETION(wait);
2812                 goto resend_cmd2;
2813         }
2814
2815 command_done:
2816         /* unlock the buffers from DMA */
2817         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2818         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2819         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2820                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2821         return return_status;
2822 }
2823
2824 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2825                            __u8 page_code, unsigned char scsi3addr[],
2826                         int cmd_type)
2827 {
2828         CommandList_struct *c;
2829         int return_status;
2830
2831         c = cmd_special_alloc(h);
2832         if (!c)
2833                 return -ENOMEM;
2834         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2835                 scsi3addr, cmd_type);
2836         if (return_status == IO_OK)
2837                 return_status = sendcmd_withirq_core(h, c, 1);
2838
2839         cmd_special_free(h, c);
2840         return return_status;
2841 }
2842
2843 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2844                                    sector_t total_size,
2845                                    unsigned int block_size,
2846                                    InquiryData_struct *inq_buff,
2847                                    drive_info_struct *drv)
2848 {
2849         int return_code;
2850         unsigned long t;
2851         unsigned char scsi3addr[8];
2852
2853         memset(inq_buff, 0, sizeof(InquiryData_struct));
2854         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2855         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2856                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2857         if (return_code == IO_OK) {
2858                 if (inq_buff->data_byte[8] == 0xFF) {
2859                         dev_warn(&h->pdev->dev,
2860                                "reading geometry failed, volume "
2861                                "does not support reading geometry\n");
2862                         drv->heads = 255;
2863                         drv->sectors = 32;      /* Sectors per track */
2864                         drv->cylinders = total_size + 1;
2865                         drv->raid_level = RAID_UNKNOWN;
2866                 } else {
2867                         drv->heads = inq_buff->data_byte[6];
2868                         drv->sectors = inq_buff->data_byte[7];
2869                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2870                         drv->cylinders += inq_buff->data_byte[5];
2871                         drv->raid_level = inq_buff->data_byte[8];
2872                 }
2873                 drv->block_size = block_size;
2874                 drv->nr_blocks = total_size + 1;
2875                 t = drv->heads * drv->sectors;
2876                 if (t > 1) {
2877                         sector_t real_size = total_size + 1;
2878                         unsigned long rem = sector_div(real_size, t);
2879                         if (rem)
2880                                 real_size++;
2881                         drv->cylinders = real_size;
2882                 }
2883         } else {                /* Get geometry failed */
2884                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2885         }
2886 }
2887
2888 static void
2889 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2890                     unsigned int *block_size)
2891 {
2892         ReadCapdata_struct *buf;
2893         int return_code;
2894         unsigned char scsi3addr[8];
2895
2896         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2897         if (!buf) {
2898                 dev_warn(&h->pdev->dev, "out of memory\n");
2899                 return;
2900         }
2901
2902         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2903         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2904                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2905         if (return_code == IO_OK) {
2906                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2907                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2908         } else {                /* read capacity command failed */
2909                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2910                 *total_size = 0;
2911                 *block_size = BLOCK_SIZE;
2912         }
2913         kfree(buf);
2914 }
2915
2916 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2917         sector_t *total_size, unsigned int *block_size)
2918 {
2919         ReadCapdata_struct_16 *buf;
2920         int return_code;
2921         unsigned char scsi3addr[8];
2922
2923         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2924         if (!buf) {
2925                 dev_warn(&h->pdev->dev, "out of memory\n");
2926                 return;
2927         }
2928
2929         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2930         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2931                 buf, sizeof(ReadCapdata_struct_16),
2932                         0, scsi3addr, TYPE_CMD);
2933         if (return_code == IO_OK) {
2934                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2935                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2936         } else {                /* read capacity command failed */
2937                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2938                 *total_size = 0;
2939                 *block_size = BLOCK_SIZE;
2940         }
2941         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2942                (unsigned long long)*total_size+1, *block_size);
2943         kfree(buf);
2944 }
2945
2946 static int cciss_revalidate(struct gendisk *disk)
2947 {
2948         ctlr_info_t *h = get_host(disk);
2949         drive_info_struct *drv = get_drv(disk);
2950         int logvol;
2951         int FOUND = 0;
2952         unsigned int block_size;
2953         sector_t total_size;
2954         InquiryData_struct *inq_buff = NULL;
2955
2956         for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2957                 if (!h->drv[logvol])
2958                         continue;
2959                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2960                         sizeof(drv->LunID)) == 0) {
2961                         FOUND = 1;
2962                         break;
2963                 }
2964         }
2965
2966         if (!FOUND)
2967                 return 1;
2968
2969         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2970         if (inq_buff == NULL) {
2971                 dev_warn(&h->pdev->dev, "out of memory\n");
2972                 return 1;
2973         }
2974         if (h->cciss_read == CCISS_READ_10) {
2975                 cciss_read_capacity(h, logvol,
2976                                         &total_size, &block_size);
2977         } else {
2978                 cciss_read_capacity_16(h, logvol,
2979                                         &total_size, &block_size);
2980         }
2981         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2982                                inq_buff, drv);
2983
2984         blk_queue_logical_block_size(drv->queue, drv->block_size);
2985         set_capacity(disk, drv->nr_blocks);
2986
2987         kfree(inq_buff);
2988         return 0;
2989 }
2990
2991 /*
2992  * Map (physical) PCI mem into (virtual) kernel space
2993  */
2994 static void __iomem *remap_pci_mem(ulong base, ulong size)
2995 {
2996         ulong page_base = ((ulong) base) & PAGE_MASK;
2997         ulong page_offs = ((ulong) base) - page_base;
2998         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2999
3000         return page_remapped ? (page_remapped + page_offs) : NULL;
3001 }
3002
3003 /*
3004  * Takes jobs of the Q and sends them to the hardware, then puts it on
3005  * the Q to wait for completion.
3006  */
3007 static void start_io(ctlr_info_t *h)
3008 {
3009         CommandList_struct *c;
3010
3011         while (!list_empty(&h->reqQ)) {
3012                 c = list_entry(h->reqQ.next, CommandList_struct, list);
3013                 /* can't do anything if fifo is full */
3014                 if ((h->access.fifo_full(h))) {
3015                         dev_warn(&h->pdev->dev, "fifo full\n");
3016                         break;
3017                 }
3018
3019                 /* Get the first entry from the Request Q */
3020                 removeQ(c);
3021                 h->Qdepth--;
3022
3023                 /* Tell the controller execute command */
3024                 h->access.submit_command(h, c);
3025
3026                 /* Put job onto the completed Q */
3027                 addQ(&h->cmpQ, c);
3028         }
3029 }
3030
3031 /* Assumes that h->lock is held. */
3032 /* Zeros out the error record and then resends the command back */
3033 /* to the controller */
3034 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
3035 {
3036         /* erase the old error information */
3037         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
3038
3039         /* add it to software queue and then send it to the controller */
3040         addQ(&h->reqQ, c);
3041         h->Qdepth++;
3042         if (h->Qdepth > h->maxQsinceinit)
3043                 h->maxQsinceinit = h->Qdepth;
3044
3045         start_io(h);
3046 }
3047
3048 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
3049         unsigned int msg_byte, unsigned int host_byte,
3050         unsigned int driver_byte)
3051 {
3052         /* inverse of macros in scsi.h */
3053         return (scsi_status_byte & 0xff) |
3054                 ((msg_byte & 0xff) << 8) |
3055                 ((host_byte & 0xff) << 16) |
3056                 ((driver_byte & 0xff) << 24);
3057 }
3058
3059 static inline int evaluate_target_status(ctlr_info_t *h,
3060                         CommandList_struct *cmd, int *retry_cmd)
3061 {
3062         unsigned char sense_key;
3063         unsigned char status_byte, msg_byte, host_byte, driver_byte;
3064         int error_value;
3065
3066         *retry_cmd = 0;
3067         /* If we get in here, it means we got "target status", that is, scsi status */
3068         status_byte = cmd->err_info->ScsiStatus;
3069         driver_byte = DRIVER_OK;
3070         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3071
3072         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3073                 host_byte = DID_PASSTHROUGH;
3074         else
3075                 host_byte = DID_OK;
3076
3077         error_value = make_status_bytes(status_byte, msg_byte,
3078                 host_byte, driver_byte);
3079
3080         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3081                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3082                         dev_warn(&h->pdev->dev, "cmd %p "
3083                                "has SCSI Status 0x%x\n",
3084                                cmd, cmd->err_info->ScsiStatus);
3085                 return error_value;
3086         }
3087
3088         /* check the sense key */
3089         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3090         /* no status or recovered error */
3091         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3092             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3093                 error_value = 0;
3094
3095         if (check_for_unit_attention(h, cmd)) {
3096                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3097                 return 0;
3098         }
3099
3100         /* Not SG_IO or similar? */
3101         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3102                 if (error_value != 0)
3103                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3104                                " sense key = 0x%x\n", cmd, sense_key);
3105                 return error_value;
3106         }
3107
3108         /* SG_IO or similar, copy sense data back */
3109         if (cmd->rq->sense) {
3110                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3111                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3112                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3113                         cmd->rq->sense_len);
3114         } else
3115                 cmd->rq->sense_len = 0;
3116
3117         return error_value;
3118 }
3119
3120 /* checks the status of the job and calls complete buffers to mark all
3121  * buffers for the completed job. Note that this function does not need
3122  * to hold the hba/queue lock.
3123  */
3124 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3125                                     int timeout)
3126 {
3127         int retry_cmd = 0;
3128         struct request *rq = cmd->rq;
3129
3130         rq->errors = 0;
3131
3132         if (timeout)
3133                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3134
3135         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3136                 goto after_error_processing;
3137
3138         switch (cmd->err_info->CommandStatus) {
3139         case CMD_TARGET_STATUS:
3140                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3141                 break;
3142         case CMD_DATA_UNDERRUN:
3143                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3144                         dev_warn(&h->pdev->dev, "cmd %p has"
3145                                " completed with data underrun "
3146                                "reported\n", cmd);
3147                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3148                 }
3149                 break;
3150         case CMD_DATA_OVERRUN:
3151                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3152                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3153                                " completed with data overrun "
3154                                "reported\n", cmd);
3155                 break;
3156         case CMD_INVALID:
3157                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3158                        "reported invalid\n", cmd);
3159                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3160                         cmd->err_info->CommandStatus, DRIVER_OK,
3161                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3162                                 DID_PASSTHROUGH : DID_ERROR);
3163                 break;
3164         case CMD_PROTOCOL_ERR:
3165                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3166                        "protocol error\n", cmd);
3167                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3168                         cmd->err_info->CommandStatus, DRIVER_OK,
3169                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3170                                 DID_PASSTHROUGH : DID_ERROR);
3171                 break;
3172         case CMD_HARDWARE_ERR:
3173                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3174                        " hardware error\n", cmd);
3175                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3176                         cmd->err_info->CommandStatus, DRIVER_OK,
3177                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3178                                 DID_PASSTHROUGH : DID_ERROR);
3179                 break;
3180         case CMD_CONNECTION_LOST:
3181                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3182                        "connection lost\n", cmd);
3183                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3184                         cmd->err_info->CommandStatus, DRIVER_OK,
3185                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3186                                 DID_PASSTHROUGH : DID_ERROR);
3187                 break;
3188         case CMD_ABORTED:
3189                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3190                        "aborted\n", cmd);
3191                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3192                         cmd->err_info->CommandStatus, DRIVER_OK,
3193                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3194                                 DID_PASSTHROUGH : DID_ABORT);
3195                 break;
3196         case CMD_ABORT_FAILED:
3197                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3198                        "abort failed\n", cmd);
3199                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3200                         cmd->err_info->CommandStatus, DRIVER_OK,
3201                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3202                                 DID_PASSTHROUGH : DID_ERROR);
3203                 break;
3204         case CMD_UNSOLICITED_ABORT:
3205                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3206                        "abort %p\n", h->ctlr, cmd);
3207                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3208                         retry_cmd = 1;
3209                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3210                         cmd->retry_count++;
3211                 } else
3212                         dev_warn(&h->pdev->dev,
3213                                 "%p retried too many times\n", cmd);
3214                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3215                         cmd->err_info->CommandStatus, DRIVER_OK,
3216                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3217                                 DID_PASSTHROUGH : DID_ABORT);
3218                 break;
3219         case CMD_TIMEOUT:
3220                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3221                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3222                         cmd->err_info->CommandStatus, DRIVER_OK,
3223                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3224                                 DID_PASSTHROUGH : DID_ERROR);
3225                 break;
3226         case CMD_UNABORTABLE:
3227                 dev_warn(&h->pdev->dev, "cmd %p unabortable\n", cmd);
3228                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3229                         cmd->err_info->CommandStatus, DRIVER_OK,
3230                         cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC ?
3231                                 DID_PASSTHROUGH : DID_ERROR);
3232                 break;
3233         default:
3234                 dev_warn(&h->pdev->dev, "cmd %p returned "
3235                        "unknown status %x\n", cmd,
3236                        cmd->err_info->CommandStatus);
3237                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3238                         cmd->err_info->CommandStatus, DRIVER_OK,
3239                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3240                                 DID_PASSTHROUGH : DID_ERROR);
3241         }
3242
3243 after_error_processing:
3244
3245         /* We need to return this command */
3246         if (retry_cmd) {
3247                 resend_cciss_cmd(h, cmd);
3248                 return;
3249         }
3250         cmd->rq->completion_data = cmd;
3251         blk_complete_request(cmd->rq);
3252 }
3253
3254 static inline u32 cciss_tag_contains_index(u32 tag)
3255 {
3256 #define DIRECT_LOOKUP_BIT 0x10
3257         return tag & DIRECT_LOOKUP_BIT;
3258 }
3259
3260 static inline u32 cciss_tag_to_index(u32 tag)
3261 {
3262 #define DIRECT_LOOKUP_SHIFT 5
3263         return tag >> DIRECT_LOOKUP_SHIFT;
3264 }
3265
3266 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag)
3267 {
3268 #define CCISS_PERF_ERROR_BITS ((1 << DIRECT_LOOKUP_SHIFT) - 1)
3269 #define CCISS_SIMPLE_ERROR_BITS 0x03
3270         if (likely(h->transMethod & CFGTBL_Trans_Performant))
3271                 return tag & ~CCISS_PERF_ERROR_BITS;
3272         return tag & ~CCISS_SIMPLE_ERROR_BITS;
3273 }
3274
3275 static inline void cciss_mark_tag_indexed(u32 *tag)
3276 {
3277         *tag |= DIRECT_LOOKUP_BIT;
3278 }
3279
3280 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3281 {
3282         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3283 }
3284
3285 /*
3286  * Get a request and submit it to the controller.
3287  */
3288 static void do_cciss_request(struct request_queue *q)
3289 {
3290         ctlr_info_t *h = q->queuedata;
3291         CommandList_struct *c;
3292         sector_t start_blk;
3293         int seg;
3294         struct request *creq;
3295         u64bit temp64;
3296         struct scatterlist *tmp_sg;
3297         SGDescriptor_struct *curr_sg;
3298         drive_info_struct *drv;
3299         int i, dir;
3300         int sg_index = 0;
3301         int chained = 0;
3302
3303       queue:
3304         creq = blk_peek_request(q);
3305         if (!creq)
3306                 goto startio;
3307
3308         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3309
3310         c = cmd_alloc(h);
3311         if (!c)
3312                 goto full;
3313
3314         blk_start_request(creq);
3315
3316         tmp_sg = h->scatter_list[c->cmdindex];
3317         spin_unlock_irq(q->queue_lock);
3318
3319         c->cmd_type = CMD_RWREQ;
3320         c->rq = creq;
3321
3322         /* fill in the request */
3323         drv = creq->rq_disk->private_data;
3324         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3325         /* got command from pool, so use the command block index instead */
3326         /* for direct lookups. */
3327         /* The first 2 bits are reserved for controller error reporting. */
3328         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3329         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3330         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3331         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3332         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3333         c->Request.Type.Attribute = ATTR_SIMPLE;
3334         c->Request.Type.Direction =
3335             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3336         c->Request.Timeout = 0; /* Don't time out */
3337         c->Request.CDB[0] =
3338             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3339         start_blk = blk_rq_pos(creq);
3340         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3341                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3342         sg_init_table(tmp_sg, h->maxsgentries);
3343         seg = blk_rq_map_sg(q, creq, tmp_sg);
3344
3345         /* get the DMA records for the setup */
3346         if (c->Request.Type.Direction == XFER_READ)
3347                 dir = PCI_DMA_FROMDEVICE;
3348         else
3349                 dir = PCI_DMA_TODEVICE;
3350
3351         curr_sg = c->SG;
3352         sg_index = 0;
3353         chained = 0;
3354
3355         for (i = 0; i < seg; i++) {
3356                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3357                         !chained && ((seg - i) > 1)) {
3358                         /* Point to next chain block. */
3359                         curr_sg = h->cmd_sg_list[c->cmdindex];
3360                         sg_index = 0;
3361                         chained = 1;
3362                 }
3363                 curr_sg[sg_index].Len = tmp_sg[i].length;
3364                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3365                                                 tmp_sg[i].offset,
3366                                                 tmp_sg[i].length, dir);
3367                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3368                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3369                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3370                 ++sg_index;
3371         }
3372         if (chained)
3373                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3374                         (seg - (h->max_cmd_sgentries - 1)) *
3375                                 sizeof(SGDescriptor_struct));
3376
3377         /* track how many SG entries we are using */
3378         if (seg > h->maxSG)
3379                 h->maxSG = seg;
3380
3381         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3382                         "chained[%d]\n",
3383                         blk_rq_sectors(creq), seg, chained);
3384
3385         c->Header.SGTotal = seg + chained;
3386         if (seg <= h->max_cmd_sgentries)
3387                 c->Header.SGList = c->Header.SGTotal;
3388         else
3389                 c->Header.SGList = h->max_cmd_sgentries;
3390         set_performant_mode(h, c);
3391
3392         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3393                 if(h->cciss_read == CCISS_READ_10) {
3394                         c->Request.CDB[1] = 0;
3395                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3396                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3397                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3398                         c->Request.CDB[5] = start_blk & 0xff;
3399                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3400                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3401                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3402                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3403                 } else {
3404                         u32 upper32 = upper_32_bits(start_blk);
3405
3406                         c->Request.CDBLen = 16;
3407                         c->Request.CDB[1]= 0;
3408                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3409                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3410                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3411                         c->Request.CDB[5]= upper32 & 0xff;
3412                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3413                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3414                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3415                         c->Request.CDB[9]= start_blk & 0xff;
3416                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3417                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3418                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3419                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3420                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3421                 }
3422         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3423                 c->Request.CDBLen = creq->cmd_len;
3424                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3425         } else {
3426                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3427                         creq->cmd_type);
3428                 BUG();
3429         }
3430
3431         spin_lock_irq(q->queue_lock);
3432
3433         addQ(&h->reqQ, c);
3434         h->Qdepth++;
3435         if (h->Qdepth > h->maxQsinceinit)
3436                 h->maxQsinceinit = h->Qdepth;
3437
3438         goto queue;
3439 full:
3440         blk_stop_queue(q);
3441 startio:
3442         /* We will already have the driver lock here so not need
3443          * to lock it.
3444          */
3445         start_io(h);
3446 }
3447
3448 static inline unsigned long get_next_completion(ctlr_info_t *h)
3449 {
3450         return h->access.command_completed(h);
3451 }
3452
3453 static inline int interrupt_pending(ctlr_info_t *h)
3454 {
3455         return h->access.intr_pending(h);
3456 }
3457
3458 static inline long interrupt_not_for_us(ctlr_info_t *h)
3459 {
3460         return ((h->access.intr_pending(h) == 0) ||
3461                 (h->interrupts_enabled == 0));
3462 }
3463
3464 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3465                         u32 raw_tag)
3466 {
3467         if (unlikely(tag_index >= h->nr_cmds)) {
3468                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3469                 return 1;
3470         }
3471         return 0;
3472 }
3473
3474 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3475                                 u32 raw_tag)
3476 {
3477         removeQ(c);
3478         if (likely(c->cmd_type == CMD_RWREQ))
3479                 complete_command(h, c, 0);
3480         else if (c->cmd_type == CMD_IOCTL_PEND)
3481                 complete(c->waiting);
3482 #ifdef CONFIG_CISS_SCSI_TAPE
3483         else if (c->cmd_type == CMD_SCSI)
3484                 complete_scsi_command(c, 0, raw_tag);
3485 #endif
3486 }
3487
3488 static inline u32 next_command(ctlr_info_t *h)
3489 {
3490         u32 a;
3491
3492         if (unlikely(!(h->transMethod & CFGTBL_Trans_Performant)))
3493                 return h->access.command_completed(h);
3494
3495         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3496                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3497                 (h->reply_pool_head)++;
3498                 h->commands_outstanding--;
3499         } else {
3500                 a = FIFO_EMPTY;
3501         }
3502         /* Check for wraparound */
3503         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3504                 h->reply_pool_head = h->reply_pool;
3505                 h->reply_pool_wraparound ^= 1;
3506         }
3507         return a;
3508 }
3509
3510 /* process completion of an indexed ("direct lookup") command */
3511 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3512 {
3513         u32 tag_index;
3514         CommandList_struct *c;
3515
3516         tag_index = cciss_tag_to_index(raw_tag);
3517         if (bad_tag(h, tag_index, raw_tag))
3518                 return next_command(h);
3519         c = h->cmd_pool + tag_index;
3520         finish_cmd(h, c, raw_tag);
3521         return next_command(h);
3522 }
3523
3524 /* process completion of a non-indexed command */
3525 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3526 {
3527         CommandList_struct *c = NULL;
3528         __u32 busaddr_masked, tag_masked;
3529
3530         tag_masked = cciss_tag_discard_error_bits(h, raw_tag);
3531         list_for_each_entry(c, &h->cmpQ, list) {
3532                 busaddr_masked = cciss_tag_discard_error_bits(h, c->busaddr);
3533                 if (busaddr_masked == tag_masked) {
3534                         finish_cmd(h, c, raw_tag);
3535                         return next_command(h);
3536                 }
3537         }
3538         bad_tag(h, h->nr_cmds + 1, raw_tag);
3539         return next_command(h);
3540 }
3541
3542 /* Some controllers, like p400, will give us one interrupt
3543  * after a soft reset, even if we turned interrupts off.
3544  * Only need to check for this in the cciss_xxx_discard_completions
3545  * functions.
3546  */
3547 static int ignore_bogus_interrupt(ctlr_info_t *h)
3548 {
3549         if (likely(!reset_devices))
3550                 return 0;
3551
3552         if (likely(h->interrupts_enabled))
3553                 return 0;
3554
3555         dev_info(&h->pdev->dev, "Received interrupt while interrupts disabled "
3556                 "(known firmware bug.)  Ignoring.\n");
3557
3558         return 1;
3559 }
3560
3561 static irqreturn_t cciss_intx_discard_completions(int irq, void *dev_id)
3562 {
3563         ctlr_info_t *h = dev_id;
3564         unsigned long flags;
3565         u32 raw_tag;
3566
3567         if (ignore_bogus_interrupt(h))
3568                 return IRQ_NONE;
3569
3570         if (interrupt_not_for_us(h))
3571                 return IRQ_NONE;
3572         spin_lock_irqsave(&h->lock, flags);
3573         while (interrupt_pending(h)) {
3574                 raw_tag = get_next_completion(h);
3575                 while (raw_tag != FIFO_EMPTY)
3576                         raw_tag = next_command(h);
3577         }
3578         spin_unlock_irqrestore(&h->lock, flags);
3579         return IRQ_HANDLED;
3580 }
3581
3582 static irqreturn_t cciss_msix_discard_completions(int irq, void *dev_id)
3583 {
3584         ctlr_info_t *h = dev_id;
3585         unsigned long flags;
3586         u32 raw_tag;
3587
3588         if (ignore_bogus_interrupt(h))
3589                 return IRQ_NONE;
3590
3591         spin_lock_irqsave(&h->lock, flags);
3592         raw_tag = get_next_completion(h);
3593         while (raw_tag != FIFO_EMPTY)
3594                 raw_tag = next_command(h);
3595         spin_unlock_irqrestore(&h->lock, flags);
3596         return IRQ_HANDLED;
3597 }
3598
3599 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3600 {
3601         ctlr_info_t *h = dev_id;
3602         unsigned long flags;
3603         u32 raw_tag;
3604
3605         if (interrupt_not_for_us(h))
3606                 return IRQ_NONE;
3607         spin_lock_irqsave(&h->lock, flags);
3608         while (interrupt_pending(h)) {
3609                 raw_tag = get_next_completion(h);
3610                 while (raw_tag != FIFO_EMPTY) {
3611                         if (cciss_tag_contains_index(raw_tag))
3612                                 raw_tag = process_indexed_cmd(h, raw_tag);
3613                         else
3614                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3615                 }
3616         }
3617         spin_unlock_irqrestore(&h->lock, flags);
3618         return IRQ_HANDLED;
3619 }
3620
3621 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3622  * check the interrupt pending register because it is not set.
3623  */
3624 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3625 {
3626         ctlr_info_t *h = dev_id;
3627         unsigned long flags;
3628         u32 raw_tag;
3629
3630         spin_lock_irqsave(&h->lock, flags);
3631         raw_tag = get_next_completion(h);
3632         while (raw_tag != FIFO_EMPTY) {
3633                 if (cciss_tag_contains_index(raw_tag))
3634                         raw_tag = process_indexed_cmd(h, raw_tag);
3635                 else
3636                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3637         }
3638         spin_unlock_irqrestore(&h->lock, flags);
3639         return IRQ_HANDLED;
3640 }
3641
3642 /**
3643  * add_to_scan_list() - add controller to rescan queue
3644  * @h:                Pointer to the controller.
3645  *
3646  * Adds the controller to the rescan queue if not already on the queue.
3647  *
3648  * returns 1 if added to the queue, 0 if skipped (could be on the
3649  * queue already, or the controller could be initializing or shutting
3650  * down).
3651  **/
3652 static int add_to_scan_list(struct ctlr_info *h)
3653 {
3654         struct ctlr_info *test_h;
3655         int found = 0;
3656         int ret = 0;
3657
3658         if (h->busy_initializing)
3659                 return 0;
3660
3661         if (!mutex_trylock(&h->busy_shutting_down))
3662                 return 0;
3663
3664         mutex_lock(&scan_mutex);
3665         list_for_each_entry(test_h, &scan_q, scan_list) {
3666                 if (test_h == h) {
3667                         found = 1;
3668                         break;
3669                 }
3670         }
3671         if (!found && !h->busy_scanning) {
3672                 INIT_COMPLETION(h->scan_wait);
3673                 list_add_tail(&h->scan_list, &scan_q);
3674                 ret = 1;
3675         }
3676         mutex_unlock(&scan_mutex);
3677         mutex_unlock(&h->busy_shutting_down);
3678
3679         return ret;
3680 }
3681
3682 /**
3683  * remove_from_scan_list() - remove controller from rescan queue
3684  * @h:                     Pointer to the controller.
3685  *
3686  * Removes the controller from the rescan queue if present. Blocks if
3687  * the controller is currently conducting a rescan.  The controller
3688  * can be in one of three states:
3689  * 1. Doesn't need a scan
3690  * 2. On the scan list, but not scanning yet (we remove it)
3691  * 3. Busy scanning (and not on the list). In this case we want to wait for
3692  *    the scan to complete to make sure the scanning thread for this
3693  *    controller is completely idle.
3694  **/
3695 static void remove_from_scan_list(struct ctlr_info *h)
3696 {
3697         struct ctlr_info *test_h, *tmp_h;
3698
3699         mutex_lock(&scan_mutex);
3700         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3701                 if (test_h == h) { /* state 2. */
3702                         list_del(&h->scan_list);
3703                         complete_all(&h->scan_wait);
3704                         mutex_unlock(&scan_mutex);
3705                         return;
3706                 }
3707         }
3708         if (h->busy_scanning) { /* state 3. */
3709                 mutex_unlock(&scan_mutex);
3710                 wait_for_completion(&h->scan_wait);
3711         } else { /* state 1, nothing to do. */
3712                 mutex_unlock(&scan_mutex);
3713         }
3714 }
3715
3716 /**
3717  * scan_thread() - kernel thread used to rescan controllers
3718  * @data:        Ignored.
3719  *
3720  * A kernel thread used scan for drive topology changes on
3721  * controllers. The thread processes only one controller at a time
3722  * using a queue.  Controllers are added to the queue using
3723  * add_to_scan_list() and removed from the queue either after done
3724  * processing or using remove_from_scan_list().
3725  *
3726  * returns 0.
3727  **/
3728 static int scan_thread(void *data)
3729 {
3730         struct ctlr_info *h;
3731
3732         while (1) {
3733                 set_current_state(TASK_INTERRUPTIBLE);
3734                 schedule();
3735                 if (kthread_should_stop())
3736                         break;
3737
3738                 while (1) {
3739                         mutex_lock(&scan_mutex);
3740                         if (list_empty(&scan_q)) {
3741                                 mutex_unlock(&scan_mutex);
3742                                 break;
3743                         }
3744
3745                         h = list_entry(scan_q.next,
3746                                        struct ctlr_info,
3747                                        scan_list);
3748                         list_del(&h->scan_list);
3749                         h->busy_scanning = 1;
3750                         mutex_unlock(&scan_mutex);
3751
3752                         rebuild_lun_table(h, 0, 0);
3753                         complete_all(&h->scan_wait);
3754                         mutex_lock(&scan_mutex);
3755                         h->busy_scanning = 0;
3756                         mutex_unlock(&scan_mutex);
3757                 }
3758         }
3759
3760         return 0;
3761 }
3762
3763 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3764 {
3765         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3766                 return 0;
3767
3768         switch (c->err_info->SenseInfo[12]) {
3769         case STATE_CHANGED:
3770                 dev_warn(&h->pdev->dev, "a state change "
3771                         "detected, command retried\n");
3772                 return 1;
3773         break;
3774         case LUN_FAILED:
3775                 dev_warn(&h->pdev->dev, "LUN failure "
3776                         "detected, action required\n");
3777                 return 1;
3778         break;
3779         case REPORT_LUNS_CHANGED:
3780                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3781         /*
3782          * Here, we could call add_to_scan_list and wake up the scan thread,
3783          * except that it's quite likely that we will get more than one
3784          * REPORT_LUNS_CHANGED condition in quick succession, which means
3785          * that those which occur after the first one will likely happen
3786          * *during* the scan_thread's rescan.  And the rescan code is not
3787          * robust enough to restart in the middle, undoing what it has already
3788          * done, and it's not clear that it's even possible to do this, since
3789          * part of what it does is notify the block layer, which starts
3790          * doing it's own i/o to read partition tables and so on, and the
3791          * driver doesn't have visibility to know what might need undoing.
3792          * In any event, if possible, it is horribly complicated to get right
3793          * so we just don't do it for now.
3794          *
3795          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3796          */
3797                 return 1;
3798         break;
3799         case POWER_OR_RESET:
3800                 dev_warn(&h->pdev->dev,
3801                         "a power on or device reset detected\n");
3802                 return 1;
3803         break;
3804         case UNIT_ATTENTION_CLEARED:
3805                 dev_warn(&h->pdev->dev,
3806                         "unit attention cleared by another initiator\n");
3807                 return 1;
3808         break;
3809         default:
3810                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3811                 return 1;
3812         }
3813 }
3814
3815 /*
3816  *  We cannot read the structure directly, for portability we must use
3817  *   the io functions.
3818  *   This is for debug only.
3819  */
3820 static void print_cfg_table(ctlr_info_t *h)
3821 {
3822         int i;
3823         char temp_name[17];
3824         CfgTable_struct *tb = h->cfgtable;
3825
3826         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3827         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3828         for (i = 0; i < 4; i++)
3829                 temp_name[i] = readb(&(tb->Signature[i]));
3830         temp_name[4] = '\0';
3831         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3832         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3833                 readl(&(tb->SpecValence)));
3834         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3835                readl(&(tb->TransportSupport)));
3836         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3837                readl(&(tb->TransportActive)));
3838         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3839                readl(&(tb->HostWrite.TransportRequest)));
3840         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3841                readl(&(tb->HostWrite.CoalIntDelay)));
3842         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3843                readl(&(tb->HostWrite.CoalIntCount)));
3844         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3845                readl(&(tb->CmdsOutMax)));
3846         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3847                 readl(&(tb->BusTypes)));
3848         for (i = 0; i < 16; i++)
3849                 temp_name[i] = readb(&(tb->ServerName[i]));
3850         temp_name[16] = '\0';
3851         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3852         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3853                 readl(&(tb->HeartBeat)));
3854 }
3855
3856 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3857 {
3858         int i, offset, mem_type, bar_type;
3859         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3860                 return 0;
3861         offset = 0;
3862         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3863                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3864                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3865                         offset += 4;
3866                 else {
3867                         mem_type = pci_resource_flags(pdev, i) &
3868                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3869                         switch (mem_type) {
3870                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3871                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3872                                 offset += 4;    /* 32 bit */
3873                                 break;
3874                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3875                                 offset += 8;
3876                                 break;
3877                         default:        /* reserved in PCI 2.2 */
3878                                 dev_warn(&pdev->dev,
3879                                        "Base address is invalid\n");
3880                                 return -1;
3881                                 break;
3882                         }
3883                 }
3884                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3885                         return i + 1;
3886         }
3887         return -1;
3888 }
3889
3890 /* Fill in bucket_map[], given nsgs (the max number of
3891  * scatter gather elements supported) and bucket[],
3892  * which is an array of 8 integers.  The bucket[] array
3893  * contains 8 different DMA transfer sizes (in 16
3894  * byte increments) which the controller uses to fetch
3895  * commands.  This function fills in bucket_map[], which
3896  * maps a given number of scatter gather elements to one of
3897  * the 8 DMA transfer sizes.  The point of it is to allow the
3898  * controller to only do as much DMA as needed to fetch the
3899  * command, with the DMA transfer size encoded in the lower
3900  * bits of the command address.
3901  */
3902 static void  calc_bucket_map(int bucket[], int num_buckets,
3903         int nsgs, int *bucket_map)
3904 {
3905         int i, j, b, size;
3906
3907         /* even a command with 0 SGs requires 4 blocks */
3908 #define MINIMUM_TRANSFER_BLOCKS 4
3909 #define NUM_BUCKETS 8
3910         /* Note, bucket_map must have nsgs+1 entries. */
3911         for (i = 0; i <= nsgs; i++) {
3912                 /* Compute size of a command with i SG entries */
3913                 size = i + MINIMUM_TRANSFER_BLOCKS;
3914                 b = num_buckets; /* Assume the biggest bucket */
3915                 /* Find the bucket that is just big enough */
3916                 for (j = 0; j < 8; j++) {
3917                         if (bucket[j] >= size) {
3918                                 b = j;
3919                                 break;
3920                         }
3921                 }
3922                 /* for a command with i SG entries, use bucket b. */
3923                 bucket_map[i] = b;
3924         }
3925 }
3926
3927 static void cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3928 {
3929         int i;
3930
3931         /* under certain very rare conditions, this can take awhile.
3932          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3933          * as we enter this code.) */
3934         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3935                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3936                         break;
3937                 usleep_range(10000, 20000);
3938         }
3939 }
3940
3941 static void cciss_enter_performant_mode(ctlr_info_t *h, u32 use_short_tags)
3942 {
3943         /* This is a bit complicated.  There are 8 registers on
3944          * the controller which we write to to tell it 8 different
3945          * sizes of commands which there may be.  It's a way of
3946          * reducing the DMA done to fetch each command.  Encoded into
3947          * each command's tag are 3 bits which communicate to the controller
3948          * which of the eight sizes that command fits within.  The size of
3949          * each command depends on how many scatter gather entries there are.
3950          * Each SG entry requires 16 bytes.  The eight registers are programmed
3951          * with the number of 16-byte blocks a command of that size requires.
3952          * The smallest command possible requires 5 such 16 byte blocks.
3953          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3954          * blocks.  Note, this only extends to the SG entries contained
3955          * within the command block, and does not extend to chained blocks
3956          * of SG elements.   bft[] contains the eight values we write to
3957          * the registers.  They are not evenly distributed, but have more
3958          * sizes for small commands, and fewer sizes for larger commands.
3959          */
3960         __u32 trans_offset;
3961         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3962                         /*
3963                          *  5 = 1 s/g entry or 4k
3964                          *  6 = 2 s/g entry or 8k
3965                          *  8 = 4 s/g entry or 16k
3966                          * 10 = 6 s/g entry or 24k
3967                          */
3968         unsigned long register_value;
3969         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3970
3971         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3972
3973         /* Controller spec: zero out this buffer. */
3974         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3975         h->reply_pool_head = h->reply_pool;
3976
3977         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3978         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3979                                 h->blockFetchTable);
3980         writel(bft[0], &h->transtable->BlockFetch0);
3981         writel(bft[1], &h->transtable->BlockFetch1);
3982         writel(bft[2], &h->transtable->BlockFetch2);
3983         writel(bft[3], &h->transtable->BlockFetch3);
3984         writel(bft[4], &h->transtable->BlockFetch4);
3985         writel(bft[5], &h->transtable->BlockFetch5);
3986         writel(bft[6], &h->transtable->BlockFetch6);
3987         writel(bft[7], &h->transtable->BlockFetch7);
3988
3989         /* size of controller ring buffer */
3990         writel(h->max_commands, &h->transtable->RepQSize);
3991         writel(1, &h->transtable->RepQCount);
3992         writel(0, &h->transtable->RepQCtrAddrLow32);
3993         writel(0, &h->transtable->RepQCtrAddrHigh32);
3994         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3995         writel(0, &h->transtable->RepQAddr0High32);
3996         writel(CFGTBL_Trans_Performant | use_short_tags,
3997                         &(h->cfgtable->HostWrite.TransportRequest));
3998
3999         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
4000         cciss_wait_for_mode_change_ack(h);
4001         register_value = readl(&(h->cfgtable->TransportActive));
4002         if (!(register_value & CFGTBL_Trans_Performant))
4003                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
4004                                         " performant mode\n");
4005 }
4006
4007 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h)
4008 {
4009         __u32 trans_support;
4010
4011         if (cciss_simple_mode)
4012                 return;
4013
4014         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
4015         /* Attempt to put controller into performant mode if supported */
4016         /* Does board support performant mode? */
4017         trans_support = readl(&(h->cfgtable->TransportSupport));
4018         if (!(trans_support & PERFORMANT_MODE))
4019                 return;
4020
4021         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
4022         /* Performant mode demands commands on a 32 byte boundary
4023          * pci_alloc_consistent aligns on page boundarys already.
4024          * Just need to check if divisible by 32
4025          */
4026         if ((sizeof(CommandList_struct) % 32) != 0) {
4027                 dev_warn(&h->pdev->dev, "%s %d %s\n",
4028                         "cciss info: command size[",
4029                         (int)sizeof(CommandList_struct),
4030                         "] not divisible by 32, no performant mode..\n");
4031                 return;
4032         }
4033
4034         /* Performant mode ring buffer and supporting data structures */
4035         h->reply_pool = (__u64 *)pci_alloc_consistent(
4036                 h->pdev, h->max_commands * sizeof(__u64),
4037                 &(h->reply_pool_dhandle));
4038
4039         /* Need a block fetch table for performant mode */
4040         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
4041                 sizeof(__u32)), GFP_KERNEL);
4042
4043         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
4044                 goto clean_up;
4045
4046         cciss_enter_performant_mode(h,
4047                 trans_support & CFGTBL_Trans_use_short_tags);
4048
4049         /* Change the access methods to the performant access methods */
4050         h->access = SA5_performant_access;
4051         h->transMethod = CFGTBL_Trans_Performant;
4052
4053         return;
4054 clean_up:
4055         kfree(h->blockFetchTable);
4056         if (h->reply_pool)
4057                 pci_free_consistent(h->pdev,
4058                                 h->max_commands * sizeof(__u64),
4059                                 h->reply_pool,
4060                                 h->reply_pool_dhandle);
4061         return;
4062
4063 } /* cciss_put_controller_into_performant_mode */
4064
4065 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
4066  * controllers that are capable. If not, we use IO-APIC mode.
4067  */
4068
4069 static void cciss_interrupt_mode(ctlr_info_t *h)
4070 {
4071 #ifdef CONFIG_PCI_MSI
4072         int err;
4073         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
4074         {0, 2}, {0, 3}
4075         };
4076
4077         /* Some boards advertise MSI but don't really support it */
4078         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
4079             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
4080                 goto default_int_mode;
4081
4082         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
4083                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
4084                 if (!err) {
4085                         h->intr[0] = cciss_msix_entries[0].vector;
4086                         h->intr[1] = cciss_msix_entries[1].vector;
4087                         h->intr[2] = cciss_msix_entries[2].vector;
4088                         h->intr[3] = cciss_msix_entries[3].vector;
4089                         h->msix_vector = 1;
4090                         return;
4091                 }
4092                 if (err > 0) {
4093                         dev_warn(&h->pdev->dev,
4094                                 "only %d MSI-X vectors available\n", err);
4095                         goto default_int_mode;
4096                 } else {
4097                         dev_warn(&h->pdev->dev,
4098                                 "MSI-X init failed %d\n", err);
4099                         goto default_int_mode;
4100                 }
4101         }
4102         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
4103                 if (!pci_enable_msi(h->pdev))
4104                         h->msi_vector = 1;
4105                 else
4106                         dev_warn(&h->pdev->dev, "MSI init failed\n");
4107         }
4108 default_int_mode:
4109 #endif                          /* CONFIG_PCI_MSI */
4110         /* if we get here we're going to use the default interrupt mode */
4111         h->intr[h->intr_mode] = h->pdev->irq;
4112         return;
4113 }
4114
4115 static int cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
4116 {
4117         int i;
4118         u32 subsystem_vendor_id, subsystem_device_id;
4119
4120         subsystem_vendor_id = pdev->subsystem_vendor;
4121         subsystem_device_id = pdev->subsystem_device;
4122         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
4123                         subsystem_vendor_id;
4124
4125         for (i = 0; i < ARRAY_SIZE(products); i++) {
4126                 /* Stand aside for hpsa driver on request */
4127                 if (cciss_allow_hpsa)
4128                         return -ENODEV;
4129                 if (*board_id == products[i].board_id)
4130                         return i;
4131         }
4132         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4133                 *board_id);
4134         return -ENODEV;
4135 }
4136
4137 static inline bool cciss_board_disabled(ctlr_info_t *h)
4138 {
4139         u16 command;
4140
4141         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4142         return ((command & PCI_COMMAND_MEMORY) == 0);
4143 }
4144
4145 static int cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4146                                      unsigned long *memory_bar)
4147 {
4148         int i;
4149
4150         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4151                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4152                         /* addressing mode bits already removed */
4153                         *memory_bar = pci_resource_start(pdev, i);
4154                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4155                                 *memory_bar);
4156                         return 0;
4157                 }
4158         dev_warn(&pdev->dev, "no memory BAR found\n");
4159         return -ENODEV;
4160 }
4161
4162 static int cciss_wait_for_board_state(struct pci_dev *pdev,
4163                                       void __iomem *vaddr, int wait_for_ready)
4164 #define BOARD_READY 1
4165 #define BOARD_NOT_READY 0
4166 {
4167         int i, iterations;
4168         u32 scratchpad;
4169
4170         if (wait_for_ready)
4171                 iterations = CCISS_BOARD_READY_ITERATIONS;
4172         else
4173                 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
4174
4175         for (i = 0; i < iterations; i++) {
4176                 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
4177                 if (wait_for_ready) {
4178                         if (scratchpad == CCISS_FIRMWARE_READY)
4179                                 return 0;
4180                 } else {
4181                         if (scratchpad != CCISS_FIRMWARE_READY)
4182                                 return 0;
4183                 }
4184                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4185         }
4186         dev_warn(&pdev->dev, "board not ready, timed out.\n");
4187         return -ENODEV;
4188 }
4189
4190 static int cciss_find_cfg_addrs(struct pci_dev *pdev, void __iomem *vaddr,
4191                                 u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4192                                 u64 *cfg_offset)
4193 {
4194         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4195         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4196         *cfg_base_addr &= (u32) 0x0000ffff;
4197         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4198         if (*cfg_base_addr_index == -1) {
4199                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4200                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4201                 return -ENODEV;
4202         }
4203         return 0;
4204 }
4205
4206 static int cciss_find_cfgtables(ctlr_info_t *h)
4207 {
4208         u64 cfg_offset;
4209         u32 cfg_base_addr;
4210         u64 cfg_base_addr_index;
4211         u32 trans_offset;
4212         int rc;
4213
4214         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4215                 &cfg_base_addr_index, &cfg_offset);
4216         if (rc)
4217                 return rc;
4218         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4219                 cfg_base_addr_index) + cfg_offset, sizeof(*h->cfgtable));
4220         if (!h->cfgtable)
4221                 return -ENOMEM;
4222         rc = write_driver_ver_to_cfgtable(h->cfgtable);
4223         if (rc)
4224                 return rc;
4225         /* Find performant mode table. */
4226         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4227         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4228                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4229                                 sizeof(*h->transtable));
4230         if (!h->transtable)
4231                 return -ENOMEM;
4232         return 0;
4233 }
4234
4235 static void cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4236 {
4237         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4238
4239         /* Limit commands in memory limited kdump scenario. */
4240         if (reset_devices && h->max_commands > 32)
4241                 h->max_commands = 32;
4242
4243         if (h->max_commands < 16) {
4244                 dev_warn(&h->pdev->dev, "Controller reports "
4245                         "max supported commands of %d, an obvious lie. "
4246                         "Using 16.  Ensure that firmware is up to date.\n",
4247                         h->max_commands);
4248                 h->max_commands = 16;
4249         }
4250 }
4251
4252 /* Interrogate the hardware for some limits:
4253  * max commands, max SG elements without chaining, and with chaining,
4254  * SG chain block size, etc.
4255  */
4256 static void cciss_find_board_params(ctlr_info_t *h)
4257 {
4258         cciss_get_max_perf_mode_cmds(h);
4259         h->nr_cmds = h->max_commands - 4 - cciss_tape_cmds;
4260         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4261         /*
4262          * Limit in-command s/g elements to 32 save dma'able memory.
4263          * Howvever spec says if 0, use 31
4264          */
4265         h->max_cmd_sgentries = 31;
4266         if (h->maxsgentries > 512) {
4267                 h->max_cmd_sgentries = 32;
4268                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4269                 h->maxsgentries--; /* save one for chain pointer */
4270         } else {
4271                 h->maxsgentries = 31; /* default to traditional values */
4272                 h->chainsize = 0;
4273         }
4274 }
4275
4276 static inline bool CISS_signature_present(ctlr_info_t *h)
4277 {
4278         if (!check_signature(h->cfgtable->Signature, "CISS", 4)) {
4279                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4280                 return false;
4281         }
4282         return true;
4283 }
4284
4285 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4286 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4287 {
4288 #ifdef CONFIG_X86
4289         u32 prefetch;
4290
4291         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4292         prefetch |= 0x100;
4293         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4294 #endif
4295 }
4296
4297 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4298  * in a prefetch beyond physical memory.
4299  */
4300 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4301 {
4302         u32 dma_prefetch;
4303         __u32 dma_refetch;
4304
4305         if (h->board_id != 0x3225103C)
4306                 return;
4307         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4308         dma_prefetch |= 0x8000;
4309         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4310         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4311         dma_refetch |= 0x1;
4312         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4313 }
4314
4315 static int cciss_pci_init(ctlr_info_t *h)
4316 {
4317         int prod_index, err;
4318
4319         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4320         if (prod_index < 0)
4321                 return -ENODEV;
4322         h->product_name = products[prod_index].product_name;
4323         h->access = *(products[prod_index].access);
4324
4325         if (cciss_board_disabled(h)) {
4326                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4327                 return -ENODEV;
4328         }
4329
4330         pci_disable_link_state(h->pdev, PCIE_LINK_STATE_L0S |
4331                                 PCIE_LINK_STATE_L1 | PCIE_LINK_STATE_CLKPM);
4332
4333         err = pci_enable_device(h->pdev);
4334         if (err) {
4335                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4336                 return err;
4337         }
4338
4339         err = pci_request_regions(h->pdev, "cciss");
4340         if (err) {
4341                 dev_warn(&h->pdev->dev,
4342                         "Cannot obtain PCI resources, aborting\n");
4343                 return err;
4344         }
4345
4346         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4347         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4348
4349 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4350  * else we use the IO-APIC interrupt assigned to us by system ROM.
4351  */
4352         cciss_interrupt_mode(h);
4353         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4354         if (err)
4355                 goto err_out_free_res;
4356         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4357         if (!h->vaddr) {
4358                 err = -ENOMEM;
4359                 goto err_out_free_res;
4360         }
4361         err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4362         if (err)
4363                 goto err_out_free_res;
4364         err = cciss_find_cfgtables(h);
4365         if (err)
4366                 goto err_out_free_res;
4367         print_cfg_table(h);
4368         cciss_find_board_params(h);
4369
4370         if (!CISS_signature_present(h)) {
4371                 err = -ENODEV;
4372                 goto err_out_free_res;
4373         }
4374         cciss_enable_scsi_prefetch(h);
4375         cciss_p600_dma_prefetch_quirk(h);
4376         err = cciss_enter_simple_mode(h);
4377         if (err)
4378                 goto err_out_free_res;
4379         cciss_put_controller_into_performant_mode(h);
4380         return 0;
4381
4382 err_out_free_res:
4383         /*
4384          * Deliberately omit pci_disable_device(): it does something nasty to
4385          * Smart Array controllers that pci_enable_device does not undo
4386          */
4387         if (h->transtable)
4388                 iounmap(h->transtable);
4389         if (h->cfgtable)
4390                 iounmap(h->cfgtable);
4391         if (h->vaddr)
4392                 iounmap(h->vaddr);
4393         pci_release_regions(h->pdev);
4394         return err;
4395 }
4396
4397 /* Function to find the first free pointer into our hba[] array
4398  * Returns -1 if no free entries are left.
4399  */
4400 static int alloc_cciss_hba(struct pci_dev *pdev)
4401 {
4402         int i;
4403
4404         for (i = 0; i < MAX_CTLR; i++) {
4405                 if (!hba[i]) {
4406                         ctlr_info_t *h;
4407
4408                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4409                         if (!h)
4410                                 goto Enomem;
4411                         hba[i] = h;
4412                         return i;
4413                 }
4414         }
4415         dev_warn(&pdev->dev, "This driver supports a maximum"
4416                " of %d controllers.\n", MAX_CTLR);
4417         return -1;
4418 Enomem:
4419         dev_warn(&pdev->dev, "out of memory.\n");
4420         return -1;
4421 }
4422
4423 static void free_hba(ctlr_info_t *h)
4424 {
4425         int i;
4426
4427         hba[h->ctlr] = NULL;
4428         for (i = 0; i < h->highest_lun + 1; i++)
4429                 if (h->gendisk[i] != NULL)
4430                         put_disk(h->gendisk[i]);
4431         kfree(h);
4432 }
4433
4434 /* Send a message CDB to the firmware. */
4435 static int cciss_message(struct pci_dev *pdev, unsigned char opcode,
4436                          unsigned char type)
4437 {
4438         typedef struct {
4439                 CommandListHeader_struct CommandHeader;
4440                 RequestBlock_struct Request;
4441                 ErrDescriptor_struct ErrorDescriptor;
4442         } Command;
4443         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4444         Command *cmd;
4445         dma_addr_t paddr64;
4446         uint32_t paddr32, tag;
4447         void __iomem *vaddr;
4448         int i, err;
4449
4450         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4451         if (vaddr == NULL)
4452                 return -ENOMEM;
4453
4454         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4455            CCISS commands, so they must be allocated from the lower 4GiB of
4456            memory. */
4457         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4458         if (err) {
4459                 iounmap(vaddr);
4460                 return -ENOMEM;
4461         }
4462
4463         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4464         if (cmd == NULL) {
4465                 iounmap(vaddr);
4466                 return -ENOMEM;
4467         }
4468
4469         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4470            although there's no guarantee, we assume that the address is at
4471            least 4-byte aligned (most likely, it's page-aligned). */
4472         paddr32 = paddr64;
4473
4474         cmd->CommandHeader.ReplyQueue = 0;
4475         cmd->CommandHeader.SGList = 0;
4476         cmd->CommandHeader.SGTotal = 0;
4477         cmd->CommandHeader.Tag.lower = paddr32;
4478         cmd->CommandHeader.Tag.upper = 0;
4479         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4480
4481         cmd->Request.CDBLen = 16;
4482         cmd->Request.Type.Type = TYPE_MSG;
4483         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4484         cmd->Request.Type.Direction = XFER_NONE;
4485         cmd->Request.Timeout = 0; /* Don't time out */
4486         cmd->Request.CDB[0] = opcode;
4487         cmd->Request.CDB[1] = type;
4488         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4489
4490         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4491         cmd->ErrorDescriptor.Addr.upper = 0;
4492         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4493
4494         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4495
4496         for (i = 0; i < 10; i++) {
4497                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4498                 if ((tag & ~3) == paddr32)
4499                         break;
4500                 msleep(CCISS_POST_RESET_NOOP_TIMEOUT_MSECS);
4501         }
4502
4503         iounmap(vaddr);
4504
4505         /* we leak the DMA buffer here ... no choice since the controller could
4506            still complete the command. */
4507         if (i == 10) {
4508                 dev_err(&pdev->dev,
4509                         "controller message %02x:%02x timed out\n",
4510                         opcode, type);
4511                 return -ETIMEDOUT;
4512         }
4513
4514         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4515
4516         if (tag & 2) {
4517                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4518                         opcode, type);
4519                 return -EIO;
4520         }
4521
4522         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4523                 opcode, type);
4524         return 0;
4525 }
4526
4527 #define cciss_noop(p) cciss_message(p, 3, 0)
4528
4529 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4530         void * __iomem vaddr, u32 use_doorbell)
4531 {
4532         u16 pmcsr;
4533         int pos;
4534
4535         if (use_doorbell) {
4536                 /* For everything after the P600, the PCI power state method
4537                  * of resetting the controller doesn't work, so we have this
4538                  * other way using the doorbell register.
4539                  */
4540                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4541                 writel(use_doorbell, vaddr + SA5_DOORBELL);
4542         } else { /* Try to do it the PCI power state way */
4543
4544                 /* Quoting from the Open CISS Specification: "The Power
4545                  * Management Control/Status Register (CSR) controls the power
4546                  * state of the device.  The normal operating state is D0,
4547                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4548                  * the controller, place the interface device in D3 then to D0,
4549                  * this causes a secondary PCI reset which will reset the
4550                  * controller." */
4551
4552                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4553                 if (pos == 0) {
4554                         dev_err(&pdev->dev,
4555                                 "cciss_controller_hard_reset: "
4556                                 "PCI PM not supported\n");
4557                         return -ENODEV;
4558                 }
4559                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4560                 /* enter the D3hot power management state */
4561                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4562                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4563                 pmcsr |= PCI_D3hot;
4564                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4565
4566                 msleep(500);
4567
4568                 /* enter the D0 power management state */
4569                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4570                 pmcsr |= PCI_D0;
4571                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4572
4573                 /*
4574                  * The P600 requires a small delay when changing states.
4575                  * Otherwise we may think the board did not reset and we bail.
4576                  * This for kdump only and is particular to the P600.
4577                  */
4578                 msleep(500);
4579         }
4580         return 0;
4581 }
4582
4583 static void init_driver_version(char *driver_version, int len)
4584 {
4585         memset(driver_version, 0, len);
4586         strncpy(driver_version, "cciss " DRIVER_NAME, len - 1);
4587 }
4588
4589 static int write_driver_ver_to_cfgtable(CfgTable_struct __iomem *cfgtable)
4590 {
4591         char *driver_version;
4592         int i, size = sizeof(cfgtable->driver_version);
4593
4594         driver_version = kmalloc(size, GFP_KERNEL);
4595         if (!driver_version)
4596                 return -ENOMEM;
4597
4598         init_driver_version(driver_version, size);
4599         for (i = 0; i < size; i++)
4600                 writeb(driver_version[i], &cfgtable->driver_version[i]);
4601         kfree(driver_version);
4602         return 0;
4603 }
4604
4605 static void read_driver_ver_from_cfgtable(CfgTable_struct __iomem *cfgtable,
4606                                           unsigned char *driver_ver)
4607 {
4608         int i;
4609
4610         for (i = 0; i < sizeof(cfgtable->driver_version); i++)
4611                 driver_ver[i] = readb(&cfgtable->driver_version[i]);
4612 }
4613
4614 static int controller_reset_failed(CfgTable_struct __iomem *cfgtable)
4615 {
4616
4617         char *driver_ver, *old_driver_ver;
4618         int rc, size = sizeof(cfgtable->driver_version);
4619
4620         old_driver_ver = kmalloc(2 * size, GFP_KERNEL);
4621         if (!old_driver_ver)
4622                 return -ENOMEM;
4623         driver_ver = old_driver_ver + size;
4624
4625         /* After a reset, the 32 bytes of "driver version" in the cfgtable
4626          * should have been changed, otherwise we know the reset failed.
4627          */
4628         init_driver_version(old_driver_ver, size);
4629         read_driver_ver_from_cfgtable(cfgtable, driver_ver);
4630         rc = !memcmp(driver_ver, old_driver_ver, size);
4631         kfree(old_driver_ver);
4632         return rc;
4633 }
4634
4635 /* This does a hard reset of the controller using PCI power management
4636  * states or using the doorbell register. */
4637 static int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4638 {
4639         u64 cfg_offset;
4640         u32 cfg_base_addr;
4641         u64 cfg_base_addr_index;
4642         void __iomem *vaddr;
4643         unsigned long paddr;
4644         u32 misc_fw_support;
4645         int rc;
4646         CfgTable_struct __iomem *cfgtable;
4647         u32 use_doorbell;
4648         u32 board_id;
4649         u16 command_register;
4650
4651         /* For controllers as old a the p600, this is very nearly
4652          * the same thing as
4653          *
4654          * pci_save_state(pci_dev);
4655          * pci_set_power_state(pci_dev, PCI_D3hot);
4656          * pci_set_power_state(pci_dev, PCI_D0);
4657          * pci_restore_state(pci_dev);
4658          *
4659          * For controllers newer than the P600, the pci power state
4660          * method of resetting doesn't work so we have another way
4661          * using the doorbell register.
4662          */
4663
4664         /* Exclude 640x boards.  These are two pci devices in one slot
4665          * which share a battery backed cache module.  One controls the
4666          * cache, the other accesses the cache through the one that controls
4667          * it.  If we reset the one controlling the cache, the other will
4668          * likely not be happy.  Just forbid resetting this conjoined mess.
4669          */
4670         cciss_lookup_board_id(pdev, &board_id);
4671         if (!ctlr_is_resettable(board_id)) {
4672                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4673                                 "due to shared cache module.");
4674                 return -ENODEV;
4675         }
4676
4677         /* if controller is soft- but not hard resettable... */
4678         if (!ctlr_is_hard_resettable(board_id))
4679                 return -ENOTSUPP; /* try soft reset later. */
4680
4681         /* Save the PCI command register */
4682         pci_read_config_word(pdev, 4, &command_register);
4683         /* Turn the board off.  This is so that later pci_restore_state()
4684          * won't turn the board on before the rest of config space is ready.
4685          */
4686         pci_disable_device(pdev);
4687         pci_save_state(pdev);
4688
4689         /* find the first memory BAR, so we can find the cfg table */
4690         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4691         if (rc)
4692                 return rc;
4693         vaddr = remap_pci_mem(paddr, 0x250);
4694         if (!vaddr)
4695                 return -ENOMEM;
4696
4697         /* find cfgtable in order to check if reset via doorbell is supported */
4698         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4699                                         &cfg_base_addr_index, &cfg_offset);
4700         if (rc)
4701                 goto unmap_vaddr;
4702         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4703                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4704         if (!cfgtable) {
4705                 rc = -ENOMEM;
4706                 goto unmap_vaddr;
4707         }
4708         rc = write_driver_ver_to_cfgtable(cfgtable);
4709         if (rc)
4710                 goto unmap_vaddr;
4711
4712         /* If reset via doorbell register is supported, use that.
4713          * There are two such methods.  Favor the newest method.
4714          */
4715         misc_fw_support = readl(&cfgtable->misc_fw_support);
4716         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET2;
4717         if (use_doorbell) {
4718                 use_doorbell = DOORBELL_CTLR_RESET2;
4719         } else {
4720                 use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4721                 if (use_doorbell) {
4722                         dev_warn(&pdev->dev, "Controller claims that "
4723                                 "'Bit 2 doorbell reset' is "
4724                                 "supported, but not 'bit 5 doorbell reset'.  "
4725                                 "Firmware update is recommended.\n");
4726                         rc = -ENOTSUPP; /* use the soft reset */
4727                         goto unmap_cfgtable;
4728                 }
4729         }
4730
4731         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4732         if (rc)
4733                 goto unmap_cfgtable;
4734         pci_restore_state(pdev);
4735         rc = pci_enable_device(pdev);
4736         if (rc) {
4737                 dev_warn(&pdev->dev, "failed to enable device.\n");
4738                 goto unmap_cfgtable;
4739         }
4740         pci_write_config_word(pdev, 4, command_register);
4741
4742         /* Some devices (notably the HP Smart Array 5i Controller)
4743            need a little pause here */
4744         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4745
4746         /* Wait for board to become not ready, then ready. */
4747         dev_info(&pdev->dev, "Waiting for board to reset.\n");
4748         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4749         if (rc) {
4750                 dev_warn(&pdev->dev, "Failed waiting for board to hard reset."
4751                                 "  Will try soft reset.\n");
4752                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4753                 goto unmap_cfgtable;
4754         }
4755         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4756         if (rc) {
4757                 dev_warn(&pdev->dev,
4758                         "failed waiting for board to become ready "
4759                         "after hard reset\n");
4760                 goto unmap_cfgtable;
4761         }
4762
4763         rc = controller_reset_failed(vaddr);
4764         if (rc < 0)
4765                 goto unmap_cfgtable;
4766         if (rc) {
4767                 dev_warn(&pdev->dev, "Unable to successfully hard reset "
4768                         "controller. Will try soft reset.\n");
4769                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4770         } else {
4771                 dev_info(&pdev->dev, "Board ready after hard reset.\n");
4772         }
4773
4774 unmap_cfgtable:
4775         iounmap(cfgtable);
4776
4777 unmap_vaddr:
4778         iounmap(vaddr);
4779         return rc;
4780 }
4781
4782 static int cciss_init_reset_devices(struct pci_dev *pdev)
4783 {
4784         int rc, i;
4785
4786         if (!reset_devices)
4787                 return 0;
4788
4789         /* Reset the controller with a PCI power-cycle or via doorbell */
4790         rc = cciss_kdump_hard_reset_controller(pdev);
4791
4792         /* -ENOTSUPP here means we cannot reset the controller
4793          * but it's already (and still) up and running in
4794          * "performant mode".  Or, it might be 640x, which can't reset
4795          * due to concerns about shared bbwc between 6402/6404 pair.
4796          */
4797         if (rc == -ENOTSUPP)
4798                 return rc; /* just try to do the kdump anyhow. */
4799         if (rc)
4800                 return -ENODEV;
4801
4802         /* Now try to get the controller to respond to a no-op */
4803         dev_warn(&pdev->dev, "Waiting for controller to respond to no-op\n");
4804         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4805                 if (cciss_noop(pdev) == 0)
4806                         break;
4807                 else
4808                         dev_warn(&pdev->dev, "no-op failed%s\n",
4809                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4810                                         "; re-trying" : ""));
4811                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4812         }
4813         return 0;
4814 }
4815
4816 static int cciss_allocate_cmd_pool(ctlr_info_t *h)
4817 {
4818         h->cmd_pool_bits = kmalloc(BITS_TO_LONGS(h->nr_cmds) *
4819                 sizeof(unsigned long), GFP_KERNEL);
4820         h->cmd_pool = pci_alloc_consistent(h->pdev,
4821                 h->nr_cmds * sizeof(CommandList_struct),
4822                 &(h->cmd_pool_dhandle));
4823         h->errinfo_pool = pci_alloc_consistent(h->pdev,
4824                 h->nr_cmds * sizeof(ErrorInfo_struct),
4825                 &(h->errinfo_pool_dhandle));
4826         if ((h->cmd_pool_bits == NULL)
4827                 || (h->cmd_pool == NULL)
4828                 || (h->errinfo_pool == NULL)) {
4829                 dev_err(&h->pdev->dev, "out of memory");
4830                 return -ENOMEM;
4831         }
4832         return 0;
4833 }
4834
4835 static int cciss_allocate_scatterlists(ctlr_info_t *h)
4836 {
4837         int i;
4838
4839         /* zero it, so that on free we need not know how many were alloc'ed */
4840         h->scatter_list = kzalloc(h->max_commands *
4841                                 sizeof(struct scatterlist *), GFP_KERNEL);
4842         if (!h->scatter_list)
4843                 return -ENOMEM;
4844
4845         for (i = 0; i < h->nr_cmds; i++) {
4846                 h->scatter_list[i] = kmalloc(sizeof(struct scatterlist) *
4847                                                 h->maxsgentries, GFP_KERNEL);
4848                 if (h->scatter_list[i] == NULL) {
4849                         dev_err(&h->pdev->dev, "could not allocate "
4850                                 "s/g lists\n");
4851                         return -ENOMEM;
4852                 }
4853         }
4854         return 0;
4855 }
4856
4857 static void cciss_free_scatterlists(ctlr_info_t *h)
4858 {
4859         int i;
4860
4861         if (h->scatter_list) {
4862                 for (i = 0; i < h->nr_cmds; i++)
4863                         kfree(h->scatter_list[i]);
4864                 kfree(h->scatter_list);
4865         }
4866 }
4867
4868 static void cciss_free_cmd_pool(ctlr_info_t *h)
4869 {
4870         kfree(h->cmd_pool_bits);
4871         if (h->cmd_pool)
4872                 pci_free_consistent(h->pdev,
4873                         h->nr_cmds * sizeof(CommandList_struct),
4874                         h->cmd_pool, h->cmd_pool_dhandle);
4875         if (h->errinfo_pool)
4876                 pci_free_consistent(h->pdev,
4877                         h->nr_cmds * sizeof(ErrorInfo_struct),
4878                         h->errinfo_pool, h->errinfo_pool_dhandle);
4879 }
4880
4881 static int cciss_request_irq(ctlr_info_t *h,
4882         irqreturn_t (*msixhandler)(int, void *),
4883         irqreturn_t (*intxhandler)(int, void *))
4884 {
4885         if (h->msix_vector || h->msi_vector) {
4886                 if (!request_irq(h->intr[h->intr_mode], msixhandler,
4887                                 0, h->devname, h))
4888                         return 0;
4889                 dev_err(&h->pdev->dev, "Unable to get msi irq %d"
4890                         " for %s\n", h->intr[h->intr_mode],
4891                         h->devname);
4892                 return -1;
4893         }
4894
4895         if (!request_irq(h->intr[h->intr_mode], intxhandler,
4896                         IRQF_SHARED, h->devname, h))
4897                 return 0;
4898         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4899                 h->intr[h->intr_mode], h->devname);
4900         return -1;
4901 }
4902
4903 static int cciss_kdump_soft_reset(ctlr_info_t *h)
4904 {
4905         if (cciss_send_reset(h, CTLR_LUNID, CCISS_RESET_TYPE_CONTROLLER)) {
4906                 dev_warn(&h->pdev->dev, "Resetting array controller failed.\n");
4907                 return -EIO;
4908         }
4909
4910         dev_info(&h->pdev->dev, "Waiting for board to soft reset.\n");
4911         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_NOT_READY)) {
4912                 dev_warn(&h->pdev->dev, "Soft reset had no effect.\n");
4913                 return -1;
4914         }
4915
4916         dev_info(&h->pdev->dev, "Board reset, awaiting READY status.\n");
4917         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY)) {
4918                 dev_warn(&h->pdev->dev, "Board failed to become ready "
4919                         "after soft reset.\n");
4920                 return -1;
4921         }
4922
4923         return 0;
4924 }
4925
4926 static void cciss_undo_allocations_after_kdump_soft_reset(ctlr_info_t *h)
4927 {
4928         int ctlr = h->ctlr;
4929
4930         free_irq(h->intr[h->intr_mode], h);
4931 #ifdef CONFIG_PCI_MSI
4932         if (h->msix_vector)
4933                 pci_disable_msix(h->pdev);
4934         else if (h->msi_vector)
4935                 pci_disable_msi(h->pdev);
4936 #endif /* CONFIG_PCI_MSI */
4937         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4938         cciss_free_scatterlists(h);
4939         cciss_free_cmd_pool(h);
4940         kfree(h->blockFetchTable);
4941         if (h->reply_pool)
4942                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
4943                                 h->reply_pool, h->reply_pool_dhandle);
4944         if (h->transtable)
4945                 iounmap(h->transtable);
4946         if (h->cfgtable)
4947                 iounmap(h->cfgtable);
4948         if (h->vaddr)
4949                 iounmap(h->vaddr);
4950         unregister_blkdev(h->major, h->devname);
4951         cciss_destroy_hba_sysfs_entry(h);
4952         pci_release_regions(h->pdev);
4953         kfree(h);
4954         hba[ctlr] = NULL;
4955 }
4956
4957 /*
4958  *  This is it.  Find all the controllers and register them.  I really hate
4959  *  stealing all these major device numbers.
4960  *  returns the number of block devices registered.
4961  */
4962 static int cciss_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4963 {
4964         int i;
4965         int j = 0;
4966         int rc;
4967         int try_soft_reset = 0;
4968         int dac, return_code;
4969         InquiryData_struct *inq_buff;
4970         ctlr_info_t *h;
4971         unsigned long flags;
4972
4973         /*
4974          * By default the cciss driver is used for all older HP Smart Array
4975          * controllers. There are module paramaters that allow a user to
4976          * override this behavior and instead use the hpsa SCSI driver. If
4977          * this is the case cciss may be loaded first from the kdump initrd
4978          * image and cause a kernel panic. So if reset_devices is true and
4979          * cciss_allow_hpsa is set just bail.
4980          */
4981         if ((reset_devices) && (cciss_allow_hpsa == 1))
4982                 return -ENODEV;
4983         rc = cciss_init_reset_devices(pdev);
4984         if (rc) {
4985                 if (rc != -ENOTSUPP)
4986                         return rc;
4987                 /* If the reset fails in a particular way (it has no way to do
4988                  * a proper hard reset, so returns -ENOTSUPP) we can try to do
4989                  * a soft reset once we get the controller configured up to the
4990                  * point that it can accept a command.
4991                  */
4992                 try_soft_reset = 1;
4993                 rc = 0;
4994         }
4995
4996 reinit_after_soft_reset:
4997
4998         i = alloc_cciss_hba(pdev);
4999         if (i < 0)
5000                 return -1;
5001
5002         h = hba[i];
5003         h->pdev = pdev;
5004         h->busy_initializing = 1;
5005         h->intr_mode = cciss_simple_mode ? SIMPLE_MODE_INT : PERF_MODE_INT;
5006         INIT_LIST_HEAD(&h->cmpQ);
5007         INIT_LIST_HEAD(&h->reqQ);
5008         mutex_init(&h->busy_shutting_down);
5009
5010         if (cciss_pci_init(h) != 0)
5011                 goto clean_no_release_regions;
5012
5013         sprintf(h->devname, "cciss%d", i);
5014         h->ctlr = i;
5015
5016         if (cciss_tape_cmds < 2)
5017                 cciss_tape_cmds = 2;
5018         if (cciss_tape_cmds > 16)
5019                 cciss_tape_cmds = 16;
5020
5021         init_completion(&h->scan_wait);
5022
5023         if (cciss_create_hba_sysfs_entry(h))
5024                 goto clean0;
5025
5026         /* configure PCI DMA stuff */
5027         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
5028                 dac = 1;
5029         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
5030                 dac = 0;
5031         else {
5032                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
5033                 goto clean1;
5034         }
5035
5036         /*
5037          * register with the major number, or get a dynamic major number
5038          * by passing 0 as argument.  This is done for greater than
5039          * 8 controller support.
5040          */
5041         if (i < MAX_CTLR_ORIG)
5042                 h->major = COMPAQ_CISS_MAJOR + i;
5043         rc = register_blkdev(h->major, h->devname);
5044         if (rc == -EBUSY || rc == -EINVAL) {
5045                 dev_err(&h->pdev->dev,
5046                        "Unable to get major number %d for %s "
5047                        "on hba %d\n", h->major, h->devname, i);
5048                 goto clean1;
5049         } else {
5050                 if (i >= MAX_CTLR_ORIG)
5051                         h->major = rc;
5052         }
5053
5054         /* make sure the board interrupts are off */
5055         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5056         rc = cciss_request_irq(h, do_cciss_msix_intr, do_cciss_intx);
5057         if (rc)
5058                 goto clean2;
5059
5060         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
5061                h->devname, pdev->device, pci_name(pdev),
5062                h->intr[h->intr_mode], dac ? "" : " not");
5063
5064         if (cciss_allocate_cmd_pool(h))
5065                 goto clean4;
5066
5067         if (cciss_allocate_scatterlists(h))
5068                 goto clean4;
5069
5070         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
5071                 h->chainsize, h->nr_cmds);
5072         if (!h->cmd_sg_list && h->chainsize > 0)
5073                 goto clean4;
5074
5075         spin_lock_init(&h->lock);
5076
5077         /* Initialize the pdev driver private data.
5078            have it point to h.  */
5079         pci_set_drvdata(pdev, h);
5080         /* command and error info recs zeroed out before
5081            they are used */
5082         bitmap_zero(h->cmd_pool_bits, h->nr_cmds);
5083
5084         h->num_luns = 0;
5085         h->highest_lun = -1;
5086         for (j = 0; j < CISS_MAX_LUN; j++) {
5087                 h->drv[j] = NULL;
5088                 h->gendisk[j] = NULL;
5089         }
5090
5091         /* At this point, the controller is ready to take commands.
5092          * Now, if reset_devices and the hard reset didn't work, try
5093          * the soft reset and see if that works.
5094          */
5095         if (try_soft_reset) {
5096
5097                 /* This is kind of gross.  We may or may not get a completion
5098                  * from the soft reset command, and if we do, then the value
5099                  * from the fifo may or may not be valid.  So, we wait 10 secs
5100                  * after the reset throwing away any completions we get during
5101                  * that time.  Unregister the interrupt handler and register
5102                  * fake ones to scoop up any residual completions.
5103                  */
5104                 spin_lock_irqsave(&h->lock, flags);
5105                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5106                 spin_unlock_irqrestore(&h->lock, flags);
5107                 free_irq(h->intr[h->intr_mode], h);
5108                 rc = cciss_request_irq(h, cciss_msix_discard_completions,
5109                                         cciss_intx_discard_completions);
5110                 if (rc) {
5111                         dev_warn(&h->pdev->dev, "Failed to request_irq after "
5112                                 "soft reset.\n");
5113                         goto clean4;
5114                 }
5115
5116                 rc = cciss_kdump_soft_reset(h);
5117                 if (rc) {
5118                         dev_warn(&h->pdev->dev, "Soft reset failed.\n");
5119                         goto clean4;
5120                 }
5121
5122                 dev_info(&h->pdev->dev, "Board READY.\n");
5123                 dev_info(&h->pdev->dev,
5124                         "Waiting for stale completions to drain.\n");
5125                 h->access.set_intr_mask(h, CCISS_INTR_ON);
5126                 msleep(10000);
5127                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5128
5129                 rc = controller_reset_failed(h->cfgtable);
5130                 if (rc)
5131                         dev_info(&h->pdev->dev,
5132                                 "Soft reset appears to have failed.\n");
5133
5134                 /* since the controller's reset, we have to go back and re-init
5135                  * everything.  Easiest to just forget what we've done and do it
5136                  * all over again.
5137                  */
5138                 cciss_undo_allocations_after_kdump_soft_reset(h);
5139                 try_soft_reset = 0;
5140                 if (rc)
5141                         /* don't go to clean4, we already unallocated */
5142                         return -ENODEV;
5143
5144                 goto reinit_after_soft_reset;
5145         }
5146
5147         cciss_scsi_setup(h);
5148
5149         /* Turn the interrupts on so we can service requests */
5150         h->access.set_intr_mask(h, CCISS_INTR_ON);
5151
5152         /* Get the firmware version */
5153         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
5154         if (inq_buff == NULL) {
5155                 dev_err(&h->pdev->dev, "out of memory\n");
5156                 goto clean4;
5157         }
5158
5159         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
5160                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
5161         if (return_code == IO_OK) {
5162                 h->firm_ver[0] = inq_buff->data_byte[32];
5163                 h->firm_ver[1] = inq_buff->data_byte[33];
5164                 h->firm_ver[2] = inq_buff->data_byte[34];
5165                 h->firm_ver[3] = inq_buff->data_byte[35];
5166         } else {         /* send command failed */
5167                 dev_warn(&h->pdev->dev, "unable to determine firmware"
5168                         " version of controller\n");
5169         }
5170         kfree(inq_buff);
5171
5172         cciss_procinit(h);
5173
5174         h->cciss_max_sectors = 8192;
5175
5176         rebuild_lun_table(h, 1, 0);
5177         cciss_engage_scsi(h);
5178         h->busy_initializing = 0;
5179         return 1;
5180
5181 clean4:
5182         cciss_free_cmd_pool(h);
5183         cciss_free_scatterlists(h);
5184         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5185         free_irq(h->intr[h->intr_mode], h);
5186 clean2:
5187         unregister_blkdev(h->major, h->devname);
5188 clean1:
5189         cciss_destroy_hba_sysfs_entry(h);
5190 clean0:
5191         pci_release_regions(pdev);
5192 clean_no_release_regions:
5193         h->busy_initializing = 0;
5194
5195         /*
5196          * Deliberately omit pci_disable_device(): it does something nasty to
5197          * Smart Array controllers that pci_enable_device does not undo
5198          */
5199         pci_set_drvdata(pdev, NULL);
5200         free_hba(h);
5201         return -1;
5202 }
5203
5204 static void cciss_shutdown(struct pci_dev *pdev)
5205 {
5206         ctlr_info_t *h;
5207         char *flush_buf;
5208         int return_code;
5209
5210         h = pci_get_drvdata(pdev);
5211         flush_buf = kzalloc(4, GFP_KERNEL);
5212         if (!flush_buf) {
5213                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
5214                 return;
5215         }
5216         /* write all data in the battery backed cache to disk */
5217         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
5218                 4, 0, CTLR_LUNID, TYPE_CMD);
5219         kfree(flush_buf);
5220         if (return_code != IO_OK)
5221                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
5222         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5223         free_irq(h->intr[h->intr_mode], h);
5224 }
5225
5226 static int cciss_enter_simple_mode(struct ctlr_info *h)
5227 {
5228         u32 trans_support;
5229
5230         trans_support = readl(&(h->cfgtable->TransportSupport));
5231         if (!(trans_support & SIMPLE_MODE))
5232                 return -ENOTSUPP;
5233
5234         h->max_commands = readl(&(h->cfgtable->CmdsOutMax));
5235         writel(CFGTBL_Trans_Simple, &(h->cfgtable->HostWrite.TransportRequest));
5236         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
5237         cciss_wait_for_mode_change_ack(h);
5238         print_cfg_table(h);
5239         if (!(readl(&(h->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
5240                 dev_warn(&h->pdev->dev, "unable to get board into simple mode\n");
5241                 return -ENODEV;
5242         }
5243         h->transMethod = CFGTBL_Trans_Simple;
5244         return 0;
5245 }
5246
5247
5248 static void cciss_remove_one(struct pci_dev *pdev)
5249 {
5250         ctlr_info_t *h;
5251         int i, j;
5252
5253         if (pci_get_drvdata(pdev) == NULL) {
5254                 dev_err(&pdev->dev, "Unable to remove device\n");
5255                 return;
5256         }
5257
5258         h = pci_get_drvdata(pdev);
5259         i = h->ctlr;
5260         if (hba[i] == NULL) {
5261                 dev_err(&pdev->dev, "device appears to already be removed\n");
5262                 return;
5263         }
5264
5265         mutex_lock(&h->busy_shutting_down);
5266
5267         remove_from_scan_list(h);
5268         remove_proc_entry(h->devname, proc_cciss);
5269         unregister_blkdev(h->major, h->devname);
5270
5271         /* remove it from the disk list */
5272         for (j = 0; j < CISS_MAX_LUN; j++) {
5273                 struct gendisk *disk = h->gendisk[j];
5274                 if (disk) {
5275                         struct request_queue *q = disk->queue;
5276
5277                         if (disk->flags & GENHD_FL_UP) {
5278                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
5279                                 del_gendisk(disk);
5280                         }
5281                         if (q)
5282                                 blk_cleanup_queue(q);
5283                 }
5284         }
5285
5286 #ifdef CONFIG_CISS_SCSI_TAPE
5287         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
5288 #endif
5289
5290         cciss_shutdown(pdev);
5291
5292 #ifdef CONFIG_PCI_MSI
5293         if (h->msix_vector)
5294                 pci_disable_msix(h->pdev);
5295         else if (h->msi_vector)
5296                 pci_disable_msi(h->pdev);
5297 #endif                          /* CONFIG_PCI_MSI */
5298
5299         iounmap(h->transtable);
5300         iounmap(h->cfgtable);
5301         iounmap(h->vaddr);
5302
5303         cciss_free_cmd_pool(h);
5304         /* Free up sg elements */
5305         for (j = 0; j < h->nr_cmds; j++)
5306                 kfree(h->scatter_list[j]);
5307         kfree(h->scatter_list);
5308         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5309         kfree(h->blockFetchTable);
5310         if (h->reply_pool)
5311                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
5312                                 h->reply_pool, h->reply_pool_dhandle);
5313         /*
5314          * Deliberately omit pci_disable_device(): it does something nasty to
5315          * Smart Array controllers that pci_enable_device does not undo
5316          */
5317         pci_release_regions(pdev);
5318         pci_set_drvdata(pdev, NULL);
5319         cciss_destroy_hba_sysfs_entry(h);
5320         mutex_unlock(&h->busy_shutting_down);
5321         free_hba(h);
5322 }
5323
5324 static struct pci_driver cciss_pci_driver = {
5325         .name = "cciss",
5326         .probe = cciss_init_one,
5327         .remove = cciss_remove_one,
5328         .id_table = cciss_pci_device_id,        /* id_table */
5329         .shutdown = cciss_shutdown,
5330 };
5331
5332 /*
5333  *  This is it.  Register the PCI driver information for the cards we control
5334  *  the OS will call our registered routines when it finds one of our cards.
5335  */
5336 static int __init cciss_init(void)
5337 {
5338         int err;
5339
5340         /*
5341          * The hardware requires that commands are aligned on a 64-bit
5342          * boundary. Given that we use pci_alloc_consistent() to allocate an
5343          * array of them, the size must be a multiple of 8 bytes.
5344          */
5345         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
5346         printk(KERN_INFO DRIVER_NAME "\n");
5347
5348         err = bus_register(&cciss_bus_type);
5349         if (err)
5350                 return err;
5351
5352         /* Start the scan thread */
5353         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
5354         if (IS_ERR(cciss_scan_thread)) {
5355                 err = PTR_ERR(cciss_scan_thread);
5356                 goto err_bus_unregister;
5357         }
5358
5359         /* Register for our PCI devices */
5360         err = pci_register_driver(&cciss_pci_driver);
5361         if (err)
5362                 goto err_thread_stop;
5363
5364         return err;
5365
5366 err_thread_stop:
5367         kthread_stop(cciss_scan_thread);
5368 err_bus_unregister:
5369         bus_unregister(&cciss_bus_type);
5370
5371         return err;
5372 }
5373
5374 static void __exit cciss_cleanup(void)
5375 {
5376         int i;
5377
5378         pci_unregister_driver(&cciss_pci_driver);
5379         /* double check that all controller entrys have been removed */
5380         for (i = 0; i < MAX_CTLR; i++) {
5381                 if (hba[i] != NULL) {
5382                         dev_warn(&hba[i]->pdev->dev,
5383                                 "had to remove controller\n");
5384                         cciss_remove_one(hba[i]->pdev);
5385                 }
5386         }
5387         kthread_stop(cciss_scan_thread);
5388         if (proc_cciss)
5389                 remove_proc_entry("driver/cciss", NULL);
5390         bus_unregister(&cciss_bus_type);
5391 }
5392
5393 module_init(cciss_init);
5394 module_exit(cciss_cleanup);