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         err = 0;
1193         err |=
1194             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1195                            sizeof(arg64.LUN_info));
1196         err |=
1197             copy_from_user(&arg64.Request, &arg32->Request,
1198                            sizeof(arg64.Request));
1199         err |=
1200             copy_from_user(&arg64.error_info, &arg32->error_info,
1201                            sizeof(arg64.error_info));
1202         err |= get_user(arg64.buf_size, &arg32->buf_size);
1203         err |= get_user(cp, &arg32->buf);
1204         arg64.buf = compat_ptr(cp);
1205         err |= copy_to_user(p, &arg64, sizeof(arg64));
1206
1207         if (err)
1208                 return -EFAULT;
1209
1210         err = cciss_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1211         if (err)
1212                 return err;
1213         err |=
1214             copy_in_user(&arg32->error_info, &p->error_info,
1215                          sizeof(arg32->error_info));
1216         if (err)
1217                 return -EFAULT;
1218         return err;
1219 }
1220
1221 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1222                                       unsigned cmd, unsigned long arg)
1223 {
1224         BIG_IOCTL32_Command_struct __user *arg32 =
1225             (BIG_IOCTL32_Command_struct __user *) arg;
1226         BIG_IOCTL_Command_struct arg64;
1227         BIG_IOCTL_Command_struct __user *p =
1228             compat_alloc_user_space(sizeof(arg64));
1229         int err;
1230         u32 cp;
1231
1232         memset(&arg64, 0, sizeof(arg64));
1233         err = 0;
1234         err |=
1235             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1236                            sizeof(arg64.LUN_info));
1237         err |=
1238             copy_from_user(&arg64.Request, &arg32->Request,
1239                            sizeof(arg64.Request));
1240         err |=
1241             copy_from_user(&arg64.error_info, &arg32->error_info,
1242                            sizeof(arg64.error_info));
1243         err |= get_user(arg64.buf_size, &arg32->buf_size);
1244         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1245         err |= get_user(cp, &arg32->buf);
1246         arg64.buf = compat_ptr(cp);
1247         err |= copy_to_user(p, &arg64, sizeof(arg64));
1248
1249         if (err)
1250                 return -EFAULT;
1251
1252         err = cciss_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1253         if (err)
1254                 return err;
1255         err |=
1256             copy_in_user(&arg32->error_info, &p->error_info,
1257                          sizeof(arg32->error_info));
1258         if (err)
1259                 return -EFAULT;
1260         return err;
1261 }
1262 #endif
1263
1264 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1265 {
1266         drive_info_struct *drv = get_drv(bdev->bd_disk);
1267
1268         if (!drv->cylinders)
1269                 return -ENXIO;
1270
1271         geo->heads = drv->heads;
1272         geo->sectors = drv->sectors;
1273         geo->cylinders = drv->cylinders;
1274         return 0;
1275 }
1276
1277 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1278 {
1279         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1280                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1281                 (void)check_for_unit_attention(h, c);
1282 }
1283
1284 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1285 {
1286         cciss_pci_info_struct pciinfo;
1287
1288         if (!argp)
1289                 return -EINVAL;
1290         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1291         pciinfo.bus = h->pdev->bus->number;
1292         pciinfo.dev_fn = h->pdev->devfn;
1293         pciinfo.board_id = h->board_id;
1294         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1295                 return -EFAULT;
1296         return 0;
1297 }
1298
1299 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1300 {
1301         cciss_coalint_struct intinfo;
1302         unsigned long flags;
1303
1304         if (!argp)
1305                 return -EINVAL;
1306         spin_lock_irqsave(&h->lock, flags);
1307         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1308         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1309         spin_unlock_irqrestore(&h->lock, flags);
1310         if (copy_to_user
1311             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1312                 return -EFAULT;
1313         return 0;
1314 }
1315
1316 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1317 {
1318         cciss_coalint_struct intinfo;
1319         unsigned long flags;
1320         int i;
1321
1322         if (!argp)
1323                 return -EINVAL;
1324         if (!capable(CAP_SYS_ADMIN))
1325                 return -EPERM;
1326         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1327                 return -EFAULT;
1328         if ((intinfo.delay == 0) && (intinfo.count == 0))
1329                 return -EINVAL;
1330         spin_lock_irqsave(&h->lock, flags);
1331         /* Update the field, and then ring the doorbell */
1332         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1333         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1334         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1335
1336         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1337                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1338                         break;
1339                 udelay(1000); /* delay and try again */
1340         }
1341         spin_unlock_irqrestore(&h->lock, flags);
1342         if (i >= MAX_IOCTL_CONFIG_WAIT)
1343                 return -EAGAIN;
1344         return 0;
1345 }
1346
1347 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1348 {
1349         NodeName_type NodeName;
1350         unsigned long flags;
1351         int i;
1352
1353         if (!argp)
1354                 return -EINVAL;
1355         spin_lock_irqsave(&h->lock, flags);
1356         for (i = 0; i < 16; i++)
1357                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1358         spin_unlock_irqrestore(&h->lock, flags);
1359         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1360                 return -EFAULT;
1361         return 0;
1362 }
1363
1364 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1365 {
1366         NodeName_type NodeName;
1367         unsigned long flags;
1368         int i;
1369
1370         if (!argp)
1371                 return -EINVAL;
1372         if (!capable(CAP_SYS_ADMIN))
1373                 return -EPERM;
1374         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1375                 return -EFAULT;
1376         spin_lock_irqsave(&h->lock, flags);
1377         /* Update the field, and then ring the doorbell */
1378         for (i = 0; i < 16; i++)
1379                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1380         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1381         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1382                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1383                         break;
1384                 udelay(1000); /* delay and try again */
1385         }
1386         spin_unlock_irqrestore(&h->lock, flags);
1387         if (i >= MAX_IOCTL_CONFIG_WAIT)
1388                 return -EAGAIN;
1389         return 0;
1390 }
1391
1392 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1393 {
1394         Heartbeat_type heartbeat;
1395         unsigned long flags;
1396
1397         if (!argp)
1398                 return -EINVAL;
1399         spin_lock_irqsave(&h->lock, flags);
1400         heartbeat = readl(&h->cfgtable->HeartBeat);
1401         spin_unlock_irqrestore(&h->lock, flags);
1402         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1403                 return -EFAULT;
1404         return 0;
1405 }
1406
1407 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1408 {
1409         BusTypes_type BusTypes;
1410         unsigned long flags;
1411
1412         if (!argp)
1413                 return -EINVAL;
1414         spin_lock_irqsave(&h->lock, flags);
1415         BusTypes = readl(&h->cfgtable->BusTypes);
1416         spin_unlock_irqrestore(&h->lock, flags);
1417         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1418                 return -EFAULT;
1419         return 0;
1420 }
1421
1422 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1423 {
1424         FirmwareVer_type firmware;
1425
1426         if (!argp)
1427                 return -EINVAL;
1428         memcpy(firmware, h->firm_ver, 4);
1429
1430         if (copy_to_user
1431             (argp, firmware, sizeof(FirmwareVer_type)))
1432                 return -EFAULT;
1433         return 0;
1434 }
1435
1436 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1437 {
1438         DriverVer_type DriverVer = DRIVER_VERSION;
1439
1440         if (!argp)
1441                 return -EINVAL;
1442         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1443                 return -EFAULT;
1444         return 0;
1445 }
1446
1447 static int cciss_getluninfo(ctlr_info_t *h,
1448         struct gendisk *disk, void __user *argp)
1449 {
1450         LogvolInfo_struct luninfo;
1451         drive_info_struct *drv = get_drv(disk);
1452
1453         if (!argp)
1454                 return -EINVAL;
1455         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1456         luninfo.num_opens = drv->usage_count;
1457         luninfo.num_parts = 0;
1458         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1459                 return -EFAULT;
1460         return 0;
1461 }
1462
1463 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1464 {
1465         IOCTL_Command_struct iocommand;
1466         CommandList_struct *c;
1467         char *buff = NULL;
1468         u64bit temp64;
1469         DECLARE_COMPLETION_ONSTACK(wait);
1470
1471         if (!argp)
1472                 return -EINVAL;
1473
1474         if (!capable(CAP_SYS_RAWIO))
1475                 return -EPERM;
1476
1477         if (copy_from_user
1478             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1479                 return -EFAULT;
1480         if ((iocommand.buf_size < 1) &&
1481             (iocommand.Request.Type.Direction != XFER_NONE)) {
1482                 return -EINVAL;
1483         }
1484         if (iocommand.buf_size > 0) {
1485                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1486                 if (buff == NULL)
1487                         return -EFAULT;
1488         }
1489         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1490                 /* Copy the data into the buffer we created */
1491                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1492                         kfree(buff);
1493                         return -EFAULT;
1494                 }
1495         } else {
1496                 memset(buff, 0, iocommand.buf_size);
1497         }
1498         c = cmd_special_alloc(h);
1499         if (!c) {
1500                 kfree(buff);
1501                 return -ENOMEM;
1502         }
1503         /* Fill in the command type */
1504         c->cmd_type = CMD_IOCTL_PEND;
1505         /* Fill in Command Header */
1506         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1507         if (iocommand.buf_size > 0) { /* buffer to fill */
1508                 c->Header.SGList = 1;
1509                 c->Header.SGTotal = 1;
1510         } else { /* no buffers to fill */
1511                 c->Header.SGList = 0;
1512                 c->Header.SGTotal = 0;
1513         }
1514         c->Header.LUN = iocommand.LUN_info;
1515         /* use the kernel address the cmd block for tag */
1516         c->Header.Tag.lower = c->busaddr;
1517
1518         /* Fill in Request block */
1519         c->Request = iocommand.Request;
1520
1521         /* Fill in the scatter gather information */
1522         if (iocommand.buf_size > 0) {
1523                 temp64.val = pci_map_single(h->pdev, buff,
1524                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1525                 c->SG[0].Addr.lower = temp64.val32.lower;
1526                 c->SG[0].Addr.upper = temp64.val32.upper;
1527                 c->SG[0].Len = iocommand.buf_size;
1528                 c->SG[0].Ext = 0;  /* we are not chaining */
1529         }
1530         c->waiting = &wait;
1531
1532         enqueue_cmd_and_start_io(h, c);
1533         wait_for_completion(&wait);
1534
1535         /* unlock the buffers from DMA */
1536         temp64.val32.lower = c->SG[0].Addr.lower;
1537         temp64.val32.upper = c->SG[0].Addr.upper;
1538         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1539                          PCI_DMA_BIDIRECTIONAL);
1540         check_ioctl_unit_attention(h, c);
1541
1542         /* Copy the error information out */
1543         iocommand.error_info = *(c->err_info);
1544         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1545                 kfree(buff);
1546                 cmd_special_free(h, c);
1547                 return -EFAULT;
1548         }
1549
1550         if (iocommand.Request.Type.Direction == XFER_READ) {
1551                 /* Copy the data out of the buffer we created */
1552                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1553                         kfree(buff);
1554                         cmd_special_free(h, c);
1555                         return -EFAULT;
1556                 }
1557         }
1558         kfree(buff);
1559         cmd_special_free(h, c);
1560         return 0;
1561 }
1562
1563 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1564 {
1565         BIG_IOCTL_Command_struct *ioc;
1566         CommandList_struct *c;
1567         unsigned char **buff = NULL;
1568         int *buff_size = NULL;
1569         u64bit temp64;
1570         BYTE sg_used = 0;
1571         int status = 0;
1572         int i;
1573         DECLARE_COMPLETION_ONSTACK(wait);
1574         __u32 left;
1575         __u32 sz;
1576         BYTE __user *data_ptr;
1577
1578         if (!argp)
1579                 return -EINVAL;
1580         if (!capable(CAP_SYS_RAWIO))
1581                 return -EPERM;
1582         ioc = kmalloc(sizeof(*ioc), GFP_KERNEL);
1583         if (!ioc) {
1584                 status = -ENOMEM;
1585                 goto cleanup1;
1586         }
1587         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1588                 status = -EFAULT;
1589                 goto cleanup1;
1590         }
1591         if ((ioc->buf_size < 1) &&
1592             (ioc->Request.Type.Direction != XFER_NONE)) {
1593                 status = -EINVAL;
1594                 goto cleanup1;
1595         }
1596         /* Check kmalloc limits  using all SGs */
1597         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1598                 status = -EINVAL;
1599                 goto cleanup1;
1600         }
1601         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1602                 status = -EINVAL;
1603                 goto cleanup1;
1604         }
1605         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1606         if (!buff) {
1607                 status = -ENOMEM;
1608                 goto cleanup1;
1609         }
1610         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1611         if (!buff_size) {
1612                 status = -ENOMEM;
1613                 goto cleanup1;
1614         }
1615         left = ioc->buf_size;
1616         data_ptr = ioc->buf;
1617         while (left) {
1618                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1619                 buff_size[sg_used] = sz;
1620                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1621                 if (buff[sg_used] == NULL) {
1622                         status = -ENOMEM;
1623                         goto cleanup1;
1624                 }
1625                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1626                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1627                                 status = -EFAULT;
1628                                 goto cleanup1;
1629                         }
1630                 } else {
1631                         memset(buff[sg_used], 0, sz);
1632                 }
1633                 left -= sz;
1634                 data_ptr += sz;
1635                 sg_used++;
1636         }
1637         c = cmd_special_alloc(h);
1638         if (!c) {
1639                 status = -ENOMEM;
1640                 goto cleanup1;
1641         }
1642         c->cmd_type = CMD_IOCTL_PEND;
1643         c->Header.ReplyQueue = 0;
1644         c->Header.SGList = sg_used;
1645         c->Header.SGTotal = sg_used;
1646         c->Header.LUN = ioc->LUN_info;
1647         c->Header.Tag.lower = c->busaddr;
1648
1649         c->Request = ioc->Request;
1650         for (i = 0; i < sg_used; i++) {
1651                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1652                                     PCI_DMA_BIDIRECTIONAL);
1653                 c->SG[i].Addr.lower = temp64.val32.lower;
1654                 c->SG[i].Addr.upper = temp64.val32.upper;
1655                 c->SG[i].Len = buff_size[i];
1656                 c->SG[i].Ext = 0;       /* we are not chaining */
1657         }
1658         c->waiting = &wait;
1659         enqueue_cmd_and_start_io(h, c);
1660         wait_for_completion(&wait);
1661         /* unlock the buffers from DMA */
1662         for (i = 0; i < sg_used; i++) {
1663                 temp64.val32.lower = c->SG[i].Addr.lower;
1664                 temp64.val32.upper = c->SG[i].Addr.upper;
1665                 pci_unmap_single(h->pdev,
1666                         (dma_addr_t) temp64.val, buff_size[i],
1667                         PCI_DMA_BIDIRECTIONAL);
1668         }
1669         check_ioctl_unit_attention(h, c);
1670         /* Copy the error information out */
1671         ioc->error_info = *(c->err_info);
1672         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1673                 cmd_special_free(h, c);
1674                 status = -EFAULT;
1675                 goto cleanup1;
1676         }
1677         if (ioc->Request.Type.Direction == XFER_READ) {
1678                 /* Copy the data out of the buffer we created */
1679                 BYTE __user *ptr = ioc->buf;
1680                 for (i = 0; i < sg_used; i++) {
1681                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1682                                 cmd_special_free(h, c);
1683                                 status = -EFAULT;
1684                                 goto cleanup1;
1685                         }
1686                         ptr += buff_size[i];
1687                 }
1688         }
1689         cmd_special_free(h, c);
1690         status = 0;
1691 cleanup1:
1692         if (buff) {
1693                 for (i = 0; i < sg_used; i++)
1694                         kfree(buff[i]);
1695                 kfree(buff);
1696         }
1697         kfree(buff_size);
1698         kfree(ioc);
1699         return status;
1700 }
1701
1702 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1703         unsigned int cmd, unsigned long arg)
1704 {
1705         struct gendisk *disk = bdev->bd_disk;
1706         ctlr_info_t *h = get_host(disk);
1707         void __user *argp = (void __user *)arg;
1708
1709         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1710                 cmd, arg);
1711         switch (cmd) {
1712         case CCISS_GETPCIINFO:
1713                 return cciss_getpciinfo(h, argp);
1714         case CCISS_GETINTINFO:
1715                 return cciss_getintinfo(h, argp);
1716         case CCISS_SETINTINFO:
1717                 return cciss_setintinfo(h, argp);
1718         case CCISS_GETNODENAME:
1719                 return cciss_getnodename(h, argp);
1720         case CCISS_SETNODENAME:
1721                 return cciss_setnodename(h, argp);
1722         case CCISS_GETHEARTBEAT:
1723                 return cciss_getheartbeat(h, argp);
1724         case CCISS_GETBUSTYPES:
1725                 return cciss_getbustypes(h, argp);
1726         case CCISS_GETFIRMVER:
1727                 return cciss_getfirmver(h, argp);
1728         case CCISS_GETDRIVVER:
1729                 return cciss_getdrivver(h, argp);
1730         case CCISS_DEREGDISK:
1731         case CCISS_REGNEWD:
1732         case CCISS_REVALIDVOLS:
1733                 return rebuild_lun_table(h, 0, 1);
1734         case CCISS_GETLUNINFO:
1735                 return cciss_getluninfo(h, disk, argp);
1736         case CCISS_PASSTHRU:
1737                 return cciss_passthru(h, argp);
1738         case CCISS_BIG_PASSTHRU:
1739                 return cciss_bigpassthru(h, argp);
1740
1741         /* scsi_cmd_blk_ioctl handles these, below, though some are not */
1742         /* very meaningful for cciss.  SG_IO is the main one people want. */
1743
1744         case SG_GET_VERSION_NUM:
1745         case SG_SET_TIMEOUT:
1746         case SG_GET_TIMEOUT:
1747         case SG_GET_RESERVED_SIZE:
1748         case SG_SET_RESERVED_SIZE:
1749         case SG_EMULATED_HOST:
1750         case SG_IO:
1751         case SCSI_IOCTL_SEND_COMMAND:
1752                 return scsi_cmd_blk_ioctl(bdev, mode, cmd, argp);
1753
1754         /* scsi_cmd_blk_ioctl would normally handle these, below, but */
1755         /* they aren't a good fit for cciss, as CD-ROMs are */
1756         /* not supported, and we don't have any bus/target/lun */
1757         /* which we present to the kernel. */
1758
1759         case CDROM_SEND_PACKET:
1760         case CDROMCLOSETRAY:
1761         case CDROMEJECT:
1762         case SCSI_IOCTL_GET_IDLUN:
1763         case SCSI_IOCTL_GET_BUS_NUMBER:
1764         default:
1765                 return -ENOTTY;
1766         }
1767 }
1768
1769 static void cciss_check_queues(ctlr_info_t *h)
1770 {
1771         int start_queue = h->next_to_run;
1772         int i;
1773
1774         /* check to see if we have maxed out the number of commands that can
1775          * be placed on the queue.  If so then exit.  We do this check here
1776          * in case the interrupt we serviced was from an ioctl and did not
1777          * free any new commands.
1778          */
1779         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1780                 return;
1781
1782         /* We have room on the queue for more commands.  Now we need to queue
1783          * them up.  We will also keep track of the next queue to run so
1784          * that every queue gets a chance to be started first.
1785          */
1786         for (i = 0; i < h->highest_lun + 1; i++) {
1787                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1788                 /* make sure the disk has been added and the drive is real
1789                  * because this can be called from the middle of init_one.
1790                  */
1791                 if (!h->drv[curr_queue])
1792                         continue;
1793                 if (!(h->drv[curr_queue]->queue) ||
1794                         !(h->drv[curr_queue]->heads))
1795                         continue;
1796                 blk_start_queue(h->gendisk[curr_queue]->queue);
1797
1798                 /* check to see if we have maxed out the number of commands
1799                  * that can be placed on the queue.
1800                  */
1801                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1802                         if (curr_queue == start_queue) {
1803                                 h->next_to_run =
1804                                     (start_queue + 1) % (h->highest_lun + 1);
1805                                 break;
1806                         } else {
1807                                 h->next_to_run = curr_queue;
1808                                 break;
1809                         }
1810                 }
1811         }
1812 }
1813
1814 static void cciss_softirq_done(struct request *rq)
1815 {
1816         CommandList_struct *c = rq->completion_data;
1817         ctlr_info_t *h = hba[c->ctlr];
1818         SGDescriptor_struct *curr_sg = c->SG;
1819         u64bit temp64;
1820         unsigned long flags;
1821         int i, ddir;
1822         int sg_index = 0;
1823
1824         if (c->Request.Type.Direction == XFER_READ)
1825                 ddir = PCI_DMA_FROMDEVICE;
1826         else
1827                 ddir = PCI_DMA_TODEVICE;
1828
1829         /* command did not need to be retried */
1830         /* unmap the DMA mapping for all the scatter gather elements */
1831         for (i = 0; i < c->Header.SGList; i++) {
1832                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1833                         cciss_unmap_sg_chain_block(h, c);
1834                         /* Point to the next block */
1835                         curr_sg = h->cmd_sg_list[c->cmdindex];
1836                         sg_index = 0;
1837                 }
1838                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1839                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1840                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1841                                 ddir);
1842                 ++sg_index;
1843         }
1844
1845         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1846
1847         /* set the residual count for pc requests */
1848         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1849                 rq->resid_len = c->err_info->ResidualCnt;
1850
1851         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1852
1853         spin_lock_irqsave(&h->lock, flags);
1854         cmd_free(h, c);
1855         cciss_check_queues(h);
1856         spin_unlock_irqrestore(&h->lock, flags);
1857 }
1858
1859 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1860         unsigned char scsi3addr[], uint32_t log_unit)
1861 {
1862         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1863                 sizeof(h->drv[log_unit]->LunID));
1864 }
1865
1866 /* This function gets the SCSI vendor, model, and revision of a logical drive
1867  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1868  * they cannot be read.
1869  */
1870 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1871                                    char *vendor, char *model, char *rev)
1872 {
1873         int rc;
1874         InquiryData_struct *inq_buf;
1875         unsigned char scsi3addr[8];
1876
1877         *vendor = '\0';
1878         *model = '\0';
1879         *rev = '\0';
1880
1881         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1882         if (!inq_buf)
1883                 return;
1884
1885         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1886         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1887                         scsi3addr, TYPE_CMD);
1888         if (rc == IO_OK) {
1889                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1890                 vendor[VENDOR_LEN] = '\0';
1891                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1892                 model[MODEL_LEN] = '\0';
1893                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1894                 rev[REV_LEN] = '\0';
1895         }
1896
1897         kfree(inq_buf);
1898         return;
1899 }
1900
1901 /* This function gets the serial number of a logical drive via
1902  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1903  * number cannot be had, for whatever reason, 16 bytes of 0xff
1904  * are returned instead.
1905  */
1906 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1907                                 unsigned char *serial_no, int buflen)
1908 {
1909 #define PAGE_83_INQ_BYTES 64
1910         int rc;
1911         unsigned char *buf;
1912         unsigned char scsi3addr[8];
1913
1914         if (buflen > 16)
1915                 buflen = 16;
1916         memset(serial_no, 0xff, buflen);
1917         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1918         if (!buf)
1919                 return;
1920         memset(serial_no, 0, buflen);
1921         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1922         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1923                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1924         if (rc == IO_OK)
1925                 memcpy(serial_no, &buf[8], buflen);
1926         kfree(buf);
1927         return;
1928 }
1929
1930 /*
1931  * cciss_add_disk sets up the block device queue for a logical drive
1932  */
1933 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1934                                 int drv_index)
1935 {
1936         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1937         if (!disk->queue)
1938                 goto init_queue_failure;
1939         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1940         disk->major = h->major;
1941         disk->first_minor = drv_index << NWD_SHIFT;
1942         disk->fops = &cciss_fops;
1943         if (cciss_create_ld_sysfs_entry(h, drv_index))
1944                 goto cleanup_queue;
1945         disk->private_data = h->drv[drv_index];
1946         disk->driverfs_dev = &h->drv[drv_index]->dev;
1947
1948         /* Set up queue information */
1949         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1950
1951         /* This is a hardware imposed limit. */
1952         blk_queue_max_segments(disk->queue, h->maxsgentries);
1953
1954         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1955
1956         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1957
1958         disk->queue->queuedata = h;
1959
1960         blk_queue_logical_block_size(disk->queue,
1961                                      h->drv[drv_index]->block_size);
1962
1963         /* Make sure all queue data is written out before */
1964         /* setting h->drv[drv_index]->queue, as setting this */
1965         /* allows the interrupt handler to start the queue */
1966         wmb();
1967         h->drv[drv_index]->queue = disk->queue;
1968         add_disk(disk);
1969         return 0;
1970
1971 cleanup_queue:
1972         blk_cleanup_queue(disk->queue);
1973         disk->queue = NULL;
1974 init_queue_failure:
1975         return -1;
1976 }
1977
1978 /* This function will check the usage_count of the drive to be updated/added.
1979  * If the usage_count is zero and it is a heretofore unknown drive, or,
1980  * the drive's capacity, geometry, or serial number has changed,
1981  * then the drive information will be updated and the disk will be
1982  * re-registered with the kernel.  If these conditions don't hold,
1983  * then it will be left alone for the next reboot.  The exception to this
1984  * is disk 0 which will always be left registered with the kernel since it
1985  * is also the controller node.  Any changes to disk 0 will show up on
1986  * the next reboot.
1987  */
1988 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1989         int first_time, int via_ioctl)
1990 {
1991         struct gendisk *disk;
1992         InquiryData_struct *inq_buff = NULL;
1993         unsigned int block_size;
1994         sector_t total_size;
1995         unsigned long flags = 0;
1996         int ret = 0;
1997         drive_info_struct *drvinfo;
1998
1999         /* Get information about the disk and modify the driver structure */
2000         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2001         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
2002         if (inq_buff == NULL || drvinfo == NULL)
2003                 goto mem_msg;
2004
2005         /* testing to see if 16-byte CDBs are already being used */
2006         if (h->cciss_read == CCISS_READ_16) {
2007                 cciss_read_capacity_16(h, drv_index,
2008                         &total_size, &block_size);
2009
2010         } else {
2011                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
2012                 /* if read_capacity returns all F's this volume is >2TB */
2013                 /* in size so we switch to 16-byte CDB's for all */
2014                 /* read/write ops */
2015                 if (total_size == 0xFFFFFFFFULL) {
2016                         cciss_read_capacity_16(h, drv_index,
2017                         &total_size, &block_size);
2018                         h->cciss_read = CCISS_READ_16;
2019                         h->cciss_write = CCISS_WRITE_16;
2020                 } else {
2021                         h->cciss_read = CCISS_READ_10;
2022                         h->cciss_write = CCISS_WRITE_10;
2023                 }
2024         }
2025
2026         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
2027                                inq_buff, drvinfo);
2028         drvinfo->block_size = block_size;
2029         drvinfo->nr_blocks = total_size + 1;
2030
2031         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
2032                                 drvinfo->model, drvinfo->rev);
2033         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
2034                         sizeof(drvinfo->serial_no));
2035         /* Save the lunid in case we deregister the disk, below. */
2036         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
2037                 sizeof(drvinfo->LunID));
2038
2039         /* Is it the same disk we already know, and nothing's changed? */
2040         if (h->drv[drv_index]->raid_level != -1 &&
2041                 ((memcmp(drvinfo->serial_no,
2042                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2043                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2044                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2045                 drvinfo->heads == h->drv[drv_index]->heads &&
2046                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2047                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2048                         /* The disk is unchanged, nothing to update */
2049                         goto freeret;
2050
2051         /* If we get here it's not the same disk, or something's changed,
2052          * so we need to * deregister it, and re-register it, if it's not
2053          * in use.
2054          * If the disk already exists then deregister it before proceeding
2055          * (unless it's the first disk (for the controller node).
2056          */
2057         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2058                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2059                 spin_lock_irqsave(&h->lock, flags);
2060                 h->drv[drv_index]->busy_configuring = 1;
2061                 spin_unlock_irqrestore(&h->lock, flags);
2062
2063                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2064                  * which keeps the interrupt handler from starting
2065                  * the queue.
2066                  */
2067                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2068         }
2069
2070         /* If the disk is in use return */
2071         if (ret)
2072                 goto freeret;
2073
2074         /* Save the new information from cciss_geometry_inquiry
2075          * and serial number inquiry.  If the disk was deregistered
2076          * above, then h->drv[drv_index] will be NULL.
2077          */
2078         if (h->drv[drv_index] == NULL) {
2079                 drvinfo->device_initialized = 0;
2080                 h->drv[drv_index] = drvinfo;
2081                 drvinfo = NULL; /* so it won't be freed below. */
2082         } else {
2083                 /* special case for cxd0 */
2084                 h->drv[drv_index]->block_size = drvinfo->block_size;
2085                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2086                 h->drv[drv_index]->heads = drvinfo->heads;
2087                 h->drv[drv_index]->sectors = drvinfo->sectors;
2088                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2089                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2090                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2091                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2092                         VENDOR_LEN + 1);
2093                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2094                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2095         }
2096
2097         ++h->num_luns;
2098         disk = h->gendisk[drv_index];
2099         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2100
2101         /* If it's not disk 0 (drv_index != 0)
2102          * or if it was disk 0, but there was previously
2103          * no actual corresponding configured logical drive
2104          * (raid_leve == -1) then we want to update the
2105          * logical drive's information.
2106          */
2107         if (drv_index || first_time) {
2108                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2109                         cciss_free_gendisk(h, drv_index);
2110                         cciss_free_drive_info(h, drv_index);
2111                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2112                                 drv_index);
2113                         --h->num_luns;
2114                 }
2115         }
2116
2117 freeret:
2118         kfree(inq_buff);
2119         kfree(drvinfo);
2120         return;
2121 mem_msg:
2122         dev_err(&h->pdev->dev, "out of memory\n");
2123         goto freeret;
2124 }
2125
2126 /* This function will find the first index of the controllers drive array
2127  * that has a null drv pointer and allocate the drive info struct and
2128  * will return that index   This is where new drives will be added.
2129  * If the index to be returned is greater than the highest_lun index for
2130  * the controller then highest_lun is set * to this new index.
2131  * If there are no available indexes or if tha allocation fails, then -1
2132  * is returned.  * "controller_node" is used to know if this is a real
2133  * logical drive, or just the controller node, which determines if this
2134  * counts towards highest_lun.
2135  */
2136 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2137 {
2138         int i;
2139         drive_info_struct *drv;
2140
2141         /* Search for an empty slot for our drive info */
2142         for (i = 0; i < CISS_MAX_LUN; i++) {
2143
2144                 /* if not cxd0 case, and it's occupied, skip it. */
2145                 if (h->drv[i] && i != 0)
2146                         continue;
2147                 /*
2148                  * If it's cxd0 case, and drv is alloc'ed already, and a
2149                  * disk is configured there, skip it.
2150                  */
2151                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2152                         continue;
2153
2154                 /*
2155                  * We've found an empty slot.  Update highest_lun
2156                  * provided this isn't just the fake cxd0 controller node.
2157                  */
2158                 if (i > h->highest_lun && !controller_node)
2159                         h->highest_lun = i;
2160
2161                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2162                 if (i == 0 && h->drv[i] != NULL)
2163                         return i;
2164
2165                 /*
2166                  * Found an empty slot, not already alloc'ed.  Allocate it.
2167                  * Mark it with raid_level == -1, so we know it's new later on.
2168                  */
2169                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2170                 if (!drv)
2171                         return -1;
2172                 drv->raid_level = -1; /* so we know it's new */
2173                 h->drv[i] = drv;
2174                 return i;
2175         }
2176         return -1;
2177 }
2178
2179 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2180 {
2181         kfree(h->drv[drv_index]);
2182         h->drv[drv_index] = NULL;
2183 }
2184
2185 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2186 {
2187         put_disk(h->gendisk[drv_index]);
2188         h->gendisk[drv_index] = NULL;
2189 }
2190
2191 /* cciss_add_gendisk finds a free hba[]->drv structure
2192  * and allocates a gendisk if needed, and sets the lunid
2193  * in the drvinfo structure.   It returns the index into
2194  * the ->drv[] array, or -1 if none are free.
2195  * is_controller_node indicates whether highest_lun should
2196  * count this disk, or if it's only being added to provide
2197  * a means to talk to the controller in case no logical
2198  * drives have yet been configured.
2199  */
2200 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2201         int controller_node)
2202 {
2203         int drv_index;
2204
2205         drv_index = cciss_alloc_drive_info(h, controller_node);
2206         if (drv_index == -1)
2207                 return -1;
2208
2209         /*Check if the gendisk needs to be allocated */
2210         if (!h->gendisk[drv_index]) {
2211                 h->gendisk[drv_index] =
2212                         alloc_disk(1 << NWD_SHIFT);
2213                 if (!h->gendisk[drv_index]) {
2214                         dev_err(&h->pdev->dev,
2215                                 "could not allocate a new disk %d\n",
2216                                 drv_index);
2217                         goto err_free_drive_info;
2218                 }
2219         }
2220         memcpy(h->drv[drv_index]->LunID, lunid,
2221                 sizeof(h->drv[drv_index]->LunID));
2222         if (cciss_create_ld_sysfs_entry(h, drv_index))
2223                 goto err_free_disk;
2224         /* Don't need to mark this busy because nobody */
2225         /* else knows about this disk yet to contend */
2226         /* for access to it. */
2227         h->drv[drv_index]->busy_configuring = 0;
2228         wmb();
2229         return drv_index;
2230
2231 err_free_disk:
2232         cciss_free_gendisk(h, drv_index);
2233 err_free_drive_info:
2234         cciss_free_drive_info(h, drv_index);
2235         return -1;
2236 }
2237
2238 /* This is for the special case of a controller which
2239  * has no logical drives.  In this case, we still need
2240  * to register a disk so the controller can be accessed
2241  * by the Array Config Utility.
2242  */
2243 static void cciss_add_controller_node(ctlr_info_t *h)
2244 {
2245         struct gendisk *disk;
2246         int drv_index;
2247
2248         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2249                 return;
2250
2251         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2252         if (drv_index == -1)
2253                 goto error;
2254         h->drv[drv_index]->block_size = 512;
2255         h->drv[drv_index]->nr_blocks = 0;
2256         h->drv[drv_index]->heads = 0;
2257         h->drv[drv_index]->sectors = 0;
2258         h->drv[drv_index]->cylinders = 0;
2259         h->drv[drv_index]->raid_level = -1;
2260         memset(h->drv[drv_index]->serial_no, 0, 16);
2261         disk = h->gendisk[drv_index];
2262         if (cciss_add_disk(h, disk, drv_index) == 0)
2263                 return;
2264         cciss_free_gendisk(h, drv_index);
2265         cciss_free_drive_info(h, drv_index);
2266 error:
2267         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2268         return;
2269 }
2270
2271 /* This function will add and remove logical drives from the Logical
2272  * drive array of the controller and maintain persistency of ordering
2273  * so that mount points are preserved until the next reboot.  This allows
2274  * for the removal of logical drives in the middle of the drive array
2275  * without a re-ordering of those drives.
2276  * INPUT
2277  * h            = The controller to perform the operations on
2278  */
2279 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2280         int via_ioctl)
2281 {
2282         int num_luns;
2283         ReportLunData_struct *ld_buff = NULL;
2284         int return_code;
2285         int listlength = 0;
2286         int i;
2287         int drv_found;
2288         int drv_index = 0;
2289         unsigned char lunid[8] = CTLR_LUNID;
2290         unsigned long flags;
2291
2292         if (!capable(CAP_SYS_RAWIO))
2293                 return -EPERM;
2294
2295         /* Set busy_configuring flag for this operation */
2296         spin_lock_irqsave(&h->lock, flags);
2297         if (h->busy_configuring) {
2298                 spin_unlock_irqrestore(&h->lock, flags);
2299                 return -EBUSY;
2300         }
2301         h->busy_configuring = 1;
2302         spin_unlock_irqrestore(&h->lock, flags);
2303
2304         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2305         if (ld_buff == NULL)
2306                 goto mem_msg;
2307
2308         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2309                                       sizeof(ReportLunData_struct),
2310                                       0, CTLR_LUNID, TYPE_CMD);
2311
2312         if (return_code == IO_OK)
2313                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2314         else {  /* reading number of logical volumes failed */
2315                 dev_warn(&h->pdev->dev,
2316                         "report logical volume command failed\n");
2317                 listlength = 0;
2318                 goto freeret;
2319         }
2320
2321         num_luns = listlength / 8;      /* 8 bytes per entry */
2322         if (num_luns > CISS_MAX_LUN) {
2323                 num_luns = CISS_MAX_LUN;
2324                 dev_warn(&h->pdev->dev, "more luns configured"
2325                        " on controller than can be handled by"
2326                        " this driver.\n");
2327         }
2328
2329         if (num_luns == 0)
2330                 cciss_add_controller_node(h);
2331
2332         /* Compare controller drive array to driver's drive array
2333          * to see if any drives are missing on the controller due
2334          * to action of Array Config Utility (user deletes drive)
2335          * and deregister logical drives which have disappeared.
2336          */
2337         for (i = 0; i <= h->highest_lun; i++) {
2338                 int j;
2339                 drv_found = 0;
2340
2341                 /* skip holes in the array from already deleted drives */
2342                 if (h->drv[i] == NULL)
2343                         continue;
2344
2345                 for (j = 0; j < num_luns; j++) {
2346                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2347                         if (memcmp(h->drv[i]->LunID, lunid,
2348                                 sizeof(lunid)) == 0) {
2349                                 drv_found = 1;
2350                                 break;
2351                         }
2352                 }
2353                 if (!drv_found) {
2354                         /* Deregister it from the OS, it's gone. */
2355                         spin_lock_irqsave(&h->lock, flags);
2356                         h->drv[i]->busy_configuring = 1;
2357                         spin_unlock_irqrestore(&h->lock, flags);
2358                         return_code = deregister_disk(h, i, 1, via_ioctl);
2359                         if (h->drv[i] != NULL)
2360                                 h->drv[i]->busy_configuring = 0;
2361                 }
2362         }
2363
2364         /* Compare controller drive array to driver's drive array.
2365          * Check for updates in the drive information and any new drives
2366          * on the controller due to ACU adding logical drives, or changing
2367          * a logical drive's size, etc.  Reregister any new/changed drives
2368          */
2369         for (i = 0; i < num_luns; i++) {
2370                 int j;
2371
2372                 drv_found = 0;
2373
2374                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2375                 /* Find if the LUN is already in the drive array
2376                  * of the driver.  If so then update its info
2377                  * if not in use.  If it does not exist then find
2378                  * the first free index and add it.
2379                  */
2380                 for (j = 0; j <= h->highest_lun; j++) {
2381                         if (h->drv[j] != NULL &&
2382                                 memcmp(h->drv[j]->LunID, lunid,
2383                                         sizeof(h->drv[j]->LunID)) == 0) {
2384                                 drv_index = j;
2385                                 drv_found = 1;
2386                                 break;
2387                         }
2388                 }
2389
2390                 /* check if the drive was found already in the array */
2391                 if (!drv_found) {
2392                         drv_index = cciss_add_gendisk(h, lunid, 0);
2393                         if (drv_index == -1)
2394                                 goto freeret;
2395                 }
2396                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2397         }               /* end for */
2398
2399 freeret:
2400         kfree(ld_buff);
2401         h->busy_configuring = 0;
2402         /* We return -1 here to tell the ACU that we have registered/updated
2403          * all of the drives that we can and to keep it from calling us
2404          * additional times.
2405          */
2406         return -1;
2407 mem_msg:
2408         dev_err(&h->pdev->dev, "out of memory\n");
2409         h->busy_configuring = 0;
2410         goto freeret;
2411 }
2412
2413 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2414 {
2415         /* zero out the disk size info */
2416         drive_info->nr_blocks = 0;
2417         drive_info->block_size = 0;
2418         drive_info->heads = 0;
2419         drive_info->sectors = 0;
2420         drive_info->cylinders = 0;
2421         drive_info->raid_level = -1;
2422         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2423         memset(drive_info->model, 0, sizeof(drive_info->model));
2424         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2425         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2426         /*
2427          * don't clear the LUNID though, we need to remember which
2428          * one this one is.
2429          */
2430 }
2431
2432 /* This function will deregister the disk and it's queue from the
2433  * kernel.  It must be called with the controller lock held and the
2434  * drv structures busy_configuring flag set.  It's parameters are:
2435  *
2436  * disk = This is the disk to be deregistered
2437  * drv  = This is the drive_info_struct associated with the disk to be
2438  *        deregistered.  It contains information about the disk used
2439  *        by the driver.
2440  * clear_all = This flag determines whether or not the disk information
2441  *             is going to be completely cleared out and the highest_lun
2442  *             reset.  Sometimes we want to clear out information about
2443  *             the disk in preparation for re-adding it.  In this case
2444  *             the highest_lun should be left unchanged and the LunID
2445  *             should not be cleared.
2446  * via_ioctl
2447  *    This indicates whether we've reached this path via ioctl.
2448  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2449  *    If this path is reached via ioctl(), then the max_usage_count will
2450  *    be 1, as the process calling ioctl() has got to have the device open.
2451  *    If we get here via sysfs, then the max usage count will be zero.
2452 */
2453 static int deregister_disk(ctlr_info_t *h, int drv_index,
2454                            int clear_all, int via_ioctl)
2455 {
2456         int i;
2457         struct gendisk *disk;
2458         drive_info_struct *drv;
2459         int recalculate_highest_lun;
2460
2461         if (!capable(CAP_SYS_RAWIO))
2462                 return -EPERM;
2463
2464         drv = h->drv[drv_index];
2465         disk = h->gendisk[drv_index];
2466
2467         /* make sure logical volume is NOT is use */
2468         if (clear_all || (h->gendisk[0] == disk)) {
2469                 if (drv->usage_count > via_ioctl)
2470                         return -EBUSY;
2471         } else if (drv->usage_count > 0)
2472                 return -EBUSY;
2473
2474         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2475
2476         /* invalidate the devices and deregister the disk.  If it is disk
2477          * zero do not deregister it but just zero out it's values.  This
2478          * allows us to delete disk zero but keep the controller registered.
2479          */
2480         if (h->gendisk[0] != disk) {
2481                 struct request_queue *q = disk->queue;
2482                 if (disk->flags & GENHD_FL_UP) {
2483                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2484                         del_gendisk(disk);
2485                 }
2486                 if (q)
2487                         blk_cleanup_queue(q);
2488                 /* If clear_all is set then we are deleting the logical
2489                  * drive, not just refreshing its info.  For drives
2490                  * other than disk 0 we will call put_disk.  We do not
2491                  * do this for disk 0 as we need it to be able to
2492                  * configure the controller.
2493                  */
2494                 if (clear_all){
2495                         /* This isn't pretty, but we need to find the
2496                          * disk in our array and NULL our the pointer.
2497                          * This is so that we will call alloc_disk if
2498                          * this index is used again later.
2499                          */
2500                         for (i=0; i < CISS_MAX_LUN; i++){
2501                                 if (h->gendisk[i] == disk) {
2502                                         h->gendisk[i] = NULL;
2503                                         break;
2504                                 }
2505                         }
2506                         put_disk(disk);
2507                 }
2508         } else {
2509                 set_capacity(disk, 0);
2510                 cciss_clear_drive_info(drv);
2511         }
2512
2513         --h->num_luns;
2514
2515         /* if it was the last disk, find the new hightest lun */
2516         if (clear_all && recalculate_highest_lun) {
2517                 int newhighest = -1;
2518                 for (i = 0; i <= h->highest_lun; i++) {
2519                         /* if the disk has size > 0, it is available */
2520                         if (h->drv[i] && h->drv[i]->heads)
2521                                 newhighest = i;
2522                 }
2523                 h->highest_lun = newhighest;
2524         }
2525         return 0;
2526 }
2527
2528 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2529                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2530                 int cmd_type)
2531 {
2532         u64bit buff_dma_handle;
2533         int status = IO_OK;
2534
2535         c->cmd_type = CMD_IOCTL_PEND;
2536         c->Header.ReplyQueue = 0;
2537         if (buff != NULL) {
2538                 c->Header.SGList = 1;
2539                 c->Header.SGTotal = 1;
2540         } else {
2541                 c->Header.SGList = 0;
2542                 c->Header.SGTotal = 0;
2543         }
2544         c->Header.Tag.lower = c->busaddr;
2545         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2546
2547         c->Request.Type.Type = cmd_type;
2548         if (cmd_type == TYPE_CMD) {
2549                 switch (cmd) {
2550                 case CISS_INQUIRY:
2551                         /* are we trying to read a vital product page */
2552                         if (page_code != 0) {
2553                                 c->Request.CDB[1] = 0x01;
2554                                 c->Request.CDB[2] = page_code;
2555                         }
2556                         c->Request.CDBLen = 6;
2557                         c->Request.Type.Attribute = ATTR_SIMPLE;
2558                         c->Request.Type.Direction = XFER_READ;
2559                         c->Request.Timeout = 0;
2560                         c->Request.CDB[0] = CISS_INQUIRY;
2561                         c->Request.CDB[4] = size & 0xFF;
2562                         break;
2563                 case CISS_REPORT_LOG:
2564                 case CISS_REPORT_PHYS:
2565                         /* Talking to controller so It's a physical command
2566                            mode = 00 target = 0.  Nothing to write.
2567                          */
2568                         c->Request.CDBLen = 12;
2569                         c->Request.Type.Attribute = ATTR_SIMPLE;
2570                         c->Request.Type.Direction = XFER_READ;
2571                         c->Request.Timeout = 0;
2572                         c->Request.CDB[0] = cmd;
2573                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2574                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2575                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2576                         c->Request.CDB[9] = size & 0xFF;
2577                         break;
2578
2579                 case CCISS_READ_CAPACITY:
2580                         c->Request.CDBLen = 10;
2581                         c->Request.Type.Attribute = ATTR_SIMPLE;
2582                         c->Request.Type.Direction = XFER_READ;
2583                         c->Request.Timeout = 0;
2584                         c->Request.CDB[0] = cmd;
2585                         break;
2586                 case CCISS_READ_CAPACITY_16:
2587                         c->Request.CDBLen = 16;
2588                         c->Request.Type.Attribute = ATTR_SIMPLE;
2589                         c->Request.Type.Direction = XFER_READ;
2590                         c->Request.Timeout = 0;
2591                         c->Request.CDB[0] = cmd;
2592                         c->Request.CDB[1] = 0x10;
2593                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2594                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2595                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2596                         c->Request.CDB[13] = size & 0xFF;
2597                         c->Request.Timeout = 0;
2598                         c->Request.CDB[0] = cmd;
2599                         break;
2600                 case CCISS_CACHE_FLUSH:
2601                         c->Request.CDBLen = 12;
2602                         c->Request.Type.Attribute = ATTR_SIMPLE;
2603                         c->Request.Type.Direction = XFER_WRITE;
2604                         c->Request.Timeout = 0;
2605                         c->Request.CDB[0] = BMIC_WRITE;
2606                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2607                         c->Request.CDB[7] = (size >> 8) & 0xFF;
2608                         c->Request.CDB[8] = size & 0xFF;
2609                         break;
2610                 case TEST_UNIT_READY:
2611                         c->Request.CDBLen = 6;
2612                         c->Request.Type.Attribute = ATTR_SIMPLE;
2613                         c->Request.Type.Direction = XFER_NONE;
2614                         c->Request.Timeout = 0;
2615                         break;
2616                 default:
2617                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2618                         return IO_ERROR;
2619                 }
2620         } else if (cmd_type == TYPE_MSG) {
2621                 switch (cmd) {
2622                 case CCISS_ABORT_MSG:
2623                         c->Request.CDBLen = 12;
2624                         c->Request.Type.Attribute = ATTR_SIMPLE;
2625                         c->Request.Type.Direction = XFER_WRITE;
2626                         c->Request.Timeout = 0;
2627                         c->Request.CDB[0] = cmd;        /* abort */
2628                         c->Request.CDB[1] = 0;  /* abort a command */
2629                         /* buff contains the tag of the command to abort */
2630                         memcpy(&c->Request.CDB[4], buff, 8);
2631                         break;
2632                 case CCISS_RESET_MSG:
2633                         c->Request.CDBLen = 16;
2634                         c->Request.Type.Attribute = ATTR_SIMPLE;
2635                         c->Request.Type.Direction = XFER_NONE;
2636                         c->Request.Timeout = 0;
2637                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2638                         c->Request.CDB[0] = cmd;        /* reset */
2639                         c->Request.CDB[1] = CCISS_RESET_TYPE_TARGET;
2640                         break;
2641                 case CCISS_NOOP_MSG:
2642                         c->Request.CDBLen = 1;
2643                         c->Request.Type.Attribute = ATTR_SIMPLE;
2644                         c->Request.Type.Direction = XFER_WRITE;
2645                         c->Request.Timeout = 0;
2646                         c->Request.CDB[0] = cmd;
2647                         break;
2648                 default:
2649                         dev_warn(&h->pdev->dev,
2650                                 "unknown message type %d\n", cmd);
2651                         return IO_ERROR;
2652                 }
2653         } else {
2654                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2655                 return IO_ERROR;
2656         }
2657         /* Fill in the scatter gather information */
2658         if (size > 0) {
2659                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2660                                                              buff, size,
2661                                                              PCI_DMA_BIDIRECTIONAL);
2662                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2663                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2664                 c->SG[0].Len = size;
2665                 c->SG[0].Ext = 0;       /* we are not chaining */
2666         }
2667         return status;
2668 }
2669
2670 static int cciss_send_reset(ctlr_info_t *h, unsigned char *scsi3addr,
2671                             u8 reset_type)
2672 {
2673         CommandList_struct *c;
2674         int return_status;
2675
2676         c = cmd_alloc(h);
2677         if (!c)
2678                 return -ENOMEM;
2679         return_status = fill_cmd(h, c, CCISS_RESET_MSG, NULL, 0, 0,
2680                 CTLR_LUNID, TYPE_MSG);
2681         c->Request.CDB[1] = reset_type; /* fill_cmd defaults to target reset */
2682         if (return_status != IO_OK) {
2683                 cmd_special_free(h, c);
2684                 return return_status;
2685         }
2686         c->waiting = NULL;
2687         enqueue_cmd_and_start_io(h, c);
2688         /* Don't wait for completion, the reset won't complete.  Don't free
2689          * the command either.  This is the last command we will send before
2690          * re-initializing everything, so it doesn't matter and won't leak.
2691          */
2692         return 0;
2693 }
2694
2695 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2696 {
2697         switch (c->err_info->ScsiStatus) {
2698         case SAM_STAT_GOOD:
2699                 return IO_OK;
2700         case SAM_STAT_CHECK_CONDITION:
2701                 switch (0xf & c->err_info->SenseInfo[2]) {
2702                 case 0: return IO_OK; /* no sense */
2703                 case 1: return IO_OK; /* recovered error */
2704                 default:
2705                         if (check_for_unit_attention(h, c))
2706                                 return IO_NEEDS_RETRY;
2707                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2708                                 "check condition, sense key = 0x%02x\n",
2709                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2710                 }
2711                 break;
2712         default:
2713                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2714                         "scsi status = 0x%02x\n",
2715                         c->Request.CDB[0], c->err_info->ScsiStatus);
2716                 break;
2717         }
2718         return IO_ERROR;
2719 }
2720
2721 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2722 {
2723         int return_status = IO_OK;
2724
2725         if (c->err_info->CommandStatus == CMD_SUCCESS)
2726                 return IO_OK;
2727
2728         switch (c->err_info->CommandStatus) {
2729         case CMD_TARGET_STATUS:
2730                 return_status = check_target_status(h, c);
2731                 break;
2732         case CMD_DATA_UNDERRUN:
2733         case CMD_DATA_OVERRUN:
2734                 /* expected for inquiry and report lun commands */
2735                 break;
2736         case CMD_INVALID:
2737                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2738                        "reported invalid\n", c->Request.CDB[0]);
2739                 return_status = IO_ERROR;
2740                 break;
2741         case CMD_PROTOCOL_ERR:
2742                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2743                        "protocol error\n", c->Request.CDB[0]);
2744                 return_status = IO_ERROR;
2745                 break;
2746         case CMD_HARDWARE_ERR:
2747                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2748                        " hardware error\n", c->Request.CDB[0]);
2749                 return_status = IO_ERROR;
2750                 break;
2751         case CMD_CONNECTION_LOST:
2752                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2753                        "connection lost\n", c->Request.CDB[0]);
2754                 return_status = IO_ERROR;
2755                 break;
2756         case CMD_ABORTED:
2757                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2758                        "aborted\n", c->Request.CDB[0]);
2759                 return_status = IO_ERROR;
2760                 break;
2761         case CMD_ABORT_FAILED:
2762                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2763                        "abort failed\n", c->Request.CDB[0]);
2764                 return_status = IO_ERROR;
2765                 break;
2766         case CMD_UNSOLICITED_ABORT:
2767                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2768                         c->Request.CDB[0]);
2769                 return_status = IO_NEEDS_RETRY;
2770                 break;
2771         case CMD_UNABORTABLE:
2772                 dev_warn(&h->pdev->dev, "cmd unabortable\n");
2773                 return_status = IO_ERROR;
2774                 break;
2775         default:
2776                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2777                        "unknown status %x\n", c->Request.CDB[0],
2778                        c->err_info->CommandStatus);
2779                 return_status = IO_ERROR;
2780         }
2781         return return_status;
2782 }
2783
2784 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2785         int attempt_retry)
2786 {
2787         DECLARE_COMPLETION_ONSTACK(wait);
2788         u64bit buff_dma_handle;
2789         int return_status = IO_OK;
2790
2791 resend_cmd2:
2792         c->waiting = &wait;
2793         enqueue_cmd_and_start_io(h, c);
2794
2795         wait_for_completion(&wait);
2796
2797         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2798                 goto command_done;
2799
2800         return_status = process_sendcmd_error(h, c);
2801
2802         if (return_status == IO_NEEDS_RETRY &&
2803                 c->retry_count < MAX_CMD_RETRIES) {
2804                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2805                         c->Request.CDB[0]);
2806                 c->retry_count++;
2807                 /* erase the old error information */
2808                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2809                 return_status = IO_OK;
2810                 INIT_COMPLETION(wait);
2811                 goto resend_cmd2;
2812         }
2813
2814 command_done:
2815         /* unlock the buffers from DMA */
2816         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2817         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2818         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2819                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2820         return return_status;
2821 }
2822
2823 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2824                            __u8 page_code, unsigned char scsi3addr[],
2825                         int cmd_type)
2826 {
2827         CommandList_struct *c;
2828         int return_status;
2829
2830         c = cmd_special_alloc(h);
2831         if (!c)
2832                 return -ENOMEM;
2833         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2834                 scsi3addr, cmd_type);
2835         if (return_status == IO_OK)
2836                 return_status = sendcmd_withirq_core(h, c, 1);
2837
2838         cmd_special_free(h, c);
2839         return return_status;
2840 }
2841
2842 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2843                                    sector_t total_size,
2844                                    unsigned int block_size,
2845                                    InquiryData_struct *inq_buff,
2846                                    drive_info_struct *drv)
2847 {
2848         int return_code;
2849         unsigned long t;
2850         unsigned char scsi3addr[8];
2851
2852         memset(inq_buff, 0, sizeof(InquiryData_struct));
2853         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2854         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2855                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2856         if (return_code == IO_OK) {
2857                 if (inq_buff->data_byte[8] == 0xFF) {
2858                         dev_warn(&h->pdev->dev,
2859                                "reading geometry failed, volume "
2860                                "does not support reading geometry\n");
2861                         drv->heads = 255;
2862                         drv->sectors = 32;      /* Sectors per track */
2863                         drv->cylinders = total_size + 1;
2864                         drv->raid_level = RAID_UNKNOWN;
2865                 } else {
2866                         drv->heads = inq_buff->data_byte[6];
2867                         drv->sectors = inq_buff->data_byte[7];
2868                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2869                         drv->cylinders += inq_buff->data_byte[5];
2870                         drv->raid_level = inq_buff->data_byte[8];
2871                 }
2872                 drv->block_size = block_size;
2873                 drv->nr_blocks = total_size + 1;
2874                 t = drv->heads * drv->sectors;
2875                 if (t > 1) {
2876                         sector_t real_size = total_size + 1;
2877                         unsigned long rem = sector_div(real_size, t);
2878                         if (rem)
2879                                 real_size++;
2880                         drv->cylinders = real_size;
2881                 }
2882         } else {                /* Get geometry failed */
2883                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2884         }
2885 }
2886
2887 static void
2888 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2889                     unsigned int *block_size)
2890 {
2891         ReadCapdata_struct *buf;
2892         int return_code;
2893         unsigned char scsi3addr[8];
2894
2895         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2896         if (!buf) {
2897                 dev_warn(&h->pdev->dev, "out of memory\n");
2898                 return;
2899         }
2900
2901         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2902         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2903                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2904         if (return_code == IO_OK) {
2905                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2906                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2907         } else {                /* read capacity command failed */
2908                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2909                 *total_size = 0;
2910                 *block_size = BLOCK_SIZE;
2911         }
2912         kfree(buf);
2913 }
2914
2915 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2916         sector_t *total_size, unsigned int *block_size)
2917 {
2918         ReadCapdata_struct_16 *buf;
2919         int return_code;
2920         unsigned char scsi3addr[8];
2921
2922         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2923         if (!buf) {
2924                 dev_warn(&h->pdev->dev, "out of memory\n");
2925                 return;
2926         }
2927
2928         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2929         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2930                 buf, sizeof(ReadCapdata_struct_16),
2931                         0, scsi3addr, TYPE_CMD);
2932         if (return_code == IO_OK) {
2933                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2934                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2935         } else {                /* read capacity command failed */
2936                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2937                 *total_size = 0;
2938                 *block_size = BLOCK_SIZE;
2939         }
2940         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2941                (unsigned long long)*total_size+1, *block_size);
2942         kfree(buf);
2943 }
2944
2945 static int cciss_revalidate(struct gendisk *disk)
2946 {
2947         ctlr_info_t *h = get_host(disk);
2948         drive_info_struct *drv = get_drv(disk);
2949         int logvol;
2950         int FOUND = 0;
2951         unsigned int block_size;
2952         sector_t total_size;
2953         InquiryData_struct *inq_buff = NULL;
2954
2955         for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2956                 if (!h->drv[logvol])
2957                         continue;
2958                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2959                         sizeof(drv->LunID)) == 0) {
2960                         FOUND = 1;
2961                         break;
2962                 }
2963         }
2964
2965         if (!FOUND)
2966                 return 1;
2967
2968         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2969         if (inq_buff == NULL) {
2970                 dev_warn(&h->pdev->dev, "out of memory\n");
2971                 return 1;
2972         }
2973         if (h->cciss_read == CCISS_READ_10) {
2974                 cciss_read_capacity(h, logvol,
2975                                         &total_size, &block_size);
2976         } else {
2977                 cciss_read_capacity_16(h, logvol,
2978                                         &total_size, &block_size);
2979         }
2980         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2981                                inq_buff, drv);
2982
2983         blk_queue_logical_block_size(drv->queue, drv->block_size);
2984         set_capacity(disk, drv->nr_blocks);
2985
2986         kfree(inq_buff);
2987         return 0;
2988 }
2989
2990 /*
2991  * Map (physical) PCI mem into (virtual) kernel space
2992  */
2993 static void __iomem *remap_pci_mem(ulong base, ulong size)
2994 {
2995         ulong page_base = ((ulong) base) & PAGE_MASK;
2996         ulong page_offs = ((ulong) base) - page_base;
2997         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2998
2999         return page_remapped ? (page_remapped + page_offs) : NULL;
3000 }
3001
3002 /*
3003  * Takes jobs of the Q and sends them to the hardware, then puts it on
3004  * the Q to wait for completion.
3005  */
3006 static void start_io(ctlr_info_t *h)
3007 {
3008         CommandList_struct *c;
3009
3010         while (!list_empty(&h->reqQ)) {
3011                 c = list_entry(h->reqQ.next, CommandList_struct, list);
3012                 /* can't do anything if fifo is full */
3013                 if ((h->access.fifo_full(h))) {
3014                         dev_warn(&h->pdev->dev, "fifo full\n");
3015                         break;
3016                 }
3017
3018                 /* Get the first entry from the Request Q */
3019                 removeQ(c);
3020                 h->Qdepth--;
3021
3022                 /* Tell the controller execute command */
3023                 h->access.submit_command(h, c);
3024
3025                 /* Put job onto the completed Q */
3026                 addQ(&h->cmpQ, c);
3027         }
3028 }
3029
3030 /* Assumes that h->lock is held. */
3031 /* Zeros out the error record and then resends the command back */
3032 /* to the controller */
3033 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
3034 {
3035         /* erase the old error information */
3036         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
3037
3038         /* add it to software queue and then send it to the controller */
3039         addQ(&h->reqQ, c);
3040         h->Qdepth++;
3041         if (h->Qdepth > h->maxQsinceinit)
3042                 h->maxQsinceinit = h->Qdepth;
3043
3044         start_io(h);
3045 }
3046
3047 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
3048         unsigned int msg_byte, unsigned int host_byte,
3049         unsigned int driver_byte)
3050 {
3051         /* inverse of macros in scsi.h */
3052         return (scsi_status_byte & 0xff) |
3053                 ((msg_byte & 0xff) << 8) |
3054                 ((host_byte & 0xff) << 16) |
3055                 ((driver_byte & 0xff) << 24);
3056 }
3057
3058 static inline int evaluate_target_status(ctlr_info_t *h,
3059                         CommandList_struct *cmd, int *retry_cmd)
3060 {
3061         unsigned char sense_key;
3062         unsigned char status_byte, msg_byte, host_byte, driver_byte;
3063         int error_value;
3064
3065         *retry_cmd = 0;
3066         /* If we get in here, it means we got "target status", that is, scsi status */
3067         status_byte = cmd->err_info->ScsiStatus;
3068         driver_byte = DRIVER_OK;
3069         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3070
3071         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3072                 host_byte = DID_PASSTHROUGH;
3073         else
3074                 host_byte = DID_OK;
3075
3076         error_value = make_status_bytes(status_byte, msg_byte,
3077                 host_byte, driver_byte);
3078
3079         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3080                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3081                         dev_warn(&h->pdev->dev, "cmd %p "
3082                                "has SCSI Status 0x%x\n",
3083                                cmd, cmd->err_info->ScsiStatus);
3084                 return error_value;
3085         }
3086
3087         /* check the sense key */
3088         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3089         /* no status or recovered error */
3090         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3091             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3092                 error_value = 0;
3093
3094         if (check_for_unit_attention(h, cmd)) {
3095                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3096                 return 0;
3097         }
3098
3099         /* Not SG_IO or similar? */
3100         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3101                 if (error_value != 0)
3102                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3103                                " sense key = 0x%x\n", cmd, sense_key);
3104                 return error_value;
3105         }
3106
3107         /* SG_IO or similar, copy sense data back */
3108         if (cmd->rq->sense) {
3109                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3110                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3111                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3112                         cmd->rq->sense_len);
3113         } else
3114                 cmd->rq->sense_len = 0;
3115
3116         return error_value;
3117 }
3118
3119 /* checks the status of the job and calls complete buffers to mark all
3120  * buffers for the completed job. Note that this function does not need
3121  * to hold the hba/queue lock.
3122  */
3123 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3124                                     int timeout)
3125 {
3126         int retry_cmd = 0;
3127         struct request *rq = cmd->rq;
3128
3129         rq->errors = 0;
3130
3131         if (timeout)
3132                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3133
3134         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3135                 goto after_error_processing;
3136
3137         switch (cmd->err_info->CommandStatus) {
3138         case CMD_TARGET_STATUS:
3139                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3140                 break;
3141         case CMD_DATA_UNDERRUN:
3142                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3143                         dev_warn(&h->pdev->dev, "cmd %p has"
3144                                " completed with data underrun "
3145                                "reported\n", cmd);
3146                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3147                 }
3148                 break;
3149         case CMD_DATA_OVERRUN:
3150                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3151                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3152                                " completed with data overrun "
3153                                "reported\n", cmd);
3154                 break;
3155         case CMD_INVALID:
3156                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3157                        "reported invalid\n", cmd);
3158                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3159                         cmd->err_info->CommandStatus, DRIVER_OK,
3160                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3161                                 DID_PASSTHROUGH : DID_ERROR);
3162                 break;
3163         case CMD_PROTOCOL_ERR:
3164                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3165                        "protocol error\n", cmd);
3166                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3167                         cmd->err_info->CommandStatus, DRIVER_OK,
3168                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3169                                 DID_PASSTHROUGH : DID_ERROR);
3170                 break;
3171         case CMD_HARDWARE_ERR:
3172                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3173                        " hardware error\n", cmd);
3174                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3175                         cmd->err_info->CommandStatus, DRIVER_OK,
3176                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3177                                 DID_PASSTHROUGH : DID_ERROR);
3178                 break;
3179         case CMD_CONNECTION_LOST:
3180                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3181                        "connection lost\n", cmd);
3182                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3183                         cmd->err_info->CommandStatus, DRIVER_OK,
3184                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3185                                 DID_PASSTHROUGH : DID_ERROR);
3186                 break;
3187         case CMD_ABORTED:
3188                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3189                        "aborted\n", cmd);
3190                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3191                         cmd->err_info->CommandStatus, DRIVER_OK,
3192                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3193                                 DID_PASSTHROUGH : DID_ABORT);
3194                 break;
3195         case CMD_ABORT_FAILED:
3196                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3197                        "abort failed\n", cmd);
3198                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3199                         cmd->err_info->CommandStatus, DRIVER_OK,
3200                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3201                                 DID_PASSTHROUGH : DID_ERROR);
3202                 break;
3203         case CMD_UNSOLICITED_ABORT:
3204                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3205                        "abort %p\n", h->ctlr, cmd);
3206                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3207                         retry_cmd = 1;
3208                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3209                         cmd->retry_count++;
3210                 } else
3211                         dev_warn(&h->pdev->dev,
3212                                 "%p retried too many times\n", cmd);
3213                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3214                         cmd->err_info->CommandStatus, DRIVER_OK,
3215                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3216                                 DID_PASSTHROUGH : DID_ABORT);
3217                 break;
3218         case CMD_TIMEOUT:
3219                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3220                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3221                         cmd->err_info->CommandStatus, DRIVER_OK,
3222                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3223                                 DID_PASSTHROUGH : DID_ERROR);
3224                 break;
3225         case CMD_UNABORTABLE:
3226                 dev_warn(&h->pdev->dev, "cmd %p unabortable\n", cmd);
3227                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3228                         cmd->err_info->CommandStatus, DRIVER_OK,
3229                         cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC ?
3230                                 DID_PASSTHROUGH : DID_ERROR);
3231                 break;
3232         default:
3233                 dev_warn(&h->pdev->dev, "cmd %p returned "
3234                        "unknown status %x\n", cmd,
3235                        cmd->err_info->CommandStatus);
3236                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3237                         cmd->err_info->CommandStatus, DRIVER_OK,
3238                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3239                                 DID_PASSTHROUGH : DID_ERROR);
3240         }
3241
3242 after_error_processing:
3243
3244         /* We need to return this command */
3245         if (retry_cmd) {
3246                 resend_cciss_cmd(h, cmd);
3247                 return;
3248         }
3249         cmd->rq->completion_data = cmd;
3250         blk_complete_request(cmd->rq);
3251 }
3252
3253 static inline u32 cciss_tag_contains_index(u32 tag)
3254 {
3255 #define DIRECT_LOOKUP_BIT 0x10
3256         return tag & DIRECT_LOOKUP_BIT;
3257 }
3258
3259 static inline u32 cciss_tag_to_index(u32 tag)
3260 {
3261 #define DIRECT_LOOKUP_SHIFT 5
3262         return tag >> DIRECT_LOOKUP_SHIFT;
3263 }
3264
3265 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag)
3266 {
3267 #define CCISS_PERF_ERROR_BITS ((1 << DIRECT_LOOKUP_SHIFT) - 1)
3268 #define CCISS_SIMPLE_ERROR_BITS 0x03
3269         if (likely(h->transMethod & CFGTBL_Trans_Performant))
3270                 return tag & ~CCISS_PERF_ERROR_BITS;
3271         return tag & ~CCISS_SIMPLE_ERROR_BITS;
3272 }
3273
3274 static inline void cciss_mark_tag_indexed(u32 *tag)
3275 {
3276         *tag |= DIRECT_LOOKUP_BIT;
3277 }
3278
3279 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3280 {
3281         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3282 }
3283
3284 /*
3285  * Get a request and submit it to the controller.
3286  */
3287 static void do_cciss_request(struct request_queue *q)
3288 {
3289         ctlr_info_t *h = q->queuedata;
3290         CommandList_struct *c;
3291         sector_t start_blk;
3292         int seg;
3293         struct request *creq;
3294         u64bit temp64;
3295         struct scatterlist *tmp_sg;
3296         SGDescriptor_struct *curr_sg;
3297         drive_info_struct *drv;
3298         int i, dir;
3299         int sg_index = 0;
3300         int chained = 0;
3301
3302       queue:
3303         creq = blk_peek_request(q);
3304         if (!creq)
3305                 goto startio;
3306
3307         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3308
3309         c = cmd_alloc(h);
3310         if (!c)
3311                 goto full;
3312
3313         blk_start_request(creq);
3314
3315         tmp_sg = h->scatter_list[c->cmdindex];
3316         spin_unlock_irq(q->queue_lock);
3317
3318         c->cmd_type = CMD_RWREQ;
3319         c->rq = creq;
3320
3321         /* fill in the request */
3322         drv = creq->rq_disk->private_data;
3323         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3324         /* got command from pool, so use the command block index instead */
3325         /* for direct lookups. */
3326         /* The first 2 bits are reserved for controller error reporting. */
3327         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3328         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3329         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3330         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3331         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3332         c->Request.Type.Attribute = ATTR_SIMPLE;
3333         c->Request.Type.Direction =
3334             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3335         c->Request.Timeout = 0; /* Don't time out */
3336         c->Request.CDB[0] =
3337             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3338         start_blk = blk_rq_pos(creq);
3339         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3340                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3341         sg_init_table(tmp_sg, h->maxsgentries);
3342         seg = blk_rq_map_sg(q, creq, tmp_sg);
3343
3344         /* get the DMA records for the setup */
3345         if (c->Request.Type.Direction == XFER_READ)
3346                 dir = PCI_DMA_FROMDEVICE;
3347         else
3348                 dir = PCI_DMA_TODEVICE;
3349
3350         curr_sg = c->SG;
3351         sg_index = 0;
3352         chained = 0;
3353
3354         for (i = 0; i < seg; i++) {
3355                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3356                         !chained && ((seg - i) > 1)) {
3357                         /* Point to next chain block. */
3358                         curr_sg = h->cmd_sg_list[c->cmdindex];
3359                         sg_index = 0;
3360                         chained = 1;
3361                 }
3362                 curr_sg[sg_index].Len = tmp_sg[i].length;
3363                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3364                                                 tmp_sg[i].offset,
3365                                                 tmp_sg[i].length, dir);
3366                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3367                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3368                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3369                 ++sg_index;
3370         }
3371         if (chained)
3372                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3373                         (seg - (h->max_cmd_sgentries - 1)) *
3374                                 sizeof(SGDescriptor_struct));
3375
3376         /* track how many SG entries we are using */
3377         if (seg > h->maxSG)
3378                 h->maxSG = seg;
3379
3380         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3381                         "chained[%d]\n",
3382                         blk_rq_sectors(creq), seg, chained);
3383
3384         c->Header.SGTotal = seg + chained;
3385         if (seg <= h->max_cmd_sgentries)
3386                 c->Header.SGList = c->Header.SGTotal;
3387         else
3388                 c->Header.SGList = h->max_cmd_sgentries;
3389         set_performant_mode(h, c);
3390
3391         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3392                 if(h->cciss_read == CCISS_READ_10) {
3393                         c->Request.CDB[1] = 0;
3394                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3395                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3396                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3397                         c->Request.CDB[5] = start_blk & 0xff;
3398                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3399                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3400                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3401                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3402                 } else {
3403                         u32 upper32 = upper_32_bits(start_blk);
3404
3405                         c->Request.CDBLen = 16;
3406                         c->Request.CDB[1]= 0;
3407                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3408                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3409                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3410                         c->Request.CDB[5]= upper32 & 0xff;
3411                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3412                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3413                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3414                         c->Request.CDB[9]= start_blk & 0xff;
3415                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3416                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3417                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3418                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3419                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3420                 }
3421         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3422                 c->Request.CDBLen = creq->cmd_len;
3423                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3424         } else {
3425                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3426                         creq->cmd_type);
3427                 BUG();
3428         }
3429
3430         spin_lock_irq(q->queue_lock);
3431
3432         addQ(&h->reqQ, c);
3433         h->Qdepth++;
3434         if (h->Qdepth > h->maxQsinceinit)
3435                 h->maxQsinceinit = h->Qdepth;
3436
3437         goto queue;
3438 full:
3439         blk_stop_queue(q);
3440 startio:
3441         /* We will already have the driver lock here so not need
3442          * to lock it.
3443          */
3444         start_io(h);
3445 }
3446
3447 static inline unsigned long get_next_completion(ctlr_info_t *h)
3448 {
3449         return h->access.command_completed(h);
3450 }
3451
3452 static inline int interrupt_pending(ctlr_info_t *h)
3453 {
3454         return h->access.intr_pending(h);
3455 }
3456
3457 static inline long interrupt_not_for_us(ctlr_info_t *h)
3458 {
3459         return ((h->access.intr_pending(h) == 0) ||
3460                 (h->interrupts_enabled == 0));
3461 }
3462
3463 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3464                         u32 raw_tag)
3465 {
3466         if (unlikely(tag_index >= h->nr_cmds)) {
3467                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3468                 return 1;
3469         }
3470         return 0;
3471 }
3472
3473 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3474                                 u32 raw_tag)
3475 {
3476         removeQ(c);
3477         if (likely(c->cmd_type == CMD_RWREQ))
3478                 complete_command(h, c, 0);
3479         else if (c->cmd_type == CMD_IOCTL_PEND)
3480                 complete(c->waiting);
3481 #ifdef CONFIG_CISS_SCSI_TAPE
3482         else if (c->cmd_type == CMD_SCSI)
3483                 complete_scsi_command(c, 0, raw_tag);
3484 #endif
3485 }
3486
3487 static inline u32 next_command(ctlr_info_t *h)
3488 {
3489         u32 a;
3490
3491         if (unlikely(!(h->transMethod & CFGTBL_Trans_Performant)))
3492                 return h->access.command_completed(h);
3493
3494         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3495                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3496                 (h->reply_pool_head)++;
3497                 h->commands_outstanding--;
3498         } else {
3499                 a = FIFO_EMPTY;
3500         }
3501         /* Check for wraparound */
3502         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3503                 h->reply_pool_head = h->reply_pool;
3504                 h->reply_pool_wraparound ^= 1;
3505         }
3506         return a;
3507 }
3508
3509 /* process completion of an indexed ("direct lookup") command */
3510 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3511 {
3512         u32 tag_index;
3513         CommandList_struct *c;
3514
3515         tag_index = cciss_tag_to_index(raw_tag);
3516         if (bad_tag(h, tag_index, raw_tag))
3517                 return next_command(h);
3518         c = h->cmd_pool + tag_index;
3519         finish_cmd(h, c, raw_tag);
3520         return next_command(h);
3521 }
3522
3523 /* process completion of a non-indexed command */
3524 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3525 {
3526         CommandList_struct *c = NULL;
3527         __u32 busaddr_masked, tag_masked;
3528
3529         tag_masked = cciss_tag_discard_error_bits(h, raw_tag);
3530         list_for_each_entry(c, &h->cmpQ, list) {
3531                 busaddr_masked = cciss_tag_discard_error_bits(h, c->busaddr);
3532                 if (busaddr_masked == tag_masked) {
3533                         finish_cmd(h, c, raw_tag);
3534                         return next_command(h);
3535                 }
3536         }
3537         bad_tag(h, h->nr_cmds + 1, raw_tag);
3538         return next_command(h);
3539 }
3540
3541 /* Some controllers, like p400, will give us one interrupt
3542  * after a soft reset, even if we turned interrupts off.
3543  * Only need to check for this in the cciss_xxx_discard_completions
3544  * functions.
3545  */
3546 static int ignore_bogus_interrupt(ctlr_info_t *h)
3547 {
3548         if (likely(!reset_devices))
3549                 return 0;
3550
3551         if (likely(h->interrupts_enabled))
3552                 return 0;
3553
3554         dev_info(&h->pdev->dev, "Received interrupt while interrupts disabled "
3555                 "(known firmware bug.)  Ignoring.\n");
3556
3557         return 1;
3558 }
3559
3560 static irqreturn_t cciss_intx_discard_completions(int irq, void *dev_id)
3561 {
3562         ctlr_info_t *h = dev_id;
3563         unsigned long flags;
3564         u32 raw_tag;
3565
3566         if (ignore_bogus_interrupt(h))
3567                 return IRQ_NONE;
3568
3569         if (interrupt_not_for_us(h))
3570                 return IRQ_NONE;
3571         spin_lock_irqsave(&h->lock, flags);
3572         while (interrupt_pending(h)) {
3573                 raw_tag = get_next_completion(h);
3574                 while (raw_tag != FIFO_EMPTY)
3575                         raw_tag = next_command(h);
3576         }
3577         spin_unlock_irqrestore(&h->lock, flags);
3578         return IRQ_HANDLED;
3579 }
3580
3581 static irqreturn_t cciss_msix_discard_completions(int irq, void *dev_id)
3582 {
3583         ctlr_info_t *h = dev_id;
3584         unsigned long flags;
3585         u32 raw_tag;
3586
3587         if (ignore_bogus_interrupt(h))
3588                 return IRQ_NONE;
3589
3590         spin_lock_irqsave(&h->lock, flags);
3591         raw_tag = get_next_completion(h);
3592         while (raw_tag != FIFO_EMPTY)
3593                 raw_tag = next_command(h);
3594         spin_unlock_irqrestore(&h->lock, flags);
3595         return IRQ_HANDLED;
3596 }
3597
3598 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3599 {
3600         ctlr_info_t *h = dev_id;
3601         unsigned long flags;
3602         u32 raw_tag;
3603
3604         if (interrupt_not_for_us(h))
3605                 return IRQ_NONE;
3606         spin_lock_irqsave(&h->lock, flags);
3607         while (interrupt_pending(h)) {
3608                 raw_tag = get_next_completion(h);
3609                 while (raw_tag != FIFO_EMPTY) {
3610                         if (cciss_tag_contains_index(raw_tag))
3611                                 raw_tag = process_indexed_cmd(h, raw_tag);
3612                         else
3613                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3614                 }
3615         }
3616         spin_unlock_irqrestore(&h->lock, flags);
3617         return IRQ_HANDLED;
3618 }
3619
3620 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3621  * check the interrupt pending register because it is not set.
3622  */
3623 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3624 {
3625         ctlr_info_t *h = dev_id;
3626         unsigned long flags;
3627         u32 raw_tag;
3628
3629         spin_lock_irqsave(&h->lock, flags);
3630         raw_tag = get_next_completion(h);
3631         while (raw_tag != FIFO_EMPTY) {
3632                 if (cciss_tag_contains_index(raw_tag))
3633                         raw_tag = process_indexed_cmd(h, raw_tag);
3634                 else
3635                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3636         }
3637         spin_unlock_irqrestore(&h->lock, flags);
3638         return IRQ_HANDLED;
3639 }
3640
3641 /**
3642  * add_to_scan_list() - add controller to rescan queue
3643  * @h:                Pointer to the controller.
3644  *
3645  * Adds the controller to the rescan queue if not already on the queue.
3646  *
3647  * returns 1 if added to the queue, 0 if skipped (could be on the
3648  * queue already, or the controller could be initializing or shutting
3649  * down).
3650  **/
3651 static int add_to_scan_list(struct ctlr_info *h)
3652 {
3653         struct ctlr_info *test_h;
3654         int found = 0;
3655         int ret = 0;
3656
3657         if (h->busy_initializing)
3658                 return 0;
3659
3660         if (!mutex_trylock(&h->busy_shutting_down))
3661                 return 0;
3662
3663         mutex_lock(&scan_mutex);
3664         list_for_each_entry(test_h, &scan_q, scan_list) {
3665                 if (test_h == h) {
3666                         found = 1;
3667                         break;
3668                 }
3669         }
3670         if (!found && !h->busy_scanning) {
3671                 INIT_COMPLETION(h->scan_wait);
3672                 list_add_tail(&h->scan_list, &scan_q);
3673                 ret = 1;
3674         }
3675         mutex_unlock(&scan_mutex);
3676         mutex_unlock(&h->busy_shutting_down);
3677
3678         return ret;
3679 }
3680
3681 /**
3682  * remove_from_scan_list() - remove controller from rescan queue
3683  * @h:                     Pointer to the controller.
3684  *
3685  * Removes the controller from the rescan queue if present. Blocks if
3686  * the controller is currently conducting a rescan.  The controller
3687  * can be in one of three states:
3688  * 1. Doesn't need a scan
3689  * 2. On the scan list, but not scanning yet (we remove it)
3690  * 3. Busy scanning (and not on the list). In this case we want to wait for
3691  *    the scan to complete to make sure the scanning thread for this
3692  *    controller is completely idle.
3693  **/
3694 static void remove_from_scan_list(struct ctlr_info *h)
3695 {
3696         struct ctlr_info *test_h, *tmp_h;
3697
3698         mutex_lock(&scan_mutex);
3699         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3700                 if (test_h == h) { /* state 2. */
3701                         list_del(&h->scan_list);
3702                         complete_all(&h->scan_wait);
3703                         mutex_unlock(&scan_mutex);
3704                         return;
3705                 }
3706         }
3707         if (h->busy_scanning) { /* state 3. */
3708                 mutex_unlock(&scan_mutex);
3709                 wait_for_completion(&h->scan_wait);
3710         } else { /* state 1, nothing to do. */
3711                 mutex_unlock(&scan_mutex);
3712         }
3713 }
3714
3715 /**
3716  * scan_thread() - kernel thread used to rescan controllers
3717  * @data:        Ignored.
3718  *
3719  * A kernel thread used scan for drive topology changes on
3720  * controllers. The thread processes only one controller at a time
3721  * using a queue.  Controllers are added to the queue using
3722  * add_to_scan_list() and removed from the queue either after done
3723  * processing or using remove_from_scan_list().
3724  *
3725  * returns 0.
3726  **/
3727 static int scan_thread(void *data)
3728 {
3729         struct ctlr_info *h;
3730
3731         while (1) {
3732                 set_current_state(TASK_INTERRUPTIBLE);
3733                 schedule();
3734                 if (kthread_should_stop())
3735                         break;
3736
3737                 while (1) {
3738                         mutex_lock(&scan_mutex);
3739                         if (list_empty(&scan_q)) {
3740                                 mutex_unlock(&scan_mutex);
3741                                 break;
3742                         }
3743
3744                         h = list_entry(scan_q.next,
3745                                        struct ctlr_info,
3746                                        scan_list);
3747                         list_del(&h->scan_list);
3748                         h->busy_scanning = 1;
3749                         mutex_unlock(&scan_mutex);
3750
3751                         rebuild_lun_table(h, 0, 0);
3752                         complete_all(&h->scan_wait);
3753                         mutex_lock(&scan_mutex);
3754                         h->busy_scanning = 0;
3755                         mutex_unlock(&scan_mutex);
3756                 }
3757         }
3758
3759         return 0;
3760 }
3761
3762 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3763 {
3764         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3765                 return 0;
3766
3767         switch (c->err_info->SenseInfo[12]) {
3768         case STATE_CHANGED:
3769                 dev_warn(&h->pdev->dev, "a state change "
3770                         "detected, command retried\n");
3771                 return 1;
3772         break;
3773         case LUN_FAILED:
3774                 dev_warn(&h->pdev->dev, "LUN failure "
3775                         "detected, action required\n");
3776                 return 1;
3777         break;
3778         case REPORT_LUNS_CHANGED:
3779                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3780         /*
3781          * Here, we could call add_to_scan_list and wake up the scan thread,
3782          * except that it's quite likely that we will get more than one
3783          * REPORT_LUNS_CHANGED condition in quick succession, which means
3784          * that those which occur after the first one will likely happen
3785          * *during* the scan_thread's rescan.  And the rescan code is not
3786          * robust enough to restart in the middle, undoing what it has already
3787          * done, and it's not clear that it's even possible to do this, since
3788          * part of what it does is notify the block layer, which starts
3789          * doing it's own i/o to read partition tables and so on, and the
3790          * driver doesn't have visibility to know what might need undoing.
3791          * In any event, if possible, it is horribly complicated to get right
3792          * so we just don't do it for now.
3793          *
3794          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3795          */
3796                 return 1;
3797         break;
3798         case POWER_OR_RESET:
3799                 dev_warn(&h->pdev->dev,
3800                         "a power on or device reset detected\n");
3801                 return 1;
3802         break;
3803         case UNIT_ATTENTION_CLEARED:
3804                 dev_warn(&h->pdev->dev,
3805                         "unit attention cleared by another initiator\n");
3806                 return 1;
3807         break;
3808         default:
3809                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3810                 return 1;
3811         }
3812 }
3813
3814 /*
3815  *  We cannot read the structure directly, for portability we must use
3816  *   the io functions.
3817  *   This is for debug only.
3818  */
3819 static void print_cfg_table(ctlr_info_t *h)
3820 {
3821         int i;
3822         char temp_name[17];
3823         CfgTable_struct *tb = h->cfgtable;
3824
3825         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3826         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3827         for (i = 0; i < 4; i++)
3828                 temp_name[i] = readb(&(tb->Signature[i]));
3829         temp_name[4] = '\0';
3830         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3831         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3832                 readl(&(tb->SpecValence)));
3833         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3834                readl(&(tb->TransportSupport)));
3835         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3836                readl(&(tb->TransportActive)));
3837         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3838                readl(&(tb->HostWrite.TransportRequest)));
3839         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3840                readl(&(tb->HostWrite.CoalIntDelay)));
3841         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3842                readl(&(tb->HostWrite.CoalIntCount)));
3843         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3844                readl(&(tb->CmdsOutMax)));
3845         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3846                 readl(&(tb->BusTypes)));
3847         for (i = 0; i < 16; i++)
3848                 temp_name[i] = readb(&(tb->ServerName[i]));
3849         temp_name[16] = '\0';
3850         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3851         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3852                 readl(&(tb->HeartBeat)));
3853 }
3854
3855 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3856 {
3857         int i, offset, mem_type, bar_type;
3858         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3859                 return 0;
3860         offset = 0;
3861         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3862                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3863                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3864                         offset += 4;
3865                 else {
3866                         mem_type = pci_resource_flags(pdev, i) &
3867                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3868                         switch (mem_type) {
3869                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3870                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3871                                 offset += 4;    /* 32 bit */
3872                                 break;
3873                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3874                                 offset += 8;
3875                                 break;
3876                         default:        /* reserved in PCI 2.2 */
3877                                 dev_warn(&pdev->dev,
3878                                        "Base address is invalid\n");
3879                                 return -1;
3880                                 break;
3881                         }
3882                 }
3883                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3884                         return i + 1;
3885         }
3886         return -1;
3887 }
3888
3889 /* Fill in bucket_map[], given nsgs (the max number of
3890  * scatter gather elements supported) and bucket[],
3891  * which is an array of 8 integers.  The bucket[] array
3892  * contains 8 different DMA transfer sizes (in 16
3893  * byte increments) which the controller uses to fetch
3894  * commands.  This function fills in bucket_map[], which
3895  * maps a given number of scatter gather elements to one of
3896  * the 8 DMA transfer sizes.  The point of it is to allow the
3897  * controller to only do as much DMA as needed to fetch the
3898  * command, with the DMA transfer size encoded in the lower
3899  * bits of the command address.
3900  */
3901 static void  calc_bucket_map(int bucket[], int num_buckets,
3902         int nsgs, int *bucket_map)
3903 {
3904         int i, j, b, size;
3905
3906         /* even a command with 0 SGs requires 4 blocks */
3907 #define MINIMUM_TRANSFER_BLOCKS 4
3908 #define NUM_BUCKETS 8
3909         /* Note, bucket_map must have nsgs+1 entries. */
3910         for (i = 0; i <= nsgs; i++) {
3911                 /* Compute size of a command with i SG entries */
3912                 size = i + MINIMUM_TRANSFER_BLOCKS;
3913                 b = num_buckets; /* Assume the biggest bucket */
3914                 /* Find the bucket that is just big enough */
3915                 for (j = 0; j < 8; j++) {
3916                         if (bucket[j] >= size) {
3917                                 b = j;
3918                                 break;
3919                         }
3920                 }
3921                 /* for a command with i SG entries, use bucket b. */
3922                 bucket_map[i] = b;
3923         }
3924 }
3925
3926 static void cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3927 {
3928         int i;
3929
3930         /* under certain very rare conditions, this can take awhile.
3931          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3932          * as we enter this code.) */
3933         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3934                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3935                         break;
3936                 usleep_range(10000, 20000);
3937         }
3938 }
3939
3940 static void cciss_enter_performant_mode(ctlr_info_t *h, u32 use_short_tags)
3941 {
3942         /* This is a bit complicated.  There are 8 registers on
3943          * the controller which we write to to tell it 8 different
3944          * sizes of commands which there may be.  It's a way of
3945          * reducing the DMA done to fetch each command.  Encoded into
3946          * each command's tag are 3 bits which communicate to the controller
3947          * which of the eight sizes that command fits within.  The size of
3948          * each command depends on how many scatter gather entries there are.
3949          * Each SG entry requires 16 bytes.  The eight registers are programmed
3950          * with the number of 16-byte blocks a command of that size requires.
3951          * The smallest command possible requires 5 such 16 byte blocks.
3952          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3953          * blocks.  Note, this only extends to the SG entries contained
3954          * within the command block, and does not extend to chained blocks
3955          * of SG elements.   bft[] contains the eight values we write to
3956          * the registers.  They are not evenly distributed, but have more
3957          * sizes for small commands, and fewer sizes for larger commands.
3958          */
3959         __u32 trans_offset;
3960         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3961                         /*
3962                          *  5 = 1 s/g entry or 4k
3963                          *  6 = 2 s/g entry or 8k
3964                          *  8 = 4 s/g entry or 16k
3965                          * 10 = 6 s/g entry or 24k
3966                          */
3967         unsigned long register_value;
3968         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3969
3970         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3971
3972         /* Controller spec: zero out this buffer. */
3973         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3974         h->reply_pool_head = h->reply_pool;
3975
3976         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3977         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3978                                 h->blockFetchTable);
3979         writel(bft[0], &h->transtable->BlockFetch0);
3980         writel(bft[1], &h->transtable->BlockFetch1);
3981         writel(bft[2], &h->transtable->BlockFetch2);
3982         writel(bft[3], &h->transtable->BlockFetch3);
3983         writel(bft[4], &h->transtable->BlockFetch4);
3984         writel(bft[5], &h->transtable->BlockFetch5);
3985         writel(bft[6], &h->transtable->BlockFetch6);
3986         writel(bft[7], &h->transtable->BlockFetch7);
3987
3988         /* size of controller ring buffer */
3989         writel(h->max_commands, &h->transtable->RepQSize);
3990         writel(1, &h->transtable->RepQCount);
3991         writel(0, &h->transtable->RepQCtrAddrLow32);
3992         writel(0, &h->transtable->RepQCtrAddrHigh32);
3993         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3994         writel(0, &h->transtable->RepQAddr0High32);
3995         writel(CFGTBL_Trans_Performant | use_short_tags,
3996                         &(h->cfgtable->HostWrite.TransportRequest));
3997
3998         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3999         cciss_wait_for_mode_change_ack(h);
4000         register_value = readl(&(h->cfgtable->TransportActive));
4001         if (!(register_value & CFGTBL_Trans_Performant))
4002                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
4003                                         " performant mode\n");
4004 }
4005
4006 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h)
4007 {
4008         __u32 trans_support;
4009
4010         if (cciss_simple_mode)
4011                 return;
4012
4013         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
4014         /* Attempt to put controller into performant mode if supported */
4015         /* Does board support performant mode? */
4016         trans_support = readl(&(h->cfgtable->TransportSupport));
4017         if (!(trans_support & PERFORMANT_MODE))
4018                 return;
4019
4020         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
4021         /* Performant mode demands commands on a 32 byte boundary
4022          * pci_alloc_consistent aligns on page boundarys already.
4023          * Just need to check if divisible by 32
4024          */
4025         if ((sizeof(CommandList_struct) % 32) != 0) {
4026                 dev_warn(&h->pdev->dev, "%s %d %s\n",
4027                         "cciss info: command size[",
4028                         (int)sizeof(CommandList_struct),
4029                         "] not divisible by 32, no performant mode..\n");
4030                 return;
4031         }
4032
4033         /* Performant mode ring buffer and supporting data structures */
4034         h->reply_pool = (__u64 *)pci_alloc_consistent(
4035                 h->pdev, h->max_commands * sizeof(__u64),
4036                 &(h->reply_pool_dhandle));
4037
4038         /* Need a block fetch table for performant mode */
4039         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
4040                 sizeof(__u32)), GFP_KERNEL);
4041
4042         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
4043                 goto clean_up;
4044
4045         cciss_enter_performant_mode(h,
4046                 trans_support & CFGTBL_Trans_use_short_tags);
4047
4048         /* Change the access methods to the performant access methods */
4049         h->access = SA5_performant_access;
4050         h->transMethod = CFGTBL_Trans_Performant;
4051
4052         return;
4053 clean_up:
4054         kfree(h->blockFetchTable);
4055         if (h->reply_pool)
4056                 pci_free_consistent(h->pdev,
4057                                 h->max_commands * sizeof(__u64),
4058                                 h->reply_pool,
4059                                 h->reply_pool_dhandle);
4060         return;
4061
4062 } /* cciss_put_controller_into_performant_mode */
4063
4064 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
4065  * controllers that are capable. If not, we use IO-APIC mode.
4066  */
4067
4068 static void cciss_interrupt_mode(ctlr_info_t *h)
4069 {
4070 #ifdef CONFIG_PCI_MSI
4071         int err;
4072         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
4073         {0, 2}, {0, 3}
4074         };
4075
4076         /* Some boards advertise MSI but don't really support it */
4077         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
4078             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
4079                 goto default_int_mode;
4080
4081         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
4082                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
4083                 if (!err) {
4084                         h->intr[0] = cciss_msix_entries[0].vector;
4085                         h->intr[1] = cciss_msix_entries[1].vector;
4086                         h->intr[2] = cciss_msix_entries[2].vector;
4087                         h->intr[3] = cciss_msix_entries[3].vector;
4088                         h->msix_vector = 1;
4089                         return;
4090                 }
4091                 if (err > 0) {
4092                         dev_warn(&h->pdev->dev,
4093                                 "only %d MSI-X vectors available\n", err);
4094                         goto default_int_mode;
4095                 } else {
4096                         dev_warn(&h->pdev->dev,
4097                                 "MSI-X init failed %d\n", err);
4098                         goto default_int_mode;
4099                 }
4100         }
4101         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
4102                 if (!pci_enable_msi(h->pdev))
4103                         h->msi_vector = 1;
4104                 else
4105                         dev_warn(&h->pdev->dev, "MSI init failed\n");
4106         }
4107 default_int_mode:
4108 #endif                          /* CONFIG_PCI_MSI */
4109         /* if we get here we're going to use the default interrupt mode */
4110         h->intr[h->intr_mode] = h->pdev->irq;
4111         return;
4112 }
4113
4114 static int cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
4115 {
4116         int i;
4117         u32 subsystem_vendor_id, subsystem_device_id;
4118
4119         subsystem_vendor_id = pdev->subsystem_vendor;
4120         subsystem_device_id = pdev->subsystem_device;
4121         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
4122                         subsystem_vendor_id;
4123
4124         for (i = 0; i < ARRAY_SIZE(products); i++) {
4125                 /* Stand aside for hpsa driver on request */
4126                 if (cciss_allow_hpsa)
4127                         return -ENODEV;
4128                 if (*board_id == products[i].board_id)
4129                         return i;
4130         }
4131         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4132                 *board_id);
4133         return -ENODEV;
4134 }
4135
4136 static inline bool cciss_board_disabled(ctlr_info_t *h)
4137 {
4138         u16 command;
4139
4140         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4141         return ((command & PCI_COMMAND_MEMORY) == 0);
4142 }
4143
4144 static int cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4145                                      unsigned long *memory_bar)
4146 {
4147         int i;
4148
4149         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4150                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4151                         /* addressing mode bits already removed */
4152                         *memory_bar = pci_resource_start(pdev, i);
4153                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4154                                 *memory_bar);
4155                         return 0;
4156                 }
4157         dev_warn(&pdev->dev, "no memory BAR found\n");
4158         return -ENODEV;
4159 }
4160
4161 static int cciss_wait_for_board_state(struct pci_dev *pdev,
4162                                       void __iomem *vaddr, int wait_for_ready)
4163 #define BOARD_READY 1
4164 #define BOARD_NOT_READY 0
4165 {
4166         int i, iterations;
4167         u32 scratchpad;
4168
4169         if (wait_for_ready)
4170                 iterations = CCISS_BOARD_READY_ITERATIONS;
4171         else
4172                 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
4173
4174         for (i = 0; i < iterations; i++) {
4175                 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
4176                 if (wait_for_ready) {
4177                         if (scratchpad == CCISS_FIRMWARE_READY)
4178                                 return 0;
4179                 } else {
4180                         if (scratchpad != CCISS_FIRMWARE_READY)
4181                                 return 0;
4182                 }
4183                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4184         }
4185         dev_warn(&pdev->dev, "board not ready, timed out.\n");
4186         return -ENODEV;
4187 }
4188
4189 static int cciss_find_cfg_addrs(struct pci_dev *pdev, void __iomem *vaddr,
4190                                 u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4191                                 u64 *cfg_offset)
4192 {
4193         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4194         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4195         *cfg_base_addr &= (u32) 0x0000ffff;
4196         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4197         if (*cfg_base_addr_index == -1) {
4198                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4199                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4200                 return -ENODEV;
4201         }
4202         return 0;
4203 }
4204
4205 static int cciss_find_cfgtables(ctlr_info_t *h)
4206 {
4207         u64 cfg_offset;
4208         u32 cfg_base_addr;
4209         u64 cfg_base_addr_index;
4210         u32 trans_offset;
4211         int rc;
4212
4213         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4214                 &cfg_base_addr_index, &cfg_offset);
4215         if (rc)
4216                 return rc;
4217         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4218                 cfg_base_addr_index) + cfg_offset, sizeof(*h->cfgtable));
4219         if (!h->cfgtable)
4220                 return -ENOMEM;
4221         rc = write_driver_ver_to_cfgtable(h->cfgtable);
4222         if (rc)
4223                 return rc;
4224         /* Find performant mode table. */
4225         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4226         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4227                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4228                                 sizeof(*h->transtable));
4229         if (!h->transtable)
4230                 return -ENOMEM;
4231         return 0;
4232 }
4233
4234 static void cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4235 {
4236         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4237
4238         /* Limit commands in memory limited kdump scenario. */
4239         if (reset_devices && h->max_commands > 32)
4240                 h->max_commands = 32;
4241
4242         if (h->max_commands < 16) {
4243                 dev_warn(&h->pdev->dev, "Controller reports "
4244                         "max supported commands of %d, an obvious lie. "
4245                         "Using 16.  Ensure that firmware is up to date.\n",
4246                         h->max_commands);
4247                 h->max_commands = 16;
4248         }
4249 }
4250
4251 /* Interrogate the hardware for some limits:
4252  * max commands, max SG elements without chaining, and with chaining,
4253  * SG chain block size, etc.
4254  */
4255 static void cciss_find_board_params(ctlr_info_t *h)
4256 {
4257         cciss_get_max_perf_mode_cmds(h);
4258         h->nr_cmds = h->max_commands - 4 - cciss_tape_cmds;
4259         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4260         /*
4261          * Limit in-command s/g elements to 32 save dma'able memory.
4262          * Howvever spec says if 0, use 31
4263          */
4264         h->max_cmd_sgentries = 31;
4265         if (h->maxsgentries > 512) {
4266                 h->max_cmd_sgentries = 32;
4267                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4268                 h->maxsgentries--; /* save one for chain pointer */
4269         } else {
4270                 h->maxsgentries = 31; /* default to traditional values */
4271                 h->chainsize = 0;
4272         }
4273 }
4274
4275 static inline bool CISS_signature_present(ctlr_info_t *h)
4276 {
4277         if (!check_signature(h->cfgtable->Signature, "CISS", 4)) {
4278                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4279                 return false;
4280         }
4281         return true;
4282 }
4283
4284 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4285 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4286 {
4287 #ifdef CONFIG_X86
4288         u32 prefetch;
4289
4290         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4291         prefetch |= 0x100;
4292         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4293 #endif
4294 }
4295
4296 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4297  * in a prefetch beyond physical memory.
4298  */
4299 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4300 {
4301         u32 dma_prefetch;
4302         __u32 dma_refetch;
4303
4304         if (h->board_id != 0x3225103C)
4305                 return;
4306         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4307         dma_prefetch |= 0x8000;
4308         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4309         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4310         dma_refetch |= 0x1;
4311         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4312 }
4313
4314 static int cciss_pci_init(ctlr_info_t *h)
4315 {
4316         int prod_index, err;
4317
4318         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4319         if (prod_index < 0)
4320                 return -ENODEV;
4321         h->product_name = products[prod_index].product_name;
4322         h->access = *(products[prod_index].access);
4323
4324         if (cciss_board_disabled(h)) {
4325                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4326                 return -ENODEV;
4327         }
4328
4329         pci_disable_link_state(h->pdev, PCIE_LINK_STATE_L0S |
4330                                 PCIE_LINK_STATE_L1 | PCIE_LINK_STATE_CLKPM);
4331
4332         err = pci_enable_device(h->pdev);
4333         if (err) {
4334                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4335                 return err;
4336         }
4337
4338         err = pci_request_regions(h->pdev, "cciss");
4339         if (err) {
4340                 dev_warn(&h->pdev->dev,
4341                         "Cannot obtain PCI resources, aborting\n");
4342                 return err;
4343         }
4344
4345         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4346         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4347
4348 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4349  * else we use the IO-APIC interrupt assigned to us by system ROM.
4350  */
4351         cciss_interrupt_mode(h);
4352         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4353         if (err)
4354                 goto err_out_free_res;
4355         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4356         if (!h->vaddr) {
4357                 err = -ENOMEM;
4358                 goto err_out_free_res;
4359         }
4360         err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4361         if (err)
4362                 goto err_out_free_res;
4363         err = cciss_find_cfgtables(h);
4364         if (err)
4365                 goto err_out_free_res;
4366         print_cfg_table(h);
4367         cciss_find_board_params(h);
4368
4369         if (!CISS_signature_present(h)) {
4370                 err = -ENODEV;
4371                 goto err_out_free_res;
4372         }
4373         cciss_enable_scsi_prefetch(h);
4374         cciss_p600_dma_prefetch_quirk(h);
4375         err = cciss_enter_simple_mode(h);
4376         if (err)
4377                 goto err_out_free_res;
4378         cciss_put_controller_into_performant_mode(h);
4379         return 0;
4380
4381 err_out_free_res:
4382         /*
4383          * Deliberately omit pci_disable_device(): it does something nasty to
4384          * Smart Array controllers that pci_enable_device does not undo
4385          */
4386         if (h->transtable)
4387                 iounmap(h->transtable);
4388         if (h->cfgtable)
4389                 iounmap(h->cfgtable);
4390         if (h->vaddr)
4391                 iounmap(h->vaddr);
4392         pci_release_regions(h->pdev);
4393         return err;
4394 }
4395
4396 /* Function to find the first free pointer into our hba[] array
4397  * Returns -1 if no free entries are left.
4398  */
4399 static int alloc_cciss_hba(struct pci_dev *pdev)
4400 {
4401         int i;
4402
4403         for (i = 0; i < MAX_CTLR; i++) {
4404                 if (!hba[i]) {
4405                         ctlr_info_t *h;
4406
4407                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4408                         if (!h)
4409                                 goto Enomem;
4410                         hba[i] = h;
4411                         return i;
4412                 }
4413         }
4414         dev_warn(&pdev->dev, "This driver supports a maximum"
4415                " of %d controllers.\n", MAX_CTLR);
4416         return -1;
4417 Enomem:
4418         dev_warn(&pdev->dev, "out of memory.\n");
4419         return -1;
4420 }
4421
4422 static void free_hba(ctlr_info_t *h)
4423 {
4424         int i;
4425
4426         hba[h->ctlr] = NULL;
4427         for (i = 0; i < h->highest_lun + 1; i++)
4428                 if (h->gendisk[i] != NULL)
4429                         put_disk(h->gendisk[i]);
4430         kfree(h);
4431 }
4432
4433 /* Send a message CDB to the firmware. */
4434 static int cciss_message(struct pci_dev *pdev, unsigned char opcode,
4435                          unsigned char type)
4436 {
4437         typedef struct {
4438                 CommandListHeader_struct CommandHeader;
4439                 RequestBlock_struct Request;
4440                 ErrDescriptor_struct ErrorDescriptor;
4441         } Command;
4442         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4443         Command *cmd;
4444         dma_addr_t paddr64;
4445         uint32_t paddr32, tag;
4446         void __iomem *vaddr;
4447         int i, err;
4448
4449         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4450         if (vaddr == NULL)
4451                 return -ENOMEM;
4452
4453         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4454            CCISS commands, so they must be allocated from the lower 4GiB of
4455            memory. */
4456         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4457         if (err) {
4458                 iounmap(vaddr);
4459                 return -ENOMEM;
4460         }
4461
4462         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4463         if (cmd == NULL) {
4464                 iounmap(vaddr);
4465                 return -ENOMEM;
4466         }
4467
4468         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4469            although there's no guarantee, we assume that the address is at
4470            least 4-byte aligned (most likely, it's page-aligned). */
4471         paddr32 = paddr64;
4472
4473         cmd->CommandHeader.ReplyQueue = 0;
4474         cmd->CommandHeader.SGList = 0;
4475         cmd->CommandHeader.SGTotal = 0;
4476         cmd->CommandHeader.Tag.lower = paddr32;
4477         cmd->CommandHeader.Tag.upper = 0;
4478         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4479
4480         cmd->Request.CDBLen = 16;
4481         cmd->Request.Type.Type = TYPE_MSG;
4482         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4483         cmd->Request.Type.Direction = XFER_NONE;
4484         cmd->Request.Timeout = 0; /* Don't time out */
4485         cmd->Request.CDB[0] = opcode;
4486         cmd->Request.CDB[1] = type;
4487         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4488
4489         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4490         cmd->ErrorDescriptor.Addr.upper = 0;
4491         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4492
4493         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4494
4495         for (i = 0; i < 10; i++) {
4496                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4497                 if ((tag & ~3) == paddr32)
4498                         break;
4499                 msleep(CCISS_POST_RESET_NOOP_TIMEOUT_MSECS);
4500         }
4501
4502         iounmap(vaddr);
4503
4504         /* we leak the DMA buffer here ... no choice since the controller could
4505            still complete the command. */
4506         if (i == 10) {
4507                 dev_err(&pdev->dev,
4508                         "controller message %02x:%02x timed out\n",
4509                         opcode, type);
4510                 return -ETIMEDOUT;
4511         }
4512
4513         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4514
4515         if (tag & 2) {
4516                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4517                         opcode, type);
4518                 return -EIO;
4519         }
4520
4521         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4522                 opcode, type);
4523         return 0;
4524 }
4525
4526 #define cciss_noop(p) cciss_message(p, 3, 0)
4527
4528 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4529         void * __iomem vaddr, u32 use_doorbell)
4530 {
4531         u16 pmcsr;
4532         int pos;
4533
4534         if (use_doorbell) {
4535                 /* For everything after the P600, the PCI power state method
4536                  * of resetting the controller doesn't work, so we have this
4537                  * other way using the doorbell register.
4538                  */
4539                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4540                 writel(use_doorbell, vaddr + SA5_DOORBELL);
4541         } else { /* Try to do it the PCI power state way */
4542
4543                 /* Quoting from the Open CISS Specification: "The Power
4544                  * Management Control/Status Register (CSR) controls the power
4545                  * state of the device.  The normal operating state is D0,
4546                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4547                  * the controller, place the interface device in D3 then to D0,
4548                  * this causes a secondary PCI reset which will reset the
4549                  * controller." */
4550
4551                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4552                 if (pos == 0) {
4553                         dev_err(&pdev->dev,
4554                                 "cciss_controller_hard_reset: "
4555                                 "PCI PM not supported\n");
4556                         return -ENODEV;
4557                 }
4558                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4559                 /* enter the D3hot power management state */
4560                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4561                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4562                 pmcsr |= PCI_D3hot;
4563                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4564
4565                 msleep(500);
4566
4567                 /* enter the D0 power management state */
4568                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4569                 pmcsr |= PCI_D0;
4570                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4571
4572                 /*
4573                  * The P600 requires a small delay when changing states.
4574                  * Otherwise we may think the board did not reset and we bail.
4575                  * This for kdump only and is particular to the P600.
4576                  */
4577                 msleep(500);
4578         }
4579         return 0;
4580 }
4581
4582 static void init_driver_version(char *driver_version, int len)
4583 {
4584         memset(driver_version, 0, len);
4585         strncpy(driver_version, "cciss " DRIVER_NAME, len - 1);
4586 }
4587
4588 static int write_driver_ver_to_cfgtable(CfgTable_struct __iomem *cfgtable)
4589 {
4590         char *driver_version;
4591         int i, size = sizeof(cfgtable->driver_version);
4592
4593         driver_version = kmalloc(size, GFP_KERNEL);
4594         if (!driver_version)
4595                 return -ENOMEM;
4596
4597         init_driver_version(driver_version, size);
4598         for (i = 0; i < size; i++)
4599                 writeb(driver_version[i], &cfgtable->driver_version[i]);
4600         kfree(driver_version);
4601         return 0;
4602 }
4603
4604 static void read_driver_ver_from_cfgtable(CfgTable_struct __iomem *cfgtable,
4605                                           unsigned char *driver_ver)
4606 {
4607         int i;
4608
4609         for (i = 0; i < sizeof(cfgtable->driver_version); i++)
4610                 driver_ver[i] = readb(&cfgtable->driver_version[i]);
4611 }
4612
4613 static int controller_reset_failed(CfgTable_struct __iomem *cfgtable)
4614 {
4615
4616         char *driver_ver, *old_driver_ver;
4617         int rc, size = sizeof(cfgtable->driver_version);
4618
4619         old_driver_ver = kmalloc(2 * size, GFP_KERNEL);
4620         if (!old_driver_ver)
4621                 return -ENOMEM;
4622         driver_ver = old_driver_ver + size;
4623
4624         /* After a reset, the 32 bytes of "driver version" in the cfgtable
4625          * should have been changed, otherwise we know the reset failed.
4626          */
4627         init_driver_version(old_driver_ver, size);
4628         read_driver_ver_from_cfgtable(cfgtable, driver_ver);
4629         rc = !memcmp(driver_ver, old_driver_ver, size);
4630         kfree(old_driver_ver);
4631         return rc;
4632 }
4633
4634 /* This does a hard reset of the controller using PCI power management
4635  * states or using the doorbell register. */
4636 static int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4637 {
4638         u64 cfg_offset;
4639         u32 cfg_base_addr;
4640         u64 cfg_base_addr_index;
4641         void __iomem *vaddr;
4642         unsigned long paddr;
4643         u32 misc_fw_support;
4644         int rc;
4645         CfgTable_struct __iomem *cfgtable;
4646         u32 use_doorbell;
4647         u32 board_id;
4648         u16 command_register;
4649
4650         /* For controllers as old a the p600, this is very nearly
4651          * the same thing as
4652          *
4653          * pci_save_state(pci_dev);
4654          * pci_set_power_state(pci_dev, PCI_D3hot);
4655          * pci_set_power_state(pci_dev, PCI_D0);
4656          * pci_restore_state(pci_dev);
4657          *
4658          * For controllers newer than the P600, the pci power state
4659          * method of resetting doesn't work so we have another way
4660          * using the doorbell register.
4661          */
4662
4663         /* Exclude 640x boards.  These are two pci devices in one slot
4664          * which share a battery backed cache module.  One controls the
4665          * cache, the other accesses the cache through the one that controls
4666          * it.  If we reset the one controlling the cache, the other will
4667          * likely not be happy.  Just forbid resetting this conjoined mess.
4668          */
4669         cciss_lookup_board_id(pdev, &board_id);
4670         if (!ctlr_is_resettable(board_id)) {
4671                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4672                                 "due to shared cache module.");
4673                 return -ENODEV;
4674         }
4675
4676         /* if controller is soft- but not hard resettable... */
4677         if (!ctlr_is_hard_resettable(board_id))
4678                 return -ENOTSUPP; /* try soft reset later. */
4679
4680         /* Save the PCI command register */
4681         pci_read_config_word(pdev, 4, &command_register);
4682         /* Turn the board off.  This is so that later pci_restore_state()
4683          * won't turn the board on before the rest of config space is ready.
4684          */
4685         pci_disable_device(pdev);
4686         pci_save_state(pdev);
4687
4688         /* find the first memory BAR, so we can find the cfg table */
4689         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4690         if (rc)
4691                 return rc;
4692         vaddr = remap_pci_mem(paddr, 0x250);
4693         if (!vaddr)
4694                 return -ENOMEM;
4695
4696         /* find cfgtable in order to check if reset via doorbell is supported */
4697         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4698                                         &cfg_base_addr_index, &cfg_offset);
4699         if (rc)
4700                 goto unmap_vaddr;
4701         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4702                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4703         if (!cfgtable) {
4704                 rc = -ENOMEM;
4705                 goto unmap_vaddr;
4706         }
4707         rc = write_driver_ver_to_cfgtable(cfgtable);
4708         if (rc)
4709                 goto unmap_vaddr;
4710
4711         /* If reset via doorbell register is supported, use that.
4712          * There are two such methods.  Favor the newest method.
4713          */
4714         misc_fw_support = readl(&cfgtable->misc_fw_support);
4715         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET2;
4716         if (use_doorbell) {
4717                 use_doorbell = DOORBELL_CTLR_RESET2;
4718         } else {
4719                 use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4720                 if (use_doorbell) {
4721                         dev_warn(&pdev->dev, "Controller claims that "
4722                                 "'Bit 2 doorbell reset' is "
4723                                 "supported, but not 'bit 5 doorbell reset'.  "
4724                                 "Firmware update is recommended.\n");
4725                         rc = -ENOTSUPP; /* use the soft reset */
4726                         goto unmap_cfgtable;
4727                 }
4728         }
4729
4730         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4731         if (rc)
4732                 goto unmap_cfgtable;
4733         pci_restore_state(pdev);
4734         rc = pci_enable_device(pdev);
4735         if (rc) {
4736                 dev_warn(&pdev->dev, "failed to enable device.\n");
4737                 goto unmap_cfgtable;
4738         }
4739         pci_write_config_word(pdev, 4, command_register);
4740
4741         /* Some devices (notably the HP Smart Array 5i Controller)
4742            need a little pause here */
4743         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4744
4745         /* Wait for board to become not ready, then ready. */
4746         dev_info(&pdev->dev, "Waiting for board to reset.\n");
4747         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4748         if (rc) {
4749                 dev_warn(&pdev->dev, "Failed waiting for board to hard reset."
4750                                 "  Will try soft reset.\n");
4751                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4752                 goto unmap_cfgtable;
4753         }
4754         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4755         if (rc) {
4756                 dev_warn(&pdev->dev,
4757                         "failed waiting for board to become ready "
4758                         "after hard reset\n");
4759                 goto unmap_cfgtable;
4760         }
4761
4762         rc = controller_reset_failed(vaddr);
4763         if (rc < 0)
4764                 goto unmap_cfgtable;
4765         if (rc) {
4766                 dev_warn(&pdev->dev, "Unable to successfully hard reset "
4767                         "controller. Will try soft reset.\n");
4768                 rc = -ENOTSUPP; /* Not expected, but try soft reset later */
4769         } else {
4770                 dev_info(&pdev->dev, "Board ready after hard reset.\n");
4771         }
4772
4773 unmap_cfgtable:
4774         iounmap(cfgtable);
4775
4776 unmap_vaddr:
4777         iounmap(vaddr);
4778         return rc;
4779 }
4780
4781 static int cciss_init_reset_devices(struct pci_dev *pdev)
4782 {
4783         int rc, i;
4784
4785         if (!reset_devices)
4786                 return 0;
4787
4788         /* Reset the controller with a PCI power-cycle or via doorbell */
4789         rc = cciss_kdump_hard_reset_controller(pdev);
4790
4791         /* -ENOTSUPP here means we cannot reset the controller
4792          * but it's already (and still) up and running in
4793          * "performant mode".  Or, it might be 640x, which can't reset
4794          * due to concerns about shared bbwc between 6402/6404 pair.
4795          */
4796         if (rc == -ENOTSUPP)
4797                 return rc; /* just try to do the kdump anyhow. */
4798         if (rc)
4799                 return -ENODEV;
4800
4801         /* Now try to get the controller to respond to a no-op */
4802         dev_warn(&pdev->dev, "Waiting for controller to respond to no-op\n");
4803         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4804                 if (cciss_noop(pdev) == 0)
4805                         break;
4806                 else
4807                         dev_warn(&pdev->dev, "no-op failed%s\n",
4808                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4809                                         "; re-trying" : ""));
4810                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4811         }
4812         return 0;
4813 }
4814
4815 static int cciss_allocate_cmd_pool(ctlr_info_t *h)
4816 {
4817         h->cmd_pool_bits = kmalloc(BITS_TO_LONGS(h->nr_cmds) *
4818                 sizeof(unsigned long), GFP_KERNEL);
4819         h->cmd_pool = pci_alloc_consistent(h->pdev,
4820                 h->nr_cmds * sizeof(CommandList_struct),
4821                 &(h->cmd_pool_dhandle));
4822         h->errinfo_pool = pci_alloc_consistent(h->pdev,
4823                 h->nr_cmds * sizeof(ErrorInfo_struct),
4824                 &(h->errinfo_pool_dhandle));
4825         if ((h->cmd_pool_bits == NULL)
4826                 || (h->cmd_pool == NULL)
4827                 || (h->errinfo_pool == NULL)) {
4828                 dev_err(&h->pdev->dev, "out of memory");
4829                 return -ENOMEM;
4830         }
4831         return 0;
4832 }
4833
4834 static int cciss_allocate_scatterlists(ctlr_info_t *h)
4835 {
4836         int i;
4837
4838         /* zero it, so that on free we need not know how many were alloc'ed */
4839         h->scatter_list = kzalloc(h->max_commands *
4840                                 sizeof(struct scatterlist *), GFP_KERNEL);
4841         if (!h->scatter_list)
4842                 return -ENOMEM;
4843
4844         for (i = 0; i < h->nr_cmds; i++) {
4845                 h->scatter_list[i] = kmalloc(sizeof(struct scatterlist) *
4846                                                 h->maxsgentries, GFP_KERNEL);
4847                 if (h->scatter_list[i] == NULL) {
4848                         dev_err(&h->pdev->dev, "could not allocate "
4849                                 "s/g lists\n");
4850                         return -ENOMEM;
4851                 }
4852         }
4853         return 0;
4854 }
4855
4856 static void cciss_free_scatterlists(ctlr_info_t *h)
4857 {
4858         int i;
4859
4860         if (h->scatter_list) {
4861                 for (i = 0; i < h->nr_cmds; i++)
4862                         kfree(h->scatter_list[i]);
4863                 kfree(h->scatter_list);
4864         }
4865 }
4866
4867 static void cciss_free_cmd_pool(ctlr_info_t *h)
4868 {
4869         kfree(h->cmd_pool_bits);
4870         if (h->cmd_pool)
4871                 pci_free_consistent(h->pdev,
4872                         h->nr_cmds * sizeof(CommandList_struct),
4873                         h->cmd_pool, h->cmd_pool_dhandle);
4874         if (h->errinfo_pool)
4875                 pci_free_consistent(h->pdev,
4876                         h->nr_cmds * sizeof(ErrorInfo_struct),
4877                         h->errinfo_pool, h->errinfo_pool_dhandle);
4878 }
4879
4880 static int cciss_request_irq(ctlr_info_t *h,
4881         irqreturn_t (*msixhandler)(int, void *),
4882         irqreturn_t (*intxhandler)(int, void *))
4883 {
4884         if (h->msix_vector || h->msi_vector) {
4885                 if (!request_irq(h->intr[h->intr_mode], msixhandler,
4886                                 0, h->devname, h))
4887                         return 0;
4888                 dev_err(&h->pdev->dev, "Unable to get msi irq %d"
4889                         " for %s\n", h->intr[h->intr_mode],
4890                         h->devname);
4891                 return -1;
4892         }
4893
4894         if (!request_irq(h->intr[h->intr_mode], intxhandler,
4895                         IRQF_SHARED, h->devname, h))
4896                 return 0;
4897         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4898                 h->intr[h->intr_mode], h->devname);
4899         return -1;
4900 }
4901
4902 static int cciss_kdump_soft_reset(ctlr_info_t *h)
4903 {
4904         if (cciss_send_reset(h, CTLR_LUNID, CCISS_RESET_TYPE_CONTROLLER)) {
4905                 dev_warn(&h->pdev->dev, "Resetting array controller failed.\n");
4906                 return -EIO;
4907         }
4908
4909         dev_info(&h->pdev->dev, "Waiting for board to soft reset.\n");
4910         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_NOT_READY)) {
4911                 dev_warn(&h->pdev->dev, "Soft reset had no effect.\n");
4912                 return -1;
4913         }
4914
4915         dev_info(&h->pdev->dev, "Board reset, awaiting READY status.\n");
4916         if (cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY)) {
4917                 dev_warn(&h->pdev->dev, "Board failed to become ready "
4918                         "after soft reset.\n");
4919                 return -1;
4920         }
4921
4922         return 0;
4923 }
4924
4925 static void cciss_undo_allocations_after_kdump_soft_reset(ctlr_info_t *h)
4926 {
4927         int ctlr = h->ctlr;
4928
4929         free_irq(h->intr[h->intr_mode], h);
4930 #ifdef CONFIG_PCI_MSI
4931         if (h->msix_vector)
4932                 pci_disable_msix(h->pdev);
4933         else if (h->msi_vector)
4934                 pci_disable_msi(h->pdev);
4935 #endif /* CONFIG_PCI_MSI */
4936         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4937         cciss_free_scatterlists(h);
4938         cciss_free_cmd_pool(h);
4939         kfree(h->blockFetchTable);
4940         if (h->reply_pool)
4941                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
4942                                 h->reply_pool, h->reply_pool_dhandle);
4943         if (h->transtable)
4944                 iounmap(h->transtable);
4945         if (h->cfgtable)
4946                 iounmap(h->cfgtable);
4947         if (h->vaddr)
4948                 iounmap(h->vaddr);
4949         unregister_blkdev(h->major, h->devname);
4950         cciss_destroy_hba_sysfs_entry(h);
4951         pci_release_regions(h->pdev);
4952         kfree(h);
4953         hba[ctlr] = NULL;
4954 }
4955
4956 /*
4957  *  This is it.  Find all the controllers and register them.  I really hate
4958  *  stealing all these major device numbers.
4959  *  returns the number of block devices registered.
4960  */
4961 static int cciss_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
4962 {
4963         int i;
4964         int j = 0;
4965         int rc;
4966         int try_soft_reset = 0;
4967         int dac, return_code;
4968         InquiryData_struct *inq_buff;
4969         ctlr_info_t *h;
4970         unsigned long flags;
4971
4972         /*
4973          * By default the cciss driver is used for all older HP Smart Array
4974          * controllers. There are module paramaters that allow a user to
4975          * override this behavior and instead use the hpsa SCSI driver. If
4976          * this is the case cciss may be loaded first from the kdump initrd
4977          * image and cause a kernel panic. So if reset_devices is true and
4978          * cciss_allow_hpsa is set just bail.
4979          */
4980         if ((reset_devices) && (cciss_allow_hpsa == 1))
4981                 return -ENODEV;
4982         rc = cciss_init_reset_devices(pdev);
4983         if (rc) {
4984                 if (rc != -ENOTSUPP)
4985                         return rc;
4986                 /* If the reset fails in a particular way (it has no way to do
4987                  * a proper hard reset, so returns -ENOTSUPP) we can try to do
4988                  * a soft reset once we get the controller configured up to the
4989                  * point that it can accept a command.
4990                  */
4991                 try_soft_reset = 1;
4992                 rc = 0;
4993         }
4994
4995 reinit_after_soft_reset:
4996
4997         i = alloc_cciss_hba(pdev);
4998         if (i < 0)
4999                 return -1;
5000
5001         h = hba[i];
5002         h->pdev = pdev;
5003         h->busy_initializing = 1;
5004         h->intr_mode = cciss_simple_mode ? SIMPLE_MODE_INT : PERF_MODE_INT;
5005         INIT_LIST_HEAD(&h->cmpQ);
5006         INIT_LIST_HEAD(&h->reqQ);
5007         mutex_init(&h->busy_shutting_down);
5008
5009         if (cciss_pci_init(h) != 0)
5010                 goto clean_no_release_regions;
5011
5012         sprintf(h->devname, "cciss%d", i);
5013         h->ctlr = i;
5014
5015         if (cciss_tape_cmds < 2)
5016                 cciss_tape_cmds = 2;
5017         if (cciss_tape_cmds > 16)
5018                 cciss_tape_cmds = 16;
5019
5020         init_completion(&h->scan_wait);
5021
5022         if (cciss_create_hba_sysfs_entry(h))
5023                 goto clean0;
5024
5025         /* configure PCI DMA stuff */
5026         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
5027                 dac = 1;
5028         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
5029                 dac = 0;
5030         else {
5031                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
5032                 goto clean1;
5033         }
5034
5035         /*
5036          * register with the major number, or get a dynamic major number
5037          * by passing 0 as argument.  This is done for greater than
5038          * 8 controller support.
5039          */
5040         if (i < MAX_CTLR_ORIG)
5041                 h->major = COMPAQ_CISS_MAJOR + i;
5042         rc = register_blkdev(h->major, h->devname);
5043         if (rc == -EBUSY || rc == -EINVAL) {
5044                 dev_err(&h->pdev->dev,
5045                        "Unable to get major number %d for %s "
5046                        "on hba %d\n", h->major, h->devname, i);
5047                 goto clean1;
5048         } else {
5049                 if (i >= MAX_CTLR_ORIG)
5050                         h->major = rc;
5051         }
5052
5053         /* make sure the board interrupts are off */
5054         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5055         rc = cciss_request_irq(h, do_cciss_msix_intr, do_cciss_intx);
5056         if (rc)
5057                 goto clean2;
5058
5059         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
5060                h->devname, pdev->device, pci_name(pdev),
5061                h->intr[h->intr_mode], dac ? "" : " not");
5062
5063         if (cciss_allocate_cmd_pool(h))
5064                 goto clean4;
5065
5066         if (cciss_allocate_scatterlists(h))
5067                 goto clean4;
5068
5069         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
5070                 h->chainsize, h->nr_cmds);
5071         if (!h->cmd_sg_list && h->chainsize > 0)
5072                 goto clean4;
5073
5074         spin_lock_init(&h->lock);
5075
5076         /* Initialize the pdev driver private data.
5077            have it point to h.  */
5078         pci_set_drvdata(pdev, h);
5079         /* command and error info recs zeroed out before
5080            they are used */
5081         bitmap_zero(h->cmd_pool_bits, h->nr_cmds);
5082
5083         h->num_luns = 0;
5084         h->highest_lun = -1;
5085         for (j = 0; j < CISS_MAX_LUN; j++) {
5086                 h->drv[j] = NULL;
5087                 h->gendisk[j] = NULL;
5088         }
5089
5090         /* At this point, the controller is ready to take commands.
5091          * Now, if reset_devices and the hard reset didn't work, try
5092          * the soft reset and see if that works.
5093          */
5094         if (try_soft_reset) {
5095
5096                 /* This is kind of gross.  We may or may not get a completion
5097                  * from the soft reset command, and if we do, then the value
5098                  * from the fifo may or may not be valid.  So, we wait 10 secs
5099                  * after the reset throwing away any completions we get during
5100                  * that time.  Unregister the interrupt handler and register
5101                  * fake ones to scoop up any residual completions.
5102                  */
5103                 spin_lock_irqsave(&h->lock, flags);
5104                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5105                 spin_unlock_irqrestore(&h->lock, flags);
5106                 free_irq(h->intr[h->intr_mode], h);
5107                 rc = cciss_request_irq(h, cciss_msix_discard_completions,
5108                                         cciss_intx_discard_completions);
5109                 if (rc) {
5110                         dev_warn(&h->pdev->dev, "Failed to request_irq after "
5111                                 "soft reset.\n");
5112                         goto clean4;
5113                 }
5114
5115                 rc = cciss_kdump_soft_reset(h);
5116                 if (rc) {
5117                         dev_warn(&h->pdev->dev, "Soft reset failed.\n");
5118                         goto clean4;
5119                 }
5120
5121                 dev_info(&h->pdev->dev, "Board READY.\n");
5122                 dev_info(&h->pdev->dev,
5123                         "Waiting for stale completions to drain.\n");
5124                 h->access.set_intr_mask(h, CCISS_INTR_ON);
5125                 msleep(10000);
5126                 h->access.set_intr_mask(h, CCISS_INTR_OFF);
5127
5128                 rc = controller_reset_failed(h->cfgtable);
5129                 if (rc)
5130                         dev_info(&h->pdev->dev,
5131                                 "Soft reset appears to have failed.\n");
5132
5133                 /* since the controller's reset, we have to go back and re-init
5134                  * everything.  Easiest to just forget what we've done and do it
5135                  * all over again.
5136                  */
5137                 cciss_undo_allocations_after_kdump_soft_reset(h);
5138                 try_soft_reset = 0;
5139                 if (rc)
5140                         /* don't go to clean4, we already unallocated */
5141                         return -ENODEV;
5142
5143                 goto reinit_after_soft_reset;
5144         }
5145
5146         cciss_scsi_setup(h);
5147
5148         /* Turn the interrupts on so we can service requests */
5149         h->access.set_intr_mask(h, CCISS_INTR_ON);
5150
5151         /* Get the firmware version */
5152         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
5153         if (inq_buff == NULL) {
5154                 dev_err(&h->pdev->dev, "out of memory\n");
5155                 goto clean4;
5156         }
5157
5158         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
5159                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
5160         if (return_code == IO_OK) {
5161                 h->firm_ver[0] = inq_buff->data_byte[32];
5162                 h->firm_ver[1] = inq_buff->data_byte[33];
5163                 h->firm_ver[2] = inq_buff->data_byte[34];
5164                 h->firm_ver[3] = inq_buff->data_byte[35];
5165         } else {         /* send command failed */
5166                 dev_warn(&h->pdev->dev, "unable to determine firmware"
5167                         " version of controller\n");
5168         }
5169         kfree(inq_buff);
5170
5171         cciss_procinit(h);
5172
5173         h->cciss_max_sectors = 8192;
5174
5175         rebuild_lun_table(h, 1, 0);
5176         cciss_engage_scsi(h);
5177         h->busy_initializing = 0;
5178         return 1;
5179
5180 clean4:
5181         cciss_free_cmd_pool(h);
5182         cciss_free_scatterlists(h);
5183         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5184         free_irq(h->intr[h->intr_mode], h);
5185 clean2:
5186         unregister_blkdev(h->major, h->devname);
5187 clean1:
5188         cciss_destroy_hba_sysfs_entry(h);
5189 clean0:
5190         pci_release_regions(pdev);
5191 clean_no_release_regions:
5192         h->busy_initializing = 0;
5193
5194         /*
5195          * Deliberately omit pci_disable_device(): it does something nasty to
5196          * Smart Array controllers that pci_enable_device does not undo
5197          */
5198         pci_set_drvdata(pdev, NULL);
5199         free_hba(h);
5200         return -1;
5201 }
5202
5203 static void cciss_shutdown(struct pci_dev *pdev)
5204 {
5205         ctlr_info_t *h;
5206         char *flush_buf;
5207         int return_code;
5208
5209         h = pci_get_drvdata(pdev);
5210         flush_buf = kzalloc(4, GFP_KERNEL);
5211         if (!flush_buf) {
5212                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
5213                 return;
5214         }
5215         /* write all data in the battery backed cache to disk */
5216         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
5217                 4, 0, CTLR_LUNID, TYPE_CMD);
5218         kfree(flush_buf);
5219         if (return_code != IO_OK)
5220                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
5221         h->access.set_intr_mask(h, CCISS_INTR_OFF);
5222         free_irq(h->intr[h->intr_mode], h);
5223 }
5224
5225 static int cciss_enter_simple_mode(struct ctlr_info *h)
5226 {
5227         u32 trans_support;
5228
5229         trans_support = readl(&(h->cfgtable->TransportSupport));
5230         if (!(trans_support & SIMPLE_MODE))
5231                 return -ENOTSUPP;
5232
5233         h->max_commands = readl(&(h->cfgtable->CmdsOutMax));
5234         writel(CFGTBL_Trans_Simple, &(h->cfgtable->HostWrite.TransportRequest));
5235         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
5236         cciss_wait_for_mode_change_ack(h);
5237         print_cfg_table(h);
5238         if (!(readl(&(h->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
5239                 dev_warn(&h->pdev->dev, "unable to get board into simple mode\n");
5240                 return -ENODEV;
5241         }
5242         h->transMethod = CFGTBL_Trans_Simple;
5243         return 0;
5244 }
5245
5246
5247 static void cciss_remove_one(struct pci_dev *pdev)
5248 {
5249         ctlr_info_t *h;
5250         int i, j;
5251
5252         if (pci_get_drvdata(pdev) == NULL) {
5253                 dev_err(&pdev->dev, "Unable to remove device\n");
5254                 return;
5255         }
5256
5257         h = pci_get_drvdata(pdev);
5258         i = h->ctlr;
5259         if (hba[i] == NULL) {
5260                 dev_err(&pdev->dev, "device appears to already be removed\n");
5261                 return;
5262         }
5263
5264         mutex_lock(&h->busy_shutting_down);
5265
5266         remove_from_scan_list(h);
5267         remove_proc_entry(h->devname, proc_cciss);
5268         unregister_blkdev(h->major, h->devname);
5269
5270         /* remove it from the disk list */
5271         for (j = 0; j < CISS_MAX_LUN; j++) {
5272                 struct gendisk *disk = h->gendisk[j];
5273                 if (disk) {
5274                         struct request_queue *q = disk->queue;
5275
5276                         if (disk->flags & GENHD_FL_UP) {
5277                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
5278                                 del_gendisk(disk);
5279                         }
5280                         if (q)
5281                                 blk_cleanup_queue(q);
5282                 }
5283         }
5284
5285 #ifdef CONFIG_CISS_SCSI_TAPE
5286         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
5287 #endif
5288
5289         cciss_shutdown(pdev);
5290
5291 #ifdef CONFIG_PCI_MSI
5292         if (h->msix_vector)
5293                 pci_disable_msix(h->pdev);
5294         else if (h->msi_vector)
5295                 pci_disable_msi(h->pdev);
5296 #endif                          /* CONFIG_PCI_MSI */
5297
5298         iounmap(h->transtable);
5299         iounmap(h->cfgtable);
5300         iounmap(h->vaddr);
5301
5302         cciss_free_cmd_pool(h);
5303         /* Free up sg elements */
5304         for (j = 0; j < h->nr_cmds; j++)
5305                 kfree(h->scatter_list[j]);
5306         kfree(h->scatter_list);
5307         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
5308         kfree(h->blockFetchTable);
5309         if (h->reply_pool)
5310                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
5311                                 h->reply_pool, h->reply_pool_dhandle);
5312         /*
5313          * Deliberately omit pci_disable_device(): it does something nasty to
5314          * Smart Array controllers that pci_enable_device does not undo
5315          */
5316         pci_release_regions(pdev);
5317         pci_set_drvdata(pdev, NULL);
5318         cciss_destroy_hba_sysfs_entry(h);
5319         mutex_unlock(&h->busy_shutting_down);
5320         free_hba(h);
5321 }
5322
5323 static struct pci_driver cciss_pci_driver = {
5324         .name = "cciss",
5325         .probe = cciss_init_one,
5326         .remove = cciss_remove_one,
5327         .id_table = cciss_pci_device_id,        /* id_table */
5328         .shutdown = cciss_shutdown,
5329 };
5330
5331 /*
5332  *  This is it.  Register the PCI driver information for the cards we control
5333  *  the OS will call our registered routines when it finds one of our cards.
5334  */
5335 static int __init cciss_init(void)
5336 {
5337         int err;
5338
5339         /*
5340          * The hardware requires that commands are aligned on a 64-bit
5341          * boundary. Given that we use pci_alloc_consistent() to allocate an
5342          * array of them, the size must be a multiple of 8 bytes.
5343          */
5344         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
5345         printk(KERN_INFO DRIVER_NAME "\n");
5346
5347         err = bus_register(&cciss_bus_type);
5348         if (err)
5349                 return err;
5350
5351         /* Start the scan thread */
5352         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
5353         if (IS_ERR(cciss_scan_thread)) {
5354                 err = PTR_ERR(cciss_scan_thread);
5355                 goto err_bus_unregister;
5356         }
5357
5358         /* Register for our PCI devices */
5359         err = pci_register_driver(&cciss_pci_driver);
5360         if (err)
5361                 goto err_thread_stop;
5362
5363         return err;
5364
5365 err_thread_stop:
5366         kthread_stop(cciss_scan_thread);
5367 err_bus_unregister:
5368         bus_unregister(&cciss_bus_type);
5369
5370         return err;
5371 }
5372
5373 static void __exit cciss_cleanup(void)
5374 {
5375         int i;
5376
5377         pci_unregister_driver(&cciss_pci_driver);
5378         /* double check that all controller entrys have been removed */
5379         for (i = 0; i < MAX_CTLR; i++) {
5380                 if (hba[i] != NULL) {
5381                         dev_warn(&hba[i]->pdev->dev,
5382                                 "had to remove controller\n");
5383                         cciss_remove_one(hba[i]->pdev);
5384                 }
5385         }
5386         kthread_stop(cciss_scan_thread);
5387         if (proc_cciss)
5388                 remove_proc_entry("driver/cciss", NULL);
5389         bus_unregister(&cciss_bus_type);
5390 }
5391
5392 module_init(cciss_init);
5393 module_exit(cciss_cleanup);