drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/delay.h>
  30 #include <linux/major.h>
  31 #include <linux/fs.h>
  32 #include <linux/bio.h>
  33 #include <linux/blkpg.h>
  34 #include <linux/timer.h>
  35 #include <linux/proc_fs.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/init.h>
  38 #include <linux/jiffies.h>
  39 #include <linux/hdreg.h>
  40 #include <linux/spinlock.h>
  41 #include <linux/compat.h>
  42 #include <linux/mutex.h>
  43 #include <asm/uaccess.h>
  44 #include <asm/io.h>
  45
  46 #include <linux/dma-mapping.h>
  47 #include <linux/blkdev.h>
  48 #include <linux/genhd.h>
  49 #include <linux/completion.h>
  50 #include <scsi/scsi.h>
  51 #include <scsi/sg.h>
  52 #include <scsi/scsi_ioctl.h>
  53 #include <linux/cdrom.h>
  54 #include <linux/scatterlist.h>
  55 #include <linux/kthread.h>
  56
  57 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  58 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  59 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  60
  61 /* Embedded module documentation macros - see modules.h */
  62 MODULE_AUTHOR("Hewlett-Packard Company");
  63 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  64 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  65 MODULE_VERSION("3.6.26");
  66 MODULE_LICENSE("GPL");
  67
  68 static DEFINE_MUTEX(cciss_mutex);
  69 static struct proc_dir_entry *proc_cciss;
  70
  71 #include "cciss_cmd.h"
  72 #include "cciss.h"
  73 #include <linux/cciss_ioctl.h>
  74
  75 /* define the PCI info for the cards we can control */
  76 static const struct pci_device_id cciss_pci_device_id[] = {
  77         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  78         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  79         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  80         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  86         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  87         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  88         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  89         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
  97         {0,}
  98 };
  99
 100 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 101
 102 /*  board_id = Subsystem Device ID & Vendor ID
 103  *  product = Marketing Name for the board
 104  *  access = Address of the struct of function pointers
 105  */
 106 static struct board_type products[] = {
 107         {0x40700E11, "Smart Array 5300", &SA5_access},
 108         {0x40800E11, "Smart Array 5i", &SA5B_access},
 109         {0x40820E11, "Smart Array 532", &SA5B_access},
 110         {0x40830E11, "Smart Array 5312", &SA5B_access},
 111         {0x409A0E11, "Smart Array 641", &SA5_access},
 112         {0x409B0E11, "Smart Array 642", &SA5_access},
 113         {0x409C0E11, "Smart Array 6400", &SA5_access},
 114         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 115         {0x40910E11, "Smart Array 6i", &SA5_access},
 116         {0x3225103C, "Smart Array P600", &SA5_access},
 117         {0x3223103C, "Smart Array P800", &SA5_access},
 118         {0x3234103C, "Smart Array P400", &SA5_access},
 119         {0x3235103C, "Smart Array P400i", &SA5_access},
 120         {0x3211103C, "Smart Array E200i", &SA5_access},
 121         {0x3212103C, "Smart Array E200", &SA5_access},
 122         {0x3213103C, "Smart Array E200i", &SA5_access},
 123         {0x3214103C, "Smart Array E200i", &SA5_access},
 124         {0x3215103C, "Smart Array E200i", &SA5_access},
 125         {0x3237103C, "Smart Array E500", &SA5_access},
 126         {0x3223103C, "Smart Array P800", &SA5_access},
 127         {0x3234103C, "Smart Array P400", &SA5_access},
 128         {0x323D103C, "Smart Array P700m", &SA5_access},
 129 };
 130
 131 /* How long to wait (in milliseconds) for board to go into simple mode */
 132 #define MAX_CONFIG_WAIT 30000
 133 #define MAX_IOCTL_CONFIG_WAIT 1000
 134
 135 /*define how many times we will try a command because of bus resets */
 136 #define MAX_CMD_RETRIES 3
 137
 138 #define MAX_CTLR        32
 139
 140 /* Originally cciss driver only supports 8 major numbers */
 141 #define MAX_CTLR_ORIG   8
 142
 143 static ctlr_info_t *hba[MAX_CTLR];
 144
 145 static struct task_struct *cciss_scan_thread;
 146 static DEFINE_MUTEX(scan_mutex);
 147 static LIST_HEAD(scan_q);
 148
 149 static void do_cciss_request(struct request_queue *q);
 150 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 151 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 152 static int cciss_open(struct block_device *bdev, fmode_t mode);
 153 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 154 static int cciss_release(struct gendisk *disk, fmode_t mode);
 155 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 156                     unsigned int cmd, unsigned long arg);
 157 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 158                        unsigned int cmd, unsigned long arg);
 159 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 160
 161 static int cciss_revalidate(struct gendisk *disk);
 162 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 163 static int deregister_disk(ctlr_info_t *h, int drv_index,
 164                            int clear_all, int via_ioctl);
 165
 166 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 167                         sector_t *total_size, unsigned int *block_size);
 168 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 169                         sector_t *total_size, unsigned int *block_size);
 170 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 171                         sector_t total_size,
 172                         unsigned int block_size, InquiryData_struct *inq_buff,
 173                                    drive_info_struct *drv);
 174 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
 175 static void start_io(ctlr_info_t *h);
 176 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 177                         __u8 page_code, unsigned char scsi3addr[],
 178                         int cmd_type);
 179 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 180         int attempt_retry);
 181 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 182
 183 static int add_to_scan_list(struct ctlr_info *h);
 184 static int scan_thread(void *data);
 185 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 186 static void cciss_hba_release(struct device *dev);
 187 static void cciss_device_release(struct device *dev);
 188 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 189 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 190 static inline u32 next_command(ctlr_info_t *h);
 191 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
 192         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 193         u64 *cfg_offset);
 194 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 195         unsigned long *memory_bar);
 196 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag);
 197 static __devinit int write_driver_ver_to_cfgtable(
 198         CfgTable_struct __iomem *cfgtable);
 199
 200 /* performant mode helper functions */
 201 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 202                                 int *bucket_map);
 203 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 204
 205 #ifdef CONFIG_PROC_FS
 206 static void cciss_procinit(ctlr_info_t *h);
 207 #else
 208 static void cciss_procinit(ctlr_info_t *h)
 209 {
 210 }
 211 #endif                          /* CONFIG_PROC_FS */
 212
 213 #ifdef CONFIG_COMPAT
 214 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 215                               unsigned, unsigned long);
 216 #endif
 217
 218 static const struct block_device_operations cciss_fops = {
 219         .owner = THIS_MODULE,
 220         .open = cciss_unlocked_open,
 221         .release = cciss_release,
 222         .ioctl = do_ioctl,
 223         .getgeo = cciss_getgeo,
 224 #ifdef CONFIG_COMPAT
 225         .compat_ioctl = cciss_compat_ioctl,
 226 #endif
 227         .revalidate_disk = cciss_revalidate,
 228 };
 229
 230 /* set_performant_mode: Modify the tag for cciss performant
 231  * set bit 0 for pull model, bits 3-1 for block fetch
 232  * register number
 233  */
 234 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 235 {
 236         if (likely(h->transMethod & CFGTBL_Trans_Performant))
 237                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 238 }
 239
 240 /*
 241  * Enqueuing and dequeuing functions for cmdlists.
 242  */
 243 static inline void addQ(struct list_head *list, CommandList_struct *c)
 244 {
 245         list_add_tail(&c->list, list);
 246 }
 247
 248 static inline void removeQ(CommandList_struct *c)
 249 {
 250         /*
 251          * After kexec/dump some commands might still
 252          * be in flight, which the firmware will try
 253          * to complete. Resetting the firmware doesn't work
 254          * with old fw revisions, so we have to mark
 255          * them off as 'stale' to prevent the driver from
 256          * falling over.
 257          */
 258         if (WARN_ON(list_empty(&c->list))) {
 259                 c->cmd_type = CMD_MSG_STALE;
 260                 return;
 261         }
 262
 263         list_del_init(&c->list);
 264 }
 265
 266 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 267         CommandList_struct *c)
 268 {
 269         unsigned long flags;
 270         set_performant_mode(h, c);
 271         spin_lock_irqsave(&h->lock, flags);
 272         addQ(&h->reqQ, c);
 273         h->Qdepth++;
 274         if (h->Qdepth > h->maxQsinceinit)
 275                 h->maxQsinceinit = h->Qdepth;
 276         start_io(h);
 277         spin_unlock_irqrestore(&h->lock, flags);
 278 }
 279
 280 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 281         int nr_cmds)
 282 {
 283         int i;
 284
 285         if (!cmd_sg_list)
 286                 return;
 287         for (i = 0; i < nr_cmds; i++) {
 288                 kfree(cmd_sg_list[i]);
 289                 cmd_sg_list[i] = NULL;
 290         }
 291         kfree(cmd_sg_list);
 292 }
 293
 294 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 295         ctlr_info_t *h, int chainsize, int nr_cmds)
 296 {
 297         int j;
 298         SGDescriptor_struct **cmd_sg_list;
 299
 300         if (chainsize <= 0)
 301                 return NULL;
 302
 303         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 304         if (!cmd_sg_list)
 305                 return NULL;
 306
 307         /* Build up chain blocks for each command */
 308         for (j = 0; j < nr_cmds; j++) {
 309                 /* Need a block of chainsized s/g elements. */
 310                 cmd_sg_list[j] = kmalloc((chainsize *
 311                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 312                 if (!cmd_sg_list[j]) {
 313                         dev_err(&h->pdev->dev, "Cannot get memory "
 314                                 "for s/g chains.\n");
 315                         goto clean;
 316                 }
 317         }
 318         return cmd_sg_list;
 319 clean:
 320         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 321         return NULL;
 322 }
 323
 324 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 325 {
 326         SGDescriptor_struct *chain_sg;
 327         u64bit temp64;
 328
 329         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 330                 return;
 331
 332         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 333         temp64.val32.lower = chain_sg->Addr.lower;
 334         temp64.val32.upper = chain_sg->Addr.upper;
 335         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 336 }
 337
 338 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 339         SGDescriptor_struct *chain_block, int len)
 340 {
 341         SGDescriptor_struct *chain_sg;
 342         u64bit temp64;
 343
 344         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 345         chain_sg->Ext = CCISS_SG_CHAIN;
 346         chain_sg->Len = len;
 347         temp64.val = pci_map_single(h->pdev, chain_block, len,
 348                                 PCI_DMA_TODEVICE);
 349         chain_sg->Addr.lower = temp64.val32.lower;
 350         chain_sg->Addr.upper = temp64.val32.upper;
 351 }
 352
 353 #include "cciss_scsi.c"         /* For SCSI tape support */
 354
 355 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 356         "UNKNOWN"
 357 };
 358 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 359
 360 #ifdef CONFIG_PROC_FS
 361
 362 /*
 363  * Report information about this controller.
 364  */
 365 #define ENG_GIG 1000000000
 366 #define ENG_GIG_FACTOR (ENG_GIG/512)
 367 #define ENGAGE_SCSI     "engage scsi"
 368
 369 static void cciss_seq_show_header(struct seq_file *seq)
 370 {
 371         ctlr_info_t *h = seq->private;
 372
 373         seq_printf(seq, "%s: HP %s Controller\n"
 374                 "Board ID: 0x%08lx\n"
 375                 "Firmware Version: %c%c%c%c\n"
 376                 "IRQ: %d\n"
 377                 "Logical drives: %d\n"
 378                 "Current Q depth: %d\n"
 379                 "Current # commands on controller: %d\n"
 380                 "Max Q depth since init: %d\n"
 381                 "Max # commands on controller since init: %d\n"
 382                 "Max SG entries since init: %d\n",
 383                 h->devname,
 384                 h->product_name,
 385                 (unsigned long)h->board_id,
 386                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 387                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 388                 h->num_luns,
 389                 h->Qdepth, h->commands_outstanding,
 390                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 391
 392 #ifdef CONFIG_CISS_SCSI_TAPE
 393         cciss_seq_tape_report(seq, h);
 394 #endif /* CONFIG_CISS_SCSI_TAPE */
 395 }
 396
 397 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 398 {
 399         ctlr_info_t *h = seq->private;
 400         unsigned long flags;
 401
 402         /* prevent displaying bogus info during configuration
 403          * or deconfiguration of a logical volume
 404          */
 405         spin_lock_irqsave(&h->lock, flags);
 406         if (h->busy_configuring) {
 407                 spin_unlock_irqrestore(&h->lock, flags);
 408                 return ERR_PTR(-EBUSY);
 409         }
 410         h->busy_configuring = 1;
 411         spin_unlock_irqrestore(&h->lock, flags);
 412
 413         if (*pos == 0)
 414                 cciss_seq_show_header(seq);
 415
 416         return pos;
 417 }
 418
 419 static int cciss_seq_show(struct seq_file *seq, void *v)
 420 {
 421         sector_t vol_sz, vol_sz_frac;
 422         ctlr_info_t *h = seq->private;
 423         unsigned ctlr = h->ctlr;
 424         loff_t *pos = v;
 425         drive_info_struct *drv = h->drv[*pos];
 426
 427         if (*pos > h->highest_lun)
 428                 return 0;
 429
 430         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 431                 return 0;
 432
 433         if (drv->heads == 0)
 434                 return 0;
 435
 436         vol_sz = drv->nr_blocks;
 437         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 438         vol_sz_frac *= 100;
 439         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 440
 441         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 442                 drv->raid_level = RAID_UNKNOWN;
 443         seq_printf(seq, "cciss/c%dd%d:"
 444                         "\t%4u.%02uGB\tRAID %s\n",
 445                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 446                         raid_label[drv->raid_level]);
 447         return 0;
 448 }
 449
 450 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 451 {
 452         ctlr_info_t *h = seq->private;
 453
 454         if (*pos > h->highest_lun)
 455                 return NULL;
 456         *pos += 1;
 457
 458         return pos;
 459 }
 460
 461 static void cciss_seq_stop(struct seq_file *seq, void *v)
 462 {
 463         ctlr_info_t *h = seq->private;
 464
 465         /* Only reset h->busy_configuring if we succeeded in setting
 466          * it during cciss_seq_start. */
 467         if (v == ERR_PTR(-EBUSY))
 468                 return;
 469
 470         h->busy_configuring = 0;
 471 }
 472
 473 static const struct seq_operations cciss_seq_ops = {
 474         .start = cciss_seq_start,
 475         .show  = cciss_seq_show,
 476         .next  = cciss_seq_next,
 477         .stop  = cciss_seq_stop,
 478 };
 479
 480 static int cciss_seq_open(struct inode *inode, struct file *file)
 481 {
 482         int ret = seq_open(file, &cciss_seq_ops);
 483         struct seq_file *seq = file->private_data;
 484
 485         if (!ret)
 486                 seq->private = PDE(inode)->data;
 487
 488         return ret;
 489 }
 490
 491 static ssize_t
 492 cciss_proc_write(struct file *file, const char __user *buf,
 493                  size_t length, loff_t *ppos)
 494 {
 495         int err;
 496         char *buffer;
 497
 498 #ifndef CONFIG_CISS_SCSI_TAPE
 499         return -EINVAL;
 500 #endif
 501
 502         if (!buf || length > PAGE_SIZE - 1)
 503                 return -EINVAL;
 504
 505         buffer = (char *)__get_free_page(GFP_KERNEL);
 506         if (!buffer)
 507                 return -ENOMEM;
 508
 509         err = -EFAULT;
 510         if (copy_from_user(buffer, buf, length))
 511                 goto out;
 512         buffer[length] = '\0';
 513
 514 #ifdef CONFIG_CISS_SCSI_TAPE
 515         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 516                 struct seq_file *seq = file->private_data;
 517                 ctlr_info_t *h = seq->private;
 518
 519                 err = cciss_engage_scsi(h);
 520                 if (err == 0)
 521                         err = length;
 522         } else
 523 #endif /* CONFIG_CISS_SCSI_TAPE */
 524                 err = -EINVAL;
 525         /* might be nice to have "disengage" too, but it's not
 526            safely possible. (only 1 module use count, lock issues.) */
 527
 528 out:
 529         free_page((unsigned long)buffer);
 530         return err;
 531 }
 532
 533 static const struct file_operations cciss_proc_fops = {
 534         .owner   = THIS_MODULE,
 535         .open    = cciss_seq_open,
 536         .read    = seq_read,
 537         .llseek  = seq_lseek,
 538         .release = seq_release,
 539         .write   = cciss_proc_write,
 540 };
 541
 542 static void __devinit cciss_procinit(ctlr_info_t *h)
 543 {
 544         struct proc_dir_entry *pde;
 545
 546         if (proc_cciss == NULL)
 547                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 548         if (!proc_cciss)
 549                 return;
 550         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 551                                         S_IROTH, proc_cciss,
 552                                         &cciss_proc_fops, h);
 553 }
 554 #endif                          /* CONFIG_PROC_FS */
 555
 556 #define MAX_PRODUCT_NAME_LEN 19
 557
 558 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 559 #define to_drv(n) container_of(n, drive_info_struct, dev)
 560
 561 /* List of controllers which cannot be reset on kexec with reset_devices */
 562 static u32 unresettable_controller[] = {
 563         0x324a103C, /* Smart Array P712m */
 564         0x324b103C, /* SmartArray P711m */
 565         0x3223103C, /* Smart Array P800 */
 566         0x3234103C, /* Smart Array P400 */
 567         0x3235103C, /* Smart Array P400i */
 568         0x3211103C, /* Smart Array E200i */
 569         0x3212103C, /* Smart Array E200 */
 570         0x3213103C, /* Smart Array E200i */
 571         0x3214103C, /* Smart Array E200i */
 572         0x3215103C, /* Smart Array E200i */
 573         0x3237103C, /* Smart Array E500 */
 574         0x323D103C, /* Smart Array P700m */
 575         0x409C0E11, /* Smart Array 6400 */
 576         0x409D0E11, /* Smart Array 6400 EM */
 577 };
 578
 579 static int ctlr_is_resettable(struct ctlr_info *h)
 580 {
 581         int i;
 582
 583         for (i = 0; i < ARRAY_SIZE(unresettable_controller); i++)
 584                 if (unresettable_controller[i] == h->board_id)
 585                         return 0;
 586         return 1;
 587 }
 588
 589 static ssize_t host_show_resettable(struct device *dev,
 590                                     struct device_attribute *attr,
 591                                     char *buf)
 592 {
 593         struct ctlr_info *h = to_hba(dev);
 594
 595         return snprintf(buf, 20, "%d\n", ctlr_is_resettable(h));
 596 }
 597 static DEVICE_ATTR(resettable, S_IRUGO, host_show_resettable, NULL);
 598
 599 static ssize_t host_store_rescan(struct device *dev,
 600                                  struct device_attribute *attr,
 601                                  const char *buf, size_t count)
 602 {
 603         struct ctlr_info *h = to_hba(dev);
 604
 605         add_to_scan_list(h);
 606         wake_up_process(cciss_scan_thread);
 607         wait_for_completion_interruptible(&h->scan_wait);
 608
 609         return count;
 610 }
 611 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 612
 613 static ssize_t dev_show_unique_id(struct device *dev,
 614                                  struct device_attribute *attr,
 615                                  char *buf)
 616 {
 617         drive_info_struct *drv = to_drv(dev);
 618         struct ctlr_info *h = to_hba(drv->dev.parent);
 619         __u8 sn[16];
 620         unsigned long flags;
 621         int ret = 0;
 622
 623         spin_lock_irqsave(&h->lock, flags);
 624         if (h->busy_configuring)
 625                 ret = -EBUSY;
 626         else
 627                 memcpy(sn, drv->serial_no, sizeof(sn));
 628         spin_unlock_irqrestore(&h->lock, flags);
 629
 630         if (ret)
 631                 return ret;
 632         else
 633                 return snprintf(buf, 16 * 2 + 2,
 634                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 635                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 636                                 sn[0], sn[1], sn[2], sn[3],
 637                                 sn[4], sn[5], sn[6], sn[7],
 638                                 sn[8], sn[9], sn[10], sn[11],
 639                                 sn[12], sn[13], sn[14], sn[15]);
 640 }
 641 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 642
 643 static ssize_t dev_show_vendor(struct device *dev,
 644                                struct device_attribute *attr,
 645                                char *buf)
 646 {
 647         drive_info_struct *drv = to_drv(dev);
 648         struct ctlr_info *h = to_hba(drv->dev.parent);
 649         char vendor[VENDOR_LEN + 1];
 650         unsigned long flags;
 651         int ret = 0;
 652
 653         spin_lock_irqsave(&h->lock, flags);
 654         if (h->busy_configuring)
 655                 ret = -EBUSY;
 656         else
 657                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 658         spin_unlock_irqrestore(&h->lock, flags);
 659
 660         if (ret)
 661                 return ret;
 662         else
 663                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 664 }
 665 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 666
 667 static ssize_t dev_show_model(struct device *dev,
 668                               struct device_attribute *attr,
 669                               char *buf)
 670 {
 671         drive_info_struct *drv = to_drv(dev);
 672         struct ctlr_info *h = to_hba(drv->dev.parent);
 673         char model[MODEL_LEN + 1];
 674         unsigned long flags;
 675         int ret = 0;
 676
 677         spin_lock_irqsave(&h->lock, flags);
 678         if (h->busy_configuring)
 679                 ret = -EBUSY;
 680         else
 681                 memcpy(model, drv->model, MODEL_LEN + 1);
 682         spin_unlock_irqrestore(&h->lock, flags);
 683
 684         if (ret)
 685                 return ret;
 686         else
 687                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 688 }
 689 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 690
 691 static ssize_t dev_show_rev(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 rev[REV_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(rev, drv->rev, REV_LEN + 1);
 706         spin_unlock_irqrestore(&h->lock, flags);
 707
 708         if (ret)
 709                 return ret;
 710         else
 711                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 712 }
 713 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 714
 715 static ssize_t cciss_show_lunid(struct device *dev,
 716                                 struct device_attribute *attr, char *buf)
 717 {
 718         drive_info_struct *drv = to_drv(dev);
 719         struct ctlr_info *h = to_hba(drv->dev.parent);
 720         unsigned long flags;
 721         unsigned char lunid[8];
 722
 723         spin_lock_irqsave(&h->lock, flags);
 724         if (h->busy_configuring) {
 725                 spin_unlock_irqrestore(&h->lock, flags);
 726                 return -EBUSY;
 727         }
 728         if (!drv->heads) {
 729                 spin_unlock_irqrestore(&h->lock, flags);
 730                 return -ENOTTY;
 731         }
 732         memcpy(lunid, drv->LunID, sizeof(lunid));
 733         spin_unlock_irqrestore(&h->lock, flags);
 734         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 735                 lunid[0], lunid[1], lunid[2], lunid[3],
 736                 lunid[4], lunid[5], lunid[6], lunid[7]);
 737 }
 738 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 739
 740 static ssize_t cciss_show_raid_level(struct device *dev,
 741                                      struct device_attribute *attr, char *buf)
 742 {
 743         drive_info_struct *drv = to_drv(dev);
 744         struct ctlr_info *h = to_hba(drv->dev.parent);
 745         int raid;
 746         unsigned long flags;
 747
 748         spin_lock_irqsave(&h->lock, flags);
 749         if (h->busy_configuring) {
 750                 spin_unlock_irqrestore(&h->lock, flags);
 751                 return -EBUSY;
 752         }
 753         raid = drv->raid_level;
 754         spin_unlock_irqrestore(&h->lock, flags);
 755         if (raid < 0 || raid > RAID_UNKNOWN)
 756                 raid = RAID_UNKNOWN;
 757
 758         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 759                         raid_label[raid]);
 760 }
 761 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 762
 763 static ssize_t cciss_show_usage_count(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         int count;
 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         count = drv->usage_count;
 777         spin_unlock_irqrestore(&h->lock, flags);
 778         return snprintf(buf, 20, "%d\n", count);
 779 }
 780 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 781
 782 static struct attribute *cciss_host_attrs[] = {
 783         &dev_attr_rescan.attr,
 784         &dev_attr_resettable.attr,
 785         NULL
 786 };
 787
 788 static struct attribute_group cciss_host_attr_group = {
 789         .attrs = cciss_host_attrs,
 790 };
 791
 792 static const struct attribute_group *cciss_host_attr_groups[] = {
 793         &cciss_host_attr_group,
 794         NULL
 795 };
 796
 797 static struct device_type cciss_host_type = {
 798         .name           = "cciss_host",
 799         .groups         = cciss_host_attr_groups,
 800         .release        = cciss_hba_release,
 801 };
 802
 803 static struct attribute *cciss_dev_attrs[] = {
 804         &dev_attr_unique_id.attr,
 805         &dev_attr_model.attr,
 806         &dev_attr_vendor.attr,
 807         &dev_attr_rev.attr,
 808         &dev_attr_lunid.attr,
 809         &dev_attr_raid_level.attr,
 810         &dev_attr_usage_count.attr,
 811         NULL
 812 };
 813
 814 static struct attribute_group cciss_dev_attr_group = {
 815         .attrs = cciss_dev_attrs,
 816 };
 817
 818 static const struct attribute_group *cciss_dev_attr_groups[] = {
 819         &cciss_dev_attr_group,
 820         NULL
 821 };
 822
 823 static struct device_type cciss_dev_type = {
 824         .name           = "cciss_device",
 825         .groups         = cciss_dev_attr_groups,
 826         .release        = cciss_device_release,
 827 };
 828
 829 static struct bus_type cciss_bus_type = {
 830         .name           = "cciss",
 831 };
 832
 833 /*
 834  * cciss_hba_release is called when the reference count
 835  * of h->dev goes to zero.
 836  */
 837 static void cciss_hba_release(struct device *dev)
 838 {
 839         /*
 840          * nothing to do, but need this to avoid a warning
 841          * about not having a release handler from lib/kref.c.
 842          */
 843 }
 844
 845 /*
 846  * Initialize sysfs entry for each controller.  This sets up and registers
 847  * the 'cciss#' directory for each individual controller under
 848  * /sys/bus/pci/devices/<dev>/.
 849  */
 850 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 851 {
 852         device_initialize(&h->dev);
 853         h->dev.type = &cciss_host_type;
 854         h->dev.bus = &cciss_bus_type;
 855         dev_set_name(&h->dev, "%s", h->devname);
 856         h->dev.parent = &h->pdev->dev;
 857
 858         return device_add(&h->dev);
 859 }
 860
 861 /*
 862  * Remove sysfs entries for an hba.
 863  */
 864 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 865 {
 866         device_del(&h->dev);
 867         put_device(&h->dev); /* final put. */
 868 }
 869
 870 /* cciss_device_release is called when the reference count
 871  * of h->drv[x]dev goes to zero.
 872  */
 873 static void cciss_device_release(struct device *dev)
 874 {
 875         drive_info_struct *drv = to_drv(dev);
 876         kfree(drv);
 877 }
 878
 879 /*
 880  * Initialize sysfs for each logical drive.  This sets up and registers
 881  * the 'c#d#' directory for each individual logical drive under
 882  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 883  * /sys/block/cciss!c#d# to this entry.
 884  */
 885 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 886                                        int drv_index)
 887 {
 888         struct device *dev;
 889
 890         if (h->drv[drv_index]->device_initialized)
 891                 return 0;
 892
 893         dev = &h->drv[drv_index]->dev;
 894         device_initialize(dev);
 895         dev->type = &cciss_dev_type;
 896         dev->bus = &cciss_bus_type;
 897         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 898         dev->parent = &h->dev;
 899         h->drv[drv_index]->device_initialized = 1;
 900         return device_add(dev);
 901 }
 902
 903 /*
 904  * Remove sysfs entries for a logical drive.
 905  */
 906 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 907         int ctlr_exiting)
 908 {
 909         struct device *dev = &h->drv[drv_index]->dev;
 910
 911         /* special case for c*d0, we only destroy it on controller exit */
 912         if (drv_index == 0 && !ctlr_exiting)
 913                 return;
 914
 915         device_del(dev);
 916         put_device(dev); /* the "final" put. */
 917         h->drv[drv_index] = NULL;
 918 }
 919
 920 /*
 921  * For operations that cannot sleep, a command block is allocated at init,
 922  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 923  * which ones are free or in use.
 924  */
 925 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 926 {
 927         CommandList_struct *c;
 928         int i;
 929         u64bit temp64;
 930         dma_addr_t cmd_dma_handle, err_dma_handle;
 931
 932         do {
 933                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 934                 if (i == h->nr_cmds)
 935                         return NULL;
 936         } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
 937                   h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 938         c = h->cmd_pool + i;
 939         memset(c, 0, sizeof(CommandList_struct));
 940         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
 941         c->err_info = h->errinfo_pool + i;
 942         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 943         err_dma_handle = h->errinfo_pool_dhandle
 944             + i * sizeof(ErrorInfo_struct);
 945         h->nr_allocs++;
 946
 947         c->cmdindex = i;
 948
 949         INIT_LIST_HEAD(&c->list);
 950         c->busaddr = (__u32) cmd_dma_handle;
 951         temp64.val = (__u64) err_dma_handle;
 952         c->ErrDesc.Addr.lower = temp64.val32.lower;
 953         c->ErrDesc.Addr.upper = temp64.val32.upper;
 954         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 955
 956         c->ctlr = h->ctlr;
 957         return c;
 958 }
 959
 960 /* allocate a command using pci_alloc_consistent, used for ioctls,
 961  * etc., not for the main i/o path.
 962  */
 963 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
 964 {
 965         CommandList_struct *c;
 966         u64bit temp64;
 967         dma_addr_t cmd_dma_handle, err_dma_handle;
 968
 969         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 970                 sizeof(CommandList_struct), &cmd_dma_handle);
 971         if (c == NULL)
 972                 return NULL;
 973         memset(c, 0, sizeof(CommandList_struct));
 974
 975         c->cmdindex = -1;
 976
 977         c->err_info = (ErrorInfo_struct *)
 978             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 979                     &err_dma_handle);
 980
 981         if (c->err_info == NULL) {
 982                 pci_free_consistent(h->pdev,
 983                         sizeof(CommandList_struct), c, cmd_dma_handle);
 984                 return NULL;
 985         }
 986         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 987
 988         INIT_LIST_HEAD(&c->list);
 989         c->busaddr = (__u32) cmd_dma_handle;
 990         temp64.val = (__u64) err_dma_handle;
 991         c->ErrDesc.Addr.lower = temp64.val32.lower;
 992         c->ErrDesc.Addr.upper = temp64.val32.upper;
 993         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 994
 995         c->ctlr = h->ctlr;
 996         return c;
 997 }
 998
 999 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
1000 {
1001         int i;
1002
1003         i = c - h->cmd_pool;
1004         clear_bit(i & (BITS_PER_LONG - 1),
1005                   h->cmd_pool_bits + (i / BITS_PER_LONG));
1006         h->nr_frees++;
1007 }
1008
1009 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
1010 {
1011         u64bit temp64;
1012
1013         temp64.val32.lower = c->ErrDesc.Addr.lower;
1014         temp64.val32.upper = c->ErrDesc.Addr.upper;
1015         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1016                             c->err_info, (dma_addr_t) temp64.val);
1017         pci_free_consistent(h->pdev, sizeof(CommandList_struct), c,
1018                 (dma_addr_t) cciss_tag_discard_error_bits(h, (u32) c->busaddr));
1019 }
1020
1021 static inline ctlr_info_t *get_host(struct gendisk *disk)
1022 {
1023         return disk->queue->queuedata;
1024 }
1025
1026 static inline drive_info_struct *get_drv(struct gendisk *disk)
1027 {
1028         return disk->private_data;
1029 }
1030
1031 /*
1032  * Open.  Make sure the device is really there.
1033  */
1034 static int cciss_open(struct block_device *bdev, fmode_t mode)
1035 {
1036         ctlr_info_t *h = get_host(bdev->bd_disk);
1037         drive_info_struct *drv = get_drv(bdev->bd_disk);
1038
1039         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1040         if (drv->busy_configuring)
1041                 return -EBUSY;
1042         /*
1043          * Root is allowed to open raw volume zero even if it's not configured
1044          * so array config can still work. Root is also allowed to open any
1045          * volume that has a LUN ID, so it can issue IOCTL to reread the
1046          * disk information.  I don't think I really like this
1047          * but I'm already using way to many device nodes to claim another one
1048          * for "raw controller".
1049          */
1050         if (drv->heads == 0) {
1051                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1052                         /* if not node 0 make sure it is a partition = 0 */
1053                         if (MINOR(bdev->bd_dev) & 0x0f) {
1054                                 return -ENXIO;
1055                                 /* if it is, make sure we have a LUN ID */
1056                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1057                                 sizeof(drv->LunID))) {
1058                                 return -ENXIO;
1059                         }
1060                 }
1061                 if (!capable(CAP_SYS_ADMIN))
1062                         return -EPERM;
1063         }
1064         drv->usage_count++;
1065         h->usage_count++;
1066         return 0;
1067 }
1068
1069 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1070 {
1071         int ret;
1072
1073         mutex_lock(&cciss_mutex);
1074         ret = cciss_open(bdev, mode);
1075         mutex_unlock(&cciss_mutex);
1076
1077         return ret;
1078 }
1079
1080 /*
1081  * Close.  Sync first.
1082  */
1083 static int cciss_release(struct gendisk *disk, fmode_t mode)
1084 {
1085         ctlr_info_t *h;
1086         drive_info_struct *drv;
1087
1088         mutex_lock(&cciss_mutex);
1089         h = get_host(disk);
1090         drv = get_drv(disk);
1091         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1092         drv->usage_count--;
1093         h->usage_count--;
1094         mutex_unlock(&cciss_mutex);
1095         return 0;
1096 }
1097
1098 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1099                     unsigned cmd, unsigned long arg)
1100 {
1101         int ret;
1102         mutex_lock(&cciss_mutex);
1103         ret = cciss_ioctl(bdev, mode, cmd, arg);
1104         mutex_unlock(&cciss_mutex);
1105         return ret;
1106 }
1107
1108 #ifdef CONFIG_COMPAT
1109
1110 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1111                                   unsigned cmd, unsigned long arg);
1112 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1113                                       unsigned cmd, unsigned long arg);
1114
1115 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1116                               unsigned cmd, unsigned long arg)
1117 {
1118         switch (cmd) {
1119         case CCISS_GETPCIINFO:
1120         case CCISS_GETINTINFO:
1121         case CCISS_SETINTINFO:
1122         case CCISS_GETNODENAME:
1123         case CCISS_SETNODENAME:
1124         case CCISS_GETHEARTBEAT:
1125         case CCISS_GETBUSTYPES:
1126         case CCISS_GETFIRMVER:
1127         case CCISS_GETDRIVVER:
1128         case CCISS_REVALIDVOLS:
1129         case CCISS_DEREGDISK:
1130         case CCISS_REGNEWDISK:
1131         case CCISS_REGNEWD:
1132         case CCISS_RESCANDISK:
1133         case CCISS_GETLUNINFO:
1134                 return do_ioctl(bdev, mode, cmd, arg);
1135
1136         case CCISS_PASSTHRU32:
1137                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1138         case CCISS_BIG_PASSTHRU32:
1139                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1140
1141         default:
1142                 return -ENOIOCTLCMD;
1143         }
1144 }
1145
1146 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1147                                   unsigned cmd, unsigned long arg)
1148 {
1149         IOCTL32_Command_struct __user *arg32 =
1150             (IOCTL32_Command_struct __user *) arg;
1151         IOCTL_Command_struct arg64;
1152         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1153         int err;
1154         u32 cp;
1155
1156         err = 0;
1157         err |=
1158             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1159                            sizeof(arg64.LUN_info));
1160         err |=
1161             copy_from_user(&arg64.Request, &arg32->Request,
1162                            sizeof(arg64.Request));
1163         err |=
1164             copy_from_user(&arg64.error_info, &arg32->error_info,
1165                            sizeof(arg64.error_info));
1166         err |= get_user(arg64.buf_size, &arg32->buf_size);
1167         err |= get_user(cp, &arg32->buf);
1168         arg64.buf = compat_ptr(cp);
1169         err |= copy_to_user(p, &arg64, sizeof(arg64));
1170
1171         if (err)
1172                 return -EFAULT;
1173
1174         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1175         if (err)
1176                 return err;
1177         err |=
1178             copy_in_user(&arg32->error_info, &p->error_info,
1179                          sizeof(arg32->error_info));
1180         if (err)
1181                 return -EFAULT;
1182         return err;
1183 }
1184
1185 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1186                                       unsigned cmd, unsigned long arg)
1187 {
1188         BIG_IOCTL32_Command_struct __user *arg32 =
1189             (BIG_IOCTL32_Command_struct __user *) arg;
1190         BIG_IOCTL_Command_struct arg64;
1191         BIG_IOCTL_Command_struct __user *p =
1192             compat_alloc_user_space(sizeof(arg64));
1193         int err;
1194         u32 cp;
1195
1196         memset(&arg64, 0, sizeof(arg64));
1197         err = 0;
1198         err |=
1199             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1200                            sizeof(arg64.LUN_info));
1201         err |=
1202             copy_from_user(&arg64.Request, &arg32->Request,
1203                            sizeof(arg64.Request));
1204         err |=
1205             copy_from_user(&arg64.error_info, &arg32->error_info,
1206                            sizeof(arg64.error_info));
1207         err |= get_user(arg64.buf_size, &arg32->buf_size);
1208         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1209         err |= get_user(cp, &arg32->buf);
1210         arg64.buf = compat_ptr(cp);
1211         err |= copy_to_user(p, &arg64, sizeof(arg64));
1212
1213         if (err)
1214                 return -EFAULT;
1215
1216         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1217         if (err)
1218                 return err;
1219         err |=
1220             copy_in_user(&arg32->error_info, &p->error_info,
1221                          sizeof(arg32->error_info));
1222         if (err)
1223                 return -EFAULT;
1224         return err;
1225 }
1226 #endif
1227
1228 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1229 {
1230         drive_info_struct *drv = get_drv(bdev->bd_disk);
1231
1232         if (!drv->cylinders)
1233                 return -ENXIO;
1234
1235         geo->heads = drv->heads;
1236         geo->sectors = drv->sectors;
1237         geo->cylinders = drv->cylinders;
1238         return 0;
1239 }
1240
1241 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1242 {
1243         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1244                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1245                 (void)check_for_unit_attention(h, c);
1246 }
1247
1248 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1249 {
1250         cciss_pci_info_struct pciinfo;
1251
1252         if (!argp)
1253                 return -EINVAL;
1254         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1255         pciinfo.bus = h->pdev->bus->number;
1256         pciinfo.dev_fn = h->pdev->devfn;
1257         pciinfo.board_id = h->board_id;
1258         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1259                 return -EFAULT;
1260         return 0;
1261 }
1262
1263 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1264 {
1265         cciss_coalint_struct intinfo;
1266
1267         if (!argp)
1268                 return -EINVAL;
1269         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1270         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1271         if (copy_to_user
1272             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1273                 return -EFAULT;
1274         return 0;
1275 }
1276
1277 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1278 {
1279         cciss_coalint_struct intinfo;
1280         unsigned long flags;
1281         int i;
1282
1283         if (!argp)
1284                 return -EINVAL;
1285         if (!capable(CAP_SYS_ADMIN))
1286                 return -EPERM;
1287         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1288                 return -EFAULT;
1289         if ((intinfo.delay == 0) && (intinfo.count == 0))
1290                 return -EINVAL;
1291         spin_lock_irqsave(&h->lock, flags);
1292         /* Update the field, and then ring the doorbell */
1293         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1294         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1295         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1296
1297         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1298                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1299                         break;
1300                 udelay(1000); /* delay and try again */
1301         }
1302         spin_unlock_irqrestore(&h->lock, flags);
1303         if (i >= MAX_IOCTL_CONFIG_WAIT)
1304                 return -EAGAIN;
1305         return 0;
1306 }
1307
1308 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1309 {
1310         NodeName_type NodeName;
1311         int i;
1312
1313         if (!argp)
1314                 return -EINVAL;
1315         for (i = 0; i < 16; i++)
1316                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1317         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1318                 return -EFAULT;
1319         return 0;
1320 }
1321
1322 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1323 {
1324         NodeName_type NodeName;
1325         unsigned long flags;
1326         int i;
1327
1328         if (!argp)
1329                 return -EINVAL;
1330         if (!capable(CAP_SYS_ADMIN))
1331                 return -EPERM;
1332         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1333                 return -EFAULT;
1334         spin_lock_irqsave(&h->lock, flags);
1335         /* Update the field, and then ring the doorbell */
1336         for (i = 0; i < 16; i++)
1337                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1338         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1339         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1340                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1341                         break;
1342                 udelay(1000); /* delay and try again */
1343         }
1344         spin_unlock_irqrestore(&h->lock, flags);
1345         if (i >= MAX_IOCTL_CONFIG_WAIT)
1346                 return -EAGAIN;
1347         return 0;
1348 }
1349
1350 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1351 {
1352         Heartbeat_type heartbeat;
1353
1354         if (!argp)
1355                 return -EINVAL;
1356         heartbeat = readl(&h->cfgtable->HeartBeat);
1357         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1358                 return -EFAULT;
1359         return 0;
1360 }
1361
1362 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1363 {
1364         BusTypes_type BusTypes;
1365
1366         if (!argp)
1367                 return -EINVAL;
1368         BusTypes = readl(&h->cfgtable->BusTypes);
1369         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1370                 return -EFAULT;
1371         return 0;
1372 }
1373
1374 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1375 {
1376         FirmwareVer_type firmware;
1377
1378         if (!argp)
1379                 return -EINVAL;
1380         memcpy(firmware, h->firm_ver, 4);
1381
1382         if (copy_to_user
1383             (argp, firmware, sizeof(FirmwareVer_type)))
1384                 return -EFAULT;
1385         return 0;
1386 }
1387
1388 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1389 {
1390         DriverVer_type DriverVer = DRIVER_VERSION;
1391
1392         if (!argp)
1393                 return -EINVAL;
1394         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1395                 return -EFAULT;
1396         return 0;
1397 }
1398
1399 static int cciss_getluninfo(ctlr_info_t *h,
1400         struct gendisk *disk, void __user *argp)
1401 {
1402         LogvolInfo_struct luninfo;
1403         drive_info_struct *drv = get_drv(disk);
1404
1405         if (!argp)
1406                 return -EINVAL;
1407         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1408         luninfo.num_opens = drv->usage_count;
1409         luninfo.num_parts = 0;
1410         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1411                 return -EFAULT;
1412         return 0;
1413 }
1414
1415 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1416 {
1417         IOCTL_Command_struct iocommand;
1418         CommandList_struct *c;
1419         char *buff = NULL;
1420         u64bit temp64;
1421         DECLARE_COMPLETION_ONSTACK(wait);
1422
1423         if (!argp)
1424                 return -EINVAL;
1425
1426         if (!capable(CAP_SYS_RAWIO))
1427                 return -EPERM;
1428
1429         if (copy_from_user
1430             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1431                 return -EFAULT;
1432         if ((iocommand.buf_size < 1) &&
1433             (iocommand.Request.Type.Direction != XFER_NONE)) {
1434                 return -EINVAL;
1435         }
1436         if (iocommand.buf_size > 0) {
1437                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1438                 if (buff == NULL)
1439                         return -EFAULT;
1440         }
1441         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1442                 /* Copy the data into the buffer we created */
1443                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1444                         kfree(buff);
1445                         return -EFAULT;
1446                 }
1447         } else {
1448                 memset(buff, 0, iocommand.buf_size);
1449         }
1450         c = cmd_special_alloc(h);
1451         if (!c) {
1452                 kfree(buff);
1453                 return -ENOMEM;
1454         }
1455         /* Fill in the command type */
1456         c->cmd_type = CMD_IOCTL_PEND;
1457         /* Fill in Command Header */
1458         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1459         if (iocommand.buf_size > 0) { /* buffer to fill */
1460                 c->Header.SGList = 1;
1461                 c->Header.SGTotal = 1;
1462         } else { /* no buffers to fill */
1463                 c->Header.SGList = 0;
1464                 c->Header.SGTotal = 0;
1465         }
1466         c->Header.LUN = iocommand.LUN_info;
1467         /* use the kernel address the cmd block for tag */
1468         c->Header.Tag.lower = c->busaddr;
1469
1470         /* Fill in Request block */
1471         c->Request = iocommand.Request;
1472
1473         /* Fill in the scatter gather information */
1474         if (iocommand.buf_size > 0) {
1475                 temp64.val = pci_map_single(h->pdev, buff,
1476                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1477                 c->SG[0].Addr.lower = temp64.val32.lower;
1478                 c->SG[0].Addr.upper = temp64.val32.upper;
1479                 c->SG[0].Len = iocommand.buf_size;
1480                 c->SG[0].Ext = 0;  /* we are not chaining */
1481         }
1482         c->waiting = &wait;
1483
1484         enqueue_cmd_and_start_io(h, c);
1485         wait_for_completion(&wait);
1486
1487         /* unlock the buffers from DMA */
1488         temp64.val32.lower = c->SG[0].Addr.lower;
1489         temp64.val32.upper = c->SG[0].Addr.upper;
1490         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1491                          PCI_DMA_BIDIRECTIONAL);
1492         check_ioctl_unit_attention(h, c);
1493
1494         /* Copy the error information out */
1495         iocommand.error_info = *(c->err_info);
1496         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1497                 kfree(buff);
1498                 cmd_special_free(h, c);
1499                 return -EFAULT;
1500         }
1501
1502         if (iocommand.Request.Type.Direction == XFER_READ) {
1503                 /* Copy the data out of the buffer we created */
1504                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1505                         kfree(buff);
1506                         cmd_special_free(h, c);
1507                         return -EFAULT;
1508                 }
1509         }
1510         kfree(buff);
1511         cmd_special_free(h, c);
1512         return 0;
1513 }
1514
1515 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1516 {
1517         BIG_IOCTL_Command_struct *ioc;
1518         CommandList_struct *c;
1519         unsigned char **buff = NULL;
1520         int *buff_size = NULL;
1521         u64bit temp64;
1522         BYTE sg_used = 0;
1523         int status = 0;
1524         int i;
1525         DECLARE_COMPLETION_ONSTACK(wait);
1526         __u32 left;
1527         __u32 sz;
1528         BYTE __user *data_ptr;
1529
1530         if (!argp)
1531                 return -EINVAL;
1532         if (!capable(CAP_SYS_RAWIO))
1533                 return -EPERM;
1534         ioc = kmalloc(sizeof(*ioc), GFP_KERNEL);
1535         if (!ioc) {
1536                 status = -ENOMEM;
1537                 goto cleanup1;
1538         }
1539         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1540                 status = -EFAULT;
1541                 goto cleanup1;
1542         }
1543         if ((ioc->buf_size < 1) &&
1544             (ioc->Request.Type.Direction != XFER_NONE)) {
1545                 status = -EINVAL;
1546                 goto cleanup1;
1547         }
1548         /* Check kmalloc limits  using all SGs */
1549         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1550                 status = -EINVAL;
1551                 goto cleanup1;
1552         }
1553         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1554                 status = -EINVAL;
1555                 goto cleanup1;
1556         }
1557         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1558         if (!buff) {
1559                 status = -ENOMEM;
1560                 goto cleanup1;
1561         }
1562         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1563         if (!buff_size) {
1564                 status = -ENOMEM;
1565                 goto cleanup1;
1566         }
1567         left = ioc->buf_size;
1568         data_ptr = ioc->buf;
1569         while (left) {
1570                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1571                 buff_size[sg_used] = sz;
1572                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1573                 if (buff[sg_used] == NULL) {
1574                         status = -ENOMEM;
1575                         goto cleanup1;
1576                 }
1577                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1578                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1579                                 status = -EFAULT;
1580                                 goto cleanup1;
1581                         }
1582                 } else {
1583                         memset(buff[sg_used], 0, sz);
1584                 }
1585                 left -= sz;
1586                 data_ptr += sz;
1587                 sg_used++;
1588         }
1589         c = cmd_special_alloc(h);
1590         if (!c) {
1591                 status = -ENOMEM;
1592                 goto cleanup1;
1593         }
1594         c->cmd_type = CMD_IOCTL_PEND;
1595         c->Header.ReplyQueue = 0;
1596         c->Header.SGList = sg_used;
1597         c->Header.SGTotal = sg_used;
1598         c->Header.LUN = ioc->LUN_info;
1599         c->Header.Tag.lower = c->busaddr;
1600
1601         c->Request = ioc->Request;
1602         for (i = 0; i < sg_used; i++) {
1603                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1604                                     PCI_DMA_BIDIRECTIONAL);
1605                 c->SG[i].Addr.lower = temp64.val32.lower;
1606                 c->SG[i].Addr.upper = temp64.val32.upper;
1607                 c->SG[i].Len = buff_size[i];
1608                 c->SG[i].Ext = 0;       /* we are not chaining */
1609         }
1610         c->waiting = &wait;
1611         enqueue_cmd_and_start_io(h, c);
1612         wait_for_completion(&wait);
1613         /* unlock the buffers from DMA */
1614         for (i = 0; i < sg_used; i++) {
1615                 temp64.val32.lower = c->SG[i].Addr.lower;
1616                 temp64.val32.upper = c->SG[i].Addr.upper;
1617                 pci_unmap_single(h->pdev,
1618                         (dma_addr_t) temp64.val, buff_size[i],
1619                         PCI_DMA_BIDIRECTIONAL);
1620         }
1621         check_ioctl_unit_attention(h, c);
1622         /* Copy the error information out */
1623         ioc->error_info = *(c->err_info);
1624         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1625                 cmd_special_free(h, c);
1626                 status = -EFAULT;
1627                 goto cleanup1;
1628         }
1629         if (ioc->Request.Type.Direction == XFER_READ) {
1630                 /* Copy the data out of the buffer we created */
1631                 BYTE __user *ptr = ioc->buf;
1632                 for (i = 0; i < sg_used; i++) {
1633                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1634                                 cmd_special_free(h, c);
1635                                 status = -EFAULT;
1636                                 goto cleanup1;
1637                         }
1638                         ptr += buff_size[i];
1639                 }
1640         }
1641         cmd_special_free(h, c);
1642         status = 0;
1643 cleanup1:
1644         if (buff) {
1645                 for (i = 0; i < sg_used; i++)
1646                         kfree(buff[i]);
1647                 kfree(buff);
1648         }
1649         kfree(buff_size);
1650         kfree(ioc);
1651         return status;
1652 }
1653
1654 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1655         unsigned int cmd, unsigned long arg)
1656 {
1657         struct gendisk *disk = bdev->bd_disk;
1658         ctlr_info_t *h = get_host(disk);
1659         void __user *argp = (void __user *)arg;
1660
1661         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1662                 cmd, arg);
1663         switch (cmd) {
1664         case CCISS_GETPCIINFO:
1665                 return cciss_getpciinfo(h, argp);
1666         case CCISS_GETINTINFO:
1667                 return cciss_getintinfo(h, argp);
1668         case CCISS_SETINTINFO:
1669                 return cciss_setintinfo(h, argp);
1670         case CCISS_GETNODENAME:
1671                 return cciss_getnodename(h, argp);
1672         case CCISS_SETNODENAME:
1673                 return cciss_setnodename(h, argp);
1674         case CCISS_GETHEARTBEAT:
1675                 return cciss_getheartbeat(h, argp);
1676         case CCISS_GETBUSTYPES:
1677                 return cciss_getbustypes(h, argp);
1678         case CCISS_GETFIRMVER:
1679                 return cciss_getfirmver(h, argp);
1680         case CCISS_GETDRIVVER:
1681                 return cciss_getdrivver(h, argp);
1682         case CCISS_DEREGDISK:
1683         case CCISS_REGNEWD:
1684         case CCISS_REVALIDVOLS:
1685                 return rebuild_lun_table(h, 0, 1);
1686         case CCISS_GETLUNINFO:
1687                 return cciss_getluninfo(h, disk, argp);
1688         case CCISS_PASSTHRU:
1689                 return cciss_passthru(h, argp);
1690         case CCISS_BIG_PASSTHRU:
1691                 return cciss_bigpassthru(h, argp);
1692
1693         /* scsi_cmd_ioctl handles these, below, though some are not */
1694         /* very meaningful for cciss.  SG_IO is the main one people want. */
1695
1696         case SG_GET_VERSION_NUM:
1697         case SG_SET_TIMEOUT:
1698         case SG_GET_TIMEOUT:
1699         case SG_GET_RESERVED_SIZE:
1700         case SG_SET_RESERVED_SIZE:
1701         case SG_EMULATED_HOST:
1702         case SG_IO:
1703         case SCSI_IOCTL_SEND_COMMAND:
1704                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1705
1706         /* scsi_cmd_ioctl would normally handle these, below, but */
1707         /* they aren't a good fit for cciss, as CD-ROMs are */
1708         /* not supported, and we don't have any bus/target/lun */
1709         /* which we present to the kernel. */
1710
1711         case CDROM_SEND_PACKET:
1712         case CDROMCLOSETRAY:
1713         case CDROMEJECT:
1714         case SCSI_IOCTL_GET_IDLUN:
1715         case SCSI_IOCTL_GET_BUS_NUMBER:
1716         default:
1717                 return -ENOTTY;
1718         }
1719 }
1720
1721 static void cciss_check_queues(ctlr_info_t *h)
1722 {
1723         int start_queue = h->next_to_run;
1724         int i;
1725
1726         /* check to see if we have maxed out the number of commands that can
1727          * be placed on the queue.  If so then exit.  We do this check here
1728          * in case the interrupt we serviced was from an ioctl and did not
1729          * free any new commands.
1730          */
1731         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1732                 return;
1733
1734         /* We have room on the queue for more commands.  Now we need to queue
1735          * them up.  We will also keep track of the next queue to run so
1736          * that every queue gets a chance to be started first.
1737          */
1738         for (i = 0; i < h->highest_lun + 1; i++) {
1739                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1740                 /* make sure the disk has been added and the drive is real
1741                  * because this can be called from the middle of init_one.
1742                  */
1743                 if (!h->drv[curr_queue])
1744                         continue;
1745                 if (!(h->drv[curr_queue]->queue) ||
1746                         !(h->drv[curr_queue]->heads))
1747                         continue;
1748                 blk_start_queue(h->gendisk[curr_queue]->queue);
1749
1750                 /* check to see if we have maxed out the number of commands
1751                  * that can be placed on the queue.
1752                  */
1753                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1754                         if (curr_queue == start_queue) {
1755                                 h->next_to_run =
1756                                     (start_queue + 1) % (h->highest_lun + 1);
1757                                 break;
1758                         } else {
1759                                 h->next_to_run = curr_queue;
1760                                 break;
1761                         }
1762                 }
1763         }
1764 }
1765
1766 static void cciss_softirq_done(struct request *rq)
1767 {
1768         CommandList_struct *c = rq->completion_data;
1769         ctlr_info_t *h = hba[c->ctlr];
1770         SGDescriptor_struct *curr_sg = c->SG;
1771         u64bit temp64;
1772         unsigned long flags;
1773         int i, ddir;
1774         int sg_index = 0;
1775
1776         if (c->Request.Type.Direction == XFER_READ)
1777                 ddir = PCI_DMA_FROMDEVICE;
1778         else
1779                 ddir = PCI_DMA_TODEVICE;
1780
1781         /* command did not need to be retried */
1782         /* unmap the DMA mapping for all the scatter gather elements */
1783         for (i = 0; i < c->Header.SGList; i++) {
1784                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1785                         cciss_unmap_sg_chain_block(h, c);
1786                         /* Point to the next block */
1787                         curr_sg = h->cmd_sg_list[c->cmdindex];
1788                         sg_index = 0;
1789                 }
1790                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1791                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1792                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1793                                 ddir);
1794                 ++sg_index;
1795         }
1796
1797         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1798
1799         /* set the residual count for pc requests */
1800         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1801                 rq->resid_len = c->err_info->ResidualCnt;
1802
1803         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1804
1805         spin_lock_irqsave(&h->lock, flags);
1806         cmd_free(h, c);
1807         cciss_check_queues(h);
1808         spin_unlock_irqrestore(&h->lock, flags);
1809 }
1810
1811 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1812         unsigned char scsi3addr[], uint32_t log_unit)
1813 {
1814         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1815                 sizeof(h->drv[log_unit]->LunID));
1816 }
1817
1818 /* This function gets the SCSI vendor, model, and revision of a logical drive
1819  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1820  * they cannot be read.
1821  */
1822 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1823                                    char *vendor, char *model, char *rev)
1824 {
1825         int rc;
1826         InquiryData_struct *inq_buf;
1827         unsigned char scsi3addr[8];
1828
1829         *vendor = '\0';
1830         *model = '\0';
1831         *rev = '\0';
1832
1833         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1834         if (!inq_buf)
1835                 return;
1836
1837         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1838         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1839                         scsi3addr, TYPE_CMD);
1840         if (rc == IO_OK) {
1841                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1842                 vendor[VENDOR_LEN] = '\0';
1843                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1844                 model[MODEL_LEN] = '\0';
1845                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1846                 rev[REV_LEN] = '\0';
1847         }
1848
1849         kfree(inq_buf);
1850         return;
1851 }
1852
1853 /* This function gets the serial number of a logical drive via
1854  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1855  * number cannot be had, for whatever reason, 16 bytes of 0xff
1856  * are returned instead.
1857  */
1858 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1859                                 unsigned char *serial_no, int buflen)
1860 {
1861 #define PAGE_83_INQ_BYTES 64
1862         int rc;
1863         unsigned char *buf;
1864         unsigned char scsi3addr[8];
1865
1866         if (buflen > 16)
1867                 buflen = 16;
1868         memset(serial_no, 0xff, buflen);
1869         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1870         if (!buf)
1871                 return;
1872         memset(serial_no, 0, buflen);
1873         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1874         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1875                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1876         if (rc == IO_OK)
1877                 memcpy(serial_no, &buf[8], buflen);
1878         kfree(buf);
1879         return;
1880 }
1881
1882 /*
1883  * cciss_add_disk sets up the block device queue for a logical drive
1884  */
1885 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1886                                 int drv_index)
1887 {
1888         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1889         if (!disk->queue)
1890                 goto init_queue_failure;
1891         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1892         disk->major = h->major;
1893         disk->first_minor = drv_index << NWD_SHIFT;
1894         disk->fops = &cciss_fops;
1895         if (cciss_create_ld_sysfs_entry(h, drv_index))
1896                 goto cleanup_queue;
1897         disk->private_data = h->drv[drv_index];
1898         disk->driverfs_dev = &h->drv[drv_index]->dev;
1899
1900         /* Set up queue information */
1901         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1902
1903         /* This is a hardware imposed limit. */
1904         blk_queue_max_segments(disk->queue, h->maxsgentries);
1905
1906         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1907
1908         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1909
1910         disk->queue->queuedata = h;
1911
1912         blk_queue_logical_block_size(disk->queue,
1913                                      h->drv[drv_index]->block_size);
1914
1915         /* Make sure all queue data is written out before */
1916         /* setting h->drv[drv_index]->queue, as setting this */
1917         /* allows the interrupt handler to start the queue */
1918         wmb();
1919         h->drv[drv_index]->queue = disk->queue;
1920         add_disk(disk);
1921         return 0;
1922
1923 cleanup_queue:
1924         blk_cleanup_queue(disk->queue);
1925         disk->queue = NULL;
1926 init_queue_failure:
1927         return -1;
1928 }
1929
1930 /* This function will check the usage_count of the drive to be updated/added.
1931  * If the usage_count is zero and it is a heretofore unknown drive, or,
1932  * the drive's capacity, geometry, or serial number has changed,
1933  * then the drive information will be updated and the disk will be
1934  * re-registered with the kernel.  If these conditions don't hold,
1935  * then it will be left alone for the next reboot.  The exception to this
1936  * is disk 0 which will always be left registered with the kernel since it
1937  * is also the controller node.  Any changes to disk 0 will show up on
1938  * the next reboot.
1939  */
1940 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1941         int first_time, int via_ioctl)
1942 {
1943         struct gendisk *disk;
1944         InquiryData_struct *inq_buff = NULL;
1945         unsigned int block_size;
1946         sector_t total_size;
1947         unsigned long flags = 0;
1948         int ret = 0;
1949         drive_info_struct *drvinfo;
1950
1951         /* Get information about the disk and modify the driver structure */
1952         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1953         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1954         if (inq_buff == NULL || drvinfo == NULL)
1955                 goto mem_msg;
1956
1957         /* testing to see if 16-byte CDBs are already being used */
1958         if (h->cciss_read == CCISS_READ_16) {
1959                 cciss_read_capacity_16(h, drv_index,
1960                         &total_size, &block_size);
1961
1962         } else {
1963                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1964                 /* if read_capacity returns all F's this volume is >2TB */
1965                 /* in size so we switch to 16-byte CDB's for all */
1966                 /* read/write ops */
1967                 if (total_size == 0xFFFFFFFFULL) {
1968                         cciss_read_capacity_16(h, drv_index,
1969                         &total_size, &block_size);
1970                         h->cciss_read = CCISS_READ_16;
1971                         h->cciss_write = CCISS_WRITE_16;
1972                 } else {
1973                         h->cciss_read = CCISS_READ_10;
1974                         h->cciss_write = CCISS_WRITE_10;
1975                 }
1976         }
1977
1978         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1979                                inq_buff, drvinfo);
1980         drvinfo->block_size = block_size;
1981         drvinfo->nr_blocks = total_size + 1;
1982
1983         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1984                                 drvinfo->model, drvinfo->rev);
1985         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1986                         sizeof(drvinfo->serial_no));
1987         /* Save the lunid in case we deregister the disk, below. */
1988         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1989                 sizeof(drvinfo->LunID));
1990
1991         /* Is it the same disk we already know, and nothing's changed? */
1992         if (h->drv[drv_index]->raid_level != -1 &&
1993                 ((memcmp(drvinfo->serial_no,
1994                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1995                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1996                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1997                 drvinfo->heads == h->drv[drv_index]->heads &&
1998                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1999                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2000                         /* The disk is unchanged, nothing to update */
2001                         goto freeret;
2002
2003         /* If we get here it's not the same disk, or something's changed,
2004          * so we need to * deregister it, and re-register it, if it's not
2005          * in use.
2006          * If the disk already exists then deregister it before proceeding
2007          * (unless it's the first disk (for the controller node).
2008          */
2009         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2010                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2011                 spin_lock_irqsave(&h->lock, flags);
2012                 h->drv[drv_index]->busy_configuring = 1;
2013                 spin_unlock_irqrestore(&h->lock, flags);
2014
2015                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2016                  * which keeps the interrupt handler from starting
2017                  * the queue.
2018                  */
2019                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2020         }
2021
2022         /* If the disk is in use return */
2023         if (ret)
2024                 goto freeret;
2025
2026         /* Save the new information from cciss_geometry_inquiry
2027          * and serial number inquiry.  If the disk was deregistered
2028          * above, then h->drv[drv_index] will be NULL.
2029          */
2030         if (h->drv[drv_index] == NULL) {
2031                 drvinfo->device_initialized = 0;
2032                 h->drv[drv_index] = drvinfo;
2033                 drvinfo = NULL; /* so it won't be freed below. */
2034         } else {
2035                 /* special case for cxd0 */
2036                 h->drv[drv_index]->block_size = drvinfo->block_size;
2037                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2038                 h->drv[drv_index]->heads = drvinfo->heads;
2039                 h->drv[drv_index]->sectors = drvinfo->sectors;
2040                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2041                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2042                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2043                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2044                         VENDOR_LEN + 1);
2045                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2046                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2047         }
2048
2049         ++h->num_luns;
2050         disk = h->gendisk[drv_index];
2051         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2052
2053         /* If it's not disk 0 (drv_index != 0)
2054          * or if it was disk 0, but there was previously
2055          * no actual corresponding configured logical drive
2056          * (raid_leve == -1) then we want to update the
2057          * logical drive's information.
2058          */
2059         if (drv_index || first_time) {
2060                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2061                         cciss_free_gendisk(h, drv_index);
2062                         cciss_free_drive_info(h, drv_index);
2063                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2064                                 drv_index);
2065                         --h->num_luns;
2066                 }
2067         }
2068
2069 freeret:
2070         kfree(inq_buff);
2071         kfree(drvinfo);
2072         return;
2073 mem_msg:
2074         dev_err(&h->pdev->dev, "out of memory\n");
2075         goto freeret;
2076 }
2077
2078 /* This function will find the first index of the controllers drive array
2079  * that has a null drv pointer and allocate the drive info struct and
2080  * will return that index   This is where new drives will be added.
2081  * If the index to be returned is greater than the highest_lun index for
2082  * the controller then highest_lun is set * to this new index.
2083  * If there are no available indexes or if tha allocation fails, then -1
2084  * is returned.  * "controller_node" is used to know if this is a real
2085  * logical drive, or just the controller node, which determines if this
2086  * counts towards highest_lun.
2087  */
2088 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2089 {
2090         int i;
2091         drive_info_struct *drv;
2092
2093         /* Search for an empty slot for our drive info */
2094         for (i = 0; i < CISS_MAX_LUN; i++) {
2095
2096                 /* if not cxd0 case, and it's occupied, skip it. */
2097                 if (h->drv[i] && i != 0)
2098                         continue;
2099                 /*
2100                  * If it's cxd0 case, and drv is alloc'ed already, and a
2101                  * disk is configured there, skip it.
2102                  */
2103                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2104                         continue;
2105
2106                 /*
2107                  * We've found an empty slot.  Update highest_lun
2108                  * provided this isn't just the fake cxd0 controller node.
2109                  */
2110                 if (i > h->highest_lun && !controller_node)
2111                         h->highest_lun = i;
2112
2113                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2114                 if (i == 0 && h->drv[i] != NULL)
2115                         return i;
2116
2117                 /*
2118                  * Found an empty slot, not already alloc'ed.  Allocate it.
2119                  * Mark it with raid_level == -1, so we know it's new later on.
2120                  */
2121                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2122                 if (!drv)
2123                         return -1;
2124                 drv->raid_level = -1; /* so we know it's new */
2125                 h->drv[i] = drv;
2126                 return i;
2127         }
2128         return -1;
2129 }
2130
2131 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2132 {
2133         kfree(h->drv[drv_index]);
2134         h->drv[drv_index] = NULL;
2135 }
2136
2137 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2138 {
2139         put_disk(h->gendisk[drv_index]);
2140         h->gendisk[drv_index] = NULL;
2141 }
2142
2143 /* cciss_add_gendisk finds a free hba[]->drv structure
2144  * and allocates a gendisk if needed, and sets the lunid
2145  * in the drvinfo structure.   It returns the index into
2146  * the ->drv[] array, or -1 if none are free.
2147  * is_controller_node indicates whether highest_lun should
2148  * count this disk, or if it's only being added to provide
2149  * a means to talk to the controller in case no logical
2150  * drives have yet been configured.
2151  */
2152 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2153         int controller_node)
2154 {
2155         int drv_index;
2156
2157         drv_index = cciss_alloc_drive_info(h, controller_node);
2158         if (drv_index == -1)
2159                 return -1;
2160
2161         /*Check if the gendisk needs to be allocated */
2162         if (!h->gendisk[drv_index]) {
2163                 h->gendisk[drv_index] =
2164                         alloc_disk(1 << NWD_SHIFT);
2165                 if (!h->gendisk[drv_index]) {
2166                         dev_err(&h->pdev->dev,
2167                                 "could not allocate a new disk %d\n",
2168                                 drv_index);
2169                         goto err_free_drive_info;
2170                 }
2171         }
2172         memcpy(h->drv[drv_index]->LunID, lunid,
2173                 sizeof(h->drv[drv_index]->LunID));
2174         if (cciss_create_ld_sysfs_entry(h, drv_index))
2175                 goto err_free_disk;
2176         /* Don't need to mark this busy because nobody */
2177         /* else knows about this disk yet to contend */
2178         /* for access to it. */
2179         h->drv[drv_index]->busy_configuring = 0;
2180         wmb();
2181         return drv_index;
2182
2183 err_free_disk:
2184         cciss_free_gendisk(h, drv_index);
2185 err_free_drive_info:
2186         cciss_free_drive_info(h, drv_index);
2187         return -1;
2188 }
2189
2190 /* This is for the special case of a controller which
2191  * has no logical drives.  In this case, we still need
2192  * to register a disk so the controller can be accessed
2193  * by the Array Config Utility.
2194  */
2195 static void cciss_add_controller_node(ctlr_info_t *h)
2196 {
2197         struct gendisk *disk;
2198         int drv_index;
2199
2200         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2201                 return;
2202
2203         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2204         if (drv_index == -1)
2205                 goto error;
2206         h->drv[drv_index]->block_size = 512;
2207         h->drv[drv_index]->nr_blocks = 0;
2208         h->drv[drv_index]->heads = 0;
2209         h->drv[drv_index]->sectors = 0;
2210         h->drv[drv_index]->cylinders = 0;
2211         h->drv[drv_index]->raid_level = -1;
2212         memset(h->drv[drv_index]->serial_no, 0, 16);
2213         disk = h->gendisk[drv_index];
2214         if (cciss_add_disk(h, disk, drv_index) == 0)
2215                 return;
2216         cciss_free_gendisk(h, drv_index);
2217         cciss_free_drive_info(h, drv_index);
2218 error:
2219         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2220         return;
2221 }
2222
2223 /* This function will add and remove logical drives from the Logical
2224  * drive array of the controller and maintain persistency of ordering
2225  * so that mount points are preserved until the next reboot.  This allows
2226  * for the removal of logical drives in the middle of the drive array
2227  * without a re-ordering of those drives.
2228  * INPUT
2229  * h            = The controller to perform the operations on
2230  */
2231 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2232         int via_ioctl)
2233 {
2234         int num_luns;
2235         ReportLunData_struct *ld_buff = NULL;
2236         int return_code;
2237         int listlength = 0;
2238         int i;
2239         int drv_found;
2240         int drv_index = 0;
2241         unsigned char lunid[8] = CTLR_LUNID;
2242         unsigned long flags;
2243
2244         if (!capable(CAP_SYS_RAWIO))
2245                 return -EPERM;
2246
2247         /* Set busy_configuring flag for this operation */
2248         spin_lock_irqsave(&h->lock, flags);
2249         if (h->busy_configuring) {
2250                 spin_unlock_irqrestore(&h->lock, flags);
2251                 return -EBUSY;
2252         }
2253         h->busy_configuring = 1;
2254         spin_unlock_irqrestore(&h->lock, flags);
2255
2256         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2257         if (ld_buff == NULL)
2258                 goto mem_msg;
2259
2260         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2261                                       sizeof(ReportLunData_struct),
2262                                       0, CTLR_LUNID, TYPE_CMD);
2263
2264         if (return_code == IO_OK)
2265                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2266         else {  /* reading number of logical volumes failed */
2267                 dev_warn(&h->pdev->dev,
2268                         "report logical volume command failed\n");
2269                 listlength = 0;
2270                 goto freeret;
2271         }
2272
2273         num_luns = listlength / 8;      /* 8 bytes per entry */
2274         if (num_luns > CISS_MAX_LUN) {
2275                 num_luns = CISS_MAX_LUN;
2276                 dev_warn(&h->pdev->dev, "more luns configured"
2277                        " on controller than can be handled by"
2278                        " this driver.\n");
2279         }
2280
2281         if (num_luns == 0)
2282                 cciss_add_controller_node(h);
2283
2284         /* Compare controller drive array to driver's drive array
2285          * to see if any drives are missing on the controller due
2286          * to action of Array Config Utility (user deletes drive)
2287          * and deregister logical drives which have disappeared.
2288          */
2289         for (i = 0; i <= h->highest_lun; i++) {
2290                 int j;
2291                 drv_found = 0;
2292
2293                 /* skip holes in the array from already deleted drives */
2294                 if (h->drv[i] == NULL)
2295                         continue;
2296
2297                 for (j = 0; j < num_luns; j++) {
2298                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2299                         if (memcmp(h->drv[i]->LunID, lunid,
2300                                 sizeof(lunid)) == 0) {
2301                                 drv_found = 1;
2302                                 break;
2303                         }
2304                 }
2305                 if (!drv_found) {
2306                         /* Deregister it from the OS, it's gone. */
2307                         spin_lock_irqsave(&h->lock, flags);
2308                         h->drv[i]->busy_configuring = 1;
2309                         spin_unlock_irqrestore(&h->lock, flags);
2310                         return_code = deregister_disk(h, i, 1, via_ioctl);
2311                         if (h->drv[i] != NULL)
2312                                 h->drv[i]->busy_configuring = 0;
2313                 }
2314         }
2315
2316         /* Compare controller drive array to driver's drive array.
2317          * Check for updates in the drive information and any new drives
2318          * on the controller due to ACU adding logical drives, or changing
2319          * a logical drive's size, etc.  Reregister any new/changed drives
2320          */
2321         for (i = 0; i < num_luns; i++) {
2322                 int j;
2323
2324                 drv_found = 0;
2325
2326                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2327                 /* Find if the LUN is already in the drive array
2328                  * of the driver.  If so then update its info
2329                  * if not in use.  If it does not exist then find
2330                  * the first free index and add it.
2331                  */
2332                 for (j = 0; j <= h->highest_lun; j++) {
2333                         if (h->drv[j] != NULL &&
2334                                 memcmp(h->drv[j]->LunID, lunid,
2335                                         sizeof(h->drv[j]->LunID)) == 0) {
2336                                 drv_index = j;
2337                                 drv_found = 1;
2338                                 break;
2339                         }
2340                 }
2341
2342                 /* check if the drive was found already in the array */
2343                 if (!drv_found) {
2344                         drv_index = cciss_add_gendisk(h, lunid, 0);
2345                         if (drv_index == -1)
2346                                 goto freeret;
2347                 }
2348                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2349         }               /* end for */
2350
2351 freeret:
2352         kfree(ld_buff);
2353         h->busy_configuring = 0;
2354         /* We return -1 here to tell the ACU that we have registered/updated
2355          * all of the drives that we can and to keep it from calling us
2356          * additional times.
2357          */
2358         return -1;
2359 mem_msg:
2360         dev_err(&h->pdev->dev, "out of memory\n");
2361         h->busy_configuring = 0;
2362         goto freeret;
2363 }
2364
2365 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2366 {
2367         /* zero out the disk size info */
2368         drive_info->nr_blocks = 0;
2369         drive_info->block_size = 0;
2370         drive_info->heads = 0;
2371         drive_info->sectors = 0;
2372         drive_info->cylinders = 0;
2373         drive_info->raid_level = -1;
2374         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2375         memset(drive_info->model, 0, sizeof(drive_info->model));
2376         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2377         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2378         /*
2379          * don't clear the LUNID though, we need to remember which
2380          * one this one is.
2381          */
2382 }
2383
2384 /* This function will deregister the disk and it's queue from the
2385  * kernel.  It must be called with the controller lock held and the
2386  * drv structures busy_configuring flag set.  It's parameters are:
2387  *
2388  * disk = This is the disk to be deregistered
2389  * drv  = This is the drive_info_struct associated with the disk to be
2390  *        deregistered.  It contains information about the disk used
2391  *        by the driver.
2392  * clear_all = This flag determines whether or not the disk information
2393  *             is going to be completely cleared out and the highest_lun
2394  *             reset.  Sometimes we want to clear out information about
2395  *             the disk in preparation for re-adding it.  In this case
2396  *             the highest_lun should be left unchanged and the LunID
2397  *             should not be cleared.
2398  * via_ioctl
2399  *    This indicates whether we've reached this path via ioctl.
2400  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2401  *    If this path is reached via ioctl(), then the max_usage_count will
2402  *    be 1, as the process calling ioctl() has got to have the device open.
2403  *    If we get here via sysfs, then the max usage count will be zero.
2404 */
2405 static int deregister_disk(ctlr_info_t *h, int drv_index,
2406                            int clear_all, int via_ioctl)
2407 {
2408         int i;
2409         struct gendisk *disk;
2410         drive_info_struct *drv;
2411         int recalculate_highest_lun;
2412
2413         if (!capable(CAP_SYS_RAWIO))
2414                 return -EPERM;
2415
2416         drv = h->drv[drv_index];
2417         disk = h->gendisk[drv_index];
2418
2419         /* make sure logical volume is NOT is use */
2420         if (clear_all || (h->gendisk[0] == disk)) {
2421                 if (drv->usage_count > via_ioctl)
2422                         return -EBUSY;
2423         } else if (drv->usage_count > 0)
2424                 return -EBUSY;
2425
2426         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2427
2428         /* invalidate the devices and deregister the disk.  If it is disk
2429          * zero do not deregister it but just zero out it's values.  This
2430          * allows us to delete disk zero but keep the controller registered.
2431          */
2432         if (h->gendisk[0] != disk) {
2433                 struct request_queue *q = disk->queue;
2434                 if (disk->flags & GENHD_FL_UP) {
2435                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2436                         del_gendisk(disk);
2437                 }
2438                 if (q)
2439                         blk_cleanup_queue(q);
2440                 /* If clear_all is set then we are deleting the logical
2441                  * drive, not just refreshing its info.  For drives
2442                  * other than disk 0 we will call put_disk.  We do not
2443                  * do this for disk 0 as we need it to be able to
2444                  * configure the controller.
2445                  */
2446                 if (clear_all){
2447                         /* This isn't pretty, but we need to find the
2448                          * disk in our array and NULL our the pointer.
2449                          * This is so that we will call alloc_disk if
2450                          * this index is used again later.
2451                          */
2452                         for (i=0; i < CISS_MAX_LUN; i++){
2453                                 if (h->gendisk[i] == disk) {
2454                                         h->gendisk[i] = NULL;
2455                                         break;
2456                                 }
2457                         }
2458                         put_disk(disk);
2459                 }
2460         } else {
2461                 set_capacity(disk, 0);
2462                 cciss_clear_drive_info(drv);
2463         }
2464
2465         --h->num_luns;
2466
2467         /* if it was the last disk, find the new hightest lun */
2468         if (clear_all && recalculate_highest_lun) {
2469                 int newhighest = -1;
2470                 for (i = 0; i <= h->highest_lun; i++) {
2471                         /* if the disk has size > 0, it is available */
2472                         if (h->drv[i] && h->drv[i]->heads)
2473                                 newhighest = i;
2474                 }
2475                 h->highest_lun = newhighest;
2476         }
2477         return 0;
2478 }
2479
2480 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2481                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2482                 int cmd_type)
2483 {
2484         u64bit buff_dma_handle;
2485         int status = IO_OK;
2486
2487         c->cmd_type = CMD_IOCTL_PEND;
2488         c->Header.ReplyQueue = 0;
2489         if (buff != NULL) {
2490                 c->Header.SGList = 1;
2491                 c->Header.SGTotal = 1;
2492         } else {
2493                 c->Header.SGList = 0;
2494                 c->Header.SGTotal = 0;
2495         }
2496         c->Header.Tag.lower = c->busaddr;
2497         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2498
2499         c->Request.Type.Type = cmd_type;
2500         if (cmd_type == TYPE_CMD) {
2501                 switch (cmd) {
2502                 case CISS_INQUIRY:
2503                         /* are we trying to read a vital product page */
2504                         if (page_code != 0) {
2505                                 c->Request.CDB[1] = 0x01;
2506                                 c->Request.CDB[2] = page_code;
2507                         }
2508                         c->Request.CDBLen = 6;
2509                         c->Request.Type.Attribute = ATTR_SIMPLE;
2510                         c->Request.Type.Direction = XFER_READ;
2511                         c->Request.Timeout = 0;
2512                         c->Request.CDB[0] = CISS_INQUIRY;
2513                         c->Request.CDB[4] = size & 0xFF;
2514                         break;
2515                 case CISS_REPORT_LOG:
2516                 case CISS_REPORT_PHYS:
2517                         /* Talking to controller so It's a physical command
2518                            mode = 00 target = 0.  Nothing to write.
2519                          */
2520                         c->Request.CDBLen = 12;
2521                         c->Request.Type.Attribute = ATTR_SIMPLE;
2522                         c->Request.Type.Direction = XFER_READ;
2523                         c->Request.Timeout = 0;
2524                         c->Request.CDB[0] = cmd;
2525                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2526                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2527                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2528                         c->Request.CDB[9] = size & 0xFF;
2529                         break;
2530
2531                 case CCISS_READ_CAPACITY:
2532                         c->Request.CDBLen = 10;
2533                         c->Request.Type.Attribute = ATTR_SIMPLE;
2534                         c->Request.Type.Direction = XFER_READ;
2535                         c->Request.Timeout = 0;
2536                         c->Request.CDB[0] = cmd;
2537                         break;
2538                 case CCISS_READ_CAPACITY_16:
2539                         c->Request.CDBLen = 16;
2540                         c->Request.Type.Attribute = ATTR_SIMPLE;
2541                         c->Request.Type.Direction = XFER_READ;
2542                         c->Request.Timeout = 0;
2543                         c->Request.CDB[0] = cmd;
2544                         c->Request.CDB[1] = 0x10;
2545                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2546                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2547                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2548                         c->Request.CDB[13] = size & 0xFF;
2549                         c->Request.Timeout = 0;
2550                         c->Request.CDB[0] = cmd;
2551                         break;
2552                 case CCISS_CACHE_FLUSH:
2553                         c->Request.CDBLen = 12;
2554                         c->Request.Type.Attribute = ATTR_SIMPLE;
2555                         c->Request.Type.Direction = XFER_WRITE;
2556                         c->Request.Timeout = 0;
2557                         c->Request.CDB[0] = BMIC_WRITE;
2558                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2559                         break;
2560                 case TEST_UNIT_READY:
2561                         c->Request.CDBLen = 6;
2562                         c->Request.Type.Attribute = ATTR_SIMPLE;
2563                         c->Request.Type.Direction = XFER_NONE;
2564                         c->Request.Timeout = 0;
2565                         break;
2566                 default:
2567                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2568                         return IO_ERROR;
2569                 }
2570         } else if (cmd_type == TYPE_MSG) {
2571                 switch (cmd) {
2572                 case CCISS_ABORT_MSG:
2573                         c->Request.CDBLen = 12;
2574                         c->Request.Type.Attribute = ATTR_SIMPLE;
2575                         c->Request.Type.Direction = XFER_WRITE;
2576                         c->Request.Timeout = 0;
2577                         c->Request.CDB[0] = cmd;        /* abort */
2578                         c->Request.CDB[1] = 0;  /* abort a command */
2579                         /* buff contains the tag of the command to abort */
2580                         memcpy(&c->Request.CDB[4], buff, 8);
2581                         break;
2582                 case CCISS_RESET_MSG:
2583                         c->Request.CDBLen = 16;
2584                         c->Request.Type.Attribute = ATTR_SIMPLE;
2585                         c->Request.Type.Direction = XFER_NONE;
2586                         c->Request.Timeout = 0;
2587                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2588                         c->Request.CDB[0] = cmd;        /* reset */
2589                         c->Request.CDB[1] = CCISS_RESET_TYPE_TARGET;
2590                         break;
2591                 case CCISS_NOOP_MSG:
2592                         c->Request.CDBLen = 1;
2593                         c->Request.Type.Attribute = ATTR_SIMPLE;
2594                         c->Request.Type.Direction = XFER_WRITE;
2595                         c->Request.Timeout = 0;
2596                         c->Request.CDB[0] = cmd;
2597                         break;
2598                 default:
2599                         dev_warn(&h->pdev->dev,
2600                                 "unknown message type %d\n", cmd);
2601                         return IO_ERROR;
2602                 }
2603         } else {
2604                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2605                 return IO_ERROR;
2606         }
2607         /* Fill in the scatter gather information */
2608         if (size > 0) {
2609                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2610                                                              buff, size,
2611                                                              PCI_DMA_BIDIRECTIONAL);
2612                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2613                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2614                 c->SG[0].Len = size;
2615                 c->SG[0].Ext = 0;       /* we are not chaining */
2616         }
2617         return status;
2618 }
2619
2620 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2621 {
2622         switch (c->err_info->ScsiStatus) {
2623         case SAM_STAT_GOOD:
2624                 return IO_OK;
2625         case SAM_STAT_CHECK_CONDITION:
2626                 switch (0xf & c->err_info->SenseInfo[2]) {
2627                 case 0: return IO_OK; /* no sense */
2628                 case 1: return IO_OK; /* recovered error */
2629                 default:
2630                         if (check_for_unit_attention(h, c))
2631                                 return IO_NEEDS_RETRY;
2632                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2633                                 "check condition, sense key = 0x%02x\n",
2634                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2635                 }
2636                 break;
2637         default:
2638                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2639                         "scsi status = 0x%02x\n",
2640                         c->Request.CDB[0], c->err_info->ScsiStatus);
2641                 break;
2642         }
2643         return IO_ERROR;
2644 }
2645
2646 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2647 {
2648         int return_status = IO_OK;
2649
2650         if (c->err_info->CommandStatus == CMD_SUCCESS)
2651                 return IO_OK;
2652
2653         switch (c->err_info->CommandStatus) {
2654         case CMD_TARGET_STATUS:
2655                 return_status = check_target_status(h, c);
2656                 break;
2657         case CMD_DATA_UNDERRUN:
2658         case CMD_DATA_OVERRUN:
2659                 /* expected for inquiry and report lun commands */
2660                 break;
2661         case CMD_INVALID:
2662                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2663                        "reported invalid\n", c->Request.CDB[0]);
2664                 return_status = IO_ERROR;
2665                 break;
2666         case CMD_PROTOCOL_ERR:
2667                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2668                        "protocol error\n", c->Request.CDB[0]);
2669                 return_status = IO_ERROR;
2670                 break;
2671         case CMD_HARDWARE_ERR:
2672                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2673                        " hardware error\n", c->Request.CDB[0]);
2674                 return_status = IO_ERROR;
2675                 break;
2676         case CMD_CONNECTION_LOST:
2677                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2678                        "connection lost\n", c->Request.CDB[0]);
2679                 return_status = IO_ERROR;
2680                 break;
2681         case CMD_ABORTED:
2682                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2683                        "aborted\n", c->Request.CDB[0]);
2684                 return_status = IO_ERROR;
2685                 break;
2686         case CMD_ABORT_FAILED:
2687                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2688                        "abort failed\n", c->Request.CDB[0]);
2689                 return_status = IO_ERROR;
2690                 break;
2691         case CMD_UNSOLICITED_ABORT:
2692                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2693                         c->Request.CDB[0]);
2694                 return_status = IO_NEEDS_RETRY;
2695                 break;
2696         case CMD_UNABORTABLE:
2697                 dev_warn(&h->pdev->dev, "cmd unabortable\n");
2698                 return_status = IO_ERROR;
2699                 break;
2700         default:
2701                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2702                        "unknown status %x\n", c->Request.CDB[0],
2703                        c->err_info->CommandStatus);
2704                 return_status = IO_ERROR;
2705         }
2706         return return_status;
2707 }
2708
2709 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2710         int attempt_retry)
2711 {
2712         DECLARE_COMPLETION_ONSTACK(wait);
2713         u64bit buff_dma_handle;
2714         int return_status = IO_OK;
2715
2716 resend_cmd2:
2717         c->waiting = &wait;
2718         enqueue_cmd_and_start_io(h, c);
2719
2720         wait_for_completion(&wait);
2721
2722         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2723                 goto command_done;
2724
2725         return_status = process_sendcmd_error(h, c);
2726
2727         if (return_status == IO_NEEDS_RETRY &&
2728                 c->retry_count < MAX_CMD_RETRIES) {
2729                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2730                         c->Request.CDB[0]);
2731                 c->retry_count++;
2732                 /* erase the old error information */
2733                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2734                 return_status = IO_OK;
2735                 INIT_COMPLETION(wait);
2736                 goto resend_cmd2;
2737         }
2738
2739 command_done:
2740         /* unlock the buffers from DMA */
2741         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2742         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2743         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2744                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2745         return return_status;
2746 }
2747
2748 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2749                            __u8 page_code, unsigned char scsi3addr[],
2750                         int cmd_type)
2751 {
2752         CommandList_struct *c;
2753         int return_status;
2754
2755         c = cmd_special_alloc(h);
2756         if (!c)
2757                 return -ENOMEM;
2758         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2759                 scsi3addr, cmd_type);
2760         if (return_status == IO_OK)
2761                 return_status = sendcmd_withirq_core(h, c, 1);
2762
2763         cmd_special_free(h, c);
2764         return return_status;
2765 }
2766
2767 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2768                                    sector_t total_size,
2769                                    unsigned int block_size,
2770                                    InquiryData_struct *inq_buff,
2771                                    drive_info_struct *drv)
2772 {
2773         int return_code;
2774         unsigned long t;
2775         unsigned char scsi3addr[8];
2776
2777         memset(inq_buff, 0, sizeof(InquiryData_struct));
2778         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2779         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2780                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2781         if (return_code == IO_OK) {
2782                 if (inq_buff->data_byte[8] == 0xFF) {
2783                         dev_warn(&h->pdev->dev,
2784                                "reading geometry failed, volume "
2785                                "does not support reading geometry\n");
2786                         drv->heads = 255;
2787                         drv->sectors = 32;      /* Sectors per track */
2788                         drv->cylinders = total_size + 1;
2789                         drv->raid_level = RAID_UNKNOWN;
2790                 } else {
2791                         drv->heads = inq_buff->data_byte[6];
2792                         drv->sectors = inq_buff->data_byte[7];
2793                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2794                         drv->cylinders += inq_buff->data_byte[5];
2795                         drv->raid_level = inq_buff->data_byte[8];
2796                 }
2797                 drv->block_size = block_size;
2798                 drv->nr_blocks = total_size + 1;
2799                 t = drv->heads * drv->sectors;
2800                 if (t > 1) {
2801                         sector_t real_size = total_size + 1;
2802                         unsigned long rem = sector_div(real_size, t);
2803                         if (rem)
2804                                 real_size++;
2805                         drv->cylinders = real_size;
2806                 }
2807         } else {                /* Get geometry failed */
2808                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2809         }
2810 }
2811
2812 static void
2813 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2814                     unsigned int *block_size)
2815 {
2816         ReadCapdata_struct *buf;
2817         int return_code;
2818         unsigned char scsi3addr[8];
2819
2820         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2821         if (!buf) {
2822                 dev_warn(&h->pdev->dev, "out of memory\n");
2823                 return;
2824         }
2825
2826         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2827         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2828                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2829         if (return_code == IO_OK) {
2830                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2831                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2832         } else {                /* read capacity command failed */
2833                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2834                 *total_size = 0;
2835                 *block_size = BLOCK_SIZE;
2836         }
2837         kfree(buf);
2838 }
2839
2840 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2841         sector_t *total_size, unsigned int *block_size)
2842 {
2843         ReadCapdata_struct_16 *buf;
2844         int return_code;
2845         unsigned char scsi3addr[8];
2846
2847         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2848         if (!buf) {
2849                 dev_warn(&h->pdev->dev, "out of memory\n");
2850                 return;
2851         }
2852
2853         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2854         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2855                 buf, sizeof(ReadCapdata_struct_16),
2856                         0, scsi3addr, TYPE_CMD);
2857         if (return_code == IO_OK) {
2858                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2859                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2860         } else {                /* read capacity command failed */
2861                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2862                 *total_size = 0;
2863                 *block_size = BLOCK_SIZE;
2864         }
2865         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2866                (unsigned long long)*total_size+1, *block_size);
2867         kfree(buf);
2868 }
2869
2870 static int cciss_revalidate(struct gendisk *disk)
2871 {
2872         ctlr_info_t *h = get_host(disk);
2873         drive_info_struct *drv = get_drv(disk);
2874         int logvol;
2875         int FOUND = 0;
2876         unsigned int block_size;
2877         sector_t total_size;
2878         InquiryData_struct *inq_buff = NULL;
2879
2880         for (logvol = 0; logvol <= h->highest_lun; logvol++) {
2881                 if (!h->drv[logvol])
2882                         continue;
2883                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2884                         sizeof(drv->LunID)) == 0) {
2885                         FOUND = 1;
2886                         break;
2887                 }
2888         }
2889
2890         if (!FOUND)
2891                 return 1;
2892
2893         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2894         if (inq_buff == NULL) {
2895                 dev_warn(&h->pdev->dev, "out of memory\n");
2896                 return 1;
2897         }
2898         if (h->cciss_read == CCISS_READ_10) {
2899                 cciss_read_capacity(h, logvol,
2900                                         &total_size, &block_size);
2901         } else {
2902                 cciss_read_capacity_16(h, logvol,
2903                                         &total_size, &block_size);
2904         }
2905         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2906                                inq_buff, drv);
2907
2908         blk_queue_logical_block_size(drv->queue, drv->block_size);
2909         set_capacity(disk, drv->nr_blocks);
2910
2911         kfree(inq_buff);
2912         return 0;
2913 }
2914
2915 /*
2916  * Map (physical) PCI mem into (virtual) kernel space
2917  */
2918 static void __iomem *remap_pci_mem(ulong base, ulong size)
2919 {
2920         ulong page_base = ((ulong) base) & PAGE_MASK;
2921         ulong page_offs = ((ulong) base) - page_base;
2922         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2923
2924         return page_remapped ? (page_remapped + page_offs) : NULL;
2925 }
2926
2927 /*
2928  * Takes jobs of the Q and sends them to the hardware, then puts it on
2929  * the Q to wait for completion.
2930  */
2931 static void start_io(ctlr_info_t *h)
2932 {
2933         CommandList_struct *c;
2934
2935         while (!list_empty(&h->reqQ)) {
2936                 c = list_entry(h->reqQ.next, CommandList_struct, list);
2937                 /* can't do anything if fifo is full */
2938                 if ((h->access.fifo_full(h))) {
2939                         dev_warn(&h->pdev->dev, "fifo full\n");
2940                         break;
2941                 }
2942
2943                 /* Get the first entry from the Request Q */
2944                 removeQ(c);
2945                 h->Qdepth--;
2946
2947                 /* Tell the controller execute command */
2948                 h->access.submit_command(h, c);
2949
2950                 /* Put job onto the completed Q */
2951                 addQ(&h->cmpQ, c);
2952         }
2953 }
2954
2955 /* Assumes that h->lock is held. */
2956 /* Zeros out the error record and then resends the command back */
2957 /* to the controller */
2958 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2959 {
2960         /* erase the old error information */
2961         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2962
2963         /* add it to software queue and then send it to the controller */
2964         addQ(&h->reqQ, c);
2965         h->Qdepth++;
2966         if (h->Qdepth > h->maxQsinceinit)
2967                 h->maxQsinceinit = h->Qdepth;
2968
2969         start_io(h);
2970 }
2971
2972 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2973         unsigned int msg_byte, unsigned int host_byte,
2974         unsigned int driver_byte)
2975 {
2976         /* inverse of macros in scsi.h */
2977         return (scsi_status_byte & 0xff) |
2978                 ((msg_byte & 0xff) << 8) |
2979                 ((host_byte & 0xff) << 16) |
2980                 ((driver_byte & 0xff) << 24);
2981 }
2982
2983 static inline int evaluate_target_status(ctlr_info_t *h,
2984                         CommandList_struct *cmd, int *retry_cmd)
2985 {
2986         unsigned char sense_key;
2987         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2988         int error_value;
2989
2990         *retry_cmd = 0;
2991         /* If we get in here, it means we got "target status", that is, scsi status */
2992         status_byte = cmd->err_info->ScsiStatus;
2993         driver_byte = DRIVER_OK;
2994         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2995
2996         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2997                 host_byte = DID_PASSTHROUGH;
2998         else
2999                 host_byte = DID_OK;
3000
3001         error_value = make_status_bytes(status_byte, msg_byte,
3002                 host_byte, driver_byte);
3003
3004         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3005                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3006                         dev_warn(&h->pdev->dev, "cmd %p "
3007                                "has SCSI Status 0x%x\n",
3008                                cmd, cmd->err_info->ScsiStatus);
3009                 return error_value;
3010         }
3011
3012         /* check the sense key */
3013         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3014         /* no status or recovered error */
3015         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3016             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3017                 error_value = 0;
3018
3019         if (check_for_unit_attention(h, cmd)) {
3020                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3021                 return 0;
3022         }
3023
3024         /* Not SG_IO or similar? */
3025         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3026                 if (error_value != 0)
3027                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3028                                " sense key = 0x%x\n", cmd, sense_key);
3029                 return error_value;
3030         }
3031
3032         /* SG_IO or similar, copy sense data back */
3033         if (cmd->rq->sense) {
3034                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3035                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3036                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3037                         cmd->rq->sense_len);
3038         } else
3039                 cmd->rq->sense_len = 0;
3040
3041         return error_value;
3042 }
3043
3044 /* checks the status of the job and calls complete buffers to mark all
3045  * buffers for the completed job. Note that this function does not need
3046  * to hold the hba/queue lock.
3047  */
3048 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3049                                     int timeout)
3050 {
3051         int retry_cmd = 0;
3052         struct request *rq = cmd->rq;
3053
3054         rq->errors = 0;
3055
3056         if (timeout)
3057                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3058
3059         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3060                 goto after_error_processing;
3061
3062         switch (cmd->err_info->CommandStatus) {
3063         case CMD_TARGET_STATUS:
3064                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3065                 break;
3066         case CMD_DATA_UNDERRUN:
3067                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3068                         dev_warn(&h->pdev->dev, "cmd %p has"
3069                                " completed with data underrun "
3070                                "reported\n", cmd);
3071                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3072                 }
3073                 break;
3074         case CMD_DATA_OVERRUN:
3075                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3076                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3077                                " completed with data overrun "
3078                                "reported\n", cmd);
3079                 break;
3080         case CMD_INVALID:
3081                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3082                        "reported invalid\n", cmd);
3083                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3084                         cmd->err_info->CommandStatus, DRIVER_OK,
3085                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3086                                 DID_PASSTHROUGH : DID_ERROR);
3087                 break;
3088         case CMD_PROTOCOL_ERR:
3089                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3090                        "protocol error\n", cmd);
3091                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3092                         cmd->err_info->CommandStatus, DRIVER_OK,
3093                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3094                                 DID_PASSTHROUGH : DID_ERROR);
3095                 break;
3096         case CMD_HARDWARE_ERR:
3097                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3098                        " hardware error\n", cmd);
3099                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3100                         cmd->err_info->CommandStatus, DRIVER_OK,
3101                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3102                                 DID_PASSTHROUGH : DID_ERROR);
3103                 break;
3104         case CMD_CONNECTION_LOST:
3105                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3106                        "connection lost\n", cmd);
3107                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3108                         cmd->err_info->CommandStatus, DRIVER_OK,
3109                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3110                                 DID_PASSTHROUGH : DID_ERROR);
3111                 break;
3112         case CMD_ABORTED:
3113                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3114                        "aborted\n", cmd);
3115                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3116                         cmd->err_info->CommandStatus, DRIVER_OK,
3117                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3118                                 DID_PASSTHROUGH : DID_ABORT);
3119                 break;
3120         case CMD_ABORT_FAILED:
3121                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3122                        "abort failed\n", cmd);
3123                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3124                         cmd->err_info->CommandStatus, DRIVER_OK,
3125                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3126                                 DID_PASSTHROUGH : DID_ERROR);
3127                 break;
3128         case CMD_UNSOLICITED_ABORT:
3129                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3130                        "abort %p\n", h->ctlr, cmd);
3131                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3132                         retry_cmd = 1;
3133                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3134                         cmd->retry_count++;
3135                 } else
3136                         dev_warn(&h->pdev->dev,
3137                                 "%p retried too many times\n", cmd);
3138                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3139                         cmd->err_info->CommandStatus, DRIVER_OK,
3140                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3141                                 DID_PASSTHROUGH : DID_ABORT);
3142                 break;
3143         case CMD_TIMEOUT:
3144                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3145                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3146                         cmd->err_info->CommandStatus, DRIVER_OK,
3147                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3148                                 DID_PASSTHROUGH : DID_ERROR);
3149                 break;
3150         case CMD_UNABORTABLE:
3151                 dev_warn(&h->pdev->dev, "cmd %p unabortable\n", cmd);
3152                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3153                         cmd->err_info->CommandStatus, DRIVER_OK,
3154                         cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC ?
3155                                 DID_PASSTHROUGH : DID_ERROR);
3156                 break;
3157         default:
3158                 dev_warn(&h->pdev->dev, "cmd %p returned "
3159                        "unknown status %x\n", cmd,
3160                        cmd->err_info->CommandStatus);
3161                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3162                         cmd->err_info->CommandStatus, DRIVER_OK,
3163                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3164                                 DID_PASSTHROUGH : DID_ERROR);
3165         }
3166
3167 after_error_processing:
3168
3169         /* We need to return this command */
3170         if (retry_cmd) {
3171                 resend_cciss_cmd(h, cmd);
3172                 return;
3173         }
3174         cmd->rq->completion_data = cmd;
3175         blk_complete_request(cmd->rq);
3176 }
3177
3178 static inline u32 cciss_tag_contains_index(u32 tag)
3179 {
3180 #define DIRECT_LOOKUP_BIT 0x10
3181         return tag & DIRECT_LOOKUP_BIT;
3182 }
3183
3184 static inline u32 cciss_tag_to_index(u32 tag)
3185 {
3186 #define DIRECT_LOOKUP_SHIFT 5
3187         return tag >> DIRECT_LOOKUP_SHIFT;
3188 }
3189
3190 static inline u32 cciss_tag_discard_error_bits(ctlr_info_t *h, u32 tag)
3191 {
3192 #define CCISS_PERF_ERROR_BITS ((1 << DIRECT_LOOKUP_SHIFT) - 1)
3193 #define CCISS_SIMPLE_ERROR_BITS 0x03
3194         if (likely(h->transMethod & CFGTBL_Trans_Performant))
3195                 return tag & ~CCISS_PERF_ERROR_BITS;
3196         return tag & ~CCISS_SIMPLE_ERROR_BITS;
3197 }
3198
3199 static inline void cciss_mark_tag_indexed(u32 *tag)
3200 {
3201         *tag |= DIRECT_LOOKUP_BIT;
3202 }
3203
3204 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3205 {
3206         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3207 }
3208
3209 /*
3210  * Get a request and submit it to the controller.
3211  */
3212 static void do_cciss_request(struct request_queue *q)
3213 {
3214         ctlr_info_t *h = q->queuedata;
3215         CommandList_struct *c;
3216         sector_t start_blk;
3217         int seg;
3218         struct request *creq;
3219         u64bit temp64;
3220         struct scatterlist *tmp_sg;
3221         SGDescriptor_struct *curr_sg;
3222         drive_info_struct *drv;
3223         int i, dir;
3224         int sg_index = 0;
3225         int chained = 0;
3226
3227       queue:
3228         creq = blk_peek_request(q);
3229         if (!creq)
3230                 goto startio;
3231
3232         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3233
3234         c = cmd_alloc(h);
3235         if (!c)
3236                 goto full;
3237
3238         blk_start_request(creq);
3239
3240         tmp_sg = h->scatter_list[c->cmdindex];
3241         spin_unlock_irq(q->queue_lock);
3242
3243         c->cmd_type = CMD_RWREQ;
3244         c->rq = creq;
3245
3246         /* fill in the request */
3247         drv = creq->rq_disk->private_data;
3248         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3249         /* got command from pool, so use the command block index instead */
3250         /* for direct lookups. */
3251         /* The first 2 bits are reserved for controller error reporting. */
3252         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3253         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3254         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3255         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3256         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3257         c->Request.Type.Attribute = ATTR_SIMPLE;
3258         c->Request.Type.Direction =
3259             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3260         c->Request.Timeout = 0; /* Don't time out */
3261         c->Request.CDB[0] =
3262             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3263         start_blk = blk_rq_pos(creq);
3264         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3265                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3266         sg_init_table(tmp_sg, h->maxsgentries);
3267         seg = blk_rq_map_sg(q, creq, tmp_sg);
3268
3269         /* get the DMA records for the setup */
3270         if (c->Request.Type.Direction == XFER_READ)
3271                 dir = PCI_DMA_FROMDEVICE;
3272         else
3273                 dir = PCI_DMA_TODEVICE;
3274
3275         curr_sg = c->SG;
3276         sg_index = 0;
3277         chained = 0;
3278
3279         for (i = 0; i < seg; i++) {
3280                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3281                         !chained && ((seg - i) > 1)) {
3282                         /* Point to next chain block. */
3283                         curr_sg = h->cmd_sg_list[c->cmdindex];
3284                         sg_index = 0;
3285                         chained = 1;
3286                 }
3287                 curr_sg[sg_index].Len = tmp_sg[i].length;
3288                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3289                                                 tmp_sg[i].offset,
3290                                                 tmp_sg[i].length, dir);
3291                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3292                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3293                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3294                 ++sg_index;
3295         }
3296         if (chained)
3297                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3298                         (seg - (h->max_cmd_sgentries - 1)) *
3299                                 sizeof(SGDescriptor_struct));
3300
3301         /* track how many SG entries we are using */
3302         if (seg > h->maxSG)
3303                 h->maxSG = seg;
3304
3305         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3306                         "chained[%d]\n",
3307                         blk_rq_sectors(creq), seg, chained);
3308
3309         c->Header.SGTotal = seg + chained;
3310         if (seg <= h->max_cmd_sgentries)
3311                 c->Header.SGList = c->Header.SGTotal;
3312         else
3313                 c->Header.SGList = h->max_cmd_sgentries;
3314         set_performant_mode(h, c);
3315
3316         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3317                 if(h->cciss_read == CCISS_READ_10) {
3318                         c->Request.CDB[1] = 0;
3319                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3320                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3321                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3322                         c->Request.CDB[5] = start_blk & 0xff;
3323                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3324                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3325                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3326                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3327                 } else {
3328                         u32 upper32 = upper_32_bits(start_blk);
3329
3330                         c->Request.CDBLen = 16;
3331                         c->Request.CDB[1]= 0;
3332                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3333                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3334                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3335                         c->Request.CDB[5]= upper32 & 0xff;
3336                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3337                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3338                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3339                         c->Request.CDB[9]= start_blk & 0xff;
3340                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3341                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3342                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3343                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3344                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3345                 }
3346         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3347                 c->Request.CDBLen = creq->cmd_len;
3348                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3349         } else {
3350                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3351                         creq->cmd_type);
3352                 BUG();
3353         }
3354
3355         spin_lock_irq(q->queue_lock);
3356
3357         addQ(&h->reqQ, c);
3358         h->Qdepth++;
3359         if (h->Qdepth > h->maxQsinceinit)
3360                 h->maxQsinceinit = h->Qdepth;
3361
3362         goto queue;
3363 full:
3364         blk_stop_queue(q);
3365 startio:
3366         /* We will already have the driver lock here so not need
3367          * to lock it.
3368          */
3369         start_io(h);
3370 }
3371
3372 static inline unsigned long get_next_completion(ctlr_info_t *h)
3373 {
3374         return h->access.command_completed(h);
3375 }
3376
3377 static inline int interrupt_pending(ctlr_info_t *h)
3378 {
3379         return h->access.intr_pending(h);
3380 }
3381
3382 static inline long interrupt_not_for_us(ctlr_info_t *h)
3383 {
3384         return ((h->access.intr_pending(h) == 0) ||
3385                 (h->interrupts_enabled == 0));
3386 }
3387
3388 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3389                         u32 raw_tag)
3390 {
3391         if (unlikely(tag_index >= h->nr_cmds)) {
3392                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3393                 return 1;
3394         }
3395         return 0;
3396 }
3397
3398 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3399                                 u32 raw_tag)
3400 {
3401         removeQ(c);
3402         if (likely(c->cmd_type == CMD_RWREQ))
3403                 complete_command(h, c, 0);
3404         else if (c->cmd_type == CMD_IOCTL_PEND)
3405                 complete(c->waiting);
3406 #ifdef CONFIG_CISS_SCSI_TAPE
3407         else if (c->cmd_type == CMD_SCSI)
3408                 complete_scsi_command(c, 0, raw_tag);
3409 #endif
3410 }
3411
3412 static inline u32 next_command(ctlr_info_t *h)
3413 {
3414         u32 a;
3415
3416         if (unlikely(!(h->transMethod & CFGTBL_Trans_Performant)))
3417                 return h->access.command_completed(h);
3418
3419         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3420                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3421                 (h->reply_pool_head)++;
3422                 h->commands_outstanding--;
3423         } else {
3424                 a = FIFO_EMPTY;
3425         }
3426         /* Check for wraparound */
3427         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3428                 h->reply_pool_head = h->reply_pool;
3429                 h->reply_pool_wraparound ^= 1;
3430         }
3431         return a;
3432 }
3433
3434 /* process completion of an indexed ("direct lookup") command */
3435 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3436 {
3437         u32 tag_index;
3438         CommandList_struct *c;
3439
3440         tag_index = cciss_tag_to_index(raw_tag);
3441         if (bad_tag(h, tag_index, raw_tag))
3442                 return next_command(h);
3443         c = h->cmd_pool + tag_index;
3444         finish_cmd(h, c, raw_tag);
3445         return next_command(h);
3446 }
3447
3448 /* process completion of a non-indexed command */
3449 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3450 {
3451         CommandList_struct *c = NULL;
3452         __u32 busaddr_masked, tag_masked;
3453
3454         tag_masked = cciss_tag_discard_error_bits(h, raw_tag);
3455         list_for_each_entry(c, &h->cmpQ, list) {
3456                 busaddr_masked = cciss_tag_discard_error_bits(h, c->busaddr);
3457                 if (busaddr_masked == tag_masked) {
3458                         finish_cmd(h, c, raw_tag);
3459                         return next_command(h);
3460                 }
3461         }
3462         bad_tag(h, h->nr_cmds + 1, raw_tag);
3463         return next_command(h);
3464 }
3465
3466 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3467 {
3468         ctlr_info_t *h = dev_id;
3469         unsigned long flags;
3470         u32 raw_tag;
3471
3472         if (interrupt_not_for_us(h))
3473                 return IRQ_NONE;
3474         spin_lock_irqsave(&h->lock, flags);
3475         while (interrupt_pending(h)) {
3476                 raw_tag = get_next_completion(h);
3477                 while (raw_tag != FIFO_EMPTY) {
3478                         if (cciss_tag_contains_index(raw_tag))
3479                                 raw_tag = process_indexed_cmd(h, raw_tag);
3480                         else
3481                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3482                 }
3483         }
3484         spin_unlock_irqrestore(&h->lock, flags);
3485         return IRQ_HANDLED;
3486 }
3487
3488 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3489  * check the interrupt pending register because it is not set.
3490  */
3491 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3492 {
3493         ctlr_info_t *h = dev_id;
3494         unsigned long flags;
3495         u32 raw_tag;
3496
3497         spin_lock_irqsave(&h->lock, flags);
3498         raw_tag = get_next_completion(h);
3499         while (raw_tag != FIFO_EMPTY) {
3500                 if (cciss_tag_contains_index(raw_tag))
3501                         raw_tag = process_indexed_cmd(h, raw_tag);
3502                 else
3503                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3504         }
3505         spin_unlock_irqrestore(&h->lock, flags);
3506         return IRQ_HANDLED;
3507 }
3508
3509 /**
3510  * add_to_scan_list() - add controller to rescan queue
3511  * @h:                Pointer to the controller.
3512  *
3513  * Adds the controller to the rescan queue if not already on the queue.
3514  *
3515  * returns 1 if added to the queue, 0 if skipped (could be on the
3516  * queue already, or the controller could be initializing or shutting
3517  * down).
3518  **/
3519 static int add_to_scan_list(struct ctlr_info *h)
3520 {
3521         struct ctlr_info *test_h;
3522         int found = 0;
3523         int ret = 0;
3524
3525         if (h->busy_initializing)
3526                 return 0;
3527
3528         if (!mutex_trylock(&h->busy_shutting_down))
3529                 return 0;
3530
3531         mutex_lock(&scan_mutex);
3532         list_for_each_entry(test_h, &scan_q, scan_list) {
3533                 if (test_h == h) {
3534                         found = 1;
3535                         break;
3536                 }
3537         }
3538         if (!found && !h->busy_scanning) {
3539                 INIT_COMPLETION(h->scan_wait);
3540                 list_add_tail(&h->scan_list, &scan_q);
3541                 ret = 1;
3542         }
3543         mutex_unlock(&scan_mutex);
3544         mutex_unlock(&h->busy_shutting_down);
3545
3546         return ret;
3547 }
3548
3549 /**
3550  * remove_from_scan_list() - remove controller from rescan queue
3551  * @h:                     Pointer to the controller.
3552  *
3553  * Removes the controller from the rescan queue if present. Blocks if
3554  * the controller is currently conducting a rescan.  The controller
3555  * can be in one of three states:
3556  * 1. Doesn't need a scan
3557  * 2. On the scan list, but not scanning yet (we remove it)
3558  * 3. Busy scanning (and not on the list). In this case we want to wait for
3559  *    the scan to complete to make sure the scanning thread for this
3560  *    controller is completely idle.
3561  **/
3562 static void remove_from_scan_list(struct ctlr_info *h)
3563 {
3564         struct ctlr_info *test_h, *tmp_h;
3565
3566         mutex_lock(&scan_mutex);
3567         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3568                 if (test_h == h) { /* state 2. */
3569                         list_del(&h->scan_list);
3570                         complete_all(&h->scan_wait);
3571                         mutex_unlock(&scan_mutex);
3572                         return;
3573                 }
3574         }
3575         if (h->busy_scanning) { /* state 3. */
3576                 mutex_unlock(&scan_mutex);
3577                 wait_for_completion(&h->scan_wait);
3578         } else { /* state 1, nothing to do. */
3579                 mutex_unlock(&scan_mutex);
3580         }
3581 }
3582
3583 /**
3584  * scan_thread() - kernel thread used to rescan controllers
3585  * @data:        Ignored.
3586  *
3587  * A kernel thread used scan for drive topology changes on
3588  * controllers. The thread processes only one controller at a time
3589  * using a queue.  Controllers are added to the queue using
3590  * add_to_scan_list() and removed from the queue either after done
3591  * processing or using remove_from_scan_list().
3592  *
3593  * returns 0.
3594  **/
3595 static int scan_thread(void *data)
3596 {
3597         struct ctlr_info *h;
3598
3599         while (1) {
3600                 set_current_state(TASK_INTERRUPTIBLE);
3601                 schedule();
3602                 if (kthread_should_stop())
3603                         break;
3604
3605                 while (1) {
3606                         mutex_lock(&scan_mutex);
3607                         if (list_empty(&scan_q)) {
3608                                 mutex_unlock(&scan_mutex);
3609                                 break;
3610                         }
3611
3612                         h = list_entry(scan_q.next,
3613                                        struct ctlr_info,
3614                                        scan_list);
3615                         list_del(&h->scan_list);
3616                         h->busy_scanning = 1;
3617                         mutex_unlock(&scan_mutex);
3618
3619                         rebuild_lun_table(h, 0, 0);
3620                         complete_all(&h->scan_wait);
3621                         mutex_lock(&scan_mutex);
3622                         h->busy_scanning = 0;
3623                         mutex_unlock(&scan_mutex);
3624                 }
3625         }
3626
3627         return 0;
3628 }
3629
3630 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3631 {
3632         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3633                 return 0;
3634
3635         switch (c->err_info->SenseInfo[12]) {
3636         case STATE_CHANGED:
3637                 dev_warn(&h->pdev->dev, "a state change "
3638                         "detected, command retried\n");
3639                 return 1;
3640         break;
3641         case LUN_FAILED:
3642                 dev_warn(&h->pdev->dev, "LUN failure "
3643                         "detected, action required\n");
3644                 return 1;
3645         break;
3646         case REPORT_LUNS_CHANGED:
3647                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3648         /*
3649          * Here, we could call add_to_scan_list and wake up the scan thread,
3650          * except that it's quite likely that we will get more than one
3651          * REPORT_LUNS_CHANGED condition in quick succession, which means
3652          * that those which occur after the first one will likely happen
3653          * *during* the scan_thread's rescan.  And the rescan code is not
3654          * robust enough to restart in the middle, undoing what it has already
3655          * done, and it's not clear that it's even possible to do this, since
3656          * part of what it does is notify the block layer, which starts
3657          * doing it's own i/o to read partition tables and so on, and the
3658          * driver doesn't have visibility to know what might need undoing.
3659          * In any event, if possible, it is horribly complicated to get right
3660          * so we just don't do it for now.
3661          *
3662          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3663          */
3664                 return 1;
3665         break;
3666         case POWER_OR_RESET:
3667                 dev_warn(&h->pdev->dev,
3668                         "a power on or device reset detected\n");
3669                 return 1;
3670         break;
3671         case UNIT_ATTENTION_CLEARED:
3672                 dev_warn(&h->pdev->dev,
3673                         "unit attention cleared by another initiator\n");
3674                 return 1;
3675         break;
3676         default:
3677                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3678                 return 1;
3679         }
3680 }
3681
3682 /*
3683  *  We cannot read the structure directly, for portability we must use
3684  *   the io functions.
3685  *   This is for debug only.
3686  */
3687 static void print_cfg_table(ctlr_info_t *h)
3688 {
3689         int i;
3690         char temp_name[17];
3691         CfgTable_struct *tb = h->cfgtable;
3692
3693         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3694         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3695         for (i = 0; i < 4; i++)
3696                 temp_name[i] = readb(&(tb->Signature[i]));
3697         temp_name[4] = '\0';
3698         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3699         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3700                 readl(&(tb->SpecValence)));
3701         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3702                readl(&(tb->TransportSupport)));
3703         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3704                readl(&(tb->TransportActive)));
3705         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3706                readl(&(tb->HostWrite.TransportRequest)));
3707         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3708                readl(&(tb->HostWrite.CoalIntDelay)));
3709         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3710                readl(&(tb->HostWrite.CoalIntCount)));
3711         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3712                readl(&(tb->CmdsOutMax)));
3713         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3714                 readl(&(tb->BusTypes)));
3715         for (i = 0; i < 16; i++)
3716                 temp_name[i] = readb(&(tb->ServerName[i]));
3717         temp_name[16] = '\0';
3718         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3719         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3720                 readl(&(tb->HeartBeat)));
3721 }
3722
3723 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3724 {
3725         int i, offset, mem_type, bar_type;
3726         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3727                 return 0;
3728         offset = 0;
3729         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3730                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3731                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3732                         offset += 4;
3733                 else {
3734                         mem_type = pci_resource_flags(pdev, i) &
3735                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3736                         switch (mem_type) {
3737                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3738                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3739                                 offset += 4;    /* 32 bit */
3740                                 break;
3741                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3742                                 offset += 8;
3743                                 break;
3744                         default:        /* reserved in PCI 2.2 */
3745                                 dev_warn(&pdev->dev,
3746                                        "Base address is invalid\n");
3747                                 return -1;
3748                                 break;
3749                         }
3750                 }
3751                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3752                         return i + 1;
3753         }
3754         return -1;
3755 }
3756
3757 /* Fill in bucket_map[], given nsgs (the max number of
3758  * scatter gather elements supported) and bucket[],
3759  * which is an array of 8 integers.  The bucket[] array
3760  * contains 8 different DMA transfer sizes (in 16
3761  * byte increments) which the controller uses to fetch
3762  * commands.  This function fills in bucket_map[], which
3763  * maps a given number of scatter gather elements to one of
3764  * the 8 DMA transfer sizes.  The point of it is to allow the
3765  * controller to only do as much DMA as needed to fetch the
3766  * command, with the DMA transfer size encoded in the lower
3767  * bits of the command address.
3768  */
3769 static void  calc_bucket_map(int bucket[], int num_buckets,
3770         int nsgs, int *bucket_map)
3771 {
3772         int i, j, b, size;
3773
3774         /* even a command with 0 SGs requires 4 blocks */
3775 #define MINIMUM_TRANSFER_BLOCKS 4
3776 #define NUM_BUCKETS 8
3777         /* Note, bucket_map must have nsgs+1 entries. */
3778         for (i = 0; i <= nsgs; i++) {
3779                 /* Compute size of a command with i SG entries */
3780                 size = i + MINIMUM_TRANSFER_BLOCKS;
3781                 b = num_buckets; /* Assume the biggest bucket */
3782                 /* Find the bucket that is just big enough */
3783                 for (j = 0; j < 8; j++) {
3784                         if (bucket[j] >= size) {
3785                                 b = j;
3786                                 break;
3787                         }
3788                 }
3789                 /* for a command with i SG entries, use bucket b. */
3790                 bucket_map[i] = b;
3791         }
3792 }
3793
3794 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3795 {
3796         int i;
3797
3798         /* under certain very rare conditions, this can take awhile.
3799          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3800          * as we enter this code.) */
3801         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3802                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3803                         break;
3804                 usleep_range(10000, 20000);
3805         }
3806 }
3807
3808 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h,
3809         u32 use_short_tags)
3810 {
3811         /* This is a bit complicated.  There are 8 registers on
3812          * the controller which we write to to tell it 8 different
3813          * sizes of commands which there may be.  It's a way of
3814          * reducing the DMA done to fetch each command.  Encoded into
3815          * each command's tag are 3 bits which communicate to the controller
3816          * which of the eight sizes that command fits within.  The size of
3817          * each command depends on how many scatter gather entries there are.
3818          * Each SG entry requires 16 bytes.  The eight registers are programmed
3819          * with the number of 16-byte blocks a command of that size requires.
3820          * The smallest command possible requires 5 such 16 byte blocks.
3821          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3822          * blocks.  Note, this only extends to the SG entries contained
3823          * within the command block, and does not extend to chained blocks
3824          * of SG elements.   bft[] contains the eight values we write to
3825          * the registers.  They are not evenly distributed, but have more
3826          * sizes for small commands, and fewer sizes for larger commands.
3827          */
3828         __u32 trans_offset;
3829         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3830                         /*
3831                          *  5 = 1 s/g entry or 4k
3832                          *  6 = 2 s/g entry or 8k
3833                          *  8 = 4 s/g entry or 16k
3834                          * 10 = 6 s/g entry or 24k
3835                          */
3836         unsigned long register_value;
3837         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3838
3839         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3840
3841         /* Controller spec: zero out this buffer. */
3842         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3843         h->reply_pool_head = h->reply_pool;
3844
3845         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3846         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3847                                 h->blockFetchTable);
3848         writel(bft[0], &h->transtable->BlockFetch0);
3849         writel(bft[1], &h->transtable->BlockFetch1);
3850         writel(bft[2], &h->transtable->BlockFetch2);
3851         writel(bft[3], &h->transtable->BlockFetch3);
3852         writel(bft[4], &h->transtable->BlockFetch4);
3853         writel(bft[5], &h->transtable->BlockFetch5);
3854         writel(bft[6], &h->transtable->BlockFetch6);
3855         writel(bft[7], &h->transtable->BlockFetch7);
3856
3857         /* size of controller ring buffer */
3858         writel(h->max_commands, &h->transtable->RepQSize);
3859         writel(1, &h->transtable->RepQCount);
3860         writel(0, &h->transtable->RepQCtrAddrLow32);
3861         writel(0, &h->transtable->RepQCtrAddrHigh32);
3862         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3863         writel(0, &h->transtable->RepQAddr0High32);
3864         writel(CFGTBL_Trans_Performant | use_short_tags,
3865                         &(h->cfgtable->HostWrite.TransportRequest));
3866
3867         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3868         cciss_wait_for_mode_change_ack(h);
3869         register_value = readl(&(h->cfgtable->TransportActive));
3870         if (!(register_value & CFGTBL_Trans_Performant))
3871                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3872                                         " performant mode\n");
3873 }
3874
3875 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3876 {
3877         __u32 trans_support;
3878
3879         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3880         /* Attempt to put controller into performant mode if supported */
3881         /* Does board support performant mode? */
3882         trans_support = readl(&(h->cfgtable->TransportSupport));
3883         if (!(trans_support & PERFORMANT_MODE))
3884                 return;
3885
3886         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3887         /* Performant mode demands commands on a 32 byte boundary
3888          * pci_alloc_consistent aligns on page boundarys already.
3889          * Just need to check if divisible by 32
3890          */
3891         if ((sizeof(CommandList_struct) % 32) != 0) {
3892                 dev_warn(&h->pdev->dev, "%s %d %s\n",
3893                         "cciss info: command size[",
3894                         (int)sizeof(CommandList_struct),
3895                         "] not divisible by 32, no performant mode..\n");
3896                 return;
3897         }
3898
3899         /* Performant mode ring buffer and supporting data structures */
3900         h->reply_pool = (__u64 *)pci_alloc_consistent(
3901                 h->pdev, h->max_commands * sizeof(__u64),
3902                 &(h->reply_pool_dhandle));
3903
3904         /* Need a block fetch table for performant mode */
3905         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3906                 sizeof(__u32)), GFP_KERNEL);
3907
3908         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3909                 goto clean_up;
3910
3911         cciss_enter_performant_mode(h,
3912                 trans_support & CFGTBL_Trans_use_short_tags);
3913
3914         /* Change the access methods to the performant access methods */
3915         h->access = SA5_performant_access;
3916         h->transMethod = CFGTBL_Trans_Performant;
3917
3918         return;
3919 clean_up:
3920         kfree(h->blockFetchTable);
3921         if (h->reply_pool)
3922                 pci_free_consistent(h->pdev,
3923                                 h->max_commands * sizeof(__u64),
3924                                 h->reply_pool,
3925                                 h->reply_pool_dhandle);
3926         return;
3927
3928 } /* cciss_put_controller_into_performant_mode */
3929
3930 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3931  * controllers that are capable. If not, we use IO-APIC mode.
3932  */
3933
3934 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3935 {
3936 #ifdef CONFIG_PCI_MSI
3937         int err;
3938         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3939         {0, 2}, {0, 3}
3940         };
3941
3942         /* Some boards advertise MSI but don't really support it */
3943         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3944             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3945                 goto default_int_mode;
3946
3947         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3948                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3949                 if (!err) {
3950                         h->intr[0] = cciss_msix_entries[0].vector;
3951                         h->intr[1] = cciss_msix_entries[1].vector;
3952                         h->intr[2] = cciss_msix_entries[2].vector;
3953                         h->intr[3] = cciss_msix_entries[3].vector;
3954                         h->msix_vector = 1;
3955                         return;
3956                 }
3957                 if (err > 0) {
3958                         dev_warn(&h->pdev->dev,
3959                                 "only %d MSI-X vectors available\n", err);
3960                         goto default_int_mode;
3961                 } else {
3962                         dev_warn(&h->pdev->dev,
3963                                 "MSI-X init failed %d\n", err);
3964                         goto default_int_mode;
3965                 }
3966         }
3967         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3968                 if (!pci_enable_msi(h->pdev))
3969                         h->msi_vector = 1;
3970                 else
3971                         dev_warn(&h->pdev->dev, "MSI init failed\n");
3972         }
3973 default_int_mode:
3974 #endif                          /* CONFIG_PCI_MSI */
3975         /* if we get here we're going to use the default interrupt mode */
3976         h->intr[PERF_MODE_INT] = h->pdev->irq;
3977         return;
3978 }
3979
3980 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3981 {
3982         int i;
3983         u32 subsystem_vendor_id, subsystem_device_id;
3984
3985         subsystem_vendor_id = pdev->subsystem_vendor;
3986         subsystem_device_id = pdev->subsystem_device;
3987         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3988                         subsystem_vendor_id;
3989
3990         for (i = 0; i < ARRAY_SIZE(products); i++)
3991                 if (*board_id == products[i].board_id)
3992                         return i;
3993         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
3994                 *board_id);
3995         return -ENODEV;
3996 }
3997
3998 static inline bool cciss_board_disabled(ctlr_info_t *h)
3999 {
4000         u16 command;
4001
4002         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4003         return ((command & PCI_COMMAND_MEMORY) == 0);
4004 }
4005
4006 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4007         unsigned long *memory_bar)
4008 {
4009         int i;
4010
4011         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4012                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4013                         /* addressing mode bits already removed */
4014                         *memory_bar = pci_resource_start(pdev, i);
4015                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4016                                 *memory_bar);
4017                         return 0;
4018                 }
4019         dev_warn(&pdev->dev, "no memory BAR found\n");
4020         return -ENODEV;
4021 }
4022
4023 static int __devinit cciss_wait_for_board_state(struct pci_dev *pdev,
4024         void __iomem *vaddr, int wait_for_ready)
4025 #define BOARD_READY 1
4026 #define BOARD_NOT_READY 0
4027 {
4028         int i, iterations;
4029         u32 scratchpad;
4030
4031         if (wait_for_ready)
4032                 iterations = CCISS_BOARD_READY_ITERATIONS;
4033         else
4034                 iterations = CCISS_BOARD_NOT_READY_ITERATIONS;
4035
4036         for (i = 0; i < iterations; i++) {
4037                 scratchpad = readl(vaddr + SA5_SCRATCHPAD_OFFSET);
4038                 if (wait_for_ready) {
4039                         if (scratchpad == CCISS_FIRMWARE_READY)
4040                                 return 0;
4041                 } else {
4042                         if (scratchpad != CCISS_FIRMWARE_READY)
4043                                 return 0;
4044                 }
4045                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4046         }
4047         dev_warn(&pdev->dev, "board not ready, timed out.\n");
4048         return -ENODEV;
4049 }
4050
4051 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4052         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4053         u64 *cfg_offset)
4054 {
4055         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4056         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4057         *cfg_base_addr &= (u32) 0x0000ffff;
4058         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4059         if (*cfg_base_addr_index == -1) {
4060                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4061                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4062                 return -ENODEV;
4063         }
4064         return 0;
4065 }
4066
4067 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4068 {
4069         u64 cfg_offset;
4070         u32 cfg_base_addr;
4071         u64 cfg_base_addr_index;
4072         u32 trans_offset;
4073         int rc;
4074
4075         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4076                 &cfg_base_addr_index, &cfg_offset);
4077         if (rc)
4078                 return rc;
4079         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4080                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4081         if (!h->cfgtable)
4082                 return -ENOMEM;
4083         rc = write_driver_ver_to_cfgtable(h->cfgtable);
4084         if (rc)
4085                 return rc;
4086         /* Find performant mode table. */
4087         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4088         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4089                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4090                                 sizeof(*h->transtable));
4091         if (!h->transtable)
4092                 return -ENOMEM;
4093         return 0;
4094 }
4095
4096 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4097 {
4098         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4099
4100         /* Limit commands in memory limited kdump scenario. */
4101         if (reset_devices && h->max_commands > 32)
4102                 h->max_commands = 32;
4103
4104         if (h->max_commands < 16) {
4105                 dev_warn(&h->pdev->dev, "Controller reports "
4106                         "max supported commands of %d, an obvious lie. "
4107                         "Using 16.  Ensure that firmware is up to date.\n",
4108                         h->max_commands);
4109                 h->max_commands = 16;
4110         }
4111 }
4112
4113 /* Interrogate the hardware for some limits:
4114  * max commands, max SG elements without chaining, and with chaining,
4115  * SG chain block size, etc.
4116  */
4117 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4118 {
4119         cciss_get_max_perf_mode_cmds(h);
4120         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4121         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4122         /*
4123          * Limit in-command s/g elements to 32 save dma'able memory.
4124          * Howvever spec says if 0, use 31
4125          */
4126         h->max_cmd_sgentries = 31;
4127         if (h->maxsgentries > 512) {
4128                 h->max_cmd_sgentries = 32;
4129                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4130                 h->maxsgentries--; /* save one for chain pointer */
4131         } else {
4132                 h->maxsgentries = 31; /* default to traditional values */
4133                 h->chainsize = 0;
4134         }
4135 }
4136
4137 static inline bool CISS_signature_present(ctlr_info_t *h)
4138 {
4139         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4140             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4141             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4142             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4143                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4144                 return false;
4145         }
4146         return true;
4147 }
4148
4149 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4150 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4151 {
4152 #ifdef CONFIG_X86
4153         u32 prefetch;
4154
4155         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4156         prefetch |= 0x100;
4157         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4158 #endif
4159 }
4160
4161 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4162  * in a prefetch beyond physical memory.
4163  */
4164 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4165 {
4166         u32 dma_prefetch;
4167         __u32 dma_refetch;
4168
4169         if (h->board_id != 0x3225103C)
4170                 return;
4171         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4172         dma_prefetch |= 0x8000;
4173         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4174         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4175         dma_refetch |= 0x1;
4176         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4177 }
4178
4179 static int __devinit cciss_pci_init(ctlr_info_t *h)
4180 {
4181         int prod_index, err;
4182
4183         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4184         if (prod_index < 0)
4185                 return -ENODEV;
4186         h->product_name = products[prod_index].product_name;
4187         h->access = *(products[prod_index].access);
4188
4189         if (cciss_board_disabled(h)) {
4190                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4191                 return -ENODEV;
4192         }
4193         err = pci_enable_device(h->pdev);
4194         if (err) {
4195                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4196                 return err;
4197         }
4198
4199         err = pci_request_regions(h->pdev, "cciss");
4200         if (err) {
4201                 dev_warn(&h->pdev->dev,
4202                         "Cannot obtain PCI resources, aborting\n");
4203                 return err;
4204         }
4205
4206         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4207         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4208
4209 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4210  * else we use the IO-APIC interrupt assigned to us by system ROM.
4211  */
4212         cciss_interrupt_mode(h);
4213         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4214         if (err)
4215                 goto err_out_free_res;
4216         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4217         if (!h->vaddr) {
4218                 err = -ENOMEM;
4219                 goto err_out_free_res;
4220         }
4221         err = cciss_wait_for_board_state(h->pdev, h->vaddr, BOARD_READY);
4222         if (err)
4223                 goto err_out_free_res;
4224         err = cciss_find_cfgtables(h);
4225         if (err)
4226                 goto err_out_free_res;
4227         print_cfg_table(h);
4228         cciss_find_board_params(h);
4229
4230         if (!CISS_signature_present(h)) {
4231                 err = -ENODEV;
4232                 goto err_out_free_res;
4233         }
4234         cciss_enable_scsi_prefetch(h);
4235         cciss_p600_dma_prefetch_quirk(h);
4236         cciss_put_controller_into_performant_mode(h);
4237         return 0;
4238
4239 err_out_free_res:
4240         /*
4241          * Deliberately omit pci_disable_device(): it does something nasty to
4242          * Smart Array controllers that pci_enable_device does not undo
4243          */
4244         if (h->transtable)
4245                 iounmap(h->transtable);
4246         if (h->cfgtable)
4247                 iounmap(h->cfgtable);
4248         if (h->vaddr)
4249                 iounmap(h->vaddr);
4250         pci_release_regions(h->pdev);
4251         return err;
4252 }
4253
4254 /* Function to find the first free pointer into our hba[] array
4255  * Returns -1 if no free entries are left.
4256  */
4257 static int alloc_cciss_hba(struct pci_dev *pdev)
4258 {
4259         int i;
4260
4261         for (i = 0; i < MAX_CTLR; i++) {
4262                 if (!hba[i]) {
4263                         ctlr_info_t *h;
4264
4265                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4266                         if (!h)
4267                                 goto Enomem;
4268                         hba[i] = h;
4269                         return i;
4270                 }
4271         }
4272         dev_warn(&pdev->dev, "This driver supports a maximum"
4273                " of %d controllers.\n", MAX_CTLR);
4274         return -1;
4275 Enomem:
4276         dev_warn(&pdev->dev, "out of memory.\n");
4277         return -1;
4278 }
4279
4280 static void free_hba(ctlr_info_t *h)
4281 {
4282         int i;
4283
4284         hba[h->ctlr] = NULL;
4285         for (i = 0; i < h->highest_lun + 1; i++)
4286                 if (h->gendisk[i] != NULL)
4287                         put_disk(h->gendisk[i]);
4288         kfree(h);
4289 }
4290
4291 /* Send a message CDB to the firmware. */
4292 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4293 {
4294         typedef struct {
4295                 CommandListHeader_struct CommandHeader;
4296                 RequestBlock_struct Request;
4297                 ErrDescriptor_struct ErrorDescriptor;
4298         } Command;
4299         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4300         Command *cmd;
4301         dma_addr_t paddr64;
4302         uint32_t paddr32, tag;
4303         void __iomem *vaddr;
4304         int i, err;
4305
4306         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4307         if (vaddr == NULL)
4308                 return -ENOMEM;
4309
4310         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4311            CCISS commands, so they must be allocated from the lower 4GiB of
4312            memory. */
4313         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4314         if (err) {
4315                 iounmap(vaddr);
4316                 return -ENOMEM;
4317         }
4318
4319         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4320         if (cmd == NULL) {
4321                 iounmap(vaddr);
4322                 return -ENOMEM;
4323         }
4324
4325         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4326            although there's no guarantee, we assume that the address is at
4327            least 4-byte aligned (most likely, it's page-aligned). */
4328         paddr32 = paddr64;
4329
4330         cmd->CommandHeader.ReplyQueue = 0;
4331         cmd->CommandHeader.SGList = 0;
4332         cmd->CommandHeader.SGTotal = 0;
4333         cmd->CommandHeader.Tag.lower = paddr32;
4334         cmd->CommandHeader.Tag.upper = 0;
4335         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4336
4337         cmd->Request.CDBLen = 16;
4338         cmd->Request.Type.Type = TYPE_MSG;
4339         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4340         cmd->Request.Type.Direction = XFER_NONE;
4341         cmd->Request.Timeout = 0; /* Don't time out */
4342         cmd->Request.CDB[0] = opcode;
4343         cmd->Request.CDB[1] = type;
4344         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4345
4346         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4347         cmd->ErrorDescriptor.Addr.upper = 0;
4348         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4349
4350         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4351
4352         for (i = 0; i < 10; i++) {
4353                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4354                 if ((tag & ~3) == paddr32)
4355                         break;
4356                 schedule_timeout_uninterruptible(HZ);
4357         }
4358
4359         iounmap(vaddr);
4360
4361         /* we leak the DMA buffer here ... no choice since the controller could
4362            still complete the command. */
4363         if (i == 10) {
4364                 dev_err(&pdev->dev,
4365                         "controller message %02x:%02x timed out\n",
4366                         opcode, type);
4367                 return -ETIMEDOUT;
4368         }
4369
4370         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4371
4372         if (tag & 2) {
4373                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4374                         opcode, type);
4375                 return -EIO;
4376         }
4377
4378         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4379                 opcode, type);
4380         return 0;
4381 }
4382
4383 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4384 #define cciss_noop(p) cciss_message(p, 3, 0)
4385
4386 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4387         void * __iomem vaddr, bool use_doorbell)
4388 {
4389         u16 pmcsr;
4390         int pos;
4391
4392         if (use_doorbell) {
4393                 /* For everything after the P600, the PCI power state method
4394                  * of resetting the controller doesn't work, so we have this
4395                  * other way using the doorbell register.
4396                  */
4397                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4398                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4399                 msleep(1000);
4400         } else { /* Try to do it the PCI power state way */
4401
4402                 /* Quoting from the Open CISS Specification: "The Power
4403                  * Management Control/Status Register (CSR) controls the power
4404                  * state of the device.  The normal operating state is D0,
4405                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4406                  * the controller, place the interface device in D3 then to D0,
4407                  * this causes a secondary PCI reset which will reset the
4408                  * controller." */
4409
4410                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4411                 if (pos == 0) {
4412                         dev_err(&pdev->dev,
4413                                 "cciss_controller_hard_reset: "
4414                                 "PCI PM not supported\n");
4415                         return -ENODEV;
4416                 }
4417                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4418                 /* enter the D3hot power management state */
4419                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4420                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4421                 pmcsr |= PCI_D3hot;
4422                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4423
4424                 msleep(500);
4425
4426                 /* enter the D0 power management state */
4427                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4428                 pmcsr |= PCI_D0;
4429                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4430
4431                 msleep(500);
4432         }
4433         return 0;
4434 }
4435
4436 static __devinit void init_driver_version(char *driver_version, int len)
4437 {
4438         memset(driver_version, 0, len);
4439         strncpy(driver_version, "cciss " DRIVER_NAME, len - 1);
4440 }
4441
4442 static __devinit int write_driver_ver_to_cfgtable(
4443         CfgTable_struct __iomem *cfgtable)
4444 {
4445         char *driver_version;
4446         int i, size = sizeof(cfgtable->driver_version);
4447
4448         driver_version = kmalloc(size, GFP_KERNEL);
4449         if (!driver_version)
4450                 return -ENOMEM;
4451
4452         init_driver_version(driver_version, size);
4453         for (i = 0; i < size; i++)
4454                 writeb(driver_version[i], &cfgtable->driver_version[i]);
4455         kfree(driver_version);
4456         return 0;
4457 }
4458
4459 static __devinit void read_driver_ver_from_cfgtable(
4460         CfgTable_struct __iomem *cfgtable, unsigned char *driver_ver)
4461 {
4462         int i;
4463
4464         for (i = 0; i < sizeof(cfgtable->driver_version); i++)
4465                 driver_ver[i] = readb(&cfgtable->driver_version[i]);
4466 }
4467
4468 static __devinit int controller_reset_failed(
4469         CfgTable_struct __iomem *cfgtable)
4470 {
4471
4472         char *driver_ver, *old_driver_ver;
4473         int rc, size = sizeof(cfgtable->driver_version);
4474
4475         old_driver_ver = kmalloc(2 * size, GFP_KERNEL);
4476         if (!old_driver_ver)
4477                 return -ENOMEM;
4478         driver_ver = old_driver_ver + size;
4479
4480         /* After a reset, the 32 bytes of "driver version" in the cfgtable
4481          * should have been changed, otherwise we know the reset failed.
4482          */
4483         init_driver_version(old_driver_ver, size);
4484         read_driver_ver_from_cfgtable(cfgtable, driver_ver);
4485         rc = !memcmp(driver_ver, old_driver_ver, size);
4486         kfree(old_driver_ver);
4487         return rc;
4488 }
4489
4490 /* This does a hard reset of the controller using PCI power management
4491  * states or using the doorbell register. */
4492 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4493 {
4494         u64 cfg_offset;
4495         u32 cfg_base_addr;
4496         u64 cfg_base_addr_index;
4497         void __iomem *vaddr;
4498         unsigned long paddr;
4499         u32 misc_fw_support;
4500         int rc;
4501         CfgTable_struct __iomem *cfgtable;
4502         bool use_doorbell;
4503         u32 board_id;
4504         u16 command_register;
4505
4506         /* For controllers as old a the p600, this is very nearly
4507          * the same thing as
4508          *
4509          * pci_save_state(pci_dev);
4510          * pci_set_power_state(pci_dev, PCI_D3hot);
4511          * pci_set_power_state(pci_dev, PCI_D0);
4512          * pci_restore_state(pci_dev);
4513          *
4514          * For controllers newer than the P600, the pci power state
4515          * method of resetting doesn't work so we have another way
4516          * using the doorbell register.
4517          */
4518
4519         /* Exclude 640x boards.  These are two pci devices in one slot
4520          * which share a battery backed cache module.  One controls the
4521          * cache, the other accesses the cache through the one that controls
4522          * it.  If we reset the one controlling the cache, the other will
4523          * likely not be happy.  Just forbid resetting this conjoined mess.
4524          */
4525         cciss_lookup_board_id(pdev, &board_id);
4526         if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4527                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4528                                 "due to shared cache module.");
4529                 return -ENODEV;
4530         }
4531
4532         /* Save the PCI command register */
4533         pci_read_config_word(pdev, 4, &command_register);
4534         /* Turn the board off.  This is so that later pci_restore_state()
4535          * won't turn the board on before the rest of config space is ready.
4536          */
4537         pci_disable_device(pdev);
4538         pci_save_state(pdev);
4539
4540         /* find the first memory BAR, so we can find the cfg table */
4541         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4542         if (rc)
4543                 return rc;
4544         vaddr = remap_pci_mem(paddr, 0x250);
4545         if (!vaddr)
4546                 return -ENOMEM;
4547
4548         /* find cfgtable in order to check if reset via doorbell is supported */
4549         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4550                                         &cfg_base_addr_index, &cfg_offset);
4551         if (rc)
4552                 goto unmap_vaddr;
4553         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4554                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4555         if (!cfgtable) {
4556                 rc = -ENOMEM;
4557                 goto unmap_vaddr;
4558         }
4559         rc = write_driver_ver_to_cfgtable(cfgtable);
4560         if (rc)
4561                 goto unmap_vaddr;
4562
4563         /* If reset via doorbell register is supported, use that. */
4564         misc_fw_support = readl(&cfgtable->misc_fw_support);
4565         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4566
4567         /* The doorbell reset seems to cause lockups on some Smart
4568          * Arrays (e.g. P410, P410i, maybe others).  Until this is
4569          * fixed or at least isolated, avoid the doorbell reset.
4570          */
4571         use_doorbell = 0;
4572
4573         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4574         if (rc)
4575                 goto unmap_cfgtable;
4576         pci_restore_state(pdev);
4577         rc = pci_enable_device(pdev);
4578         if (rc) {
4579                 dev_warn(&pdev->dev, "failed to enable device.\n");
4580                 goto unmap_cfgtable;
4581         }
4582         pci_write_config_word(pdev, 4, command_register);
4583
4584         /* Some devices (notably the HP Smart Array 5i Controller)
4585            need a little pause here */
4586         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4587
4588         /* Wait for board to become not ready, then ready. */
4589         dev_info(&pdev->dev, "Waiting for board to reset.\n");
4590         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_NOT_READY);
4591         if (rc) /* Don't bail, might be E500, etc. which can't be reset */
4592                 dev_warn(&pdev->dev,
4593                         "failed waiting for board to reset\n");
4594         rc = cciss_wait_for_board_state(pdev, vaddr, BOARD_READY);
4595         if (rc) {
4596                 dev_warn(&pdev->dev,
4597                         "failed waiting for board to become ready\n");
4598                 goto unmap_cfgtable;
4599         }
4600
4601         rc = controller_reset_failed(vaddr);
4602         if (rc < 0)
4603                 goto unmap_cfgtable;
4604         if (rc) {
4605                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4606                         " Ignoring controller.\n");
4607                 rc = -ENODEV;
4608                 goto unmap_cfgtable;
4609         } else {
4610                 dev_info(&pdev->dev, "board ready.\n");
4611         }
4612
4613         dev_info(&pdev->dev, "board ready.\n");
4614
4615 unmap_cfgtable:
4616         iounmap(cfgtable);
4617
4618 unmap_vaddr:
4619         iounmap(vaddr);
4620         return rc;
4621 }
4622
4623 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4624 {
4625         int rc, i;
4626
4627         if (!reset_devices)
4628                 return 0;
4629
4630         /* Reset the controller with a PCI power-cycle or via doorbell */
4631         rc = cciss_kdump_hard_reset_controller(pdev);
4632
4633         /* -ENOTSUPP here means we cannot reset the controller
4634          * but it's already (and still) up and running in
4635          * "performant mode".  Or, it might be 640x, which can't reset
4636          * due to concerns about shared bbwc between 6402/6404 pair.
4637          */
4638         if (rc == -ENOTSUPP)
4639                 return 0; /* just try to do the kdump anyhow. */
4640         if (rc)
4641                 return -ENODEV;
4642
4643         /* Now try to get the controller to respond to a no-op */
4644         dev_warn(&pdev->dev, "Waiting for controller to respond to no-op\n");
4645         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4646                 if (cciss_noop(pdev) == 0)
4647                         break;
4648                 else
4649                         dev_warn(&pdev->dev, "no-op failed%s\n",
4650                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4651                                         "; re-trying" : ""));
4652                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4653         }
4654         return 0;
4655 }
4656
4657 static __devinit int cciss_allocate_cmd_pool(ctlr_info_t *h)
4658 {
4659         h->cmd_pool_bits = kmalloc(
4660                 DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG) *
4661                 sizeof(unsigned long), GFP_KERNEL);
4662         h->cmd_pool = pci_alloc_consistent(h->pdev,
4663                 h->nr_cmds * sizeof(CommandList_struct),
4664                 &(h->cmd_pool_dhandle));
4665         h->errinfo_pool = pci_alloc_consistent(h->pdev,
4666                 h->nr_cmds * sizeof(ErrorInfo_struct),
4667                 &(h->errinfo_pool_dhandle));
4668         if ((h->cmd_pool_bits == NULL)
4669                 || (h->cmd_pool == NULL)
4670                 || (h->errinfo_pool == NULL)) {
4671                 dev_err(&h->pdev->dev, "out of memory");
4672                 return -ENOMEM;
4673         }
4674         return 0;
4675 }
4676
4677 static __devinit int cciss_allocate_scatterlists(ctlr_info_t *h)
4678 {
4679         int i;
4680
4681         /* zero it, so that on free we need not know how many were alloc'ed */
4682         h->scatter_list = kzalloc(h->max_commands *
4683                                 sizeof(struct scatterlist *), GFP_KERNEL);
4684         if (!h->scatter_list)
4685                 return -ENOMEM;
4686
4687         for (i = 0; i < h->nr_cmds; i++) {
4688                 h->scatter_list[i] = kmalloc(sizeof(struct scatterlist) *
4689                                                 h->maxsgentries, GFP_KERNEL);
4690                 if (h->scatter_list[i] == NULL) {
4691                         dev_err(&h->pdev->dev, "could not allocate "
4692                                 "s/g lists\n");
4693                         return -ENOMEM;
4694                 }
4695         }
4696         return 0;
4697 }
4698
4699 static void cciss_free_scatterlists(ctlr_info_t *h)
4700 {
4701         int i;
4702
4703         if (h->scatter_list) {
4704                 for (i = 0; i < h->nr_cmds; i++)
4705                         kfree(h->scatter_list[i]);
4706                 kfree(h->scatter_list);
4707         }
4708 }
4709
4710 static void cciss_free_cmd_pool(ctlr_info_t *h)
4711 {
4712         kfree(h->cmd_pool_bits);
4713         if (h->cmd_pool)
4714                 pci_free_consistent(h->pdev,
4715                         h->nr_cmds * sizeof(CommandList_struct),
4716                         h->cmd_pool, h->cmd_pool_dhandle);
4717         if (h->errinfo_pool)
4718                 pci_free_consistent(h->pdev,
4719                         h->nr_cmds * sizeof(ErrorInfo_struct),
4720                         h->errinfo_pool, h->errinfo_pool_dhandle);
4721 }
4722
4723 static int cciss_request_irq(ctlr_info_t *h,
4724         irqreturn_t (*msixhandler)(int, void *),
4725         irqreturn_t (*intxhandler)(int, void *))
4726 {
4727         if (h->msix_vector || h->msi_vector) {
4728                 if (!request_irq(h->intr[PERF_MODE_INT], msixhandler,
4729                                 IRQF_DISABLED, h->devname, h))
4730                         return 0;
4731                 dev_err(&h->pdev->dev, "Unable to get msi irq %d"
4732                         " for %s\n", h->intr[PERF_MODE_INT],
4733                         h->devname);
4734                 return -1;
4735         }
4736
4737         if (!request_irq(h->intr[PERF_MODE_INT], intxhandler,
4738                         IRQF_DISABLED, h->devname, h))
4739                 return 0;
4740         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4741                 h->intr[PERF_MODE_INT], h->devname);
4742         return -1;
4743 }
4744
4745 /*
4746  *  This is it.  Find all the controllers and register them.  I really hate
4747  *  stealing all these major device numbers.
4748  *  returns the number of block devices registered.
4749  */
4750 static int __devinit cciss_init_one(struct pci_dev *pdev,
4751                                     const struct pci_device_id *ent)
4752 {
4753         int i;
4754         int j = 0;
4755         int rc;
4756         int dac, return_code;
4757         InquiryData_struct *inq_buff;
4758         ctlr_info_t *h;
4759
4760         rc = cciss_init_reset_devices(pdev);
4761         if (rc)
4762                 return rc;
4763         i = alloc_cciss_hba(pdev);
4764         if (i < 0)
4765                 return -1;
4766
4767         h = hba[i];
4768         h->pdev = pdev;
4769         h->busy_initializing = 1;
4770         INIT_LIST_HEAD(&h->cmpQ);
4771         INIT_LIST_HEAD(&h->reqQ);
4772         mutex_init(&h->busy_shutting_down);
4773
4774         if (cciss_pci_init(h) != 0)
4775                 goto clean_no_release_regions;
4776
4777         sprintf(h->devname, "cciss%d", i);
4778         h->ctlr = i;
4779
4780         init_completion(&h->scan_wait);
4781
4782         if (cciss_create_hba_sysfs_entry(h))
4783                 goto clean0;
4784
4785         /* configure PCI DMA stuff */
4786         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4787                 dac = 1;
4788         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4789                 dac = 0;
4790         else {
4791                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4792                 goto clean1;
4793         }
4794
4795         /*
4796          * register with the major number, or get a dynamic major number
4797          * by passing 0 as argument.  This is done for greater than
4798          * 8 controller support.
4799          */
4800         if (i < MAX_CTLR_ORIG)
4801                 h->major = COMPAQ_CISS_MAJOR + i;
4802         rc = register_blkdev(h->major, h->devname);
4803         if (rc == -EBUSY || rc == -EINVAL) {
4804                 dev_err(&h->pdev->dev,
4805                        "Unable to get major number %d for %s "
4806                        "on hba %d\n", h->major, h->devname, i);
4807                 goto clean1;
4808         } else {
4809                 if (i >= MAX_CTLR_ORIG)
4810                         h->major = rc;
4811         }
4812
4813         /* make sure the board interrupts are off */
4814         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4815         rc = cciss_request_irq(h, do_cciss_msix_intr, do_cciss_intx);
4816         if (rc)
4817                 goto clean2;
4818
4819         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4820                h->devname, pdev->device, pci_name(pdev),
4821                h->intr[PERF_MODE_INT], dac ? "" : " not");
4822
4823         if (cciss_allocate_cmd_pool(h))
4824                 goto clean4;
4825
4826         if (cciss_allocate_scatterlists(h))
4827                 goto clean4;
4828
4829         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4830                 h->chainsize, h->nr_cmds);
4831         if (!h->cmd_sg_list && h->chainsize > 0)
4832                 goto clean4;
4833
4834         spin_lock_init(&h->lock);
4835
4836         /* Initialize the pdev driver private data.
4837            have it point to h.  */
4838         pci_set_drvdata(pdev, h);
4839         /* command and error info recs zeroed out before
4840            they are used */
4841         memset(h->cmd_pool_bits, 0,
4842                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4843                         * sizeof(unsigned long));
4844
4845         h->num_luns = 0;
4846         h->highest_lun = -1;
4847         for (j = 0; j < CISS_MAX_LUN; j++) {
4848                 h->drv[j] = NULL;
4849                 h->gendisk[j] = NULL;
4850         }
4851
4852         cciss_scsi_setup(h);
4853
4854         /* Turn the interrupts on so we can service requests */
4855         h->access.set_intr_mask(h, CCISS_INTR_ON);
4856
4857         /* Get the firmware version */
4858         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4859         if (inq_buff == NULL) {
4860                 dev_err(&h->pdev->dev, "out of memory\n");
4861                 goto clean4;
4862         }
4863
4864         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4865                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4866         if (return_code == IO_OK) {
4867                 h->firm_ver[0] = inq_buff->data_byte[32];
4868                 h->firm_ver[1] = inq_buff->data_byte[33];
4869                 h->firm_ver[2] = inq_buff->data_byte[34];
4870                 h->firm_ver[3] = inq_buff->data_byte[35];
4871         } else {         /* send command failed */
4872                 dev_warn(&h->pdev->dev, "unable to determine firmware"
4873                         " version of controller\n");
4874         }
4875         kfree(inq_buff);
4876
4877         cciss_procinit(h);
4878
4879         h->cciss_max_sectors = 8192;
4880
4881         rebuild_lun_table(h, 1, 0);
4882         h->busy_initializing = 0;
4883         return 1;
4884
4885 clean4:
4886         cciss_free_cmd_pool(h);
4887         cciss_free_scatterlists(h);
4888         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4889         free_irq(h->intr[PERF_MODE_INT], h);
4890 clean2:
4891         unregister_blkdev(h->major, h->devname);
4892 clean1:
4893         cciss_destroy_hba_sysfs_entry(h);
4894 clean0:
4895         pci_release_regions(pdev);
4896 clean_no_release_regions:
4897         h->busy_initializing = 0;
4898
4899         /*
4900          * Deliberately omit pci_disable_device(): it does something nasty to
4901          * Smart Array controllers that pci_enable_device does not undo
4902          */
4903         pci_set_drvdata(pdev, NULL);
4904         free_hba(h);
4905         return -1;
4906 }
4907
4908 static void cciss_shutdown(struct pci_dev *pdev)
4909 {
4910         ctlr_info_t *h;
4911         char *flush_buf;
4912         int return_code;
4913
4914         h = pci_get_drvdata(pdev);
4915         flush_buf = kzalloc(4, GFP_KERNEL);
4916         if (!flush_buf) {
4917                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
4918                 return;
4919         }
4920         /* write all data in the battery backed cache to disk */
4921         memset(flush_buf, 0, 4);
4922         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4923                 4, 0, CTLR_LUNID, TYPE_CMD);
4924         kfree(flush_buf);
4925         if (return_code != IO_OK)
4926                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
4927         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4928         free_irq(h->intr[PERF_MODE_INT], h);
4929 }
4930
4931 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4932 {
4933         ctlr_info_t *h;
4934         int i, j;
4935
4936         if (pci_get_drvdata(pdev) == NULL) {
4937                 dev_err(&pdev->dev, "Unable to remove device\n");
4938                 return;
4939         }
4940
4941         h = pci_get_drvdata(pdev);
4942         i = h->ctlr;
4943         if (hba[i] == NULL) {
4944                 dev_err(&pdev->dev, "device appears to already be removed\n");
4945                 return;
4946         }
4947
4948         mutex_lock(&h->busy_shutting_down);
4949
4950         remove_from_scan_list(h);
4951         remove_proc_entry(h->devname, proc_cciss);
4952         unregister_blkdev(h->major, h->devname);
4953
4954         /* remove it from the disk list */
4955         for (j = 0; j < CISS_MAX_LUN; j++) {
4956                 struct gendisk *disk = h->gendisk[j];
4957                 if (disk) {
4958                         struct request_queue *q = disk->queue;
4959
4960                         if (disk->flags & GENHD_FL_UP) {
4961                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
4962                                 del_gendisk(disk);
4963                         }
4964                         if (q)
4965                                 blk_cleanup_queue(q);
4966                 }
4967         }
4968
4969 #ifdef CONFIG_CISS_SCSI_TAPE
4970         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
4971 #endif
4972
4973         cciss_shutdown(pdev);
4974
4975 #ifdef CONFIG_PCI_MSI
4976         if (h->msix_vector)
4977                 pci_disable_msix(h->pdev);
4978         else if (h->msi_vector)
4979                 pci_disable_msi(h->pdev);
4980 #endif                          /* CONFIG_PCI_MSI */
4981
4982         iounmap(h->transtable);
4983         iounmap(h->cfgtable);
4984         iounmap(h->vaddr);
4985
4986         cciss_free_cmd_pool(h);
4987         /* Free up sg elements */
4988         for (j = 0; j < h->nr_cmds; j++)
4989                 kfree(h->scatter_list[j]);
4990         kfree(h->scatter_list);
4991         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4992         kfree(h->blockFetchTable);
4993         if (h->reply_pool)
4994                 pci_free_consistent(h->pdev, h->max_commands * sizeof(__u64),
4995                                 h->reply_pool, h->reply_pool_dhandle);
4996         /*
4997          * Deliberately omit pci_disable_device(): it does something nasty to
4998          * Smart Array controllers that pci_enable_device does not undo
4999          */
5000         pci_release_regions(pdev);
5001         pci_set_drvdata(pdev, NULL);
5002         cciss_destroy_hba_sysfs_entry(h);
5003         mutex_unlock(&h->busy_shutting_down);
5004         free_hba(h);
5005 }
5006
5007 static struct pci_driver cciss_pci_driver = {
5008         .name = "cciss",
5009         .probe = cciss_init_one,
5010         .remove = __devexit_p(cciss_remove_one),
5011         .id_table = cciss_pci_device_id,        /* id_table */
5012         .shutdown = cciss_shutdown,
5013 };
5014
5015 /*
5016  *  This is it.  Register the PCI driver information for the cards we control
5017  *  the OS will call our registered routines when it finds one of our cards.
5018  */
5019 static int __init cciss_init(void)
5020 {
5021         int err;
5022
5023         /*
5024          * The hardware requires that commands are aligned on a 64-bit
5025          * boundary. Given that we use pci_alloc_consistent() to allocate an
5026          * array of them, the size must be a multiple of 8 bytes.
5027          */
5028         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
5029         printk(KERN_INFO DRIVER_NAME "\n");
5030
5031         err = bus_register(&cciss_bus_type);
5032         if (err)
5033                 return err;
5034
5035         /* Start the scan thread */
5036         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
5037         if (IS_ERR(cciss_scan_thread)) {
5038                 err = PTR_ERR(cciss_scan_thread);
5039                 goto err_bus_unregister;
5040         }
5041
5042         /* Register for our PCI devices */
5043         err = pci_register_driver(&cciss_pci_driver);
5044         if (err)
5045                 goto err_thread_stop;
5046
5047         return err;
5048
5049 err_thread_stop:
5050         kthread_stop(cciss_scan_thread);
5051 err_bus_unregister:
5052         bus_unregister(&cciss_bus_type);
5053
5054         return err;
5055 }
5056
5057 static void __exit cciss_cleanup(void)
5058 {
5059         int i;
5060
5061         pci_unregister_driver(&cciss_pci_driver);
5062         /* double check that all controller entrys have been removed */
5063         for (i = 0; i < MAX_CTLR; i++) {
5064                 if (hba[i] != NULL) {
5065                         dev_warn(&hba[i]->pdev->dev,
5066                                 "had to remove controller\n");
5067                         cciss_remove_one(hba[i]->pdev);
5068                 }
5069         }
5070         kthread_stop(cciss_scan_thread);
5071         if (proc_cciss)
5072                 remove_proc_entry("driver/cciss", NULL);
5073         bus_unregister(&cciss_bus_type);
5074 }
5075
5076 module_init(cciss_init);
5077 module_exit(cciss_cleanup);