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 int cciss_allow_hpsa;
  70 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  71 MODULE_PARM_DESC(cciss_allow_hpsa,
  72         "Prevent cciss driver from accessing hardware known to be "
  73         " supported by the hpsa driver");
  74
  75 #include "cciss_cmd.h"
  76 #include "cciss.h"
  77 #include <linux/cciss_ioctl.h>
  78
  79 /* define the PCI info for the cards we can control */
  80 static const struct pci_device_id cciss_pci_device_id[] = {
  81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3350},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3351},
 110         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3352},
 111         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3353},
 112         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3354},
 113         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSF,     0x103C, 0x3355},
 114         {0,}
 115 };
 116
 117 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 118
 119 /*  board_id = Subsystem Device ID & Vendor ID
 120  *  product = Marketing Name for the board
 121  *  access = Address of the struct of function pointers
 122  */
 123 static struct board_type products[] = {
 124         {0x40700E11, "Smart Array 5300", &SA5_access},
 125         {0x40800E11, "Smart Array 5i", &SA5B_access},
 126         {0x40820E11, "Smart Array 532", &SA5B_access},
 127         {0x40830E11, "Smart Array 5312", &SA5B_access},
 128         {0x409A0E11, "Smart Array 641", &SA5_access},
 129         {0x409B0E11, "Smart Array 642", &SA5_access},
 130         {0x409C0E11, "Smart Array 6400", &SA5_access},
 131         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 132         {0x40910E11, "Smart Array 6i", &SA5_access},
 133         {0x3225103C, "Smart Array P600", &SA5_access},
 134         {0x3235103C, "Smart Array P400i", &SA5_access},
 135         {0x3211103C, "Smart Array E200i", &SA5_access},
 136         {0x3212103C, "Smart Array E200", &SA5_access},
 137         {0x3213103C, "Smart Array E200i", &SA5_access},
 138         {0x3214103C, "Smart Array E200i", &SA5_access},
 139         {0x3215103C, "Smart Array E200i", &SA5_access},
 140         {0x3237103C, "Smart Array E500", &SA5_access},
 141 /* controllers below this line are also supported by the hpsa driver. */
 142 #define HPSA_BOUNDARY 0x3223103C
 143         {0x3223103C, "Smart Array P800", &SA5_access},
 144         {0x3234103C, "Smart Array P400", &SA5_access},
 145         {0x323D103C, "Smart Array P700m", &SA5_access},
 146         {0x3241103C, "Smart Array P212", &SA5_access},
 147         {0x3243103C, "Smart Array P410", &SA5_access},
 148         {0x3245103C, "Smart Array P410i", &SA5_access},
 149         {0x3247103C, "Smart Array P411", &SA5_access},
 150         {0x3249103C, "Smart Array P812", &SA5_access},
 151         {0x324A103C, "Smart Array P712m", &SA5_access},
 152         {0x324B103C, "Smart Array P711m", &SA5_access},
 153         {0x3350103C, "Smart Array", &SA5_access},
 154         {0x3351103C, "Smart Array", &SA5_access},
 155         {0x3352103C, "Smart Array", &SA5_access},
 156         {0x3353103C, "Smart Array", &SA5_access},
 157         {0x3354103C, "Smart Array", &SA5_access},
 158         {0x3355103C, "Smart Array", &SA5_access},
 159 };
 160
 161 /* How long to wait (in milliseconds) for board to go into simple mode */
 162 #define MAX_CONFIG_WAIT 30000
 163 #define MAX_IOCTL_CONFIG_WAIT 1000
 164
 165 /*define how many times we will try a command because of bus resets */
 166 #define MAX_CMD_RETRIES 3
 167
 168 #define MAX_CTLR        32
 169
 170 /* Originally cciss driver only supports 8 major numbers */
 171 #define MAX_CTLR_ORIG   8
 172
 173 static ctlr_info_t *hba[MAX_CTLR];
 174
 175 static struct task_struct *cciss_scan_thread;
 176 static DEFINE_MUTEX(scan_mutex);
 177 static LIST_HEAD(scan_q);
 178
 179 static void do_cciss_request(struct request_queue *q);
 180 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 181 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 182 static int cciss_open(struct block_device *bdev, fmode_t mode);
 183 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 184 static int cciss_release(struct gendisk *disk, fmode_t mode);
 185 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 186                     unsigned int cmd, unsigned long arg);
 187 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 188                        unsigned int cmd, unsigned long arg);
 189 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 190
 191 static int cciss_revalidate(struct gendisk *disk);
 192 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 193 static int deregister_disk(ctlr_info_t *h, int drv_index,
 194                            int clear_all, int via_ioctl);
 195
 196 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 197                         sector_t *total_size, unsigned int *block_size);
 198 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 199                         sector_t *total_size, unsigned int *block_size);
 200 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 201                         sector_t total_size,
 202                         unsigned int block_size, InquiryData_struct *inq_buff,
 203                                    drive_info_struct *drv);
 204 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
 205 static void start_io(ctlr_info_t *h);
 206 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 207                         __u8 page_code, unsigned char scsi3addr[],
 208                         int cmd_type);
 209 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 210         int attempt_retry);
 211 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 212
 213 static int add_to_scan_list(struct ctlr_info *h);
 214 static int scan_thread(void *data);
 215 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 216 static void cciss_hba_release(struct device *dev);
 217 static void cciss_device_release(struct device *dev);
 218 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 219 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 220 static inline u32 next_command(ctlr_info_t *h);
 221 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
 222         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 223         u64 *cfg_offset);
 224 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 225         unsigned long *memory_bar);
 226
 227
 228 /* performant mode helper functions */
 229 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 230                                 int *bucket_map);
 231 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 232
 233 #ifdef CONFIG_PROC_FS
 234 static void cciss_procinit(ctlr_info_t *h);
 235 #else
 236 static void cciss_procinit(ctlr_info_t *h)
 237 {
 238 }
 239 #endif                          /* CONFIG_PROC_FS */
 240
 241 #ifdef CONFIG_COMPAT
 242 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 243                               unsigned, unsigned long);
 244 #endif
 245
 246 static const struct block_device_operations cciss_fops = {
 247         .owner = THIS_MODULE,
 248         .open = cciss_unlocked_open,
 249         .release = cciss_release,
 250         .ioctl = do_ioctl,
 251         .getgeo = cciss_getgeo,
 252 #ifdef CONFIG_COMPAT
 253         .compat_ioctl = cciss_compat_ioctl,
 254 #endif
 255         .revalidate_disk = cciss_revalidate,
 256 };
 257
 258 /* set_performant_mode: Modify the tag for cciss performant
 259  * set bit 0 for pull model, bits 3-1 for block fetch
 260  * register number
 261  */
 262 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 263 {
 264         if (likely(h->transMethod == CFGTBL_Trans_Performant))
 265                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 266 }
 267
 268 /*
 269  * Enqueuing and dequeuing functions for cmdlists.
 270  */
 271 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 272 {
 273         hlist_add_head(&c->list, list);
 274 }
 275
 276 static inline void removeQ(CommandList_struct *c)
 277 {
 278         /*
 279          * After kexec/dump some commands might still
 280          * be in flight, which the firmware will try
 281          * to complete. Resetting the firmware doesn't work
 282          * with old fw revisions, so we have to mark
 283          * them off as 'stale' to prevent the driver from
 284          * falling over.
 285          */
 286         if (WARN_ON(hlist_unhashed(&c->list))) {
 287                 c->cmd_type = CMD_MSG_STALE;
 288                 return;
 289         }
 290
 291         hlist_del_init(&c->list);
 292 }
 293
 294 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 295         CommandList_struct *c)
 296 {
 297         unsigned long flags;
 298         set_performant_mode(h, c);
 299         spin_lock_irqsave(&h->lock, flags);
 300         addQ(&h->reqQ, c);
 301         h->Qdepth++;
 302         if (h->Qdepth > h->maxQsinceinit)
 303                 h->maxQsinceinit = h->Qdepth;
 304         start_io(h);
 305         spin_unlock_irqrestore(&h->lock, flags);
 306 }
 307
 308 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 309         int nr_cmds)
 310 {
 311         int i;
 312
 313         if (!cmd_sg_list)
 314                 return;
 315         for (i = 0; i < nr_cmds; i++) {
 316                 kfree(cmd_sg_list[i]);
 317                 cmd_sg_list[i] = NULL;
 318         }
 319         kfree(cmd_sg_list);
 320 }
 321
 322 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 323         ctlr_info_t *h, int chainsize, int nr_cmds)
 324 {
 325         int j;
 326         SGDescriptor_struct **cmd_sg_list;
 327
 328         if (chainsize <= 0)
 329                 return NULL;
 330
 331         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 332         if (!cmd_sg_list)
 333                 return NULL;
 334
 335         /* Build up chain blocks for each command */
 336         for (j = 0; j < nr_cmds; j++) {
 337                 /* Need a block of chainsized s/g elements. */
 338                 cmd_sg_list[j] = kmalloc((chainsize *
 339                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 340                 if (!cmd_sg_list[j]) {
 341                         dev_err(&h->pdev->dev, "Cannot get memory "
 342                                 "for s/g chains.\n");
 343                         goto clean;
 344                 }
 345         }
 346         return cmd_sg_list;
 347 clean:
 348         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 349         return NULL;
 350 }
 351
 352 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 353 {
 354         SGDescriptor_struct *chain_sg;
 355         u64bit temp64;
 356
 357         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 358                 return;
 359
 360         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 361         temp64.val32.lower = chain_sg->Addr.lower;
 362         temp64.val32.upper = chain_sg->Addr.upper;
 363         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 364 }
 365
 366 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 367         SGDescriptor_struct *chain_block, int len)
 368 {
 369         SGDescriptor_struct *chain_sg;
 370         u64bit temp64;
 371
 372         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 373         chain_sg->Ext = CCISS_SG_CHAIN;
 374         chain_sg->Len = len;
 375         temp64.val = pci_map_single(h->pdev, chain_block, len,
 376                                 PCI_DMA_TODEVICE);
 377         chain_sg->Addr.lower = temp64.val32.lower;
 378         chain_sg->Addr.upper = temp64.val32.upper;
 379 }
 380
 381 #include "cciss_scsi.c"         /* For SCSI tape support */
 382
 383 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 384         "UNKNOWN"
 385 };
 386 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 387
 388 #ifdef CONFIG_PROC_FS
 389
 390 /*
 391  * Report information about this controller.
 392  */
 393 #define ENG_GIG 1000000000
 394 #define ENG_GIG_FACTOR (ENG_GIG/512)
 395 #define ENGAGE_SCSI     "engage scsi"
 396
 397 static struct proc_dir_entry *proc_cciss;
 398
 399 static void cciss_seq_show_header(struct seq_file *seq)
 400 {
 401         ctlr_info_t *h = seq->private;
 402
 403         seq_printf(seq, "%s: HP %s Controller\n"
 404                 "Board ID: 0x%08lx\n"
 405                 "Firmware Version: %c%c%c%c\n"
 406                 "IRQ: %d\n"
 407                 "Logical drives: %d\n"
 408                 "Current Q depth: %d\n"
 409                 "Current # commands on controller: %d\n"
 410                 "Max Q depth since init: %d\n"
 411                 "Max # commands on controller since init: %d\n"
 412                 "Max SG entries since init: %d\n",
 413                 h->devname,
 414                 h->product_name,
 415                 (unsigned long)h->board_id,
 416                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 417                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 418                 h->num_luns,
 419                 h->Qdepth, h->commands_outstanding,
 420                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 421
 422 #ifdef CONFIG_CISS_SCSI_TAPE
 423         cciss_seq_tape_report(seq, h);
 424 #endif /* CONFIG_CISS_SCSI_TAPE */
 425 }
 426
 427 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 428 {
 429         ctlr_info_t *h = seq->private;
 430         unsigned long flags;
 431
 432         /* prevent displaying bogus info during configuration
 433          * or deconfiguration of a logical volume
 434          */
 435         spin_lock_irqsave(&h->lock, flags);
 436         if (h->busy_configuring) {
 437                 spin_unlock_irqrestore(&h->lock, flags);
 438                 return ERR_PTR(-EBUSY);
 439         }
 440         h->busy_configuring = 1;
 441         spin_unlock_irqrestore(&h->lock, flags);
 442
 443         if (*pos == 0)
 444                 cciss_seq_show_header(seq);
 445
 446         return pos;
 447 }
 448
 449 static int cciss_seq_show(struct seq_file *seq, void *v)
 450 {
 451         sector_t vol_sz, vol_sz_frac;
 452         ctlr_info_t *h = seq->private;
 453         unsigned ctlr = h->ctlr;
 454         loff_t *pos = v;
 455         drive_info_struct *drv = h->drv[*pos];
 456
 457         if (*pos > h->highest_lun)
 458                 return 0;
 459
 460         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 461                 return 0;
 462
 463         if (drv->heads == 0)
 464                 return 0;
 465
 466         vol_sz = drv->nr_blocks;
 467         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 468         vol_sz_frac *= 100;
 469         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 470
 471         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 472                 drv->raid_level = RAID_UNKNOWN;
 473         seq_printf(seq, "cciss/c%dd%d:"
 474                         "\t%4u.%02uGB\tRAID %s\n",
 475                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 476                         raid_label[drv->raid_level]);
 477         return 0;
 478 }
 479
 480 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 481 {
 482         ctlr_info_t *h = seq->private;
 483
 484         if (*pos > h->highest_lun)
 485                 return NULL;
 486         *pos += 1;
 487
 488         return pos;
 489 }
 490
 491 static void cciss_seq_stop(struct seq_file *seq, void *v)
 492 {
 493         ctlr_info_t *h = seq->private;
 494
 495         /* Only reset h->busy_configuring if we succeeded in setting
 496          * it during cciss_seq_start. */
 497         if (v == ERR_PTR(-EBUSY))
 498                 return;
 499
 500         h->busy_configuring = 0;
 501 }
 502
 503 static const struct seq_operations cciss_seq_ops = {
 504         .start = cciss_seq_start,
 505         .show  = cciss_seq_show,
 506         .next  = cciss_seq_next,
 507         .stop  = cciss_seq_stop,
 508 };
 509
 510 static int cciss_seq_open(struct inode *inode, struct file *file)
 511 {
 512         int ret = seq_open(file, &cciss_seq_ops);
 513         struct seq_file *seq = file->private_data;
 514
 515         if (!ret)
 516                 seq->private = PDE(inode)->data;
 517
 518         return ret;
 519 }
 520
 521 static ssize_t
 522 cciss_proc_write(struct file *file, const char __user *buf,
 523                  size_t length, loff_t *ppos)
 524 {
 525         int err;
 526         char *buffer;
 527
 528 #ifndef CONFIG_CISS_SCSI_TAPE
 529         return -EINVAL;
 530 #endif
 531
 532         if (!buf || length > PAGE_SIZE - 1)
 533                 return -EINVAL;
 534
 535         buffer = (char *)__get_free_page(GFP_KERNEL);
 536         if (!buffer)
 537                 return -ENOMEM;
 538
 539         err = -EFAULT;
 540         if (copy_from_user(buffer, buf, length))
 541                 goto out;
 542         buffer[length] = '\0';
 543
 544 #ifdef CONFIG_CISS_SCSI_TAPE
 545         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 546                 struct seq_file *seq = file->private_data;
 547                 ctlr_info_t *h = seq->private;
 548
 549                 err = cciss_engage_scsi(h);
 550                 if (err == 0)
 551                         err = length;
 552         } else
 553 #endif /* CONFIG_CISS_SCSI_TAPE */
 554                 err = -EINVAL;
 555         /* might be nice to have "disengage" too, but it's not
 556            safely possible. (only 1 module use count, lock issues.) */
 557
 558 out:
 559         free_page((unsigned long)buffer);
 560         return err;
 561 }
 562
 563 static const struct file_operations cciss_proc_fops = {
 564         .owner   = THIS_MODULE,
 565         .open    = cciss_seq_open,
 566         .read    = seq_read,
 567         .llseek  = seq_lseek,
 568         .release = seq_release,
 569         .write   = cciss_proc_write,
 570 };
 571
 572 static void __devinit cciss_procinit(ctlr_info_t *h)
 573 {
 574         struct proc_dir_entry *pde;
 575
 576         if (proc_cciss == NULL)
 577                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 578         if (!proc_cciss)
 579                 return;
 580         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 581                                         S_IROTH, proc_cciss,
 582                                         &cciss_proc_fops, h);
 583 }
 584 #endif                          /* CONFIG_PROC_FS */
 585
 586 #define MAX_PRODUCT_NAME_LEN 19
 587
 588 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 589 #define to_drv(n) container_of(n, drive_info_struct, dev)
 590
 591 static ssize_t host_store_rescan(struct device *dev,
 592                                  struct device_attribute *attr,
 593                                  const char *buf, size_t count)
 594 {
 595         struct ctlr_info *h = to_hba(dev);
 596
 597         add_to_scan_list(h);
 598         wake_up_process(cciss_scan_thread);
 599         wait_for_completion_interruptible(&h->scan_wait);
 600
 601         return count;
 602 }
 603 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 604
 605 static ssize_t dev_show_unique_id(struct device *dev,
 606                                  struct device_attribute *attr,
 607                                  char *buf)
 608 {
 609         drive_info_struct *drv = to_drv(dev);
 610         struct ctlr_info *h = to_hba(drv->dev.parent);
 611         __u8 sn[16];
 612         unsigned long flags;
 613         int ret = 0;
 614
 615         spin_lock_irqsave(&h->lock, flags);
 616         if (h->busy_configuring)
 617                 ret = -EBUSY;
 618         else
 619                 memcpy(sn, drv->serial_no, sizeof(sn));
 620         spin_unlock_irqrestore(&h->lock, flags);
 621
 622         if (ret)
 623                 return ret;
 624         else
 625                 return snprintf(buf, 16 * 2 + 2,
 626                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 627                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 628                                 sn[0], sn[1], sn[2], sn[3],
 629                                 sn[4], sn[5], sn[6], sn[7],
 630                                 sn[8], sn[9], sn[10], sn[11],
 631                                 sn[12], sn[13], sn[14], sn[15]);
 632 }
 633 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 634
 635 static ssize_t dev_show_vendor(struct device *dev,
 636                                struct device_attribute *attr,
 637                                char *buf)
 638 {
 639         drive_info_struct *drv = to_drv(dev);
 640         struct ctlr_info *h = to_hba(drv->dev.parent);
 641         char vendor[VENDOR_LEN + 1];
 642         unsigned long flags;
 643         int ret = 0;
 644
 645         spin_lock_irqsave(&h->lock, flags);
 646         if (h->busy_configuring)
 647                 ret = -EBUSY;
 648         else
 649                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 650         spin_unlock_irqrestore(&h->lock, flags);
 651
 652         if (ret)
 653                 return ret;
 654         else
 655                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 656 }
 657 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 658
 659 static ssize_t dev_show_model(struct device *dev,
 660                               struct device_attribute *attr,
 661                               char *buf)
 662 {
 663         drive_info_struct *drv = to_drv(dev);
 664         struct ctlr_info *h = to_hba(drv->dev.parent);
 665         char model[MODEL_LEN + 1];
 666         unsigned long flags;
 667         int ret = 0;
 668
 669         spin_lock_irqsave(&h->lock, flags);
 670         if (h->busy_configuring)
 671                 ret = -EBUSY;
 672         else
 673                 memcpy(model, drv->model, MODEL_LEN + 1);
 674         spin_unlock_irqrestore(&h->lock, flags);
 675
 676         if (ret)
 677                 return ret;
 678         else
 679                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 680 }
 681 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 682
 683 static ssize_t dev_show_rev(struct device *dev,
 684                             struct device_attribute *attr,
 685                             char *buf)
 686 {
 687         drive_info_struct *drv = to_drv(dev);
 688         struct ctlr_info *h = to_hba(drv->dev.parent);
 689         char rev[REV_LEN + 1];
 690         unsigned long flags;
 691         int ret = 0;
 692
 693         spin_lock_irqsave(&h->lock, flags);
 694         if (h->busy_configuring)
 695                 ret = -EBUSY;
 696         else
 697                 memcpy(rev, drv->rev, REV_LEN + 1);
 698         spin_unlock_irqrestore(&h->lock, flags);
 699
 700         if (ret)
 701                 return ret;
 702         else
 703                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 704 }
 705 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 706
 707 static ssize_t cciss_show_lunid(struct device *dev,
 708                                 struct device_attribute *attr, char *buf)
 709 {
 710         drive_info_struct *drv = to_drv(dev);
 711         struct ctlr_info *h = to_hba(drv->dev.parent);
 712         unsigned long flags;
 713         unsigned char lunid[8];
 714
 715         spin_lock_irqsave(&h->lock, flags);
 716         if (h->busy_configuring) {
 717                 spin_unlock_irqrestore(&h->lock, flags);
 718                 return -EBUSY;
 719         }
 720         if (!drv->heads) {
 721                 spin_unlock_irqrestore(&h->lock, flags);
 722                 return -ENOTTY;
 723         }
 724         memcpy(lunid, drv->LunID, sizeof(lunid));
 725         spin_unlock_irqrestore(&h->lock, flags);
 726         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 727                 lunid[0], lunid[1], lunid[2], lunid[3],
 728                 lunid[4], lunid[5], lunid[6], lunid[7]);
 729 }
 730 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 731
 732 static ssize_t cciss_show_raid_level(struct device *dev,
 733                                      struct device_attribute *attr, char *buf)
 734 {
 735         drive_info_struct *drv = to_drv(dev);
 736         struct ctlr_info *h = to_hba(drv->dev.parent);
 737         int raid;
 738         unsigned long flags;
 739
 740         spin_lock_irqsave(&h->lock, flags);
 741         if (h->busy_configuring) {
 742                 spin_unlock_irqrestore(&h->lock, flags);
 743                 return -EBUSY;
 744         }
 745         raid = drv->raid_level;
 746         spin_unlock_irqrestore(&h->lock, flags);
 747         if (raid < 0 || raid > RAID_UNKNOWN)
 748                 raid = RAID_UNKNOWN;
 749
 750         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 751                         raid_label[raid]);
 752 }
 753 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 754
 755 static ssize_t cciss_show_usage_count(struct device *dev,
 756                                       struct device_attribute *attr, char *buf)
 757 {
 758         drive_info_struct *drv = to_drv(dev);
 759         struct ctlr_info *h = to_hba(drv->dev.parent);
 760         unsigned long flags;
 761         int count;
 762
 763         spin_lock_irqsave(&h->lock, flags);
 764         if (h->busy_configuring) {
 765                 spin_unlock_irqrestore(&h->lock, flags);
 766                 return -EBUSY;
 767         }
 768         count = drv->usage_count;
 769         spin_unlock_irqrestore(&h->lock, flags);
 770         return snprintf(buf, 20, "%d\n", count);
 771 }
 772 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 773
 774 static struct attribute *cciss_host_attrs[] = {
 775         &dev_attr_rescan.attr,
 776         NULL
 777 };
 778
 779 static struct attribute_group cciss_host_attr_group = {
 780         .attrs = cciss_host_attrs,
 781 };
 782
 783 static const struct attribute_group *cciss_host_attr_groups[] = {
 784         &cciss_host_attr_group,
 785         NULL
 786 };
 787
 788 static struct device_type cciss_host_type = {
 789         .name           = "cciss_host",
 790         .groups         = cciss_host_attr_groups,
 791         .release        = cciss_hba_release,
 792 };
 793
 794 static struct attribute *cciss_dev_attrs[] = {
 795         &dev_attr_unique_id.attr,
 796         &dev_attr_model.attr,
 797         &dev_attr_vendor.attr,
 798         &dev_attr_rev.attr,
 799         &dev_attr_lunid.attr,
 800         &dev_attr_raid_level.attr,
 801         &dev_attr_usage_count.attr,
 802         NULL
 803 };
 804
 805 static struct attribute_group cciss_dev_attr_group = {
 806         .attrs = cciss_dev_attrs,
 807 };
 808
 809 static const struct attribute_group *cciss_dev_attr_groups[] = {
 810         &cciss_dev_attr_group,
 811         NULL
 812 };
 813
 814 static struct device_type cciss_dev_type = {
 815         .name           = "cciss_device",
 816         .groups         = cciss_dev_attr_groups,
 817         .release        = cciss_device_release,
 818 };
 819
 820 static struct bus_type cciss_bus_type = {
 821         .name           = "cciss",
 822 };
 823
 824 /*
 825  * cciss_hba_release is called when the reference count
 826  * of h->dev goes to zero.
 827  */
 828 static void cciss_hba_release(struct device *dev)
 829 {
 830         /*
 831          * nothing to do, but need this to avoid a warning
 832          * about not having a release handler from lib/kref.c.
 833          */
 834 }
 835
 836 /*
 837  * Initialize sysfs entry for each controller.  This sets up and registers
 838  * the 'cciss#' directory for each individual controller under
 839  * /sys/bus/pci/devices/<dev>/.
 840  */
 841 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 842 {
 843         device_initialize(&h->dev);
 844         h->dev.type = &cciss_host_type;
 845         h->dev.bus = &cciss_bus_type;
 846         dev_set_name(&h->dev, "%s", h->devname);
 847         h->dev.parent = &h->pdev->dev;
 848
 849         return device_add(&h->dev);
 850 }
 851
 852 /*
 853  * Remove sysfs entries for an hba.
 854  */
 855 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 856 {
 857         device_del(&h->dev);
 858         put_device(&h->dev); /* final put. */
 859 }
 860
 861 /* cciss_device_release is called when the reference count
 862  * of h->drv[x]dev goes to zero.
 863  */
 864 static void cciss_device_release(struct device *dev)
 865 {
 866         drive_info_struct *drv = to_drv(dev);
 867         kfree(drv);
 868 }
 869
 870 /*
 871  * Initialize sysfs for each logical drive.  This sets up and registers
 872  * the 'c#d#' directory for each individual logical drive under
 873  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 874  * /sys/block/cciss!c#d# to this entry.
 875  */
 876 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 877                                        int drv_index)
 878 {
 879         struct device *dev;
 880
 881         if (h->drv[drv_index]->device_initialized)
 882                 return 0;
 883
 884         dev = &h->drv[drv_index]->dev;
 885         device_initialize(dev);
 886         dev->type = &cciss_dev_type;
 887         dev->bus = &cciss_bus_type;
 888         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 889         dev->parent = &h->dev;
 890         h->drv[drv_index]->device_initialized = 1;
 891         return device_add(dev);
 892 }
 893
 894 /*
 895  * Remove sysfs entries for a logical drive.
 896  */
 897 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 898         int ctlr_exiting)
 899 {
 900         struct device *dev = &h->drv[drv_index]->dev;
 901
 902         /* special case for c*d0, we only destroy it on controller exit */
 903         if (drv_index == 0 && !ctlr_exiting)
 904                 return;
 905
 906         device_del(dev);
 907         put_device(dev); /* the "final" put. */
 908         h->drv[drv_index] = NULL;
 909 }
 910
 911 /*
 912  * For operations that cannot sleep, a command block is allocated at init,
 913  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 914  * which ones are free or in use.
 915  */
 916 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 917 {
 918         CommandList_struct *c;
 919         int i;
 920         u64bit temp64;
 921         dma_addr_t cmd_dma_handle, err_dma_handle;
 922
 923         do {
 924                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 925                 if (i == h->nr_cmds)
 926                         return NULL;
 927         } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
 928                   h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 929         c = h->cmd_pool + i;
 930         memset(c, 0, sizeof(CommandList_struct));
 931         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
 932         c->err_info = h->errinfo_pool + i;
 933         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 934         err_dma_handle = h->errinfo_pool_dhandle
 935             + i * sizeof(ErrorInfo_struct);
 936         h->nr_allocs++;
 937
 938         c->cmdindex = i;
 939
 940         INIT_HLIST_NODE(&c->list);
 941         c->busaddr = (__u32) cmd_dma_handle;
 942         temp64.val = (__u64) err_dma_handle;
 943         c->ErrDesc.Addr.lower = temp64.val32.lower;
 944         c->ErrDesc.Addr.upper = temp64.val32.upper;
 945         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 946
 947         c->ctlr = h->ctlr;
 948         return c;
 949 }
 950
 951 /* allocate a command using pci_alloc_consistent, used for ioctls,
 952  * etc., not for the main i/o path.
 953  */
 954 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
 955 {
 956         CommandList_struct *c;
 957         u64bit temp64;
 958         dma_addr_t cmd_dma_handle, err_dma_handle;
 959
 960         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 961                 sizeof(CommandList_struct), &cmd_dma_handle);
 962         if (c == NULL)
 963                 return NULL;
 964         memset(c, 0, sizeof(CommandList_struct));
 965
 966         c->cmdindex = -1;
 967
 968         c->err_info = (ErrorInfo_struct *)
 969             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 970                     &err_dma_handle);
 971
 972         if (c->err_info == NULL) {
 973                 pci_free_consistent(h->pdev,
 974                         sizeof(CommandList_struct), c, cmd_dma_handle);
 975                 return NULL;
 976         }
 977         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 978
 979         INIT_HLIST_NODE(&c->list);
 980         c->busaddr = (__u32) cmd_dma_handle;
 981         temp64.val = (__u64) err_dma_handle;
 982         c->ErrDesc.Addr.lower = temp64.val32.lower;
 983         c->ErrDesc.Addr.upper = temp64.val32.upper;
 984         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 985
 986         c->ctlr = h->ctlr;
 987         return c;
 988 }
 989
 990 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
 991 {
 992         int i;
 993
 994         i = c - h->cmd_pool;
 995         clear_bit(i & (BITS_PER_LONG - 1),
 996                   h->cmd_pool_bits + (i / BITS_PER_LONG));
 997         h->nr_frees++;
 998 }
 999
1000 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
1001 {
1002         u64bit temp64;
1003
1004         temp64.val32.lower = c->ErrDesc.Addr.lower;
1005         temp64.val32.upper = c->ErrDesc.Addr.upper;
1006         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1007                             c->err_info, (dma_addr_t) temp64.val);
1008         pci_free_consistent(h->pdev, sizeof(CommandList_struct),
1009                             c, (dma_addr_t) c->busaddr);
1010 }
1011
1012 static inline ctlr_info_t *get_host(struct gendisk *disk)
1013 {
1014         return disk->queue->queuedata;
1015 }
1016
1017 static inline drive_info_struct *get_drv(struct gendisk *disk)
1018 {
1019         return disk->private_data;
1020 }
1021
1022 /*
1023  * Open.  Make sure the device is really there.
1024  */
1025 static int cciss_open(struct block_device *bdev, fmode_t mode)
1026 {
1027         ctlr_info_t *h = get_host(bdev->bd_disk);
1028         drive_info_struct *drv = get_drv(bdev->bd_disk);
1029
1030         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1031         if (drv->busy_configuring)
1032                 return -EBUSY;
1033         /*
1034          * Root is allowed to open raw volume zero even if it's not configured
1035          * so array config can still work. Root is also allowed to open any
1036          * volume that has a LUN ID, so it can issue IOCTL to reread the
1037          * disk information.  I don't think I really like this
1038          * but I'm already using way to many device nodes to claim another one
1039          * for "raw controller".
1040          */
1041         if (drv->heads == 0) {
1042                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1043                         /* if not node 0 make sure it is a partition = 0 */
1044                         if (MINOR(bdev->bd_dev) & 0x0f) {
1045                                 return -ENXIO;
1046                                 /* if it is, make sure we have a LUN ID */
1047                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1048                                 sizeof(drv->LunID))) {
1049                                 return -ENXIO;
1050                         }
1051                 }
1052                 if (!capable(CAP_SYS_ADMIN))
1053                         return -EPERM;
1054         }
1055         drv->usage_count++;
1056         h->usage_count++;
1057         return 0;
1058 }
1059
1060 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1061 {
1062         int ret;
1063
1064         mutex_lock(&cciss_mutex);
1065         ret = cciss_open(bdev, mode);
1066         mutex_unlock(&cciss_mutex);
1067
1068         return ret;
1069 }
1070
1071 /*
1072  * Close.  Sync first.
1073  */
1074 static int cciss_release(struct gendisk *disk, fmode_t mode)
1075 {
1076         ctlr_info_t *h;
1077         drive_info_struct *drv;
1078
1079         mutex_lock(&cciss_mutex);
1080         h = get_host(disk);
1081         drv = get_drv(disk);
1082         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1083         drv->usage_count--;
1084         h->usage_count--;
1085         mutex_unlock(&cciss_mutex);
1086         return 0;
1087 }
1088
1089 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1090                     unsigned cmd, unsigned long arg)
1091 {
1092         int ret;
1093         mutex_lock(&cciss_mutex);
1094         ret = cciss_ioctl(bdev, mode, cmd, arg);
1095         mutex_unlock(&cciss_mutex);
1096         return ret;
1097 }
1098
1099 #ifdef CONFIG_COMPAT
1100
1101 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1102                                   unsigned cmd, unsigned long arg);
1103 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1104                                       unsigned cmd, unsigned long arg);
1105
1106 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1107                               unsigned cmd, unsigned long arg)
1108 {
1109         switch (cmd) {
1110         case CCISS_GETPCIINFO:
1111         case CCISS_GETINTINFO:
1112         case CCISS_SETINTINFO:
1113         case CCISS_GETNODENAME:
1114         case CCISS_SETNODENAME:
1115         case CCISS_GETHEARTBEAT:
1116         case CCISS_GETBUSTYPES:
1117         case CCISS_GETFIRMVER:
1118         case CCISS_GETDRIVVER:
1119         case CCISS_REVALIDVOLS:
1120         case CCISS_DEREGDISK:
1121         case CCISS_REGNEWDISK:
1122         case CCISS_REGNEWD:
1123         case CCISS_RESCANDISK:
1124         case CCISS_GETLUNINFO:
1125                 return do_ioctl(bdev, mode, cmd, arg);
1126
1127         case CCISS_PASSTHRU32:
1128                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1129         case CCISS_BIG_PASSTHRU32:
1130                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1131
1132         default:
1133                 return -ENOIOCTLCMD;
1134         }
1135 }
1136
1137 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1138                                   unsigned cmd, unsigned long arg)
1139 {
1140         IOCTL32_Command_struct __user *arg32 =
1141             (IOCTL32_Command_struct __user *) arg;
1142         IOCTL_Command_struct arg64;
1143         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1144         int err;
1145         u32 cp;
1146
1147         err = 0;
1148         err |=
1149             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1150                            sizeof(arg64.LUN_info));
1151         err |=
1152             copy_from_user(&arg64.Request, &arg32->Request,
1153                            sizeof(arg64.Request));
1154         err |=
1155             copy_from_user(&arg64.error_info, &arg32->error_info,
1156                            sizeof(arg64.error_info));
1157         err |= get_user(arg64.buf_size, &arg32->buf_size);
1158         err |= get_user(cp, &arg32->buf);
1159         arg64.buf = compat_ptr(cp);
1160         err |= copy_to_user(p, &arg64, sizeof(arg64));
1161
1162         if (err)
1163                 return -EFAULT;
1164
1165         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1166         if (err)
1167                 return err;
1168         err |=
1169             copy_in_user(&arg32->error_info, &p->error_info,
1170                          sizeof(arg32->error_info));
1171         if (err)
1172                 return -EFAULT;
1173         return err;
1174 }
1175
1176 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1177                                       unsigned cmd, unsigned long arg)
1178 {
1179         BIG_IOCTL32_Command_struct __user *arg32 =
1180             (BIG_IOCTL32_Command_struct __user *) arg;
1181         BIG_IOCTL_Command_struct arg64;
1182         BIG_IOCTL_Command_struct __user *p =
1183             compat_alloc_user_space(sizeof(arg64));
1184         int err;
1185         u32 cp;
1186
1187         memset(&arg64, 0, sizeof(arg64));
1188         err = 0;
1189         err |=
1190             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1191                            sizeof(arg64.LUN_info));
1192         err |=
1193             copy_from_user(&arg64.Request, &arg32->Request,
1194                            sizeof(arg64.Request));
1195         err |=
1196             copy_from_user(&arg64.error_info, &arg32->error_info,
1197                            sizeof(arg64.error_info));
1198         err |= get_user(arg64.buf_size, &arg32->buf_size);
1199         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1200         err |= get_user(cp, &arg32->buf);
1201         arg64.buf = compat_ptr(cp);
1202         err |= copy_to_user(p, &arg64, sizeof(arg64));
1203
1204         if (err)
1205                 return -EFAULT;
1206
1207         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1208         if (err)
1209                 return err;
1210         err |=
1211             copy_in_user(&arg32->error_info, &p->error_info,
1212                          sizeof(arg32->error_info));
1213         if (err)
1214                 return -EFAULT;
1215         return err;
1216 }
1217 #endif
1218
1219 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1220 {
1221         drive_info_struct *drv = get_drv(bdev->bd_disk);
1222
1223         if (!drv->cylinders)
1224                 return -ENXIO;
1225
1226         geo->heads = drv->heads;
1227         geo->sectors = drv->sectors;
1228         geo->cylinders = drv->cylinders;
1229         return 0;
1230 }
1231
1232 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1233 {
1234         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1235                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1236                 (void)check_for_unit_attention(h, c);
1237 }
1238
1239 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1240 {
1241         cciss_pci_info_struct pciinfo;
1242
1243         if (!argp)
1244                 return -EINVAL;
1245         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1246         pciinfo.bus = h->pdev->bus->number;
1247         pciinfo.dev_fn = h->pdev->devfn;
1248         pciinfo.board_id = h->board_id;
1249         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1250                 return -EFAULT;
1251         return 0;
1252 }
1253
1254 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1255 {
1256         cciss_coalint_struct intinfo;
1257
1258         if (!argp)
1259                 return -EINVAL;
1260         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1261         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1262         if (copy_to_user
1263             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1264                 return -EFAULT;
1265         return 0;
1266 }
1267
1268 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1269 {
1270         cciss_coalint_struct intinfo;
1271         unsigned long flags;
1272         int i;
1273
1274         if (!argp)
1275                 return -EINVAL;
1276         if (!capable(CAP_SYS_ADMIN))
1277                 return -EPERM;
1278         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1279                 return -EFAULT;
1280         if ((intinfo.delay == 0) && (intinfo.count == 0))
1281                 return -EINVAL;
1282         spin_lock_irqsave(&h->lock, flags);
1283         /* Update the field, and then ring the doorbell */
1284         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1285         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1286         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1287
1288         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1289                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1290                         break;
1291                 udelay(1000); /* delay and try again */
1292         }
1293         spin_unlock_irqrestore(&h->lock, flags);
1294         if (i >= MAX_IOCTL_CONFIG_WAIT)
1295                 return -EAGAIN;
1296         return 0;
1297 }
1298
1299 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1300 {
1301         NodeName_type NodeName;
1302         int i;
1303
1304         if (!argp)
1305                 return -EINVAL;
1306         for (i = 0; i < 16; i++)
1307                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1308         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1309                 return -EFAULT;
1310         return 0;
1311 }
1312
1313 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1314 {
1315         NodeName_type NodeName;
1316         unsigned long flags;
1317         int i;
1318
1319         if (!argp)
1320                 return -EINVAL;
1321         if (!capable(CAP_SYS_ADMIN))
1322                 return -EPERM;
1323         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1324                 return -EFAULT;
1325         spin_lock_irqsave(&h->lock, flags);
1326         /* Update the field, and then ring the doorbell */
1327         for (i = 0; i < 16; i++)
1328                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1329         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1330         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1331                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1332                         break;
1333                 udelay(1000); /* delay and try again */
1334         }
1335         spin_unlock_irqrestore(&h->lock, flags);
1336         if (i >= MAX_IOCTL_CONFIG_WAIT)
1337                 return -EAGAIN;
1338         return 0;
1339 }
1340
1341 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1342 {
1343         Heartbeat_type heartbeat;
1344
1345         if (!argp)
1346                 return -EINVAL;
1347         heartbeat = readl(&h->cfgtable->HeartBeat);
1348         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1349                 return -EFAULT;
1350         return 0;
1351 }
1352
1353 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1354 {
1355         BusTypes_type BusTypes;
1356
1357         if (!argp)
1358                 return -EINVAL;
1359         BusTypes = readl(&h->cfgtable->BusTypes);
1360         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1361                 return -EFAULT;
1362         return 0;
1363 }
1364
1365 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1366 {
1367         FirmwareVer_type firmware;
1368
1369         if (!argp)
1370                 return -EINVAL;
1371         memcpy(firmware, h->firm_ver, 4);
1372
1373         if (copy_to_user
1374             (argp, firmware, sizeof(FirmwareVer_type)))
1375                 return -EFAULT;
1376         return 0;
1377 }
1378
1379 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1380 {
1381         DriverVer_type DriverVer = DRIVER_VERSION;
1382
1383         if (!argp)
1384                 return -EINVAL;
1385         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1386                 return -EFAULT;
1387         return 0;
1388 }
1389
1390 static int cciss_getluninfo(ctlr_info_t *h,
1391         struct gendisk *disk, void __user *argp)
1392 {
1393         LogvolInfo_struct luninfo;
1394         drive_info_struct *drv = get_drv(disk);
1395
1396         if (!argp)
1397                 return -EINVAL;
1398         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1399         luninfo.num_opens = drv->usage_count;
1400         luninfo.num_parts = 0;
1401         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1402                 return -EFAULT;
1403         return 0;
1404 }
1405
1406 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1407 {
1408         IOCTL_Command_struct iocommand;
1409         CommandList_struct *c;
1410         char *buff = NULL;
1411         u64bit temp64;
1412         DECLARE_COMPLETION_ONSTACK(wait);
1413
1414         if (!argp)
1415                 return -EINVAL;
1416
1417         if (!capable(CAP_SYS_RAWIO))
1418                 return -EPERM;
1419
1420         if (copy_from_user
1421             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1422                 return -EFAULT;
1423         if ((iocommand.buf_size < 1) &&
1424             (iocommand.Request.Type.Direction != XFER_NONE)) {
1425                 return -EINVAL;
1426         }
1427         if (iocommand.buf_size > 0) {
1428                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1429                 if (buff == NULL)
1430                         return -EFAULT;
1431         }
1432         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1433                 /* Copy the data into the buffer we created */
1434                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1435                         kfree(buff);
1436                         return -EFAULT;
1437                 }
1438         } else {
1439                 memset(buff, 0, iocommand.buf_size);
1440         }
1441         c = cmd_special_alloc(h);
1442         if (!c) {
1443                 kfree(buff);
1444                 return -ENOMEM;
1445         }
1446         /* Fill in the command type */
1447         c->cmd_type = CMD_IOCTL_PEND;
1448         /* Fill in Command Header */
1449         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1450         if (iocommand.buf_size > 0) { /* buffer to fill */
1451                 c->Header.SGList = 1;
1452                 c->Header.SGTotal = 1;
1453         } else { /* no buffers to fill */
1454                 c->Header.SGList = 0;
1455                 c->Header.SGTotal = 0;
1456         }
1457         c->Header.LUN = iocommand.LUN_info;
1458         /* use the kernel address the cmd block for tag */
1459         c->Header.Tag.lower = c->busaddr;
1460
1461         /* Fill in Request block */
1462         c->Request = iocommand.Request;
1463
1464         /* Fill in the scatter gather information */
1465         if (iocommand.buf_size > 0) {
1466                 temp64.val = pci_map_single(h->pdev, buff,
1467                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1468                 c->SG[0].Addr.lower = temp64.val32.lower;
1469                 c->SG[0].Addr.upper = temp64.val32.upper;
1470                 c->SG[0].Len = iocommand.buf_size;
1471                 c->SG[0].Ext = 0;  /* we are not chaining */
1472         }
1473         c->waiting = &wait;
1474
1475         enqueue_cmd_and_start_io(h, c);
1476         wait_for_completion(&wait);
1477
1478         /* unlock the buffers from DMA */
1479         temp64.val32.lower = c->SG[0].Addr.lower;
1480         temp64.val32.upper = c->SG[0].Addr.upper;
1481         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1482                          PCI_DMA_BIDIRECTIONAL);
1483         check_ioctl_unit_attention(h, c);
1484
1485         /* Copy the error information out */
1486         iocommand.error_info = *(c->err_info);
1487         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1488                 kfree(buff);
1489                 cmd_special_free(h, c);
1490                 return -EFAULT;
1491         }
1492
1493         if (iocommand.Request.Type.Direction == XFER_READ) {
1494                 /* Copy the data out of the buffer we created */
1495                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1496                         kfree(buff);
1497                         cmd_special_free(h, c);
1498                         return -EFAULT;
1499                 }
1500         }
1501         kfree(buff);
1502         cmd_special_free(h, c);
1503         return 0;
1504 }
1505
1506 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1507 {
1508         BIG_IOCTL_Command_struct *ioc;
1509         CommandList_struct *c;
1510         unsigned char **buff = NULL;
1511         int *buff_size = NULL;
1512         u64bit temp64;
1513         BYTE sg_used = 0;
1514         int status = 0;
1515         int i;
1516         DECLARE_COMPLETION_ONSTACK(wait);
1517         __u32 left;
1518         __u32 sz;
1519         BYTE __user *data_ptr;
1520
1521         if (!argp)
1522                 return -EINVAL;
1523         if (!capable(CAP_SYS_RAWIO))
1524                 return -EPERM;
1525         ioc = (BIG_IOCTL_Command_struct *)
1526             kmalloc(sizeof(*ioc), GFP_KERNEL);
1527         if (!ioc) {
1528                 status = -ENOMEM;
1529                 goto cleanup1;
1530         }
1531         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1532                 status = -EFAULT;
1533                 goto cleanup1;
1534         }
1535         if ((ioc->buf_size < 1) &&
1536             (ioc->Request.Type.Direction != XFER_NONE)) {
1537                 status = -EINVAL;
1538                 goto cleanup1;
1539         }
1540         /* Check kmalloc limits  using all SGs */
1541         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1542                 status = -EINVAL;
1543                 goto cleanup1;
1544         }
1545         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1546                 status = -EINVAL;
1547                 goto cleanup1;
1548         }
1549         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1550         if (!buff) {
1551                 status = -ENOMEM;
1552                 goto cleanup1;
1553         }
1554         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1555         if (!buff_size) {
1556                 status = -ENOMEM;
1557                 goto cleanup1;
1558         }
1559         left = ioc->buf_size;
1560         data_ptr = ioc->buf;
1561         while (left) {
1562                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1563                 buff_size[sg_used] = sz;
1564                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1565                 if (buff[sg_used] == NULL) {
1566                         status = -ENOMEM;
1567                         goto cleanup1;
1568                 }
1569                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1570                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1571                                 status = -EFAULT;
1572                                 goto cleanup1;
1573                         }
1574                 } else {
1575                         memset(buff[sg_used], 0, sz);
1576                 }
1577                 left -= sz;
1578                 data_ptr += sz;
1579                 sg_used++;
1580         }
1581         c = cmd_special_alloc(h);
1582         if (!c) {
1583                 status = -ENOMEM;
1584                 goto cleanup1;
1585         }
1586         c->cmd_type = CMD_IOCTL_PEND;
1587         c->Header.ReplyQueue = 0;
1588         c->Header.SGList = sg_used;
1589         c->Header.SGTotal = sg_used;
1590         c->Header.LUN = ioc->LUN_info;
1591         c->Header.Tag.lower = c->busaddr;
1592
1593         c->Request = ioc->Request;
1594         for (i = 0; i < sg_used; i++) {
1595                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1596                                     PCI_DMA_BIDIRECTIONAL);
1597                 c->SG[i].Addr.lower = temp64.val32.lower;
1598                 c->SG[i].Addr.upper = temp64.val32.upper;
1599                 c->SG[i].Len = buff_size[i];
1600                 c->SG[i].Ext = 0;       /* we are not chaining */
1601         }
1602         c->waiting = &wait;
1603         enqueue_cmd_and_start_io(h, c);
1604         wait_for_completion(&wait);
1605         /* unlock the buffers from DMA */
1606         for (i = 0; i < sg_used; i++) {
1607                 temp64.val32.lower = c->SG[i].Addr.lower;
1608                 temp64.val32.upper = c->SG[i].Addr.upper;
1609                 pci_unmap_single(h->pdev,
1610                         (dma_addr_t) temp64.val, buff_size[i],
1611                         PCI_DMA_BIDIRECTIONAL);
1612         }
1613         check_ioctl_unit_attention(h, c);
1614         /* Copy the error information out */
1615         ioc->error_info = *(c->err_info);
1616         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1617                 cmd_special_free(h, c);
1618                 status = -EFAULT;
1619                 goto cleanup1;
1620         }
1621         if (ioc->Request.Type.Direction == XFER_READ) {
1622                 /* Copy the data out of the buffer we created */
1623                 BYTE __user *ptr = ioc->buf;
1624                 for (i = 0; i < sg_used; i++) {
1625                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1626                                 cmd_special_free(h, c);
1627                                 status = -EFAULT;
1628                                 goto cleanup1;
1629                         }
1630                         ptr += buff_size[i];
1631                 }
1632         }
1633         cmd_special_free(h, c);
1634         status = 0;
1635 cleanup1:
1636         if (buff) {
1637                 for (i = 0; i < sg_used; i++)
1638                         kfree(buff[i]);
1639                 kfree(buff);
1640         }
1641         kfree(buff_size);
1642         kfree(ioc);
1643         return status;
1644 }
1645
1646 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1647         unsigned int cmd, unsigned long arg)
1648 {
1649         struct gendisk *disk = bdev->bd_disk;
1650         ctlr_info_t *h = get_host(disk);
1651         void __user *argp = (void __user *)arg;
1652
1653         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1654                 cmd, arg);
1655         switch (cmd) {
1656         case CCISS_GETPCIINFO:
1657                 return cciss_getpciinfo(h, argp);
1658         case CCISS_GETINTINFO:
1659                 return cciss_getintinfo(h, argp);
1660         case CCISS_SETINTINFO:
1661                 return cciss_setintinfo(h, argp);
1662         case CCISS_GETNODENAME:
1663                 return cciss_getnodename(h, argp);
1664         case CCISS_SETNODENAME:
1665                 return cciss_setnodename(h, argp);
1666         case CCISS_GETHEARTBEAT:
1667                 return cciss_getheartbeat(h, argp);
1668         case CCISS_GETBUSTYPES:
1669                 return cciss_getbustypes(h, argp);
1670         case CCISS_GETFIRMVER:
1671                 return cciss_getfirmver(h, argp);
1672         case CCISS_GETDRIVVER:
1673                 return cciss_getdrivver(h, argp);
1674         case CCISS_DEREGDISK:
1675         case CCISS_REGNEWD:
1676         case CCISS_REVALIDVOLS:
1677                 return rebuild_lun_table(h, 0, 1);
1678         case CCISS_GETLUNINFO:
1679                 return cciss_getluninfo(h, disk, argp);
1680         case CCISS_PASSTHRU:
1681                 return cciss_passthru(h, argp);
1682         case CCISS_BIG_PASSTHRU:
1683                 return cciss_bigpassthru(h, argp);
1684
1685         /* scsi_cmd_ioctl handles these, below, though some are not */
1686         /* very meaningful for cciss.  SG_IO is the main one people want. */
1687
1688         case SG_GET_VERSION_NUM:
1689         case SG_SET_TIMEOUT:
1690         case SG_GET_TIMEOUT:
1691         case SG_GET_RESERVED_SIZE:
1692         case SG_SET_RESERVED_SIZE:
1693         case SG_EMULATED_HOST:
1694         case SG_IO:
1695         case SCSI_IOCTL_SEND_COMMAND:
1696                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1697
1698         /* scsi_cmd_ioctl would normally handle these, below, but */
1699         /* they aren't a good fit for cciss, as CD-ROMs are */
1700         /* not supported, and we don't have any bus/target/lun */
1701         /* which we present to the kernel. */
1702
1703         case CDROM_SEND_PACKET:
1704         case CDROMCLOSETRAY:
1705         case CDROMEJECT:
1706         case SCSI_IOCTL_GET_IDLUN:
1707         case SCSI_IOCTL_GET_BUS_NUMBER:
1708         default:
1709                 return -ENOTTY;
1710         }
1711 }
1712
1713 static void cciss_check_queues(ctlr_info_t *h)
1714 {
1715         int start_queue = h->next_to_run;
1716         int i;
1717
1718         /* check to see if we have maxed out the number of commands that can
1719          * be placed on the queue.  If so then exit.  We do this check here
1720          * in case the interrupt we serviced was from an ioctl and did not
1721          * free any new commands.
1722          */
1723         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1724                 return;
1725
1726         /* We have room on the queue for more commands.  Now we need to queue
1727          * them up.  We will also keep track of the next queue to run so
1728          * that every queue gets a chance to be started first.
1729          */
1730         for (i = 0; i < h->highest_lun + 1; i++) {
1731                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1732                 /* make sure the disk has been added and the drive is real
1733                  * because this can be called from the middle of init_one.
1734                  */
1735                 if (!h->drv[curr_queue])
1736                         continue;
1737                 if (!(h->drv[curr_queue]->queue) ||
1738                         !(h->drv[curr_queue]->heads))
1739                         continue;
1740                 blk_start_queue(h->gendisk[curr_queue]->queue);
1741
1742                 /* check to see if we have maxed out the number of commands
1743                  * that can be placed on the queue.
1744                  */
1745                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1746                         if (curr_queue == start_queue) {
1747                                 h->next_to_run =
1748                                     (start_queue + 1) % (h->highest_lun + 1);
1749                                 break;
1750                         } else {
1751                                 h->next_to_run = curr_queue;
1752                                 break;
1753                         }
1754                 }
1755         }
1756 }
1757
1758 static void cciss_softirq_done(struct request *rq)
1759 {
1760         CommandList_struct *c = rq->completion_data;
1761         ctlr_info_t *h = hba[c->ctlr];
1762         SGDescriptor_struct *curr_sg = c->SG;
1763         u64bit temp64;
1764         unsigned long flags;
1765         int i, ddir;
1766         int sg_index = 0;
1767
1768         if (c->Request.Type.Direction == XFER_READ)
1769                 ddir = PCI_DMA_FROMDEVICE;
1770         else
1771                 ddir = PCI_DMA_TODEVICE;
1772
1773         /* command did not need to be retried */
1774         /* unmap the DMA mapping for all the scatter gather elements */
1775         for (i = 0; i < c->Header.SGList; i++) {
1776                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1777                         cciss_unmap_sg_chain_block(h, c);
1778                         /* Point to the next block */
1779                         curr_sg = h->cmd_sg_list[c->cmdindex];
1780                         sg_index = 0;
1781                 }
1782                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1783                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1784                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1785                                 ddir);
1786                 ++sg_index;
1787         }
1788
1789         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1790
1791         /* set the residual count for pc requests */
1792         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1793                 rq->resid_len = c->err_info->ResidualCnt;
1794
1795         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1796
1797         spin_lock_irqsave(&h->lock, flags);
1798         cmd_free(h, c);
1799         cciss_check_queues(h);
1800         spin_unlock_irqrestore(&h->lock, flags);
1801 }
1802
1803 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1804         unsigned char scsi3addr[], uint32_t log_unit)
1805 {
1806         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1807                 sizeof(h->drv[log_unit]->LunID));
1808 }
1809
1810 /* This function gets the SCSI vendor, model, and revision of a logical drive
1811  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1812  * they cannot be read.
1813  */
1814 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1815                                    char *vendor, char *model, char *rev)
1816 {
1817         int rc;
1818         InquiryData_struct *inq_buf;
1819         unsigned char scsi3addr[8];
1820
1821         *vendor = '\0';
1822         *model = '\0';
1823         *rev = '\0';
1824
1825         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1826         if (!inq_buf)
1827                 return;
1828
1829         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1830         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1831                         scsi3addr, TYPE_CMD);
1832         if (rc == IO_OK) {
1833                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1834                 vendor[VENDOR_LEN] = '\0';
1835                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1836                 model[MODEL_LEN] = '\0';
1837                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1838                 rev[REV_LEN] = '\0';
1839         }
1840
1841         kfree(inq_buf);
1842         return;
1843 }
1844
1845 /* This function gets the serial number of a logical drive via
1846  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1847  * number cannot be had, for whatever reason, 16 bytes of 0xff
1848  * are returned instead.
1849  */
1850 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1851                                 unsigned char *serial_no, int buflen)
1852 {
1853 #define PAGE_83_INQ_BYTES 64
1854         int rc;
1855         unsigned char *buf;
1856         unsigned char scsi3addr[8];
1857
1858         if (buflen > 16)
1859                 buflen = 16;
1860         memset(serial_no, 0xff, buflen);
1861         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1862         if (!buf)
1863                 return;
1864         memset(serial_no, 0, buflen);
1865         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1866         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1867                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1868         if (rc == IO_OK)
1869                 memcpy(serial_no, &buf[8], buflen);
1870         kfree(buf);
1871         return;
1872 }
1873
1874 /*
1875  * cciss_add_disk sets up the block device queue for a logical drive
1876  */
1877 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1878                                 int drv_index)
1879 {
1880         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1881         if (!disk->queue)
1882                 goto init_queue_failure;
1883         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1884         disk->major = h->major;
1885         disk->first_minor = drv_index << NWD_SHIFT;
1886         disk->fops = &cciss_fops;
1887         if (cciss_create_ld_sysfs_entry(h, drv_index))
1888                 goto cleanup_queue;
1889         disk->private_data = h->drv[drv_index];
1890         disk->driverfs_dev = &h->drv[drv_index]->dev;
1891
1892         /* Set up queue information */
1893         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1894
1895         /* This is a hardware imposed limit. */
1896         blk_queue_max_segments(disk->queue, h->maxsgentries);
1897
1898         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1899
1900         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1901
1902         disk->queue->queuedata = h;
1903
1904         blk_queue_logical_block_size(disk->queue,
1905                                      h->drv[drv_index]->block_size);
1906
1907         /* Make sure all queue data is written out before */
1908         /* setting h->drv[drv_index]->queue, as setting this */
1909         /* allows the interrupt handler to start the queue */
1910         wmb();
1911         h->drv[drv_index]->queue = disk->queue;
1912         add_disk(disk);
1913         return 0;
1914
1915 cleanup_queue:
1916         blk_cleanup_queue(disk->queue);
1917         disk->queue = NULL;
1918 init_queue_failure:
1919         return -1;
1920 }
1921
1922 /* This function will check the usage_count of the drive to be updated/added.
1923  * If the usage_count is zero and it is a heretofore unknown drive, or,
1924  * the drive's capacity, geometry, or serial number has changed,
1925  * then the drive information will be updated and the disk will be
1926  * re-registered with the kernel.  If these conditions don't hold,
1927  * then it will be left alone for the next reboot.  The exception to this
1928  * is disk 0 which will always be left registered with the kernel since it
1929  * is also the controller node.  Any changes to disk 0 will show up on
1930  * the next reboot.
1931  */
1932 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1933         int first_time, int via_ioctl)
1934 {
1935         struct gendisk *disk;
1936         InquiryData_struct *inq_buff = NULL;
1937         unsigned int block_size;
1938         sector_t total_size;
1939         unsigned long flags = 0;
1940         int ret = 0;
1941         drive_info_struct *drvinfo;
1942
1943         /* Get information about the disk and modify the driver structure */
1944         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1945         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1946         if (inq_buff == NULL || drvinfo == NULL)
1947                 goto mem_msg;
1948
1949         /* testing to see if 16-byte CDBs are already being used */
1950         if (h->cciss_read == CCISS_READ_16) {
1951                 cciss_read_capacity_16(h, drv_index,
1952                         &total_size, &block_size);
1953
1954         } else {
1955                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1956                 /* if read_capacity returns all F's this volume is >2TB */
1957                 /* in size so we switch to 16-byte CDB's for all */
1958                 /* read/write ops */
1959                 if (total_size == 0xFFFFFFFFULL) {
1960                         cciss_read_capacity_16(h, drv_index,
1961                         &total_size, &block_size);
1962                         h->cciss_read = CCISS_READ_16;
1963                         h->cciss_write = CCISS_WRITE_16;
1964                 } else {
1965                         h->cciss_read = CCISS_READ_10;
1966                         h->cciss_write = CCISS_WRITE_10;
1967                 }
1968         }
1969
1970         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1971                                inq_buff, drvinfo);
1972         drvinfo->block_size = block_size;
1973         drvinfo->nr_blocks = total_size + 1;
1974
1975         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1976                                 drvinfo->model, drvinfo->rev);
1977         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1978                         sizeof(drvinfo->serial_no));
1979         /* Save the lunid in case we deregister the disk, below. */
1980         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1981                 sizeof(drvinfo->LunID));
1982
1983         /* Is it the same disk we already know, and nothing's changed? */
1984         if (h->drv[drv_index]->raid_level != -1 &&
1985                 ((memcmp(drvinfo->serial_no,
1986                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1987                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1988                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1989                 drvinfo->heads == h->drv[drv_index]->heads &&
1990                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1991                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1992                         /* The disk is unchanged, nothing to update */
1993                         goto freeret;
1994
1995         /* If we get here it's not the same disk, or something's changed,
1996          * so we need to * deregister it, and re-register it, if it's not
1997          * in use.
1998          * If the disk already exists then deregister it before proceeding
1999          * (unless it's the first disk (for the controller node).
2000          */
2001         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2002                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2003                 spin_lock_irqsave(&h->lock, flags);
2004                 h->drv[drv_index]->busy_configuring = 1;
2005                 spin_unlock_irqrestore(&h->lock, flags);
2006
2007                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2008                  * which keeps the interrupt handler from starting
2009                  * the queue.
2010                  */
2011                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2012         }
2013
2014         /* If the disk is in use return */
2015         if (ret)
2016                 goto freeret;
2017
2018         /* Save the new information from cciss_geometry_inquiry
2019          * and serial number inquiry.  If the disk was deregistered
2020          * above, then h->drv[drv_index] will be NULL.
2021          */
2022         if (h->drv[drv_index] == NULL) {
2023                 drvinfo->device_initialized = 0;
2024                 h->drv[drv_index] = drvinfo;
2025                 drvinfo = NULL; /* so it won't be freed below. */
2026         } else {
2027                 /* special case for cxd0 */
2028                 h->drv[drv_index]->block_size = drvinfo->block_size;
2029                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2030                 h->drv[drv_index]->heads = drvinfo->heads;
2031                 h->drv[drv_index]->sectors = drvinfo->sectors;
2032                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2033                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2034                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2035                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2036                         VENDOR_LEN + 1);
2037                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2038                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2039         }
2040
2041         ++h->num_luns;
2042         disk = h->gendisk[drv_index];
2043         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2044
2045         /* If it's not disk 0 (drv_index != 0)
2046          * or if it was disk 0, but there was previously
2047          * no actual corresponding configured logical drive
2048          * (raid_leve == -1) then we want to update the
2049          * logical drive's information.
2050          */
2051         if (drv_index || first_time) {
2052                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2053                         cciss_free_gendisk(h, drv_index);
2054                         cciss_free_drive_info(h, drv_index);
2055                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2056                                 drv_index);
2057                         --h->num_luns;
2058                 }
2059         }
2060
2061 freeret:
2062         kfree(inq_buff);
2063         kfree(drvinfo);
2064         return;
2065 mem_msg:
2066         dev_err(&h->pdev->dev, "out of memory\n");
2067         goto freeret;
2068 }
2069
2070 /* This function will find the first index of the controllers drive array
2071  * that has a null drv pointer and allocate the drive info struct and
2072  * will return that index   This is where new drives will be added.
2073  * If the index to be returned is greater than the highest_lun index for
2074  * the controller then highest_lun is set * to this new index.
2075  * If there are no available indexes or if tha allocation fails, then -1
2076  * is returned.  * "controller_node" is used to know if this is a real
2077  * logical drive, or just the controller node, which determines if this
2078  * counts towards highest_lun.
2079  */
2080 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2081 {
2082         int i;
2083         drive_info_struct *drv;
2084
2085         /* Search for an empty slot for our drive info */
2086         for (i = 0; i < CISS_MAX_LUN; i++) {
2087
2088                 /* if not cxd0 case, and it's occupied, skip it. */
2089                 if (h->drv[i] && i != 0)
2090                         continue;
2091                 /*
2092                  * If it's cxd0 case, and drv is alloc'ed already, and a
2093                  * disk is configured there, skip it.
2094                  */
2095                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2096                         continue;
2097
2098                 /*
2099                  * We've found an empty slot.  Update highest_lun
2100                  * provided this isn't just the fake cxd0 controller node.
2101                  */
2102                 if (i > h->highest_lun && !controller_node)
2103                         h->highest_lun = i;
2104
2105                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2106                 if (i == 0 && h->drv[i] != NULL)
2107                         return i;
2108
2109                 /*
2110                  * Found an empty slot, not already alloc'ed.  Allocate it.
2111                  * Mark it with raid_level == -1, so we know it's new later on.
2112                  */
2113                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2114                 if (!drv)
2115                         return -1;
2116                 drv->raid_level = -1; /* so we know it's new */
2117                 h->drv[i] = drv;
2118                 return i;
2119         }
2120         return -1;
2121 }
2122
2123 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2124 {
2125         kfree(h->drv[drv_index]);
2126         h->drv[drv_index] = NULL;
2127 }
2128
2129 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2130 {
2131         put_disk(h->gendisk[drv_index]);
2132         h->gendisk[drv_index] = NULL;
2133 }
2134
2135 /* cciss_add_gendisk finds a free hba[]->drv structure
2136  * and allocates a gendisk if needed, and sets the lunid
2137  * in the drvinfo structure.   It returns the index into
2138  * the ->drv[] array, or -1 if none are free.
2139  * is_controller_node indicates whether highest_lun should
2140  * count this disk, or if it's only being added to provide
2141  * a means to talk to the controller in case no logical
2142  * drives have yet been configured.
2143  */
2144 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2145         int controller_node)
2146 {
2147         int drv_index;
2148
2149         drv_index = cciss_alloc_drive_info(h, controller_node);
2150         if (drv_index == -1)
2151                 return -1;
2152
2153         /*Check if the gendisk needs to be allocated */
2154         if (!h->gendisk[drv_index]) {
2155                 h->gendisk[drv_index] =
2156                         alloc_disk(1 << NWD_SHIFT);
2157                 if (!h->gendisk[drv_index]) {
2158                         dev_err(&h->pdev->dev,
2159                                 "could not allocate a new disk %d\n",
2160                                 drv_index);
2161                         goto err_free_drive_info;
2162                 }
2163         }
2164         memcpy(h->drv[drv_index]->LunID, lunid,
2165                 sizeof(h->drv[drv_index]->LunID));
2166         if (cciss_create_ld_sysfs_entry(h, drv_index))
2167                 goto err_free_disk;
2168         /* Don't need to mark this busy because nobody */
2169         /* else knows about this disk yet to contend */
2170         /* for access to it. */
2171         h->drv[drv_index]->busy_configuring = 0;
2172         wmb();
2173         return drv_index;
2174
2175 err_free_disk:
2176         cciss_free_gendisk(h, drv_index);
2177 err_free_drive_info:
2178         cciss_free_drive_info(h, drv_index);
2179         return -1;
2180 }
2181
2182 /* This is for the special case of a controller which
2183  * has no logical drives.  In this case, we still need
2184  * to register a disk so the controller can be accessed
2185  * by the Array Config Utility.
2186  */
2187 static void cciss_add_controller_node(ctlr_info_t *h)
2188 {
2189         struct gendisk *disk;
2190         int drv_index;
2191
2192         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2193                 return;
2194
2195         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2196         if (drv_index == -1)
2197                 goto error;
2198         h->drv[drv_index]->block_size = 512;
2199         h->drv[drv_index]->nr_blocks = 0;
2200         h->drv[drv_index]->heads = 0;
2201         h->drv[drv_index]->sectors = 0;
2202         h->drv[drv_index]->cylinders = 0;
2203         h->drv[drv_index]->raid_level = -1;
2204         memset(h->drv[drv_index]->serial_no, 0, 16);
2205         disk = h->gendisk[drv_index];
2206         if (cciss_add_disk(h, disk, drv_index) == 0)
2207                 return;
2208         cciss_free_gendisk(h, drv_index);
2209         cciss_free_drive_info(h, drv_index);
2210 error:
2211         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2212         return;
2213 }
2214
2215 /* This function will add and remove logical drives from the Logical
2216  * drive array of the controller and maintain persistency of ordering
2217  * so that mount points are preserved until the next reboot.  This allows
2218  * for the removal of logical drives in the middle of the drive array
2219  * without a re-ordering of those drives.
2220  * INPUT
2221  * h            = The controller to perform the operations on
2222  */
2223 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2224         int via_ioctl)
2225 {
2226         int num_luns;
2227         ReportLunData_struct *ld_buff = NULL;
2228         int return_code;
2229         int listlength = 0;
2230         int i;
2231         int drv_found;
2232         int drv_index = 0;
2233         unsigned char lunid[8] = CTLR_LUNID;
2234         unsigned long flags;
2235
2236         if (!capable(CAP_SYS_RAWIO))
2237                 return -EPERM;
2238
2239         /* Set busy_configuring flag for this operation */
2240         spin_lock_irqsave(&h->lock, flags);
2241         if (h->busy_configuring) {
2242                 spin_unlock_irqrestore(&h->lock, flags);
2243                 return -EBUSY;
2244         }
2245         h->busy_configuring = 1;
2246         spin_unlock_irqrestore(&h->lock, flags);
2247
2248         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2249         if (ld_buff == NULL)
2250                 goto mem_msg;
2251
2252         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2253                                       sizeof(ReportLunData_struct),
2254                                       0, CTLR_LUNID, TYPE_CMD);
2255
2256         if (return_code == IO_OK)
2257                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2258         else {  /* reading number of logical volumes failed */
2259                 dev_warn(&h->pdev->dev,
2260                         "report logical volume command failed\n");
2261                 listlength = 0;
2262                 goto freeret;
2263         }
2264
2265         num_luns = listlength / 8;      /* 8 bytes per entry */
2266         if (num_luns > CISS_MAX_LUN) {
2267                 num_luns = CISS_MAX_LUN;
2268                 dev_warn(&h->pdev->dev, "more luns configured"
2269                        " on controller than can be handled by"
2270                        " this driver.\n");
2271         }
2272
2273         if (num_luns == 0)
2274                 cciss_add_controller_node(h);
2275
2276         /* Compare controller drive array to driver's drive array
2277          * to see if any drives are missing on the controller due
2278          * to action of Array Config Utility (user deletes drive)
2279          * and deregister logical drives which have disappeared.
2280          */
2281         for (i = 0; i <= h->highest_lun; i++) {
2282                 int j;
2283                 drv_found = 0;
2284
2285                 /* skip holes in the array from already deleted drives */
2286                 if (h->drv[i] == NULL)
2287                         continue;
2288
2289                 for (j = 0; j < num_luns; j++) {
2290                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2291                         if (memcmp(h->drv[i]->LunID, lunid,
2292                                 sizeof(lunid)) == 0) {
2293                                 drv_found = 1;
2294                                 break;
2295                         }
2296                 }
2297                 if (!drv_found) {
2298                         /* Deregister it from the OS, it's gone. */
2299                         spin_lock_irqsave(&h->lock, flags);
2300                         h->drv[i]->busy_configuring = 1;
2301                         spin_unlock_irqrestore(&h->lock, flags);
2302                         return_code = deregister_disk(h, i, 1, via_ioctl);
2303                         if (h->drv[i] != NULL)
2304                                 h->drv[i]->busy_configuring = 0;
2305                 }
2306         }
2307
2308         /* Compare controller drive array to driver's drive array.
2309          * Check for updates in the drive information and any new drives
2310          * on the controller due to ACU adding logical drives, or changing
2311          * a logical drive's size, etc.  Reregister any new/changed drives
2312          */
2313         for (i = 0; i < num_luns; i++) {
2314                 int j;
2315
2316                 drv_found = 0;
2317
2318                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2319                 /* Find if the LUN is already in the drive array
2320                  * of the driver.  If so then update its info
2321                  * if not in use.  If it does not exist then find
2322                  * the first free index and add it.
2323                  */
2324                 for (j = 0; j <= h->highest_lun; j++) {
2325                         if (h->drv[j] != NULL &&
2326                                 memcmp(h->drv[j]->LunID, lunid,
2327                                         sizeof(h->drv[j]->LunID)) == 0) {
2328                                 drv_index = j;
2329                                 drv_found = 1;
2330                                 break;
2331                         }
2332                 }
2333
2334                 /* check if the drive was found already in the array */
2335                 if (!drv_found) {
2336                         drv_index = cciss_add_gendisk(h, lunid, 0);
2337                         if (drv_index == -1)
2338                                 goto freeret;
2339                 }
2340                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2341         }               /* end for */
2342
2343 freeret:
2344         kfree(ld_buff);
2345         h->busy_configuring = 0;
2346         /* We return -1 here to tell the ACU that we have registered/updated
2347          * all of the drives that we can and to keep it from calling us
2348          * additional times.
2349          */
2350         return -1;
2351 mem_msg:
2352         dev_err(&h->pdev->dev, "out of memory\n");
2353         h->busy_configuring = 0;
2354         goto freeret;
2355 }
2356
2357 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2358 {
2359         /* zero out the disk size info */
2360         drive_info->nr_blocks = 0;
2361         drive_info->block_size = 0;
2362         drive_info->heads = 0;
2363         drive_info->sectors = 0;
2364         drive_info->cylinders = 0;
2365         drive_info->raid_level = -1;
2366         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2367         memset(drive_info->model, 0, sizeof(drive_info->model));
2368         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2369         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2370         /*
2371          * don't clear the LUNID though, we need to remember which
2372          * one this one is.
2373          */
2374 }
2375
2376 /* This function will deregister the disk and it's queue from the
2377  * kernel.  It must be called with the controller lock held and the
2378  * drv structures busy_configuring flag set.  It's parameters are:
2379  *
2380  * disk = This is the disk to be deregistered
2381  * drv  = This is the drive_info_struct associated with the disk to be
2382  *        deregistered.  It contains information about the disk used
2383  *        by the driver.
2384  * clear_all = This flag determines whether or not the disk information
2385  *             is going to be completely cleared out and the highest_lun
2386  *             reset.  Sometimes we want to clear out information about
2387  *             the disk in preparation for re-adding it.  In this case
2388  *             the highest_lun should be left unchanged and the LunID
2389  *             should not be cleared.
2390  * via_ioctl
2391  *    This indicates whether we've reached this path via ioctl.
2392  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2393  *    If this path is reached via ioctl(), then the max_usage_count will
2394  *    be 1, as the process calling ioctl() has got to have the device open.
2395  *    If we get here via sysfs, then the max usage count will be zero.
2396 */
2397 static int deregister_disk(ctlr_info_t *h, int drv_index,
2398                            int clear_all, int via_ioctl)
2399 {
2400         int i;
2401         struct gendisk *disk;
2402         drive_info_struct *drv;
2403         int recalculate_highest_lun;
2404
2405         if (!capable(CAP_SYS_RAWIO))
2406                 return -EPERM;
2407
2408         drv = h->drv[drv_index];
2409         disk = h->gendisk[drv_index];
2410
2411         /* make sure logical volume is NOT is use */
2412         if (clear_all || (h->gendisk[0] == disk)) {
2413                 if (drv->usage_count > via_ioctl)
2414                         return -EBUSY;
2415         } else if (drv->usage_count > 0)
2416                 return -EBUSY;
2417
2418         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2419
2420         /* invalidate the devices and deregister the disk.  If it is disk
2421          * zero do not deregister it but just zero out it's values.  This
2422          * allows us to delete disk zero but keep the controller registered.
2423          */
2424         if (h->gendisk[0] != disk) {
2425                 struct request_queue *q = disk->queue;
2426                 if (disk->flags & GENHD_FL_UP) {
2427                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2428                         del_gendisk(disk);
2429                 }
2430                 if (q)
2431                         blk_cleanup_queue(q);
2432                 /* If clear_all is set then we are deleting the logical
2433                  * drive, not just refreshing its info.  For drives
2434                  * other than disk 0 we will call put_disk.  We do not
2435                  * do this for disk 0 as we need it to be able to
2436                  * configure the controller.
2437                  */
2438                 if (clear_all){
2439                         /* This isn't pretty, but we need to find the
2440                          * disk in our array and NULL our the pointer.
2441                          * This is so that we will call alloc_disk if
2442                          * this index is used again later.
2443                          */
2444                         for (i=0; i < CISS_MAX_LUN; i++){
2445                                 if (h->gendisk[i] == disk) {
2446                                         h->gendisk[i] = NULL;
2447                                         break;
2448                                 }
2449                         }
2450                         put_disk(disk);
2451                 }
2452         } else {
2453                 set_capacity(disk, 0);
2454                 cciss_clear_drive_info(drv);
2455         }
2456
2457         --h->num_luns;
2458
2459         /* if it was the last disk, find the new hightest lun */
2460         if (clear_all && recalculate_highest_lun) {
2461                 int newhighest = -1;
2462                 for (i = 0; i <= h->highest_lun; i++) {
2463                         /* if the disk has size > 0, it is available */
2464                         if (h->drv[i] && h->drv[i]->heads)
2465                                 newhighest = i;
2466                 }
2467                 h->highest_lun = newhighest;
2468         }
2469         return 0;
2470 }
2471
2472 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2473                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2474                 int cmd_type)
2475 {
2476         u64bit buff_dma_handle;
2477         int status = IO_OK;
2478
2479         c->cmd_type = CMD_IOCTL_PEND;
2480         c->Header.ReplyQueue = 0;
2481         if (buff != NULL) {
2482                 c->Header.SGList = 1;
2483                 c->Header.SGTotal = 1;
2484         } else {
2485                 c->Header.SGList = 0;
2486                 c->Header.SGTotal = 0;
2487         }
2488         c->Header.Tag.lower = c->busaddr;
2489         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2490
2491         c->Request.Type.Type = cmd_type;
2492         if (cmd_type == TYPE_CMD) {
2493                 switch (cmd) {
2494                 case CISS_INQUIRY:
2495                         /* are we trying to read a vital product page */
2496                         if (page_code != 0) {
2497                                 c->Request.CDB[1] = 0x01;
2498                                 c->Request.CDB[2] = page_code;
2499                         }
2500                         c->Request.CDBLen = 6;
2501                         c->Request.Type.Attribute = ATTR_SIMPLE;
2502                         c->Request.Type.Direction = XFER_READ;
2503                         c->Request.Timeout = 0;
2504                         c->Request.CDB[0] = CISS_INQUIRY;
2505                         c->Request.CDB[4] = size & 0xFF;
2506                         break;
2507                 case CISS_REPORT_LOG:
2508                 case CISS_REPORT_PHYS:
2509                         /* Talking to controller so It's a physical command
2510                            mode = 00 target = 0.  Nothing to write.
2511                          */
2512                         c->Request.CDBLen = 12;
2513                         c->Request.Type.Attribute = ATTR_SIMPLE;
2514                         c->Request.Type.Direction = XFER_READ;
2515                         c->Request.Timeout = 0;
2516                         c->Request.CDB[0] = cmd;
2517                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2518                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2519                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2520                         c->Request.CDB[9] = size & 0xFF;
2521                         break;
2522
2523                 case CCISS_READ_CAPACITY:
2524                         c->Request.CDBLen = 10;
2525                         c->Request.Type.Attribute = ATTR_SIMPLE;
2526                         c->Request.Type.Direction = XFER_READ;
2527                         c->Request.Timeout = 0;
2528                         c->Request.CDB[0] = cmd;
2529                         break;
2530                 case CCISS_READ_CAPACITY_16:
2531                         c->Request.CDBLen = 16;
2532                         c->Request.Type.Attribute = ATTR_SIMPLE;
2533                         c->Request.Type.Direction = XFER_READ;
2534                         c->Request.Timeout = 0;
2535                         c->Request.CDB[0] = cmd;
2536                         c->Request.CDB[1] = 0x10;
2537                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2538                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2539                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2540                         c->Request.CDB[13] = size & 0xFF;
2541                         c->Request.Timeout = 0;
2542                         c->Request.CDB[0] = cmd;
2543                         break;
2544                 case CCISS_CACHE_FLUSH:
2545                         c->Request.CDBLen = 12;
2546                         c->Request.Type.Attribute = ATTR_SIMPLE;
2547                         c->Request.Type.Direction = XFER_WRITE;
2548                         c->Request.Timeout = 0;
2549                         c->Request.CDB[0] = BMIC_WRITE;
2550                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2551                         break;
2552                 case TEST_UNIT_READY:
2553                         c->Request.CDBLen = 6;
2554                         c->Request.Type.Attribute = ATTR_SIMPLE;
2555                         c->Request.Type.Direction = XFER_NONE;
2556                         c->Request.Timeout = 0;
2557                         break;
2558                 default:
2559                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2560                         return IO_ERROR;
2561                 }
2562         } else if (cmd_type == TYPE_MSG) {
2563                 switch (cmd) {
2564                 case 0: /* ABORT message */
2565                         c->Request.CDBLen = 12;
2566                         c->Request.Type.Attribute = ATTR_SIMPLE;
2567                         c->Request.Type.Direction = XFER_WRITE;
2568                         c->Request.Timeout = 0;
2569                         c->Request.CDB[0] = cmd;        /* abort */
2570                         c->Request.CDB[1] = 0;  /* abort a command */
2571                         /* buff contains the tag of the command to abort */
2572                         memcpy(&c->Request.CDB[4], buff, 8);
2573                         break;
2574                 case 1: /* RESET message */
2575                         c->Request.CDBLen = 16;
2576                         c->Request.Type.Attribute = ATTR_SIMPLE;
2577                         c->Request.Type.Direction = XFER_NONE;
2578                         c->Request.Timeout = 0;
2579                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2580                         c->Request.CDB[0] = cmd;        /* reset */
2581                         c->Request.CDB[1] = 0x03;       /* reset a target */
2582                         break;
2583                 case 3: /* No-Op message */
2584                         c->Request.CDBLen = 1;
2585                         c->Request.Type.Attribute = ATTR_SIMPLE;
2586                         c->Request.Type.Direction = XFER_WRITE;
2587                         c->Request.Timeout = 0;
2588                         c->Request.CDB[0] = cmd;
2589                         break;
2590                 default:
2591                         dev_warn(&h->pdev->dev,
2592                                 "unknown message type %d\n", cmd);
2593                         return IO_ERROR;
2594                 }
2595         } else {
2596                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2597                 return IO_ERROR;
2598         }
2599         /* Fill in the scatter gather information */
2600         if (size > 0) {
2601                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2602                                                              buff, size,
2603                                                              PCI_DMA_BIDIRECTIONAL);
2604                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2605                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2606                 c->SG[0].Len = size;
2607                 c->SG[0].Ext = 0;       /* we are not chaining */
2608         }
2609         return status;
2610 }
2611
2612 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2613 {
2614         switch (c->err_info->ScsiStatus) {
2615         case SAM_STAT_GOOD:
2616                 return IO_OK;
2617         case SAM_STAT_CHECK_CONDITION:
2618                 switch (0xf & c->err_info->SenseInfo[2]) {
2619                 case 0: return IO_OK; /* no sense */
2620                 case 1: return IO_OK; /* recovered error */
2621                 default:
2622                         if (check_for_unit_attention(h, c))
2623                                 return IO_NEEDS_RETRY;
2624                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2625                                 "check condition, sense key = 0x%02x\n",
2626                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2627                 }
2628                 break;
2629         default:
2630                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2631                         "scsi status = 0x%02x\n",
2632                         c->Request.CDB[0], c->err_info->ScsiStatus);
2633                 break;
2634         }
2635         return IO_ERROR;
2636 }
2637
2638 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2639 {
2640         int return_status = IO_OK;
2641
2642         if (c->err_info->CommandStatus == CMD_SUCCESS)
2643                 return IO_OK;
2644
2645         switch (c->err_info->CommandStatus) {
2646         case CMD_TARGET_STATUS:
2647                 return_status = check_target_status(h, c);
2648                 break;
2649         case CMD_DATA_UNDERRUN:
2650         case CMD_DATA_OVERRUN:
2651                 /* expected for inquiry and report lun commands */
2652                 break;
2653         case CMD_INVALID:
2654                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2655                        "reported invalid\n", c->Request.CDB[0]);
2656                 return_status = IO_ERROR;
2657                 break;
2658         case CMD_PROTOCOL_ERR:
2659                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2660                        "protocol error\n", c->Request.CDB[0]);
2661                 return_status = IO_ERROR;
2662                 break;
2663         case CMD_HARDWARE_ERR:
2664                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2665                        " hardware error\n", c->Request.CDB[0]);
2666                 return_status = IO_ERROR;
2667                 break;
2668         case CMD_CONNECTION_LOST:
2669                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2670                        "connection lost\n", c->Request.CDB[0]);
2671                 return_status = IO_ERROR;
2672                 break;
2673         case CMD_ABORTED:
2674                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2675                        "aborted\n", c->Request.CDB[0]);
2676                 return_status = IO_ERROR;
2677                 break;
2678         case CMD_ABORT_FAILED:
2679                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2680                        "abort failed\n", c->Request.CDB[0]);
2681                 return_status = IO_ERROR;
2682                 break;
2683         case CMD_UNSOLICITED_ABORT:
2684                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2685                         c->Request.CDB[0]);
2686                 return_status = IO_NEEDS_RETRY;
2687                 break;
2688         default:
2689                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2690                        "unknown status %x\n", c->Request.CDB[0],
2691                        c->err_info->CommandStatus);
2692                 return_status = IO_ERROR;
2693         }
2694         return return_status;
2695 }
2696
2697 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2698         int attempt_retry)
2699 {
2700         DECLARE_COMPLETION_ONSTACK(wait);
2701         u64bit buff_dma_handle;
2702         int return_status = IO_OK;
2703
2704 resend_cmd2:
2705         c->waiting = &wait;
2706         enqueue_cmd_and_start_io(h, c);
2707
2708         wait_for_completion(&wait);
2709
2710         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2711                 goto command_done;
2712
2713         return_status = process_sendcmd_error(h, c);
2714
2715         if (return_status == IO_NEEDS_RETRY &&
2716                 c->retry_count < MAX_CMD_RETRIES) {
2717                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2718                         c->Request.CDB[0]);
2719                 c->retry_count++;
2720                 /* erase the old error information */
2721                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2722                 return_status = IO_OK;
2723                 INIT_COMPLETION(wait);
2724                 goto resend_cmd2;
2725         }
2726
2727 command_done:
2728         /* unlock the buffers from DMA */
2729         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2730         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2731         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2732                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2733         return return_status;
2734 }
2735
2736 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2737                            __u8 page_code, unsigned char scsi3addr[],
2738                         int cmd_type)
2739 {
2740         CommandList_struct *c;
2741         int return_status;
2742
2743         c = cmd_special_alloc(h);
2744         if (!c)
2745                 return -ENOMEM;
2746         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2747                 scsi3addr, cmd_type);
2748         if (return_status == IO_OK)
2749                 return_status = sendcmd_withirq_core(h, c, 1);
2750
2751         cmd_special_free(h, c);
2752         return return_status;
2753 }
2754
2755 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2756                                    sector_t total_size,
2757                                    unsigned int block_size,
2758                                    InquiryData_struct *inq_buff,
2759                                    drive_info_struct *drv)
2760 {
2761         int return_code;
2762         unsigned long t;
2763         unsigned char scsi3addr[8];
2764
2765         memset(inq_buff, 0, sizeof(InquiryData_struct));
2766         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2767         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2768                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2769         if (return_code == IO_OK) {
2770                 if (inq_buff->data_byte[8] == 0xFF) {
2771                         dev_warn(&h->pdev->dev,
2772                                "reading geometry failed, volume "
2773                                "does not support reading geometry\n");
2774                         drv->heads = 255;
2775                         drv->sectors = 32;      /* Sectors per track */
2776                         drv->cylinders = total_size + 1;
2777                         drv->raid_level = RAID_UNKNOWN;
2778                 } else {
2779                         drv->heads = inq_buff->data_byte[6];
2780                         drv->sectors = inq_buff->data_byte[7];
2781                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2782                         drv->cylinders += inq_buff->data_byte[5];
2783                         drv->raid_level = inq_buff->data_byte[8];
2784                 }
2785                 drv->block_size = block_size;
2786                 drv->nr_blocks = total_size + 1;
2787                 t = drv->heads * drv->sectors;
2788                 if (t > 1) {
2789                         sector_t real_size = total_size + 1;
2790                         unsigned long rem = sector_div(real_size, t);
2791                         if (rem)
2792                                 real_size++;
2793                         drv->cylinders = real_size;
2794                 }
2795         } else {                /* Get geometry failed */
2796                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2797         }
2798 }
2799
2800 static void
2801 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2802                     unsigned int *block_size)
2803 {
2804         ReadCapdata_struct *buf;
2805         int return_code;
2806         unsigned char scsi3addr[8];
2807
2808         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2809         if (!buf) {
2810                 dev_warn(&h->pdev->dev, "out of memory\n");
2811                 return;
2812         }
2813
2814         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2815         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2816                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2817         if (return_code == IO_OK) {
2818                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2819                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2820         } else {                /* read capacity command failed */
2821                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2822                 *total_size = 0;
2823                 *block_size = BLOCK_SIZE;
2824         }
2825         kfree(buf);
2826 }
2827
2828 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2829         sector_t *total_size, unsigned int *block_size)
2830 {
2831         ReadCapdata_struct_16 *buf;
2832         int return_code;
2833         unsigned char scsi3addr[8];
2834
2835         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2836         if (!buf) {
2837                 dev_warn(&h->pdev->dev, "out of memory\n");
2838                 return;
2839         }
2840
2841         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2842         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2843                 buf, sizeof(ReadCapdata_struct_16),
2844                         0, scsi3addr, TYPE_CMD);
2845         if (return_code == IO_OK) {
2846                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2847                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2848         } else {                /* read capacity command failed */
2849                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2850                 *total_size = 0;
2851                 *block_size = BLOCK_SIZE;
2852         }
2853         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2854                (unsigned long long)*total_size+1, *block_size);
2855         kfree(buf);
2856 }
2857
2858 static int cciss_revalidate(struct gendisk *disk)
2859 {
2860         ctlr_info_t *h = get_host(disk);
2861         drive_info_struct *drv = get_drv(disk);
2862         int logvol;
2863         int FOUND = 0;
2864         unsigned int block_size;
2865         sector_t total_size;
2866         InquiryData_struct *inq_buff = NULL;
2867
2868         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2869                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2870                         sizeof(drv->LunID)) == 0) {
2871                         FOUND = 1;
2872                         break;
2873                 }
2874         }
2875
2876         if (!FOUND)
2877                 return 1;
2878
2879         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2880         if (inq_buff == NULL) {
2881                 dev_warn(&h->pdev->dev, "out of memory\n");
2882                 return 1;
2883         }
2884         if (h->cciss_read == CCISS_READ_10) {
2885                 cciss_read_capacity(h, logvol,
2886                                         &total_size, &block_size);
2887         } else {
2888                 cciss_read_capacity_16(h, logvol,
2889                                         &total_size, &block_size);
2890         }
2891         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2892                                inq_buff, drv);
2893
2894         blk_queue_logical_block_size(drv->queue, drv->block_size);
2895         set_capacity(disk, drv->nr_blocks);
2896
2897         kfree(inq_buff);
2898         return 0;
2899 }
2900
2901 /*
2902  * Map (physical) PCI mem into (virtual) kernel space
2903  */
2904 static void __iomem *remap_pci_mem(ulong base, ulong size)
2905 {
2906         ulong page_base = ((ulong) base) & PAGE_MASK;
2907         ulong page_offs = ((ulong) base) - page_base;
2908         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2909
2910         return page_remapped ? (page_remapped + page_offs) : NULL;
2911 }
2912
2913 /*
2914  * Takes jobs of the Q and sends them to the hardware, then puts it on
2915  * the Q to wait for completion.
2916  */
2917 static void start_io(ctlr_info_t *h)
2918 {
2919         CommandList_struct *c;
2920
2921         while (!hlist_empty(&h->reqQ)) {
2922                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2923                 /* can't do anything if fifo is full */
2924                 if ((h->access.fifo_full(h))) {
2925                         dev_warn(&h->pdev->dev, "fifo full\n");
2926                         break;
2927                 }
2928
2929                 /* Get the first entry from the Request Q */
2930                 removeQ(c);
2931                 h->Qdepth--;
2932
2933                 /* Tell the controller execute command */
2934                 h->access.submit_command(h, c);
2935
2936                 /* Put job onto the completed Q */
2937                 addQ(&h->cmpQ, c);
2938         }
2939 }
2940
2941 /* Assumes that h->lock is held. */
2942 /* Zeros out the error record and then resends the command back */
2943 /* to the controller */
2944 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2945 {
2946         /* erase the old error information */
2947         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2948
2949         /* add it to software queue and then send it to the controller */
2950         addQ(&h->reqQ, c);
2951         h->Qdepth++;
2952         if (h->Qdepth > h->maxQsinceinit)
2953                 h->maxQsinceinit = h->Qdepth;
2954
2955         start_io(h);
2956 }
2957
2958 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2959         unsigned int msg_byte, unsigned int host_byte,
2960         unsigned int driver_byte)
2961 {
2962         /* inverse of macros in scsi.h */
2963         return (scsi_status_byte & 0xff) |
2964                 ((msg_byte & 0xff) << 8) |
2965                 ((host_byte & 0xff) << 16) |
2966                 ((driver_byte & 0xff) << 24);
2967 }
2968
2969 static inline int evaluate_target_status(ctlr_info_t *h,
2970                         CommandList_struct *cmd, int *retry_cmd)
2971 {
2972         unsigned char sense_key;
2973         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2974         int error_value;
2975
2976         *retry_cmd = 0;
2977         /* If we get in here, it means we got "target status", that is, scsi status */
2978         status_byte = cmd->err_info->ScsiStatus;
2979         driver_byte = DRIVER_OK;
2980         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2981
2982         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2983                 host_byte = DID_PASSTHROUGH;
2984         else
2985                 host_byte = DID_OK;
2986
2987         error_value = make_status_bytes(status_byte, msg_byte,
2988                 host_byte, driver_byte);
2989
2990         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2991                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
2992                         dev_warn(&h->pdev->dev, "cmd %p "
2993                                "has SCSI Status 0x%x\n",
2994                                cmd, cmd->err_info->ScsiStatus);
2995                 return error_value;
2996         }
2997
2998         /* check the sense key */
2999         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3000         /* no status or recovered error */
3001         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3002             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3003                 error_value = 0;
3004
3005         if (check_for_unit_attention(h, cmd)) {
3006                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3007                 return 0;
3008         }
3009
3010         /* Not SG_IO or similar? */
3011         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3012                 if (error_value != 0)
3013                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3014                                " sense key = 0x%x\n", cmd, sense_key);
3015                 return error_value;
3016         }
3017
3018         /* SG_IO or similar, copy sense data back */
3019         if (cmd->rq->sense) {
3020                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3021                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3022                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3023                         cmd->rq->sense_len);
3024         } else
3025                 cmd->rq->sense_len = 0;
3026
3027         return error_value;
3028 }
3029
3030 /* checks the status of the job and calls complete buffers to mark all
3031  * buffers for the completed job. Note that this function does not need
3032  * to hold the hba/queue lock.
3033  */
3034 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3035                                     int timeout)
3036 {
3037         int retry_cmd = 0;
3038         struct request *rq = cmd->rq;
3039
3040         rq->errors = 0;
3041
3042         if (timeout)
3043                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3044
3045         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3046                 goto after_error_processing;
3047
3048         switch (cmd->err_info->CommandStatus) {
3049         case CMD_TARGET_STATUS:
3050                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3051                 break;
3052         case CMD_DATA_UNDERRUN:
3053                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3054                         dev_warn(&h->pdev->dev, "cmd %p has"
3055                                " completed with data underrun "
3056                                "reported\n", cmd);
3057                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3058                 }
3059                 break;
3060         case CMD_DATA_OVERRUN:
3061                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3062                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3063                                " completed with data overrun "
3064                                "reported\n", cmd);
3065                 break;
3066         case CMD_INVALID:
3067                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3068                        "reported invalid\n", cmd);
3069                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3070                         cmd->err_info->CommandStatus, DRIVER_OK,
3071                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3072                                 DID_PASSTHROUGH : DID_ERROR);
3073                 break;
3074         case CMD_PROTOCOL_ERR:
3075                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3076                        "protocol error\n", cmd);
3077                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3078                         cmd->err_info->CommandStatus, DRIVER_OK,
3079                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3080                                 DID_PASSTHROUGH : DID_ERROR);
3081                 break;
3082         case CMD_HARDWARE_ERR:
3083                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3084                        " hardware error\n", cmd);
3085                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3086                         cmd->err_info->CommandStatus, DRIVER_OK,
3087                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3088                                 DID_PASSTHROUGH : DID_ERROR);
3089                 break;
3090         case CMD_CONNECTION_LOST:
3091                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3092                        "connection lost\n", cmd);
3093                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3094                         cmd->err_info->CommandStatus, DRIVER_OK,
3095                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3096                                 DID_PASSTHROUGH : DID_ERROR);
3097                 break;
3098         case CMD_ABORTED:
3099                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3100                        "aborted\n", cmd);
3101                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3102                         cmd->err_info->CommandStatus, DRIVER_OK,
3103                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3104                                 DID_PASSTHROUGH : DID_ABORT);
3105                 break;
3106         case CMD_ABORT_FAILED:
3107                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3108                        "abort failed\n", cmd);
3109                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3110                         cmd->err_info->CommandStatus, DRIVER_OK,
3111                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3112                                 DID_PASSTHROUGH : DID_ERROR);
3113                 break;
3114         case CMD_UNSOLICITED_ABORT:
3115                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3116                        "abort %p\n", h->ctlr, cmd);
3117                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3118                         retry_cmd = 1;
3119                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3120                         cmd->retry_count++;
3121                 } else
3122                         dev_warn(&h->pdev->dev,
3123                                 "%p retried too many times\n", cmd);
3124                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3125                         cmd->err_info->CommandStatus, DRIVER_OK,
3126                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3127                                 DID_PASSTHROUGH : DID_ABORT);
3128                 break;
3129         case CMD_TIMEOUT:
3130                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3131                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3132                         cmd->err_info->CommandStatus, DRIVER_OK,
3133                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3134                                 DID_PASSTHROUGH : DID_ERROR);
3135                 break;
3136         default:
3137                 dev_warn(&h->pdev->dev, "cmd %p returned "
3138                        "unknown status %x\n", cmd,
3139                        cmd->err_info->CommandStatus);
3140                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3141                         cmd->err_info->CommandStatus, DRIVER_OK,
3142                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3143                                 DID_PASSTHROUGH : DID_ERROR);
3144         }
3145
3146 after_error_processing:
3147
3148         /* We need to return this command */
3149         if (retry_cmd) {
3150                 resend_cciss_cmd(h, cmd);
3151                 return;
3152         }
3153         cmd->rq->completion_data = cmd;
3154         blk_complete_request(cmd->rq);
3155 }
3156
3157 static inline u32 cciss_tag_contains_index(u32 tag)
3158 {
3159 #define DIRECT_LOOKUP_BIT 0x10
3160         return tag & DIRECT_LOOKUP_BIT;
3161 }
3162
3163 static inline u32 cciss_tag_to_index(u32 tag)
3164 {
3165 #define DIRECT_LOOKUP_SHIFT 5
3166         return tag >> DIRECT_LOOKUP_SHIFT;
3167 }
3168
3169 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3170 {
3171 #define CCISS_ERROR_BITS 0x03
3172         return tag & ~CCISS_ERROR_BITS;
3173 }
3174
3175 static inline void cciss_mark_tag_indexed(u32 *tag)
3176 {
3177         *tag |= DIRECT_LOOKUP_BIT;
3178 }
3179
3180 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3181 {
3182         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3183 }
3184
3185 /*
3186  * Get a request and submit it to the controller.
3187  */
3188 static void do_cciss_request(struct request_queue *q)
3189 {
3190         ctlr_info_t *h = q->queuedata;
3191         CommandList_struct *c;
3192         sector_t start_blk;
3193         int seg;
3194         struct request *creq;
3195         u64bit temp64;
3196         struct scatterlist *tmp_sg;
3197         SGDescriptor_struct *curr_sg;
3198         drive_info_struct *drv;
3199         int i, dir;
3200         int sg_index = 0;
3201         int chained = 0;
3202
3203         /* We call start_io here in case there is a command waiting on the
3204          * queue that has not been sent.
3205          */
3206         if (blk_queue_plugged(q))
3207                 goto startio;
3208
3209       queue:
3210         creq = blk_peek_request(q);
3211         if (!creq)
3212                 goto startio;
3213
3214         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3215
3216         c = cmd_alloc(h);
3217         if (!c)
3218                 goto full;
3219
3220         blk_start_request(creq);
3221
3222         tmp_sg = h->scatter_list[c->cmdindex];
3223         spin_unlock_irq(q->queue_lock);
3224
3225         c->cmd_type = CMD_RWREQ;
3226         c->rq = creq;
3227
3228         /* fill in the request */
3229         drv = creq->rq_disk->private_data;
3230         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3231         /* got command from pool, so use the command block index instead */
3232         /* for direct lookups. */
3233         /* The first 2 bits are reserved for controller error reporting. */
3234         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3235         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3236         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3237         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3238         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3239         c->Request.Type.Attribute = ATTR_SIMPLE;
3240         c->Request.Type.Direction =
3241             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3242         c->Request.Timeout = 0; /* Don't time out */
3243         c->Request.CDB[0] =
3244             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3245         start_blk = blk_rq_pos(creq);
3246         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3247                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3248         sg_init_table(tmp_sg, h->maxsgentries);
3249         seg = blk_rq_map_sg(q, creq, tmp_sg);
3250
3251         /* get the DMA records for the setup */
3252         if (c->Request.Type.Direction == XFER_READ)
3253                 dir = PCI_DMA_FROMDEVICE;
3254         else
3255                 dir = PCI_DMA_TODEVICE;
3256
3257         curr_sg = c->SG;
3258         sg_index = 0;
3259         chained = 0;
3260
3261         for (i = 0; i < seg; i++) {
3262                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3263                         !chained && ((seg - i) > 1)) {
3264                         /* Point to next chain block. */
3265                         curr_sg = h->cmd_sg_list[c->cmdindex];
3266                         sg_index = 0;
3267                         chained = 1;
3268                 }
3269                 curr_sg[sg_index].Len = tmp_sg[i].length;
3270                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3271                                                 tmp_sg[i].offset,
3272                                                 tmp_sg[i].length, dir);
3273                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3274                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3275                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3276                 ++sg_index;
3277         }
3278         if (chained)
3279                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3280                         (seg - (h->max_cmd_sgentries - 1)) *
3281                                 sizeof(SGDescriptor_struct));
3282
3283         /* track how many SG entries we are using */
3284         if (seg > h->maxSG)
3285                 h->maxSG = seg;
3286
3287         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3288                         "chained[%d]\n",
3289                         blk_rq_sectors(creq), seg, chained);
3290
3291         c->Header.SGTotal = seg + chained;
3292         if (seg <= h->max_cmd_sgentries)
3293                 c->Header.SGList = c->Header.SGTotal;
3294         else
3295                 c->Header.SGList = h->max_cmd_sgentries;
3296         set_performant_mode(h, c);
3297
3298         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3299                 if(h->cciss_read == CCISS_READ_10) {
3300                         c->Request.CDB[1] = 0;
3301                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3302                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3303                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3304                         c->Request.CDB[5] = start_blk & 0xff;
3305                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3306                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3307                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3308                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3309                 } else {
3310                         u32 upper32 = upper_32_bits(start_blk);
3311
3312                         c->Request.CDBLen = 16;
3313                         c->Request.CDB[1]= 0;
3314                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3315                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3316                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3317                         c->Request.CDB[5]= upper32 & 0xff;
3318                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3319                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3320                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3321                         c->Request.CDB[9]= start_blk & 0xff;
3322                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3323                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3324                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3325                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3326                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3327                 }
3328         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3329                 c->Request.CDBLen = creq->cmd_len;
3330                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3331         } else {
3332                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3333                         creq->cmd_type);
3334                 BUG();
3335         }
3336
3337         spin_lock_irq(q->queue_lock);
3338
3339         addQ(&h->reqQ, c);
3340         h->Qdepth++;
3341         if (h->Qdepth > h->maxQsinceinit)
3342                 h->maxQsinceinit = h->Qdepth;
3343
3344         goto queue;
3345 full:
3346         blk_stop_queue(q);
3347 startio:
3348         /* We will already have the driver lock here so not need
3349          * to lock it.
3350          */
3351         start_io(h);
3352 }
3353
3354 static inline unsigned long get_next_completion(ctlr_info_t *h)
3355 {
3356         return h->access.command_completed(h);
3357 }
3358
3359 static inline int interrupt_pending(ctlr_info_t *h)
3360 {
3361         return h->access.intr_pending(h);
3362 }
3363
3364 static inline long interrupt_not_for_us(ctlr_info_t *h)
3365 {
3366         return ((h->access.intr_pending(h) == 0) ||
3367                 (h->interrupts_enabled == 0));
3368 }
3369
3370 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3371                         u32 raw_tag)
3372 {
3373         if (unlikely(tag_index >= h->nr_cmds)) {
3374                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3375                 return 1;
3376         }
3377         return 0;
3378 }
3379
3380 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3381                                 u32 raw_tag)
3382 {
3383         removeQ(c);
3384         if (likely(c->cmd_type == CMD_RWREQ))
3385                 complete_command(h, c, 0);
3386         else if (c->cmd_type == CMD_IOCTL_PEND)
3387                 complete(c->waiting);
3388 #ifdef CONFIG_CISS_SCSI_TAPE
3389         else if (c->cmd_type == CMD_SCSI)
3390                 complete_scsi_command(c, 0, raw_tag);
3391 #endif
3392 }
3393
3394 static inline u32 next_command(ctlr_info_t *h)
3395 {
3396         u32 a;
3397
3398         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3399                 return h->access.command_completed(h);
3400
3401         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3402                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3403                 (h->reply_pool_head)++;
3404                 h->commands_outstanding--;
3405         } else {
3406                 a = FIFO_EMPTY;
3407         }
3408         /* Check for wraparound */
3409         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3410                 h->reply_pool_head = h->reply_pool;
3411                 h->reply_pool_wraparound ^= 1;
3412         }
3413         return a;
3414 }
3415
3416 /* process completion of an indexed ("direct lookup") command */
3417 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3418 {
3419         u32 tag_index;
3420         CommandList_struct *c;
3421
3422         tag_index = cciss_tag_to_index(raw_tag);
3423         if (bad_tag(h, tag_index, raw_tag))
3424                 return next_command(h);
3425         c = h->cmd_pool + tag_index;
3426         finish_cmd(h, c, raw_tag);
3427         return next_command(h);
3428 }
3429
3430 /* process completion of a non-indexed command */
3431 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3432 {
3433         u32 tag;
3434         CommandList_struct *c = NULL;
3435         struct hlist_node *tmp;
3436         __u32 busaddr_masked, tag_masked;
3437
3438         tag = cciss_tag_discard_error_bits(raw_tag);
3439         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3440                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3441                 tag_masked = cciss_tag_discard_error_bits(tag);
3442                 if (busaddr_masked == tag_masked) {
3443                         finish_cmd(h, c, raw_tag);
3444                         return next_command(h);
3445                 }
3446         }
3447         bad_tag(h, h->nr_cmds + 1, raw_tag);
3448         return next_command(h);
3449 }
3450
3451 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3452 {
3453         ctlr_info_t *h = dev_id;
3454         unsigned long flags;
3455         u32 raw_tag;
3456
3457         if (interrupt_not_for_us(h))
3458                 return IRQ_NONE;
3459         spin_lock_irqsave(&h->lock, flags);
3460         while (interrupt_pending(h)) {
3461                 raw_tag = get_next_completion(h);
3462                 while (raw_tag != FIFO_EMPTY) {
3463                         if (cciss_tag_contains_index(raw_tag))
3464                                 raw_tag = process_indexed_cmd(h, raw_tag);
3465                         else
3466                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3467                 }
3468         }
3469         spin_unlock_irqrestore(&h->lock, flags);
3470         return IRQ_HANDLED;
3471 }
3472
3473 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3474  * check the interrupt pending register because it is not set.
3475  */
3476 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3477 {
3478         ctlr_info_t *h = dev_id;
3479         unsigned long flags;
3480         u32 raw_tag;
3481
3482         spin_lock_irqsave(&h->lock, flags);
3483         raw_tag = get_next_completion(h);
3484         while (raw_tag != FIFO_EMPTY) {
3485                 if (cciss_tag_contains_index(raw_tag))
3486                         raw_tag = process_indexed_cmd(h, raw_tag);
3487                 else
3488                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3489         }
3490         spin_unlock_irqrestore(&h->lock, flags);
3491         return IRQ_HANDLED;
3492 }
3493
3494 /**
3495  * add_to_scan_list() - add controller to rescan queue
3496  * @h:                Pointer to the controller.
3497  *
3498  * Adds the controller to the rescan queue if not already on the queue.
3499  *
3500  * returns 1 if added to the queue, 0 if skipped (could be on the
3501  * queue already, or the controller could be initializing or shutting
3502  * down).
3503  **/
3504 static int add_to_scan_list(struct ctlr_info *h)
3505 {
3506         struct ctlr_info *test_h;
3507         int found = 0;
3508         int ret = 0;
3509
3510         if (h->busy_initializing)
3511                 return 0;
3512
3513         if (!mutex_trylock(&h->busy_shutting_down))
3514                 return 0;
3515
3516         mutex_lock(&scan_mutex);
3517         list_for_each_entry(test_h, &scan_q, scan_list) {
3518                 if (test_h == h) {
3519                         found = 1;
3520                         break;
3521                 }
3522         }
3523         if (!found && !h->busy_scanning) {
3524                 INIT_COMPLETION(h->scan_wait);
3525                 list_add_tail(&h->scan_list, &scan_q);
3526                 ret = 1;
3527         }
3528         mutex_unlock(&scan_mutex);
3529         mutex_unlock(&h->busy_shutting_down);
3530
3531         return ret;
3532 }
3533
3534 /**
3535  * remove_from_scan_list() - remove controller from rescan queue
3536  * @h:                     Pointer to the controller.
3537  *
3538  * Removes the controller from the rescan queue if present. Blocks if
3539  * the controller is currently conducting a rescan.  The controller
3540  * can be in one of three states:
3541  * 1. Doesn't need a scan
3542  * 2. On the scan list, but not scanning yet (we remove it)
3543  * 3. Busy scanning (and not on the list). In this case we want to wait for
3544  *    the scan to complete to make sure the scanning thread for this
3545  *    controller is completely idle.
3546  **/
3547 static void remove_from_scan_list(struct ctlr_info *h)
3548 {
3549         struct ctlr_info *test_h, *tmp_h;
3550
3551         mutex_lock(&scan_mutex);
3552         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3553                 if (test_h == h) { /* state 2. */
3554                         list_del(&h->scan_list);
3555                         complete_all(&h->scan_wait);
3556                         mutex_unlock(&scan_mutex);
3557                         return;
3558                 }
3559         }
3560         if (h->busy_scanning) { /* state 3. */
3561                 mutex_unlock(&scan_mutex);
3562                 wait_for_completion(&h->scan_wait);
3563         } else { /* state 1, nothing to do. */
3564                 mutex_unlock(&scan_mutex);
3565         }
3566 }
3567
3568 /**
3569  * scan_thread() - kernel thread used to rescan controllers
3570  * @data:        Ignored.
3571  *
3572  * A kernel thread used scan for drive topology changes on
3573  * controllers. The thread processes only one controller at a time
3574  * using a queue.  Controllers are added to the queue using
3575  * add_to_scan_list() and removed from the queue either after done
3576  * processing or using remove_from_scan_list().
3577  *
3578  * returns 0.
3579  **/
3580 static int scan_thread(void *data)
3581 {
3582         struct ctlr_info *h;
3583
3584         while (1) {
3585                 set_current_state(TASK_INTERRUPTIBLE);
3586                 schedule();
3587                 if (kthread_should_stop())
3588                         break;
3589
3590                 while (1) {
3591                         mutex_lock(&scan_mutex);
3592                         if (list_empty(&scan_q)) {
3593                                 mutex_unlock(&scan_mutex);
3594                                 break;
3595                         }
3596
3597                         h = list_entry(scan_q.next,
3598                                        struct ctlr_info,
3599                                        scan_list);
3600                         list_del(&h->scan_list);
3601                         h->busy_scanning = 1;
3602                         mutex_unlock(&scan_mutex);
3603
3604                         rebuild_lun_table(h, 0, 0);
3605                         complete_all(&h->scan_wait);
3606                         mutex_lock(&scan_mutex);
3607                         h->busy_scanning = 0;
3608                         mutex_unlock(&scan_mutex);
3609                 }
3610         }
3611
3612         return 0;
3613 }
3614
3615 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3616 {
3617         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3618                 return 0;
3619
3620         switch (c->err_info->SenseInfo[12]) {
3621         case STATE_CHANGED:
3622                 dev_warn(&h->pdev->dev, "a state change "
3623                         "detected, command retried\n");
3624                 return 1;
3625         break;
3626         case LUN_FAILED:
3627                 dev_warn(&h->pdev->dev, "LUN failure "
3628                         "detected, action required\n");
3629                 return 1;
3630         break;
3631         case REPORT_LUNS_CHANGED:
3632                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3633         /*
3634          * Here, we could call add_to_scan_list and wake up the scan thread,
3635          * except that it's quite likely that we will get more than one
3636          * REPORT_LUNS_CHANGED condition in quick succession, which means
3637          * that those which occur after the first one will likely happen
3638          * *during* the scan_thread's rescan.  And the rescan code is not
3639          * robust enough to restart in the middle, undoing what it has already
3640          * done, and it's not clear that it's even possible to do this, since
3641          * part of what it does is notify the block layer, which starts
3642          * doing it's own i/o to read partition tables and so on, and the
3643          * driver doesn't have visibility to know what might need undoing.
3644          * In any event, if possible, it is horribly complicated to get right
3645          * so we just don't do it for now.
3646          *
3647          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3648          */
3649                 return 1;
3650         break;
3651         case POWER_OR_RESET:
3652                 dev_warn(&h->pdev->dev,
3653                         "a power on or device reset detected\n");
3654                 return 1;
3655         break;
3656         case UNIT_ATTENTION_CLEARED:
3657                 dev_warn(&h->pdev->dev,
3658                         "unit attention cleared by another initiator\n");
3659                 return 1;
3660         break;
3661         default:
3662                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3663                 return 1;
3664         }
3665 }
3666
3667 /*
3668  *  We cannot read the structure directly, for portability we must use
3669  *   the io functions.
3670  *   This is for debug only.
3671  */
3672 static void print_cfg_table(ctlr_info_t *h)
3673 {
3674         int i;
3675         char temp_name[17];
3676         CfgTable_struct *tb = h->cfgtable;
3677
3678         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3679         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3680         for (i = 0; i < 4; i++)
3681                 temp_name[i] = readb(&(tb->Signature[i]));
3682         temp_name[4] = '\0';
3683         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3684         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3685                 readl(&(tb->SpecValence)));
3686         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3687                readl(&(tb->TransportSupport)));
3688         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3689                readl(&(tb->TransportActive)));
3690         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3691                readl(&(tb->HostWrite.TransportRequest)));
3692         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3693                readl(&(tb->HostWrite.CoalIntDelay)));
3694         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3695                readl(&(tb->HostWrite.CoalIntCount)));
3696         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3697                readl(&(tb->CmdsOutMax)));
3698         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3699                 readl(&(tb->BusTypes)));
3700         for (i = 0; i < 16; i++)
3701                 temp_name[i] = readb(&(tb->ServerName[i]));
3702         temp_name[16] = '\0';
3703         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3704         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3705                 readl(&(tb->HeartBeat)));
3706 }
3707
3708 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3709 {
3710         int i, offset, mem_type, bar_type;
3711         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3712                 return 0;
3713         offset = 0;
3714         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3715                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3716                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3717                         offset += 4;
3718                 else {
3719                         mem_type = pci_resource_flags(pdev, i) &
3720                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3721                         switch (mem_type) {
3722                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3723                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3724                                 offset += 4;    /* 32 bit */
3725                                 break;
3726                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3727                                 offset += 8;
3728                                 break;
3729                         default:        /* reserved in PCI 2.2 */
3730                                 dev_warn(&pdev->dev,
3731                                        "Base address is invalid\n");
3732                                 return -1;
3733                                 break;
3734                         }
3735                 }
3736                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3737                         return i + 1;
3738         }
3739         return -1;
3740 }
3741
3742 /* Fill in bucket_map[], given nsgs (the max number of
3743  * scatter gather elements supported) and bucket[],
3744  * which is an array of 8 integers.  The bucket[] array
3745  * contains 8 different DMA transfer sizes (in 16
3746  * byte increments) which the controller uses to fetch
3747  * commands.  This function fills in bucket_map[], which
3748  * maps a given number of scatter gather elements to one of
3749  * the 8 DMA transfer sizes.  The point of it is to allow the
3750  * controller to only do as much DMA as needed to fetch the
3751  * command, with the DMA transfer size encoded in the lower
3752  * bits of the command address.
3753  */
3754 static void  calc_bucket_map(int bucket[], int num_buckets,
3755         int nsgs, int *bucket_map)
3756 {
3757         int i, j, b, size;
3758
3759         /* even a command with 0 SGs requires 4 blocks */
3760 #define MINIMUM_TRANSFER_BLOCKS 4
3761 #define NUM_BUCKETS 8
3762         /* Note, bucket_map must have nsgs+1 entries. */
3763         for (i = 0; i <= nsgs; i++) {
3764                 /* Compute size of a command with i SG entries */
3765                 size = i + MINIMUM_TRANSFER_BLOCKS;
3766                 b = num_buckets; /* Assume the biggest bucket */
3767                 /* Find the bucket that is just big enough */
3768                 for (j = 0; j < 8; j++) {
3769                         if (bucket[j] >= size) {
3770                                 b = j;
3771                                 break;
3772                         }
3773                 }
3774                 /* for a command with i SG entries, use bucket b. */
3775                 bucket_map[i] = b;
3776         }
3777 }
3778
3779 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3780 {
3781         int i;
3782
3783         /* under certain very rare conditions, this can take awhile.
3784          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3785          * as we enter this code.) */
3786         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3787                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3788                         break;
3789                 msleep(10);
3790         }
3791 }
3792
3793 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3794 {
3795         /* This is a bit complicated.  There are 8 registers on
3796          * the controller which we write to to tell it 8 different
3797          * sizes of commands which there may be.  It's a way of
3798          * reducing the DMA done to fetch each command.  Encoded into
3799          * each command's tag are 3 bits which communicate to the controller
3800          * which of the eight sizes that command fits within.  The size of
3801          * each command depends on how many scatter gather entries there are.
3802          * Each SG entry requires 16 bytes.  The eight registers are programmed
3803          * with the number of 16-byte blocks a command of that size requires.
3804          * The smallest command possible requires 5 such 16 byte blocks.
3805          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3806          * blocks.  Note, this only extends to the SG entries contained
3807          * within the command block, and does not extend to chained blocks
3808          * of SG elements.   bft[] contains the eight values we write to
3809          * the registers.  They are not evenly distributed, but have more
3810          * sizes for small commands, and fewer sizes for larger commands.
3811          */
3812         __u32 trans_offset;
3813         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3814                         /*
3815                          *  5 = 1 s/g entry or 4k
3816                          *  6 = 2 s/g entry or 8k
3817                          *  8 = 4 s/g entry or 16k
3818                          * 10 = 6 s/g entry or 24k
3819                          */
3820         unsigned long register_value;
3821         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3822
3823         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3824
3825         /* Controller spec: zero out this buffer. */
3826         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3827         h->reply_pool_head = h->reply_pool;
3828
3829         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3830         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3831                                 h->blockFetchTable);
3832         writel(bft[0], &h->transtable->BlockFetch0);
3833         writel(bft[1], &h->transtable->BlockFetch1);
3834         writel(bft[2], &h->transtable->BlockFetch2);
3835         writel(bft[3], &h->transtable->BlockFetch3);
3836         writel(bft[4], &h->transtable->BlockFetch4);
3837         writel(bft[5], &h->transtable->BlockFetch5);
3838         writel(bft[6], &h->transtable->BlockFetch6);
3839         writel(bft[7], &h->transtable->BlockFetch7);
3840
3841         /* size of controller ring buffer */
3842         writel(h->max_commands, &h->transtable->RepQSize);
3843         writel(1, &h->transtable->RepQCount);
3844         writel(0, &h->transtable->RepQCtrAddrLow32);
3845         writel(0, &h->transtable->RepQCtrAddrHigh32);
3846         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3847         writel(0, &h->transtable->RepQAddr0High32);
3848         writel(CFGTBL_Trans_Performant,
3849                         &(h->cfgtable->HostWrite.TransportRequest));
3850
3851         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3852         cciss_wait_for_mode_change_ack(h);
3853         register_value = readl(&(h->cfgtable->TransportActive));
3854         if (!(register_value & CFGTBL_Trans_Performant))
3855                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3856                                         " performant mode\n");
3857 }
3858
3859 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3860 {
3861         __u32 trans_support;
3862
3863         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3864         /* Attempt to put controller into performant mode if supported */
3865         /* Does board support performant mode? */
3866         trans_support = readl(&(h->cfgtable->TransportSupport));
3867         if (!(trans_support & PERFORMANT_MODE))
3868                 return;
3869
3870         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3871         /* Performant mode demands commands on a 32 byte boundary
3872          * pci_alloc_consistent aligns on page boundarys already.
3873          * Just need to check if divisible by 32
3874          */
3875         if ((sizeof(CommandList_struct) % 32) != 0) {
3876                 dev_warn(&h->pdev->dev, "%s %d %s\n",
3877                         "cciss info: command size[",
3878                         (int)sizeof(CommandList_struct),
3879                         "] not divisible by 32, no performant mode..\n");
3880                 return;
3881         }
3882
3883         /* Performant mode ring buffer and supporting data structures */
3884         h->reply_pool = (__u64 *)pci_alloc_consistent(
3885                 h->pdev, h->max_commands * sizeof(__u64),
3886                 &(h->reply_pool_dhandle));
3887
3888         /* Need a block fetch table for performant mode */
3889         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3890                 sizeof(__u32)), GFP_KERNEL);
3891
3892         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3893                 goto clean_up;
3894
3895         cciss_enter_performant_mode(h);
3896
3897         /* Change the access methods to the performant access methods */
3898         h->access = SA5_performant_access;
3899         h->transMethod = CFGTBL_Trans_Performant;
3900
3901         return;
3902 clean_up:
3903         kfree(h->blockFetchTable);
3904         if (h->reply_pool)
3905                 pci_free_consistent(h->pdev,
3906                                 h->max_commands * sizeof(__u64),
3907                                 h->reply_pool,
3908                                 h->reply_pool_dhandle);
3909         return;
3910
3911 } /* cciss_put_controller_into_performant_mode */
3912
3913 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3914  * controllers that are capable. If not, we use IO-APIC mode.
3915  */
3916
3917 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3918 {
3919 #ifdef CONFIG_PCI_MSI
3920         int err;
3921         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3922         {0, 2}, {0, 3}
3923         };
3924
3925         /* Some boards advertise MSI but don't really support it */
3926         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3927             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3928                 goto default_int_mode;
3929
3930         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3931                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3932                 if (!err) {
3933                         h->intr[0] = cciss_msix_entries[0].vector;
3934                         h->intr[1] = cciss_msix_entries[1].vector;
3935                         h->intr[2] = cciss_msix_entries[2].vector;
3936                         h->intr[3] = cciss_msix_entries[3].vector;
3937                         h->msix_vector = 1;
3938                         return;
3939                 }
3940                 if (err > 0) {
3941                         dev_warn(&h->pdev->dev,
3942                                 "only %d MSI-X vectors available\n", err);
3943                         goto default_int_mode;
3944                 } else {
3945                         dev_warn(&h->pdev->dev,
3946                                 "MSI-X init failed %d\n", err);
3947                         goto default_int_mode;
3948                 }
3949         }
3950         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3951                 if (!pci_enable_msi(h->pdev))
3952                         h->msi_vector = 1;
3953                 else
3954                         dev_warn(&h->pdev->dev, "MSI init failed\n");
3955         }
3956 default_int_mode:
3957 #endif                          /* CONFIG_PCI_MSI */
3958         /* if we get here we're going to use the default interrupt mode */
3959         h->intr[PERF_MODE_INT] = h->pdev->irq;
3960         return;
3961 }
3962
3963 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3964 {
3965         int i;
3966         u32 subsystem_vendor_id, subsystem_device_id;
3967
3968         subsystem_vendor_id = pdev->subsystem_vendor;
3969         subsystem_device_id = pdev->subsystem_device;
3970         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3971                         subsystem_vendor_id;
3972
3973         for (i = 0; i < ARRAY_SIZE(products); i++) {
3974                 /* Stand aside for hpsa driver on request */
3975                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3976                         return -ENODEV;
3977                 if (*board_id == products[i].board_id)
3978                         return i;
3979         }
3980         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
3981                 *board_id);
3982         return -ENODEV;
3983 }
3984
3985 static inline bool cciss_board_disabled(ctlr_info_t *h)
3986 {
3987         u16 command;
3988
3989         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
3990         return ((command & PCI_COMMAND_MEMORY) == 0);
3991 }
3992
3993 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
3994         unsigned long *memory_bar)
3995 {
3996         int i;
3997
3998         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
3999                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4000                         /* addressing mode bits already removed */
4001                         *memory_bar = pci_resource_start(pdev, i);
4002                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4003                                 *memory_bar);
4004                         return 0;
4005                 }
4006         dev_warn(&pdev->dev, "no memory BAR found\n");
4007         return -ENODEV;
4008 }
4009
4010 static int __devinit cciss_wait_for_board_ready(ctlr_info_t *h)
4011 {
4012         int i;
4013         u32 scratchpad;
4014
4015         for (i = 0; i < CCISS_BOARD_READY_ITERATIONS; i++) {
4016                 scratchpad = readl(h->vaddr + SA5_SCRATCHPAD_OFFSET);
4017                 if (scratchpad == CCISS_FIRMWARE_READY)
4018                         return 0;
4019                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4020         }
4021         dev_warn(&h->pdev->dev, "board not ready, timed out.\n");
4022         return -ENODEV;
4023 }
4024
4025 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4026         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4027         u64 *cfg_offset)
4028 {
4029         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4030         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4031         *cfg_base_addr &= (u32) 0x0000ffff;
4032         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4033         if (*cfg_base_addr_index == -1) {
4034                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4035                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4036                 return -ENODEV;
4037         }
4038         return 0;
4039 }
4040
4041 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4042 {
4043         u64 cfg_offset;
4044         u32 cfg_base_addr;
4045         u64 cfg_base_addr_index;
4046         u32 trans_offset;
4047         int rc;
4048
4049         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4050                 &cfg_base_addr_index, &cfg_offset);
4051         if (rc)
4052                 return rc;
4053         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4054                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4055         if (!h->cfgtable)
4056                 return -ENOMEM;
4057         /* Find performant mode table. */
4058         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4059         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4060                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4061                                 sizeof(*h->transtable));
4062         if (!h->transtable)
4063                 return -ENOMEM;
4064         return 0;
4065 }
4066
4067 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4068 {
4069         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4070         if (h->max_commands < 16) {
4071                 dev_warn(&h->pdev->dev, "Controller reports "
4072                         "max supported commands of %d, an obvious lie. "
4073                         "Using 16.  Ensure that firmware is up to date.\n",
4074                         h->max_commands);
4075                 h->max_commands = 16;
4076         }
4077 }
4078
4079 /* Interrogate the hardware for some limits:
4080  * max commands, max SG elements without chaining, and with chaining,
4081  * SG chain block size, etc.
4082  */
4083 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4084 {
4085         cciss_get_max_perf_mode_cmds(h);
4086         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4087         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4088         /*
4089          * Limit in-command s/g elements to 32 save dma'able memory.
4090          * Howvever spec says if 0, use 31
4091          */
4092         h->max_cmd_sgentries = 31;
4093         if (h->maxsgentries > 512) {
4094                 h->max_cmd_sgentries = 32;
4095                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4096                 h->maxsgentries--; /* save one for chain pointer */
4097         } else {
4098                 h->maxsgentries = 31; /* default to traditional values */
4099                 h->chainsize = 0;
4100         }
4101 }
4102
4103 static inline bool CISS_signature_present(ctlr_info_t *h)
4104 {
4105         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4106             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4107             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4108             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4109                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4110                 return false;
4111         }
4112         return true;
4113 }
4114
4115 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4116 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4117 {
4118 #ifdef CONFIG_X86
4119         u32 prefetch;
4120
4121         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4122         prefetch |= 0x100;
4123         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4124 #endif
4125 }
4126
4127 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4128  * in a prefetch beyond physical memory.
4129  */
4130 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4131 {
4132         u32 dma_prefetch;
4133         __u32 dma_refetch;
4134
4135         if (h->board_id != 0x3225103C)
4136                 return;
4137         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4138         dma_prefetch |= 0x8000;
4139         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4140         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4141         dma_refetch |= 0x1;
4142         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4143 }
4144
4145 static int __devinit cciss_pci_init(ctlr_info_t *h)
4146 {
4147         int prod_index, err;
4148
4149         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4150         if (prod_index < 0)
4151                 return -ENODEV;
4152         h->product_name = products[prod_index].product_name;
4153         h->access = *(products[prod_index].access);
4154
4155         if (cciss_board_disabled(h)) {
4156                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4157                 return -ENODEV;
4158         }
4159         err = pci_enable_device(h->pdev);
4160         if (err) {
4161                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4162                 return err;
4163         }
4164
4165         err = pci_request_regions(h->pdev, "cciss");
4166         if (err) {
4167                 dev_warn(&h->pdev->dev,
4168                         "Cannot obtain PCI resources, aborting\n");
4169                 return err;
4170         }
4171
4172         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4173         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4174
4175 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4176  * else we use the IO-APIC interrupt assigned to us by system ROM.
4177  */
4178         cciss_interrupt_mode(h);
4179         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4180         if (err)
4181                 goto err_out_free_res;
4182         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4183         if (!h->vaddr) {
4184                 err = -ENOMEM;
4185                 goto err_out_free_res;
4186         }
4187         err = cciss_wait_for_board_ready(h);
4188         if (err)
4189                 goto err_out_free_res;
4190         err = cciss_find_cfgtables(h);
4191         if (err)
4192                 goto err_out_free_res;
4193         print_cfg_table(h);
4194         cciss_find_board_params(h);
4195
4196         if (!CISS_signature_present(h)) {
4197                 err = -ENODEV;
4198                 goto err_out_free_res;
4199         }
4200         cciss_enable_scsi_prefetch(h);
4201         cciss_p600_dma_prefetch_quirk(h);
4202         cciss_put_controller_into_performant_mode(h);
4203         return 0;
4204
4205 err_out_free_res:
4206         /*
4207          * Deliberately omit pci_disable_device(): it does something nasty to
4208          * Smart Array controllers that pci_enable_device does not undo
4209          */
4210         if (h->transtable)
4211                 iounmap(h->transtable);
4212         if (h->cfgtable)
4213                 iounmap(h->cfgtable);
4214         if (h->vaddr)
4215                 iounmap(h->vaddr);
4216         pci_release_regions(h->pdev);
4217         return err;
4218 }
4219
4220 /* Function to find the first free pointer into our hba[] array
4221  * Returns -1 if no free entries are left.
4222  */
4223 static int alloc_cciss_hba(struct pci_dev *pdev)
4224 {
4225         int i;
4226
4227         for (i = 0; i < MAX_CTLR; i++) {
4228                 if (!hba[i]) {
4229                         ctlr_info_t *h;
4230
4231                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4232                         if (!h)
4233                                 goto Enomem;
4234                         hba[i] = h;
4235                         return i;
4236                 }
4237         }
4238         dev_warn(&pdev->dev, "This driver supports a maximum"
4239                " of %d controllers.\n", MAX_CTLR);
4240         return -1;
4241 Enomem:
4242         dev_warn(&pdev->dev, "out of memory.\n");
4243         return -1;
4244 }
4245
4246 static void free_hba(ctlr_info_t *h)
4247 {
4248         int i;
4249
4250         hba[h->ctlr] = NULL;
4251         for (i = 0; i < h->highest_lun + 1; i++)
4252                 if (h->gendisk[i] != NULL)
4253                         put_disk(h->gendisk[i]);
4254         kfree(h);
4255 }
4256
4257 /* Send a message CDB to the firmware. */
4258 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4259 {
4260         typedef struct {
4261                 CommandListHeader_struct CommandHeader;
4262                 RequestBlock_struct Request;
4263                 ErrDescriptor_struct ErrorDescriptor;
4264         } Command;
4265         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4266         Command *cmd;
4267         dma_addr_t paddr64;
4268         uint32_t paddr32, tag;
4269         void __iomem *vaddr;
4270         int i, err;
4271
4272         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4273         if (vaddr == NULL)
4274                 return -ENOMEM;
4275
4276         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4277            CCISS commands, so they must be allocated from the lower 4GiB of
4278            memory. */
4279         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4280         if (err) {
4281                 iounmap(vaddr);
4282                 return -ENOMEM;
4283         }
4284
4285         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4286         if (cmd == NULL) {
4287                 iounmap(vaddr);
4288                 return -ENOMEM;
4289         }
4290
4291         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4292            although there's no guarantee, we assume that the address is at
4293            least 4-byte aligned (most likely, it's page-aligned). */
4294         paddr32 = paddr64;
4295
4296         cmd->CommandHeader.ReplyQueue = 0;
4297         cmd->CommandHeader.SGList = 0;
4298         cmd->CommandHeader.SGTotal = 0;
4299         cmd->CommandHeader.Tag.lower = paddr32;
4300         cmd->CommandHeader.Tag.upper = 0;
4301         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4302
4303         cmd->Request.CDBLen = 16;
4304         cmd->Request.Type.Type = TYPE_MSG;
4305         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4306         cmd->Request.Type.Direction = XFER_NONE;
4307         cmd->Request.Timeout = 0; /* Don't time out */
4308         cmd->Request.CDB[0] = opcode;
4309         cmd->Request.CDB[1] = type;
4310         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4311
4312         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4313         cmd->ErrorDescriptor.Addr.upper = 0;
4314         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4315
4316         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4317
4318         for (i = 0; i < 10; i++) {
4319                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4320                 if ((tag & ~3) == paddr32)
4321                         break;
4322                 schedule_timeout_uninterruptible(HZ);
4323         }
4324
4325         iounmap(vaddr);
4326
4327         /* we leak the DMA buffer here ... no choice since the controller could
4328            still complete the command. */
4329         if (i == 10) {
4330                 dev_err(&pdev->dev,
4331                         "controller message %02x:%02x timed out\n",
4332                         opcode, type);
4333                 return -ETIMEDOUT;
4334         }
4335
4336         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4337
4338         if (tag & 2) {
4339                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4340                         opcode, type);
4341                 return -EIO;
4342         }
4343
4344         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4345                 opcode, type);
4346         return 0;
4347 }
4348
4349 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4350 #define cciss_noop(p) cciss_message(p, 3, 0)
4351
4352 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4353 {
4354 /* the #defines are stolen from drivers/pci/msi.h. */
4355 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4356 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4357
4358         int pos;
4359         u16 control = 0;
4360
4361         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4362         if (pos) {
4363                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4364                 if (control & PCI_MSI_FLAGS_ENABLE) {
4365                         dev_info(&pdev->dev, "resetting MSI\n");
4366                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4367                 }
4368         }
4369
4370         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4371         if (pos) {
4372                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4373                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4374                         dev_info(&pdev->dev, "resetting MSI-X\n");
4375                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4376                 }
4377         }
4378
4379         return 0;
4380 }
4381
4382 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4383         void * __iomem vaddr, bool use_doorbell)
4384 {
4385         u16 pmcsr;
4386         int pos;
4387
4388         if (use_doorbell) {
4389                 /* For everything after the P600, the PCI power state method
4390                  * of resetting the controller doesn't work, so we have this
4391                  * other way using the doorbell register.
4392                  */
4393                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4394                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4395                 msleep(1000);
4396         } else { /* Try to do it the PCI power state way */
4397
4398                 /* Quoting from the Open CISS Specification: "The Power
4399                  * Management Control/Status Register (CSR) controls the power
4400                  * state of the device.  The normal operating state is D0,
4401                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4402                  * the controller, place the interface device in D3 then to D0,
4403                  * this causes a secondary PCI reset which will reset the
4404                  * controller." */
4405
4406                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4407                 if (pos == 0) {
4408                         dev_err(&pdev->dev,
4409                                 "cciss_controller_hard_reset: "
4410                                 "PCI PM not supported\n");
4411                         return -ENODEV;
4412                 }
4413                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4414                 /* enter the D3hot power management state */
4415                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4416                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4417                 pmcsr |= PCI_D3hot;
4418                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4419
4420                 msleep(500);
4421
4422                 /* enter the D0 power management state */
4423                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4424                 pmcsr |= PCI_D0;
4425                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4426
4427                 msleep(500);
4428         }
4429         return 0;
4430 }
4431
4432 /* This does a hard reset of the controller using PCI power management
4433  * states or using the doorbell register. */
4434 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4435 {
4436         u16 saved_config_space[32];
4437         u64 cfg_offset;
4438         u32 cfg_base_addr;
4439         u64 cfg_base_addr_index;
4440         void __iomem *vaddr;
4441         unsigned long paddr;
4442         u32 misc_fw_support, active_transport;
4443         int rc, i;
4444         CfgTable_struct __iomem *cfgtable;
4445         bool use_doorbell;
4446         u32 board_id;
4447
4448         /* For controllers as old a the p600, this is very nearly
4449          * the same thing as
4450          *
4451          * pci_save_state(pci_dev);
4452          * pci_set_power_state(pci_dev, PCI_D3hot);
4453          * pci_set_power_state(pci_dev, PCI_D0);
4454          * pci_restore_state(pci_dev);
4455          *
4456          * but we can't use these nice canned kernel routines on
4457          * kexec, because they also check the MSI/MSI-X state in PCI
4458          * configuration space and do the wrong thing when it is
4459          * set/cleared.  Also, the pci_save/restore_state functions
4460          * violate the ordering requirements for restoring the
4461          * configuration space from the CCISS document (see the
4462          * comment below).  So we roll our own ....
4463          *
4464          * For controllers newer than the P600, the pci power state
4465          * method of resetting doesn't work so we have another way
4466          * using the doorbell register.
4467          */
4468
4469         /* Exclude 640x boards.  These are two pci devices in one slot
4470          * which share a battery backed cache module.  One controls the
4471          * cache, the other accesses the cache through the one that controls
4472          * it.  If we reset the one controlling the cache, the other will
4473          * likely not be happy.  Just forbid resetting this conjoined mess.
4474          */
4475         cciss_lookup_board_id(pdev, &board_id);
4476         if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4477                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4478                                 "due to shared cache module.");
4479                 return -ENODEV;
4480         }
4481
4482         for (i = 0; i < 32; i++)
4483                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4484
4485         /* find the first memory BAR, so we can find the cfg table */
4486         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4487         if (rc)
4488                 return rc;
4489         vaddr = remap_pci_mem(paddr, 0x250);
4490         if (!vaddr)
4491                 return -ENOMEM;
4492
4493         /* find cfgtable in order to check if reset via doorbell is supported */
4494         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4495                                         &cfg_base_addr_index, &cfg_offset);
4496         if (rc)
4497                 goto unmap_vaddr;
4498         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4499                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4500         if (!cfgtable) {
4501                 rc = -ENOMEM;
4502                 goto unmap_vaddr;
4503         }
4504
4505         /* If reset via doorbell register is supported, use that. */
4506         misc_fw_support = readl(&cfgtable->misc_fw_support);
4507         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4508
4509         /* The doorbell reset seems to cause lockups on some Smart
4510          * Arrays (e.g. P410, P410i, maybe others).  Until this is
4511          * fixed or at least isolated, avoid the doorbell reset.
4512          */
4513         use_doorbell = 0;
4514
4515         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4516         if (rc)
4517                 goto unmap_cfgtable;
4518
4519         /* Restore the PCI configuration space.  The Open CISS
4520          * Specification says, "Restore the PCI Configuration
4521          * Registers, offsets 00h through 60h. It is important to
4522          * restore the command register, 16-bits at offset 04h,
4523          * last. Do not restore the configuration status register,
4524          * 16-bits at offset 06h."  Note that the offset is 2*i.
4525          */
4526         for (i = 0; i < 32; i++) {
4527                 if (i == 2 || i == 3)
4528                         continue;
4529                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4530         }
4531         wmb();
4532         pci_write_config_word(pdev, 4, saved_config_space[2]);
4533
4534         /* Some devices (notably the HP Smart Array 5i Controller)
4535            need a little pause here */
4536         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4537
4538         /* Controller should be in simple mode at this point.  If it's not,
4539          * It means we're on one of those controllers which doesn't support
4540          * the doorbell reset method and on which the PCI power management reset
4541          * method doesn't work (P800, for example.)
4542          * In those cases, don't try to proceed, as it generally doesn't work.
4543          */
4544         active_transport = readl(&cfgtable->TransportActive);
4545         if (active_transport & PERFORMANT_MODE) {
4546                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4547                         " Ignoring controller.\n");
4548                 rc = -ENODEV;
4549         }
4550
4551 unmap_cfgtable:
4552         iounmap(cfgtable);
4553
4554 unmap_vaddr:
4555         iounmap(vaddr);
4556         return rc;
4557 }
4558
4559 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4560 {
4561         int rc, i;
4562
4563         if (!reset_devices)
4564                 return 0;
4565
4566         /* Reset the controller with a PCI power-cycle or via doorbell */
4567         rc = cciss_kdump_hard_reset_controller(pdev);
4568
4569         /* -ENOTSUPP here means we cannot reset the controller
4570          * but it's already (and still) up and running in
4571          * "performant mode".  Or, it might be 640x, which can't reset
4572          * due to concerns about shared bbwc between 6402/6404 pair.
4573          */
4574         if (rc == -ENOTSUPP)
4575                 return 0; /* just try to do the kdump anyhow. */
4576         if (rc)
4577                 return -ENODEV;
4578         if (cciss_reset_msi(pdev))
4579                 return -ENODEV;
4580
4581         /* Now try to get the controller to respond to a no-op */
4582         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4583                 if (cciss_noop(pdev) == 0)
4584                         break;
4585                 else
4586                         dev_warn(&pdev->dev, "no-op failed%s\n",
4587                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4588                                         "; re-trying" : ""));
4589                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4590         }
4591         return 0;
4592 }
4593
4594 /*
4595  *  This is it.  Find all the controllers and register them.  I really hate
4596  *  stealing all these major device numbers.
4597  *  returns the number of block devices registered.
4598  */
4599 static int __devinit cciss_init_one(struct pci_dev *pdev,
4600                                     const struct pci_device_id *ent)
4601 {
4602         int i;
4603         int j = 0;
4604         int k = 0;
4605         int rc;
4606         int dac, return_code;
4607         InquiryData_struct *inq_buff;
4608         ctlr_info_t *h;
4609
4610         rc = cciss_init_reset_devices(pdev);
4611         if (rc)
4612                 return rc;
4613         i = alloc_cciss_hba(pdev);
4614         if (i < 0)
4615                 return -1;
4616
4617         h = hba[i];
4618         h->pdev = pdev;
4619         h->busy_initializing = 1;
4620         INIT_HLIST_HEAD(&h->cmpQ);
4621         INIT_HLIST_HEAD(&h->reqQ);
4622         mutex_init(&h->busy_shutting_down);
4623
4624         if (cciss_pci_init(h) != 0)
4625                 goto clean_no_release_regions;
4626
4627         sprintf(h->devname, "cciss%d", i);
4628         h->ctlr = i;
4629
4630         init_completion(&h->scan_wait);
4631
4632         if (cciss_create_hba_sysfs_entry(h))
4633                 goto clean0;
4634
4635         /* configure PCI DMA stuff */
4636         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4637                 dac = 1;
4638         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4639                 dac = 0;
4640         else {
4641                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4642                 goto clean1;
4643         }
4644
4645         /*
4646          * register with the major number, or get a dynamic major number
4647          * by passing 0 as argument.  This is done for greater than
4648          * 8 controller support.
4649          */
4650         if (i < MAX_CTLR_ORIG)
4651                 h->major = COMPAQ_CISS_MAJOR + i;
4652         rc = register_blkdev(h->major, h->devname);
4653         if (rc == -EBUSY || rc == -EINVAL) {
4654                 dev_err(&h->pdev->dev,
4655                        "Unable to get major number %d for %s "
4656                        "on hba %d\n", h->major, h->devname, i);
4657                 goto clean1;
4658         } else {
4659                 if (i >= MAX_CTLR_ORIG)
4660                         h->major = rc;
4661         }
4662
4663         /* make sure the board interrupts are off */
4664         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4665         if (h->msi_vector || h->msix_vector) {
4666                 if (request_irq(h->intr[PERF_MODE_INT],
4667                                 do_cciss_msix_intr,
4668                                 IRQF_DISABLED, h->devname, h)) {
4669                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4670                                h->intr[PERF_MODE_INT], h->devname);
4671                         goto clean2;
4672                 }
4673         } else {
4674                 if (request_irq(h->intr[PERF_MODE_INT], do_cciss_intx,
4675                                 IRQF_DISABLED, h->devname, h)) {
4676                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4677                                h->intr[PERF_MODE_INT], h->devname);
4678                         goto clean2;
4679                 }
4680         }
4681
4682         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4683                h->devname, pdev->device, pci_name(pdev),
4684                h->intr[PERF_MODE_INT], dac ? "" : " not");
4685
4686         h->cmd_pool_bits =
4687             kmalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4688                         * sizeof(unsigned long), GFP_KERNEL);
4689         h->cmd_pool = (CommandList_struct *)
4690             pci_alloc_consistent(h->pdev,
4691                     h->nr_cmds * sizeof(CommandList_struct),
4692                     &(h->cmd_pool_dhandle));
4693         h->errinfo_pool = (ErrorInfo_struct *)
4694             pci_alloc_consistent(h->pdev,
4695                     h->nr_cmds * sizeof(ErrorInfo_struct),
4696                     &(h->errinfo_pool_dhandle));
4697         if ((h->cmd_pool_bits == NULL)
4698             || (h->cmd_pool == NULL)
4699             || (h->errinfo_pool == NULL)) {
4700                 dev_err(&h->pdev->dev, "out of memory");
4701                 goto clean4;
4702         }
4703
4704         /* Need space for temp scatter list */
4705         h->scatter_list = kmalloc(h->max_commands *
4706                                                 sizeof(struct scatterlist *),
4707                                                 GFP_KERNEL);
4708         if (!h->scatter_list)
4709                 goto clean4;
4710
4711         for (k = 0; k < h->nr_cmds; k++) {
4712                 h->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4713                                                         h->maxsgentries,
4714                                                         GFP_KERNEL);
4715                 if (h->scatter_list[k] == NULL) {
4716                         dev_err(&h->pdev->dev,
4717                                 "could not allocate s/g lists\n");
4718                         goto clean4;
4719                 }
4720         }
4721         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4722                 h->chainsize, h->nr_cmds);
4723         if (!h->cmd_sg_list && h->chainsize > 0)
4724                 goto clean4;
4725
4726         spin_lock_init(&h->lock);
4727
4728         /* Initialize the pdev driver private data.
4729            have it point to h.  */
4730         pci_set_drvdata(pdev, h);
4731         /* command and error info recs zeroed out before
4732            they are used */
4733         memset(h->cmd_pool_bits, 0,
4734                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4735                         * sizeof(unsigned long));
4736
4737         h->num_luns = 0;
4738         h->highest_lun = -1;
4739         for (j = 0; j < CISS_MAX_LUN; j++) {
4740                 h->drv[j] = NULL;
4741                 h->gendisk[j] = NULL;
4742         }
4743
4744         cciss_scsi_setup(h);
4745
4746         /* Turn the interrupts on so we can service requests */
4747         h->access.set_intr_mask(h, CCISS_INTR_ON);
4748
4749         /* Get the firmware version */
4750         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4751         if (inq_buff == NULL) {
4752                 dev_err(&h->pdev->dev, "out of memory\n");
4753                 goto clean4;
4754         }
4755
4756         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4757                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4758         if (return_code == IO_OK) {
4759                 h->firm_ver[0] = inq_buff->data_byte[32];
4760                 h->firm_ver[1] = inq_buff->data_byte[33];
4761                 h->firm_ver[2] = inq_buff->data_byte[34];
4762                 h->firm_ver[3] = inq_buff->data_byte[35];
4763         } else {         /* send command failed */
4764                 dev_warn(&h->pdev->dev, "unable to determine firmware"
4765                         " version of controller\n");
4766         }
4767         kfree(inq_buff);
4768
4769         cciss_procinit(h);
4770
4771         h->cciss_max_sectors = 8192;
4772
4773         rebuild_lun_table(h, 1, 0);
4774         h->busy_initializing = 0;
4775         return 1;
4776
4777 clean4:
4778         kfree(h->cmd_pool_bits);
4779         /* Free up sg elements */
4780         for (k-- ; k >= 0; k--)
4781                 kfree(h->scatter_list[k]);
4782         kfree(h->scatter_list);
4783         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4784         if (h->cmd_pool)
4785                 pci_free_consistent(h->pdev,
4786                                     h->nr_cmds * sizeof(CommandList_struct),
4787                                     h->cmd_pool, h->cmd_pool_dhandle);
4788         if (h->errinfo_pool)
4789                 pci_free_consistent(h->pdev,
4790                                     h->nr_cmds * sizeof(ErrorInfo_struct),
4791                                     h->errinfo_pool,
4792                                     h->errinfo_pool_dhandle);
4793         free_irq(h->intr[PERF_MODE_INT], h);
4794 clean2:
4795         unregister_blkdev(h->major, h->devname);
4796 clean1:
4797         cciss_destroy_hba_sysfs_entry(h);
4798 clean0:
4799         pci_release_regions(pdev);
4800 clean_no_release_regions:
4801         h->busy_initializing = 0;
4802
4803         /*
4804          * Deliberately omit pci_disable_device(): it does something nasty to
4805          * Smart Array controllers that pci_enable_device does not undo
4806          */
4807         pci_set_drvdata(pdev, NULL);
4808         free_hba(h);
4809         return -1;
4810 }
4811
4812 static void cciss_shutdown(struct pci_dev *pdev)
4813 {
4814         ctlr_info_t *h;
4815         char *flush_buf;
4816         int return_code;
4817
4818         h = pci_get_drvdata(pdev);
4819         flush_buf = kzalloc(4, GFP_KERNEL);
4820         if (!flush_buf) {
4821                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
4822                 return;
4823         }
4824         /* write all data in the battery backed cache to disk */
4825         memset(flush_buf, 0, 4);
4826         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4827                 4, 0, CTLR_LUNID, TYPE_CMD);
4828         kfree(flush_buf);
4829         if (return_code != IO_OK)
4830                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
4831         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4832         free_irq(h->intr[PERF_MODE_INT], h);
4833 }
4834
4835 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4836 {
4837         ctlr_info_t *h;
4838         int i, j;
4839
4840         if (pci_get_drvdata(pdev) == NULL) {
4841                 dev_err(&pdev->dev, "Unable to remove device\n");
4842                 return;
4843         }
4844
4845         h = pci_get_drvdata(pdev);
4846         i = h->ctlr;
4847         if (hba[i] == NULL) {
4848                 dev_err(&pdev->dev, "device appears to already be removed\n");
4849                 return;
4850         }
4851
4852         mutex_lock(&h->busy_shutting_down);
4853
4854         remove_from_scan_list(h);
4855         remove_proc_entry(h->devname, proc_cciss);
4856         unregister_blkdev(h->major, h->devname);
4857
4858         /* remove it from the disk list */
4859         for (j = 0; j < CISS_MAX_LUN; j++) {
4860                 struct gendisk *disk = h->gendisk[j];
4861                 if (disk) {
4862                         struct request_queue *q = disk->queue;
4863
4864                         if (disk->flags & GENHD_FL_UP) {
4865                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
4866                                 del_gendisk(disk);
4867                         }
4868                         if (q)
4869                                 blk_cleanup_queue(q);
4870                 }
4871         }
4872
4873 #ifdef CONFIG_CISS_SCSI_TAPE
4874         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
4875 #endif
4876
4877         cciss_shutdown(pdev);
4878
4879 #ifdef CONFIG_PCI_MSI
4880         if (h->msix_vector)
4881                 pci_disable_msix(h->pdev);
4882         else if (h->msi_vector)
4883                 pci_disable_msi(h->pdev);
4884 #endif                          /* CONFIG_PCI_MSI */
4885
4886         iounmap(h->transtable);
4887         iounmap(h->cfgtable);
4888         iounmap(h->vaddr);
4889
4890         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(CommandList_struct),
4891                             h->cmd_pool, h->cmd_pool_dhandle);
4892         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(ErrorInfo_struct),
4893                             h->errinfo_pool, h->errinfo_pool_dhandle);
4894         kfree(h->cmd_pool_bits);
4895         /* Free up sg elements */
4896         for (j = 0; j < h->nr_cmds; j++)
4897                 kfree(h->scatter_list[j]);
4898         kfree(h->scatter_list);
4899         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4900         /*
4901          * Deliberately omit pci_disable_device(): it does something nasty to
4902          * Smart Array controllers that pci_enable_device does not undo
4903          */
4904         pci_release_regions(pdev);
4905         pci_set_drvdata(pdev, NULL);
4906         cciss_destroy_hba_sysfs_entry(h);
4907         mutex_unlock(&h->busy_shutting_down);
4908         free_hba(h);
4909 }
4910
4911 static struct pci_driver cciss_pci_driver = {
4912         .name = "cciss",
4913         .probe = cciss_init_one,
4914         .remove = __devexit_p(cciss_remove_one),
4915         .id_table = cciss_pci_device_id,        /* id_table */
4916         .shutdown = cciss_shutdown,
4917 };
4918
4919 /*
4920  *  This is it.  Register the PCI driver information for the cards we control
4921  *  the OS will call our registered routines when it finds one of our cards.
4922  */
4923 static int __init cciss_init(void)
4924 {
4925         int err;
4926
4927         /*
4928          * The hardware requires that commands are aligned on a 64-bit
4929          * boundary. Given that we use pci_alloc_consistent() to allocate an
4930          * array of them, the size must be a multiple of 8 bytes.
4931          */
4932         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4933         printk(KERN_INFO DRIVER_NAME "\n");
4934
4935         err = bus_register(&cciss_bus_type);
4936         if (err)
4937                 return err;
4938
4939         /* Start the scan thread */
4940         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4941         if (IS_ERR(cciss_scan_thread)) {
4942                 err = PTR_ERR(cciss_scan_thread);
4943                 goto err_bus_unregister;
4944         }
4945
4946         /* Register for our PCI devices */
4947         err = pci_register_driver(&cciss_pci_driver);
4948         if (err)
4949                 goto err_thread_stop;
4950
4951         return err;
4952
4953 err_thread_stop:
4954         kthread_stop(cciss_scan_thread);
4955 err_bus_unregister:
4956         bus_unregister(&cciss_bus_type);
4957
4958         return err;
4959 }
4960
4961 static void __exit cciss_cleanup(void)
4962 {
4963         int i;
4964
4965         pci_unregister_driver(&cciss_pci_driver);
4966         /* double check that all controller entrys have been removed */
4967         for (i = 0; i < MAX_CTLR; i++) {
4968                 if (hba[i] != NULL) {
4969                         dev_warn(&hba[i]->pdev->dev,
4970                                 "had to remove controller\n");
4971                         cciss_remove_one(hba[i]->pdev);
4972                 }
4973         }
4974         kthread_stop(cciss_scan_thread);
4975         remove_proc_entry("driver/cciss", NULL);
4976         bus_unregister(&cciss_bus_type);
4977 }
4978
4979 module_init(cciss_init);
4980 module_exit(cciss_cleanup);