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         err = 0;
1188         err |=
1189             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1190                            sizeof(arg64.LUN_info));
1191         err |=
1192             copy_from_user(&arg64.Request, &arg32->Request,
1193                            sizeof(arg64.Request));
1194         err |=
1195             copy_from_user(&arg64.error_info, &arg32->error_info,
1196                            sizeof(arg64.error_info));
1197         err |= get_user(arg64.buf_size, &arg32->buf_size);
1198         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1199         err |= get_user(cp, &arg32->buf);
1200         arg64.buf = compat_ptr(cp);
1201         err |= copy_to_user(p, &arg64, sizeof(arg64));
1202
1203         if (err)
1204                 return -EFAULT;
1205
1206         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1207         if (err)
1208                 return err;
1209         err |=
1210             copy_in_user(&arg32->error_info, &p->error_info,
1211                          sizeof(arg32->error_info));
1212         if (err)
1213                 return -EFAULT;
1214         return err;
1215 }
1216 #endif
1217
1218 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1219 {
1220         drive_info_struct *drv = get_drv(bdev->bd_disk);
1221
1222         if (!drv->cylinders)
1223                 return -ENXIO;
1224
1225         geo->heads = drv->heads;
1226         geo->sectors = drv->sectors;
1227         geo->cylinders = drv->cylinders;
1228         return 0;
1229 }
1230
1231 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1232 {
1233         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1234                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1235                 (void)check_for_unit_attention(h, c);
1236 }
1237
1238 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1239 {
1240         cciss_pci_info_struct pciinfo;
1241
1242         if (!argp)
1243                 return -EINVAL;
1244         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1245         pciinfo.bus = h->pdev->bus->number;
1246         pciinfo.dev_fn = h->pdev->devfn;
1247         pciinfo.board_id = h->board_id;
1248         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1249                 return -EFAULT;
1250         return 0;
1251 }
1252
1253 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1254 {
1255         cciss_coalint_struct intinfo;
1256
1257         if (!argp)
1258                 return -EINVAL;
1259         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1260         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1261         if (copy_to_user
1262             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1263                 return -EFAULT;
1264         return 0;
1265 }
1266
1267 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1268 {
1269         cciss_coalint_struct intinfo;
1270         unsigned long flags;
1271         int i;
1272
1273         if (!argp)
1274                 return -EINVAL;
1275         if (!capable(CAP_SYS_ADMIN))
1276                 return -EPERM;
1277         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1278                 return -EFAULT;
1279         if ((intinfo.delay == 0) && (intinfo.count == 0))
1280                 return -EINVAL;
1281         spin_lock_irqsave(&h->lock, flags);
1282         /* Update the field, and then ring the doorbell */
1283         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1284         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1285         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1286
1287         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1288                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1289                         break;
1290                 udelay(1000); /* delay and try again */
1291         }
1292         spin_unlock_irqrestore(&h->lock, flags);
1293         if (i >= MAX_IOCTL_CONFIG_WAIT)
1294                 return -EAGAIN;
1295         return 0;
1296 }
1297
1298 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1299 {
1300         NodeName_type NodeName;
1301         int i;
1302
1303         if (!argp)
1304                 return -EINVAL;
1305         for (i = 0; i < 16; i++)
1306                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1307         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1308                 return -EFAULT;
1309         return 0;
1310 }
1311
1312 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1313 {
1314         NodeName_type NodeName;
1315         unsigned long flags;
1316         int i;
1317
1318         if (!argp)
1319                 return -EINVAL;
1320         if (!capable(CAP_SYS_ADMIN))
1321                 return -EPERM;
1322         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1323                 return -EFAULT;
1324         spin_lock_irqsave(&h->lock, flags);
1325         /* Update the field, and then ring the doorbell */
1326         for (i = 0; i < 16; i++)
1327                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1328         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1329         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1330                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1331                         break;
1332                 udelay(1000); /* delay and try again */
1333         }
1334         spin_unlock_irqrestore(&h->lock, flags);
1335         if (i >= MAX_IOCTL_CONFIG_WAIT)
1336                 return -EAGAIN;
1337         return 0;
1338 }
1339
1340 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1341 {
1342         Heartbeat_type heartbeat;
1343
1344         if (!argp)
1345                 return -EINVAL;
1346         heartbeat = readl(&h->cfgtable->HeartBeat);
1347         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1348                 return -EFAULT;
1349         return 0;
1350 }
1351
1352 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1353 {
1354         BusTypes_type BusTypes;
1355
1356         if (!argp)
1357                 return -EINVAL;
1358         BusTypes = readl(&h->cfgtable->BusTypes);
1359         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1360                 return -EFAULT;
1361         return 0;
1362 }
1363
1364 static int cciss_getfirmver(ctlr_info_t *h, void __user *argp)
1365 {
1366         FirmwareVer_type firmware;
1367
1368         if (!argp)
1369                 return -EINVAL;
1370         memcpy(firmware, h->firm_ver, 4);
1371
1372         if (copy_to_user
1373             (argp, firmware, sizeof(FirmwareVer_type)))
1374                 return -EFAULT;
1375         return 0;
1376 }
1377
1378 static int cciss_getdrivver(ctlr_info_t *h, void __user *argp)
1379 {
1380         DriverVer_type DriverVer = DRIVER_VERSION;
1381
1382         if (!argp)
1383                 return -EINVAL;
1384         if (copy_to_user(argp, &DriverVer, sizeof(DriverVer_type)))
1385                 return -EFAULT;
1386         return 0;
1387 }
1388
1389 static int cciss_getluninfo(ctlr_info_t *h,
1390         struct gendisk *disk, void __user *argp)
1391 {
1392         LogvolInfo_struct luninfo;
1393         drive_info_struct *drv = get_drv(disk);
1394
1395         if (!argp)
1396                 return -EINVAL;
1397         memcpy(&luninfo.LunID, drv->LunID, sizeof(luninfo.LunID));
1398         luninfo.num_opens = drv->usage_count;
1399         luninfo.num_parts = 0;
1400         if (copy_to_user(argp, &luninfo, sizeof(LogvolInfo_struct)))
1401                 return -EFAULT;
1402         return 0;
1403 }
1404
1405 static int cciss_passthru(ctlr_info_t *h, void __user *argp)
1406 {
1407         IOCTL_Command_struct iocommand;
1408         CommandList_struct *c;
1409         char *buff = NULL;
1410         u64bit temp64;
1411         DECLARE_COMPLETION_ONSTACK(wait);
1412
1413         if (!argp)
1414                 return -EINVAL;
1415
1416         if (!capable(CAP_SYS_RAWIO))
1417                 return -EPERM;
1418
1419         if (copy_from_user
1420             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1421                 return -EFAULT;
1422         if ((iocommand.buf_size < 1) &&
1423             (iocommand.Request.Type.Direction != XFER_NONE)) {
1424                 return -EINVAL;
1425         }
1426         if (iocommand.buf_size > 0) {
1427                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1428                 if (buff == NULL)
1429                         return -EFAULT;
1430         }
1431         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1432                 /* Copy the data into the buffer we created */
1433                 if (copy_from_user(buff, iocommand.buf, iocommand.buf_size)) {
1434                         kfree(buff);
1435                         return -EFAULT;
1436                 }
1437         } else {
1438                 memset(buff, 0, iocommand.buf_size);
1439         }
1440         c = cmd_special_alloc(h);
1441         if (!c) {
1442                 kfree(buff);
1443                 return -ENOMEM;
1444         }
1445         /* Fill in the command type */
1446         c->cmd_type = CMD_IOCTL_PEND;
1447         /* Fill in Command Header */
1448         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1449         if (iocommand.buf_size > 0) { /* buffer to fill */
1450                 c->Header.SGList = 1;
1451                 c->Header.SGTotal = 1;
1452         } else { /* no buffers to fill */
1453                 c->Header.SGList = 0;
1454                 c->Header.SGTotal = 0;
1455         }
1456         c->Header.LUN = iocommand.LUN_info;
1457         /* use the kernel address the cmd block for tag */
1458         c->Header.Tag.lower = c->busaddr;
1459
1460         /* Fill in Request block */
1461         c->Request = iocommand.Request;
1462
1463         /* Fill in the scatter gather information */
1464         if (iocommand.buf_size > 0) {
1465                 temp64.val = pci_map_single(h->pdev, buff,
1466                         iocommand.buf_size, PCI_DMA_BIDIRECTIONAL);
1467                 c->SG[0].Addr.lower = temp64.val32.lower;
1468                 c->SG[0].Addr.upper = temp64.val32.upper;
1469                 c->SG[0].Len = iocommand.buf_size;
1470                 c->SG[0].Ext = 0;  /* we are not chaining */
1471         }
1472         c->waiting = &wait;
1473
1474         enqueue_cmd_and_start_io(h, c);
1475         wait_for_completion(&wait);
1476
1477         /* unlock the buffers from DMA */
1478         temp64.val32.lower = c->SG[0].Addr.lower;
1479         temp64.val32.upper = c->SG[0].Addr.upper;
1480         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val, iocommand.buf_size,
1481                          PCI_DMA_BIDIRECTIONAL);
1482         check_ioctl_unit_attention(h, c);
1483
1484         /* Copy the error information out */
1485         iocommand.error_info = *(c->err_info);
1486         if (copy_to_user(argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1487                 kfree(buff);
1488                 cmd_special_free(h, c);
1489                 return -EFAULT;
1490         }
1491
1492         if (iocommand.Request.Type.Direction == XFER_READ) {
1493                 /* Copy the data out of the buffer we created */
1494                 if (copy_to_user(iocommand.buf, buff, iocommand.buf_size)) {
1495                         kfree(buff);
1496                         cmd_special_free(h, c);
1497                         return -EFAULT;
1498                 }
1499         }
1500         kfree(buff);
1501         cmd_special_free(h, c);
1502         return 0;
1503 }
1504
1505 static int cciss_bigpassthru(ctlr_info_t *h, void __user *argp)
1506 {
1507         BIG_IOCTL_Command_struct *ioc;
1508         CommandList_struct *c;
1509         unsigned char **buff = NULL;
1510         int *buff_size = NULL;
1511         u64bit temp64;
1512         BYTE sg_used = 0;
1513         int status = 0;
1514         int i;
1515         DECLARE_COMPLETION_ONSTACK(wait);
1516         __u32 left;
1517         __u32 sz;
1518         BYTE __user *data_ptr;
1519
1520         if (!argp)
1521                 return -EINVAL;
1522         if (!capable(CAP_SYS_RAWIO))
1523                 return -EPERM;
1524         ioc = (BIG_IOCTL_Command_struct *)
1525             kmalloc(sizeof(*ioc), GFP_KERNEL);
1526         if (!ioc) {
1527                 status = -ENOMEM;
1528                 goto cleanup1;
1529         }
1530         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1531                 status = -EFAULT;
1532                 goto cleanup1;
1533         }
1534         if ((ioc->buf_size < 1) &&
1535             (ioc->Request.Type.Direction != XFER_NONE)) {
1536                 status = -EINVAL;
1537                 goto cleanup1;
1538         }
1539         /* Check kmalloc limits  using all SGs */
1540         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1541                 status = -EINVAL;
1542                 goto cleanup1;
1543         }
1544         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1545                 status = -EINVAL;
1546                 goto cleanup1;
1547         }
1548         buff = kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1549         if (!buff) {
1550                 status = -ENOMEM;
1551                 goto cleanup1;
1552         }
1553         buff_size = kmalloc(MAXSGENTRIES * sizeof(int), GFP_KERNEL);
1554         if (!buff_size) {
1555                 status = -ENOMEM;
1556                 goto cleanup1;
1557         }
1558         left = ioc->buf_size;
1559         data_ptr = ioc->buf;
1560         while (left) {
1561                 sz = (left > ioc->malloc_size) ? ioc->malloc_size : left;
1562                 buff_size[sg_used] = sz;
1563                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1564                 if (buff[sg_used] == NULL) {
1565                         status = -ENOMEM;
1566                         goto cleanup1;
1567                 }
1568                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1569                         if (copy_from_user(buff[sg_used], data_ptr, sz)) {
1570                                 status = -EFAULT;
1571                                 goto cleanup1;
1572                         }
1573                 } else {
1574                         memset(buff[sg_used], 0, sz);
1575                 }
1576                 left -= sz;
1577                 data_ptr += sz;
1578                 sg_used++;
1579         }
1580         c = cmd_special_alloc(h);
1581         if (!c) {
1582                 status = -ENOMEM;
1583                 goto cleanup1;
1584         }
1585         c->cmd_type = CMD_IOCTL_PEND;
1586         c->Header.ReplyQueue = 0;
1587         c->Header.SGList = sg_used;
1588         c->Header.SGTotal = sg_used;
1589         c->Header.LUN = ioc->LUN_info;
1590         c->Header.Tag.lower = c->busaddr;
1591
1592         c->Request = ioc->Request;
1593         for (i = 0; i < sg_used; i++) {
1594                 temp64.val = pci_map_single(h->pdev, buff[i], buff_size[i],
1595                                     PCI_DMA_BIDIRECTIONAL);
1596                 c->SG[i].Addr.lower = temp64.val32.lower;
1597                 c->SG[i].Addr.upper = temp64.val32.upper;
1598                 c->SG[i].Len = buff_size[i];
1599                 c->SG[i].Ext = 0;       /* we are not chaining */
1600         }
1601         c->waiting = &wait;
1602         enqueue_cmd_and_start_io(h, c);
1603         wait_for_completion(&wait);
1604         /* unlock the buffers from DMA */
1605         for (i = 0; i < sg_used; i++) {
1606                 temp64.val32.lower = c->SG[i].Addr.lower;
1607                 temp64.val32.upper = c->SG[i].Addr.upper;
1608                 pci_unmap_single(h->pdev,
1609                         (dma_addr_t) temp64.val, buff_size[i],
1610                         PCI_DMA_BIDIRECTIONAL);
1611         }
1612         check_ioctl_unit_attention(h, c);
1613         /* Copy the error information out */
1614         ioc->error_info = *(c->err_info);
1615         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1616                 cmd_special_free(h, c);
1617                 status = -EFAULT;
1618                 goto cleanup1;
1619         }
1620         if (ioc->Request.Type.Direction == XFER_READ) {
1621                 /* Copy the data out of the buffer we created */
1622                 BYTE __user *ptr = ioc->buf;
1623                 for (i = 0; i < sg_used; i++) {
1624                         if (copy_to_user(ptr, buff[i], buff_size[i])) {
1625                                 cmd_special_free(h, c);
1626                                 status = -EFAULT;
1627                                 goto cleanup1;
1628                         }
1629                         ptr += buff_size[i];
1630                 }
1631         }
1632         cmd_special_free(h, c);
1633         status = 0;
1634 cleanup1:
1635         if (buff) {
1636                 for (i = 0; i < sg_used; i++)
1637                         kfree(buff[i]);
1638                 kfree(buff);
1639         }
1640         kfree(buff_size);
1641         kfree(ioc);
1642         return status;
1643 }
1644
1645 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1646         unsigned int cmd, unsigned long arg)
1647 {
1648         struct gendisk *disk = bdev->bd_disk;
1649         ctlr_info_t *h = get_host(disk);
1650         void __user *argp = (void __user *)arg;
1651
1652         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1653                 cmd, arg);
1654         switch (cmd) {
1655         case CCISS_GETPCIINFO:
1656                 return cciss_getpciinfo(h, argp);
1657         case CCISS_GETINTINFO:
1658                 return cciss_getintinfo(h, argp);
1659         case CCISS_SETINTINFO:
1660                 return cciss_setintinfo(h, argp);
1661         case CCISS_GETNODENAME:
1662                 return cciss_getnodename(h, argp);
1663         case CCISS_SETNODENAME:
1664                 return cciss_setnodename(h, argp);
1665         case CCISS_GETHEARTBEAT:
1666                 return cciss_getheartbeat(h, argp);
1667         case CCISS_GETBUSTYPES:
1668                 return cciss_getbustypes(h, argp);
1669         case CCISS_GETFIRMVER:
1670                 return cciss_getfirmver(h, argp);
1671         case CCISS_GETDRIVVER:
1672                 return cciss_getdrivver(h, argp);
1673         case CCISS_DEREGDISK:
1674         case CCISS_REGNEWD:
1675         case CCISS_REVALIDVOLS:
1676                 return rebuild_lun_table(h, 0, 1);
1677         case CCISS_GETLUNINFO:
1678                 return cciss_getluninfo(h, disk, argp);
1679         case CCISS_PASSTHRU:
1680                 return cciss_passthru(h, argp);
1681         case CCISS_BIG_PASSTHRU:
1682                 return cciss_bigpassthru(h, argp);
1683
1684         /* scsi_cmd_ioctl handles these, below, though some are not */
1685         /* very meaningful for cciss.  SG_IO is the main one people want. */
1686
1687         case SG_GET_VERSION_NUM:
1688         case SG_SET_TIMEOUT:
1689         case SG_GET_TIMEOUT:
1690         case SG_GET_RESERVED_SIZE:
1691         case SG_SET_RESERVED_SIZE:
1692         case SG_EMULATED_HOST:
1693         case SG_IO:
1694         case SCSI_IOCTL_SEND_COMMAND:
1695                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1696
1697         /* scsi_cmd_ioctl would normally handle these, below, but */
1698         /* they aren't a good fit for cciss, as CD-ROMs are */
1699         /* not supported, and we don't have any bus/target/lun */
1700         /* which we present to the kernel. */
1701
1702         case CDROM_SEND_PACKET:
1703         case CDROMCLOSETRAY:
1704         case CDROMEJECT:
1705         case SCSI_IOCTL_GET_IDLUN:
1706         case SCSI_IOCTL_GET_BUS_NUMBER:
1707         default:
1708                 return -ENOTTY;
1709         }
1710 }
1711
1712 static void cciss_check_queues(ctlr_info_t *h)
1713 {
1714         int start_queue = h->next_to_run;
1715         int i;
1716
1717         /* check to see if we have maxed out the number of commands that can
1718          * be placed on the queue.  If so then exit.  We do this check here
1719          * in case the interrupt we serviced was from an ioctl and did not
1720          * free any new commands.
1721          */
1722         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1723                 return;
1724
1725         /* We have room on the queue for more commands.  Now we need to queue
1726          * them up.  We will also keep track of the next queue to run so
1727          * that every queue gets a chance to be started first.
1728          */
1729         for (i = 0; i < h->highest_lun + 1; i++) {
1730                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1731                 /* make sure the disk has been added and the drive is real
1732                  * because this can be called from the middle of init_one.
1733                  */
1734                 if (!h->drv[curr_queue])
1735                         continue;
1736                 if (!(h->drv[curr_queue]->queue) ||
1737                         !(h->drv[curr_queue]->heads))
1738                         continue;
1739                 blk_start_queue(h->gendisk[curr_queue]->queue);
1740
1741                 /* check to see if we have maxed out the number of commands
1742                  * that can be placed on the queue.
1743                  */
1744                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1745                         if (curr_queue == start_queue) {
1746                                 h->next_to_run =
1747                                     (start_queue + 1) % (h->highest_lun + 1);
1748                                 break;
1749                         } else {
1750                                 h->next_to_run = curr_queue;
1751                                 break;
1752                         }
1753                 }
1754         }
1755 }
1756
1757 static void cciss_softirq_done(struct request *rq)
1758 {
1759         CommandList_struct *c = rq->completion_data;
1760         ctlr_info_t *h = hba[c->ctlr];
1761         SGDescriptor_struct *curr_sg = c->SG;
1762         u64bit temp64;
1763         unsigned long flags;
1764         int i, ddir;
1765         int sg_index = 0;
1766
1767         if (c->Request.Type.Direction == XFER_READ)
1768                 ddir = PCI_DMA_FROMDEVICE;
1769         else
1770                 ddir = PCI_DMA_TODEVICE;
1771
1772         /* command did not need to be retried */
1773         /* unmap the DMA mapping for all the scatter gather elements */
1774         for (i = 0; i < c->Header.SGList; i++) {
1775                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1776                         cciss_unmap_sg_chain_block(h, c);
1777                         /* Point to the next block */
1778                         curr_sg = h->cmd_sg_list[c->cmdindex];
1779                         sg_index = 0;
1780                 }
1781                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1782                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1783                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1784                                 ddir);
1785                 ++sg_index;
1786         }
1787
1788         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1789
1790         /* set the residual count for pc requests */
1791         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1792                 rq->resid_len = c->err_info->ResidualCnt;
1793
1794         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1795
1796         spin_lock_irqsave(&h->lock, flags);
1797         cmd_free(h, c);
1798         cciss_check_queues(h);
1799         spin_unlock_irqrestore(&h->lock, flags);
1800 }
1801
1802 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1803         unsigned char scsi3addr[], uint32_t log_unit)
1804 {
1805         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1806                 sizeof(h->drv[log_unit]->LunID));
1807 }
1808
1809 /* This function gets the SCSI vendor, model, and revision of a logical drive
1810  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1811  * they cannot be read.
1812  */
1813 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1814                                    char *vendor, char *model, char *rev)
1815 {
1816         int rc;
1817         InquiryData_struct *inq_buf;
1818         unsigned char scsi3addr[8];
1819
1820         *vendor = '\0';
1821         *model = '\0';
1822         *rev = '\0';
1823
1824         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1825         if (!inq_buf)
1826                 return;
1827
1828         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1829         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1830                         scsi3addr, TYPE_CMD);
1831         if (rc == IO_OK) {
1832                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1833                 vendor[VENDOR_LEN] = '\0';
1834                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1835                 model[MODEL_LEN] = '\0';
1836                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1837                 rev[REV_LEN] = '\0';
1838         }
1839
1840         kfree(inq_buf);
1841         return;
1842 }
1843
1844 /* This function gets the serial number of a logical drive via
1845  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1846  * number cannot be had, for whatever reason, 16 bytes of 0xff
1847  * are returned instead.
1848  */
1849 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1850                                 unsigned char *serial_no, int buflen)
1851 {
1852 #define PAGE_83_INQ_BYTES 64
1853         int rc;
1854         unsigned char *buf;
1855         unsigned char scsi3addr[8];
1856
1857         if (buflen > 16)
1858                 buflen = 16;
1859         memset(serial_no, 0xff, buflen);
1860         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1861         if (!buf)
1862                 return;
1863         memset(serial_no, 0, buflen);
1864         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1865         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1866                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1867         if (rc == IO_OK)
1868                 memcpy(serial_no, &buf[8], buflen);
1869         kfree(buf);
1870         return;
1871 }
1872
1873 /*
1874  * cciss_add_disk sets up the block device queue for a logical drive
1875  */
1876 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1877                                 int drv_index)
1878 {
1879         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1880         if (!disk->queue)
1881                 goto init_queue_failure;
1882         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1883         disk->major = h->major;
1884         disk->first_minor = drv_index << NWD_SHIFT;
1885         disk->fops = &cciss_fops;
1886         if (cciss_create_ld_sysfs_entry(h, drv_index))
1887                 goto cleanup_queue;
1888         disk->private_data = h->drv[drv_index];
1889         disk->driverfs_dev = &h->drv[drv_index]->dev;
1890
1891         /* Set up queue information */
1892         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1893
1894         /* This is a hardware imposed limit. */
1895         blk_queue_max_segments(disk->queue, h->maxsgentries);
1896
1897         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1898
1899         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1900
1901         disk->queue->queuedata = h;
1902
1903         blk_queue_logical_block_size(disk->queue,
1904                                      h->drv[drv_index]->block_size);
1905
1906         /* Make sure all queue data is written out before */
1907         /* setting h->drv[drv_index]->queue, as setting this */
1908         /* allows the interrupt handler to start the queue */
1909         wmb();
1910         h->drv[drv_index]->queue = disk->queue;
1911         add_disk(disk);
1912         return 0;
1913
1914 cleanup_queue:
1915         blk_cleanup_queue(disk->queue);
1916         disk->queue = NULL;
1917 init_queue_failure:
1918         return -1;
1919 }
1920
1921 /* This function will check the usage_count of the drive to be updated/added.
1922  * If the usage_count is zero and it is a heretofore unknown drive, or,
1923  * the drive's capacity, geometry, or serial number has changed,
1924  * then the drive information will be updated and the disk will be
1925  * re-registered with the kernel.  If these conditions don't hold,
1926  * then it will be left alone for the next reboot.  The exception to this
1927  * is disk 0 which will always be left registered with the kernel since it
1928  * is also the controller node.  Any changes to disk 0 will show up on
1929  * the next reboot.
1930  */
1931 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1932         int first_time, int via_ioctl)
1933 {
1934         struct gendisk *disk;
1935         InquiryData_struct *inq_buff = NULL;
1936         unsigned int block_size;
1937         sector_t total_size;
1938         unsigned long flags = 0;
1939         int ret = 0;
1940         drive_info_struct *drvinfo;
1941
1942         /* Get information about the disk and modify the driver structure */
1943         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1944         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1945         if (inq_buff == NULL || drvinfo == NULL)
1946                 goto mem_msg;
1947
1948         /* testing to see if 16-byte CDBs are already being used */
1949         if (h->cciss_read == CCISS_READ_16) {
1950                 cciss_read_capacity_16(h, drv_index,
1951                         &total_size, &block_size);
1952
1953         } else {
1954                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1955                 /* if read_capacity returns all F's this volume is >2TB */
1956                 /* in size so we switch to 16-byte CDB's for all */
1957                 /* read/write ops */
1958                 if (total_size == 0xFFFFFFFFULL) {
1959                         cciss_read_capacity_16(h, drv_index,
1960                         &total_size, &block_size);
1961                         h->cciss_read = CCISS_READ_16;
1962                         h->cciss_write = CCISS_WRITE_16;
1963                 } else {
1964                         h->cciss_read = CCISS_READ_10;
1965                         h->cciss_write = CCISS_WRITE_10;
1966                 }
1967         }
1968
1969         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1970                                inq_buff, drvinfo);
1971         drvinfo->block_size = block_size;
1972         drvinfo->nr_blocks = total_size + 1;
1973
1974         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1975                                 drvinfo->model, drvinfo->rev);
1976         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1977                         sizeof(drvinfo->serial_no));
1978         /* Save the lunid in case we deregister the disk, below. */
1979         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1980                 sizeof(drvinfo->LunID));
1981
1982         /* Is it the same disk we already know, and nothing's changed? */
1983         if (h->drv[drv_index]->raid_level != -1 &&
1984                 ((memcmp(drvinfo->serial_no,
1985                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1986                 drvinfo->block_size == h->drv[drv_index]->block_size &&
1987                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
1988                 drvinfo->heads == h->drv[drv_index]->heads &&
1989                 drvinfo->sectors == h->drv[drv_index]->sectors &&
1990                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
1991                         /* The disk is unchanged, nothing to update */
1992                         goto freeret;
1993
1994         /* If we get here it's not the same disk, or something's changed,
1995          * so we need to * deregister it, and re-register it, if it's not
1996          * in use.
1997          * If the disk already exists then deregister it before proceeding
1998          * (unless it's the first disk (for the controller node).
1999          */
2000         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2001                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2002                 spin_lock_irqsave(&h->lock, flags);
2003                 h->drv[drv_index]->busy_configuring = 1;
2004                 spin_unlock_irqrestore(&h->lock, flags);
2005
2006                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2007                  * which keeps the interrupt handler from starting
2008                  * the queue.
2009                  */
2010                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2011         }
2012
2013         /* If the disk is in use return */
2014         if (ret)
2015                 goto freeret;
2016
2017         /* Save the new information from cciss_geometry_inquiry
2018          * and serial number inquiry.  If the disk was deregistered
2019          * above, then h->drv[drv_index] will be NULL.
2020          */
2021         if (h->drv[drv_index] == NULL) {
2022                 drvinfo->device_initialized = 0;
2023                 h->drv[drv_index] = drvinfo;
2024                 drvinfo = NULL; /* so it won't be freed below. */
2025         } else {
2026                 /* special case for cxd0 */
2027                 h->drv[drv_index]->block_size = drvinfo->block_size;
2028                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2029                 h->drv[drv_index]->heads = drvinfo->heads;
2030                 h->drv[drv_index]->sectors = drvinfo->sectors;
2031                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2032                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2033                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2034                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2035                         VENDOR_LEN + 1);
2036                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2037                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2038         }
2039
2040         ++h->num_luns;
2041         disk = h->gendisk[drv_index];
2042         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2043
2044         /* If it's not disk 0 (drv_index != 0)
2045          * or if it was disk 0, but there was previously
2046          * no actual corresponding configured logical drive
2047          * (raid_leve == -1) then we want to update the
2048          * logical drive's information.
2049          */
2050         if (drv_index || first_time) {
2051                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2052                         cciss_free_gendisk(h, drv_index);
2053                         cciss_free_drive_info(h, drv_index);
2054                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2055                                 drv_index);
2056                         --h->num_luns;
2057                 }
2058         }
2059
2060 freeret:
2061         kfree(inq_buff);
2062         kfree(drvinfo);
2063         return;
2064 mem_msg:
2065         dev_err(&h->pdev->dev, "out of memory\n");
2066         goto freeret;
2067 }
2068
2069 /* This function will find the first index of the controllers drive array
2070  * that has a null drv pointer and allocate the drive info struct and
2071  * will return that index   This is where new drives will be added.
2072  * If the index to be returned is greater than the highest_lun index for
2073  * the controller then highest_lun is set * to this new index.
2074  * If there are no available indexes or if tha allocation fails, then -1
2075  * is returned.  * "controller_node" is used to know if this is a real
2076  * logical drive, or just the controller node, which determines if this
2077  * counts towards highest_lun.
2078  */
2079 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2080 {
2081         int i;
2082         drive_info_struct *drv;
2083
2084         /* Search for an empty slot for our drive info */
2085         for (i = 0; i < CISS_MAX_LUN; i++) {
2086
2087                 /* if not cxd0 case, and it's occupied, skip it. */
2088                 if (h->drv[i] && i != 0)
2089                         continue;
2090                 /*
2091                  * If it's cxd0 case, and drv is alloc'ed already, and a
2092                  * disk is configured there, skip it.
2093                  */
2094                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2095                         continue;
2096
2097                 /*
2098                  * We've found an empty slot.  Update highest_lun
2099                  * provided this isn't just the fake cxd0 controller node.
2100                  */
2101                 if (i > h->highest_lun && !controller_node)
2102                         h->highest_lun = i;
2103
2104                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2105                 if (i == 0 && h->drv[i] != NULL)
2106                         return i;
2107
2108                 /*
2109                  * Found an empty slot, not already alloc'ed.  Allocate it.
2110                  * Mark it with raid_level == -1, so we know it's new later on.
2111                  */
2112                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2113                 if (!drv)
2114                         return -1;
2115                 drv->raid_level = -1; /* so we know it's new */
2116                 h->drv[i] = drv;
2117                 return i;
2118         }
2119         return -1;
2120 }
2121
2122 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2123 {
2124         kfree(h->drv[drv_index]);
2125         h->drv[drv_index] = NULL;
2126 }
2127
2128 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2129 {
2130         put_disk(h->gendisk[drv_index]);
2131         h->gendisk[drv_index] = NULL;
2132 }
2133
2134 /* cciss_add_gendisk finds a free hba[]->drv structure
2135  * and allocates a gendisk if needed, and sets the lunid
2136  * in the drvinfo structure.   It returns the index into
2137  * the ->drv[] array, or -1 if none are free.
2138  * is_controller_node indicates whether highest_lun should
2139  * count this disk, or if it's only being added to provide
2140  * a means to talk to the controller in case no logical
2141  * drives have yet been configured.
2142  */
2143 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2144         int controller_node)
2145 {
2146         int drv_index;
2147
2148         drv_index = cciss_alloc_drive_info(h, controller_node);
2149         if (drv_index == -1)
2150                 return -1;
2151
2152         /*Check if the gendisk needs to be allocated */
2153         if (!h->gendisk[drv_index]) {
2154                 h->gendisk[drv_index] =
2155                         alloc_disk(1 << NWD_SHIFT);
2156                 if (!h->gendisk[drv_index]) {
2157                         dev_err(&h->pdev->dev,
2158                                 "could not allocate a new disk %d\n",
2159                                 drv_index);
2160                         goto err_free_drive_info;
2161                 }
2162         }
2163         memcpy(h->drv[drv_index]->LunID, lunid,
2164                 sizeof(h->drv[drv_index]->LunID));
2165         if (cciss_create_ld_sysfs_entry(h, drv_index))
2166                 goto err_free_disk;
2167         /* Don't need to mark this busy because nobody */
2168         /* else knows about this disk yet to contend */
2169         /* for access to it. */
2170         h->drv[drv_index]->busy_configuring = 0;
2171         wmb();
2172         return drv_index;
2173
2174 err_free_disk:
2175         cciss_free_gendisk(h, drv_index);
2176 err_free_drive_info:
2177         cciss_free_drive_info(h, drv_index);
2178         return -1;
2179 }
2180
2181 /* This is for the special case of a controller which
2182  * has no logical drives.  In this case, we still need
2183  * to register a disk so the controller can be accessed
2184  * by the Array Config Utility.
2185  */
2186 static void cciss_add_controller_node(ctlr_info_t *h)
2187 {
2188         struct gendisk *disk;
2189         int drv_index;
2190
2191         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2192                 return;
2193
2194         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2195         if (drv_index == -1)
2196                 goto error;
2197         h->drv[drv_index]->block_size = 512;
2198         h->drv[drv_index]->nr_blocks = 0;
2199         h->drv[drv_index]->heads = 0;
2200         h->drv[drv_index]->sectors = 0;
2201         h->drv[drv_index]->cylinders = 0;
2202         h->drv[drv_index]->raid_level = -1;
2203         memset(h->drv[drv_index]->serial_no, 0, 16);
2204         disk = h->gendisk[drv_index];
2205         if (cciss_add_disk(h, disk, drv_index) == 0)
2206                 return;
2207         cciss_free_gendisk(h, drv_index);
2208         cciss_free_drive_info(h, drv_index);
2209 error:
2210         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2211         return;
2212 }
2213
2214 /* This function will add and remove logical drives from the Logical
2215  * drive array of the controller and maintain persistency of ordering
2216  * so that mount points are preserved until the next reboot.  This allows
2217  * for the removal of logical drives in the middle of the drive array
2218  * without a re-ordering of those drives.
2219  * INPUT
2220  * h            = The controller to perform the operations on
2221  */
2222 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2223         int via_ioctl)
2224 {
2225         int num_luns;
2226         ReportLunData_struct *ld_buff = NULL;
2227         int return_code;
2228         int listlength = 0;
2229         int i;
2230         int drv_found;
2231         int drv_index = 0;
2232         unsigned char lunid[8] = CTLR_LUNID;
2233         unsigned long flags;
2234
2235         if (!capable(CAP_SYS_RAWIO))
2236                 return -EPERM;
2237
2238         /* Set busy_configuring flag for this operation */
2239         spin_lock_irqsave(&h->lock, flags);
2240         if (h->busy_configuring) {
2241                 spin_unlock_irqrestore(&h->lock, flags);
2242                 return -EBUSY;
2243         }
2244         h->busy_configuring = 1;
2245         spin_unlock_irqrestore(&h->lock, flags);
2246
2247         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2248         if (ld_buff == NULL)
2249                 goto mem_msg;
2250
2251         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2252                                       sizeof(ReportLunData_struct),
2253                                       0, CTLR_LUNID, TYPE_CMD);
2254
2255         if (return_code == IO_OK)
2256                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2257         else {  /* reading number of logical volumes failed */
2258                 dev_warn(&h->pdev->dev,
2259                         "report logical volume command failed\n");
2260                 listlength = 0;
2261                 goto freeret;
2262         }
2263
2264         num_luns = listlength / 8;      /* 8 bytes per entry */
2265         if (num_luns > CISS_MAX_LUN) {
2266                 num_luns = CISS_MAX_LUN;
2267                 dev_warn(&h->pdev->dev, "more luns configured"
2268                        " on controller than can be handled by"
2269                        " this driver.\n");
2270         }
2271
2272         if (num_luns == 0)
2273                 cciss_add_controller_node(h);
2274
2275         /* Compare controller drive array to driver's drive array
2276          * to see if any drives are missing on the controller due
2277          * to action of Array Config Utility (user deletes drive)
2278          * and deregister logical drives which have disappeared.
2279          */
2280         for (i = 0; i <= h->highest_lun; i++) {
2281                 int j;
2282                 drv_found = 0;
2283
2284                 /* skip holes in the array from already deleted drives */
2285                 if (h->drv[i] == NULL)
2286                         continue;
2287
2288                 for (j = 0; j < num_luns; j++) {
2289                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2290                         if (memcmp(h->drv[i]->LunID, lunid,
2291                                 sizeof(lunid)) == 0) {
2292                                 drv_found = 1;
2293                                 break;
2294                         }
2295                 }
2296                 if (!drv_found) {
2297                         /* Deregister it from the OS, it's gone. */
2298                         spin_lock_irqsave(&h->lock, flags);
2299                         h->drv[i]->busy_configuring = 1;
2300                         spin_unlock_irqrestore(&h->lock, flags);
2301                         return_code = deregister_disk(h, i, 1, via_ioctl);
2302                         if (h->drv[i] != NULL)
2303                                 h->drv[i]->busy_configuring = 0;
2304                 }
2305         }
2306
2307         /* Compare controller drive array to driver's drive array.
2308          * Check for updates in the drive information and any new drives
2309          * on the controller due to ACU adding logical drives, or changing
2310          * a logical drive's size, etc.  Reregister any new/changed drives
2311          */
2312         for (i = 0; i < num_luns; i++) {
2313                 int j;
2314
2315                 drv_found = 0;
2316
2317                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2318                 /* Find if the LUN is already in the drive array
2319                  * of the driver.  If so then update its info
2320                  * if not in use.  If it does not exist then find
2321                  * the first free index and add it.
2322                  */
2323                 for (j = 0; j <= h->highest_lun; j++) {
2324                         if (h->drv[j] != NULL &&
2325                                 memcmp(h->drv[j]->LunID, lunid,
2326                                         sizeof(h->drv[j]->LunID)) == 0) {
2327                                 drv_index = j;
2328                                 drv_found = 1;
2329                                 break;
2330                         }
2331                 }
2332
2333                 /* check if the drive was found already in the array */
2334                 if (!drv_found) {
2335                         drv_index = cciss_add_gendisk(h, lunid, 0);
2336                         if (drv_index == -1)
2337                                 goto freeret;
2338                 }
2339                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2340         }               /* end for */
2341
2342 freeret:
2343         kfree(ld_buff);
2344         h->busy_configuring = 0;
2345         /* We return -1 here to tell the ACU that we have registered/updated
2346          * all of the drives that we can and to keep it from calling us
2347          * additional times.
2348          */
2349         return -1;
2350 mem_msg:
2351         dev_err(&h->pdev->dev, "out of memory\n");
2352         h->busy_configuring = 0;
2353         goto freeret;
2354 }
2355
2356 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2357 {
2358         /* zero out the disk size info */
2359         drive_info->nr_blocks = 0;
2360         drive_info->block_size = 0;
2361         drive_info->heads = 0;
2362         drive_info->sectors = 0;
2363         drive_info->cylinders = 0;
2364         drive_info->raid_level = -1;
2365         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2366         memset(drive_info->model, 0, sizeof(drive_info->model));
2367         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2368         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2369         /*
2370          * don't clear the LUNID though, we need to remember which
2371          * one this one is.
2372          */
2373 }
2374
2375 /* This function will deregister the disk and it's queue from the
2376  * kernel.  It must be called with the controller lock held and the
2377  * drv structures busy_configuring flag set.  It's parameters are:
2378  *
2379  * disk = This is the disk to be deregistered
2380  * drv  = This is the drive_info_struct associated with the disk to be
2381  *        deregistered.  It contains information about the disk used
2382  *        by the driver.
2383  * clear_all = This flag determines whether or not the disk information
2384  *             is going to be completely cleared out and the highest_lun
2385  *             reset.  Sometimes we want to clear out information about
2386  *             the disk in preparation for re-adding it.  In this case
2387  *             the highest_lun should be left unchanged and the LunID
2388  *             should not be cleared.
2389  * via_ioctl
2390  *    This indicates whether we've reached this path via ioctl.
2391  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2392  *    If this path is reached via ioctl(), then the max_usage_count will
2393  *    be 1, as the process calling ioctl() has got to have the device open.
2394  *    If we get here via sysfs, then the max usage count will be zero.
2395 */
2396 static int deregister_disk(ctlr_info_t *h, int drv_index,
2397                            int clear_all, int via_ioctl)
2398 {
2399         int i;
2400         struct gendisk *disk;
2401         drive_info_struct *drv;
2402         int recalculate_highest_lun;
2403
2404         if (!capable(CAP_SYS_RAWIO))
2405                 return -EPERM;
2406
2407         drv = h->drv[drv_index];
2408         disk = h->gendisk[drv_index];
2409
2410         /* make sure logical volume is NOT is use */
2411         if (clear_all || (h->gendisk[0] == disk)) {
2412                 if (drv->usage_count > via_ioctl)
2413                         return -EBUSY;
2414         } else if (drv->usage_count > 0)
2415                 return -EBUSY;
2416
2417         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2418
2419         /* invalidate the devices and deregister the disk.  If it is disk
2420          * zero do not deregister it but just zero out it's values.  This
2421          * allows us to delete disk zero but keep the controller registered.
2422          */
2423         if (h->gendisk[0] != disk) {
2424                 struct request_queue *q = disk->queue;
2425                 if (disk->flags & GENHD_FL_UP) {
2426                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2427                         del_gendisk(disk);
2428                 }
2429                 if (q)
2430                         blk_cleanup_queue(q);
2431                 /* If clear_all is set then we are deleting the logical
2432                  * drive, not just refreshing its info.  For drives
2433                  * other than disk 0 we will call put_disk.  We do not
2434                  * do this for disk 0 as we need it to be able to
2435                  * configure the controller.
2436                  */
2437                 if (clear_all){
2438                         /* This isn't pretty, but we need to find the
2439                          * disk in our array and NULL our the pointer.
2440                          * This is so that we will call alloc_disk if
2441                          * this index is used again later.
2442                          */
2443                         for (i=0; i < CISS_MAX_LUN; i++){
2444                                 if (h->gendisk[i] == disk) {
2445                                         h->gendisk[i] = NULL;
2446                                         break;
2447                                 }
2448                         }
2449                         put_disk(disk);
2450                 }
2451         } else {
2452                 set_capacity(disk, 0);
2453                 cciss_clear_drive_info(drv);
2454         }
2455
2456         --h->num_luns;
2457
2458         /* if it was the last disk, find the new hightest lun */
2459         if (clear_all && recalculate_highest_lun) {
2460                 int newhighest = -1;
2461                 for (i = 0; i <= h->highest_lun; i++) {
2462                         /* if the disk has size > 0, it is available */
2463                         if (h->drv[i] && h->drv[i]->heads)
2464                                 newhighest = i;
2465                 }
2466                 h->highest_lun = newhighest;
2467         }
2468         return 0;
2469 }
2470
2471 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2472                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2473                 int cmd_type)
2474 {
2475         u64bit buff_dma_handle;
2476         int status = IO_OK;
2477
2478         c->cmd_type = CMD_IOCTL_PEND;
2479         c->Header.ReplyQueue = 0;
2480         if (buff != NULL) {
2481                 c->Header.SGList = 1;
2482                 c->Header.SGTotal = 1;
2483         } else {
2484                 c->Header.SGList = 0;
2485                 c->Header.SGTotal = 0;
2486         }
2487         c->Header.Tag.lower = c->busaddr;
2488         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2489
2490         c->Request.Type.Type = cmd_type;
2491         if (cmd_type == TYPE_CMD) {
2492                 switch (cmd) {
2493                 case CISS_INQUIRY:
2494                         /* are we trying to read a vital product page */
2495                         if (page_code != 0) {
2496                                 c->Request.CDB[1] = 0x01;
2497                                 c->Request.CDB[2] = page_code;
2498                         }
2499                         c->Request.CDBLen = 6;
2500                         c->Request.Type.Attribute = ATTR_SIMPLE;
2501                         c->Request.Type.Direction = XFER_READ;
2502                         c->Request.Timeout = 0;
2503                         c->Request.CDB[0] = CISS_INQUIRY;
2504                         c->Request.CDB[4] = size & 0xFF;
2505                         break;
2506                 case CISS_REPORT_LOG:
2507                 case CISS_REPORT_PHYS:
2508                         /* Talking to controller so It's a physical command
2509                            mode = 00 target = 0.  Nothing to write.
2510                          */
2511                         c->Request.CDBLen = 12;
2512                         c->Request.Type.Attribute = ATTR_SIMPLE;
2513                         c->Request.Type.Direction = XFER_READ;
2514                         c->Request.Timeout = 0;
2515                         c->Request.CDB[0] = cmd;
2516                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2517                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2518                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2519                         c->Request.CDB[9] = size & 0xFF;
2520                         break;
2521
2522                 case CCISS_READ_CAPACITY:
2523                         c->Request.CDBLen = 10;
2524                         c->Request.Type.Attribute = ATTR_SIMPLE;
2525                         c->Request.Type.Direction = XFER_READ;
2526                         c->Request.Timeout = 0;
2527                         c->Request.CDB[0] = cmd;
2528                         break;
2529                 case CCISS_READ_CAPACITY_16:
2530                         c->Request.CDBLen = 16;
2531                         c->Request.Type.Attribute = ATTR_SIMPLE;
2532                         c->Request.Type.Direction = XFER_READ;
2533                         c->Request.Timeout = 0;
2534                         c->Request.CDB[0] = cmd;
2535                         c->Request.CDB[1] = 0x10;
2536                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2537                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2538                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2539                         c->Request.CDB[13] = size & 0xFF;
2540                         c->Request.Timeout = 0;
2541                         c->Request.CDB[0] = cmd;
2542                         break;
2543                 case CCISS_CACHE_FLUSH:
2544                         c->Request.CDBLen = 12;
2545                         c->Request.Type.Attribute = ATTR_SIMPLE;
2546                         c->Request.Type.Direction = XFER_WRITE;
2547                         c->Request.Timeout = 0;
2548                         c->Request.CDB[0] = BMIC_WRITE;
2549                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2550                         break;
2551                 case TEST_UNIT_READY:
2552                         c->Request.CDBLen = 6;
2553                         c->Request.Type.Attribute = ATTR_SIMPLE;
2554                         c->Request.Type.Direction = XFER_NONE;
2555                         c->Request.Timeout = 0;
2556                         break;
2557                 default:
2558                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2559                         return IO_ERROR;
2560                 }
2561         } else if (cmd_type == TYPE_MSG) {
2562                 switch (cmd) {
2563                 case 0: /* ABORT message */
2564                         c->Request.CDBLen = 12;
2565                         c->Request.Type.Attribute = ATTR_SIMPLE;
2566                         c->Request.Type.Direction = XFER_WRITE;
2567                         c->Request.Timeout = 0;
2568                         c->Request.CDB[0] = cmd;        /* abort */
2569                         c->Request.CDB[1] = 0;  /* abort a command */
2570                         /* buff contains the tag of the command to abort */
2571                         memcpy(&c->Request.CDB[4], buff, 8);
2572                         break;
2573                 case 1: /* RESET message */
2574                         c->Request.CDBLen = 16;
2575                         c->Request.Type.Attribute = ATTR_SIMPLE;
2576                         c->Request.Type.Direction = XFER_NONE;
2577                         c->Request.Timeout = 0;
2578                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2579                         c->Request.CDB[0] = cmd;        /* reset */
2580                         c->Request.CDB[1] = 0x03;       /* reset a target */
2581                         break;
2582                 case 3: /* No-Op message */
2583                         c->Request.CDBLen = 1;
2584                         c->Request.Type.Attribute = ATTR_SIMPLE;
2585                         c->Request.Type.Direction = XFER_WRITE;
2586                         c->Request.Timeout = 0;
2587                         c->Request.CDB[0] = cmd;
2588                         break;
2589                 default:
2590                         dev_warn(&h->pdev->dev,
2591                                 "unknown message type %d\n", cmd);
2592                         return IO_ERROR;
2593                 }
2594         } else {
2595                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2596                 return IO_ERROR;
2597         }
2598         /* Fill in the scatter gather information */
2599         if (size > 0) {
2600                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2601                                                              buff, size,
2602                                                              PCI_DMA_BIDIRECTIONAL);
2603                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2604                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2605                 c->SG[0].Len = size;
2606                 c->SG[0].Ext = 0;       /* we are not chaining */
2607         }
2608         return status;
2609 }
2610
2611 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2612 {
2613         switch (c->err_info->ScsiStatus) {
2614         case SAM_STAT_GOOD:
2615                 return IO_OK;
2616         case SAM_STAT_CHECK_CONDITION:
2617                 switch (0xf & c->err_info->SenseInfo[2]) {
2618                 case 0: return IO_OK; /* no sense */
2619                 case 1: return IO_OK; /* recovered error */
2620                 default:
2621                         if (check_for_unit_attention(h, c))
2622                                 return IO_NEEDS_RETRY;
2623                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2624                                 "check condition, sense key = 0x%02x\n",
2625                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2626                 }
2627                 break;
2628         default:
2629                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2630                         "scsi status = 0x%02x\n",
2631                         c->Request.CDB[0], c->err_info->ScsiStatus);
2632                 break;
2633         }
2634         return IO_ERROR;
2635 }
2636
2637 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2638 {
2639         int return_status = IO_OK;
2640
2641         if (c->err_info->CommandStatus == CMD_SUCCESS)
2642                 return IO_OK;
2643
2644         switch (c->err_info->CommandStatus) {
2645         case CMD_TARGET_STATUS:
2646                 return_status = check_target_status(h, c);
2647                 break;
2648         case CMD_DATA_UNDERRUN:
2649         case CMD_DATA_OVERRUN:
2650                 /* expected for inquiry and report lun commands */
2651                 break;
2652         case CMD_INVALID:
2653                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2654                        "reported invalid\n", c->Request.CDB[0]);
2655                 return_status = IO_ERROR;
2656                 break;
2657         case CMD_PROTOCOL_ERR:
2658                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2659                        "protocol error\n", c->Request.CDB[0]);
2660                 return_status = IO_ERROR;
2661                 break;
2662         case CMD_HARDWARE_ERR:
2663                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2664                        " hardware error\n", c->Request.CDB[0]);
2665                 return_status = IO_ERROR;
2666                 break;
2667         case CMD_CONNECTION_LOST:
2668                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2669                        "connection lost\n", c->Request.CDB[0]);
2670                 return_status = IO_ERROR;
2671                 break;
2672         case CMD_ABORTED:
2673                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2674                        "aborted\n", c->Request.CDB[0]);
2675                 return_status = IO_ERROR;
2676                 break;
2677         case CMD_ABORT_FAILED:
2678                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2679                        "abort failed\n", c->Request.CDB[0]);
2680                 return_status = IO_ERROR;
2681                 break;
2682         case CMD_UNSOLICITED_ABORT:
2683                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2684                         c->Request.CDB[0]);
2685                 return_status = IO_NEEDS_RETRY;
2686                 break;
2687         default:
2688                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2689                        "unknown status %x\n", c->Request.CDB[0],
2690                        c->err_info->CommandStatus);
2691                 return_status = IO_ERROR;
2692         }
2693         return return_status;
2694 }
2695
2696 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2697         int attempt_retry)
2698 {
2699         DECLARE_COMPLETION_ONSTACK(wait);
2700         u64bit buff_dma_handle;
2701         int return_status = IO_OK;
2702
2703 resend_cmd2:
2704         c->waiting = &wait;
2705         enqueue_cmd_and_start_io(h, c);
2706
2707         wait_for_completion(&wait);
2708
2709         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2710                 goto command_done;
2711
2712         return_status = process_sendcmd_error(h, c);
2713
2714         if (return_status == IO_NEEDS_RETRY &&
2715                 c->retry_count < MAX_CMD_RETRIES) {
2716                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2717                         c->Request.CDB[0]);
2718                 c->retry_count++;
2719                 /* erase the old error information */
2720                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2721                 return_status = IO_OK;
2722                 INIT_COMPLETION(wait);
2723                 goto resend_cmd2;
2724         }
2725
2726 command_done:
2727         /* unlock the buffers from DMA */
2728         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2729         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2730         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2731                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2732         return return_status;
2733 }
2734
2735 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2736                            __u8 page_code, unsigned char scsi3addr[],
2737                         int cmd_type)
2738 {
2739         CommandList_struct *c;
2740         int return_status;
2741
2742         c = cmd_special_alloc(h);
2743         if (!c)
2744                 return -ENOMEM;
2745         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2746                 scsi3addr, cmd_type);
2747         if (return_status == IO_OK)
2748                 return_status = sendcmd_withirq_core(h, c, 1);
2749
2750         cmd_special_free(h, c);
2751         return return_status;
2752 }
2753
2754 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2755                                    sector_t total_size,
2756                                    unsigned int block_size,
2757                                    InquiryData_struct *inq_buff,
2758                                    drive_info_struct *drv)
2759 {
2760         int return_code;
2761         unsigned long t;
2762         unsigned char scsi3addr[8];
2763
2764         memset(inq_buff, 0, sizeof(InquiryData_struct));
2765         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2766         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2767                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2768         if (return_code == IO_OK) {
2769                 if (inq_buff->data_byte[8] == 0xFF) {
2770                         dev_warn(&h->pdev->dev,
2771                                "reading geometry failed, volume "
2772                                "does not support reading geometry\n");
2773                         drv->heads = 255;
2774                         drv->sectors = 32;      /* Sectors per track */
2775                         drv->cylinders = total_size + 1;
2776                         drv->raid_level = RAID_UNKNOWN;
2777                 } else {
2778                         drv->heads = inq_buff->data_byte[6];
2779                         drv->sectors = inq_buff->data_byte[7];
2780                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2781                         drv->cylinders += inq_buff->data_byte[5];
2782                         drv->raid_level = inq_buff->data_byte[8];
2783                 }
2784                 drv->block_size = block_size;
2785                 drv->nr_blocks = total_size + 1;
2786                 t = drv->heads * drv->sectors;
2787                 if (t > 1) {
2788                         sector_t real_size = total_size + 1;
2789                         unsigned long rem = sector_div(real_size, t);
2790                         if (rem)
2791                                 real_size++;
2792                         drv->cylinders = real_size;
2793                 }
2794         } else {                /* Get geometry failed */
2795                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2796         }
2797 }
2798
2799 static void
2800 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2801                     unsigned int *block_size)
2802 {
2803         ReadCapdata_struct *buf;
2804         int return_code;
2805         unsigned char scsi3addr[8];
2806
2807         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2808         if (!buf) {
2809                 dev_warn(&h->pdev->dev, "out of memory\n");
2810                 return;
2811         }
2812
2813         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2814         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2815                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2816         if (return_code == IO_OK) {
2817                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2818                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2819         } else {                /* read capacity command failed */
2820                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2821                 *total_size = 0;
2822                 *block_size = BLOCK_SIZE;
2823         }
2824         kfree(buf);
2825 }
2826
2827 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2828         sector_t *total_size, unsigned int *block_size)
2829 {
2830         ReadCapdata_struct_16 *buf;
2831         int return_code;
2832         unsigned char scsi3addr[8];
2833
2834         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2835         if (!buf) {
2836                 dev_warn(&h->pdev->dev, "out of memory\n");
2837                 return;
2838         }
2839
2840         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2841         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2842                 buf, sizeof(ReadCapdata_struct_16),
2843                         0, scsi3addr, TYPE_CMD);
2844         if (return_code == IO_OK) {
2845                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2846                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2847         } else {                /* read capacity command failed */
2848                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2849                 *total_size = 0;
2850                 *block_size = BLOCK_SIZE;
2851         }
2852         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2853                (unsigned long long)*total_size+1, *block_size);
2854         kfree(buf);
2855 }
2856
2857 static int cciss_revalidate(struct gendisk *disk)
2858 {
2859         ctlr_info_t *h = get_host(disk);
2860         drive_info_struct *drv = get_drv(disk);
2861         int logvol;
2862         int FOUND = 0;
2863         unsigned int block_size;
2864         sector_t total_size;
2865         InquiryData_struct *inq_buff = NULL;
2866
2867         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2868                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2869                         sizeof(drv->LunID)) == 0) {
2870                         FOUND = 1;
2871                         break;
2872                 }
2873         }
2874
2875         if (!FOUND)
2876                 return 1;
2877
2878         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2879         if (inq_buff == NULL) {
2880                 dev_warn(&h->pdev->dev, "out of memory\n");
2881                 return 1;
2882         }
2883         if (h->cciss_read == CCISS_READ_10) {
2884                 cciss_read_capacity(h, logvol,
2885                                         &total_size, &block_size);
2886         } else {
2887                 cciss_read_capacity_16(h, logvol,
2888                                         &total_size, &block_size);
2889         }
2890         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2891                                inq_buff, drv);
2892
2893         blk_queue_logical_block_size(drv->queue, drv->block_size);
2894         set_capacity(disk, drv->nr_blocks);
2895
2896         kfree(inq_buff);
2897         return 0;
2898 }
2899
2900 /*
2901  * Map (physical) PCI mem into (virtual) kernel space
2902  */
2903 static void __iomem *remap_pci_mem(ulong base, ulong size)
2904 {
2905         ulong page_base = ((ulong) base) & PAGE_MASK;
2906         ulong page_offs = ((ulong) base) - page_base;
2907         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2908
2909         return page_remapped ? (page_remapped + page_offs) : NULL;
2910 }
2911
2912 /*
2913  * Takes jobs of the Q and sends them to the hardware, then puts it on
2914  * the Q to wait for completion.
2915  */
2916 static void start_io(ctlr_info_t *h)
2917 {
2918         CommandList_struct *c;
2919
2920         while (!hlist_empty(&h->reqQ)) {
2921                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2922                 /* can't do anything if fifo is full */
2923                 if ((h->access.fifo_full(h))) {
2924                         dev_warn(&h->pdev->dev, "fifo full\n");
2925                         break;
2926                 }
2927
2928                 /* Get the first entry from the Request Q */
2929                 removeQ(c);
2930                 h->Qdepth--;
2931
2932                 /* Tell the controller execute command */
2933                 h->access.submit_command(h, c);
2934
2935                 /* Put job onto the completed Q */
2936                 addQ(&h->cmpQ, c);
2937         }
2938 }
2939
2940 /* Assumes that h->lock is held. */
2941 /* Zeros out the error record and then resends the command back */
2942 /* to the controller */
2943 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2944 {
2945         /* erase the old error information */
2946         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2947
2948         /* add it to software queue and then send it to the controller */
2949         addQ(&h->reqQ, c);
2950         h->Qdepth++;
2951         if (h->Qdepth > h->maxQsinceinit)
2952                 h->maxQsinceinit = h->Qdepth;
2953
2954         start_io(h);
2955 }
2956
2957 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2958         unsigned int msg_byte, unsigned int host_byte,
2959         unsigned int driver_byte)
2960 {
2961         /* inverse of macros in scsi.h */
2962         return (scsi_status_byte & 0xff) |
2963                 ((msg_byte & 0xff) << 8) |
2964                 ((host_byte & 0xff) << 16) |
2965                 ((driver_byte & 0xff) << 24);
2966 }
2967
2968 static inline int evaluate_target_status(ctlr_info_t *h,
2969                         CommandList_struct *cmd, int *retry_cmd)
2970 {
2971         unsigned char sense_key;
2972         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2973         int error_value;
2974
2975         *retry_cmd = 0;
2976         /* If we get in here, it means we got "target status", that is, scsi status */
2977         status_byte = cmd->err_info->ScsiStatus;
2978         driver_byte = DRIVER_OK;
2979         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2980
2981         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2982                 host_byte = DID_PASSTHROUGH;
2983         else
2984                 host_byte = DID_OK;
2985
2986         error_value = make_status_bytes(status_byte, msg_byte,
2987                 host_byte, driver_byte);
2988
2989         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2990                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
2991                         dev_warn(&h->pdev->dev, "cmd %p "
2992                                "has SCSI Status 0x%x\n",
2993                                cmd, cmd->err_info->ScsiStatus);
2994                 return error_value;
2995         }
2996
2997         /* check the sense key */
2998         sense_key = 0xf & cmd->err_info->SenseInfo[2];
2999         /* no status or recovered error */
3000         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3001             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3002                 error_value = 0;
3003
3004         if (check_for_unit_attention(h, cmd)) {
3005                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3006                 return 0;
3007         }
3008
3009         /* Not SG_IO or similar? */
3010         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3011                 if (error_value != 0)
3012                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3013                                " sense key = 0x%x\n", cmd, sense_key);
3014                 return error_value;
3015         }
3016
3017         /* SG_IO or similar, copy sense data back */
3018         if (cmd->rq->sense) {
3019                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3020                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3021                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3022                         cmd->rq->sense_len);
3023         } else
3024                 cmd->rq->sense_len = 0;
3025
3026         return error_value;
3027 }
3028
3029 /* checks the status of the job and calls complete buffers to mark all
3030  * buffers for the completed job. Note that this function does not need
3031  * to hold the hba/queue lock.
3032  */
3033 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3034                                     int timeout)
3035 {
3036         int retry_cmd = 0;
3037         struct request *rq = cmd->rq;
3038
3039         rq->errors = 0;
3040
3041         if (timeout)
3042                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3043
3044         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3045                 goto after_error_processing;
3046
3047         switch (cmd->err_info->CommandStatus) {
3048         case CMD_TARGET_STATUS:
3049                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3050                 break;
3051         case CMD_DATA_UNDERRUN:
3052                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3053                         dev_warn(&h->pdev->dev, "cmd %p has"
3054                                " completed with data underrun "
3055                                "reported\n", cmd);
3056                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3057                 }
3058                 break;
3059         case CMD_DATA_OVERRUN:
3060                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3061                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3062                                " completed with data overrun "
3063                                "reported\n", cmd);
3064                 break;
3065         case CMD_INVALID:
3066                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3067                        "reported invalid\n", cmd);
3068                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3069                         cmd->err_info->CommandStatus, DRIVER_OK,
3070                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3071                                 DID_PASSTHROUGH : DID_ERROR);
3072                 break;
3073         case CMD_PROTOCOL_ERR:
3074                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3075                        "protocol error\n", cmd);
3076                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3077                         cmd->err_info->CommandStatus, DRIVER_OK,
3078                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3079                                 DID_PASSTHROUGH : DID_ERROR);
3080                 break;
3081         case CMD_HARDWARE_ERR:
3082                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3083                        " hardware error\n", cmd);
3084                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3085                         cmd->err_info->CommandStatus, DRIVER_OK,
3086                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3087                                 DID_PASSTHROUGH : DID_ERROR);
3088                 break;
3089         case CMD_CONNECTION_LOST:
3090                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3091                        "connection lost\n", cmd);
3092                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3093                         cmd->err_info->CommandStatus, DRIVER_OK,
3094                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3095                                 DID_PASSTHROUGH : DID_ERROR);
3096                 break;
3097         case CMD_ABORTED:
3098                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3099                        "aborted\n", cmd);
3100                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3101                         cmd->err_info->CommandStatus, DRIVER_OK,
3102                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3103                                 DID_PASSTHROUGH : DID_ABORT);
3104                 break;
3105         case CMD_ABORT_FAILED:
3106                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3107                        "abort failed\n", cmd);
3108                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3109                         cmd->err_info->CommandStatus, DRIVER_OK,
3110                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3111                                 DID_PASSTHROUGH : DID_ERROR);
3112                 break;
3113         case CMD_UNSOLICITED_ABORT:
3114                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3115                        "abort %p\n", h->ctlr, cmd);
3116                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3117                         retry_cmd = 1;
3118                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3119                         cmd->retry_count++;
3120                 } else
3121                         dev_warn(&h->pdev->dev,
3122                                 "%p retried too many times\n", cmd);
3123                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3124                         cmd->err_info->CommandStatus, DRIVER_OK,
3125                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3126                                 DID_PASSTHROUGH : DID_ABORT);
3127                 break;
3128         case CMD_TIMEOUT:
3129                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3130                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3131                         cmd->err_info->CommandStatus, DRIVER_OK,
3132                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3133                                 DID_PASSTHROUGH : DID_ERROR);
3134                 break;
3135         default:
3136                 dev_warn(&h->pdev->dev, "cmd %p returned "
3137                        "unknown status %x\n", cmd,
3138                        cmd->err_info->CommandStatus);
3139                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3140                         cmd->err_info->CommandStatus, DRIVER_OK,
3141                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3142                                 DID_PASSTHROUGH : DID_ERROR);
3143         }
3144
3145 after_error_processing:
3146
3147         /* We need to return this command */
3148         if (retry_cmd) {
3149                 resend_cciss_cmd(h, cmd);
3150                 return;
3151         }
3152         cmd->rq->completion_data = cmd;
3153         blk_complete_request(cmd->rq);
3154 }
3155
3156 static inline u32 cciss_tag_contains_index(u32 tag)
3157 {
3158 #define DIRECT_LOOKUP_BIT 0x10
3159         return tag & DIRECT_LOOKUP_BIT;
3160 }
3161
3162 static inline u32 cciss_tag_to_index(u32 tag)
3163 {
3164 #define DIRECT_LOOKUP_SHIFT 5
3165         return tag >> DIRECT_LOOKUP_SHIFT;
3166 }
3167
3168 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3169 {
3170 #define CCISS_ERROR_BITS 0x03
3171         return tag & ~CCISS_ERROR_BITS;
3172 }
3173
3174 static inline void cciss_mark_tag_indexed(u32 *tag)
3175 {
3176         *tag |= DIRECT_LOOKUP_BIT;
3177 }
3178
3179 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3180 {
3181         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3182 }
3183
3184 /*
3185  * Get a request and submit it to the controller.
3186  */
3187 static void do_cciss_request(struct request_queue *q)
3188 {
3189         ctlr_info_t *h = q->queuedata;
3190         CommandList_struct *c;
3191         sector_t start_blk;
3192         int seg;
3193         struct request *creq;
3194         u64bit temp64;
3195         struct scatterlist *tmp_sg;
3196         SGDescriptor_struct *curr_sg;
3197         drive_info_struct *drv;
3198         int i, dir;
3199         int sg_index = 0;
3200         int chained = 0;
3201
3202         /* We call start_io here in case there is a command waiting on the
3203          * queue that has not been sent.
3204          */
3205         if (blk_queue_plugged(q))
3206                 goto startio;
3207
3208       queue:
3209         creq = blk_peek_request(q);
3210         if (!creq)
3211                 goto startio;
3212
3213         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3214
3215         c = cmd_alloc(h);
3216         if (!c)
3217                 goto full;
3218
3219         blk_start_request(creq);
3220
3221         tmp_sg = h->scatter_list[c->cmdindex];
3222         spin_unlock_irq(q->queue_lock);
3223
3224         c->cmd_type = CMD_RWREQ;
3225         c->rq = creq;
3226
3227         /* fill in the request */
3228         drv = creq->rq_disk->private_data;
3229         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3230         /* got command from pool, so use the command block index instead */
3231         /* for direct lookups. */
3232         /* The first 2 bits are reserved for controller error reporting. */
3233         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3234         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3235         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3236         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3237         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3238         c->Request.Type.Attribute = ATTR_SIMPLE;
3239         c->Request.Type.Direction =
3240             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3241         c->Request.Timeout = 0; /* Don't time out */
3242         c->Request.CDB[0] =
3243             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3244         start_blk = blk_rq_pos(creq);
3245         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3246                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3247         sg_init_table(tmp_sg, h->maxsgentries);
3248         seg = blk_rq_map_sg(q, creq, tmp_sg);
3249
3250         /* get the DMA records for the setup */
3251         if (c->Request.Type.Direction == XFER_READ)
3252                 dir = PCI_DMA_FROMDEVICE;
3253         else
3254                 dir = PCI_DMA_TODEVICE;
3255
3256         curr_sg = c->SG;
3257         sg_index = 0;
3258         chained = 0;
3259
3260         for (i = 0; i < seg; i++) {
3261                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3262                         !chained && ((seg - i) > 1)) {
3263                         /* Point to next chain block. */
3264                         curr_sg = h->cmd_sg_list[c->cmdindex];
3265                         sg_index = 0;
3266                         chained = 1;
3267                 }
3268                 curr_sg[sg_index].Len = tmp_sg[i].length;
3269                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3270                                                 tmp_sg[i].offset,
3271                                                 tmp_sg[i].length, dir);
3272                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3273                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3274                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3275                 ++sg_index;
3276         }
3277         if (chained)
3278                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3279                         (seg - (h->max_cmd_sgentries - 1)) *
3280                                 sizeof(SGDescriptor_struct));
3281
3282         /* track how many SG entries we are using */
3283         if (seg > h->maxSG)
3284                 h->maxSG = seg;
3285
3286         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3287                         "chained[%d]\n",
3288                         blk_rq_sectors(creq), seg, chained);
3289
3290         c->Header.SGTotal = seg + chained;
3291         if (seg <= h->max_cmd_sgentries)
3292                 c->Header.SGList = c->Header.SGTotal;
3293         else
3294                 c->Header.SGList = h->max_cmd_sgentries;
3295         set_performant_mode(h, c);
3296
3297         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3298                 if(h->cciss_read == CCISS_READ_10) {
3299                         c->Request.CDB[1] = 0;
3300                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3301                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3302                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3303                         c->Request.CDB[5] = start_blk & 0xff;
3304                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3305                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3306                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3307                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3308                 } else {
3309                         u32 upper32 = upper_32_bits(start_blk);
3310
3311                         c->Request.CDBLen = 16;
3312                         c->Request.CDB[1]= 0;
3313                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3314                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3315                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3316                         c->Request.CDB[5]= upper32 & 0xff;
3317                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3318                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3319                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3320                         c->Request.CDB[9]= start_blk & 0xff;
3321                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3322                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3323                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3324                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3325                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3326                 }
3327         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3328                 c->Request.CDBLen = creq->cmd_len;
3329                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3330         } else {
3331                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3332                         creq->cmd_type);
3333                 BUG();
3334         }
3335
3336         spin_lock_irq(q->queue_lock);
3337
3338         addQ(&h->reqQ, c);
3339         h->Qdepth++;
3340         if (h->Qdepth > h->maxQsinceinit)
3341                 h->maxQsinceinit = h->Qdepth;
3342
3343         goto queue;
3344 full:
3345         blk_stop_queue(q);
3346 startio:
3347         /* We will already have the driver lock here so not need
3348          * to lock it.
3349          */
3350         start_io(h);
3351 }
3352
3353 static inline unsigned long get_next_completion(ctlr_info_t *h)
3354 {
3355         return h->access.command_completed(h);
3356 }
3357
3358 static inline int interrupt_pending(ctlr_info_t *h)
3359 {
3360         return h->access.intr_pending(h);
3361 }
3362
3363 static inline long interrupt_not_for_us(ctlr_info_t *h)
3364 {
3365         return ((h->access.intr_pending(h) == 0) ||
3366                 (h->interrupts_enabled == 0));
3367 }
3368
3369 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3370                         u32 raw_tag)
3371 {
3372         if (unlikely(tag_index >= h->nr_cmds)) {
3373                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3374                 return 1;
3375         }
3376         return 0;
3377 }
3378
3379 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3380                                 u32 raw_tag)
3381 {
3382         removeQ(c);
3383         if (likely(c->cmd_type == CMD_RWREQ))
3384                 complete_command(h, c, 0);
3385         else if (c->cmd_type == CMD_IOCTL_PEND)
3386                 complete(c->waiting);
3387 #ifdef CONFIG_CISS_SCSI_TAPE
3388         else if (c->cmd_type == CMD_SCSI)
3389                 complete_scsi_command(c, 0, raw_tag);
3390 #endif
3391 }
3392
3393 static inline u32 next_command(ctlr_info_t *h)
3394 {
3395         u32 a;
3396
3397         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3398                 return h->access.command_completed(h);
3399
3400         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3401                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3402                 (h->reply_pool_head)++;
3403                 h->commands_outstanding--;
3404         } else {
3405                 a = FIFO_EMPTY;
3406         }
3407         /* Check for wraparound */
3408         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3409                 h->reply_pool_head = h->reply_pool;
3410                 h->reply_pool_wraparound ^= 1;
3411         }
3412         return a;
3413 }
3414
3415 /* process completion of an indexed ("direct lookup") command */
3416 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3417 {
3418         u32 tag_index;
3419         CommandList_struct *c;
3420
3421         tag_index = cciss_tag_to_index(raw_tag);
3422         if (bad_tag(h, tag_index, raw_tag))
3423                 return next_command(h);
3424         c = h->cmd_pool + tag_index;
3425         finish_cmd(h, c, raw_tag);
3426         return next_command(h);
3427 }
3428
3429 /* process completion of a non-indexed command */
3430 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3431 {
3432         u32 tag;
3433         CommandList_struct *c = NULL;
3434         struct hlist_node *tmp;
3435         __u32 busaddr_masked, tag_masked;
3436
3437         tag = cciss_tag_discard_error_bits(raw_tag);
3438         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3439                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3440                 tag_masked = cciss_tag_discard_error_bits(tag);
3441                 if (busaddr_masked == tag_masked) {
3442                         finish_cmd(h, c, raw_tag);
3443                         return next_command(h);
3444                 }
3445         }
3446         bad_tag(h, h->nr_cmds + 1, raw_tag);
3447         return next_command(h);
3448 }
3449
3450 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3451 {
3452         ctlr_info_t *h = dev_id;
3453         unsigned long flags;
3454         u32 raw_tag;
3455
3456         if (interrupt_not_for_us(h))
3457                 return IRQ_NONE;
3458         spin_lock_irqsave(&h->lock, flags);
3459         while (interrupt_pending(h)) {
3460                 raw_tag = get_next_completion(h);
3461                 while (raw_tag != FIFO_EMPTY) {
3462                         if (cciss_tag_contains_index(raw_tag))
3463                                 raw_tag = process_indexed_cmd(h, raw_tag);
3464                         else
3465                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3466                 }
3467         }
3468         spin_unlock_irqrestore(&h->lock, flags);
3469         return IRQ_HANDLED;
3470 }
3471
3472 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3473  * check the interrupt pending register because it is not set.
3474  */
3475 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3476 {
3477         ctlr_info_t *h = dev_id;
3478         unsigned long flags;
3479         u32 raw_tag;
3480
3481         spin_lock_irqsave(&h->lock, flags);
3482         raw_tag = get_next_completion(h);
3483         while (raw_tag != FIFO_EMPTY) {
3484                 if (cciss_tag_contains_index(raw_tag))
3485                         raw_tag = process_indexed_cmd(h, raw_tag);
3486                 else
3487                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3488         }
3489         spin_unlock_irqrestore(&h->lock, flags);
3490         return IRQ_HANDLED;
3491 }
3492
3493 /**
3494  * add_to_scan_list() - add controller to rescan queue
3495  * @h:                Pointer to the controller.
3496  *
3497  * Adds the controller to the rescan queue if not already on the queue.
3498  *
3499  * returns 1 if added to the queue, 0 if skipped (could be on the
3500  * queue already, or the controller could be initializing or shutting
3501  * down).
3502  **/
3503 static int add_to_scan_list(struct ctlr_info *h)
3504 {
3505         struct ctlr_info *test_h;
3506         int found = 0;
3507         int ret = 0;
3508
3509         if (h->busy_initializing)
3510                 return 0;
3511
3512         if (!mutex_trylock(&h->busy_shutting_down))
3513                 return 0;
3514
3515         mutex_lock(&scan_mutex);
3516         list_for_each_entry(test_h, &scan_q, scan_list) {
3517                 if (test_h == h) {
3518                         found = 1;
3519                         break;
3520                 }
3521         }
3522         if (!found && !h->busy_scanning) {
3523                 INIT_COMPLETION(h->scan_wait);
3524                 list_add_tail(&h->scan_list, &scan_q);
3525                 ret = 1;
3526         }
3527         mutex_unlock(&scan_mutex);
3528         mutex_unlock(&h->busy_shutting_down);
3529
3530         return ret;
3531 }
3532
3533 /**
3534  * remove_from_scan_list() - remove controller from rescan queue
3535  * @h:                     Pointer to the controller.
3536  *
3537  * Removes the controller from the rescan queue if present. Blocks if
3538  * the controller is currently conducting a rescan.  The controller
3539  * can be in one of three states:
3540  * 1. Doesn't need a scan
3541  * 2. On the scan list, but not scanning yet (we remove it)
3542  * 3. Busy scanning (and not on the list). In this case we want to wait for
3543  *    the scan to complete to make sure the scanning thread for this
3544  *    controller is completely idle.
3545  **/
3546 static void remove_from_scan_list(struct ctlr_info *h)
3547 {
3548         struct ctlr_info *test_h, *tmp_h;
3549
3550         mutex_lock(&scan_mutex);
3551         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3552                 if (test_h == h) { /* state 2. */
3553                         list_del(&h->scan_list);
3554                         complete_all(&h->scan_wait);
3555                         mutex_unlock(&scan_mutex);
3556                         return;
3557                 }
3558         }
3559         if (h->busy_scanning) { /* state 3. */
3560                 mutex_unlock(&scan_mutex);
3561                 wait_for_completion(&h->scan_wait);
3562         } else { /* state 1, nothing to do. */
3563                 mutex_unlock(&scan_mutex);
3564         }
3565 }
3566
3567 /**
3568  * scan_thread() - kernel thread used to rescan controllers
3569  * @data:        Ignored.
3570  *
3571  * A kernel thread used scan for drive topology changes on
3572  * controllers. The thread processes only one controller at a time
3573  * using a queue.  Controllers are added to the queue using
3574  * add_to_scan_list() and removed from the queue either after done
3575  * processing or using remove_from_scan_list().
3576  *
3577  * returns 0.
3578  **/
3579 static int scan_thread(void *data)
3580 {
3581         struct ctlr_info *h;
3582
3583         while (1) {
3584                 set_current_state(TASK_INTERRUPTIBLE);
3585                 schedule();
3586                 if (kthread_should_stop())
3587                         break;
3588
3589                 while (1) {
3590                         mutex_lock(&scan_mutex);
3591                         if (list_empty(&scan_q)) {
3592                                 mutex_unlock(&scan_mutex);
3593                                 break;
3594                         }
3595
3596                         h = list_entry(scan_q.next,
3597                                        struct ctlr_info,
3598                                        scan_list);
3599                         list_del(&h->scan_list);
3600                         h->busy_scanning = 1;
3601                         mutex_unlock(&scan_mutex);
3602
3603                         rebuild_lun_table(h, 0, 0);
3604                         complete_all(&h->scan_wait);
3605                         mutex_lock(&scan_mutex);
3606                         h->busy_scanning = 0;
3607                         mutex_unlock(&scan_mutex);
3608                 }
3609         }
3610
3611         return 0;
3612 }
3613
3614 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3615 {
3616         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3617                 return 0;
3618
3619         switch (c->err_info->SenseInfo[12]) {
3620         case STATE_CHANGED:
3621                 dev_warn(&h->pdev->dev, "a state change "
3622                         "detected, command retried\n");
3623                 return 1;
3624         break;
3625         case LUN_FAILED:
3626                 dev_warn(&h->pdev->dev, "LUN failure "
3627                         "detected, action required\n");
3628                 return 1;
3629         break;
3630         case REPORT_LUNS_CHANGED:
3631                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3632         /*
3633          * Here, we could call add_to_scan_list and wake up the scan thread,
3634          * except that it's quite likely that we will get more than one
3635          * REPORT_LUNS_CHANGED condition in quick succession, which means
3636          * that those which occur after the first one will likely happen
3637          * *during* the scan_thread's rescan.  And the rescan code is not
3638          * robust enough to restart in the middle, undoing what it has already
3639          * done, and it's not clear that it's even possible to do this, since
3640          * part of what it does is notify the block layer, which starts
3641          * doing it's own i/o to read partition tables and so on, and the
3642          * driver doesn't have visibility to know what might need undoing.
3643          * In any event, if possible, it is horribly complicated to get right
3644          * so we just don't do it for now.
3645          *
3646          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3647          */
3648                 return 1;
3649         break;
3650         case POWER_OR_RESET:
3651                 dev_warn(&h->pdev->dev,
3652                         "a power on or device reset detected\n");
3653                 return 1;
3654         break;
3655         case UNIT_ATTENTION_CLEARED:
3656                 dev_warn(&h->pdev->dev,
3657                         "unit attention cleared by another initiator\n");
3658                 return 1;
3659         break;
3660         default:
3661                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3662                 return 1;
3663         }
3664 }
3665
3666 /*
3667  *  We cannot read the structure directly, for portability we must use
3668  *   the io functions.
3669  *   This is for debug only.
3670  */
3671 static void print_cfg_table(ctlr_info_t *h)
3672 {
3673         int i;
3674         char temp_name[17];
3675         CfgTable_struct *tb = h->cfgtable;
3676
3677         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3678         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3679         for (i = 0; i < 4; i++)
3680                 temp_name[i] = readb(&(tb->Signature[i]));
3681         temp_name[4] = '\0';
3682         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3683         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3684                 readl(&(tb->SpecValence)));
3685         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3686                readl(&(tb->TransportSupport)));
3687         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3688                readl(&(tb->TransportActive)));
3689         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3690                readl(&(tb->HostWrite.TransportRequest)));
3691         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3692                readl(&(tb->HostWrite.CoalIntDelay)));
3693         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3694                readl(&(tb->HostWrite.CoalIntCount)));
3695         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3696                readl(&(tb->CmdsOutMax)));
3697         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3698                 readl(&(tb->BusTypes)));
3699         for (i = 0; i < 16; i++)
3700                 temp_name[i] = readb(&(tb->ServerName[i]));
3701         temp_name[16] = '\0';
3702         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3703         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3704                 readl(&(tb->HeartBeat)));
3705 }
3706
3707 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3708 {
3709         int i, offset, mem_type, bar_type;
3710         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3711                 return 0;
3712         offset = 0;
3713         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3714                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3715                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3716                         offset += 4;
3717                 else {
3718                         mem_type = pci_resource_flags(pdev, i) &
3719                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3720                         switch (mem_type) {
3721                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3722                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3723                                 offset += 4;    /* 32 bit */
3724                                 break;
3725                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3726                                 offset += 8;
3727                                 break;
3728                         default:        /* reserved in PCI 2.2 */
3729                                 dev_warn(&pdev->dev,
3730                                        "Base address is invalid\n");
3731                                 return -1;
3732                                 break;
3733                         }
3734                 }
3735                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3736                         return i + 1;
3737         }
3738         return -1;
3739 }
3740
3741 /* Fill in bucket_map[], given nsgs (the max number of
3742  * scatter gather elements supported) and bucket[],
3743  * which is an array of 8 integers.  The bucket[] array
3744  * contains 8 different DMA transfer sizes (in 16
3745  * byte increments) which the controller uses to fetch
3746  * commands.  This function fills in bucket_map[], which
3747  * maps a given number of scatter gather elements to one of
3748  * the 8 DMA transfer sizes.  The point of it is to allow the
3749  * controller to only do as much DMA as needed to fetch the
3750  * command, with the DMA transfer size encoded in the lower
3751  * bits of the command address.
3752  */
3753 static void  calc_bucket_map(int bucket[], int num_buckets,
3754         int nsgs, int *bucket_map)
3755 {
3756         int i, j, b, size;
3757
3758         /* even a command with 0 SGs requires 4 blocks */
3759 #define MINIMUM_TRANSFER_BLOCKS 4
3760 #define NUM_BUCKETS 8
3761         /* Note, bucket_map must have nsgs+1 entries. */
3762         for (i = 0; i <= nsgs; i++) {
3763                 /* Compute size of a command with i SG entries */
3764                 size = i + MINIMUM_TRANSFER_BLOCKS;
3765                 b = num_buckets; /* Assume the biggest bucket */
3766                 /* Find the bucket that is just big enough */
3767                 for (j = 0; j < 8; j++) {
3768                         if (bucket[j] >= size) {
3769                                 b = j;
3770                                 break;
3771                         }
3772                 }
3773                 /* for a command with i SG entries, use bucket b. */
3774                 bucket_map[i] = b;
3775         }
3776 }
3777
3778 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3779 {
3780         int i;
3781
3782         /* under certain very rare conditions, this can take awhile.
3783          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3784          * as we enter this code.) */
3785         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3786                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3787                         break;
3788                 msleep(10);
3789         }
3790 }
3791
3792 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3793 {
3794         /* This is a bit complicated.  There are 8 registers on
3795          * the controller which we write to to tell it 8 different
3796          * sizes of commands which there may be.  It's a way of
3797          * reducing the DMA done to fetch each command.  Encoded into
3798          * each command's tag are 3 bits which communicate to the controller
3799          * which of the eight sizes that command fits within.  The size of
3800          * each command depends on how many scatter gather entries there are.
3801          * Each SG entry requires 16 bytes.  The eight registers are programmed
3802          * with the number of 16-byte blocks a command of that size requires.
3803          * The smallest command possible requires 5 such 16 byte blocks.
3804          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3805          * blocks.  Note, this only extends to the SG entries contained
3806          * within the command block, and does not extend to chained blocks
3807          * of SG elements.   bft[] contains the eight values we write to
3808          * the registers.  They are not evenly distributed, but have more
3809          * sizes for small commands, and fewer sizes for larger commands.
3810          */
3811         __u32 trans_offset;
3812         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3813                         /*
3814                          *  5 = 1 s/g entry or 4k
3815                          *  6 = 2 s/g entry or 8k
3816                          *  8 = 4 s/g entry or 16k
3817                          * 10 = 6 s/g entry or 24k
3818                          */
3819         unsigned long register_value;
3820         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3821
3822         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3823
3824         /* Controller spec: zero out this buffer. */
3825         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3826         h->reply_pool_head = h->reply_pool;
3827
3828         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3829         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3830                                 h->blockFetchTable);
3831         writel(bft[0], &h->transtable->BlockFetch0);
3832         writel(bft[1], &h->transtable->BlockFetch1);
3833         writel(bft[2], &h->transtable->BlockFetch2);
3834         writel(bft[3], &h->transtable->BlockFetch3);
3835         writel(bft[4], &h->transtable->BlockFetch4);
3836         writel(bft[5], &h->transtable->BlockFetch5);
3837         writel(bft[6], &h->transtable->BlockFetch6);
3838         writel(bft[7], &h->transtable->BlockFetch7);
3839
3840         /* size of controller ring buffer */
3841         writel(h->max_commands, &h->transtable->RepQSize);
3842         writel(1, &h->transtable->RepQCount);
3843         writel(0, &h->transtable->RepQCtrAddrLow32);
3844         writel(0, &h->transtable->RepQCtrAddrHigh32);
3845         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3846         writel(0, &h->transtable->RepQAddr0High32);
3847         writel(CFGTBL_Trans_Performant,
3848                         &(h->cfgtable->HostWrite.TransportRequest));
3849
3850         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3851         cciss_wait_for_mode_change_ack(h);
3852         register_value = readl(&(h->cfgtable->TransportActive));
3853         if (!(register_value & CFGTBL_Trans_Performant))
3854                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3855                                         " performant mode\n");
3856 }
3857
3858 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3859 {
3860         __u32 trans_support;
3861
3862         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3863         /* Attempt to put controller into performant mode if supported */
3864         /* Does board support performant mode? */
3865         trans_support = readl(&(h->cfgtable->TransportSupport));
3866         if (!(trans_support & PERFORMANT_MODE))
3867                 return;
3868
3869         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3870         /* Performant mode demands commands on a 32 byte boundary
3871          * pci_alloc_consistent aligns on page boundarys already.
3872          * Just need to check if divisible by 32
3873          */
3874         if ((sizeof(CommandList_struct) % 32) != 0) {
3875                 dev_warn(&h->pdev->dev, "%s %d %s\n",
3876                         "cciss info: command size[",
3877                         (int)sizeof(CommandList_struct),
3878                         "] not divisible by 32, no performant mode..\n");
3879                 return;
3880         }
3881
3882         /* Performant mode ring buffer and supporting data structures */
3883         h->reply_pool = (__u64 *)pci_alloc_consistent(
3884                 h->pdev, h->max_commands * sizeof(__u64),
3885                 &(h->reply_pool_dhandle));
3886
3887         /* Need a block fetch table for performant mode */
3888         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3889                 sizeof(__u32)), GFP_KERNEL);
3890
3891         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3892                 goto clean_up;
3893
3894         cciss_enter_performant_mode(h);
3895
3896         /* Change the access methods to the performant access methods */
3897         h->access = SA5_performant_access;
3898         h->transMethod = CFGTBL_Trans_Performant;
3899
3900         return;
3901 clean_up:
3902         kfree(h->blockFetchTable);
3903         if (h->reply_pool)
3904                 pci_free_consistent(h->pdev,
3905                                 h->max_commands * sizeof(__u64),
3906                                 h->reply_pool,
3907                                 h->reply_pool_dhandle);
3908         return;
3909
3910 } /* cciss_put_controller_into_performant_mode */
3911
3912 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3913  * controllers that are capable. If not, we use IO-APIC mode.
3914  */
3915
3916 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3917 {
3918 #ifdef CONFIG_PCI_MSI
3919         int err;
3920         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3921         {0, 2}, {0, 3}
3922         };
3923
3924         /* Some boards advertise MSI but don't really support it */
3925         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3926             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3927                 goto default_int_mode;
3928
3929         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3930                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3931                 if (!err) {
3932                         h->intr[0] = cciss_msix_entries[0].vector;
3933                         h->intr[1] = cciss_msix_entries[1].vector;
3934                         h->intr[2] = cciss_msix_entries[2].vector;
3935                         h->intr[3] = cciss_msix_entries[3].vector;
3936                         h->msix_vector = 1;
3937                         return;
3938                 }
3939                 if (err > 0) {
3940                         dev_warn(&h->pdev->dev,
3941                                 "only %d MSI-X vectors available\n", err);
3942                         goto default_int_mode;
3943                 } else {
3944                         dev_warn(&h->pdev->dev,
3945                                 "MSI-X init failed %d\n", err);
3946                         goto default_int_mode;
3947                 }
3948         }
3949         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3950                 if (!pci_enable_msi(h->pdev))
3951                         h->msi_vector = 1;
3952                 else
3953                         dev_warn(&h->pdev->dev, "MSI init failed\n");
3954         }
3955 default_int_mode:
3956 #endif                          /* CONFIG_PCI_MSI */
3957         /* if we get here we're going to use the default interrupt mode */
3958         h->intr[PERF_MODE_INT] = h->pdev->irq;
3959         return;
3960 }
3961
3962 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3963 {
3964         int i;
3965         u32 subsystem_vendor_id, subsystem_device_id;
3966
3967         subsystem_vendor_id = pdev->subsystem_vendor;
3968         subsystem_device_id = pdev->subsystem_device;
3969         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3970                         subsystem_vendor_id;
3971
3972         for (i = 0; i < ARRAY_SIZE(products); i++) {
3973                 /* Stand aside for hpsa driver on request */
3974                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3975                         return -ENODEV;
3976                 if (*board_id == products[i].board_id)
3977                         return i;
3978         }
3979         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
3980                 *board_id);
3981         return -ENODEV;
3982 }
3983
3984 static inline bool cciss_board_disabled(ctlr_info_t *h)
3985 {
3986         u16 command;
3987
3988         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
3989         return ((command & PCI_COMMAND_MEMORY) == 0);
3990 }
3991
3992 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
3993         unsigned long *memory_bar)
3994 {
3995         int i;
3996
3997         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
3998                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
3999                         /* addressing mode bits already removed */
4000                         *memory_bar = pci_resource_start(pdev, i);
4001                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4002                                 *memory_bar);
4003                         return 0;
4004                 }
4005         dev_warn(&pdev->dev, "no memory BAR found\n");
4006         return -ENODEV;
4007 }
4008
4009 static int __devinit cciss_wait_for_board_ready(ctlr_info_t *h)
4010 {
4011         int i;
4012         u32 scratchpad;
4013
4014         for (i = 0; i < CCISS_BOARD_READY_ITERATIONS; i++) {
4015                 scratchpad = readl(h->vaddr + SA5_SCRATCHPAD_OFFSET);
4016                 if (scratchpad == CCISS_FIRMWARE_READY)
4017                         return 0;
4018                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4019         }
4020         dev_warn(&h->pdev->dev, "board not ready, timed out.\n");
4021         return -ENODEV;
4022 }
4023
4024 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4025         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4026         u64 *cfg_offset)
4027 {
4028         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4029         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4030         *cfg_base_addr &= (u32) 0x0000ffff;
4031         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4032         if (*cfg_base_addr_index == -1) {
4033                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4034                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4035                 return -ENODEV;
4036         }
4037         return 0;
4038 }
4039
4040 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4041 {
4042         u64 cfg_offset;
4043         u32 cfg_base_addr;
4044         u64 cfg_base_addr_index;
4045         u32 trans_offset;
4046         int rc;
4047
4048         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4049                 &cfg_base_addr_index, &cfg_offset);
4050         if (rc)
4051                 return rc;
4052         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4053                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4054         if (!h->cfgtable)
4055                 return -ENOMEM;
4056         /* Find performant mode table. */
4057         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4058         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4059                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4060                                 sizeof(*h->transtable));
4061         if (!h->transtable)
4062                 return -ENOMEM;
4063         return 0;
4064 }
4065
4066 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4067 {
4068         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4069         if (h->max_commands < 16) {
4070                 dev_warn(&h->pdev->dev, "Controller reports "
4071                         "max supported commands of %d, an obvious lie. "
4072                         "Using 16.  Ensure that firmware is up to date.\n",
4073                         h->max_commands);
4074                 h->max_commands = 16;
4075         }
4076 }
4077
4078 /* Interrogate the hardware for some limits:
4079  * max commands, max SG elements without chaining, and with chaining,
4080  * SG chain block size, etc.
4081  */
4082 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4083 {
4084         cciss_get_max_perf_mode_cmds(h);
4085         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4086         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4087         /*
4088          * Limit in-command s/g elements to 32 save dma'able memory.
4089          * Howvever spec says if 0, use 31
4090          */
4091         h->max_cmd_sgentries = 31;
4092         if (h->maxsgentries > 512) {
4093                 h->max_cmd_sgentries = 32;
4094                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4095                 h->maxsgentries--; /* save one for chain pointer */
4096         } else {
4097                 h->maxsgentries = 31; /* default to traditional values */
4098                 h->chainsize = 0;
4099         }
4100 }
4101
4102 static inline bool CISS_signature_present(ctlr_info_t *h)
4103 {
4104         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4105             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4106             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4107             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4108                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4109                 return false;
4110         }
4111         return true;
4112 }
4113
4114 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4115 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4116 {
4117 #ifdef CONFIG_X86
4118         u32 prefetch;
4119
4120         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4121         prefetch |= 0x100;
4122         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4123 #endif
4124 }
4125
4126 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4127  * in a prefetch beyond physical memory.
4128  */
4129 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4130 {
4131         u32 dma_prefetch;
4132         __u32 dma_refetch;
4133
4134         if (h->board_id != 0x3225103C)
4135                 return;
4136         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4137         dma_prefetch |= 0x8000;
4138         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4139         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4140         dma_refetch |= 0x1;
4141         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4142 }
4143
4144 static int __devinit cciss_pci_init(ctlr_info_t *h)
4145 {
4146         int prod_index, err;
4147
4148         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4149         if (prod_index < 0)
4150                 return -ENODEV;
4151         h->product_name = products[prod_index].product_name;
4152         h->access = *(products[prod_index].access);
4153
4154         if (cciss_board_disabled(h)) {
4155                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4156                 return -ENODEV;
4157         }
4158         err = pci_enable_device(h->pdev);
4159         if (err) {
4160                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4161                 return err;
4162         }
4163
4164         err = pci_request_regions(h->pdev, "cciss");
4165         if (err) {
4166                 dev_warn(&h->pdev->dev,
4167                         "Cannot obtain PCI resources, aborting\n");
4168                 return err;
4169         }
4170
4171         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4172         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4173
4174 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4175  * else we use the IO-APIC interrupt assigned to us by system ROM.
4176  */
4177         cciss_interrupt_mode(h);
4178         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4179         if (err)
4180                 goto err_out_free_res;
4181         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4182         if (!h->vaddr) {
4183                 err = -ENOMEM;
4184                 goto err_out_free_res;
4185         }
4186         err = cciss_wait_for_board_ready(h);
4187         if (err)
4188                 goto err_out_free_res;
4189         err = cciss_find_cfgtables(h);
4190         if (err)
4191                 goto err_out_free_res;
4192         print_cfg_table(h);
4193         cciss_find_board_params(h);
4194
4195         if (!CISS_signature_present(h)) {
4196                 err = -ENODEV;
4197                 goto err_out_free_res;
4198         }
4199         cciss_enable_scsi_prefetch(h);
4200         cciss_p600_dma_prefetch_quirk(h);
4201         cciss_put_controller_into_performant_mode(h);
4202         return 0;
4203
4204 err_out_free_res:
4205         /*
4206          * Deliberately omit pci_disable_device(): it does something nasty to
4207          * Smart Array controllers that pci_enable_device does not undo
4208          */
4209         if (h->transtable)
4210                 iounmap(h->transtable);
4211         if (h->cfgtable)
4212                 iounmap(h->cfgtable);
4213         if (h->vaddr)
4214                 iounmap(h->vaddr);
4215         pci_release_regions(h->pdev);
4216         return err;
4217 }
4218
4219 /* Function to find the first free pointer into our hba[] array
4220  * Returns -1 if no free entries are left.
4221  */
4222 static int alloc_cciss_hba(struct pci_dev *pdev)
4223 {
4224         int i;
4225
4226         for (i = 0; i < MAX_CTLR; i++) {
4227                 if (!hba[i]) {
4228                         ctlr_info_t *h;
4229
4230                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4231                         if (!h)
4232                                 goto Enomem;
4233                         hba[i] = h;
4234                         return i;
4235                 }
4236         }
4237         dev_warn(&pdev->dev, "This driver supports a maximum"
4238                " of %d controllers.\n", MAX_CTLR);
4239         return -1;
4240 Enomem:
4241         dev_warn(&pdev->dev, "out of memory.\n");
4242         return -1;
4243 }
4244
4245 static void free_hba(ctlr_info_t *h)
4246 {
4247         int i;
4248
4249         hba[h->ctlr] = NULL;
4250         for (i = 0; i < h->highest_lun + 1; i++)
4251                 if (h->gendisk[i] != NULL)
4252                         put_disk(h->gendisk[i]);
4253         kfree(h);
4254 }
4255
4256 /* Send a message CDB to the firmware. */
4257 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4258 {
4259         typedef struct {
4260                 CommandListHeader_struct CommandHeader;
4261                 RequestBlock_struct Request;
4262                 ErrDescriptor_struct ErrorDescriptor;
4263         } Command;
4264         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4265         Command *cmd;
4266         dma_addr_t paddr64;
4267         uint32_t paddr32, tag;
4268         void __iomem *vaddr;
4269         int i, err;
4270
4271         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4272         if (vaddr == NULL)
4273                 return -ENOMEM;
4274
4275         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4276            CCISS commands, so they must be allocated from the lower 4GiB of
4277            memory. */
4278         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4279         if (err) {
4280                 iounmap(vaddr);
4281                 return -ENOMEM;
4282         }
4283
4284         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4285         if (cmd == NULL) {
4286                 iounmap(vaddr);
4287                 return -ENOMEM;
4288         }
4289
4290         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4291            although there's no guarantee, we assume that the address is at
4292            least 4-byte aligned (most likely, it's page-aligned). */
4293         paddr32 = paddr64;
4294
4295         cmd->CommandHeader.ReplyQueue = 0;
4296         cmd->CommandHeader.SGList = 0;
4297         cmd->CommandHeader.SGTotal = 0;
4298         cmd->CommandHeader.Tag.lower = paddr32;
4299         cmd->CommandHeader.Tag.upper = 0;
4300         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4301
4302         cmd->Request.CDBLen = 16;
4303         cmd->Request.Type.Type = TYPE_MSG;
4304         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4305         cmd->Request.Type.Direction = XFER_NONE;
4306         cmd->Request.Timeout = 0; /* Don't time out */
4307         cmd->Request.CDB[0] = opcode;
4308         cmd->Request.CDB[1] = type;
4309         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4310
4311         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4312         cmd->ErrorDescriptor.Addr.upper = 0;
4313         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4314
4315         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4316
4317         for (i = 0; i < 10; i++) {
4318                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4319                 if ((tag & ~3) == paddr32)
4320                         break;
4321                 schedule_timeout_uninterruptible(HZ);
4322         }
4323
4324         iounmap(vaddr);
4325
4326         /* we leak the DMA buffer here ... no choice since the controller could
4327            still complete the command. */
4328         if (i == 10) {
4329                 dev_err(&pdev->dev,
4330                         "controller message %02x:%02x timed out\n",
4331                         opcode, type);
4332                 return -ETIMEDOUT;
4333         }
4334
4335         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4336
4337         if (tag & 2) {
4338                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4339                         opcode, type);
4340                 return -EIO;
4341         }
4342
4343         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4344                 opcode, type);
4345         return 0;
4346 }
4347
4348 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4349 #define cciss_noop(p) cciss_message(p, 3, 0)
4350
4351 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4352 {
4353 /* the #defines are stolen from drivers/pci/msi.h. */
4354 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4355 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4356
4357         int pos;
4358         u16 control = 0;
4359
4360         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4361         if (pos) {
4362                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4363                 if (control & PCI_MSI_FLAGS_ENABLE) {
4364                         dev_info(&pdev->dev, "resetting MSI\n");
4365                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4366                 }
4367         }
4368
4369         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4370         if (pos) {
4371                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4372                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4373                         dev_info(&pdev->dev, "resetting MSI-X\n");
4374                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4375                 }
4376         }
4377
4378         return 0;
4379 }
4380
4381 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4382         void * __iomem vaddr, bool use_doorbell)
4383 {
4384         u16 pmcsr;
4385         int pos;
4386
4387         if (use_doorbell) {
4388                 /* For everything after the P600, the PCI power state method
4389                  * of resetting the controller doesn't work, so we have this
4390                  * other way using the doorbell register.
4391                  */
4392                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4393                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4394                 msleep(1000);
4395         } else { /* Try to do it the PCI power state way */
4396
4397                 /* Quoting from the Open CISS Specification: "The Power
4398                  * Management Control/Status Register (CSR) controls the power
4399                  * state of the device.  The normal operating state is D0,
4400                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4401                  * the controller, place the interface device in D3 then to D0,
4402                  * this causes a secondary PCI reset which will reset the
4403                  * controller." */
4404
4405                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4406                 if (pos == 0) {
4407                         dev_err(&pdev->dev,
4408                                 "cciss_controller_hard_reset: "
4409                                 "PCI PM not supported\n");
4410                         return -ENODEV;
4411                 }
4412                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4413                 /* enter the D3hot power management state */
4414                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4415                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4416                 pmcsr |= PCI_D3hot;
4417                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4418
4419                 msleep(500);
4420
4421                 /* enter the D0 power management state */
4422                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4423                 pmcsr |= PCI_D0;
4424                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4425
4426                 msleep(500);
4427         }
4428         return 0;
4429 }
4430
4431 /* This does a hard reset of the controller using PCI power management
4432  * states or using the doorbell register. */
4433 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4434 {
4435         u16 saved_config_space[32];
4436         u64 cfg_offset;
4437         u32 cfg_base_addr;
4438         u64 cfg_base_addr_index;
4439         void __iomem *vaddr;
4440         unsigned long paddr;
4441         u32 misc_fw_support, active_transport;
4442         int rc, i;
4443         CfgTable_struct __iomem *cfgtable;
4444         bool use_doorbell;
4445         u32 board_id;
4446
4447         /* For controllers as old a the p600, this is very nearly
4448          * the same thing as
4449          *
4450          * pci_save_state(pci_dev);
4451          * pci_set_power_state(pci_dev, PCI_D3hot);
4452          * pci_set_power_state(pci_dev, PCI_D0);
4453          * pci_restore_state(pci_dev);
4454          *
4455          * but we can't use these nice canned kernel routines on
4456          * kexec, because they also check the MSI/MSI-X state in PCI
4457          * configuration space and do the wrong thing when it is
4458          * set/cleared.  Also, the pci_save/restore_state functions
4459          * violate the ordering requirements for restoring the
4460          * configuration space from the CCISS document (see the
4461          * comment below).  So we roll our own ....
4462          *
4463          * For controllers newer than the P600, the pci power state
4464          * method of resetting doesn't work so we have another way
4465          * using the doorbell register.
4466          */
4467
4468         /* Exclude 640x boards.  These are two pci devices in one slot
4469          * which share a battery backed cache module.  One controls the
4470          * cache, the other accesses the cache through the one that controls
4471          * it.  If we reset the one controlling the cache, the other will
4472          * likely not be happy.  Just forbid resetting this conjoined mess.
4473          */
4474         cciss_lookup_board_id(pdev, &board_id);
4475         if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4476                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4477                                 "due to shared cache module.");
4478                 return -ENODEV;
4479         }
4480
4481         for (i = 0; i < 32; i++)
4482                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4483
4484         /* find the first memory BAR, so we can find the cfg table */
4485         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4486         if (rc)
4487                 return rc;
4488         vaddr = remap_pci_mem(paddr, 0x250);
4489         if (!vaddr)
4490                 return -ENOMEM;
4491
4492         /* find cfgtable in order to check if reset via doorbell is supported */
4493         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4494                                         &cfg_base_addr_index, &cfg_offset);
4495         if (rc)
4496                 goto unmap_vaddr;
4497         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4498                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4499         if (!cfgtable) {
4500                 rc = -ENOMEM;
4501                 goto unmap_vaddr;
4502         }
4503
4504         /* If reset via doorbell register is supported, use that. */
4505         misc_fw_support = readl(&cfgtable->misc_fw_support);
4506         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4507
4508         /* The doorbell reset seems to cause lockups on some Smart
4509          * Arrays (e.g. P410, P410i, maybe others).  Until this is
4510          * fixed or at least isolated, avoid the doorbell reset.
4511          */
4512         use_doorbell = 0;
4513
4514         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4515         if (rc)
4516                 goto unmap_cfgtable;
4517
4518         /* Restore the PCI configuration space.  The Open CISS
4519          * Specification says, "Restore the PCI Configuration
4520          * Registers, offsets 00h through 60h. It is important to
4521          * restore the command register, 16-bits at offset 04h,
4522          * last. Do not restore the configuration status register,
4523          * 16-bits at offset 06h."  Note that the offset is 2*i.
4524          */
4525         for (i = 0; i < 32; i++) {
4526                 if (i == 2 || i == 3)
4527                         continue;
4528                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4529         }
4530         wmb();
4531         pci_write_config_word(pdev, 4, saved_config_space[2]);
4532
4533         /* Some devices (notably the HP Smart Array 5i Controller)
4534            need a little pause here */
4535         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4536
4537         /* Controller should be in simple mode at this point.  If it's not,
4538          * It means we're on one of those controllers which doesn't support
4539          * the doorbell reset method and on which the PCI power management reset
4540          * method doesn't work (P800, for example.)
4541          * In those cases, don't try to proceed, as it generally doesn't work.
4542          */
4543         active_transport = readl(&cfgtable->TransportActive);
4544         if (active_transport & PERFORMANT_MODE) {
4545                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4546                         " Ignoring controller.\n");
4547                 rc = -ENODEV;
4548         }
4549
4550 unmap_cfgtable:
4551         iounmap(cfgtable);
4552
4553 unmap_vaddr:
4554         iounmap(vaddr);
4555         return rc;
4556 }
4557
4558 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4559 {
4560         int rc, i;
4561
4562         if (!reset_devices)
4563                 return 0;
4564
4565         /* Reset the controller with a PCI power-cycle or via doorbell */
4566         rc = cciss_kdump_hard_reset_controller(pdev);
4567
4568         /* -ENOTSUPP here means we cannot reset the controller
4569          * but it's already (and still) up and running in
4570          * "performant mode".  Or, it might be 640x, which can't reset
4571          * due to concerns about shared bbwc between 6402/6404 pair.
4572          */
4573         if (rc == -ENOTSUPP)
4574                 return 0; /* just try to do the kdump anyhow. */
4575         if (rc)
4576                 return -ENODEV;
4577         if (cciss_reset_msi(pdev))
4578                 return -ENODEV;
4579
4580         /* Now try to get the controller to respond to a no-op */
4581         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4582                 if (cciss_noop(pdev) == 0)
4583                         break;
4584                 else
4585                         dev_warn(&pdev->dev, "no-op failed%s\n",
4586                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4587                                         "; re-trying" : ""));
4588                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4589         }
4590         return 0;
4591 }
4592
4593 /*
4594  *  This is it.  Find all the controllers and register them.  I really hate
4595  *  stealing all these major device numbers.
4596  *  returns the number of block devices registered.
4597  */
4598 static int __devinit cciss_init_one(struct pci_dev *pdev,
4599                                     const struct pci_device_id *ent)
4600 {
4601         int i;
4602         int j = 0;
4603         int k = 0;
4604         int rc;
4605         int dac, return_code;
4606         InquiryData_struct *inq_buff;
4607         ctlr_info_t *h;
4608
4609         rc = cciss_init_reset_devices(pdev);
4610         if (rc)
4611                 return rc;
4612         i = alloc_cciss_hba(pdev);
4613         if (i < 0)
4614                 return -1;
4615
4616         h = hba[i];
4617         h->pdev = pdev;
4618         h->busy_initializing = 1;
4619         INIT_HLIST_HEAD(&h->cmpQ);
4620         INIT_HLIST_HEAD(&h->reqQ);
4621         mutex_init(&h->busy_shutting_down);
4622
4623         if (cciss_pci_init(h) != 0)
4624                 goto clean_no_release_regions;
4625
4626         sprintf(h->devname, "cciss%d", i);
4627         h->ctlr = i;
4628
4629         init_completion(&h->scan_wait);
4630
4631         if (cciss_create_hba_sysfs_entry(h))
4632                 goto clean0;
4633
4634         /* configure PCI DMA stuff */
4635         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4636                 dac = 1;
4637         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4638                 dac = 0;
4639         else {
4640                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4641                 goto clean1;
4642         }
4643
4644         /*
4645          * register with the major number, or get a dynamic major number
4646          * by passing 0 as argument.  This is done for greater than
4647          * 8 controller support.
4648          */
4649         if (i < MAX_CTLR_ORIG)
4650                 h->major = COMPAQ_CISS_MAJOR + i;
4651         rc = register_blkdev(h->major, h->devname);
4652         if (rc == -EBUSY || rc == -EINVAL) {
4653                 dev_err(&h->pdev->dev,
4654                        "Unable to get major number %d for %s "
4655                        "on hba %d\n", h->major, h->devname, i);
4656                 goto clean1;
4657         } else {
4658                 if (i >= MAX_CTLR_ORIG)
4659                         h->major = rc;
4660         }
4661
4662         /* make sure the board interrupts are off */
4663         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4664         if (h->msi_vector || h->msix_vector) {
4665                 if (request_irq(h->intr[PERF_MODE_INT],
4666                                 do_cciss_msix_intr,
4667                                 IRQF_DISABLED, h->devname, h)) {
4668                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4669                                h->intr[PERF_MODE_INT], h->devname);
4670                         goto clean2;
4671                 }
4672         } else {
4673                 if (request_irq(h->intr[PERF_MODE_INT], do_cciss_intx,
4674                                 IRQF_DISABLED, h->devname, h)) {
4675                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4676                                h->intr[PERF_MODE_INT], h->devname);
4677                         goto clean2;
4678                 }
4679         }
4680
4681         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4682                h->devname, pdev->device, pci_name(pdev),
4683                h->intr[PERF_MODE_INT], dac ? "" : " not");
4684
4685         h->cmd_pool_bits =
4686             kmalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4687                         * sizeof(unsigned long), GFP_KERNEL);
4688         h->cmd_pool = (CommandList_struct *)
4689             pci_alloc_consistent(h->pdev,
4690                     h->nr_cmds * sizeof(CommandList_struct),
4691                     &(h->cmd_pool_dhandle));
4692         h->errinfo_pool = (ErrorInfo_struct *)
4693             pci_alloc_consistent(h->pdev,
4694                     h->nr_cmds * sizeof(ErrorInfo_struct),
4695                     &(h->errinfo_pool_dhandle));
4696         if ((h->cmd_pool_bits == NULL)
4697             || (h->cmd_pool == NULL)
4698             || (h->errinfo_pool == NULL)) {
4699                 dev_err(&h->pdev->dev, "out of memory");
4700                 goto clean4;
4701         }
4702
4703         /* Need space for temp scatter list */
4704         h->scatter_list = kmalloc(h->max_commands *
4705                                                 sizeof(struct scatterlist *),
4706                                                 GFP_KERNEL);
4707         if (!h->scatter_list)
4708                 goto clean4;
4709
4710         for (k = 0; k < h->nr_cmds; k++) {
4711                 h->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4712                                                         h->maxsgentries,
4713                                                         GFP_KERNEL);
4714                 if (h->scatter_list[k] == NULL) {
4715                         dev_err(&h->pdev->dev,
4716                                 "could not allocate s/g lists\n");
4717                         goto clean4;
4718                 }
4719         }
4720         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4721                 h->chainsize, h->nr_cmds);
4722         if (!h->cmd_sg_list && h->chainsize > 0)
4723                 goto clean4;
4724
4725         spin_lock_init(&h->lock);
4726
4727         /* Initialize the pdev driver private data.
4728            have it point to h.  */
4729         pci_set_drvdata(pdev, h);
4730         /* command and error info recs zeroed out before
4731            they are used */
4732         memset(h->cmd_pool_bits, 0,
4733                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4734                         * sizeof(unsigned long));
4735
4736         h->num_luns = 0;
4737         h->highest_lun = -1;
4738         for (j = 0; j < CISS_MAX_LUN; j++) {
4739                 h->drv[j] = NULL;
4740                 h->gendisk[j] = NULL;
4741         }
4742
4743         cciss_scsi_setup(h);
4744
4745         /* Turn the interrupts on so we can service requests */
4746         h->access.set_intr_mask(h, CCISS_INTR_ON);
4747
4748         /* Get the firmware version */
4749         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4750         if (inq_buff == NULL) {
4751                 dev_err(&h->pdev->dev, "out of memory\n");
4752                 goto clean4;
4753         }
4754
4755         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4756                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4757         if (return_code == IO_OK) {
4758                 h->firm_ver[0] = inq_buff->data_byte[32];
4759                 h->firm_ver[1] = inq_buff->data_byte[33];
4760                 h->firm_ver[2] = inq_buff->data_byte[34];
4761                 h->firm_ver[3] = inq_buff->data_byte[35];
4762         } else {         /* send command failed */
4763                 dev_warn(&h->pdev->dev, "unable to determine firmware"
4764                         " version of controller\n");
4765         }
4766         kfree(inq_buff);
4767
4768         cciss_procinit(h);
4769
4770         h->cciss_max_sectors = 8192;
4771
4772         rebuild_lun_table(h, 1, 0);
4773         h->busy_initializing = 0;
4774         return 1;
4775
4776 clean4:
4777         kfree(h->cmd_pool_bits);
4778         /* Free up sg elements */
4779         for (k-- ; k >= 0; k--)
4780                 kfree(h->scatter_list[k]);
4781         kfree(h->scatter_list);
4782         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4783         if (h->cmd_pool)
4784                 pci_free_consistent(h->pdev,
4785                                     h->nr_cmds * sizeof(CommandList_struct),
4786                                     h->cmd_pool, h->cmd_pool_dhandle);
4787         if (h->errinfo_pool)
4788                 pci_free_consistent(h->pdev,
4789                                     h->nr_cmds * sizeof(ErrorInfo_struct),
4790                                     h->errinfo_pool,
4791                                     h->errinfo_pool_dhandle);
4792         free_irq(h->intr[PERF_MODE_INT], h);
4793 clean2:
4794         unregister_blkdev(h->major, h->devname);
4795 clean1:
4796         cciss_destroy_hba_sysfs_entry(h);
4797 clean0:
4798         pci_release_regions(pdev);
4799 clean_no_release_regions:
4800         h->busy_initializing = 0;
4801
4802         /*
4803          * Deliberately omit pci_disable_device(): it does something nasty to
4804          * Smart Array controllers that pci_enable_device does not undo
4805          */
4806         pci_set_drvdata(pdev, NULL);
4807         free_hba(h);
4808         return -1;
4809 }
4810
4811 static void cciss_shutdown(struct pci_dev *pdev)
4812 {
4813         ctlr_info_t *h;
4814         char *flush_buf;
4815         int return_code;
4816
4817         h = pci_get_drvdata(pdev);
4818         flush_buf = kzalloc(4, GFP_KERNEL);
4819         if (!flush_buf) {
4820                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
4821                 return;
4822         }
4823         /* write all data in the battery backed cache to disk */
4824         memset(flush_buf, 0, 4);
4825         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4826                 4, 0, CTLR_LUNID, TYPE_CMD);
4827         kfree(flush_buf);
4828         if (return_code != IO_OK)
4829                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
4830         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4831         free_irq(h->intr[PERF_MODE_INT], h);
4832 }
4833
4834 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4835 {
4836         ctlr_info_t *h;
4837         int i, j;
4838
4839         if (pci_get_drvdata(pdev) == NULL) {
4840                 dev_err(&pdev->dev, "Unable to remove device\n");
4841                 return;
4842         }
4843
4844         h = pci_get_drvdata(pdev);
4845         i = h->ctlr;
4846         if (hba[i] == NULL) {
4847                 dev_err(&pdev->dev, "device appears to already be removed\n");
4848                 return;
4849         }
4850
4851         mutex_lock(&h->busy_shutting_down);
4852
4853         remove_from_scan_list(h);
4854         remove_proc_entry(h->devname, proc_cciss);
4855         unregister_blkdev(h->major, h->devname);
4856
4857         /* remove it from the disk list */
4858         for (j = 0; j < CISS_MAX_LUN; j++) {
4859                 struct gendisk *disk = h->gendisk[j];
4860                 if (disk) {
4861                         struct request_queue *q = disk->queue;
4862
4863                         if (disk->flags & GENHD_FL_UP) {
4864                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
4865                                 del_gendisk(disk);
4866                         }
4867                         if (q)
4868                                 blk_cleanup_queue(q);
4869                 }
4870         }
4871
4872 #ifdef CONFIG_CISS_SCSI_TAPE
4873         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
4874 #endif
4875
4876         cciss_shutdown(pdev);
4877
4878 #ifdef CONFIG_PCI_MSI
4879         if (h->msix_vector)
4880                 pci_disable_msix(h->pdev);
4881         else if (h->msi_vector)
4882                 pci_disable_msi(h->pdev);
4883 #endif                          /* CONFIG_PCI_MSI */
4884
4885         iounmap(h->transtable);
4886         iounmap(h->cfgtable);
4887         iounmap(h->vaddr);
4888
4889         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(CommandList_struct),
4890                             h->cmd_pool, h->cmd_pool_dhandle);
4891         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(ErrorInfo_struct),
4892                             h->errinfo_pool, h->errinfo_pool_dhandle);
4893         kfree(h->cmd_pool_bits);
4894         /* Free up sg elements */
4895         for (j = 0; j < h->nr_cmds; j++)
4896                 kfree(h->scatter_list[j]);
4897         kfree(h->scatter_list);
4898         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4899         /*
4900          * Deliberately omit pci_disable_device(): it does something nasty to
4901          * Smart Array controllers that pci_enable_device does not undo
4902          */
4903         pci_release_regions(pdev);
4904         pci_set_drvdata(pdev, NULL);
4905         cciss_destroy_hba_sysfs_entry(h);
4906         mutex_unlock(&h->busy_shutting_down);
4907         free_hba(h);
4908 }
4909
4910 static struct pci_driver cciss_pci_driver = {
4911         .name = "cciss",
4912         .probe = cciss_init_one,
4913         .remove = __devexit_p(cciss_remove_one),
4914         .id_table = cciss_pci_device_id,        /* id_table */
4915         .shutdown = cciss_shutdown,
4916 };
4917
4918 /*
4919  *  This is it.  Register the PCI driver information for the cards we control
4920  *  the OS will call our registered routines when it finds one of our cards.
4921  */
4922 static int __init cciss_init(void)
4923 {
4924         int err;
4925
4926         /*
4927          * The hardware requires that commands are aligned on a 64-bit
4928          * boundary. Given that we use pci_alloc_consistent() to allocate an
4929          * array of them, the size must be a multiple of 8 bytes.
4930          */
4931         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4932         printk(KERN_INFO DRIVER_NAME "\n");
4933
4934         err = bus_register(&cciss_bus_type);
4935         if (err)
4936                 return err;
4937
4938         /* Start the scan thread */
4939         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4940         if (IS_ERR(cciss_scan_thread)) {
4941                 err = PTR_ERR(cciss_scan_thread);
4942                 goto err_bus_unregister;
4943         }
4944
4945         /* Register for our PCI devices */
4946         err = pci_register_driver(&cciss_pci_driver);
4947         if (err)
4948                 goto err_thread_stop;
4949
4950         return err;
4951
4952 err_thread_stop:
4953         kthread_stop(cciss_scan_thread);
4954 err_bus_unregister:
4955         bus_unregister(&cciss_bus_type);
4956
4957         return err;
4958 }
4959
4960 static void __exit cciss_cleanup(void)
4961 {
4962         int i;
4963
4964         pci_unregister_driver(&cciss_pci_driver);
4965         /* double check that all controller entrys have been removed */
4966         for (i = 0; i < MAX_CTLR; i++) {
4967                 if (hba[i] != NULL) {
4968                         dev_warn(&hba[i]->pdev->dev,
4969                                 "had to remove controller\n");
4970                         cciss_remove_one(hba[i]->pdev);
4971                 }
4972         }
4973         kthread_stop(cciss_scan_thread);
4974         remove_proc_entry("driver/cciss", NULL);
4975         bus_unregister(&cciss_bus_type);
4976 }
4977
4978 module_init(cciss_init);
4979 module_exit(cciss_cleanup);