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/smp_lock.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/compat.h>
  43 #include <linux/mutex.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/io.h>
  46
  47 #include <linux/dma-mapping.h>
  48 #include <linux/blkdev.h>
  49 #include <linux/genhd.h>
  50 #include <linux/completion.h>
  51 #include <scsi/scsi.h>
  52 #include <scsi/sg.h>
  53 #include <scsi/scsi_ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/scatterlist.h>
  56 #include <linux/kthread.h>
  57
  58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  59 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  61
  62 /* Embedded module documentation macros - see modules.h */
  63 MODULE_AUTHOR("Hewlett-Packard Company");
  64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  65 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  66 MODULE_VERSION("3.6.26");
  67 MODULE_LICENSE("GPL");
  68
  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_CISSE,     0x103C, 0x3250},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3251},
 110         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3252},
 111         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3253},
 112         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3254},
 113         {0,}
 114 };
 115
 116 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 117
 118 /*  board_id = Subsystem Device ID & Vendor ID
 119  *  product = Marketing Name for the board
 120  *  access = Address of the struct of function pointers
 121  */
 122 static struct board_type products[] = {
 123         {0x40700E11, "Smart Array 5300", &SA5_access},
 124         {0x40800E11, "Smart Array 5i", &SA5B_access},
 125         {0x40820E11, "Smart Array 532", &SA5B_access},
 126         {0x40830E11, "Smart Array 5312", &SA5B_access},
 127         {0x409A0E11, "Smart Array 641", &SA5_access},
 128         {0x409B0E11, "Smart Array 642", &SA5_access},
 129         {0x409C0E11, "Smart Array 6400", &SA5_access},
 130         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 131         {0x40910E11, "Smart Array 6i", &SA5_access},
 132         {0x3225103C, "Smart Array P600", &SA5_access},
 133         {0x3235103C, "Smart Array P400i", &SA5_access},
 134         {0x3211103C, "Smart Array E200i", &SA5_access},
 135         {0x3212103C, "Smart Array E200", &SA5_access},
 136         {0x3213103C, "Smart Array E200i", &SA5_access},
 137         {0x3214103C, "Smart Array E200i", &SA5_access},
 138         {0x3215103C, "Smart Array E200i", &SA5_access},
 139         {0x3237103C, "Smart Array E500", &SA5_access},
 140 /* controllers below this line are also supported by the hpsa driver. */
 141 #define HPSA_BOUNDARY 0x3223103C
 142         {0x3223103C, "Smart Array P800", &SA5_access},
 143         {0x3234103C, "Smart Array P400", &SA5_access},
 144         {0x323D103C, "Smart Array P700m", &SA5_access},
 145         {0x3241103C, "Smart Array P212", &SA5_access},
 146         {0x3243103C, "Smart Array P410", &SA5_access},
 147         {0x3245103C, "Smart Array P410i", &SA5_access},
 148         {0x3247103C, "Smart Array P411", &SA5_access},
 149         {0x3249103C, "Smart Array P812", &SA5_access},
 150         {0x324A103C, "Smart Array P712m", &SA5_access},
 151         {0x324B103C, "Smart Array P711m", &SA5_access},
 152         {0x3250103C, "Smart Array", &SA5_access},
 153         {0x3251103C, "Smart Array", &SA5_access},
 154         {0x3252103C, "Smart Array", &SA5_access},
 155         {0x3253103C, "Smart Array", &SA5_access},
 156         {0x3254103C, "Smart Array", &SA5_access},
 157 };
 158
 159 /* How long to wait (in milliseconds) for board to go into simple mode */
 160 #define MAX_CONFIG_WAIT 30000
 161 #define MAX_IOCTL_CONFIG_WAIT 1000
 162
 163 /*define how many times we will try a command because of bus resets */
 164 #define MAX_CMD_RETRIES 3
 165
 166 #define MAX_CTLR        32
 167
 168 /* Originally cciss driver only supports 8 major numbers */
 169 #define MAX_CTLR_ORIG   8
 170
 171 static ctlr_info_t *hba[MAX_CTLR];
 172
 173 static struct task_struct *cciss_scan_thread;
 174 static DEFINE_MUTEX(scan_mutex);
 175 static LIST_HEAD(scan_q);
 176
 177 static void do_cciss_request(struct request_queue *q);
 178 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 179 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 180 static int cciss_open(struct block_device *bdev, fmode_t mode);
 181 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 182 static int cciss_release(struct gendisk *disk, fmode_t mode);
 183 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 184                     unsigned int cmd, unsigned long arg);
 185 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 186                        unsigned int cmd, unsigned long arg);
 187 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 188
 189 static int cciss_revalidate(struct gendisk *disk);
 190 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 191 static int deregister_disk(ctlr_info_t *h, int drv_index,
 192                            int clear_all, int via_ioctl);
 193
 194 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 195                         sector_t *total_size, unsigned int *block_size);
 196 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 197                         sector_t *total_size, unsigned int *block_size);
 198 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 199                         sector_t total_size,
 200                         unsigned int block_size, InquiryData_struct *inq_buff,
 201                                    drive_info_struct *drv);
 202 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
 203 static void start_io(ctlr_info_t *h);
 204 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 205                         __u8 page_code, unsigned char scsi3addr[],
 206                         int cmd_type);
 207 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 208         int attempt_retry);
 209 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 210
 211 static int add_to_scan_list(struct ctlr_info *h);
 212 static int scan_thread(void *data);
 213 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 214 static void cciss_hba_release(struct device *dev);
 215 static void cciss_device_release(struct device *dev);
 216 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 217 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 218 static inline u32 next_command(ctlr_info_t *h);
 219 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
 220         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 221         u64 *cfg_offset);
 222 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 223         unsigned long *memory_bar);
 224
 225
 226 /* performant mode helper functions */
 227 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 228                                 int *bucket_map);
 229 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 230
 231 #ifdef CONFIG_PROC_FS
 232 static void cciss_procinit(ctlr_info_t *h);
 233 #else
 234 static void cciss_procinit(ctlr_info_t *h)
 235 {
 236 }
 237 #endif                          /* CONFIG_PROC_FS */
 238
 239 #ifdef CONFIG_COMPAT
 240 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 241                               unsigned, unsigned long);
 242 #endif
 243
 244 static const struct block_device_operations cciss_fops = {
 245         .owner = THIS_MODULE,
 246         .open = cciss_unlocked_open,
 247         .release = cciss_release,
 248         .ioctl = do_ioctl,
 249         .getgeo = cciss_getgeo,
 250 #ifdef CONFIG_COMPAT
 251         .compat_ioctl = cciss_compat_ioctl,
 252 #endif
 253         .revalidate_disk = cciss_revalidate,
 254 };
 255
 256 /* set_performant_mode: Modify the tag for cciss performant
 257  * set bit 0 for pull model, bits 3-1 for block fetch
 258  * register number
 259  */
 260 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 261 {
 262         if (likely(h->transMethod == CFGTBL_Trans_Performant))
 263                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 264 }
 265
 266 /*
 267  * Enqueuing and dequeuing functions for cmdlists.
 268  */
 269 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 270 {
 271         hlist_add_head(&c->list, list);
 272 }
 273
 274 static inline void removeQ(CommandList_struct *c)
 275 {
 276         /*
 277          * After kexec/dump some commands might still
 278          * be in flight, which the firmware will try
 279          * to complete. Resetting the firmware doesn't work
 280          * with old fw revisions, so we have to mark
 281          * them off as 'stale' to prevent the driver from
 282          * falling over.
 283          */
 284         if (WARN_ON(hlist_unhashed(&c->list))) {
 285                 c->cmd_type = CMD_MSG_STALE;
 286                 return;
 287         }
 288
 289         hlist_del_init(&c->list);
 290 }
 291
 292 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 293         CommandList_struct *c)
 294 {
 295         unsigned long flags;
 296         set_performant_mode(h, c);
 297         spin_lock_irqsave(&h->lock, flags);
 298         addQ(&h->reqQ, c);
 299         h->Qdepth++;
 300         start_io(h);
 301         spin_unlock_irqrestore(&h->lock, flags);
 302 }
 303
 304 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 305         int nr_cmds)
 306 {
 307         int i;
 308
 309         if (!cmd_sg_list)
 310                 return;
 311         for (i = 0; i < nr_cmds; i++) {
 312                 kfree(cmd_sg_list[i]);
 313                 cmd_sg_list[i] = NULL;
 314         }
 315         kfree(cmd_sg_list);
 316 }
 317
 318 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 319         ctlr_info_t *h, int chainsize, int nr_cmds)
 320 {
 321         int j;
 322         SGDescriptor_struct **cmd_sg_list;
 323
 324         if (chainsize <= 0)
 325                 return NULL;
 326
 327         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 328         if (!cmd_sg_list)
 329                 return NULL;
 330
 331         /* Build up chain blocks for each command */
 332         for (j = 0; j < nr_cmds; j++) {
 333                 /* Need a block of chainsized s/g elements. */
 334                 cmd_sg_list[j] = kmalloc((chainsize *
 335                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 336                 if (!cmd_sg_list[j]) {
 337                         dev_err(&h->pdev->dev, "Cannot get memory "
 338                                 "for s/g chains.\n");
 339                         goto clean;
 340                 }
 341         }
 342         return cmd_sg_list;
 343 clean:
 344         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 345         return NULL;
 346 }
 347
 348 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 349 {
 350         SGDescriptor_struct *chain_sg;
 351         u64bit temp64;
 352
 353         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 354                 return;
 355
 356         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 357         temp64.val32.lower = chain_sg->Addr.lower;
 358         temp64.val32.upper = chain_sg->Addr.upper;
 359         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 360 }
 361
 362 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 363         SGDescriptor_struct *chain_block, int len)
 364 {
 365         SGDescriptor_struct *chain_sg;
 366         u64bit temp64;
 367
 368         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 369         chain_sg->Ext = CCISS_SG_CHAIN;
 370         chain_sg->Len = len;
 371         temp64.val = pci_map_single(h->pdev, chain_block, len,
 372                                 PCI_DMA_TODEVICE);
 373         chain_sg->Addr.lower = temp64.val32.lower;
 374         chain_sg->Addr.upper = temp64.val32.upper;
 375 }
 376
 377 #include "cciss_scsi.c"         /* For SCSI tape support */
 378
 379 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 380         "UNKNOWN"
 381 };
 382 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 383
 384 #ifdef CONFIG_PROC_FS
 385
 386 /*
 387  * Report information about this controller.
 388  */
 389 #define ENG_GIG 1000000000
 390 #define ENG_GIG_FACTOR (ENG_GIG/512)
 391 #define ENGAGE_SCSI     "engage scsi"
 392
 393 static struct proc_dir_entry *proc_cciss;
 394
 395 static void cciss_seq_show_header(struct seq_file *seq)
 396 {
 397         ctlr_info_t *h = seq->private;
 398
 399         seq_printf(seq, "%s: HP %s Controller\n"
 400                 "Board ID: 0x%08lx\n"
 401                 "Firmware Version: %c%c%c%c\n"
 402                 "IRQ: %d\n"
 403                 "Logical drives: %d\n"
 404                 "Current Q depth: %d\n"
 405                 "Current # commands on controller: %d\n"
 406                 "Max Q depth since init: %d\n"
 407                 "Max # commands on controller since init: %d\n"
 408                 "Max SG entries since init: %d\n",
 409                 h->devname,
 410                 h->product_name,
 411                 (unsigned long)h->board_id,
 412                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 413                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 414                 h->num_luns,
 415                 h->Qdepth, h->commands_outstanding,
 416                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 417
 418 #ifdef CONFIG_CISS_SCSI_TAPE
 419         cciss_seq_tape_report(seq, h);
 420 #endif /* CONFIG_CISS_SCSI_TAPE */
 421 }
 422
 423 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 424 {
 425         ctlr_info_t *h = seq->private;
 426         unsigned long flags;
 427
 428         /* prevent displaying bogus info during configuration
 429          * or deconfiguration of a logical volume
 430          */
 431         spin_lock_irqsave(&h->lock, flags);
 432         if (h->busy_configuring) {
 433                 spin_unlock_irqrestore(&h->lock, flags);
 434                 return ERR_PTR(-EBUSY);
 435         }
 436         h->busy_configuring = 1;
 437         spin_unlock_irqrestore(&h->lock, flags);
 438
 439         if (*pos == 0)
 440                 cciss_seq_show_header(seq);
 441
 442         return pos;
 443 }
 444
 445 static int cciss_seq_show(struct seq_file *seq, void *v)
 446 {
 447         sector_t vol_sz, vol_sz_frac;
 448         ctlr_info_t *h = seq->private;
 449         unsigned ctlr = h->ctlr;
 450         loff_t *pos = v;
 451         drive_info_struct *drv = h->drv[*pos];
 452
 453         if (*pos > h->highest_lun)
 454                 return 0;
 455
 456         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 457                 return 0;
 458
 459         if (drv->heads == 0)
 460                 return 0;
 461
 462         vol_sz = drv->nr_blocks;
 463         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 464         vol_sz_frac *= 100;
 465         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 466
 467         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 468                 drv->raid_level = RAID_UNKNOWN;
 469         seq_printf(seq, "cciss/c%dd%d:"
 470                         "\t%4u.%02uGB\tRAID %s\n",
 471                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 472                         raid_label[drv->raid_level]);
 473         return 0;
 474 }
 475
 476 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 477 {
 478         ctlr_info_t *h = seq->private;
 479
 480         if (*pos > h->highest_lun)
 481                 return NULL;
 482         *pos += 1;
 483
 484         return pos;
 485 }
 486
 487 static void cciss_seq_stop(struct seq_file *seq, void *v)
 488 {
 489         ctlr_info_t *h = seq->private;
 490
 491         /* Only reset h->busy_configuring if we succeeded in setting
 492          * it during cciss_seq_start. */
 493         if (v == ERR_PTR(-EBUSY))
 494                 return;
 495
 496         h->busy_configuring = 0;
 497 }
 498
 499 static const struct seq_operations cciss_seq_ops = {
 500         .start = cciss_seq_start,
 501         .show  = cciss_seq_show,
 502         .next  = cciss_seq_next,
 503         .stop  = cciss_seq_stop,
 504 };
 505
 506 static int cciss_seq_open(struct inode *inode, struct file *file)
 507 {
 508         int ret = seq_open(file, &cciss_seq_ops);
 509         struct seq_file *seq = file->private_data;
 510
 511         if (!ret)
 512                 seq->private = PDE(inode)->data;
 513
 514         return ret;
 515 }
 516
 517 static ssize_t
 518 cciss_proc_write(struct file *file, const char __user *buf,
 519                  size_t length, loff_t *ppos)
 520 {
 521         int err;
 522         char *buffer;
 523
 524 #ifndef CONFIG_CISS_SCSI_TAPE
 525         return -EINVAL;
 526 #endif
 527
 528         if (!buf || length > PAGE_SIZE - 1)
 529                 return -EINVAL;
 530
 531         buffer = (char *)__get_free_page(GFP_KERNEL);
 532         if (!buffer)
 533                 return -ENOMEM;
 534
 535         err = -EFAULT;
 536         if (copy_from_user(buffer, buf, length))
 537                 goto out;
 538         buffer[length] = '\0';
 539
 540 #ifdef CONFIG_CISS_SCSI_TAPE
 541         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 542                 struct seq_file *seq = file->private_data;
 543                 ctlr_info_t *h = seq->private;
 544
 545                 err = cciss_engage_scsi(h);
 546                 if (err == 0)
 547                         err = length;
 548         } else
 549 #endif /* CONFIG_CISS_SCSI_TAPE */
 550                 err = -EINVAL;
 551         /* might be nice to have "disengage" too, but it's not
 552            safely possible. (only 1 module use count, lock issues.) */
 553
 554 out:
 555         free_page((unsigned long)buffer);
 556         return err;
 557 }
 558
 559 static const struct file_operations cciss_proc_fops = {
 560         .owner   = THIS_MODULE,
 561         .open    = cciss_seq_open,
 562         .read    = seq_read,
 563         .llseek  = seq_lseek,
 564         .release = seq_release,
 565         .write   = cciss_proc_write,
 566 };
 567
 568 static void __devinit cciss_procinit(ctlr_info_t *h)
 569 {
 570         struct proc_dir_entry *pde;
 571
 572         if (proc_cciss == NULL)
 573                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 574         if (!proc_cciss)
 575                 return;
 576         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 577                                         S_IROTH, proc_cciss,
 578                                         &cciss_proc_fops, h);
 579 }
 580 #endif                          /* CONFIG_PROC_FS */
 581
 582 #define MAX_PRODUCT_NAME_LEN 19
 583
 584 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 585 #define to_drv(n) container_of(n, drive_info_struct, dev)
 586
 587 static ssize_t host_store_rescan(struct device *dev,
 588                                  struct device_attribute *attr,
 589                                  const char *buf, size_t count)
 590 {
 591         struct ctlr_info *h = to_hba(dev);
 592
 593         add_to_scan_list(h);
 594         wake_up_process(cciss_scan_thread);
 595         wait_for_completion_interruptible(&h->scan_wait);
 596
 597         return count;
 598 }
 599 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 600
 601 static ssize_t dev_show_unique_id(struct device *dev,
 602                                  struct device_attribute *attr,
 603                                  char *buf)
 604 {
 605         drive_info_struct *drv = to_drv(dev);
 606         struct ctlr_info *h = to_hba(drv->dev.parent);
 607         __u8 sn[16];
 608         unsigned long flags;
 609         int ret = 0;
 610
 611         spin_lock_irqsave(&h->lock, flags);
 612         if (h->busy_configuring)
 613                 ret = -EBUSY;
 614         else
 615                 memcpy(sn, drv->serial_no, sizeof(sn));
 616         spin_unlock_irqrestore(&h->lock, flags);
 617
 618         if (ret)
 619                 return ret;
 620         else
 621                 return snprintf(buf, 16 * 2 + 2,
 622                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 623                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 624                                 sn[0], sn[1], sn[2], sn[3],
 625                                 sn[4], sn[5], sn[6], sn[7],
 626                                 sn[8], sn[9], sn[10], sn[11],
 627                                 sn[12], sn[13], sn[14], sn[15]);
 628 }
 629 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 630
 631 static ssize_t dev_show_vendor(struct device *dev,
 632                                struct device_attribute *attr,
 633                                char *buf)
 634 {
 635         drive_info_struct *drv = to_drv(dev);
 636         struct ctlr_info *h = to_hba(drv->dev.parent);
 637         char vendor[VENDOR_LEN + 1];
 638         unsigned long flags;
 639         int ret = 0;
 640
 641         spin_lock_irqsave(&h->lock, flags);
 642         if (h->busy_configuring)
 643                 ret = -EBUSY;
 644         else
 645                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 646         spin_unlock_irqrestore(&h->lock, flags);
 647
 648         if (ret)
 649                 return ret;
 650         else
 651                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 652 }
 653 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 654
 655 static ssize_t dev_show_model(struct device *dev,
 656                               struct device_attribute *attr,
 657                               char *buf)
 658 {
 659         drive_info_struct *drv = to_drv(dev);
 660         struct ctlr_info *h = to_hba(drv->dev.parent);
 661         char model[MODEL_LEN + 1];
 662         unsigned long flags;
 663         int ret = 0;
 664
 665         spin_lock_irqsave(&h->lock, flags);
 666         if (h->busy_configuring)
 667                 ret = -EBUSY;
 668         else
 669                 memcpy(model, drv->model, MODEL_LEN + 1);
 670         spin_unlock_irqrestore(&h->lock, flags);
 671
 672         if (ret)
 673                 return ret;
 674         else
 675                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 676 }
 677 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 678
 679 static ssize_t dev_show_rev(struct device *dev,
 680                             struct device_attribute *attr,
 681                             char *buf)
 682 {
 683         drive_info_struct *drv = to_drv(dev);
 684         struct ctlr_info *h = to_hba(drv->dev.parent);
 685         char rev[REV_LEN + 1];
 686         unsigned long flags;
 687         int ret = 0;
 688
 689         spin_lock_irqsave(&h->lock, flags);
 690         if (h->busy_configuring)
 691                 ret = -EBUSY;
 692         else
 693                 memcpy(rev, drv->rev, REV_LEN + 1);
 694         spin_unlock_irqrestore(&h->lock, flags);
 695
 696         if (ret)
 697                 return ret;
 698         else
 699                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 700 }
 701 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 702
 703 static ssize_t cciss_show_lunid(struct device *dev,
 704                                 struct device_attribute *attr, char *buf)
 705 {
 706         drive_info_struct *drv = to_drv(dev);
 707         struct ctlr_info *h = to_hba(drv->dev.parent);
 708         unsigned long flags;
 709         unsigned char lunid[8];
 710
 711         spin_lock_irqsave(&h->lock, flags);
 712         if (h->busy_configuring) {
 713                 spin_unlock_irqrestore(&h->lock, flags);
 714                 return -EBUSY;
 715         }
 716         if (!drv->heads) {
 717                 spin_unlock_irqrestore(&h->lock, flags);
 718                 return -ENOTTY;
 719         }
 720         memcpy(lunid, drv->LunID, sizeof(lunid));
 721         spin_unlock_irqrestore(&h->lock, flags);
 722         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 723                 lunid[0], lunid[1], lunid[2], lunid[3],
 724                 lunid[4], lunid[5], lunid[6], lunid[7]);
 725 }
 726 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 727
 728 static ssize_t cciss_show_raid_level(struct device *dev,
 729                                      struct device_attribute *attr, char *buf)
 730 {
 731         drive_info_struct *drv = to_drv(dev);
 732         struct ctlr_info *h = to_hba(drv->dev.parent);
 733         int raid;
 734         unsigned long flags;
 735
 736         spin_lock_irqsave(&h->lock, flags);
 737         if (h->busy_configuring) {
 738                 spin_unlock_irqrestore(&h->lock, flags);
 739                 return -EBUSY;
 740         }
 741         raid = drv->raid_level;
 742         spin_unlock_irqrestore(&h->lock, flags);
 743         if (raid < 0 || raid > RAID_UNKNOWN)
 744                 raid = RAID_UNKNOWN;
 745
 746         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 747                         raid_label[raid]);
 748 }
 749 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 750
 751 static ssize_t cciss_show_usage_count(struct device *dev,
 752                                       struct device_attribute *attr, char *buf)
 753 {
 754         drive_info_struct *drv = to_drv(dev);
 755         struct ctlr_info *h = to_hba(drv->dev.parent);
 756         unsigned long flags;
 757         int count;
 758
 759         spin_lock_irqsave(&h->lock, flags);
 760         if (h->busy_configuring) {
 761                 spin_unlock_irqrestore(&h->lock, flags);
 762                 return -EBUSY;
 763         }
 764         count = drv->usage_count;
 765         spin_unlock_irqrestore(&h->lock, flags);
 766         return snprintf(buf, 20, "%d\n", count);
 767 }
 768 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 769
 770 static struct attribute *cciss_host_attrs[] = {
 771         &dev_attr_rescan.attr,
 772         NULL
 773 };
 774
 775 static struct attribute_group cciss_host_attr_group = {
 776         .attrs = cciss_host_attrs,
 777 };
 778
 779 static const struct attribute_group *cciss_host_attr_groups[] = {
 780         &cciss_host_attr_group,
 781         NULL
 782 };
 783
 784 static struct device_type cciss_host_type = {
 785         .name           = "cciss_host",
 786         .groups         = cciss_host_attr_groups,
 787         .release        = cciss_hba_release,
 788 };
 789
 790 static struct attribute *cciss_dev_attrs[] = {
 791         &dev_attr_unique_id.attr,
 792         &dev_attr_model.attr,
 793         &dev_attr_vendor.attr,
 794         &dev_attr_rev.attr,
 795         &dev_attr_lunid.attr,
 796         &dev_attr_raid_level.attr,
 797         &dev_attr_usage_count.attr,
 798         NULL
 799 };
 800
 801 static struct attribute_group cciss_dev_attr_group = {
 802         .attrs = cciss_dev_attrs,
 803 };
 804
 805 static const struct attribute_group *cciss_dev_attr_groups[] = {
 806         &cciss_dev_attr_group,
 807         NULL
 808 };
 809
 810 static struct device_type cciss_dev_type = {
 811         .name           = "cciss_device",
 812         .groups         = cciss_dev_attr_groups,
 813         .release        = cciss_device_release,
 814 };
 815
 816 static struct bus_type cciss_bus_type = {
 817         .name           = "cciss",
 818 };
 819
 820 /*
 821  * cciss_hba_release is called when the reference count
 822  * of h->dev goes to zero.
 823  */
 824 static void cciss_hba_release(struct device *dev)
 825 {
 826         /*
 827          * nothing to do, but need this to avoid a warning
 828          * about not having a release handler from lib/kref.c.
 829          */
 830 }
 831
 832 /*
 833  * Initialize sysfs entry for each controller.  This sets up and registers
 834  * the 'cciss#' directory for each individual controller under
 835  * /sys/bus/pci/devices/<dev>/.
 836  */
 837 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 838 {
 839         device_initialize(&h->dev);
 840         h->dev.type = &cciss_host_type;
 841         h->dev.bus = &cciss_bus_type;
 842         dev_set_name(&h->dev, "%s", h->devname);
 843         h->dev.parent = &h->pdev->dev;
 844
 845         return device_add(&h->dev);
 846 }
 847
 848 /*
 849  * Remove sysfs entries for an hba.
 850  */
 851 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 852 {
 853         device_del(&h->dev);
 854         put_device(&h->dev); /* final put. */
 855 }
 856
 857 /* cciss_device_release is called when the reference count
 858  * of h->drv[x]dev goes to zero.
 859  */
 860 static void cciss_device_release(struct device *dev)
 861 {
 862         drive_info_struct *drv = to_drv(dev);
 863         kfree(drv);
 864 }
 865
 866 /*
 867  * Initialize sysfs for each logical drive.  This sets up and registers
 868  * the 'c#d#' directory for each individual logical drive under
 869  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 870  * /sys/block/cciss!c#d# to this entry.
 871  */
 872 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 873                                        int drv_index)
 874 {
 875         struct device *dev;
 876
 877         if (h->drv[drv_index]->device_initialized)
 878                 return 0;
 879
 880         dev = &h->drv[drv_index]->dev;
 881         device_initialize(dev);
 882         dev->type = &cciss_dev_type;
 883         dev->bus = &cciss_bus_type;
 884         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 885         dev->parent = &h->dev;
 886         h->drv[drv_index]->device_initialized = 1;
 887         return device_add(dev);
 888 }
 889
 890 /*
 891  * Remove sysfs entries for a logical drive.
 892  */
 893 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 894         int ctlr_exiting)
 895 {
 896         struct device *dev = &h->drv[drv_index]->dev;
 897
 898         /* special case for c*d0, we only destroy it on controller exit */
 899         if (drv_index == 0 && !ctlr_exiting)
 900                 return;
 901
 902         device_del(dev);
 903         put_device(dev); /* the "final" put. */
 904         h->drv[drv_index] = NULL;
 905 }
 906
 907 /*
 908  * For operations that cannot sleep, a command block is allocated at init,
 909  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 910  * which ones are free or in use.
 911  */
 912 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 913 {
 914         CommandList_struct *c;
 915         int i;
 916         u64bit temp64;
 917         dma_addr_t cmd_dma_handle, err_dma_handle;
 918
 919         do {
 920                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 921                 if (i == h->nr_cmds)
 922                         return NULL;
 923         } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
 924                   h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 925         c = h->cmd_pool + i;
 926         memset(c, 0, sizeof(CommandList_struct));
 927         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
 928         c->err_info = h->errinfo_pool + i;
 929         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 930         err_dma_handle = h->errinfo_pool_dhandle
 931             + i * sizeof(ErrorInfo_struct);
 932         h->nr_allocs++;
 933
 934         c->cmdindex = i;
 935
 936         INIT_HLIST_NODE(&c->list);
 937         c->busaddr = (__u32) cmd_dma_handle;
 938         temp64.val = (__u64) err_dma_handle;
 939         c->ErrDesc.Addr.lower = temp64.val32.lower;
 940         c->ErrDesc.Addr.upper = temp64.val32.upper;
 941         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 942
 943         c->ctlr = h->ctlr;
 944         return c;
 945 }
 946
 947 /* allocate a command using pci_alloc_consistent, used for ioctls,
 948  * etc., not for the main i/o path.
 949  */
 950 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
 951 {
 952         CommandList_struct *c;
 953         u64bit temp64;
 954         dma_addr_t cmd_dma_handle, err_dma_handle;
 955
 956         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 957                 sizeof(CommandList_struct), &cmd_dma_handle);
 958         if (c == NULL)
 959                 return NULL;
 960         memset(c, 0, sizeof(CommandList_struct));
 961
 962         c->cmdindex = -1;
 963
 964         c->err_info = (ErrorInfo_struct *)
 965             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 966                     &err_dma_handle);
 967
 968         if (c->err_info == NULL) {
 969                 pci_free_consistent(h->pdev,
 970                         sizeof(CommandList_struct), c, cmd_dma_handle);
 971                 return NULL;
 972         }
 973         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 974
 975         INIT_HLIST_NODE(&c->list);
 976         c->busaddr = (__u32) cmd_dma_handle;
 977         temp64.val = (__u64) err_dma_handle;
 978         c->ErrDesc.Addr.lower = temp64.val32.lower;
 979         c->ErrDesc.Addr.upper = temp64.val32.upper;
 980         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 981
 982         c->ctlr = h->ctlr;
 983         return c;
 984 }
 985
 986 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
 987 {
 988         int i;
 989
 990         i = c - h->cmd_pool;
 991         clear_bit(i & (BITS_PER_LONG - 1),
 992                   h->cmd_pool_bits + (i / BITS_PER_LONG));
 993         h->nr_frees++;
 994 }
 995
 996 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
 997 {
 998         u64bit temp64;
 999
1000         temp64.val32.lower = c->ErrDesc.Addr.lower;
1001         temp64.val32.upper = c->ErrDesc.Addr.upper;
1002         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1003                             c->err_info, (dma_addr_t) temp64.val);
1004         pci_free_consistent(h->pdev, sizeof(CommandList_struct),
1005                             c, (dma_addr_t) c->busaddr);
1006 }
1007
1008 static inline ctlr_info_t *get_host(struct gendisk *disk)
1009 {
1010         return disk->queue->queuedata;
1011 }
1012
1013 static inline drive_info_struct *get_drv(struct gendisk *disk)
1014 {
1015         return disk->private_data;
1016 }
1017
1018 /*
1019  * Open.  Make sure the device is really there.
1020  */
1021 static int cciss_open(struct block_device *bdev, fmode_t mode)
1022 {
1023         ctlr_info_t *h = get_host(bdev->bd_disk);
1024         drive_info_struct *drv = get_drv(bdev->bd_disk);
1025
1026         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1027         if (drv->busy_configuring)
1028                 return -EBUSY;
1029         /*
1030          * Root is allowed to open raw volume zero even if it's not configured
1031          * so array config can still work. Root is also allowed to open any
1032          * volume that has a LUN ID, so it can issue IOCTL to reread the
1033          * disk information.  I don't think I really like this
1034          * but I'm already using way to many device nodes to claim another one
1035          * for "raw controller".
1036          */
1037         if (drv->heads == 0) {
1038                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1039                         /* if not node 0 make sure it is a partition = 0 */
1040                         if (MINOR(bdev->bd_dev) & 0x0f) {
1041                                 return -ENXIO;
1042                                 /* if it is, make sure we have a LUN ID */
1043                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1044                                 sizeof(drv->LunID))) {
1045                                 return -ENXIO;
1046                         }
1047                 }
1048                 if (!capable(CAP_SYS_ADMIN))
1049                         return -EPERM;
1050         }
1051         drv->usage_count++;
1052         h->usage_count++;
1053         return 0;
1054 }
1055
1056 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1057 {
1058         int ret;
1059
1060         lock_kernel();
1061         ret = cciss_open(bdev, mode);
1062         unlock_kernel();
1063
1064         return ret;
1065 }
1066
1067 /*
1068  * Close.  Sync first.
1069  */
1070 static int cciss_release(struct gendisk *disk, fmode_t mode)
1071 {
1072         ctlr_info_t *h;
1073         drive_info_struct *drv;
1074
1075         lock_kernel();
1076         h = get_host(disk);
1077         drv = get_drv(disk);
1078         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1079         drv->usage_count--;
1080         h->usage_count--;
1081         unlock_kernel();
1082         return 0;
1083 }
1084
1085 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1086                     unsigned cmd, unsigned long arg)
1087 {
1088         int ret;
1089         lock_kernel();
1090         ret = cciss_ioctl(bdev, mode, cmd, arg);
1091         unlock_kernel();
1092         return ret;
1093 }
1094
1095 #ifdef CONFIG_COMPAT
1096
1097 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1098                                   unsigned cmd, unsigned long arg);
1099 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1100                                       unsigned cmd, unsigned long arg);
1101
1102 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1103                               unsigned cmd, unsigned long arg)
1104 {
1105         switch (cmd) {
1106         case CCISS_GETPCIINFO:
1107         case CCISS_GETINTINFO:
1108         case CCISS_SETINTINFO:
1109         case CCISS_GETNODENAME:
1110         case CCISS_SETNODENAME:
1111         case CCISS_GETHEARTBEAT:
1112         case CCISS_GETBUSTYPES:
1113         case CCISS_GETFIRMVER:
1114         case CCISS_GETDRIVVER:
1115         case CCISS_REVALIDVOLS:
1116         case CCISS_DEREGDISK:
1117         case CCISS_REGNEWDISK:
1118         case CCISS_REGNEWD:
1119         case CCISS_RESCANDISK:
1120         case CCISS_GETLUNINFO:
1121                 return do_ioctl(bdev, mode, cmd, arg);
1122
1123         case CCISS_PASSTHRU32:
1124                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1125         case CCISS_BIG_PASSTHRU32:
1126                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1127
1128         default:
1129                 return -ENOIOCTLCMD;
1130         }
1131 }
1132
1133 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1134                                   unsigned cmd, unsigned long arg)
1135 {
1136         IOCTL32_Command_struct __user *arg32 =
1137             (IOCTL32_Command_struct __user *) arg;
1138         IOCTL_Command_struct arg64;
1139         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1140         int err;
1141         u32 cp;
1142
1143         err = 0;
1144         err |=
1145             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1146                            sizeof(arg64.LUN_info));
1147         err |=
1148             copy_from_user(&arg64.Request, &arg32->Request,
1149                            sizeof(arg64.Request));
1150         err |=
1151             copy_from_user(&arg64.error_info, &arg32->error_info,
1152                            sizeof(arg64.error_info));
1153         err |= get_user(arg64.buf_size, &arg32->buf_size);
1154         err |= get_user(cp, &arg32->buf);
1155         arg64.buf = compat_ptr(cp);
1156         err |= copy_to_user(p, &arg64, sizeof(arg64));
1157
1158         if (err)
1159                 return -EFAULT;
1160
1161         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1162         if (err)
1163                 return err;
1164         err |=
1165             copy_in_user(&arg32->error_info, &p->error_info,
1166                          sizeof(arg32->error_info));
1167         if (err)
1168                 return -EFAULT;
1169         return err;
1170 }
1171
1172 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1173                                       unsigned cmd, unsigned long arg)
1174 {
1175         BIG_IOCTL32_Command_struct __user *arg32 =
1176             (BIG_IOCTL32_Command_struct __user *) arg;
1177         BIG_IOCTL_Command_struct arg64;
1178         BIG_IOCTL_Command_struct __user *p =
1179             compat_alloc_user_space(sizeof(arg64));
1180         int err;
1181         u32 cp;
1182
1183         err = 0;
1184         err |=
1185             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1186                            sizeof(arg64.LUN_info));
1187         err |=
1188             copy_from_user(&arg64.Request, &arg32->Request,
1189                            sizeof(arg64.Request));
1190         err |=
1191             copy_from_user(&arg64.error_info, &arg32->error_info,
1192                            sizeof(arg64.error_info));
1193         err |= get_user(arg64.buf_size, &arg32->buf_size);
1194         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1195         err |= get_user(cp, &arg32->buf);
1196         arg64.buf = compat_ptr(cp);
1197         err |= copy_to_user(p, &arg64, sizeof(arg64));
1198
1199         if (err)
1200                 return -EFAULT;
1201
1202         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1203         if (err)
1204                 return err;
1205         err |=
1206             copy_in_user(&arg32->error_info, &p->error_info,
1207                          sizeof(arg32->error_info));
1208         if (err)
1209                 return -EFAULT;
1210         return err;
1211 }
1212 #endif
1213
1214 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1215 {
1216         drive_info_struct *drv = get_drv(bdev->bd_disk);
1217
1218         if (!drv->cylinders)
1219                 return -ENXIO;
1220
1221         geo->heads = drv->heads;
1222         geo->sectors = drv->sectors;
1223         geo->cylinders = drv->cylinders;
1224         return 0;
1225 }
1226
1227 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1228 {
1229         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1230                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1231                 (void)check_for_unit_attention(h, c);
1232 }
1233
1234 static int cciss_getpciinfo(ctlr_info_t *h, void __user *argp)
1235 {
1236         cciss_pci_info_struct pciinfo;
1237
1238         if (!argp)
1239                 return -EINVAL;
1240         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1241         pciinfo.bus = h->pdev->bus->number;
1242         pciinfo.dev_fn = h->pdev->devfn;
1243         pciinfo.board_id = h->board_id;
1244         if (copy_to_user(argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1245                 return -EFAULT;
1246         return 0;
1247 }
1248
1249 static int cciss_getintinfo(ctlr_info_t *h, void __user *argp)
1250 {
1251         cciss_coalint_struct intinfo;
1252
1253         if (!argp)
1254                 return -EINVAL;
1255         intinfo.delay = readl(&h->cfgtable->HostWrite.CoalIntDelay);
1256         intinfo.count = readl(&h->cfgtable->HostWrite.CoalIntCount);
1257         if (copy_to_user
1258             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1259                 return -EFAULT;
1260         return 0;
1261 }
1262
1263 static int cciss_setintinfo(ctlr_info_t *h, void __user *argp)
1264 {
1265         cciss_coalint_struct intinfo;
1266         unsigned long flags;
1267         int i;
1268
1269         if (!argp)
1270                 return -EINVAL;
1271         if (!capable(CAP_SYS_ADMIN))
1272                 return -EPERM;
1273         if (copy_from_user(&intinfo, argp, sizeof(intinfo)))
1274                 return -EFAULT;
1275         if ((intinfo.delay == 0) && (intinfo.count == 0))
1276                 return -EINVAL;
1277         spin_lock_irqsave(&h->lock, flags);
1278         /* Update the field, and then ring the doorbell */
1279         writel(intinfo.delay, &(h->cfgtable->HostWrite.CoalIntDelay));
1280         writel(intinfo.count, &(h->cfgtable->HostWrite.CoalIntCount));
1281         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1282
1283         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1284                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1285                         break;
1286                 udelay(1000); /* delay and try again */
1287         }
1288         spin_unlock_irqrestore(&h->lock, flags);
1289         if (i >= MAX_IOCTL_CONFIG_WAIT)
1290                 return -EAGAIN;
1291         return 0;
1292 }
1293
1294 static int cciss_getnodename(ctlr_info_t *h, void __user *argp)
1295 {
1296         NodeName_type NodeName;
1297         int i;
1298
1299         if (!argp)
1300                 return -EINVAL;
1301         for (i = 0; i < 16; i++)
1302                 NodeName[i] = readb(&h->cfgtable->ServerName[i]);
1303         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1304                 return -EFAULT;
1305         return 0;
1306 }
1307
1308 static int cciss_setnodename(ctlr_info_t *h, void __user *argp)
1309 {
1310         NodeName_type NodeName;
1311         unsigned long flags;
1312         int i;
1313
1314         if (!argp)
1315                 return -EINVAL;
1316         if (!capable(CAP_SYS_ADMIN))
1317                 return -EPERM;
1318         if (copy_from_user(NodeName, argp, sizeof(NodeName_type)))
1319                 return -EFAULT;
1320         spin_lock_irqsave(&h->lock, flags);
1321         /* Update the field, and then ring the doorbell */
1322         for (i = 0; i < 16; i++)
1323                 writeb(NodeName[i], &h->cfgtable->ServerName[i]);
1324         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1325         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1326                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
1327                         break;
1328                 udelay(1000); /* delay and try again */
1329         }
1330         spin_unlock_irqrestore(&h->lock, flags);
1331         if (i >= MAX_IOCTL_CONFIG_WAIT)
1332                 return -EAGAIN;
1333         return 0;
1334 }
1335
1336 static int cciss_getheartbeat(ctlr_info_t *h, void __user *argp)
1337 {
1338         Heartbeat_type heartbeat;
1339
1340         if (!argp)
1341                 return -EINVAL;
1342         heartbeat = readl(&h->cfgtable->HeartBeat);
1343         if (copy_to_user(argp, &heartbeat, sizeof(Heartbeat_type)))
1344                 return -EFAULT;
1345         return 0;
1346 }
1347
1348 static int cciss_getbustypes(ctlr_info_t *h, void __user *argp)
1349 {
1350         BusTypes_type BusTypes;
1351
1352         if (!argp)
1353                 return -EINVAL;
1354         BusTypes = readl(&h->cfgtable->BusTypes);
1355         if (copy_to_user(argp, &BusTypes, sizeof(BusTypes_type)))
1356                 return -EFAULT;
1357         return 0;
1358 }
1359
1360 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1361                        unsigned int cmd, unsigned long arg)
1362 {
1363         struct gendisk *disk = bdev->bd_disk;
1364         ctlr_info_t *h = get_host(disk);
1365         drive_info_struct *drv = get_drv(disk);
1366         void __user *argp = (void __user *)arg;
1367
1368         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1369                 cmd, arg);
1370         switch (cmd) {
1371         case CCISS_GETPCIINFO:
1372                 return cciss_getpciinfo(h, argp);
1373         case CCISS_GETINTINFO:
1374                 return cciss_getintinfo(h, argp);
1375         case CCISS_SETINTINFO:
1376                 return cciss_setintinfo(h, argp);
1377         case CCISS_GETNODENAME:
1378                 return cciss_getnodename(h, argp);
1379         case CCISS_SETNODENAME:
1380                 return cciss_setnodename(h, argp);
1381         case CCISS_GETHEARTBEAT:
1382                 return cciss_getheartbeat(h, argp);
1383         case CCISS_GETBUSTYPES:
1384                 return cciss_getbustypes(h, argp);
1385         case CCISS_GETFIRMVER:
1386                 {
1387                         FirmwareVer_type firmware;
1388
1389                         if (!arg)
1390                                 return -EINVAL;
1391                         memcpy(firmware, h->firm_ver, 4);
1392
1393                         if (copy_to_user
1394                             (argp, firmware, sizeof(FirmwareVer_type)))
1395                                 return -EFAULT;
1396                         return 0;
1397                 }
1398         case CCISS_GETDRIVVER:
1399                 {
1400                         DriverVer_type DriverVer = DRIVER_VERSION;
1401
1402                         if (!arg)
1403                                 return -EINVAL;
1404
1405                         if (copy_to_user
1406                             (argp, &DriverVer, sizeof(DriverVer_type)))
1407                                 return -EFAULT;
1408                         return 0;
1409                 }
1410
1411         case CCISS_DEREGDISK:
1412         case CCISS_REGNEWD:
1413         case CCISS_REVALIDVOLS:
1414                 return rebuild_lun_table(h, 0, 1);
1415
1416         case CCISS_GETLUNINFO:{
1417                         LogvolInfo_struct luninfo;
1418
1419                         memcpy(&luninfo.LunID, drv->LunID,
1420                                 sizeof(luninfo.LunID));
1421                         luninfo.num_opens = drv->usage_count;
1422                         luninfo.num_parts = 0;
1423                         if (copy_to_user(argp, &luninfo,
1424                                          sizeof(LogvolInfo_struct)))
1425                                 return -EFAULT;
1426                         return 0;
1427                 }
1428         case CCISS_PASSTHRU:
1429                 {
1430                         IOCTL_Command_struct iocommand;
1431                         CommandList_struct *c;
1432                         char *buff = NULL;
1433                         u64bit temp64;
1434                         DECLARE_COMPLETION_ONSTACK(wait);
1435
1436                         if (!arg)
1437                                 return -EINVAL;
1438
1439                         if (!capable(CAP_SYS_RAWIO))
1440                                 return -EPERM;
1441
1442                         if (copy_from_user
1443                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1444                                 return -EFAULT;
1445                         if ((iocommand.buf_size < 1) &&
1446                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1447                                 return -EINVAL;
1448                         }
1449 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1450                         /* Check kmalloc limits */
1451                         if (iocommand.buf_size > 128000)
1452                                 return -EINVAL;
1453 #endif
1454                         if (iocommand.buf_size > 0) {
1455                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1456                                 if (buff == NULL)
1457                                         return -EFAULT;
1458                         }
1459                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1460                                 /* Copy the data into the buffer we created */
1461                                 if (copy_from_user
1462                                     (buff, iocommand.buf, iocommand.buf_size)) {
1463                                         kfree(buff);
1464                                         return -EFAULT;
1465                                 }
1466                         } else {
1467                                 memset(buff, 0, iocommand.buf_size);
1468                         }
1469                         c = cmd_special_alloc(h);
1470                         if (!c) {
1471                                 kfree(buff);
1472                                 return -ENOMEM;
1473                         }
1474                         /* Fill in the command type */
1475                         c->cmd_type = CMD_IOCTL_PEND;
1476                         /* Fill in Command Header */
1477                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1478                         if (iocommand.buf_size > 0) /* buffer to fill */
1479                         {
1480                                 c->Header.SGList = 1;
1481                                 c->Header.SGTotal = 1;
1482                         } else /* no buffers to fill */
1483                         {
1484                                 c->Header.SGList = 0;
1485                                 c->Header.SGTotal = 0;
1486                         }
1487                         c->Header.LUN = iocommand.LUN_info;
1488                         /* use the kernel address the cmd block for tag */
1489                         c->Header.Tag.lower = c->busaddr;
1490
1491                         /* Fill in Request block */
1492                         c->Request = iocommand.Request;
1493
1494                         /* Fill in the scatter gather information */
1495                         if (iocommand.buf_size > 0) {
1496                                 temp64.val = pci_map_single(h->pdev, buff,
1497                                         iocommand.buf_size,
1498                                         PCI_DMA_BIDIRECTIONAL);
1499                                 c->SG[0].Addr.lower = temp64.val32.lower;
1500                                 c->SG[0].Addr.upper = temp64.val32.upper;
1501                                 c->SG[0].Len = iocommand.buf_size;
1502                                 c->SG[0].Ext = 0;  /* we are not chaining */
1503                         }
1504                         c->waiting = &wait;
1505
1506                         enqueue_cmd_and_start_io(h, c);
1507                         wait_for_completion(&wait);
1508
1509                         /* unlock the buffers from DMA */
1510                         temp64.val32.lower = c->SG[0].Addr.lower;
1511                         temp64.val32.upper = c->SG[0].Addr.upper;
1512                         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val,
1513                                          iocommand.buf_size,
1514                                          PCI_DMA_BIDIRECTIONAL);
1515
1516                         check_ioctl_unit_attention(h, c);
1517
1518                         /* Copy the error information out */
1519                         iocommand.error_info = *(c->err_info);
1520                         if (copy_to_user
1521                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1522                                 kfree(buff);
1523                                 cmd_special_free(h, c);
1524                                 return -EFAULT;
1525                         }
1526
1527                         if (iocommand.Request.Type.Direction == XFER_READ) {
1528                                 /* Copy the data out of the buffer we created */
1529                                 if (copy_to_user
1530                                     (iocommand.buf, buff, iocommand.buf_size)) {
1531                                         kfree(buff);
1532                                         cmd_special_free(h, c);
1533                                         return -EFAULT;
1534                                 }
1535                         }
1536                         kfree(buff);
1537                         cmd_special_free(h, c);
1538                         return 0;
1539                 }
1540         case CCISS_BIG_PASSTHRU:{
1541                         BIG_IOCTL_Command_struct *ioc;
1542                         CommandList_struct *c;
1543                         unsigned char **buff = NULL;
1544                         int *buff_size = NULL;
1545                         u64bit temp64;
1546                         BYTE sg_used = 0;
1547                         int status = 0;
1548                         int i;
1549                         DECLARE_COMPLETION_ONSTACK(wait);
1550                         __u32 left;
1551                         __u32 sz;
1552                         BYTE __user *data_ptr;
1553
1554                         if (!arg)
1555                                 return -EINVAL;
1556                         if (!capable(CAP_SYS_RAWIO))
1557                                 return -EPERM;
1558                         ioc = (BIG_IOCTL_Command_struct *)
1559                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1560                         if (!ioc) {
1561                                 status = -ENOMEM;
1562                                 goto cleanup1;
1563                         }
1564                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1565                                 status = -EFAULT;
1566                                 goto cleanup1;
1567                         }
1568                         if ((ioc->buf_size < 1) &&
1569                             (ioc->Request.Type.Direction != XFER_NONE)) {
1570                                 status = -EINVAL;
1571                                 goto cleanup1;
1572                         }
1573                         /* Check kmalloc limits  using all SGs */
1574                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1575                                 status = -EINVAL;
1576                                 goto cleanup1;
1577                         }
1578                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1579                                 status = -EINVAL;
1580                                 goto cleanup1;
1581                         }
1582                         buff =
1583                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1584                         if (!buff) {
1585                                 status = -ENOMEM;
1586                                 goto cleanup1;
1587                         }
1588                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1589                                                    GFP_KERNEL);
1590                         if (!buff_size) {
1591                                 status = -ENOMEM;
1592                                 goto cleanup1;
1593                         }
1594                         left = ioc->buf_size;
1595                         data_ptr = ioc->buf;
1596                         while (left) {
1597                                 sz = (left >
1598                                       ioc->malloc_size) ? ioc->
1599                                     malloc_size : left;
1600                                 buff_size[sg_used] = sz;
1601                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1602                                 if (buff[sg_used] == NULL) {
1603                                         status = -ENOMEM;
1604                                         goto cleanup1;
1605                                 }
1606                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1607                                         if (copy_from_user
1608                                             (buff[sg_used], data_ptr, sz)) {
1609                                                 status = -EFAULT;
1610                                                 goto cleanup1;
1611                                         }
1612                                 } else {
1613                                         memset(buff[sg_used], 0, sz);
1614                                 }
1615                                 left -= sz;
1616                                 data_ptr += sz;
1617                                 sg_used++;
1618                         }
1619                         c = cmd_special_alloc(h);
1620                         if (!c) {
1621                                 status = -ENOMEM;
1622                                 goto cleanup1;
1623                         }
1624                         c->cmd_type = CMD_IOCTL_PEND;
1625                         c->Header.ReplyQueue = 0;
1626
1627                         if (ioc->buf_size > 0) {
1628                                 c->Header.SGList = sg_used;
1629                                 c->Header.SGTotal = sg_used;
1630                         } else {
1631                                 c->Header.SGList = 0;
1632                                 c->Header.SGTotal = 0;
1633                         }
1634                         c->Header.LUN = ioc->LUN_info;
1635                         c->Header.Tag.lower = c->busaddr;
1636
1637                         c->Request = ioc->Request;
1638                         if (ioc->buf_size > 0) {
1639                                 for (i = 0; i < sg_used; i++) {
1640                                         temp64.val =
1641                                             pci_map_single(h->pdev, buff[i],
1642                                                     buff_size[i],
1643                                                     PCI_DMA_BIDIRECTIONAL);
1644                                         c->SG[i].Addr.lower =
1645                                             temp64.val32.lower;
1646                                         c->SG[i].Addr.upper =
1647                                             temp64.val32.upper;
1648                                         c->SG[i].Len = buff_size[i];
1649                                         c->SG[i].Ext = 0;       /* we are not chaining */
1650                                 }
1651                         }
1652                         c->waiting = &wait;
1653                         enqueue_cmd_and_start_io(h, c);
1654                         wait_for_completion(&wait);
1655                         /* unlock the buffers from DMA */
1656                         for (i = 0; i < sg_used; i++) {
1657                                 temp64.val32.lower = c->SG[i].Addr.lower;
1658                                 temp64.val32.upper = c->SG[i].Addr.upper;
1659                                 pci_unmap_single(h->pdev,
1660                                         (dma_addr_t) temp64.val, buff_size[i],
1661                                         PCI_DMA_BIDIRECTIONAL);
1662                         }
1663                         check_ioctl_unit_attention(h, c);
1664                         /* Copy the error information out */
1665                         ioc->error_info = *(c->err_info);
1666                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1667                                 cmd_special_free(h, c);
1668                                 status = -EFAULT;
1669                                 goto cleanup1;
1670                         }
1671                         if (ioc->Request.Type.Direction == XFER_READ) {
1672                                 /* Copy the data out of the buffer we created */
1673                                 BYTE __user *ptr = ioc->buf;
1674                                 for (i = 0; i < sg_used; i++) {
1675                                         if (copy_to_user
1676                                             (ptr, buff[i], buff_size[i])) {
1677                                                 cmd_special_free(h, c);
1678                                                 status = -EFAULT;
1679                                                 goto cleanup1;
1680                                         }
1681                                         ptr += buff_size[i];
1682                                 }
1683                         }
1684                         cmd_special_free(h, c);
1685                         status = 0;
1686                       cleanup1:
1687                         if (buff) {
1688                                 for (i = 0; i < sg_used; i++)
1689                                         kfree(buff[i]);
1690                                 kfree(buff);
1691                         }
1692                         kfree(buff_size);
1693                         kfree(ioc);
1694                         return status;
1695                 }
1696
1697         /* scsi_cmd_ioctl handles these, below, though some are not */
1698         /* very meaningful for cciss.  SG_IO is the main one people want. */
1699
1700         case SG_GET_VERSION_NUM:
1701         case SG_SET_TIMEOUT:
1702         case SG_GET_TIMEOUT:
1703         case SG_GET_RESERVED_SIZE:
1704         case SG_SET_RESERVED_SIZE:
1705         case SG_EMULATED_HOST:
1706         case SG_IO:
1707         case SCSI_IOCTL_SEND_COMMAND:
1708                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1709
1710         /* scsi_cmd_ioctl would normally handle these, below, but */
1711         /* they aren't a good fit for cciss, as CD-ROMs are */
1712         /* not supported, and we don't have any bus/target/lun */
1713         /* which we present to the kernel. */
1714
1715         case CDROM_SEND_PACKET:
1716         case CDROMCLOSETRAY:
1717         case CDROMEJECT:
1718         case SCSI_IOCTL_GET_IDLUN:
1719         case SCSI_IOCTL_GET_BUS_NUMBER:
1720         default:
1721                 return -ENOTTY;
1722         }
1723 }
1724
1725 static void cciss_check_queues(ctlr_info_t *h)
1726 {
1727         int start_queue = h->next_to_run;
1728         int i;
1729
1730         /* check to see if we have maxed out the number of commands that can
1731          * be placed on the queue.  If so then exit.  We do this check here
1732          * in case the interrupt we serviced was from an ioctl and did not
1733          * free any new commands.
1734          */
1735         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1736                 return;
1737
1738         /* We have room on the queue for more commands.  Now we need to queue
1739          * them up.  We will also keep track of the next queue to run so
1740          * that every queue gets a chance to be started first.
1741          */
1742         for (i = 0; i < h->highest_lun + 1; i++) {
1743                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1744                 /* make sure the disk has been added and the drive is real
1745                  * because this can be called from the middle of init_one.
1746                  */
1747                 if (!h->drv[curr_queue])
1748                         continue;
1749                 if (!(h->drv[curr_queue]->queue) ||
1750                         !(h->drv[curr_queue]->heads))
1751                         continue;
1752                 blk_start_queue(h->gendisk[curr_queue]->queue);
1753
1754                 /* check to see if we have maxed out the number of commands
1755                  * that can be placed on the queue.
1756                  */
1757                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1758                         if (curr_queue == start_queue) {
1759                                 h->next_to_run =
1760                                     (start_queue + 1) % (h->highest_lun + 1);
1761                                 break;
1762                         } else {
1763                                 h->next_to_run = curr_queue;
1764                                 break;
1765                         }
1766                 }
1767         }
1768 }
1769
1770 static void cciss_softirq_done(struct request *rq)
1771 {
1772         CommandList_struct *c = rq->completion_data;
1773         ctlr_info_t *h = hba[c->ctlr];
1774         SGDescriptor_struct *curr_sg = c->SG;
1775         u64bit temp64;
1776         unsigned long flags;
1777         int i, ddir;
1778         int sg_index = 0;
1779
1780         if (c->Request.Type.Direction == XFER_READ)
1781                 ddir = PCI_DMA_FROMDEVICE;
1782         else
1783                 ddir = PCI_DMA_TODEVICE;
1784
1785         /* command did not need to be retried */
1786         /* unmap the DMA mapping for all the scatter gather elements */
1787         for (i = 0; i < c->Header.SGList; i++) {
1788                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1789                         cciss_unmap_sg_chain_block(h, c);
1790                         /* Point to the next block */
1791                         curr_sg = h->cmd_sg_list[c->cmdindex];
1792                         sg_index = 0;
1793                 }
1794                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1795                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1796                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1797                                 ddir);
1798                 ++sg_index;
1799         }
1800
1801         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1802
1803         /* set the residual count for pc requests */
1804         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1805                 rq->resid_len = c->err_info->ResidualCnt;
1806
1807         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1808
1809         spin_lock_irqsave(&h->lock, flags);
1810         cmd_free(h, c);
1811         cciss_check_queues(h);
1812         spin_unlock_irqrestore(&h->lock, flags);
1813 }
1814
1815 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1816         unsigned char scsi3addr[], uint32_t log_unit)
1817 {
1818         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1819                 sizeof(h->drv[log_unit]->LunID));
1820 }
1821
1822 /* This function gets the SCSI vendor, model, and revision of a logical drive
1823  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1824  * they cannot be read.
1825  */
1826 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1827                                    char *vendor, char *model, char *rev)
1828 {
1829         int rc;
1830         InquiryData_struct *inq_buf;
1831         unsigned char scsi3addr[8];
1832
1833         *vendor = '\0';
1834         *model = '\0';
1835         *rev = '\0';
1836
1837         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1838         if (!inq_buf)
1839                 return;
1840
1841         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1842         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1843                         scsi3addr, TYPE_CMD);
1844         if (rc == IO_OK) {
1845                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1846                 vendor[VENDOR_LEN] = '\0';
1847                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1848                 model[MODEL_LEN] = '\0';
1849                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1850                 rev[REV_LEN] = '\0';
1851         }
1852
1853         kfree(inq_buf);
1854         return;
1855 }
1856
1857 /* This function gets the serial number of a logical drive via
1858  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1859  * number cannot be had, for whatever reason, 16 bytes of 0xff
1860  * are returned instead.
1861  */
1862 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1863                                 unsigned char *serial_no, int buflen)
1864 {
1865 #define PAGE_83_INQ_BYTES 64
1866         int rc;
1867         unsigned char *buf;
1868         unsigned char scsi3addr[8];
1869
1870         if (buflen > 16)
1871                 buflen = 16;
1872         memset(serial_no, 0xff, buflen);
1873         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1874         if (!buf)
1875                 return;
1876         memset(serial_no, 0, buflen);
1877         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1878         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1879                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1880         if (rc == IO_OK)
1881                 memcpy(serial_no, &buf[8], buflen);
1882         kfree(buf);
1883         return;
1884 }
1885
1886 /*
1887  * cciss_add_disk sets up the block device queue for a logical drive
1888  */
1889 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1890                                 int drv_index)
1891 {
1892         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1893         if (!disk->queue)
1894                 goto init_queue_failure;
1895         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1896         disk->major = h->major;
1897         disk->first_minor = drv_index << NWD_SHIFT;
1898         disk->fops = &cciss_fops;
1899         if (cciss_create_ld_sysfs_entry(h, drv_index))
1900                 goto cleanup_queue;
1901         disk->private_data = h->drv[drv_index];
1902         disk->driverfs_dev = &h->drv[drv_index]->dev;
1903
1904         /* Set up queue information */
1905         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1906
1907         /* This is a hardware imposed limit. */
1908         blk_queue_max_segments(disk->queue, h->maxsgentries);
1909
1910         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1911
1912         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1913
1914         disk->queue->queuedata = h;
1915
1916         blk_queue_logical_block_size(disk->queue,
1917                                      h->drv[drv_index]->block_size);
1918
1919         /* Make sure all queue data is written out before */
1920         /* setting h->drv[drv_index]->queue, as setting this */
1921         /* allows the interrupt handler to start the queue */
1922         wmb();
1923         h->drv[drv_index]->queue = disk->queue;
1924         add_disk(disk);
1925         return 0;
1926
1927 cleanup_queue:
1928         blk_cleanup_queue(disk->queue);
1929         disk->queue = NULL;
1930 init_queue_failure:
1931         return -1;
1932 }
1933
1934 /* This function will check the usage_count of the drive to be updated/added.
1935  * If the usage_count is zero and it is a heretofore unknown drive, or,
1936  * the drive's capacity, geometry, or serial number has changed,
1937  * then the drive information will be updated and the disk will be
1938  * re-registered with the kernel.  If these conditions don't hold,
1939  * then it will be left alone for the next reboot.  The exception to this
1940  * is disk 0 which will always be left registered with the kernel since it
1941  * is also the controller node.  Any changes to disk 0 will show up on
1942  * the next reboot.
1943  */
1944 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1945         int first_time, int via_ioctl)
1946 {
1947         struct gendisk *disk;
1948         InquiryData_struct *inq_buff = NULL;
1949         unsigned int block_size;
1950         sector_t total_size;
1951         unsigned long flags = 0;
1952         int ret = 0;
1953         drive_info_struct *drvinfo;
1954
1955         /* Get information about the disk and modify the driver structure */
1956         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1957         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1958         if (inq_buff == NULL || drvinfo == NULL)
1959                 goto mem_msg;
1960
1961         /* testing to see if 16-byte CDBs are already being used */
1962         if (h->cciss_read == CCISS_READ_16) {
1963                 cciss_read_capacity_16(h, drv_index,
1964                         &total_size, &block_size);
1965
1966         } else {
1967                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1968                 /* if read_capacity returns all F's this volume is >2TB */
1969                 /* in size so we switch to 16-byte CDB's for all */
1970                 /* read/write ops */
1971                 if (total_size == 0xFFFFFFFFULL) {
1972                         cciss_read_capacity_16(h, drv_index,
1973                         &total_size, &block_size);
1974                         h->cciss_read = CCISS_READ_16;
1975                         h->cciss_write = CCISS_WRITE_16;
1976                 } else {
1977                         h->cciss_read = CCISS_READ_10;
1978                         h->cciss_write = CCISS_WRITE_10;
1979                 }
1980         }
1981
1982         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1983                                inq_buff, drvinfo);
1984         drvinfo->block_size = block_size;
1985         drvinfo->nr_blocks = total_size + 1;
1986
1987         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1988                                 drvinfo->model, drvinfo->rev);
1989         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1990                         sizeof(drvinfo->serial_no));
1991         /* Save the lunid in case we deregister the disk, below. */
1992         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1993                 sizeof(drvinfo->LunID));
1994
1995         /* Is it the same disk we already know, and nothing's changed? */
1996         if (h->drv[drv_index]->raid_level != -1 &&
1997                 ((memcmp(drvinfo->serial_no,
1998                                 h->drv[drv_index]->serial_no, 16) == 0) &&
1999                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2000                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2001                 drvinfo->heads == h->drv[drv_index]->heads &&
2002                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2003                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2004                         /* The disk is unchanged, nothing to update */
2005                         goto freeret;
2006
2007         /* If we get here it's not the same disk, or something's changed,
2008          * so we need to * deregister it, and re-register it, if it's not
2009          * in use.
2010          * If the disk already exists then deregister it before proceeding
2011          * (unless it's the first disk (for the controller node).
2012          */
2013         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2014                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2015                 spin_lock_irqsave(&h->lock, flags);
2016                 h->drv[drv_index]->busy_configuring = 1;
2017                 spin_unlock_irqrestore(&h->lock, flags);
2018
2019                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2020                  * which keeps the interrupt handler from starting
2021                  * the queue.
2022                  */
2023                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2024         }
2025
2026         /* If the disk is in use return */
2027         if (ret)
2028                 goto freeret;
2029
2030         /* Save the new information from cciss_geometry_inquiry
2031          * and serial number inquiry.  If the disk was deregistered
2032          * above, then h->drv[drv_index] will be NULL.
2033          */
2034         if (h->drv[drv_index] == NULL) {
2035                 drvinfo->device_initialized = 0;
2036                 h->drv[drv_index] = drvinfo;
2037                 drvinfo = NULL; /* so it won't be freed below. */
2038         } else {
2039                 /* special case for cxd0 */
2040                 h->drv[drv_index]->block_size = drvinfo->block_size;
2041                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2042                 h->drv[drv_index]->heads = drvinfo->heads;
2043                 h->drv[drv_index]->sectors = drvinfo->sectors;
2044                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2045                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2046                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2047                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2048                         VENDOR_LEN + 1);
2049                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2050                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2051         }
2052
2053         ++h->num_luns;
2054         disk = h->gendisk[drv_index];
2055         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2056
2057         /* If it's not disk 0 (drv_index != 0)
2058          * or if it was disk 0, but there was previously
2059          * no actual corresponding configured logical drive
2060          * (raid_leve == -1) then we want to update the
2061          * logical drive's information.
2062          */
2063         if (drv_index || first_time) {
2064                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2065                         cciss_free_gendisk(h, drv_index);
2066                         cciss_free_drive_info(h, drv_index);
2067                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2068                                 drv_index);
2069                         --h->num_luns;
2070                 }
2071         }
2072
2073 freeret:
2074         kfree(inq_buff);
2075         kfree(drvinfo);
2076         return;
2077 mem_msg:
2078         dev_err(&h->pdev->dev, "out of memory\n");
2079         goto freeret;
2080 }
2081
2082 /* This function will find the first index of the controllers drive array
2083  * that has a null drv pointer and allocate the drive info struct and
2084  * will return that index   This is where new drives will be added.
2085  * If the index to be returned is greater than the highest_lun index for
2086  * the controller then highest_lun is set * to this new index.
2087  * If there are no available indexes or if tha allocation fails, then -1
2088  * is returned.  * "controller_node" is used to know if this is a real
2089  * logical drive, or just the controller node, which determines if this
2090  * counts towards highest_lun.
2091  */
2092 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2093 {
2094         int i;
2095         drive_info_struct *drv;
2096
2097         /* Search for an empty slot for our drive info */
2098         for (i = 0; i < CISS_MAX_LUN; i++) {
2099
2100                 /* if not cxd0 case, and it's occupied, skip it. */
2101                 if (h->drv[i] && i != 0)
2102                         continue;
2103                 /*
2104                  * If it's cxd0 case, and drv is alloc'ed already, and a
2105                  * disk is configured there, skip it.
2106                  */
2107                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2108                         continue;
2109
2110                 /*
2111                  * We've found an empty slot.  Update highest_lun
2112                  * provided this isn't just the fake cxd0 controller node.
2113                  */
2114                 if (i > h->highest_lun && !controller_node)
2115                         h->highest_lun = i;
2116
2117                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2118                 if (i == 0 && h->drv[i] != NULL)
2119                         return i;
2120
2121                 /*
2122                  * Found an empty slot, not already alloc'ed.  Allocate it.
2123                  * Mark it with raid_level == -1, so we know it's new later on.
2124                  */
2125                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2126                 if (!drv)
2127                         return -1;
2128                 drv->raid_level = -1; /* so we know it's new */
2129                 h->drv[i] = drv;
2130                 return i;
2131         }
2132         return -1;
2133 }
2134
2135 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2136 {
2137         kfree(h->drv[drv_index]);
2138         h->drv[drv_index] = NULL;
2139 }
2140
2141 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2142 {
2143         put_disk(h->gendisk[drv_index]);
2144         h->gendisk[drv_index] = NULL;
2145 }
2146
2147 /* cciss_add_gendisk finds a free hba[]->drv structure
2148  * and allocates a gendisk if needed, and sets the lunid
2149  * in the drvinfo structure.   It returns the index into
2150  * the ->drv[] array, or -1 if none are free.
2151  * is_controller_node indicates whether highest_lun should
2152  * count this disk, or if it's only being added to provide
2153  * a means to talk to the controller in case no logical
2154  * drives have yet been configured.
2155  */
2156 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2157         int controller_node)
2158 {
2159         int drv_index;
2160
2161         drv_index = cciss_alloc_drive_info(h, controller_node);
2162         if (drv_index == -1)
2163                 return -1;
2164
2165         /*Check if the gendisk needs to be allocated */
2166         if (!h->gendisk[drv_index]) {
2167                 h->gendisk[drv_index] =
2168                         alloc_disk(1 << NWD_SHIFT);
2169                 if (!h->gendisk[drv_index]) {
2170                         dev_err(&h->pdev->dev,
2171                                 "could not allocate a new disk %d\n",
2172                                 drv_index);
2173                         goto err_free_drive_info;
2174                 }
2175         }
2176         memcpy(h->drv[drv_index]->LunID, lunid,
2177                 sizeof(h->drv[drv_index]->LunID));
2178         if (cciss_create_ld_sysfs_entry(h, drv_index))
2179                 goto err_free_disk;
2180         /* Don't need to mark this busy because nobody */
2181         /* else knows about this disk yet to contend */
2182         /* for access to it. */
2183         h->drv[drv_index]->busy_configuring = 0;
2184         wmb();
2185         return drv_index;
2186
2187 err_free_disk:
2188         cciss_free_gendisk(h, drv_index);
2189 err_free_drive_info:
2190         cciss_free_drive_info(h, drv_index);
2191         return -1;
2192 }
2193
2194 /* This is for the special case of a controller which
2195  * has no logical drives.  In this case, we still need
2196  * to register a disk so the controller can be accessed
2197  * by the Array Config Utility.
2198  */
2199 static void cciss_add_controller_node(ctlr_info_t *h)
2200 {
2201         struct gendisk *disk;
2202         int drv_index;
2203
2204         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2205                 return;
2206
2207         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2208         if (drv_index == -1)
2209                 goto error;
2210         h->drv[drv_index]->block_size = 512;
2211         h->drv[drv_index]->nr_blocks = 0;
2212         h->drv[drv_index]->heads = 0;
2213         h->drv[drv_index]->sectors = 0;
2214         h->drv[drv_index]->cylinders = 0;
2215         h->drv[drv_index]->raid_level = -1;
2216         memset(h->drv[drv_index]->serial_no, 0, 16);
2217         disk = h->gendisk[drv_index];
2218         if (cciss_add_disk(h, disk, drv_index) == 0)
2219                 return;
2220         cciss_free_gendisk(h, drv_index);
2221         cciss_free_drive_info(h, drv_index);
2222 error:
2223         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2224         return;
2225 }
2226
2227 /* This function will add and remove logical drives from the Logical
2228  * drive array of the controller and maintain persistency of ordering
2229  * so that mount points are preserved until the next reboot.  This allows
2230  * for the removal of logical drives in the middle of the drive array
2231  * without a re-ordering of those drives.
2232  * INPUT
2233  * h            = The controller to perform the operations on
2234  */
2235 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2236         int via_ioctl)
2237 {
2238         int num_luns;
2239         ReportLunData_struct *ld_buff = NULL;
2240         int return_code;
2241         int listlength = 0;
2242         int i;
2243         int drv_found;
2244         int drv_index = 0;
2245         unsigned char lunid[8] = CTLR_LUNID;
2246         unsigned long flags;
2247
2248         if (!capable(CAP_SYS_RAWIO))
2249                 return -EPERM;
2250
2251         /* Set busy_configuring flag for this operation */
2252         spin_lock_irqsave(&h->lock, flags);
2253         if (h->busy_configuring) {
2254                 spin_unlock_irqrestore(&h->lock, flags);
2255                 return -EBUSY;
2256         }
2257         h->busy_configuring = 1;
2258         spin_unlock_irqrestore(&h->lock, flags);
2259
2260         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2261         if (ld_buff == NULL)
2262                 goto mem_msg;
2263
2264         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2265                                       sizeof(ReportLunData_struct),
2266                                       0, CTLR_LUNID, TYPE_CMD);
2267
2268         if (return_code == IO_OK)
2269                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2270         else {  /* reading number of logical volumes failed */
2271                 dev_warn(&h->pdev->dev,
2272                         "report logical volume command failed\n");
2273                 listlength = 0;
2274                 goto freeret;
2275         }
2276
2277         num_luns = listlength / 8;      /* 8 bytes per entry */
2278         if (num_luns > CISS_MAX_LUN) {
2279                 num_luns = CISS_MAX_LUN;
2280                 dev_warn(&h->pdev->dev, "more luns configured"
2281                        " on controller than can be handled by"
2282                        " this driver.\n");
2283         }
2284
2285         if (num_luns == 0)
2286                 cciss_add_controller_node(h);
2287
2288         /* Compare controller drive array to driver's drive array
2289          * to see if any drives are missing on the controller due
2290          * to action of Array Config Utility (user deletes drive)
2291          * and deregister logical drives which have disappeared.
2292          */
2293         for (i = 0; i <= h->highest_lun; i++) {
2294                 int j;
2295                 drv_found = 0;
2296
2297                 /* skip holes in the array from already deleted drives */
2298                 if (h->drv[i] == NULL)
2299                         continue;
2300
2301                 for (j = 0; j < num_luns; j++) {
2302                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2303                         if (memcmp(h->drv[i]->LunID, lunid,
2304                                 sizeof(lunid)) == 0) {
2305                                 drv_found = 1;
2306                                 break;
2307                         }
2308                 }
2309                 if (!drv_found) {
2310                         /* Deregister it from the OS, it's gone. */
2311                         spin_lock_irqsave(&h->lock, flags);
2312                         h->drv[i]->busy_configuring = 1;
2313                         spin_unlock_irqrestore(&h->lock, flags);
2314                         return_code = deregister_disk(h, i, 1, via_ioctl);
2315                         if (h->drv[i] != NULL)
2316                                 h->drv[i]->busy_configuring = 0;
2317                 }
2318         }
2319
2320         /* Compare controller drive array to driver's drive array.
2321          * Check for updates in the drive information and any new drives
2322          * on the controller due to ACU adding logical drives, or changing
2323          * a logical drive's size, etc.  Reregister any new/changed drives
2324          */
2325         for (i = 0; i < num_luns; i++) {
2326                 int j;
2327
2328                 drv_found = 0;
2329
2330                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2331                 /* Find if the LUN is already in the drive array
2332                  * of the driver.  If so then update its info
2333                  * if not in use.  If it does not exist then find
2334                  * the first free index and add it.
2335                  */
2336                 for (j = 0; j <= h->highest_lun; j++) {
2337                         if (h->drv[j] != NULL &&
2338                                 memcmp(h->drv[j]->LunID, lunid,
2339                                         sizeof(h->drv[j]->LunID)) == 0) {
2340                                 drv_index = j;
2341                                 drv_found = 1;
2342                                 break;
2343                         }
2344                 }
2345
2346                 /* check if the drive was found already in the array */
2347                 if (!drv_found) {
2348                         drv_index = cciss_add_gendisk(h, lunid, 0);
2349                         if (drv_index == -1)
2350                                 goto freeret;
2351                 }
2352                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2353         }               /* end for */
2354
2355 freeret:
2356         kfree(ld_buff);
2357         h->busy_configuring = 0;
2358         /* We return -1 here to tell the ACU that we have registered/updated
2359          * all of the drives that we can and to keep it from calling us
2360          * additional times.
2361          */
2362         return -1;
2363 mem_msg:
2364         dev_err(&h->pdev->dev, "out of memory\n");
2365         h->busy_configuring = 0;
2366         goto freeret;
2367 }
2368
2369 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2370 {
2371         /* zero out the disk size info */
2372         drive_info->nr_blocks = 0;
2373         drive_info->block_size = 0;
2374         drive_info->heads = 0;
2375         drive_info->sectors = 0;
2376         drive_info->cylinders = 0;
2377         drive_info->raid_level = -1;
2378         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2379         memset(drive_info->model, 0, sizeof(drive_info->model));
2380         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2381         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2382         /*
2383          * don't clear the LUNID though, we need to remember which
2384          * one this one is.
2385          */
2386 }
2387
2388 /* This function will deregister the disk and it's queue from the
2389  * kernel.  It must be called with the controller lock held and the
2390  * drv structures busy_configuring flag set.  It's parameters are:
2391  *
2392  * disk = This is the disk to be deregistered
2393  * drv  = This is the drive_info_struct associated with the disk to be
2394  *        deregistered.  It contains information about the disk used
2395  *        by the driver.
2396  * clear_all = This flag determines whether or not the disk information
2397  *             is going to be completely cleared out and the highest_lun
2398  *             reset.  Sometimes we want to clear out information about
2399  *             the disk in preparation for re-adding it.  In this case
2400  *             the highest_lun should be left unchanged and the LunID
2401  *             should not be cleared.
2402  * via_ioctl
2403  *    This indicates whether we've reached this path via ioctl.
2404  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2405  *    If this path is reached via ioctl(), then the max_usage_count will
2406  *    be 1, as the process calling ioctl() has got to have the device open.
2407  *    If we get here via sysfs, then the max usage count will be zero.
2408 */
2409 static int deregister_disk(ctlr_info_t *h, int drv_index,
2410                            int clear_all, int via_ioctl)
2411 {
2412         int i;
2413         struct gendisk *disk;
2414         drive_info_struct *drv;
2415         int recalculate_highest_lun;
2416
2417         if (!capable(CAP_SYS_RAWIO))
2418                 return -EPERM;
2419
2420         drv = h->drv[drv_index];
2421         disk = h->gendisk[drv_index];
2422
2423         /* make sure logical volume is NOT is use */
2424         if (clear_all || (h->gendisk[0] == disk)) {
2425                 if (drv->usage_count > via_ioctl)
2426                         return -EBUSY;
2427         } else if (drv->usage_count > 0)
2428                 return -EBUSY;
2429
2430         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2431
2432         /* invalidate the devices and deregister the disk.  If it is disk
2433          * zero do not deregister it but just zero out it's values.  This
2434          * allows us to delete disk zero but keep the controller registered.
2435          */
2436         if (h->gendisk[0] != disk) {
2437                 struct request_queue *q = disk->queue;
2438                 if (disk->flags & GENHD_FL_UP) {
2439                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2440                         del_gendisk(disk);
2441                 }
2442                 if (q)
2443                         blk_cleanup_queue(q);
2444                 /* If clear_all is set then we are deleting the logical
2445                  * drive, not just refreshing its info.  For drives
2446                  * other than disk 0 we will call put_disk.  We do not
2447                  * do this for disk 0 as we need it to be able to
2448                  * configure the controller.
2449                  */
2450                 if (clear_all){
2451                         /* This isn't pretty, but we need to find the
2452                          * disk in our array and NULL our the pointer.
2453                          * This is so that we will call alloc_disk if
2454                          * this index is used again later.
2455                          */
2456                         for (i=0; i < CISS_MAX_LUN; i++){
2457                                 if (h->gendisk[i] == disk) {
2458                                         h->gendisk[i] = NULL;
2459                                         break;
2460                                 }
2461                         }
2462                         put_disk(disk);
2463                 }
2464         } else {
2465                 set_capacity(disk, 0);
2466                 cciss_clear_drive_info(drv);
2467         }
2468
2469         --h->num_luns;
2470
2471         /* if it was the last disk, find the new hightest lun */
2472         if (clear_all && recalculate_highest_lun) {
2473                 int newhighest = -1;
2474                 for (i = 0; i <= h->highest_lun; i++) {
2475                         /* if the disk has size > 0, it is available */
2476                         if (h->drv[i] && h->drv[i]->heads)
2477                                 newhighest = i;
2478                 }
2479                 h->highest_lun = newhighest;
2480         }
2481         return 0;
2482 }
2483
2484 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2485                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2486                 int cmd_type)
2487 {
2488         u64bit buff_dma_handle;
2489         int status = IO_OK;
2490
2491         c->cmd_type = CMD_IOCTL_PEND;
2492         c->Header.ReplyQueue = 0;
2493         if (buff != NULL) {
2494                 c->Header.SGList = 1;
2495                 c->Header.SGTotal = 1;
2496         } else {
2497                 c->Header.SGList = 0;
2498                 c->Header.SGTotal = 0;
2499         }
2500         c->Header.Tag.lower = c->busaddr;
2501         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2502
2503         c->Request.Type.Type = cmd_type;
2504         if (cmd_type == TYPE_CMD) {
2505                 switch (cmd) {
2506                 case CISS_INQUIRY:
2507                         /* are we trying to read a vital product page */
2508                         if (page_code != 0) {
2509                                 c->Request.CDB[1] = 0x01;
2510                                 c->Request.CDB[2] = page_code;
2511                         }
2512                         c->Request.CDBLen = 6;
2513                         c->Request.Type.Attribute = ATTR_SIMPLE;
2514                         c->Request.Type.Direction = XFER_READ;
2515                         c->Request.Timeout = 0;
2516                         c->Request.CDB[0] = CISS_INQUIRY;
2517                         c->Request.CDB[4] = size & 0xFF;
2518                         break;
2519                 case CISS_REPORT_LOG:
2520                 case CISS_REPORT_PHYS:
2521                         /* Talking to controller so It's a physical command
2522                            mode = 00 target = 0.  Nothing to write.
2523                          */
2524                         c->Request.CDBLen = 12;
2525                         c->Request.Type.Attribute = ATTR_SIMPLE;
2526                         c->Request.Type.Direction = XFER_READ;
2527                         c->Request.Timeout = 0;
2528                         c->Request.CDB[0] = cmd;
2529                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2530                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2531                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2532                         c->Request.CDB[9] = size & 0xFF;
2533                         break;
2534
2535                 case CCISS_READ_CAPACITY:
2536                         c->Request.CDBLen = 10;
2537                         c->Request.Type.Attribute = ATTR_SIMPLE;
2538                         c->Request.Type.Direction = XFER_READ;
2539                         c->Request.Timeout = 0;
2540                         c->Request.CDB[0] = cmd;
2541                         break;
2542                 case CCISS_READ_CAPACITY_16:
2543                         c->Request.CDBLen = 16;
2544                         c->Request.Type.Attribute = ATTR_SIMPLE;
2545                         c->Request.Type.Direction = XFER_READ;
2546                         c->Request.Timeout = 0;
2547                         c->Request.CDB[0] = cmd;
2548                         c->Request.CDB[1] = 0x10;
2549                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2550                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2551                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2552                         c->Request.CDB[13] = size & 0xFF;
2553                         c->Request.Timeout = 0;
2554                         c->Request.CDB[0] = cmd;
2555                         break;
2556                 case CCISS_CACHE_FLUSH:
2557                         c->Request.CDBLen = 12;
2558                         c->Request.Type.Attribute = ATTR_SIMPLE;
2559                         c->Request.Type.Direction = XFER_WRITE;
2560                         c->Request.Timeout = 0;
2561                         c->Request.CDB[0] = BMIC_WRITE;
2562                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2563                         break;
2564                 case TEST_UNIT_READY:
2565                         c->Request.CDBLen = 6;
2566                         c->Request.Type.Attribute = ATTR_SIMPLE;
2567                         c->Request.Type.Direction = XFER_NONE;
2568                         c->Request.Timeout = 0;
2569                         break;
2570                 default:
2571                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2572                         return IO_ERROR;
2573                 }
2574         } else if (cmd_type == TYPE_MSG) {
2575                 switch (cmd) {
2576                 case 0: /* ABORT message */
2577                         c->Request.CDBLen = 12;
2578                         c->Request.Type.Attribute = ATTR_SIMPLE;
2579                         c->Request.Type.Direction = XFER_WRITE;
2580                         c->Request.Timeout = 0;
2581                         c->Request.CDB[0] = cmd;        /* abort */
2582                         c->Request.CDB[1] = 0;  /* abort a command */
2583                         /* buff contains the tag of the command to abort */
2584                         memcpy(&c->Request.CDB[4], buff, 8);
2585                         break;
2586                 case 1: /* RESET message */
2587                         c->Request.CDBLen = 16;
2588                         c->Request.Type.Attribute = ATTR_SIMPLE;
2589                         c->Request.Type.Direction = XFER_NONE;
2590                         c->Request.Timeout = 0;
2591                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2592                         c->Request.CDB[0] = cmd;        /* reset */
2593                         c->Request.CDB[1] = 0x03;       /* reset a target */
2594                         break;
2595                 case 3: /* No-Op message */
2596                         c->Request.CDBLen = 1;
2597                         c->Request.Type.Attribute = ATTR_SIMPLE;
2598                         c->Request.Type.Direction = XFER_WRITE;
2599                         c->Request.Timeout = 0;
2600                         c->Request.CDB[0] = cmd;
2601                         break;
2602                 default:
2603                         dev_warn(&h->pdev->dev,
2604                                 "unknown message type %d\n", cmd);
2605                         return IO_ERROR;
2606                 }
2607         } else {
2608                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2609                 return IO_ERROR;
2610         }
2611         /* Fill in the scatter gather information */
2612         if (size > 0) {
2613                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2614                                                              buff, size,
2615                                                              PCI_DMA_BIDIRECTIONAL);
2616                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2617                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2618                 c->SG[0].Len = size;
2619                 c->SG[0].Ext = 0;       /* we are not chaining */
2620         }
2621         return status;
2622 }
2623
2624 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2625 {
2626         switch (c->err_info->ScsiStatus) {
2627         case SAM_STAT_GOOD:
2628                 return IO_OK;
2629         case SAM_STAT_CHECK_CONDITION:
2630                 switch (0xf & c->err_info->SenseInfo[2]) {
2631                 case 0: return IO_OK; /* no sense */
2632                 case 1: return IO_OK; /* recovered error */
2633                 default:
2634                         if (check_for_unit_attention(h, c))
2635                                 return IO_NEEDS_RETRY;
2636                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2637                                 "check condition, sense key = 0x%02x\n",
2638                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2639                 }
2640                 break;
2641         default:
2642                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2643                         "scsi status = 0x%02x\n",
2644                         c->Request.CDB[0], c->err_info->ScsiStatus);
2645                 break;
2646         }
2647         return IO_ERROR;
2648 }
2649
2650 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2651 {
2652         int return_status = IO_OK;
2653
2654         if (c->err_info->CommandStatus == CMD_SUCCESS)
2655                 return IO_OK;
2656
2657         switch (c->err_info->CommandStatus) {
2658         case CMD_TARGET_STATUS:
2659                 return_status = check_target_status(h, c);
2660                 break;
2661         case CMD_DATA_UNDERRUN:
2662         case CMD_DATA_OVERRUN:
2663                 /* expected for inquiry and report lun commands */
2664                 break;
2665         case CMD_INVALID:
2666                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2667                        "reported invalid\n", c->Request.CDB[0]);
2668                 return_status = IO_ERROR;
2669                 break;
2670         case CMD_PROTOCOL_ERR:
2671                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2672                        "protocol error\n", c->Request.CDB[0]);
2673                 return_status = IO_ERROR;
2674                 break;
2675         case CMD_HARDWARE_ERR:
2676                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2677                        " hardware error\n", c->Request.CDB[0]);
2678                 return_status = IO_ERROR;
2679                 break;
2680         case CMD_CONNECTION_LOST:
2681                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2682                        "connection lost\n", c->Request.CDB[0]);
2683                 return_status = IO_ERROR;
2684                 break;
2685         case CMD_ABORTED:
2686                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2687                        "aborted\n", c->Request.CDB[0]);
2688                 return_status = IO_ERROR;
2689                 break;
2690         case CMD_ABORT_FAILED:
2691                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2692                        "abort failed\n", c->Request.CDB[0]);
2693                 return_status = IO_ERROR;
2694                 break;
2695         case CMD_UNSOLICITED_ABORT:
2696                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2697                         c->Request.CDB[0]);
2698                 return_status = IO_NEEDS_RETRY;
2699                 break;
2700         default:
2701                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2702                        "unknown status %x\n", c->Request.CDB[0],
2703                        c->err_info->CommandStatus);
2704                 return_status = IO_ERROR;
2705         }
2706         return return_status;
2707 }
2708
2709 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2710         int attempt_retry)
2711 {
2712         DECLARE_COMPLETION_ONSTACK(wait);
2713         u64bit buff_dma_handle;
2714         int return_status = IO_OK;
2715
2716 resend_cmd2:
2717         c->waiting = &wait;
2718         enqueue_cmd_and_start_io(h, c);
2719
2720         wait_for_completion(&wait);
2721
2722         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2723                 goto command_done;
2724
2725         return_status = process_sendcmd_error(h, c);
2726
2727         if (return_status == IO_NEEDS_RETRY &&
2728                 c->retry_count < MAX_CMD_RETRIES) {
2729                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2730                         c->Request.CDB[0]);
2731                 c->retry_count++;
2732                 /* erase the old error information */
2733                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2734                 return_status = IO_OK;
2735                 INIT_COMPLETION(wait);
2736                 goto resend_cmd2;
2737         }
2738
2739 command_done:
2740         /* unlock the buffers from DMA */
2741         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2742         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2743         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2744                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2745         return return_status;
2746 }
2747
2748 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2749                            __u8 page_code, unsigned char scsi3addr[],
2750                         int cmd_type)
2751 {
2752         CommandList_struct *c;
2753         int return_status;
2754
2755         c = cmd_special_alloc(h);
2756         if (!c)
2757                 return -ENOMEM;
2758         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2759                 scsi3addr, cmd_type);
2760         if (return_status == IO_OK)
2761                 return_status = sendcmd_withirq_core(h, c, 1);
2762
2763         cmd_special_free(h, c);
2764         return return_status;
2765 }
2766
2767 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2768                                    sector_t total_size,
2769                                    unsigned int block_size,
2770                                    InquiryData_struct *inq_buff,
2771                                    drive_info_struct *drv)
2772 {
2773         int return_code;
2774         unsigned long t;
2775         unsigned char scsi3addr[8];
2776
2777         memset(inq_buff, 0, sizeof(InquiryData_struct));
2778         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2779         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2780                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2781         if (return_code == IO_OK) {
2782                 if (inq_buff->data_byte[8] == 0xFF) {
2783                         dev_warn(&h->pdev->dev,
2784                                "reading geometry failed, volume "
2785                                "does not support reading geometry\n");
2786                         drv->heads = 255;
2787                         drv->sectors = 32;      /* Sectors per track */
2788                         drv->cylinders = total_size + 1;
2789                         drv->raid_level = RAID_UNKNOWN;
2790                 } else {
2791                         drv->heads = inq_buff->data_byte[6];
2792                         drv->sectors = inq_buff->data_byte[7];
2793                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2794                         drv->cylinders += inq_buff->data_byte[5];
2795                         drv->raid_level = inq_buff->data_byte[8];
2796                 }
2797                 drv->block_size = block_size;
2798                 drv->nr_blocks = total_size + 1;
2799                 t = drv->heads * drv->sectors;
2800                 if (t > 1) {
2801                         sector_t real_size = total_size + 1;
2802                         unsigned long rem = sector_div(real_size, t);
2803                         if (rem)
2804                                 real_size++;
2805                         drv->cylinders = real_size;
2806                 }
2807         } else {                /* Get geometry failed */
2808                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2809         }
2810 }
2811
2812 static void
2813 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2814                     unsigned int *block_size)
2815 {
2816         ReadCapdata_struct *buf;
2817         int return_code;
2818         unsigned char scsi3addr[8];
2819
2820         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2821         if (!buf) {
2822                 dev_warn(&h->pdev->dev, "out of memory\n");
2823                 return;
2824         }
2825
2826         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2827         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2828                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2829         if (return_code == IO_OK) {
2830                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2831                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2832         } else {                /* read capacity command failed */
2833                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2834                 *total_size = 0;
2835                 *block_size = BLOCK_SIZE;
2836         }
2837         kfree(buf);
2838 }
2839
2840 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2841         sector_t *total_size, unsigned int *block_size)
2842 {
2843         ReadCapdata_struct_16 *buf;
2844         int return_code;
2845         unsigned char scsi3addr[8];
2846
2847         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2848         if (!buf) {
2849                 dev_warn(&h->pdev->dev, "out of memory\n");
2850                 return;
2851         }
2852
2853         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2854         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2855                 buf, sizeof(ReadCapdata_struct_16),
2856                         0, scsi3addr, TYPE_CMD);
2857         if (return_code == IO_OK) {
2858                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2859                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2860         } else {                /* read capacity command failed */
2861                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2862                 *total_size = 0;
2863                 *block_size = BLOCK_SIZE;
2864         }
2865         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2866                (unsigned long long)*total_size+1, *block_size);
2867         kfree(buf);
2868 }
2869
2870 static int cciss_revalidate(struct gendisk *disk)
2871 {
2872         ctlr_info_t *h = get_host(disk);
2873         drive_info_struct *drv = get_drv(disk);
2874         int logvol;
2875         int FOUND = 0;
2876         unsigned int block_size;
2877         sector_t total_size;
2878         InquiryData_struct *inq_buff = NULL;
2879
2880         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2881                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2882                         sizeof(drv->LunID)) == 0) {
2883                         FOUND = 1;
2884                         break;
2885                 }
2886         }
2887
2888         if (!FOUND)
2889                 return 1;
2890
2891         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2892         if (inq_buff == NULL) {
2893                 dev_warn(&h->pdev->dev, "out of memory\n");
2894                 return 1;
2895         }
2896         if (h->cciss_read == CCISS_READ_10) {
2897                 cciss_read_capacity(h, logvol,
2898                                         &total_size, &block_size);
2899         } else {
2900                 cciss_read_capacity_16(h, logvol,
2901                                         &total_size, &block_size);
2902         }
2903         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2904                                inq_buff, drv);
2905
2906         blk_queue_logical_block_size(drv->queue, drv->block_size);
2907         set_capacity(disk, drv->nr_blocks);
2908
2909         kfree(inq_buff);
2910         return 0;
2911 }
2912
2913 /*
2914  * Map (physical) PCI mem into (virtual) kernel space
2915  */
2916 static void __iomem *remap_pci_mem(ulong base, ulong size)
2917 {
2918         ulong page_base = ((ulong) base) & PAGE_MASK;
2919         ulong page_offs = ((ulong) base) - page_base;
2920         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2921
2922         return page_remapped ? (page_remapped + page_offs) : NULL;
2923 }
2924
2925 /*
2926  * Takes jobs of the Q and sends them to the hardware, then puts it on
2927  * the Q to wait for completion.
2928  */
2929 static void start_io(ctlr_info_t *h)
2930 {
2931         CommandList_struct *c;
2932
2933         while (!hlist_empty(&h->reqQ)) {
2934                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2935                 /* can't do anything if fifo is full */
2936                 if ((h->access.fifo_full(h))) {
2937                         dev_warn(&h->pdev->dev, "fifo full\n");
2938                         break;
2939                 }
2940
2941                 /* Get the first entry from the Request Q */
2942                 removeQ(c);
2943                 h->Qdepth--;
2944
2945                 /* Tell the controller execute command */
2946                 h->access.submit_command(h, c);
2947
2948                 /* Put job onto the completed Q */
2949                 addQ(&h->cmpQ, c);
2950         }
2951 }
2952
2953 /* Assumes that h->lock is held. */
2954 /* Zeros out the error record and then resends the command back */
2955 /* to the controller */
2956 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2957 {
2958         /* erase the old error information */
2959         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2960
2961         /* add it to software queue and then send it to the controller */
2962         addQ(&h->reqQ, c);
2963         h->Qdepth++;
2964         if (h->Qdepth > h->maxQsinceinit)
2965                 h->maxQsinceinit = h->Qdepth;
2966
2967         start_io(h);
2968 }
2969
2970 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2971         unsigned int msg_byte, unsigned int host_byte,
2972         unsigned int driver_byte)
2973 {
2974         /* inverse of macros in scsi.h */
2975         return (scsi_status_byte & 0xff) |
2976                 ((msg_byte & 0xff) << 8) |
2977                 ((host_byte & 0xff) << 16) |
2978                 ((driver_byte & 0xff) << 24);
2979 }
2980
2981 static inline int evaluate_target_status(ctlr_info_t *h,
2982                         CommandList_struct *cmd, int *retry_cmd)
2983 {
2984         unsigned char sense_key;
2985         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2986         int error_value;
2987
2988         *retry_cmd = 0;
2989         /* If we get in here, it means we got "target status", that is, scsi status */
2990         status_byte = cmd->err_info->ScsiStatus;
2991         driver_byte = DRIVER_OK;
2992         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2993
2994         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2995                 host_byte = DID_PASSTHROUGH;
2996         else
2997                 host_byte = DID_OK;
2998
2999         error_value = make_status_bytes(status_byte, msg_byte,
3000                 host_byte, driver_byte);
3001
3002         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3003                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3004                         dev_warn(&h->pdev->dev, "cmd %p "
3005                                "has SCSI Status 0x%x\n",
3006                                cmd, cmd->err_info->ScsiStatus);
3007                 return error_value;
3008         }
3009
3010         /* check the sense key */
3011         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3012         /* no status or recovered error */
3013         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3014             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3015                 error_value = 0;
3016
3017         if (check_for_unit_attention(h, cmd)) {
3018                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3019                 return 0;
3020         }
3021
3022         /* Not SG_IO or similar? */
3023         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3024                 if (error_value != 0)
3025                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3026                                " sense key = 0x%x\n", cmd, sense_key);
3027                 return error_value;
3028         }
3029
3030         /* SG_IO or similar, copy sense data back */
3031         if (cmd->rq->sense) {
3032                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3033                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3034                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3035                         cmd->rq->sense_len);
3036         } else
3037                 cmd->rq->sense_len = 0;
3038
3039         return error_value;
3040 }
3041
3042 /* checks the status of the job and calls complete buffers to mark all
3043  * buffers for the completed job. Note that this function does not need
3044  * to hold the hba/queue lock.
3045  */
3046 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3047                                     int timeout)
3048 {
3049         int retry_cmd = 0;
3050         struct request *rq = cmd->rq;
3051
3052         rq->errors = 0;
3053
3054         if (timeout)
3055                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3056
3057         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3058                 goto after_error_processing;
3059
3060         switch (cmd->err_info->CommandStatus) {
3061         case CMD_TARGET_STATUS:
3062                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3063                 break;
3064         case CMD_DATA_UNDERRUN:
3065                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3066                         dev_warn(&h->pdev->dev, "cmd %p has"
3067                                " completed with data underrun "
3068                                "reported\n", cmd);
3069                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3070                 }
3071                 break;
3072         case CMD_DATA_OVERRUN:
3073                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3074                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3075                                " completed with data overrun "
3076                                "reported\n", cmd);
3077                 break;
3078         case CMD_INVALID:
3079                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3080                        "reported invalid\n", cmd);
3081                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3082                         cmd->err_info->CommandStatus, DRIVER_OK,
3083                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3084                                 DID_PASSTHROUGH : DID_ERROR);
3085                 break;
3086         case CMD_PROTOCOL_ERR:
3087                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3088                        "protocol error\n", cmd);
3089                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3090                         cmd->err_info->CommandStatus, DRIVER_OK,
3091                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3092                                 DID_PASSTHROUGH : DID_ERROR);
3093                 break;
3094         case CMD_HARDWARE_ERR:
3095                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3096                        " hardware error\n", cmd);
3097                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3098                         cmd->err_info->CommandStatus, DRIVER_OK,
3099                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3100                                 DID_PASSTHROUGH : DID_ERROR);
3101                 break;
3102         case CMD_CONNECTION_LOST:
3103                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3104                        "connection lost\n", cmd);
3105                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3106                         cmd->err_info->CommandStatus, DRIVER_OK,
3107                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3108                                 DID_PASSTHROUGH : DID_ERROR);
3109                 break;
3110         case CMD_ABORTED:
3111                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3112                        "aborted\n", cmd);
3113                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3114                         cmd->err_info->CommandStatus, DRIVER_OK,
3115                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3116                                 DID_PASSTHROUGH : DID_ABORT);
3117                 break;
3118         case CMD_ABORT_FAILED:
3119                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3120                        "abort failed\n", cmd);
3121                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3122                         cmd->err_info->CommandStatus, DRIVER_OK,
3123                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3124                                 DID_PASSTHROUGH : DID_ERROR);
3125                 break;
3126         case CMD_UNSOLICITED_ABORT:
3127                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3128                        "abort %p\n", h->ctlr, cmd);
3129                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3130                         retry_cmd = 1;
3131                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3132                         cmd->retry_count++;
3133                 } else
3134                         dev_warn(&h->pdev->dev,
3135                                 "%p retried too many times\n", cmd);
3136                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3137                         cmd->err_info->CommandStatus, DRIVER_OK,
3138                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3139                                 DID_PASSTHROUGH : DID_ABORT);
3140                 break;
3141         case CMD_TIMEOUT:
3142                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3143                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3144                         cmd->err_info->CommandStatus, DRIVER_OK,
3145                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3146                                 DID_PASSTHROUGH : DID_ERROR);
3147                 break;
3148         default:
3149                 dev_warn(&h->pdev->dev, "cmd %p returned "
3150                        "unknown status %x\n", cmd,
3151                        cmd->err_info->CommandStatus);
3152                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3153                         cmd->err_info->CommandStatus, DRIVER_OK,
3154                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3155                                 DID_PASSTHROUGH : DID_ERROR);
3156         }
3157
3158 after_error_processing:
3159
3160         /* We need to return this command */
3161         if (retry_cmd) {
3162                 resend_cciss_cmd(h, cmd);
3163                 return;
3164         }
3165         cmd->rq->completion_data = cmd;
3166         blk_complete_request(cmd->rq);
3167 }
3168
3169 static inline u32 cciss_tag_contains_index(u32 tag)
3170 {
3171 #define DIRECT_LOOKUP_BIT 0x10
3172         return tag & DIRECT_LOOKUP_BIT;
3173 }
3174
3175 static inline u32 cciss_tag_to_index(u32 tag)
3176 {
3177 #define DIRECT_LOOKUP_SHIFT 5
3178         return tag >> DIRECT_LOOKUP_SHIFT;
3179 }
3180
3181 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3182 {
3183 #define CCISS_ERROR_BITS 0x03
3184         return tag & ~CCISS_ERROR_BITS;
3185 }
3186
3187 static inline void cciss_mark_tag_indexed(u32 *tag)
3188 {
3189         *tag |= DIRECT_LOOKUP_BIT;
3190 }
3191
3192 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3193 {
3194         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3195 }
3196
3197 /*
3198  * Get a request and submit it to the controller.
3199  */
3200 static void do_cciss_request(struct request_queue *q)
3201 {
3202         ctlr_info_t *h = q->queuedata;
3203         CommandList_struct *c;
3204         sector_t start_blk;
3205         int seg;
3206         struct request *creq;
3207         u64bit temp64;
3208         struct scatterlist *tmp_sg;
3209         SGDescriptor_struct *curr_sg;
3210         drive_info_struct *drv;
3211         int i, dir;
3212         int sg_index = 0;
3213         int chained = 0;
3214
3215         /* We call start_io here in case there is a command waiting on the
3216          * queue that has not been sent.
3217          */
3218         if (blk_queue_plugged(q))
3219                 goto startio;
3220
3221       queue:
3222         creq = blk_peek_request(q);
3223         if (!creq)
3224                 goto startio;
3225
3226         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3227
3228         c = cmd_alloc(h);
3229         if (!c)
3230                 goto full;
3231
3232         blk_start_request(creq);
3233
3234         tmp_sg = h->scatter_list[c->cmdindex];
3235         spin_unlock_irq(q->queue_lock);
3236
3237         c->cmd_type = CMD_RWREQ;
3238         c->rq = creq;
3239
3240         /* fill in the request */
3241         drv = creq->rq_disk->private_data;
3242         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3243         /* got command from pool, so use the command block index instead */
3244         /* for direct lookups. */
3245         /* The first 2 bits are reserved for controller error reporting. */
3246         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3247         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3248         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3249         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3250         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3251         c->Request.Type.Attribute = ATTR_SIMPLE;
3252         c->Request.Type.Direction =
3253             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3254         c->Request.Timeout = 0; /* Don't time out */
3255         c->Request.CDB[0] =
3256             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3257         start_blk = blk_rq_pos(creq);
3258         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3259                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3260         sg_init_table(tmp_sg, h->maxsgentries);
3261         seg = blk_rq_map_sg(q, creq, tmp_sg);
3262
3263         /* get the DMA records for the setup */
3264         if (c->Request.Type.Direction == XFER_READ)
3265                 dir = PCI_DMA_FROMDEVICE;
3266         else
3267                 dir = PCI_DMA_TODEVICE;
3268
3269         curr_sg = c->SG;
3270         sg_index = 0;
3271         chained = 0;
3272
3273         for (i = 0; i < seg; i++) {
3274                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3275                         !chained && ((seg - i) > 1)) {
3276                         /* Point to next chain block. */
3277                         curr_sg = h->cmd_sg_list[c->cmdindex];
3278                         sg_index = 0;
3279                         chained = 1;
3280                 }
3281                 curr_sg[sg_index].Len = tmp_sg[i].length;
3282                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3283                                                 tmp_sg[i].offset,
3284                                                 tmp_sg[i].length, dir);
3285                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3286                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3287                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3288                 ++sg_index;
3289         }
3290         if (chained)
3291                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3292                         (seg - (h->max_cmd_sgentries - 1)) *
3293                                 sizeof(SGDescriptor_struct));
3294
3295         /* track how many SG entries we are using */
3296         if (seg > h->maxSG)
3297                 h->maxSG = seg;
3298
3299         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3300                         "chained[%d]\n",
3301                         blk_rq_sectors(creq), seg, chained);
3302
3303         c->Header.SGTotal = seg + chained;
3304         if (seg <= h->max_cmd_sgentries)
3305                 c->Header.SGList = c->Header.SGTotal;
3306         else
3307                 c->Header.SGList = h->max_cmd_sgentries;
3308         set_performant_mode(h, c);
3309
3310         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3311                 if(h->cciss_read == CCISS_READ_10) {
3312                         c->Request.CDB[1] = 0;
3313                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3314                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3315                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3316                         c->Request.CDB[5] = start_blk & 0xff;
3317                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3318                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3319                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3320                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3321                 } else {
3322                         u32 upper32 = upper_32_bits(start_blk);
3323
3324                         c->Request.CDBLen = 16;
3325                         c->Request.CDB[1]= 0;
3326                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3327                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3328                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3329                         c->Request.CDB[5]= upper32 & 0xff;
3330                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3331                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3332                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3333                         c->Request.CDB[9]= start_blk & 0xff;
3334                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3335                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3336                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3337                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3338                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3339                 }
3340         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3341                 c->Request.CDBLen = creq->cmd_len;
3342                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3343         } else {
3344                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3345                         creq->cmd_type);
3346                 BUG();
3347         }
3348
3349         spin_lock_irq(q->queue_lock);
3350
3351         addQ(&h->reqQ, c);
3352         h->Qdepth++;
3353         if (h->Qdepth > h->maxQsinceinit)
3354                 h->maxQsinceinit = h->Qdepth;
3355
3356         goto queue;
3357 full:
3358         blk_stop_queue(q);
3359 startio:
3360         /* We will already have the driver lock here so not need
3361          * to lock it.
3362          */
3363         start_io(h);
3364 }
3365
3366 static inline unsigned long get_next_completion(ctlr_info_t *h)
3367 {
3368         return h->access.command_completed(h);
3369 }
3370
3371 static inline int interrupt_pending(ctlr_info_t *h)
3372 {
3373         return h->access.intr_pending(h);
3374 }
3375
3376 static inline long interrupt_not_for_us(ctlr_info_t *h)
3377 {
3378         return ((h->access.intr_pending(h) == 0) ||
3379                 (h->interrupts_enabled == 0));
3380 }
3381
3382 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3383                         u32 raw_tag)
3384 {
3385         if (unlikely(tag_index >= h->nr_cmds)) {
3386                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3387                 return 1;
3388         }
3389         return 0;
3390 }
3391
3392 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3393                                 u32 raw_tag)
3394 {
3395         removeQ(c);
3396         if (likely(c->cmd_type == CMD_RWREQ))
3397                 complete_command(h, c, 0);
3398         else if (c->cmd_type == CMD_IOCTL_PEND)
3399                 complete(c->waiting);
3400 #ifdef CONFIG_CISS_SCSI_TAPE
3401         else if (c->cmd_type == CMD_SCSI)
3402                 complete_scsi_command(c, 0, raw_tag);
3403 #endif
3404 }
3405
3406 static inline u32 next_command(ctlr_info_t *h)
3407 {
3408         u32 a;
3409
3410         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3411                 return h->access.command_completed(h);
3412
3413         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3414                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3415                 (h->reply_pool_head)++;
3416                 h->commands_outstanding--;
3417         } else {
3418                 a = FIFO_EMPTY;
3419         }
3420         /* Check for wraparound */
3421         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3422                 h->reply_pool_head = h->reply_pool;
3423                 h->reply_pool_wraparound ^= 1;
3424         }
3425         return a;
3426 }
3427
3428 /* process completion of an indexed ("direct lookup") command */
3429 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3430 {
3431         u32 tag_index;
3432         CommandList_struct *c;
3433
3434         tag_index = cciss_tag_to_index(raw_tag);
3435         if (bad_tag(h, tag_index, raw_tag))
3436                 return next_command(h);
3437         c = h->cmd_pool + tag_index;
3438         finish_cmd(h, c, raw_tag);
3439         return next_command(h);
3440 }
3441
3442 /* process completion of a non-indexed command */
3443 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3444 {
3445         u32 tag;
3446         CommandList_struct *c = NULL;
3447         struct hlist_node *tmp;
3448         __u32 busaddr_masked, tag_masked;
3449
3450         tag = cciss_tag_discard_error_bits(raw_tag);
3451         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3452                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3453                 tag_masked = cciss_tag_discard_error_bits(tag);
3454                 if (busaddr_masked == tag_masked) {
3455                         finish_cmd(h, c, raw_tag);
3456                         return next_command(h);
3457                 }
3458         }
3459         bad_tag(h, h->nr_cmds + 1, raw_tag);
3460         return next_command(h);
3461 }
3462
3463 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3464 {
3465         ctlr_info_t *h = dev_id;
3466         unsigned long flags;
3467         u32 raw_tag;
3468
3469         if (interrupt_not_for_us(h))
3470                 return IRQ_NONE;
3471         spin_lock_irqsave(&h->lock, flags);
3472         while (interrupt_pending(h)) {
3473                 raw_tag = get_next_completion(h);
3474                 while (raw_tag != FIFO_EMPTY) {
3475                         if (cciss_tag_contains_index(raw_tag))
3476                                 raw_tag = process_indexed_cmd(h, raw_tag);
3477                         else
3478                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3479                 }
3480         }
3481         spin_unlock_irqrestore(&h->lock, flags);
3482         return IRQ_HANDLED;
3483 }
3484
3485 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3486  * check the interrupt pending register because it is not set.
3487  */
3488 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3489 {
3490         ctlr_info_t *h = dev_id;
3491         unsigned long flags;
3492         u32 raw_tag;
3493
3494         spin_lock_irqsave(&h->lock, flags);
3495         raw_tag = get_next_completion(h);
3496         while (raw_tag != FIFO_EMPTY) {
3497                 if (cciss_tag_contains_index(raw_tag))
3498                         raw_tag = process_indexed_cmd(h, raw_tag);
3499                 else
3500                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3501         }
3502         spin_unlock_irqrestore(&h->lock, flags);
3503         return IRQ_HANDLED;
3504 }
3505
3506 /**
3507  * add_to_scan_list() - add controller to rescan queue
3508  * @h:                Pointer to the controller.
3509  *
3510  * Adds the controller to the rescan queue if not already on the queue.
3511  *
3512  * returns 1 if added to the queue, 0 if skipped (could be on the
3513  * queue already, or the controller could be initializing or shutting
3514  * down).
3515  **/
3516 static int add_to_scan_list(struct ctlr_info *h)
3517 {
3518         struct ctlr_info *test_h;
3519         int found = 0;
3520         int ret = 0;
3521
3522         if (h->busy_initializing)
3523                 return 0;
3524
3525         if (!mutex_trylock(&h->busy_shutting_down))
3526                 return 0;
3527
3528         mutex_lock(&scan_mutex);
3529         list_for_each_entry(test_h, &scan_q, scan_list) {
3530                 if (test_h == h) {
3531                         found = 1;
3532                         break;
3533                 }
3534         }
3535         if (!found && !h->busy_scanning) {
3536                 INIT_COMPLETION(h->scan_wait);
3537                 list_add_tail(&h->scan_list, &scan_q);
3538                 ret = 1;
3539         }
3540         mutex_unlock(&scan_mutex);
3541         mutex_unlock(&h->busy_shutting_down);
3542
3543         return ret;
3544 }
3545
3546 /**
3547  * remove_from_scan_list() - remove controller from rescan queue
3548  * @h:                     Pointer to the controller.
3549  *
3550  * Removes the controller from the rescan queue if present. Blocks if
3551  * the controller is currently conducting a rescan.  The controller
3552  * can be in one of three states:
3553  * 1. Doesn't need a scan
3554  * 2. On the scan list, but not scanning yet (we remove it)
3555  * 3. Busy scanning (and not on the list). In this case we want to wait for
3556  *    the scan to complete to make sure the scanning thread for this
3557  *    controller is completely idle.
3558  **/
3559 static void remove_from_scan_list(struct ctlr_info *h)
3560 {
3561         struct ctlr_info *test_h, *tmp_h;
3562
3563         mutex_lock(&scan_mutex);
3564         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3565                 if (test_h == h) { /* state 2. */
3566                         list_del(&h->scan_list);
3567                         complete_all(&h->scan_wait);
3568                         mutex_unlock(&scan_mutex);
3569                         return;
3570                 }
3571         }
3572         if (h->busy_scanning) { /* state 3. */
3573                 mutex_unlock(&scan_mutex);
3574                 wait_for_completion(&h->scan_wait);
3575         } else { /* state 1, nothing to do. */
3576                 mutex_unlock(&scan_mutex);
3577         }
3578 }
3579
3580 /**
3581  * scan_thread() - kernel thread used to rescan controllers
3582  * @data:        Ignored.
3583  *
3584  * A kernel thread used scan for drive topology changes on
3585  * controllers. The thread processes only one controller at a time
3586  * using a queue.  Controllers are added to the queue using
3587  * add_to_scan_list() and removed from the queue either after done
3588  * processing or using remove_from_scan_list().
3589  *
3590  * returns 0.
3591  **/
3592 static int scan_thread(void *data)
3593 {
3594         struct ctlr_info *h;
3595
3596         while (1) {
3597                 set_current_state(TASK_INTERRUPTIBLE);
3598                 schedule();
3599                 if (kthread_should_stop())
3600                         break;
3601
3602                 while (1) {
3603                         mutex_lock(&scan_mutex);
3604                         if (list_empty(&scan_q)) {
3605                                 mutex_unlock(&scan_mutex);
3606                                 break;
3607                         }
3608
3609                         h = list_entry(scan_q.next,
3610                                        struct ctlr_info,
3611                                        scan_list);
3612                         list_del(&h->scan_list);
3613                         h->busy_scanning = 1;
3614                         mutex_unlock(&scan_mutex);
3615
3616                         rebuild_lun_table(h, 0, 0);
3617                         complete_all(&h->scan_wait);
3618                         mutex_lock(&scan_mutex);
3619                         h->busy_scanning = 0;
3620                         mutex_unlock(&scan_mutex);
3621                 }
3622         }
3623
3624         return 0;
3625 }
3626
3627 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3628 {
3629         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3630                 return 0;
3631
3632         switch (c->err_info->SenseInfo[12]) {
3633         case STATE_CHANGED:
3634                 dev_warn(&h->pdev->dev, "a state change "
3635                         "detected, command retried\n");
3636                 return 1;
3637         break;
3638         case LUN_FAILED:
3639                 dev_warn(&h->pdev->dev, "LUN failure "
3640                         "detected, action required\n");
3641                 return 1;
3642         break;
3643         case REPORT_LUNS_CHANGED:
3644                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3645         /*
3646          * Here, we could call add_to_scan_list and wake up the scan thread,
3647          * except that it's quite likely that we will get more than one
3648          * REPORT_LUNS_CHANGED condition in quick succession, which means
3649          * that those which occur after the first one will likely happen
3650          * *during* the scan_thread's rescan.  And the rescan code is not
3651          * robust enough to restart in the middle, undoing what it has already
3652          * done, and it's not clear that it's even possible to do this, since
3653          * part of what it does is notify the block layer, which starts
3654          * doing it's own i/o to read partition tables and so on, and the
3655          * driver doesn't have visibility to know what might need undoing.
3656          * In any event, if possible, it is horribly complicated to get right
3657          * so we just don't do it for now.
3658          *
3659          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3660          */
3661                 return 1;
3662         break;
3663         case POWER_OR_RESET:
3664                 dev_warn(&h->pdev->dev,
3665                         "a power on or device reset detected\n");
3666                 return 1;
3667         break;
3668         case UNIT_ATTENTION_CLEARED:
3669                 dev_warn(&h->pdev->dev,
3670                         "unit attention cleared by another initiator\n");
3671                 return 1;
3672         break;
3673         default:
3674                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3675                 return 1;
3676         }
3677 }
3678
3679 /*
3680  *  We cannot read the structure directly, for portability we must use
3681  *   the io functions.
3682  *   This is for debug only.
3683  */
3684 static void print_cfg_table(ctlr_info_t *h)
3685 {
3686         int i;
3687         char temp_name[17];
3688         CfgTable_struct *tb = h->cfgtable;
3689
3690         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3691         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3692         for (i = 0; i < 4; i++)
3693                 temp_name[i] = readb(&(tb->Signature[i]));
3694         temp_name[4] = '\0';
3695         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3696         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3697                 readl(&(tb->SpecValence)));
3698         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3699                readl(&(tb->TransportSupport)));
3700         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3701                readl(&(tb->TransportActive)));
3702         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3703                readl(&(tb->HostWrite.TransportRequest)));
3704         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3705                readl(&(tb->HostWrite.CoalIntDelay)));
3706         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3707                readl(&(tb->HostWrite.CoalIntCount)));
3708         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3709                readl(&(tb->CmdsOutMax)));
3710         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3711                 readl(&(tb->BusTypes)));
3712         for (i = 0; i < 16; i++)
3713                 temp_name[i] = readb(&(tb->ServerName[i]));
3714         temp_name[16] = '\0';
3715         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3716         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3717                 readl(&(tb->HeartBeat)));
3718 }
3719
3720 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3721 {
3722         int i, offset, mem_type, bar_type;
3723         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3724                 return 0;
3725         offset = 0;
3726         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3727                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3728                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3729                         offset += 4;
3730                 else {
3731                         mem_type = pci_resource_flags(pdev, i) &
3732                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3733                         switch (mem_type) {
3734                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3735                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3736                                 offset += 4;    /* 32 bit */
3737                                 break;
3738                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3739                                 offset += 8;
3740                                 break;
3741                         default:        /* reserved in PCI 2.2 */
3742                                 dev_warn(&pdev->dev,
3743                                        "Base address is invalid\n");
3744                                 return -1;
3745                                 break;
3746                         }
3747                 }
3748                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3749                         return i + 1;
3750         }
3751         return -1;
3752 }
3753
3754 /* Fill in bucket_map[], given nsgs (the max number of
3755  * scatter gather elements supported) and bucket[],
3756  * which is an array of 8 integers.  The bucket[] array
3757  * contains 8 different DMA transfer sizes (in 16
3758  * byte increments) which the controller uses to fetch
3759  * commands.  This function fills in bucket_map[], which
3760  * maps a given number of scatter gather elements to one of
3761  * the 8 DMA transfer sizes.  The point of it is to allow the
3762  * controller to only do as much DMA as needed to fetch the
3763  * command, with the DMA transfer size encoded in the lower
3764  * bits of the command address.
3765  */
3766 static void  calc_bucket_map(int bucket[], int num_buckets,
3767         int nsgs, int *bucket_map)
3768 {
3769         int i, j, b, size;
3770
3771         /* even a command with 0 SGs requires 4 blocks */
3772 #define MINIMUM_TRANSFER_BLOCKS 4
3773 #define NUM_BUCKETS 8
3774         /* Note, bucket_map must have nsgs+1 entries. */
3775         for (i = 0; i <= nsgs; i++) {
3776                 /* Compute size of a command with i SG entries */
3777                 size = i + MINIMUM_TRANSFER_BLOCKS;
3778                 b = num_buckets; /* Assume the biggest bucket */
3779                 /* Find the bucket that is just big enough */
3780                 for (j = 0; j < 8; j++) {
3781                         if (bucket[j] >= size) {
3782                                 b = j;
3783                                 break;
3784                         }
3785                 }
3786                 /* for a command with i SG entries, use bucket b. */
3787                 bucket_map[i] = b;
3788         }
3789 }
3790
3791 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3792 {
3793         int i;
3794
3795         /* under certain very rare conditions, this can take awhile.
3796          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3797          * as we enter this code.) */
3798         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3799                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3800                         break;
3801                 msleep(10);
3802         }
3803 }
3804
3805 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3806 {
3807         /* This is a bit complicated.  There are 8 registers on
3808          * the controller which we write to to tell it 8 different
3809          * sizes of commands which there may be.  It's a way of
3810          * reducing the DMA done to fetch each command.  Encoded into
3811          * each command's tag are 3 bits which communicate to the controller
3812          * which of the eight sizes that command fits within.  The size of
3813          * each command depends on how many scatter gather entries there are.
3814          * Each SG entry requires 16 bytes.  The eight registers are programmed
3815          * with the number of 16-byte blocks a command of that size requires.
3816          * The smallest command possible requires 5 such 16 byte blocks.
3817          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3818          * blocks.  Note, this only extends to the SG entries contained
3819          * within the command block, and does not extend to chained blocks
3820          * of SG elements.   bft[] contains the eight values we write to
3821          * the registers.  They are not evenly distributed, but have more
3822          * sizes for small commands, and fewer sizes for larger commands.
3823          */
3824         __u32 trans_offset;
3825         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3826                         /*
3827                          *  5 = 1 s/g entry or 4k
3828                          *  6 = 2 s/g entry or 8k
3829                          *  8 = 4 s/g entry or 16k
3830                          * 10 = 6 s/g entry or 24k
3831                          */
3832         unsigned long register_value;
3833         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3834
3835         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3836
3837         /* Controller spec: zero out this buffer. */
3838         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3839         h->reply_pool_head = h->reply_pool;
3840
3841         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3842         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3843                                 h->blockFetchTable);
3844         writel(bft[0], &h->transtable->BlockFetch0);
3845         writel(bft[1], &h->transtable->BlockFetch1);
3846         writel(bft[2], &h->transtable->BlockFetch2);
3847         writel(bft[3], &h->transtable->BlockFetch3);
3848         writel(bft[4], &h->transtable->BlockFetch4);
3849         writel(bft[5], &h->transtable->BlockFetch5);
3850         writel(bft[6], &h->transtable->BlockFetch6);
3851         writel(bft[7], &h->transtable->BlockFetch7);
3852
3853         /* size of controller ring buffer */
3854         writel(h->max_commands, &h->transtable->RepQSize);
3855         writel(1, &h->transtable->RepQCount);
3856         writel(0, &h->transtable->RepQCtrAddrLow32);
3857         writel(0, &h->transtable->RepQCtrAddrHigh32);
3858         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3859         writel(0, &h->transtable->RepQAddr0High32);
3860         writel(CFGTBL_Trans_Performant,
3861                         &(h->cfgtable->HostWrite.TransportRequest));
3862
3863         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3864         cciss_wait_for_mode_change_ack(h);
3865         register_value = readl(&(h->cfgtable->TransportActive));
3866         if (!(register_value & CFGTBL_Trans_Performant))
3867                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3868                                         " performant mode\n");
3869 }
3870
3871 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3872 {
3873         __u32 trans_support;
3874
3875         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3876         /* Attempt to put controller into performant mode if supported */
3877         /* Does board support performant mode? */
3878         trans_support = readl(&(h->cfgtable->TransportSupport));
3879         if (!(trans_support & PERFORMANT_MODE))
3880                 return;
3881
3882         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3883         /* Performant mode demands commands on a 32 byte boundary
3884          * pci_alloc_consistent aligns on page boundarys already.
3885          * Just need to check if divisible by 32
3886          */
3887         if ((sizeof(CommandList_struct) % 32) != 0) {
3888                 dev_warn(&h->pdev->dev, "%s %d %s\n",
3889                         "cciss info: command size[",
3890                         (int)sizeof(CommandList_struct),
3891                         "] not divisible by 32, no performant mode..\n");
3892                 return;
3893         }
3894
3895         /* Performant mode ring buffer and supporting data structures */
3896         h->reply_pool = (__u64 *)pci_alloc_consistent(
3897                 h->pdev, h->max_commands * sizeof(__u64),
3898                 &(h->reply_pool_dhandle));
3899
3900         /* Need a block fetch table for performant mode */
3901         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3902                 sizeof(__u32)), GFP_KERNEL);
3903
3904         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3905                 goto clean_up;
3906
3907         cciss_enter_performant_mode(h);
3908
3909         /* Change the access methods to the performant access methods */
3910         h->access = SA5_performant_access;
3911         h->transMethod = CFGTBL_Trans_Performant;
3912
3913         return;
3914 clean_up:
3915         kfree(h->blockFetchTable);
3916         if (h->reply_pool)
3917                 pci_free_consistent(h->pdev,
3918                                 h->max_commands * sizeof(__u64),
3919                                 h->reply_pool,
3920                                 h->reply_pool_dhandle);
3921         return;
3922
3923 } /* cciss_put_controller_into_performant_mode */
3924
3925 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3926  * controllers that are capable. If not, we use IO-APIC mode.
3927  */
3928
3929 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3930 {
3931 #ifdef CONFIG_PCI_MSI
3932         int err;
3933         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3934         {0, 2}, {0, 3}
3935         };
3936
3937         /* Some boards advertise MSI but don't really support it */
3938         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3939             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3940                 goto default_int_mode;
3941
3942         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3943                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3944                 if (!err) {
3945                         h->intr[0] = cciss_msix_entries[0].vector;
3946                         h->intr[1] = cciss_msix_entries[1].vector;
3947                         h->intr[2] = cciss_msix_entries[2].vector;
3948                         h->intr[3] = cciss_msix_entries[3].vector;
3949                         h->msix_vector = 1;
3950                         return;
3951                 }
3952                 if (err > 0) {
3953                         dev_warn(&h->pdev->dev,
3954                                 "only %d MSI-X vectors available\n", err);
3955                         goto default_int_mode;
3956                 } else {
3957                         dev_warn(&h->pdev->dev,
3958                                 "MSI-X init failed %d\n", err);
3959                         goto default_int_mode;
3960                 }
3961         }
3962         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3963                 if (!pci_enable_msi(h->pdev))
3964                         h->msi_vector = 1;
3965                 else
3966                         dev_warn(&h->pdev->dev, "MSI init failed\n");
3967         }
3968 default_int_mode:
3969 #endif                          /* CONFIG_PCI_MSI */
3970         /* if we get here we're going to use the default interrupt mode */
3971         h->intr[PERF_MODE_INT] = h->pdev->irq;
3972         return;
3973 }
3974
3975 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3976 {
3977         int i;
3978         u32 subsystem_vendor_id, subsystem_device_id;
3979
3980         subsystem_vendor_id = pdev->subsystem_vendor;
3981         subsystem_device_id = pdev->subsystem_device;
3982         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3983                         subsystem_vendor_id;
3984
3985         for (i = 0; i < ARRAY_SIZE(products); i++) {
3986                 /* Stand aside for hpsa driver on request */
3987                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3988                         return -ENODEV;
3989                 if (*board_id == products[i].board_id)
3990                         return i;
3991         }
3992         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
3993                 *board_id);
3994         return -ENODEV;
3995 }
3996
3997 static inline bool cciss_board_disabled(ctlr_info_t *h)
3998 {
3999         u16 command;
4000
4001         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4002         return ((command & PCI_COMMAND_MEMORY) == 0);
4003 }
4004
4005 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4006         unsigned long *memory_bar)
4007 {
4008         int i;
4009
4010         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4011                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4012                         /* addressing mode bits already removed */
4013                         *memory_bar = pci_resource_start(pdev, i);
4014                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4015                                 *memory_bar);
4016                         return 0;
4017                 }
4018         dev_warn(&pdev->dev, "no memory BAR found\n");
4019         return -ENODEV;
4020 }
4021
4022 static int __devinit cciss_wait_for_board_ready(ctlr_info_t *h)
4023 {
4024         int i;
4025         u32 scratchpad;
4026
4027         for (i = 0; i < CCISS_BOARD_READY_ITERATIONS; i++) {
4028                 scratchpad = readl(h->vaddr + SA5_SCRATCHPAD_OFFSET);
4029                 if (scratchpad == CCISS_FIRMWARE_READY)
4030                         return 0;
4031                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4032         }
4033         dev_warn(&h->pdev->dev, "board not ready, timed out.\n");
4034         return -ENODEV;
4035 }
4036
4037 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4038         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4039         u64 *cfg_offset)
4040 {
4041         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4042         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4043         *cfg_base_addr &= (u32) 0x0000ffff;
4044         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4045         if (*cfg_base_addr_index == -1) {
4046                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4047                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4048                 return -ENODEV;
4049         }
4050         return 0;
4051 }
4052
4053 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4054 {
4055         u64 cfg_offset;
4056         u32 cfg_base_addr;
4057         u64 cfg_base_addr_index;
4058         u32 trans_offset;
4059         int rc;
4060
4061         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4062                 &cfg_base_addr_index, &cfg_offset);
4063         if (rc)
4064                 return rc;
4065         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4066                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4067         if (!h->cfgtable)
4068                 return -ENOMEM;
4069         /* Find performant mode table. */
4070         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4071         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4072                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4073                                 sizeof(*h->transtable));
4074         if (!h->transtable)
4075                 return -ENOMEM;
4076         return 0;
4077 }
4078
4079 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4080 {
4081         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4082         if (h->max_commands < 16) {
4083                 dev_warn(&h->pdev->dev, "Controller reports "
4084                         "max supported commands of %d, an obvious lie. "
4085                         "Using 16.  Ensure that firmware is up to date.\n",
4086                         h->max_commands);
4087                 h->max_commands = 16;
4088         }
4089 }
4090
4091 /* Interrogate the hardware for some limits:
4092  * max commands, max SG elements without chaining, and with chaining,
4093  * SG chain block size, etc.
4094  */
4095 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4096 {
4097         cciss_get_max_perf_mode_cmds(h);
4098         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4099         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4100         /*
4101          * Limit in-command s/g elements to 32 save dma'able memory.
4102          * Howvever spec says if 0, use 31
4103          */
4104         h->max_cmd_sgentries = 31;
4105         if (h->maxsgentries > 512) {
4106                 h->max_cmd_sgentries = 32;
4107                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4108                 h->maxsgentries--; /* save one for chain pointer */
4109         } else {
4110                 h->maxsgentries = 31; /* default to traditional values */
4111                 h->chainsize = 0;
4112         }
4113 }
4114
4115 static inline bool CISS_signature_present(ctlr_info_t *h)
4116 {
4117         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4118             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4119             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4120             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4121                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4122                 return false;
4123         }
4124         return true;
4125 }
4126
4127 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4128 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4129 {
4130 #ifdef CONFIG_X86
4131         u32 prefetch;
4132
4133         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4134         prefetch |= 0x100;
4135         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4136 #endif
4137 }
4138
4139 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4140  * in a prefetch beyond physical memory.
4141  */
4142 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4143 {
4144         u32 dma_prefetch;
4145         __u32 dma_refetch;
4146
4147         if (h->board_id != 0x3225103C)
4148                 return;
4149         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4150         dma_prefetch |= 0x8000;
4151         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4152         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4153         dma_refetch |= 0x1;
4154         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4155 }
4156
4157 static int __devinit cciss_pci_init(ctlr_info_t *h)
4158 {
4159         int prod_index, err;
4160
4161         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4162         if (prod_index < 0)
4163                 return -ENODEV;
4164         h->product_name = products[prod_index].product_name;
4165         h->access = *(products[prod_index].access);
4166
4167         if (cciss_board_disabled(h)) {
4168                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4169                 return -ENODEV;
4170         }
4171         err = pci_enable_device(h->pdev);
4172         if (err) {
4173                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4174                 return err;
4175         }
4176
4177         err = pci_request_regions(h->pdev, "cciss");
4178         if (err) {
4179                 dev_warn(&h->pdev->dev,
4180                         "Cannot obtain PCI resources, aborting\n");
4181                 return err;
4182         }
4183
4184         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4185         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4186
4187 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4188  * else we use the IO-APIC interrupt assigned to us by system ROM.
4189  */
4190         cciss_interrupt_mode(h);
4191         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4192         if (err)
4193                 goto err_out_free_res;
4194         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4195         if (!h->vaddr) {
4196                 err = -ENOMEM;
4197                 goto err_out_free_res;
4198         }
4199         err = cciss_wait_for_board_ready(h);
4200         if (err)
4201                 goto err_out_free_res;
4202         err = cciss_find_cfgtables(h);
4203         if (err)
4204                 goto err_out_free_res;
4205         print_cfg_table(h);
4206         cciss_find_board_params(h);
4207
4208         if (!CISS_signature_present(h)) {
4209                 err = -ENODEV;
4210                 goto err_out_free_res;
4211         }
4212         cciss_enable_scsi_prefetch(h);
4213         cciss_p600_dma_prefetch_quirk(h);
4214         cciss_put_controller_into_performant_mode(h);
4215         return 0;
4216
4217 err_out_free_res:
4218         /*
4219          * Deliberately omit pci_disable_device(): it does something nasty to
4220          * Smart Array controllers that pci_enable_device does not undo
4221          */
4222         if (h->transtable)
4223                 iounmap(h->transtable);
4224         if (h->cfgtable)
4225                 iounmap(h->cfgtable);
4226         if (h->vaddr)
4227                 iounmap(h->vaddr);
4228         pci_release_regions(h->pdev);
4229         return err;
4230 }
4231
4232 /* Function to find the first free pointer into our hba[] array
4233  * Returns -1 if no free entries are left.
4234  */
4235 static int alloc_cciss_hba(struct pci_dev *pdev)
4236 {
4237         int i;
4238
4239         for (i = 0; i < MAX_CTLR; i++) {
4240                 if (!hba[i]) {
4241                         ctlr_info_t *h;
4242
4243                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4244                         if (!h)
4245                                 goto Enomem;
4246                         hba[i] = h;
4247                         return i;
4248                 }
4249         }
4250         dev_warn(&pdev->dev, "This driver supports a maximum"
4251                " of %d controllers.\n", MAX_CTLR);
4252         return -1;
4253 Enomem:
4254         dev_warn(&pdev->dev, "out of memory.\n");
4255         return -1;
4256 }
4257
4258 static void free_hba(ctlr_info_t *h)
4259 {
4260         int i;
4261
4262         hba[h->ctlr] = NULL;
4263         for (i = 0; i < h->highest_lun + 1; i++)
4264                 if (h->gendisk[i] != NULL)
4265                         put_disk(h->gendisk[i]);
4266         kfree(h);
4267 }
4268
4269 /* Send a message CDB to the firmware. */
4270 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4271 {
4272         typedef struct {
4273                 CommandListHeader_struct CommandHeader;
4274                 RequestBlock_struct Request;
4275                 ErrDescriptor_struct ErrorDescriptor;
4276         } Command;
4277         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4278         Command *cmd;
4279         dma_addr_t paddr64;
4280         uint32_t paddr32, tag;
4281         void __iomem *vaddr;
4282         int i, err;
4283
4284         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4285         if (vaddr == NULL)
4286                 return -ENOMEM;
4287
4288         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4289            CCISS commands, so they must be allocated from the lower 4GiB of
4290            memory. */
4291         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4292         if (err) {
4293                 iounmap(vaddr);
4294                 return -ENOMEM;
4295         }
4296
4297         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4298         if (cmd == NULL) {
4299                 iounmap(vaddr);
4300                 return -ENOMEM;
4301         }
4302
4303         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4304            although there's no guarantee, we assume that the address is at
4305            least 4-byte aligned (most likely, it's page-aligned). */
4306         paddr32 = paddr64;
4307
4308         cmd->CommandHeader.ReplyQueue = 0;
4309         cmd->CommandHeader.SGList = 0;
4310         cmd->CommandHeader.SGTotal = 0;
4311         cmd->CommandHeader.Tag.lower = paddr32;
4312         cmd->CommandHeader.Tag.upper = 0;
4313         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4314
4315         cmd->Request.CDBLen = 16;
4316         cmd->Request.Type.Type = TYPE_MSG;
4317         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4318         cmd->Request.Type.Direction = XFER_NONE;
4319         cmd->Request.Timeout = 0; /* Don't time out */
4320         cmd->Request.CDB[0] = opcode;
4321         cmd->Request.CDB[1] = type;
4322         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4323
4324         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4325         cmd->ErrorDescriptor.Addr.upper = 0;
4326         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4327
4328         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4329
4330         for (i = 0; i < 10; i++) {
4331                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4332                 if ((tag & ~3) == paddr32)
4333                         break;
4334                 schedule_timeout_uninterruptible(HZ);
4335         }
4336
4337         iounmap(vaddr);
4338
4339         /* we leak the DMA buffer here ... no choice since the controller could
4340            still complete the command. */
4341         if (i == 10) {
4342                 dev_err(&pdev->dev,
4343                         "controller message %02x:%02x timed out\n",
4344                         opcode, type);
4345                 return -ETIMEDOUT;
4346         }
4347
4348         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4349
4350         if (tag & 2) {
4351                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4352                         opcode, type);
4353                 return -EIO;
4354         }
4355
4356         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4357                 opcode, type);
4358         return 0;
4359 }
4360
4361 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4362 #define cciss_noop(p) cciss_message(p, 3, 0)
4363
4364 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4365 {
4366 /* the #defines are stolen from drivers/pci/msi.h. */
4367 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4368 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4369
4370         int pos;
4371         u16 control = 0;
4372
4373         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4374         if (pos) {
4375                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4376                 if (control & PCI_MSI_FLAGS_ENABLE) {
4377                         dev_info(&pdev->dev, "resetting MSI\n");
4378                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4379                 }
4380         }
4381
4382         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4383         if (pos) {
4384                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4385                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4386                         dev_info(&pdev->dev, "resetting MSI-X\n");
4387                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4388                 }
4389         }
4390
4391         return 0;
4392 }
4393
4394 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4395         void * __iomem vaddr, bool use_doorbell)
4396 {
4397         u16 pmcsr;
4398         int pos;
4399
4400         if (use_doorbell) {
4401                 /* For everything after the P600, the PCI power state method
4402                  * of resetting the controller doesn't work, so we have this
4403                  * other way using the doorbell register.
4404                  */
4405                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4406                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4407                 msleep(1000);
4408         } else { /* Try to do it the PCI power state way */
4409
4410                 /* Quoting from the Open CISS Specification: "The Power
4411                  * Management Control/Status Register (CSR) controls the power
4412                  * state of the device.  The normal operating state is D0,
4413                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4414                  * the controller, place the interface device in D3 then to D0,
4415                  * this causes a secondary PCI reset which will reset the
4416                  * controller." */
4417
4418                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4419                 if (pos == 0) {
4420                         dev_err(&pdev->dev,
4421                                 "cciss_controller_hard_reset: "
4422                                 "PCI PM not supported\n");
4423                         return -ENODEV;
4424                 }
4425                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4426                 /* enter the D3hot power management state */
4427                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4428                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4429                 pmcsr |= PCI_D3hot;
4430                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4431
4432                 msleep(500);
4433
4434                 /* enter the D0 power management state */
4435                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4436                 pmcsr |= PCI_D0;
4437                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4438
4439                 msleep(500);
4440         }
4441         return 0;
4442 }
4443
4444 /* This does a hard reset of the controller using PCI power management
4445  * states or using the doorbell register. */
4446 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4447 {
4448         u16 saved_config_space[32];
4449         u64 cfg_offset;
4450         u32 cfg_base_addr;
4451         u64 cfg_base_addr_index;
4452         void __iomem *vaddr;
4453         unsigned long paddr;
4454         u32 misc_fw_support, active_transport;
4455         int rc, i;
4456         CfgTable_struct __iomem *cfgtable;
4457         bool use_doorbell;
4458         u32 board_id;
4459
4460         /* For controllers as old a the p600, this is very nearly
4461          * the same thing as
4462          *
4463          * pci_save_state(pci_dev);
4464          * pci_set_power_state(pci_dev, PCI_D3hot);
4465          * pci_set_power_state(pci_dev, PCI_D0);
4466          * pci_restore_state(pci_dev);
4467          *
4468          * but we can't use these nice canned kernel routines on
4469          * kexec, because they also check the MSI/MSI-X state in PCI
4470          * configuration space and do the wrong thing when it is
4471          * set/cleared.  Also, the pci_save/restore_state functions
4472          * violate the ordering requirements for restoring the
4473          * configuration space from the CCISS document (see the
4474          * comment below).  So we roll our own ....
4475          *
4476          * For controllers newer than the P600, the pci power state
4477          * method of resetting doesn't work so we have another way
4478          * using the doorbell register.
4479          */
4480
4481         /* Exclude 640x boards.  These are two pci devices in one slot
4482          * which share a battery backed cache module.  One controls the
4483          * cache, the other accesses the cache through the one that controls
4484          * it.  If we reset the one controlling the cache, the other will
4485          * likely not be happy.  Just forbid resetting this conjoined mess.
4486          */
4487         cciss_lookup_board_id(pdev, &board_id);
4488         if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4489                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4490                                 "due to shared cache module.");
4491                 return -ENODEV;
4492         }
4493
4494         for (i = 0; i < 32; i++)
4495                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4496
4497         /* find the first memory BAR, so we can find the cfg table */
4498         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4499         if (rc)
4500                 return rc;
4501         vaddr = remap_pci_mem(paddr, 0x250);
4502         if (!vaddr)
4503                 return -ENOMEM;
4504
4505         /* find cfgtable in order to check if reset via doorbell is supported */
4506         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4507                                         &cfg_base_addr_index, &cfg_offset);
4508         if (rc)
4509                 goto unmap_vaddr;
4510         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4511                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4512         if (!cfgtable) {
4513                 rc = -ENOMEM;
4514                 goto unmap_vaddr;
4515         }
4516
4517         /* If reset via doorbell register is supported, use that. */
4518         misc_fw_support = readl(&cfgtable->misc_fw_support);
4519         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4520
4521         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4522         if (rc)
4523                 goto unmap_cfgtable;
4524
4525         /* Restore the PCI configuration space.  The Open CISS
4526          * Specification says, "Restore the PCI Configuration
4527          * Registers, offsets 00h through 60h. It is important to
4528          * restore the command register, 16-bits at offset 04h,
4529          * last. Do not restore the configuration status register,
4530          * 16-bits at offset 06h."  Note that the offset is 2*i.
4531          */
4532         for (i = 0; i < 32; i++) {
4533                 if (i == 2 || i == 3)
4534                         continue;
4535                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4536         }
4537         wmb();
4538         pci_write_config_word(pdev, 4, saved_config_space[2]);
4539
4540         /* Some devices (notably the HP Smart Array 5i Controller)
4541            need a little pause here */
4542         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4543
4544         /* Controller should be in simple mode at this point.  If it's not,
4545          * It means we're on one of those controllers which doesn't support
4546          * the doorbell reset method and on which the PCI power management reset
4547          * method doesn't work (P800, for example.)
4548          * In those cases, don't try to proceed, as it generally doesn't work.
4549          */
4550         active_transport = readl(&cfgtable->TransportActive);
4551         if (active_transport & PERFORMANT_MODE) {
4552                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4553                         " Ignoring controller.\n");
4554                 rc = -ENODEV;
4555         }
4556
4557 unmap_cfgtable:
4558         iounmap(cfgtable);
4559
4560 unmap_vaddr:
4561         iounmap(vaddr);
4562         return rc;
4563 }
4564
4565 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4566 {
4567         int rc, i;
4568
4569         if (!reset_devices)
4570                 return 0;
4571
4572         /* Reset the controller with a PCI power-cycle or via doorbell */
4573         rc = cciss_kdump_hard_reset_controller(pdev);
4574
4575         /* -ENOTSUPP here means we cannot reset the controller
4576          * but it's already (and still) up and running in
4577          * "performant mode".  Or, it might be 640x, which can't reset
4578          * due to concerns about shared bbwc between 6402/6404 pair.
4579          */
4580         if (rc == -ENOTSUPP)
4581                 return 0; /* just try to do the kdump anyhow. */
4582         if (rc)
4583                 return -ENODEV;
4584         if (cciss_reset_msi(pdev))
4585                 return -ENODEV;
4586
4587         /* Now try to get the controller to respond to a no-op */
4588         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4589                 if (cciss_noop(pdev) == 0)
4590                         break;
4591                 else
4592                         dev_warn(&pdev->dev, "no-op failed%s\n",
4593                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4594                                         "; re-trying" : ""));
4595                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4596         }
4597         return 0;
4598 }
4599
4600 /*
4601  *  This is it.  Find all the controllers and register them.  I really hate
4602  *  stealing all these major device numbers.
4603  *  returns the number of block devices registered.
4604  */
4605 static int __devinit cciss_init_one(struct pci_dev *pdev,
4606                                     const struct pci_device_id *ent)
4607 {
4608         int i;
4609         int j = 0;
4610         int k = 0;
4611         int rc;
4612         int dac, return_code;
4613         InquiryData_struct *inq_buff;
4614         ctlr_info_t *h;
4615
4616         rc = cciss_init_reset_devices(pdev);
4617         if (rc)
4618                 return rc;
4619         i = alloc_cciss_hba(pdev);
4620         if (i < 0)
4621                 return -1;
4622
4623         h = hba[i];
4624         h->pdev = pdev;
4625         h->busy_initializing = 1;
4626         INIT_HLIST_HEAD(&h->cmpQ);
4627         INIT_HLIST_HEAD(&h->reqQ);
4628         mutex_init(&h->busy_shutting_down);
4629
4630         if (cciss_pci_init(h) != 0)
4631                 goto clean_no_release_regions;
4632
4633         sprintf(h->devname, "cciss%d", i);
4634         h->ctlr = i;
4635
4636         init_completion(&h->scan_wait);
4637
4638         if (cciss_create_hba_sysfs_entry(h))
4639                 goto clean0;
4640
4641         /* configure PCI DMA stuff */
4642         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4643                 dac = 1;
4644         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4645                 dac = 0;
4646         else {
4647                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4648                 goto clean1;
4649         }
4650
4651         /*
4652          * register with the major number, or get a dynamic major number
4653          * by passing 0 as argument.  This is done for greater than
4654          * 8 controller support.
4655          */
4656         if (i < MAX_CTLR_ORIG)
4657                 h->major = COMPAQ_CISS_MAJOR + i;
4658         rc = register_blkdev(h->major, h->devname);
4659         if (rc == -EBUSY || rc == -EINVAL) {
4660                 dev_err(&h->pdev->dev,
4661                        "Unable to get major number %d for %s "
4662                        "on hba %d\n", h->major, h->devname, i);
4663                 goto clean1;
4664         } else {
4665                 if (i >= MAX_CTLR_ORIG)
4666                         h->major = rc;
4667         }
4668
4669         /* make sure the board interrupts are off */
4670         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4671         if (h->msi_vector || h->msix_vector) {
4672                 if (request_irq(h->intr[PERF_MODE_INT],
4673                                 do_cciss_msix_intr,
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         } else {
4680                 if (request_irq(h->intr[PERF_MODE_INT], do_cciss_intx,
4681                                 IRQF_DISABLED, h->devname, h)) {
4682                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4683                                h->intr[PERF_MODE_INT], h->devname);
4684                         goto clean2;
4685                 }
4686         }
4687
4688         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4689                h->devname, pdev->device, pci_name(pdev),
4690                h->intr[PERF_MODE_INT], dac ? "" : " not");
4691
4692         h->cmd_pool_bits =
4693             kmalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4694                         * sizeof(unsigned long), GFP_KERNEL);
4695         h->cmd_pool = (CommandList_struct *)
4696             pci_alloc_consistent(h->pdev,
4697                     h->nr_cmds * sizeof(CommandList_struct),
4698                     &(h->cmd_pool_dhandle));
4699         h->errinfo_pool = (ErrorInfo_struct *)
4700             pci_alloc_consistent(h->pdev,
4701                     h->nr_cmds * sizeof(ErrorInfo_struct),
4702                     &(h->errinfo_pool_dhandle));
4703         if ((h->cmd_pool_bits == NULL)
4704             || (h->cmd_pool == NULL)
4705             || (h->errinfo_pool == NULL)) {
4706                 dev_err(&h->pdev->dev, "out of memory");
4707                 goto clean4;
4708         }
4709
4710         /* Need space for temp scatter list */
4711         h->scatter_list = kmalloc(h->max_commands *
4712                                                 sizeof(struct scatterlist *),
4713                                                 GFP_KERNEL);
4714         for (k = 0; k < h->nr_cmds; k++) {
4715                 h->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4716                                                         h->maxsgentries,
4717                                                         GFP_KERNEL);
4718                 if (h->scatter_list[k] == NULL) {
4719                         dev_err(&h->pdev->dev,
4720                                 "could not allocate s/g lists\n");
4721                         goto clean4;
4722                 }
4723         }
4724         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4725                 h->chainsize, h->nr_cmds);
4726         if (!h->cmd_sg_list && h->chainsize > 0)
4727                 goto clean4;
4728
4729         spin_lock_init(&h->lock);
4730
4731         /* Initialize the pdev driver private data.
4732            have it point to h.  */
4733         pci_set_drvdata(pdev, h);
4734         /* command and error info recs zeroed out before
4735            they are used */
4736         memset(h->cmd_pool_bits, 0,
4737                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4738                         * sizeof(unsigned long));
4739
4740         h->num_luns = 0;
4741         h->highest_lun = -1;
4742         for (j = 0; j < CISS_MAX_LUN; j++) {
4743                 h->drv[j] = NULL;
4744                 h->gendisk[j] = NULL;
4745         }
4746
4747         cciss_scsi_setup(h);
4748
4749         /* Turn the interrupts on so we can service requests */
4750         h->access.set_intr_mask(h, CCISS_INTR_ON);
4751
4752         /* Get the firmware version */
4753         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4754         if (inq_buff == NULL) {
4755                 dev_err(&h->pdev->dev, "out of memory\n");
4756                 goto clean4;
4757         }
4758
4759         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4760                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4761         if (return_code == IO_OK) {
4762                 h->firm_ver[0] = inq_buff->data_byte[32];
4763                 h->firm_ver[1] = inq_buff->data_byte[33];
4764                 h->firm_ver[2] = inq_buff->data_byte[34];
4765                 h->firm_ver[3] = inq_buff->data_byte[35];
4766         } else {         /* send command failed */
4767                 dev_warn(&h->pdev->dev, "unable to determine firmware"
4768                         " version of controller\n");
4769         }
4770         kfree(inq_buff);
4771
4772         cciss_procinit(h);
4773
4774         h->cciss_max_sectors = 8192;
4775
4776         rebuild_lun_table(h, 1, 0);
4777         h->busy_initializing = 0;
4778         return 1;
4779
4780 clean4:
4781         kfree(h->cmd_pool_bits);
4782         /* Free up sg elements */
4783         for (k = 0; k < h->nr_cmds; k++)
4784                 kfree(h->scatter_list[k]);
4785         kfree(h->scatter_list);
4786         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4787         if (h->cmd_pool)
4788                 pci_free_consistent(h->pdev,
4789                                     h->nr_cmds * sizeof(CommandList_struct),
4790                                     h->cmd_pool, h->cmd_pool_dhandle);
4791         if (h->errinfo_pool)
4792                 pci_free_consistent(h->pdev,
4793                                     h->nr_cmds * sizeof(ErrorInfo_struct),
4794                                     h->errinfo_pool,
4795                                     h->errinfo_pool_dhandle);
4796         free_irq(h->intr[PERF_MODE_INT], h);
4797 clean2:
4798         unregister_blkdev(h->major, h->devname);
4799 clean1:
4800         cciss_destroy_hba_sysfs_entry(h);
4801 clean0:
4802         pci_release_regions(pdev);
4803 clean_no_release_regions:
4804         h->busy_initializing = 0;
4805
4806         /*
4807          * Deliberately omit pci_disable_device(): it does something nasty to
4808          * Smart Array controllers that pci_enable_device does not undo
4809          */
4810         pci_set_drvdata(pdev, NULL);
4811         free_hba(h);
4812         return -1;
4813 }
4814
4815 static void cciss_shutdown(struct pci_dev *pdev)
4816 {
4817         ctlr_info_t *h;
4818         char *flush_buf;
4819         int return_code;
4820
4821         h = pci_get_drvdata(pdev);
4822         flush_buf = kzalloc(4, GFP_KERNEL);
4823         if (!flush_buf) {
4824                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
4825                 return;
4826         }
4827         /* write all data in the battery backed cache to disk */
4828         memset(flush_buf, 0, 4);
4829         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4830                 4, 0, CTLR_LUNID, TYPE_CMD);
4831         kfree(flush_buf);
4832         if (return_code != IO_OK)
4833                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
4834         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4835         free_irq(h->intr[PERF_MODE_INT], h);
4836 }
4837
4838 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4839 {
4840         ctlr_info_t *h;
4841         int i, j;
4842
4843         if (pci_get_drvdata(pdev) == NULL) {
4844                 dev_err(&pdev->dev, "Unable to remove device\n");
4845                 return;
4846         }
4847
4848         h = pci_get_drvdata(pdev);
4849         i = h->ctlr;
4850         if (hba[i] == NULL) {
4851                 dev_err(&pdev->dev, "device appears to already be removed\n");
4852                 return;
4853         }
4854
4855         mutex_lock(&h->busy_shutting_down);
4856
4857         remove_from_scan_list(h);
4858         remove_proc_entry(h->devname, proc_cciss);
4859         unregister_blkdev(h->major, h->devname);
4860
4861         /* remove it from the disk list */
4862         for (j = 0; j < CISS_MAX_LUN; j++) {
4863                 struct gendisk *disk = h->gendisk[j];
4864                 if (disk) {
4865                         struct request_queue *q = disk->queue;
4866
4867                         if (disk->flags & GENHD_FL_UP) {
4868                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
4869                                 del_gendisk(disk);
4870                         }
4871                         if (q)
4872                                 blk_cleanup_queue(q);
4873                 }
4874         }
4875
4876 #ifdef CONFIG_CISS_SCSI_TAPE
4877         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
4878 #endif
4879
4880         cciss_shutdown(pdev);
4881
4882 #ifdef CONFIG_PCI_MSI
4883         if (h->msix_vector)
4884                 pci_disable_msix(h->pdev);
4885         else if (h->msi_vector)
4886                 pci_disable_msi(h->pdev);
4887 #endif                          /* CONFIG_PCI_MSI */
4888
4889         iounmap(h->transtable);
4890         iounmap(h->cfgtable);
4891         iounmap(h->vaddr);
4892
4893         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(CommandList_struct),
4894                             h->cmd_pool, h->cmd_pool_dhandle);
4895         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(ErrorInfo_struct),
4896                             h->errinfo_pool, h->errinfo_pool_dhandle);
4897         kfree(h->cmd_pool_bits);
4898         /* Free up sg elements */
4899         for (j = 0; j < h->nr_cmds; j++)
4900                 kfree(h->scatter_list[j]);
4901         kfree(h->scatter_list);
4902         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4903         /*
4904          * Deliberately omit pci_disable_device(): it does something nasty to
4905          * Smart Array controllers that pci_enable_device does not undo
4906          */
4907         pci_release_regions(pdev);
4908         pci_set_drvdata(pdev, NULL);
4909         cciss_destroy_hba_sysfs_entry(h);
4910         mutex_unlock(&h->busy_shutting_down);
4911         free_hba(h);
4912 }
4913
4914 static struct pci_driver cciss_pci_driver = {
4915         .name = "cciss",
4916         .probe = cciss_init_one,
4917         .remove = __devexit_p(cciss_remove_one),
4918         .id_table = cciss_pci_device_id,        /* id_table */
4919         .shutdown = cciss_shutdown,
4920 };
4921
4922 /*
4923  *  This is it.  Register the PCI driver information for the cards we control
4924  *  the OS will call our registered routines when it finds one of our cards.
4925  */
4926 static int __init cciss_init(void)
4927 {
4928         int err;
4929
4930         /*
4931          * The hardware requires that commands are aligned on a 64-bit
4932          * boundary. Given that we use pci_alloc_consistent() to allocate an
4933          * array of them, the size must be a multiple of 8 bytes.
4934          */
4935         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4936         printk(KERN_INFO DRIVER_NAME "\n");
4937
4938         err = bus_register(&cciss_bus_type);
4939         if (err)
4940                 return err;
4941
4942         /* Start the scan thread */
4943         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4944         if (IS_ERR(cciss_scan_thread)) {
4945                 err = PTR_ERR(cciss_scan_thread);
4946                 goto err_bus_unregister;
4947         }
4948
4949         /* Register for our PCI devices */
4950         err = pci_register_driver(&cciss_pci_driver);
4951         if (err)
4952                 goto err_thread_stop;
4953
4954         return err;
4955
4956 err_thread_stop:
4957         kthread_stop(cciss_scan_thread);
4958 err_bus_unregister:
4959         bus_unregister(&cciss_bus_type);
4960
4961         return err;
4962 }
4963
4964 static void __exit cciss_cleanup(void)
4965 {
4966         int i;
4967
4968         pci_unregister_driver(&cciss_pci_driver);
4969         /* double check that all controller entrys have been removed */
4970         for (i = 0; i < MAX_CTLR; i++) {
4971                 if (hba[i] != NULL) {
4972                         dev_warn(&hba[i]->pdev->dev,
4973                                 "had to remove controller\n");
4974                         cciss_remove_one(hba[i]->pdev);
4975                 }
4976         }
4977         kthread_stop(cciss_scan_thread);
4978         remove_proc_entry("driver/cciss", NULL);
4979         bus_unregister(&cciss_bus_type);
4980 }
4981
4982 module_init(cciss_init);
4983 module_exit(cciss_cleanup);