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 list_head *list, CommandList_struct *c)
 270 {
 271         list_add_tail(&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(list_empty(&c->list))) {
 285                 c->cmd_type = CMD_MSG_STALE;
 286                 return;
 287         }
 288
 289         list_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         if (h->Qdepth > h->maxQsinceinit)
 301                 h->maxQsinceinit = h->Qdepth;
 302         start_io(h);
 303         spin_unlock_irqrestore(&h->lock, flags);
 304 }
 305
 306 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 307         int nr_cmds)
 308 {
 309         int i;
 310
 311         if (!cmd_sg_list)
 312                 return;
 313         for (i = 0; i < nr_cmds; i++) {
 314                 kfree(cmd_sg_list[i]);
 315                 cmd_sg_list[i] = NULL;
 316         }
 317         kfree(cmd_sg_list);
 318 }
 319
 320 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 321         ctlr_info_t *h, int chainsize, int nr_cmds)
 322 {
 323         int j;
 324         SGDescriptor_struct **cmd_sg_list;
 325
 326         if (chainsize <= 0)
 327                 return NULL;
 328
 329         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 330         if (!cmd_sg_list)
 331                 return NULL;
 332
 333         /* Build up chain blocks for each command */
 334         for (j = 0; j < nr_cmds; j++) {
 335                 /* Need a block of chainsized s/g elements. */
 336                 cmd_sg_list[j] = kmalloc((chainsize *
 337                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 338                 if (!cmd_sg_list[j]) {
 339                         dev_err(&h->pdev->dev, "Cannot get memory "
 340                                 "for s/g chains.\n");
 341                         goto clean;
 342                 }
 343         }
 344         return cmd_sg_list;
 345 clean:
 346         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 347         return NULL;
 348 }
 349
 350 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 351 {
 352         SGDescriptor_struct *chain_sg;
 353         u64bit temp64;
 354
 355         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 356                 return;
 357
 358         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 359         temp64.val32.lower = chain_sg->Addr.lower;
 360         temp64.val32.upper = chain_sg->Addr.upper;
 361         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 362 }
 363
 364 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 365         SGDescriptor_struct *chain_block, int len)
 366 {
 367         SGDescriptor_struct *chain_sg;
 368         u64bit temp64;
 369
 370         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 371         chain_sg->Ext = CCISS_SG_CHAIN;
 372         chain_sg->Len = len;
 373         temp64.val = pci_map_single(h->pdev, chain_block, len,
 374                                 PCI_DMA_TODEVICE);
 375         chain_sg->Addr.lower = temp64.val32.lower;
 376         chain_sg->Addr.upper = temp64.val32.upper;
 377 }
 378
 379 #include "cciss_scsi.c"         /* For SCSI tape support */
 380
 381 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 382         "UNKNOWN"
 383 };
 384 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 385
 386 #ifdef CONFIG_PROC_FS
 387
 388 /*
 389  * Report information about this controller.
 390  */
 391 #define ENG_GIG 1000000000
 392 #define ENG_GIG_FACTOR (ENG_GIG/512)
 393 #define ENGAGE_SCSI     "engage scsi"
 394
 395 static struct proc_dir_entry *proc_cciss;
 396
 397 static void cciss_seq_show_header(struct seq_file *seq)
 398 {
 399         ctlr_info_t *h = seq->private;
 400
 401         seq_printf(seq, "%s: HP %s Controller\n"
 402                 "Board ID: 0x%08lx\n"
 403                 "Firmware Version: %c%c%c%c\n"
 404                 "IRQ: %d\n"
 405                 "Logical drives: %d\n"
 406                 "Current Q depth: %d\n"
 407                 "Current # commands on controller: %d\n"
 408                 "Max Q depth since init: %d\n"
 409                 "Max # commands on controller since init: %d\n"
 410                 "Max SG entries since init: %d\n",
 411                 h->devname,
 412                 h->product_name,
 413                 (unsigned long)h->board_id,
 414                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 415                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 416                 h->num_luns,
 417                 h->Qdepth, h->commands_outstanding,
 418                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 419
 420 #ifdef CONFIG_CISS_SCSI_TAPE
 421         cciss_seq_tape_report(seq, h);
 422 #endif /* CONFIG_CISS_SCSI_TAPE */
 423 }
 424
 425 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 426 {
 427         ctlr_info_t *h = seq->private;
 428         unsigned long flags;
 429
 430         /* prevent displaying bogus info during configuration
 431          * or deconfiguration of a logical volume
 432          */
 433         spin_lock_irqsave(&h->lock, flags);
 434         if (h->busy_configuring) {
 435                 spin_unlock_irqrestore(&h->lock, flags);
 436                 return ERR_PTR(-EBUSY);
 437         }
 438         h->busy_configuring = 1;
 439         spin_unlock_irqrestore(&h->lock, flags);
 440
 441         if (*pos == 0)
 442                 cciss_seq_show_header(seq);
 443
 444         return pos;
 445 }
 446
 447 static int cciss_seq_show(struct seq_file *seq, void *v)
 448 {
 449         sector_t vol_sz, vol_sz_frac;
 450         ctlr_info_t *h = seq->private;
 451         unsigned ctlr = h->ctlr;
 452         loff_t *pos = v;
 453         drive_info_struct *drv = h->drv[*pos];
 454
 455         if (*pos > h->highest_lun)
 456                 return 0;
 457
 458         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 459                 return 0;
 460
 461         if (drv->heads == 0)
 462                 return 0;
 463
 464         vol_sz = drv->nr_blocks;
 465         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 466         vol_sz_frac *= 100;
 467         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 468
 469         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 470                 drv->raid_level = RAID_UNKNOWN;
 471         seq_printf(seq, "cciss/c%dd%d:"
 472                         "\t%4u.%02uGB\tRAID %s\n",
 473                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 474                         raid_label[drv->raid_level]);
 475         return 0;
 476 }
 477
 478 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 479 {
 480         ctlr_info_t *h = seq->private;
 481
 482         if (*pos > h->highest_lun)
 483                 return NULL;
 484         *pos += 1;
 485
 486         return pos;
 487 }
 488
 489 static void cciss_seq_stop(struct seq_file *seq, void *v)
 490 {
 491         ctlr_info_t *h = seq->private;
 492
 493         /* Only reset h->busy_configuring if we succeeded in setting
 494          * it during cciss_seq_start. */
 495         if (v == ERR_PTR(-EBUSY))
 496                 return;
 497
 498         h->busy_configuring = 0;
 499 }
 500
 501 static const struct seq_operations cciss_seq_ops = {
 502         .start = cciss_seq_start,
 503         .show  = cciss_seq_show,
 504         .next  = cciss_seq_next,
 505         .stop  = cciss_seq_stop,
 506 };
 507
 508 static int cciss_seq_open(struct inode *inode, struct file *file)
 509 {
 510         int ret = seq_open(file, &cciss_seq_ops);
 511         struct seq_file *seq = file->private_data;
 512
 513         if (!ret)
 514                 seq->private = PDE(inode)->data;
 515
 516         return ret;
 517 }
 518
 519 static ssize_t
 520 cciss_proc_write(struct file *file, const char __user *buf,
 521                  size_t length, loff_t *ppos)
 522 {
 523         int err;
 524         char *buffer;
 525
 526 #ifndef CONFIG_CISS_SCSI_TAPE
 527         return -EINVAL;
 528 #endif
 529
 530         if (!buf || length > PAGE_SIZE - 1)
 531                 return -EINVAL;
 532
 533         buffer = (char *)__get_free_page(GFP_KERNEL);
 534         if (!buffer)
 535                 return -ENOMEM;
 536
 537         err = -EFAULT;
 538         if (copy_from_user(buffer, buf, length))
 539                 goto out;
 540         buffer[length] = '\0';
 541
 542 #ifdef CONFIG_CISS_SCSI_TAPE
 543         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 544                 struct seq_file *seq = file->private_data;
 545                 ctlr_info_t *h = seq->private;
 546
 547                 err = cciss_engage_scsi(h);
 548                 if (err == 0)
 549                         err = length;
 550         } else
 551 #endif /* CONFIG_CISS_SCSI_TAPE */
 552                 err = -EINVAL;
 553         /* might be nice to have "disengage" too, but it's not
 554            safely possible. (only 1 module use count, lock issues.) */
 555
 556 out:
 557         free_page((unsigned long)buffer);
 558         return err;
 559 }
 560
 561 static const struct file_operations cciss_proc_fops = {
 562         .owner   = THIS_MODULE,
 563         .open    = cciss_seq_open,
 564         .read    = seq_read,
 565         .llseek  = seq_lseek,
 566         .release = seq_release,
 567         .write   = cciss_proc_write,
 568 };
 569
 570 static void __devinit cciss_procinit(ctlr_info_t *h)
 571 {
 572         struct proc_dir_entry *pde;
 573
 574         if (proc_cciss == NULL)
 575                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 576         if (!proc_cciss)
 577                 return;
 578         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 579                                         S_IROTH, proc_cciss,
 580                                         &cciss_proc_fops, h);
 581 }
 582 #endif                          /* CONFIG_PROC_FS */
 583
 584 #define MAX_PRODUCT_NAME_LEN 19
 585
 586 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 587 #define to_drv(n) container_of(n, drive_info_struct, dev)
 588
 589 static ssize_t host_store_rescan(struct device *dev,
 590                                  struct device_attribute *attr,
 591                                  const char *buf, size_t count)
 592 {
 593         struct ctlr_info *h = to_hba(dev);
 594
 595         add_to_scan_list(h);
 596         wake_up_process(cciss_scan_thread);
 597         wait_for_completion_interruptible(&h->scan_wait);
 598
 599         return count;
 600 }
 601 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 602
 603 static ssize_t dev_show_unique_id(struct device *dev,
 604                                  struct device_attribute *attr,
 605                                  char *buf)
 606 {
 607         drive_info_struct *drv = to_drv(dev);
 608         struct ctlr_info *h = to_hba(drv->dev.parent);
 609         __u8 sn[16];
 610         unsigned long flags;
 611         int ret = 0;
 612
 613         spin_lock_irqsave(&h->lock, flags);
 614         if (h->busy_configuring)
 615                 ret = -EBUSY;
 616         else
 617                 memcpy(sn, drv->serial_no, sizeof(sn));
 618         spin_unlock_irqrestore(&h->lock, flags);
 619
 620         if (ret)
 621                 return ret;
 622         else
 623                 return snprintf(buf, 16 * 2 + 2,
 624                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 625                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 626                                 sn[0], sn[1], sn[2], sn[3],
 627                                 sn[4], sn[5], sn[6], sn[7],
 628                                 sn[8], sn[9], sn[10], sn[11],
 629                                 sn[12], sn[13], sn[14], sn[15]);
 630 }
 631 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 632
 633 static ssize_t dev_show_vendor(struct device *dev,
 634                                struct device_attribute *attr,
 635                                char *buf)
 636 {
 637         drive_info_struct *drv = to_drv(dev);
 638         struct ctlr_info *h = to_hba(drv->dev.parent);
 639         char vendor[VENDOR_LEN + 1];
 640         unsigned long flags;
 641         int ret = 0;
 642
 643         spin_lock_irqsave(&h->lock, flags);
 644         if (h->busy_configuring)
 645                 ret = -EBUSY;
 646         else
 647                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 648         spin_unlock_irqrestore(&h->lock, flags);
 649
 650         if (ret)
 651                 return ret;
 652         else
 653                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 654 }
 655 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 656
 657 static ssize_t dev_show_model(struct device *dev,
 658                               struct device_attribute *attr,
 659                               char *buf)
 660 {
 661         drive_info_struct *drv = to_drv(dev);
 662         struct ctlr_info *h = to_hba(drv->dev.parent);
 663         char model[MODEL_LEN + 1];
 664         unsigned long flags;
 665         int ret = 0;
 666
 667         spin_lock_irqsave(&h->lock, flags);
 668         if (h->busy_configuring)
 669                 ret = -EBUSY;
 670         else
 671                 memcpy(model, drv->model, MODEL_LEN + 1);
 672         spin_unlock_irqrestore(&h->lock, flags);
 673
 674         if (ret)
 675                 return ret;
 676         else
 677                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 678 }
 679 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 680
 681 static ssize_t dev_show_rev(struct device *dev,
 682                             struct device_attribute *attr,
 683                             char *buf)
 684 {
 685         drive_info_struct *drv = to_drv(dev);
 686         struct ctlr_info *h = to_hba(drv->dev.parent);
 687         char rev[REV_LEN + 1];
 688         unsigned long flags;
 689         int ret = 0;
 690
 691         spin_lock_irqsave(&h->lock, flags);
 692         if (h->busy_configuring)
 693                 ret = -EBUSY;
 694         else
 695                 memcpy(rev, drv->rev, REV_LEN + 1);
 696         spin_unlock_irqrestore(&h->lock, flags);
 697
 698         if (ret)
 699                 return ret;
 700         else
 701                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 702 }
 703 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 704
 705 static ssize_t cciss_show_lunid(struct device *dev,
 706                                 struct device_attribute *attr, char *buf)
 707 {
 708         drive_info_struct *drv = to_drv(dev);
 709         struct ctlr_info *h = to_hba(drv->dev.parent);
 710         unsigned long flags;
 711         unsigned char lunid[8];
 712
 713         spin_lock_irqsave(&h->lock, flags);
 714         if (h->busy_configuring) {
 715                 spin_unlock_irqrestore(&h->lock, flags);
 716                 return -EBUSY;
 717         }
 718         if (!drv->heads) {
 719                 spin_unlock_irqrestore(&h->lock, flags);
 720                 return -ENOTTY;
 721         }
 722         memcpy(lunid, drv->LunID, sizeof(lunid));
 723         spin_unlock_irqrestore(&h->lock, flags);
 724         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 725                 lunid[0], lunid[1], lunid[2], lunid[3],
 726                 lunid[4], lunid[5], lunid[6], lunid[7]);
 727 }
 728 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 729
 730 static ssize_t cciss_show_raid_level(struct device *dev,
 731                                      struct device_attribute *attr, char *buf)
 732 {
 733         drive_info_struct *drv = to_drv(dev);
 734         struct ctlr_info *h = to_hba(drv->dev.parent);
 735         int raid;
 736         unsigned long flags;
 737
 738         spin_lock_irqsave(&h->lock, flags);
 739         if (h->busy_configuring) {
 740                 spin_unlock_irqrestore(&h->lock, flags);
 741                 return -EBUSY;
 742         }
 743         raid = drv->raid_level;
 744         spin_unlock_irqrestore(&h->lock, flags);
 745         if (raid < 0 || raid > RAID_UNKNOWN)
 746                 raid = RAID_UNKNOWN;
 747
 748         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 749                         raid_label[raid]);
 750 }
 751 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 752
 753 static ssize_t cciss_show_usage_count(struct device *dev,
 754                                       struct device_attribute *attr, char *buf)
 755 {
 756         drive_info_struct *drv = to_drv(dev);
 757         struct ctlr_info *h = to_hba(drv->dev.parent);
 758         unsigned long flags;
 759         int count;
 760
 761         spin_lock_irqsave(&h->lock, flags);
 762         if (h->busy_configuring) {
 763                 spin_unlock_irqrestore(&h->lock, flags);
 764                 return -EBUSY;
 765         }
 766         count = drv->usage_count;
 767         spin_unlock_irqrestore(&h->lock, flags);
 768         return snprintf(buf, 20, "%d\n", count);
 769 }
 770 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 771
 772 static struct attribute *cciss_host_attrs[] = {
 773         &dev_attr_rescan.attr,
 774         NULL
 775 };
 776
 777 static struct attribute_group cciss_host_attr_group = {
 778         .attrs = cciss_host_attrs,
 779 };
 780
 781 static const struct attribute_group *cciss_host_attr_groups[] = {
 782         &cciss_host_attr_group,
 783         NULL
 784 };
 785
 786 static struct device_type cciss_host_type = {
 787         .name           = "cciss_host",
 788         .groups         = cciss_host_attr_groups,
 789         .release        = cciss_hba_release,
 790 };
 791
 792 static struct attribute *cciss_dev_attrs[] = {
 793         &dev_attr_unique_id.attr,
 794         &dev_attr_model.attr,
 795         &dev_attr_vendor.attr,
 796         &dev_attr_rev.attr,
 797         &dev_attr_lunid.attr,
 798         &dev_attr_raid_level.attr,
 799         &dev_attr_usage_count.attr,
 800         NULL
 801 };
 802
 803 static struct attribute_group cciss_dev_attr_group = {
 804         .attrs = cciss_dev_attrs,
 805 };
 806
 807 static const struct attribute_group *cciss_dev_attr_groups[] = {
 808         &cciss_dev_attr_group,
 809         NULL
 810 };
 811
 812 static struct device_type cciss_dev_type = {
 813         .name           = "cciss_device",
 814         .groups         = cciss_dev_attr_groups,
 815         .release        = cciss_device_release,
 816 };
 817
 818 static struct bus_type cciss_bus_type = {
 819         .name           = "cciss",
 820 };
 821
 822 /*
 823  * cciss_hba_release is called when the reference count
 824  * of h->dev goes to zero.
 825  */
 826 static void cciss_hba_release(struct device *dev)
 827 {
 828         /*
 829          * nothing to do, but need this to avoid a warning
 830          * about not having a release handler from lib/kref.c.
 831          */
 832 }
 833
 834 /*
 835  * Initialize sysfs entry for each controller.  This sets up and registers
 836  * the 'cciss#' directory for each individual controller under
 837  * /sys/bus/pci/devices/<dev>/.
 838  */
 839 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 840 {
 841         device_initialize(&h->dev);
 842         h->dev.type = &cciss_host_type;
 843         h->dev.bus = &cciss_bus_type;
 844         dev_set_name(&h->dev, "%s", h->devname);
 845         h->dev.parent = &h->pdev->dev;
 846
 847         return device_add(&h->dev);
 848 }
 849
 850 /*
 851  * Remove sysfs entries for an hba.
 852  */
 853 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 854 {
 855         device_del(&h->dev);
 856         put_device(&h->dev); /* final put. */
 857 }
 858
 859 /* cciss_device_release is called when the reference count
 860  * of h->drv[x]dev goes to zero.
 861  */
 862 static void cciss_device_release(struct device *dev)
 863 {
 864         drive_info_struct *drv = to_drv(dev);
 865         kfree(drv);
 866 }
 867
 868 /*
 869  * Initialize sysfs for each logical drive.  This sets up and registers
 870  * the 'c#d#' directory for each individual logical drive under
 871  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 872  * /sys/block/cciss!c#d# to this entry.
 873  */
 874 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 875                                        int drv_index)
 876 {
 877         struct device *dev;
 878
 879         if (h->drv[drv_index]->device_initialized)
 880                 return 0;
 881
 882         dev = &h->drv[drv_index]->dev;
 883         device_initialize(dev);
 884         dev->type = &cciss_dev_type;
 885         dev->bus = &cciss_bus_type;
 886         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 887         dev->parent = &h->dev;
 888         h->drv[drv_index]->device_initialized = 1;
 889         return device_add(dev);
 890 }
 891
 892 /*
 893  * Remove sysfs entries for a logical drive.
 894  */
 895 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 896         int ctlr_exiting)
 897 {
 898         struct device *dev = &h->drv[drv_index]->dev;
 899
 900         /* special case for c*d0, we only destroy it on controller exit */
 901         if (drv_index == 0 && !ctlr_exiting)
 902                 return;
 903
 904         device_del(dev);
 905         put_device(dev); /* the "final" put. */
 906         h->drv[drv_index] = NULL;
 907 }
 908
 909 /*
 910  * For operations that cannot sleep, a command block is allocated at init,
 911  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 912  * which ones are free or in use.
 913  */
 914 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 915 {
 916         CommandList_struct *c;
 917         int i;
 918         u64bit temp64;
 919         dma_addr_t cmd_dma_handle, err_dma_handle;
 920
 921         do {
 922                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 923                 if (i == h->nr_cmds)
 924                         return NULL;
 925         } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
 926                   h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 927         c = h->cmd_pool + i;
 928         memset(c, 0, sizeof(CommandList_struct));
 929         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
 930         c->err_info = h->errinfo_pool + i;
 931         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 932         err_dma_handle = h->errinfo_pool_dhandle
 933             + i * sizeof(ErrorInfo_struct);
 934         h->nr_allocs++;
 935
 936         c->cmdindex = i;
 937
 938         INIT_LIST_HEAD(&c->list);
 939         c->busaddr = (__u32) cmd_dma_handle;
 940         temp64.val = (__u64) err_dma_handle;
 941         c->ErrDesc.Addr.lower = temp64.val32.lower;
 942         c->ErrDesc.Addr.upper = temp64.val32.upper;
 943         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 944
 945         c->ctlr = h->ctlr;
 946         return c;
 947 }
 948
 949 /* allocate a command using pci_alloc_consistent, used for ioctls,
 950  * etc., not for the main i/o path.
 951  */
 952 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
 953 {
 954         CommandList_struct *c;
 955         u64bit temp64;
 956         dma_addr_t cmd_dma_handle, err_dma_handle;
 957
 958         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 959                 sizeof(CommandList_struct), &cmd_dma_handle);
 960         if (c == NULL)
 961                 return NULL;
 962         memset(c, 0, sizeof(CommandList_struct));
 963
 964         c->cmdindex = -1;
 965
 966         c->err_info = (ErrorInfo_struct *)
 967             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 968                     &err_dma_handle);
 969
 970         if (c->err_info == NULL) {
 971                 pci_free_consistent(h->pdev,
 972                         sizeof(CommandList_struct), c, cmd_dma_handle);
 973                 return NULL;
 974         }
 975         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 976
 977         INIT_LIST_HEAD(&c->list);
 978         c->busaddr = (__u32) cmd_dma_handle;
 979         temp64.val = (__u64) err_dma_handle;
 980         c->ErrDesc.Addr.lower = temp64.val32.lower;
 981         c->ErrDesc.Addr.upper = temp64.val32.upper;
 982         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 983
 984         c->ctlr = h->ctlr;
 985         return c;
 986 }
 987
 988 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
 989 {
 990         int i;
 991
 992         i = c - h->cmd_pool;
 993         clear_bit(i & (BITS_PER_LONG - 1),
 994                   h->cmd_pool_bits + (i / BITS_PER_LONG));
 995         h->nr_frees++;
 996 }
 997
 998 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
 999 {
1000         u64bit temp64;
1001
1002         temp64.val32.lower = c->ErrDesc.Addr.lower;
1003         temp64.val32.upper = c->ErrDesc.Addr.upper;
1004         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1005                             c->err_info, (dma_addr_t) temp64.val);
1006         pci_free_consistent(h->pdev, sizeof(CommandList_struct),
1007                             c, (dma_addr_t) c->busaddr);
1008 }
1009
1010 static inline ctlr_info_t *get_host(struct gendisk *disk)
1011 {
1012         return disk->queue->queuedata;
1013 }
1014
1015 static inline drive_info_struct *get_drv(struct gendisk *disk)
1016 {
1017         return disk->private_data;
1018 }
1019
1020 /*
1021  * Open.  Make sure the device is really there.
1022  */
1023 static int cciss_open(struct block_device *bdev, fmode_t mode)
1024 {
1025         ctlr_info_t *h = get_host(bdev->bd_disk);
1026         drive_info_struct *drv = get_drv(bdev->bd_disk);
1027
1028         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1029         if (drv->busy_configuring)
1030                 return -EBUSY;
1031         /*
1032          * Root is allowed to open raw volume zero even if it's not configured
1033          * so array config can still work. Root is also allowed to open any
1034          * volume that has a LUN ID, so it can issue IOCTL to reread the
1035          * disk information.  I don't think I really like this
1036          * but I'm already using way to many device nodes to claim another one
1037          * for "raw controller".
1038          */
1039         if (drv->heads == 0) {
1040                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1041                         /* if not node 0 make sure it is a partition = 0 */
1042                         if (MINOR(bdev->bd_dev) & 0x0f) {
1043                                 return -ENXIO;
1044                                 /* if it is, make sure we have a LUN ID */
1045                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1046                                 sizeof(drv->LunID))) {
1047                                 return -ENXIO;
1048                         }
1049                 }
1050                 if (!capable(CAP_SYS_ADMIN))
1051                         return -EPERM;
1052         }
1053         drv->usage_count++;
1054         h->usage_count++;
1055         return 0;
1056 }
1057
1058 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1059 {
1060         int ret;
1061
1062         lock_kernel();
1063         ret = cciss_open(bdev, mode);
1064         unlock_kernel();
1065
1066         return ret;
1067 }
1068
1069 /*
1070  * Close.  Sync first.
1071  */
1072 static int cciss_release(struct gendisk *disk, fmode_t mode)
1073 {
1074         ctlr_info_t *h;
1075         drive_info_struct *drv;
1076
1077         lock_kernel();
1078         h = get_host(disk);
1079         drv = get_drv(disk);
1080         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1081         drv->usage_count--;
1082         h->usage_count--;
1083         unlock_kernel();
1084         return 0;
1085 }
1086
1087 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1088                     unsigned cmd, unsigned long arg)
1089 {
1090         int ret;
1091         lock_kernel();
1092         ret = cciss_ioctl(bdev, mode, cmd, arg);
1093         unlock_kernel();
1094         return ret;
1095 }
1096
1097 #ifdef CONFIG_COMPAT
1098
1099 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1100                                   unsigned cmd, unsigned long arg);
1101 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1102                                       unsigned cmd, unsigned long arg);
1103
1104 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1105                               unsigned cmd, unsigned long arg)
1106 {
1107         switch (cmd) {
1108         case CCISS_GETPCIINFO:
1109         case CCISS_GETINTINFO:
1110         case CCISS_SETINTINFO:
1111         case CCISS_GETNODENAME:
1112         case CCISS_SETNODENAME:
1113         case CCISS_GETHEARTBEAT:
1114         case CCISS_GETBUSTYPES:
1115         case CCISS_GETFIRMVER:
1116         case CCISS_GETDRIVVER:
1117         case CCISS_REVALIDVOLS:
1118         case CCISS_DEREGDISK:
1119         case CCISS_REGNEWDISK:
1120         case CCISS_REGNEWD:
1121         case CCISS_RESCANDISK:
1122         case CCISS_GETLUNINFO:
1123                 return do_ioctl(bdev, mode, cmd, arg);
1124
1125         case CCISS_PASSTHRU32:
1126                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1127         case CCISS_BIG_PASSTHRU32:
1128                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1129
1130         default:
1131                 return -ENOIOCTLCMD;
1132         }
1133 }
1134
1135 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1136                                   unsigned cmd, unsigned long arg)
1137 {
1138         IOCTL32_Command_struct __user *arg32 =
1139             (IOCTL32_Command_struct __user *) arg;
1140         IOCTL_Command_struct arg64;
1141         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1142         int err;
1143         u32 cp;
1144
1145         err = 0;
1146         err |=
1147             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1148                            sizeof(arg64.LUN_info));
1149         err |=
1150             copy_from_user(&arg64.Request, &arg32->Request,
1151                            sizeof(arg64.Request));
1152         err |=
1153             copy_from_user(&arg64.error_info, &arg32->error_info,
1154                            sizeof(arg64.error_info));
1155         err |= get_user(arg64.buf_size, &arg32->buf_size);
1156         err |= get_user(cp, &arg32->buf);
1157         arg64.buf = compat_ptr(cp);
1158         err |= copy_to_user(p, &arg64, sizeof(arg64));
1159
1160         if (err)
1161                 return -EFAULT;
1162
1163         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1164         if (err)
1165                 return err;
1166         err |=
1167             copy_in_user(&arg32->error_info, &p->error_info,
1168                          sizeof(arg32->error_info));
1169         if (err)
1170                 return -EFAULT;
1171         return err;
1172 }
1173
1174 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1175                                       unsigned cmd, unsigned long arg)
1176 {
1177         BIG_IOCTL32_Command_struct __user *arg32 =
1178             (BIG_IOCTL32_Command_struct __user *) arg;
1179         BIG_IOCTL_Command_struct arg64;
1180         BIG_IOCTL_Command_struct __user *p =
1181             compat_alloc_user_space(sizeof(arg64));
1182         int err;
1183         u32 cp;
1184
1185         err = 0;
1186         err |=
1187             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1188                            sizeof(arg64.LUN_info));
1189         err |=
1190             copy_from_user(&arg64.Request, &arg32->Request,
1191                            sizeof(arg64.Request));
1192         err |=
1193             copy_from_user(&arg64.error_info, &arg32->error_info,
1194                            sizeof(arg64.error_info));
1195         err |= get_user(arg64.buf_size, &arg32->buf_size);
1196         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1197         err |= get_user(cp, &arg32->buf);
1198         arg64.buf = compat_ptr(cp);
1199         err |= copy_to_user(p, &arg64, sizeof(arg64));
1200
1201         if (err)
1202                 return -EFAULT;
1203
1204         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1205         if (err)
1206                 return err;
1207         err |=
1208             copy_in_user(&arg32->error_info, &p->error_info,
1209                          sizeof(arg32->error_info));
1210         if (err)
1211                 return -EFAULT;
1212         return err;
1213 }
1214 #endif
1215
1216 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1217 {
1218         drive_info_struct *drv = get_drv(bdev->bd_disk);
1219
1220         if (!drv->cylinders)
1221                 return -ENXIO;
1222
1223         geo->heads = drv->heads;
1224         geo->sectors = drv->sectors;
1225         geo->cylinders = drv->cylinders;
1226         return 0;
1227 }
1228
1229 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1230 {
1231         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1232                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1233                 (void)check_for_unit_attention(h, c);
1234 }
1235 /*
1236  * ioctl
1237  */
1238 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1239                        unsigned int cmd, unsigned long arg)
1240 {
1241         struct gendisk *disk = bdev->bd_disk;
1242         ctlr_info_t *h = get_host(disk);
1243         drive_info_struct *drv = get_drv(disk);
1244         void __user *argp = (void __user *)arg;
1245
1246         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1247                 cmd, arg);
1248         switch (cmd) {
1249         case CCISS_GETPCIINFO:
1250                 {
1251                         cciss_pci_info_struct pciinfo;
1252
1253                         if (!arg)
1254                                 return -EINVAL;
1255                         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1256                         pciinfo.bus = h->pdev->bus->number;
1257                         pciinfo.dev_fn = h->pdev->devfn;
1258                         pciinfo.board_id = h->board_id;
1259                         if (copy_to_user
1260                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1261                                 return -EFAULT;
1262                         return 0;
1263                 }
1264         case CCISS_GETINTINFO:
1265                 {
1266                         cciss_coalint_struct intinfo;
1267                         if (!arg)
1268                                 return -EINVAL;
1269                         intinfo.delay =
1270                             readl(&h->cfgtable->HostWrite.CoalIntDelay);
1271                         intinfo.count =
1272                             readl(&h->cfgtable->HostWrite.CoalIntCount);
1273                         if (copy_to_user
1274                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1275                                 return -EFAULT;
1276                         return 0;
1277                 }
1278         case CCISS_SETINTINFO:
1279                 {
1280                         cciss_coalint_struct intinfo;
1281                         unsigned long flags;
1282                         int i;
1283
1284                         if (!arg)
1285                                 return -EINVAL;
1286                         if (!capable(CAP_SYS_ADMIN))
1287                                 return -EPERM;
1288                         if (copy_from_user
1289                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1290                                 return -EFAULT;
1291                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1292                                 return -EINVAL;
1293                         spin_lock_irqsave(&h->lock, flags);
1294                         /* Update the field, and then ring the doorbell */
1295                         writel(intinfo.delay,
1296                                &(h->cfgtable->HostWrite.CoalIntDelay));
1297                         writel(intinfo.count,
1298                                &(h->cfgtable->HostWrite.CoalIntCount));
1299                         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1300
1301                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1302                                 if (!(readl(h->vaddr + SA5_DOORBELL)
1303                                       & CFGTBL_ChangeReq))
1304                                         break;
1305                                 /* delay and try again */
1306                                 udelay(1000);
1307                         }
1308                         spin_unlock_irqrestore(&h->lock, flags);
1309                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1310                                 return -EAGAIN;
1311                         return 0;
1312                 }
1313         case CCISS_GETNODENAME:
1314                 {
1315                         NodeName_type NodeName;
1316                         int i;
1317
1318                         if (!arg)
1319                                 return -EINVAL;
1320                         for (i = 0; i < 16; i++)
1321                                 NodeName[i] =
1322                                     readb(&h->cfgtable->ServerName[i]);
1323                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1324                                 return -EFAULT;
1325                         return 0;
1326                 }
1327         case CCISS_SETNODENAME:
1328                 {
1329                         NodeName_type NodeName;
1330                         unsigned long flags;
1331                         int i;
1332
1333                         if (!arg)
1334                                 return -EINVAL;
1335                         if (!capable(CAP_SYS_ADMIN))
1336                                 return -EPERM;
1337
1338                         if (copy_from_user
1339                             (NodeName, argp, sizeof(NodeName_type)))
1340                                 return -EFAULT;
1341
1342                         spin_lock_irqsave(&h->lock, flags);
1343
1344                         /* Update the field, and then ring the doorbell */
1345                         for (i = 0; i < 16; i++)
1346                                 writeb(NodeName[i],
1347                                        &h->cfgtable->ServerName[i]);
1348
1349                         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1350
1351                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1352                                 if (!(readl(h->vaddr + SA5_DOORBELL)
1353                                       & CFGTBL_ChangeReq))
1354                                         break;
1355                                 /* delay and try again */
1356                                 udelay(1000);
1357                         }
1358                         spin_unlock_irqrestore(&h->lock, flags);
1359                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1360                                 return -EAGAIN;
1361                         return 0;
1362                 }
1363
1364         case CCISS_GETHEARTBEAT:
1365                 {
1366                         Heartbeat_type heartbeat;
1367
1368                         if (!arg)
1369                                 return -EINVAL;
1370                         heartbeat = readl(&h->cfgtable->HeartBeat);
1371                         if (copy_to_user
1372                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1373                                 return -EFAULT;
1374                         return 0;
1375                 }
1376         case CCISS_GETBUSTYPES:
1377                 {
1378                         BusTypes_type BusTypes;
1379
1380                         if (!arg)
1381                                 return -EINVAL;
1382                         BusTypes = readl(&h->cfgtable->BusTypes);
1383                         if (copy_to_user
1384                             (argp, &BusTypes, sizeof(BusTypes_type)))
1385                                 return -EFAULT;
1386                         return 0;
1387                 }
1388         case CCISS_GETFIRMVER:
1389                 {
1390                         FirmwareVer_type firmware;
1391
1392                         if (!arg)
1393                                 return -EINVAL;
1394                         memcpy(firmware, h->firm_ver, 4);
1395
1396                         if (copy_to_user
1397                             (argp, firmware, sizeof(FirmwareVer_type)))
1398                                 return -EFAULT;
1399                         return 0;
1400                 }
1401         case CCISS_GETDRIVVER:
1402                 {
1403                         DriverVer_type DriverVer = DRIVER_VERSION;
1404
1405                         if (!arg)
1406                                 return -EINVAL;
1407
1408                         if (copy_to_user
1409                             (argp, &DriverVer, sizeof(DriverVer_type)))
1410                                 return -EFAULT;
1411                         return 0;
1412                 }
1413
1414         case CCISS_DEREGDISK:
1415         case CCISS_REGNEWD:
1416         case CCISS_REVALIDVOLS:
1417                 return rebuild_lun_table(h, 0, 1);
1418
1419         case CCISS_GETLUNINFO:{
1420                         LogvolInfo_struct luninfo;
1421
1422                         memcpy(&luninfo.LunID, drv->LunID,
1423                                 sizeof(luninfo.LunID));
1424                         luninfo.num_opens = drv->usage_count;
1425                         luninfo.num_parts = 0;
1426                         if (copy_to_user(argp, &luninfo,
1427                                          sizeof(LogvolInfo_struct)))
1428                                 return -EFAULT;
1429                         return 0;
1430                 }
1431         case CCISS_PASSTHRU:
1432                 {
1433                         IOCTL_Command_struct iocommand;
1434                         CommandList_struct *c;
1435                         char *buff = NULL;
1436                         u64bit temp64;
1437                         DECLARE_COMPLETION_ONSTACK(wait);
1438
1439                         if (!arg)
1440                                 return -EINVAL;
1441
1442                         if (!capable(CAP_SYS_RAWIO))
1443                                 return -EPERM;
1444
1445                         if (copy_from_user
1446                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1447                                 return -EFAULT;
1448                         if ((iocommand.buf_size < 1) &&
1449                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1450                                 return -EINVAL;
1451                         }
1452 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1453                         /* Check kmalloc limits */
1454                         if (iocommand.buf_size > 128000)
1455                                 return -EINVAL;
1456 #endif
1457                         if (iocommand.buf_size > 0) {
1458                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1459                                 if (buff == NULL)
1460                                         return -EFAULT;
1461                         }
1462                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1463                                 /* Copy the data into the buffer we created */
1464                                 if (copy_from_user
1465                                     (buff, iocommand.buf, iocommand.buf_size)) {
1466                                         kfree(buff);
1467                                         return -EFAULT;
1468                                 }
1469                         } else {
1470                                 memset(buff, 0, iocommand.buf_size);
1471                         }
1472                         c = cmd_special_alloc(h);
1473                         if (!c) {
1474                                 kfree(buff);
1475                                 return -ENOMEM;
1476                         }
1477                         /* Fill in the command type */
1478                         c->cmd_type = CMD_IOCTL_PEND;
1479                         /* Fill in Command Header */
1480                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1481                         if (iocommand.buf_size > 0) /* buffer to fill */
1482                         {
1483                                 c->Header.SGList = 1;
1484                                 c->Header.SGTotal = 1;
1485                         } else /* no buffers to fill */
1486                         {
1487                                 c->Header.SGList = 0;
1488                                 c->Header.SGTotal = 0;
1489                         }
1490                         c->Header.LUN = iocommand.LUN_info;
1491                         /* use the kernel address the cmd block for tag */
1492                         c->Header.Tag.lower = c->busaddr;
1493
1494                         /* Fill in Request block */
1495                         c->Request = iocommand.Request;
1496
1497                         /* Fill in the scatter gather information */
1498                         if (iocommand.buf_size > 0) {
1499                                 temp64.val = pci_map_single(h->pdev, buff,
1500                                         iocommand.buf_size,
1501                                         PCI_DMA_BIDIRECTIONAL);
1502                                 c->SG[0].Addr.lower = temp64.val32.lower;
1503                                 c->SG[0].Addr.upper = temp64.val32.upper;
1504                                 c->SG[0].Len = iocommand.buf_size;
1505                                 c->SG[0].Ext = 0;  /* we are not chaining */
1506                         }
1507                         c->waiting = &wait;
1508
1509                         enqueue_cmd_and_start_io(h, c);
1510                         wait_for_completion(&wait);
1511
1512                         /* unlock the buffers from DMA */
1513                         temp64.val32.lower = c->SG[0].Addr.lower;
1514                         temp64.val32.upper = c->SG[0].Addr.upper;
1515                         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val,
1516                                          iocommand.buf_size,
1517                                          PCI_DMA_BIDIRECTIONAL);
1518
1519                         check_ioctl_unit_attention(h, c);
1520
1521                         /* Copy the error information out */
1522                         iocommand.error_info = *(c->err_info);
1523                         if (copy_to_user
1524                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1525                                 kfree(buff);
1526                                 cmd_special_free(h, c);
1527                                 return -EFAULT;
1528                         }
1529
1530                         if (iocommand.Request.Type.Direction == XFER_READ) {
1531                                 /* Copy the data out of the buffer we created */
1532                                 if (copy_to_user
1533                                     (iocommand.buf, buff, iocommand.buf_size)) {
1534                                         kfree(buff);
1535                                         cmd_special_free(h, c);
1536                                         return -EFAULT;
1537                                 }
1538                         }
1539                         kfree(buff);
1540                         cmd_special_free(h, c);
1541                         return 0;
1542                 }
1543         case CCISS_BIG_PASSTHRU:{
1544                         BIG_IOCTL_Command_struct *ioc;
1545                         CommandList_struct *c;
1546                         unsigned char **buff = NULL;
1547                         int *buff_size = NULL;
1548                         u64bit temp64;
1549                         BYTE sg_used = 0;
1550                         int status = 0;
1551                         int i;
1552                         DECLARE_COMPLETION_ONSTACK(wait);
1553                         __u32 left;
1554                         __u32 sz;
1555                         BYTE __user *data_ptr;
1556
1557                         if (!arg)
1558                                 return -EINVAL;
1559                         if (!capable(CAP_SYS_RAWIO))
1560                                 return -EPERM;
1561                         ioc = (BIG_IOCTL_Command_struct *)
1562                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1563                         if (!ioc) {
1564                                 status = -ENOMEM;
1565                                 goto cleanup1;
1566                         }
1567                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1568                                 status = -EFAULT;
1569                                 goto cleanup1;
1570                         }
1571                         if ((ioc->buf_size < 1) &&
1572                             (ioc->Request.Type.Direction != XFER_NONE)) {
1573                                 status = -EINVAL;
1574                                 goto cleanup1;
1575                         }
1576                         /* Check kmalloc limits  using all SGs */
1577                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1578                                 status = -EINVAL;
1579                                 goto cleanup1;
1580                         }
1581                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1582                                 status = -EINVAL;
1583                                 goto cleanup1;
1584                         }
1585                         buff =
1586                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1587                         if (!buff) {
1588                                 status = -ENOMEM;
1589                                 goto cleanup1;
1590                         }
1591                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1592                                                    GFP_KERNEL);
1593                         if (!buff_size) {
1594                                 status = -ENOMEM;
1595                                 goto cleanup1;
1596                         }
1597                         left = ioc->buf_size;
1598                         data_ptr = ioc->buf;
1599                         while (left) {
1600                                 sz = (left >
1601                                       ioc->malloc_size) ? ioc->
1602                                     malloc_size : left;
1603                                 buff_size[sg_used] = sz;
1604                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1605                                 if (buff[sg_used] == NULL) {
1606                                         status = -ENOMEM;
1607                                         goto cleanup1;
1608                                 }
1609                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1610                                         if (copy_from_user
1611                                             (buff[sg_used], data_ptr, sz)) {
1612                                                 status = -EFAULT;
1613                                                 goto cleanup1;
1614                                         }
1615                                 } else {
1616                                         memset(buff[sg_used], 0, sz);
1617                                 }
1618                                 left -= sz;
1619                                 data_ptr += sz;
1620                                 sg_used++;
1621                         }
1622                         c = cmd_special_alloc(h);
1623                         if (!c) {
1624                                 status = -ENOMEM;
1625                                 goto cleanup1;
1626                         }
1627                         c->cmd_type = CMD_IOCTL_PEND;
1628                         c->Header.ReplyQueue = 0;
1629
1630                         if (ioc->buf_size > 0) {
1631                                 c->Header.SGList = sg_used;
1632                                 c->Header.SGTotal = sg_used;
1633                         } else {
1634                                 c->Header.SGList = 0;
1635                                 c->Header.SGTotal = 0;
1636                         }
1637                         c->Header.LUN = ioc->LUN_info;
1638                         c->Header.Tag.lower = c->busaddr;
1639
1640                         c->Request = ioc->Request;
1641                         if (ioc->buf_size > 0) {
1642                                 for (i = 0; i < sg_used; i++) {
1643                                         temp64.val =
1644                                             pci_map_single(h->pdev, buff[i],
1645                                                     buff_size[i],
1646                                                     PCI_DMA_BIDIRECTIONAL);
1647                                         c->SG[i].Addr.lower =
1648                                             temp64.val32.lower;
1649                                         c->SG[i].Addr.upper =
1650                                             temp64.val32.upper;
1651                                         c->SG[i].Len = buff_size[i];
1652                                         c->SG[i].Ext = 0;       /* we are not chaining */
1653                                 }
1654                         }
1655                         c->waiting = &wait;
1656                         enqueue_cmd_and_start_io(h, c);
1657                         wait_for_completion(&wait);
1658                         /* unlock the buffers from DMA */
1659                         for (i = 0; i < sg_used; i++) {
1660                                 temp64.val32.lower = c->SG[i].Addr.lower;
1661                                 temp64.val32.upper = c->SG[i].Addr.upper;
1662                                 pci_unmap_single(h->pdev,
1663                                         (dma_addr_t) temp64.val, buff_size[i],
1664                                         PCI_DMA_BIDIRECTIONAL);
1665                         }
1666                         check_ioctl_unit_attention(h, c);
1667                         /* Copy the error information out */
1668                         ioc->error_info = *(c->err_info);
1669                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1670                                 cmd_special_free(h, c);
1671                                 status = -EFAULT;
1672                                 goto cleanup1;
1673                         }
1674                         if (ioc->Request.Type.Direction == XFER_READ) {
1675                                 /* Copy the data out of the buffer we created */
1676                                 BYTE __user *ptr = ioc->buf;
1677                                 for (i = 0; i < sg_used; i++) {
1678                                         if (copy_to_user
1679                                             (ptr, buff[i], buff_size[i])) {
1680                                                 cmd_special_free(h, c);
1681                                                 status = -EFAULT;
1682                                                 goto cleanup1;
1683                                         }
1684                                         ptr += buff_size[i];
1685                                 }
1686                         }
1687                         cmd_special_free(h, c);
1688                         status = 0;
1689                       cleanup1:
1690                         if (buff) {
1691                                 for (i = 0; i < sg_used; i++)
1692                                         kfree(buff[i]);
1693                                 kfree(buff);
1694                         }
1695                         kfree(buff_size);
1696                         kfree(ioc);
1697                         return status;
1698                 }
1699
1700         /* scsi_cmd_ioctl handles these, below, though some are not */
1701         /* very meaningful for cciss.  SG_IO is the main one people want. */
1702
1703         case SG_GET_VERSION_NUM:
1704         case SG_SET_TIMEOUT:
1705         case SG_GET_TIMEOUT:
1706         case SG_GET_RESERVED_SIZE:
1707         case SG_SET_RESERVED_SIZE:
1708         case SG_EMULATED_HOST:
1709         case SG_IO:
1710         case SCSI_IOCTL_SEND_COMMAND:
1711                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1712
1713         /* scsi_cmd_ioctl would normally handle these, below, but */
1714         /* they aren't a good fit for cciss, as CD-ROMs are */
1715         /* not supported, and we don't have any bus/target/lun */
1716         /* which we present to the kernel. */
1717
1718         case CDROM_SEND_PACKET:
1719         case CDROMCLOSETRAY:
1720         case CDROMEJECT:
1721         case SCSI_IOCTL_GET_IDLUN:
1722         case SCSI_IOCTL_GET_BUS_NUMBER:
1723         default:
1724                 return -ENOTTY;
1725         }
1726 }
1727
1728 static void cciss_check_queues(ctlr_info_t *h)
1729 {
1730         int start_queue = h->next_to_run;
1731         int i;
1732
1733         /* check to see if we have maxed out the number of commands that can
1734          * be placed on the queue.  If so then exit.  We do this check here
1735          * in case the interrupt we serviced was from an ioctl and did not
1736          * free any new commands.
1737          */
1738         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1739                 return;
1740
1741         /* We have room on the queue for more commands.  Now we need to queue
1742          * them up.  We will also keep track of the next queue to run so
1743          * that every queue gets a chance to be started first.
1744          */
1745         for (i = 0; i < h->highest_lun + 1; i++) {
1746                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1747                 /* make sure the disk has been added and the drive is real
1748                  * because this can be called from the middle of init_one.
1749                  */
1750                 if (!h->drv[curr_queue])
1751                         continue;
1752                 if (!(h->drv[curr_queue]->queue) ||
1753                         !(h->drv[curr_queue]->heads))
1754                         continue;
1755                 blk_start_queue(h->gendisk[curr_queue]->queue);
1756
1757                 /* check to see if we have maxed out the number of commands
1758                  * that can be placed on the queue.
1759                  */
1760                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1761                         if (curr_queue == start_queue) {
1762                                 h->next_to_run =
1763                                     (start_queue + 1) % (h->highest_lun + 1);
1764                                 break;
1765                         } else {
1766                                 h->next_to_run = curr_queue;
1767                                 break;
1768                         }
1769                 }
1770         }
1771 }
1772
1773 static void cciss_softirq_done(struct request *rq)
1774 {
1775         CommandList_struct *c = rq->completion_data;
1776         ctlr_info_t *h = hba[c->ctlr];
1777         SGDescriptor_struct *curr_sg = c->SG;
1778         u64bit temp64;
1779         unsigned long flags;
1780         int i, ddir;
1781         int sg_index = 0;
1782
1783         if (c->Request.Type.Direction == XFER_READ)
1784                 ddir = PCI_DMA_FROMDEVICE;
1785         else
1786                 ddir = PCI_DMA_TODEVICE;
1787
1788         /* command did not need to be retried */
1789         /* unmap the DMA mapping for all the scatter gather elements */
1790         for (i = 0; i < c->Header.SGList; i++) {
1791                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1792                         cciss_unmap_sg_chain_block(h, c);
1793                         /* Point to the next block */
1794                         curr_sg = h->cmd_sg_list[c->cmdindex];
1795                         sg_index = 0;
1796                 }
1797                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1798                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1799                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1800                                 ddir);
1801                 ++sg_index;
1802         }
1803
1804         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1805
1806         /* set the residual count for pc requests */
1807         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1808                 rq->resid_len = c->err_info->ResidualCnt;
1809
1810         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1811
1812         spin_lock_irqsave(&h->lock, flags);
1813         cmd_free(h, c);
1814         cciss_check_queues(h);
1815         spin_unlock_irqrestore(&h->lock, flags);
1816 }
1817
1818 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1819         unsigned char scsi3addr[], uint32_t log_unit)
1820 {
1821         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1822                 sizeof(h->drv[log_unit]->LunID));
1823 }
1824
1825 /* This function gets the SCSI vendor, model, and revision of a logical drive
1826  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1827  * they cannot be read.
1828  */
1829 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1830                                    char *vendor, char *model, char *rev)
1831 {
1832         int rc;
1833         InquiryData_struct *inq_buf;
1834         unsigned char scsi3addr[8];
1835
1836         *vendor = '\0';
1837         *model = '\0';
1838         *rev = '\0';
1839
1840         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1841         if (!inq_buf)
1842                 return;
1843
1844         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1845         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1846                         scsi3addr, TYPE_CMD);
1847         if (rc == IO_OK) {
1848                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1849                 vendor[VENDOR_LEN] = '\0';
1850                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1851                 model[MODEL_LEN] = '\0';
1852                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1853                 rev[REV_LEN] = '\0';
1854         }
1855
1856         kfree(inq_buf);
1857         return;
1858 }
1859
1860 /* This function gets the serial number of a logical drive via
1861  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1862  * number cannot be had, for whatever reason, 16 bytes of 0xff
1863  * are returned instead.
1864  */
1865 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1866                                 unsigned char *serial_no, int buflen)
1867 {
1868 #define PAGE_83_INQ_BYTES 64
1869         int rc;
1870         unsigned char *buf;
1871         unsigned char scsi3addr[8];
1872
1873         if (buflen > 16)
1874                 buflen = 16;
1875         memset(serial_no, 0xff, buflen);
1876         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1877         if (!buf)
1878                 return;
1879         memset(serial_no, 0, buflen);
1880         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1881         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1882                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1883         if (rc == IO_OK)
1884                 memcpy(serial_no, &buf[8], buflen);
1885         kfree(buf);
1886         return;
1887 }
1888
1889 /*
1890  * cciss_add_disk sets up the block device queue for a logical drive
1891  */
1892 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1893                                 int drv_index)
1894 {
1895         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1896         if (!disk->queue)
1897                 goto init_queue_failure;
1898         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1899         disk->major = h->major;
1900         disk->first_minor = drv_index << NWD_SHIFT;
1901         disk->fops = &cciss_fops;
1902         if (cciss_create_ld_sysfs_entry(h, drv_index))
1903                 goto cleanup_queue;
1904         disk->private_data = h->drv[drv_index];
1905         disk->driverfs_dev = &h->drv[drv_index]->dev;
1906
1907         /* Set up queue information */
1908         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1909
1910         /* This is a hardware imposed limit. */
1911         blk_queue_max_segments(disk->queue, h->maxsgentries);
1912
1913         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1914
1915         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1916
1917         disk->queue->queuedata = h;
1918
1919         blk_queue_logical_block_size(disk->queue,
1920                                      h->drv[drv_index]->block_size);
1921
1922         /* Make sure all queue data is written out before */
1923         /* setting h->drv[drv_index]->queue, as setting this */
1924         /* allows the interrupt handler to start the queue */
1925         wmb();
1926         h->drv[drv_index]->queue = disk->queue;
1927         add_disk(disk);
1928         return 0;
1929
1930 cleanup_queue:
1931         blk_cleanup_queue(disk->queue);
1932         disk->queue = NULL;
1933 init_queue_failure:
1934         return -1;
1935 }
1936
1937 /* This function will check the usage_count of the drive to be updated/added.
1938  * If the usage_count is zero and it is a heretofore unknown drive, or,
1939  * the drive's capacity, geometry, or serial number has changed,
1940  * then the drive information will be updated and the disk will be
1941  * re-registered with the kernel.  If these conditions don't hold,
1942  * then it will be left alone for the next reboot.  The exception to this
1943  * is disk 0 which will always be left registered with the kernel since it
1944  * is also the controller node.  Any changes to disk 0 will show up on
1945  * the next reboot.
1946  */
1947 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1948         int first_time, int via_ioctl)
1949 {
1950         struct gendisk *disk;
1951         InquiryData_struct *inq_buff = NULL;
1952         unsigned int block_size;
1953         sector_t total_size;
1954         unsigned long flags = 0;
1955         int ret = 0;
1956         drive_info_struct *drvinfo;
1957
1958         /* Get information about the disk and modify the driver structure */
1959         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1960         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1961         if (inq_buff == NULL || drvinfo == NULL)
1962                 goto mem_msg;
1963
1964         /* testing to see if 16-byte CDBs are already being used */
1965         if (h->cciss_read == CCISS_READ_16) {
1966                 cciss_read_capacity_16(h, drv_index,
1967                         &total_size, &block_size);
1968
1969         } else {
1970                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1971                 /* if read_capacity returns all F's this volume is >2TB */
1972                 /* in size so we switch to 16-byte CDB's for all */
1973                 /* read/write ops */
1974                 if (total_size == 0xFFFFFFFFULL) {
1975                         cciss_read_capacity_16(h, drv_index,
1976                         &total_size, &block_size);
1977                         h->cciss_read = CCISS_READ_16;
1978                         h->cciss_write = CCISS_WRITE_16;
1979                 } else {
1980                         h->cciss_read = CCISS_READ_10;
1981                         h->cciss_write = CCISS_WRITE_10;
1982                 }
1983         }
1984
1985         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1986                                inq_buff, drvinfo);
1987         drvinfo->block_size = block_size;
1988         drvinfo->nr_blocks = total_size + 1;
1989
1990         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1991                                 drvinfo->model, drvinfo->rev);
1992         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1993                         sizeof(drvinfo->serial_no));
1994         /* Save the lunid in case we deregister the disk, below. */
1995         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1996                 sizeof(drvinfo->LunID));
1997
1998         /* Is it the same disk we already know, and nothing's changed? */
1999         if (h->drv[drv_index]->raid_level != -1 &&
2000                 ((memcmp(drvinfo->serial_no,
2001                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2002                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2003                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2004                 drvinfo->heads == h->drv[drv_index]->heads &&
2005                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2006                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2007                         /* The disk is unchanged, nothing to update */
2008                         goto freeret;
2009
2010         /* If we get here it's not the same disk, or something's changed,
2011          * so we need to * deregister it, and re-register it, if it's not
2012          * in use.
2013          * If the disk already exists then deregister it before proceeding
2014          * (unless it's the first disk (for the controller node).
2015          */
2016         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2017                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2018                 spin_lock_irqsave(&h->lock, flags);
2019                 h->drv[drv_index]->busy_configuring = 1;
2020                 spin_unlock_irqrestore(&h->lock, flags);
2021
2022                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2023                  * which keeps the interrupt handler from starting
2024                  * the queue.
2025                  */
2026                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2027         }
2028
2029         /* If the disk is in use return */
2030         if (ret)
2031                 goto freeret;
2032
2033         /* Save the new information from cciss_geometry_inquiry
2034          * and serial number inquiry.  If the disk was deregistered
2035          * above, then h->drv[drv_index] will be NULL.
2036          */
2037         if (h->drv[drv_index] == NULL) {
2038                 drvinfo->device_initialized = 0;
2039                 h->drv[drv_index] = drvinfo;
2040                 drvinfo = NULL; /* so it won't be freed below. */
2041         } else {
2042                 /* special case for cxd0 */
2043                 h->drv[drv_index]->block_size = drvinfo->block_size;
2044                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2045                 h->drv[drv_index]->heads = drvinfo->heads;
2046                 h->drv[drv_index]->sectors = drvinfo->sectors;
2047                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2048                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2049                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2050                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2051                         VENDOR_LEN + 1);
2052                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2053                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2054         }
2055
2056         ++h->num_luns;
2057         disk = h->gendisk[drv_index];
2058         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2059
2060         /* If it's not disk 0 (drv_index != 0)
2061          * or if it was disk 0, but there was previously
2062          * no actual corresponding configured logical drive
2063          * (raid_leve == -1) then we want to update the
2064          * logical drive's information.
2065          */
2066         if (drv_index || first_time) {
2067                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2068                         cciss_free_gendisk(h, drv_index);
2069                         cciss_free_drive_info(h, drv_index);
2070                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2071                                 drv_index);
2072                         --h->num_luns;
2073                 }
2074         }
2075
2076 freeret:
2077         kfree(inq_buff);
2078         kfree(drvinfo);
2079         return;
2080 mem_msg:
2081         dev_err(&h->pdev->dev, "out of memory\n");
2082         goto freeret;
2083 }
2084
2085 /* This function will find the first index of the controllers drive array
2086  * that has a null drv pointer and allocate the drive info struct and
2087  * will return that index   This is where new drives will be added.
2088  * If the index to be returned is greater than the highest_lun index for
2089  * the controller then highest_lun is set * to this new index.
2090  * If there are no available indexes or if tha allocation fails, then -1
2091  * is returned.  * "controller_node" is used to know if this is a real
2092  * logical drive, or just the controller node, which determines if this
2093  * counts towards highest_lun.
2094  */
2095 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2096 {
2097         int i;
2098         drive_info_struct *drv;
2099
2100         /* Search for an empty slot for our drive info */
2101         for (i = 0; i < CISS_MAX_LUN; i++) {
2102
2103                 /* if not cxd0 case, and it's occupied, skip it. */
2104                 if (h->drv[i] && i != 0)
2105                         continue;
2106                 /*
2107                  * If it's cxd0 case, and drv is alloc'ed already, and a
2108                  * disk is configured there, skip it.
2109                  */
2110                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2111                         continue;
2112
2113                 /*
2114                  * We've found an empty slot.  Update highest_lun
2115                  * provided this isn't just the fake cxd0 controller node.
2116                  */
2117                 if (i > h->highest_lun && !controller_node)
2118                         h->highest_lun = i;
2119
2120                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2121                 if (i == 0 && h->drv[i] != NULL)
2122                         return i;
2123
2124                 /*
2125                  * Found an empty slot, not already alloc'ed.  Allocate it.
2126                  * Mark it with raid_level == -1, so we know it's new later on.
2127                  */
2128                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2129                 if (!drv)
2130                         return -1;
2131                 drv->raid_level = -1; /* so we know it's new */
2132                 h->drv[i] = drv;
2133                 return i;
2134         }
2135         return -1;
2136 }
2137
2138 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2139 {
2140         kfree(h->drv[drv_index]);
2141         h->drv[drv_index] = NULL;
2142 }
2143
2144 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2145 {
2146         put_disk(h->gendisk[drv_index]);
2147         h->gendisk[drv_index] = NULL;
2148 }
2149
2150 /* cciss_add_gendisk finds a free hba[]->drv structure
2151  * and allocates a gendisk if needed, and sets the lunid
2152  * in the drvinfo structure.   It returns the index into
2153  * the ->drv[] array, or -1 if none are free.
2154  * is_controller_node indicates whether highest_lun should
2155  * count this disk, or if it's only being added to provide
2156  * a means to talk to the controller in case no logical
2157  * drives have yet been configured.
2158  */
2159 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2160         int controller_node)
2161 {
2162         int drv_index;
2163
2164         drv_index = cciss_alloc_drive_info(h, controller_node);
2165         if (drv_index == -1)
2166                 return -1;
2167
2168         /*Check if the gendisk needs to be allocated */
2169         if (!h->gendisk[drv_index]) {
2170                 h->gendisk[drv_index] =
2171                         alloc_disk(1 << NWD_SHIFT);
2172                 if (!h->gendisk[drv_index]) {
2173                         dev_err(&h->pdev->dev,
2174                                 "could not allocate a new disk %d\n",
2175                                 drv_index);
2176                         goto err_free_drive_info;
2177                 }
2178         }
2179         memcpy(h->drv[drv_index]->LunID, lunid,
2180                 sizeof(h->drv[drv_index]->LunID));
2181         if (cciss_create_ld_sysfs_entry(h, drv_index))
2182                 goto err_free_disk;
2183         /* Don't need to mark this busy because nobody */
2184         /* else knows about this disk yet to contend */
2185         /* for access to it. */
2186         h->drv[drv_index]->busy_configuring = 0;
2187         wmb();
2188         return drv_index;
2189
2190 err_free_disk:
2191         cciss_free_gendisk(h, drv_index);
2192 err_free_drive_info:
2193         cciss_free_drive_info(h, drv_index);
2194         return -1;
2195 }
2196
2197 /* This is for the special case of a controller which
2198  * has no logical drives.  In this case, we still need
2199  * to register a disk so the controller can be accessed
2200  * by the Array Config Utility.
2201  */
2202 static void cciss_add_controller_node(ctlr_info_t *h)
2203 {
2204         struct gendisk *disk;
2205         int drv_index;
2206
2207         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2208                 return;
2209
2210         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2211         if (drv_index == -1)
2212                 goto error;
2213         h->drv[drv_index]->block_size = 512;
2214         h->drv[drv_index]->nr_blocks = 0;
2215         h->drv[drv_index]->heads = 0;
2216         h->drv[drv_index]->sectors = 0;
2217         h->drv[drv_index]->cylinders = 0;
2218         h->drv[drv_index]->raid_level = -1;
2219         memset(h->drv[drv_index]->serial_no, 0, 16);
2220         disk = h->gendisk[drv_index];
2221         if (cciss_add_disk(h, disk, drv_index) == 0)
2222                 return;
2223         cciss_free_gendisk(h, drv_index);
2224         cciss_free_drive_info(h, drv_index);
2225 error:
2226         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2227         return;
2228 }
2229
2230 /* This function will add and remove logical drives from the Logical
2231  * drive array of the controller and maintain persistency of ordering
2232  * so that mount points are preserved until the next reboot.  This allows
2233  * for the removal of logical drives in the middle of the drive array
2234  * without a re-ordering of those drives.
2235  * INPUT
2236  * h            = The controller to perform the operations on
2237  */
2238 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2239         int via_ioctl)
2240 {
2241         int num_luns;
2242         ReportLunData_struct *ld_buff = NULL;
2243         int return_code;
2244         int listlength = 0;
2245         int i;
2246         int drv_found;
2247         int drv_index = 0;
2248         unsigned char lunid[8] = CTLR_LUNID;
2249         unsigned long flags;
2250
2251         if (!capable(CAP_SYS_RAWIO))
2252                 return -EPERM;
2253
2254         /* Set busy_configuring flag for this operation */
2255         spin_lock_irqsave(&h->lock, flags);
2256         if (h->busy_configuring) {
2257                 spin_unlock_irqrestore(&h->lock, flags);
2258                 return -EBUSY;
2259         }
2260         h->busy_configuring = 1;
2261         spin_unlock_irqrestore(&h->lock, flags);
2262
2263         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2264         if (ld_buff == NULL)
2265                 goto mem_msg;
2266
2267         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2268                                       sizeof(ReportLunData_struct),
2269                                       0, CTLR_LUNID, TYPE_CMD);
2270
2271         if (return_code == IO_OK)
2272                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2273         else {  /* reading number of logical volumes failed */
2274                 dev_warn(&h->pdev->dev,
2275                         "report logical volume command failed\n");
2276                 listlength = 0;
2277                 goto freeret;
2278         }
2279
2280         num_luns = listlength / 8;      /* 8 bytes per entry */
2281         if (num_luns > CISS_MAX_LUN) {
2282                 num_luns = CISS_MAX_LUN;
2283                 dev_warn(&h->pdev->dev, "more luns configured"
2284                        " on controller than can be handled by"
2285                        " this driver.\n");
2286         }
2287
2288         if (num_luns == 0)
2289                 cciss_add_controller_node(h);
2290
2291         /* Compare controller drive array to driver's drive array
2292          * to see if any drives are missing on the controller due
2293          * to action of Array Config Utility (user deletes drive)
2294          * and deregister logical drives which have disappeared.
2295          */
2296         for (i = 0; i <= h->highest_lun; i++) {
2297                 int j;
2298                 drv_found = 0;
2299
2300                 /* skip holes in the array from already deleted drives */
2301                 if (h->drv[i] == NULL)
2302                         continue;
2303
2304                 for (j = 0; j < num_luns; j++) {
2305                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2306                         if (memcmp(h->drv[i]->LunID, lunid,
2307                                 sizeof(lunid)) == 0) {
2308                                 drv_found = 1;
2309                                 break;
2310                         }
2311                 }
2312                 if (!drv_found) {
2313                         /* Deregister it from the OS, it's gone. */
2314                         spin_lock_irqsave(&h->lock, flags);
2315                         h->drv[i]->busy_configuring = 1;
2316                         spin_unlock_irqrestore(&h->lock, flags);
2317                         return_code = deregister_disk(h, i, 1, via_ioctl);
2318                         if (h->drv[i] != NULL)
2319                                 h->drv[i]->busy_configuring = 0;
2320                 }
2321         }
2322
2323         /* Compare controller drive array to driver's drive array.
2324          * Check for updates in the drive information and any new drives
2325          * on the controller due to ACU adding logical drives, or changing
2326          * a logical drive's size, etc.  Reregister any new/changed drives
2327          */
2328         for (i = 0; i < num_luns; i++) {
2329                 int j;
2330
2331                 drv_found = 0;
2332
2333                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2334                 /* Find if the LUN is already in the drive array
2335                  * of the driver.  If so then update its info
2336                  * if not in use.  If it does not exist then find
2337                  * the first free index and add it.
2338                  */
2339                 for (j = 0; j <= h->highest_lun; j++) {
2340                         if (h->drv[j] != NULL &&
2341                                 memcmp(h->drv[j]->LunID, lunid,
2342                                         sizeof(h->drv[j]->LunID)) == 0) {
2343                                 drv_index = j;
2344                                 drv_found = 1;
2345                                 break;
2346                         }
2347                 }
2348
2349                 /* check if the drive was found already in the array */
2350                 if (!drv_found) {
2351                         drv_index = cciss_add_gendisk(h, lunid, 0);
2352                         if (drv_index == -1)
2353                                 goto freeret;
2354                 }
2355                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2356         }               /* end for */
2357
2358 freeret:
2359         kfree(ld_buff);
2360         h->busy_configuring = 0;
2361         /* We return -1 here to tell the ACU that we have registered/updated
2362          * all of the drives that we can and to keep it from calling us
2363          * additional times.
2364          */
2365         return -1;
2366 mem_msg:
2367         dev_err(&h->pdev->dev, "out of memory\n");
2368         h->busy_configuring = 0;
2369         goto freeret;
2370 }
2371
2372 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2373 {
2374         /* zero out the disk size info */
2375         drive_info->nr_blocks = 0;
2376         drive_info->block_size = 0;
2377         drive_info->heads = 0;
2378         drive_info->sectors = 0;
2379         drive_info->cylinders = 0;
2380         drive_info->raid_level = -1;
2381         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2382         memset(drive_info->model, 0, sizeof(drive_info->model));
2383         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2384         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2385         /*
2386          * don't clear the LUNID though, we need to remember which
2387          * one this one is.
2388          */
2389 }
2390
2391 /* This function will deregister the disk and it's queue from the
2392  * kernel.  It must be called with the controller lock held and the
2393  * drv structures busy_configuring flag set.  It's parameters are:
2394  *
2395  * disk = This is the disk to be deregistered
2396  * drv  = This is the drive_info_struct associated with the disk to be
2397  *        deregistered.  It contains information about the disk used
2398  *        by the driver.
2399  * clear_all = This flag determines whether or not the disk information
2400  *             is going to be completely cleared out and the highest_lun
2401  *             reset.  Sometimes we want to clear out information about
2402  *             the disk in preparation for re-adding it.  In this case
2403  *             the highest_lun should be left unchanged and the LunID
2404  *             should not be cleared.
2405  * via_ioctl
2406  *    This indicates whether we've reached this path via ioctl.
2407  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2408  *    If this path is reached via ioctl(), then the max_usage_count will
2409  *    be 1, as the process calling ioctl() has got to have the device open.
2410  *    If we get here via sysfs, then the max usage count will be zero.
2411 */
2412 static int deregister_disk(ctlr_info_t *h, int drv_index,
2413                            int clear_all, int via_ioctl)
2414 {
2415         int i;
2416         struct gendisk *disk;
2417         drive_info_struct *drv;
2418         int recalculate_highest_lun;
2419
2420         if (!capable(CAP_SYS_RAWIO))
2421                 return -EPERM;
2422
2423         drv = h->drv[drv_index];
2424         disk = h->gendisk[drv_index];
2425
2426         /* make sure logical volume is NOT is use */
2427         if (clear_all || (h->gendisk[0] == disk)) {
2428                 if (drv->usage_count > via_ioctl)
2429                         return -EBUSY;
2430         } else if (drv->usage_count > 0)
2431                 return -EBUSY;
2432
2433         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2434
2435         /* invalidate the devices and deregister the disk.  If it is disk
2436          * zero do not deregister it but just zero out it's values.  This
2437          * allows us to delete disk zero but keep the controller registered.
2438          */
2439         if (h->gendisk[0] != disk) {
2440                 struct request_queue *q = disk->queue;
2441                 if (disk->flags & GENHD_FL_UP) {
2442                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2443                         del_gendisk(disk);
2444                 }
2445                 if (q)
2446                         blk_cleanup_queue(q);
2447                 /* If clear_all is set then we are deleting the logical
2448                  * drive, not just refreshing its info.  For drives
2449                  * other than disk 0 we will call put_disk.  We do not
2450                  * do this for disk 0 as we need it to be able to
2451                  * configure the controller.
2452                  */
2453                 if (clear_all){
2454                         /* This isn't pretty, but we need to find the
2455                          * disk in our array and NULL our the pointer.
2456                          * This is so that we will call alloc_disk if
2457                          * this index is used again later.
2458                          */
2459                         for (i=0; i < CISS_MAX_LUN; i++){
2460                                 if (h->gendisk[i] == disk) {
2461                                         h->gendisk[i] = NULL;
2462                                         break;
2463                                 }
2464                         }
2465                         put_disk(disk);
2466                 }
2467         } else {
2468                 set_capacity(disk, 0);
2469                 cciss_clear_drive_info(drv);
2470         }
2471
2472         --h->num_luns;
2473
2474         /* if it was the last disk, find the new hightest lun */
2475         if (clear_all && recalculate_highest_lun) {
2476                 int newhighest = -1;
2477                 for (i = 0; i <= h->highest_lun; i++) {
2478                         /* if the disk has size > 0, it is available */
2479                         if (h->drv[i] && h->drv[i]->heads)
2480                                 newhighest = i;
2481                 }
2482                 h->highest_lun = newhighest;
2483         }
2484         return 0;
2485 }
2486
2487 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2488                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2489                 int cmd_type)
2490 {
2491         u64bit buff_dma_handle;
2492         int status = IO_OK;
2493
2494         c->cmd_type = CMD_IOCTL_PEND;
2495         c->Header.ReplyQueue = 0;
2496         if (buff != NULL) {
2497                 c->Header.SGList = 1;
2498                 c->Header.SGTotal = 1;
2499         } else {
2500                 c->Header.SGList = 0;
2501                 c->Header.SGTotal = 0;
2502         }
2503         c->Header.Tag.lower = c->busaddr;
2504         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2505
2506         c->Request.Type.Type = cmd_type;
2507         if (cmd_type == TYPE_CMD) {
2508                 switch (cmd) {
2509                 case CISS_INQUIRY:
2510                         /* are we trying to read a vital product page */
2511                         if (page_code != 0) {
2512                                 c->Request.CDB[1] = 0x01;
2513                                 c->Request.CDB[2] = page_code;
2514                         }
2515                         c->Request.CDBLen = 6;
2516                         c->Request.Type.Attribute = ATTR_SIMPLE;
2517                         c->Request.Type.Direction = XFER_READ;
2518                         c->Request.Timeout = 0;
2519                         c->Request.CDB[0] = CISS_INQUIRY;
2520                         c->Request.CDB[4] = size & 0xFF;
2521                         break;
2522                 case CISS_REPORT_LOG:
2523                 case CISS_REPORT_PHYS:
2524                         /* Talking to controller so It's a physical command
2525                            mode = 00 target = 0.  Nothing to write.
2526                          */
2527                         c->Request.CDBLen = 12;
2528                         c->Request.Type.Attribute = ATTR_SIMPLE;
2529                         c->Request.Type.Direction = XFER_READ;
2530                         c->Request.Timeout = 0;
2531                         c->Request.CDB[0] = cmd;
2532                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2533                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2534                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2535                         c->Request.CDB[9] = size & 0xFF;
2536                         break;
2537
2538                 case CCISS_READ_CAPACITY:
2539                         c->Request.CDBLen = 10;
2540                         c->Request.Type.Attribute = ATTR_SIMPLE;
2541                         c->Request.Type.Direction = XFER_READ;
2542                         c->Request.Timeout = 0;
2543                         c->Request.CDB[0] = cmd;
2544                         break;
2545                 case CCISS_READ_CAPACITY_16:
2546                         c->Request.CDBLen = 16;
2547                         c->Request.Type.Attribute = ATTR_SIMPLE;
2548                         c->Request.Type.Direction = XFER_READ;
2549                         c->Request.Timeout = 0;
2550                         c->Request.CDB[0] = cmd;
2551                         c->Request.CDB[1] = 0x10;
2552                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2553                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2554                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2555                         c->Request.CDB[13] = size & 0xFF;
2556                         c->Request.Timeout = 0;
2557                         c->Request.CDB[0] = cmd;
2558                         break;
2559                 case CCISS_CACHE_FLUSH:
2560                         c->Request.CDBLen = 12;
2561                         c->Request.Type.Attribute = ATTR_SIMPLE;
2562                         c->Request.Type.Direction = XFER_WRITE;
2563                         c->Request.Timeout = 0;
2564                         c->Request.CDB[0] = BMIC_WRITE;
2565                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2566                         break;
2567                 case TEST_UNIT_READY:
2568                         c->Request.CDBLen = 6;
2569                         c->Request.Type.Attribute = ATTR_SIMPLE;
2570                         c->Request.Type.Direction = XFER_NONE;
2571                         c->Request.Timeout = 0;
2572                         break;
2573                 default:
2574                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2575                         return IO_ERROR;
2576                 }
2577         } else if (cmd_type == TYPE_MSG) {
2578                 switch (cmd) {
2579                 case 0: /* ABORT message */
2580                         c->Request.CDBLen = 12;
2581                         c->Request.Type.Attribute = ATTR_SIMPLE;
2582                         c->Request.Type.Direction = XFER_WRITE;
2583                         c->Request.Timeout = 0;
2584                         c->Request.CDB[0] = cmd;        /* abort */
2585                         c->Request.CDB[1] = 0;  /* abort a command */
2586                         /* buff contains the tag of the command to abort */
2587                         memcpy(&c->Request.CDB[4], buff, 8);
2588                         break;
2589                 case 1: /* RESET message */
2590                         c->Request.CDBLen = 16;
2591                         c->Request.Type.Attribute = ATTR_SIMPLE;
2592                         c->Request.Type.Direction = XFER_NONE;
2593                         c->Request.Timeout = 0;
2594                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2595                         c->Request.CDB[0] = cmd;        /* reset */
2596                         c->Request.CDB[1] = 0x03;       /* reset a target */
2597                         break;
2598                 case 3: /* No-Op message */
2599                         c->Request.CDBLen = 1;
2600                         c->Request.Type.Attribute = ATTR_SIMPLE;
2601                         c->Request.Type.Direction = XFER_WRITE;
2602                         c->Request.Timeout = 0;
2603                         c->Request.CDB[0] = cmd;
2604                         break;
2605                 default:
2606                         dev_warn(&h->pdev->dev,
2607                                 "unknown message type %d\n", cmd);
2608                         return IO_ERROR;
2609                 }
2610         } else {
2611                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2612                 return IO_ERROR;
2613         }
2614         /* Fill in the scatter gather information */
2615         if (size > 0) {
2616                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2617                                                              buff, size,
2618                                                              PCI_DMA_BIDIRECTIONAL);
2619                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2620                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2621                 c->SG[0].Len = size;
2622                 c->SG[0].Ext = 0;       /* we are not chaining */
2623         }
2624         return status;
2625 }
2626
2627 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2628 {
2629         switch (c->err_info->ScsiStatus) {
2630         case SAM_STAT_GOOD:
2631                 return IO_OK;
2632         case SAM_STAT_CHECK_CONDITION:
2633                 switch (0xf & c->err_info->SenseInfo[2]) {
2634                 case 0: return IO_OK; /* no sense */
2635                 case 1: return IO_OK; /* recovered error */
2636                 default:
2637                         if (check_for_unit_attention(h, c))
2638                                 return IO_NEEDS_RETRY;
2639                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2640                                 "check condition, sense key = 0x%02x\n",
2641                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2642                 }
2643                 break;
2644         default:
2645                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2646                         "scsi status = 0x%02x\n",
2647                         c->Request.CDB[0], c->err_info->ScsiStatus);
2648                 break;
2649         }
2650         return IO_ERROR;
2651 }
2652
2653 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2654 {
2655         int return_status = IO_OK;
2656
2657         if (c->err_info->CommandStatus == CMD_SUCCESS)
2658                 return IO_OK;
2659
2660         switch (c->err_info->CommandStatus) {
2661         case CMD_TARGET_STATUS:
2662                 return_status = check_target_status(h, c);
2663                 break;
2664         case CMD_DATA_UNDERRUN:
2665         case CMD_DATA_OVERRUN:
2666                 /* expected for inquiry and report lun commands */
2667                 break;
2668         case CMD_INVALID:
2669                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2670                        "reported invalid\n", c->Request.CDB[0]);
2671                 return_status = IO_ERROR;
2672                 break;
2673         case CMD_PROTOCOL_ERR:
2674                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2675                        "protocol error\n", c->Request.CDB[0]);
2676                 return_status = IO_ERROR;
2677                 break;
2678         case CMD_HARDWARE_ERR:
2679                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2680                        " hardware error\n", c->Request.CDB[0]);
2681                 return_status = IO_ERROR;
2682                 break;
2683         case CMD_CONNECTION_LOST:
2684                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2685                        "connection lost\n", c->Request.CDB[0]);
2686                 return_status = IO_ERROR;
2687                 break;
2688         case CMD_ABORTED:
2689                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2690                        "aborted\n", c->Request.CDB[0]);
2691                 return_status = IO_ERROR;
2692                 break;
2693         case CMD_ABORT_FAILED:
2694                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2695                        "abort failed\n", c->Request.CDB[0]);
2696                 return_status = IO_ERROR;
2697                 break;
2698         case CMD_UNSOLICITED_ABORT:
2699                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2700                         c->Request.CDB[0]);
2701                 return_status = IO_NEEDS_RETRY;
2702                 break;
2703         default:
2704                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2705                        "unknown status %x\n", c->Request.CDB[0],
2706                        c->err_info->CommandStatus);
2707                 return_status = IO_ERROR;
2708         }
2709         return return_status;
2710 }
2711
2712 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2713         int attempt_retry)
2714 {
2715         DECLARE_COMPLETION_ONSTACK(wait);
2716         u64bit buff_dma_handle;
2717         int return_status = IO_OK;
2718
2719 resend_cmd2:
2720         c->waiting = &wait;
2721         enqueue_cmd_and_start_io(h, c);
2722
2723         wait_for_completion(&wait);
2724
2725         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2726                 goto command_done;
2727
2728         return_status = process_sendcmd_error(h, c);
2729
2730         if (return_status == IO_NEEDS_RETRY &&
2731                 c->retry_count < MAX_CMD_RETRIES) {
2732                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2733                         c->Request.CDB[0]);
2734                 c->retry_count++;
2735                 /* erase the old error information */
2736                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2737                 return_status = IO_OK;
2738                 INIT_COMPLETION(wait);
2739                 goto resend_cmd2;
2740         }
2741
2742 command_done:
2743         /* unlock the buffers from DMA */
2744         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2745         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2746         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2747                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2748         return return_status;
2749 }
2750
2751 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2752                            __u8 page_code, unsigned char scsi3addr[],
2753                         int cmd_type)
2754 {
2755         CommandList_struct *c;
2756         int return_status;
2757
2758         c = cmd_special_alloc(h);
2759         if (!c)
2760                 return -ENOMEM;
2761         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2762                 scsi3addr, cmd_type);
2763         if (return_status == IO_OK)
2764                 return_status = sendcmd_withirq_core(h, c, 1);
2765
2766         cmd_special_free(h, c);
2767         return return_status;
2768 }
2769
2770 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2771                                    sector_t total_size,
2772                                    unsigned int block_size,
2773                                    InquiryData_struct *inq_buff,
2774                                    drive_info_struct *drv)
2775 {
2776         int return_code;
2777         unsigned long t;
2778         unsigned char scsi3addr[8];
2779
2780         memset(inq_buff, 0, sizeof(InquiryData_struct));
2781         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2782         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2783                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2784         if (return_code == IO_OK) {
2785                 if (inq_buff->data_byte[8] == 0xFF) {
2786                         dev_warn(&h->pdev->dev,
2787                                "reading geometry failed, volume "
2788                                "does not support reading geometry\n");
2789                         drv->heads = 255;
2790                         drv->sectors = 32;      /* Sectors per track */
2791                         drv->cylinders = total_size + 1;
2792                         drv->raid_level = RAID_UNKNOWN;
2793                 } else {
2794                         drv->heads = inq_buff->data_byte[6];
2795                         drv->sectors = inq_buff->data_byte[7];
2796                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2797                         drv->cylinders += inq_buff->data_byte[5];
2798                         drv->raid_level = inq_buff->data_byte[8];
2799                 }
2800                 drv->block_size = block_size;
2801                 drv->nr_blocks = total_size + 1;
2802                 t = drv->heads * drv->sectors;
2803                 if (t > 1) {
2804                         sector_t real_size = total_size + 1;
2805                         unsigned long rem = sector_div(real_size, t);
2806                         if (rem)
2807                                 real_size++;
2808                         drv->cylinders = real_size;
2809                 }
2810         } else {                /* Get geometry failed */
2811                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2812         }
2813 }
2814
2815 static void
2816 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2817                     unsigned int *block_size)
2818 {
2819         ReadCapdata_struct *buf;
2820         int return_code;
2821         unsigned char scsi3addr[8];
2822
2823         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2824         if (!buf) {
2825                 dev_warn(&h->pdev->dev, "out of memory\n");
2826                 return;
2827         }
2828
2829         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2830         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2831                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2832         if (return_code == IO_OK) {
2833                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2834                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2835         } else {                /* read capacity command failed */
2836                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2837                 *total_size = 0;
2838                 *block_size = BLOCK_SIZE;
2839         }
2840         kfree(buf);
2841 }
2842
2843 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2844         sector_t *total_size, unsigned int *block_size)
2845 {
2846         ReadCapdata_struct_16 *buf;
2847         int return_code;
2848         unsigned char scsi3addr[8];
2849
2850         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2851         if (!buf) {
2852                 dev_warn(&h->pdev->dev, "out of memory\n");
2853                 return;
2854         }
2855
2856         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2857         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2858                 buf, sizeof(ReadCapdata_struct_16),
2859                         0, scsi3addr, TYPE_CMD);
2860         if (return_code == IO_OK) {
2861                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2862                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2863         } else {                /* read capacity command failed */
2864                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2865                 *total_size = 0;
2866                 *block_size = BLOCK_SIZE;
2867         }
2868         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2869                (unsigned long long)*total_size+1, *block_size);
2870         kfree(buf);
2871 }
2872
2873 static int cciss_revalidate(struct gendisk *disk)
2874 {
2875         ctlr_info_t *h = get_host(disk);
2876         drive_info_struct *drv = get_drv(disk);
2877         int logvol;
2878         int FOUND = 0;
2879         unsigned int block_size;
2880         sector_t total_size;
2881         InquiryData_struct *inq_buff = NULL;
2882
2883         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2884                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2885                         sizeof(drv->LunID)) == 0) {
2886                         FOUND = 1;
2887                         break;
2888                 }
2889         }
2890
2891         if (!FOUND)
2892                 return 1;
2893
2894         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2895         if (inq_buff == NULL) {
2896                 dev_warn(&h->pdev->dev, "out of memory\n");
2897                 return 1;
2898         }
2899         if (h->cciss_read == CCISS_READ_10) {
2900                 cciss_read_capacity(h, logvol,
2901                                         &total_size, &block_size);
2902         } else {
2903                 cciss_read_capacity_16(h, logvol,
2904                                         &total_size, &block_size);
2905         }
2906         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2907                                inq_buff, drv);
2908
2909         blk_queue_logical_block_size(drv->queue, drv->block_size);
2910         set_capacity(disk, drv->nr_blocks);
2911
2912         kfree(inq_buff);
2913         return 0;
2914 }
2915
2916 /*
2917  * Map (physical) PCI mem into (virtual) kernel space
2918  */
2919 static void __iomem *remap_pci_mem(ulong base, ulong size)
2920 {
2921         ulong page_base = ((ulong) base) & PAGE_MASK;
2922         ulong page_offs = ((ulong) base) - page_base;
2923         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2924
2925         return page_remapped ? (page_remapped + page_offs) : NULL;
2926 }
2927
2928 /*
2929  * Takes jobs of the Q and sends them to the hardware, then puts it on
2930  * the Q to wait for completion.
2931  */
2932 static void start_io(ctlr_info_t *h)
2933 {
2934         CommandList_struct *c;
2935
2936         while (!list_empty(&h->reqQ)) {
2937                 c = list_entry(h->reqQ.next, CommandList_struct, list);
2938                 /* can't do anything if fifo is full */
2939                 if ((h->access.fifo_full(h))) {
2940                         dev_warn(&h->pdev->dev, "fifo full\n");
2941                         break;
2942                 }
2943
2944                 /* Get the first entry from the Request Q */
2945                 removeQ(c);
2946                 h->Qdepth--;
2947
2948                 /* Tell the controller execute command */
2949                 h->access.submit_command(h, c);
2950
2951                 /* Put job onto the completed Q */
2952                 addQ(&h->cmpQ, c);
2953         }
2954 }
2955
2956 /* Assumes that h->lock is held. */
2957 /* Zeros out the error record and then resends the command back */
2958 /* to the controller */
2959 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2960 {
2961         /* erase the old error information */
2962         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2963
2964         /* add it to software queue and then send it to the controller */
2965         addQ(&h->reqQ, c);
2966         h->Qdepth++;
2967         if (h->Qdepth > h->maxQsinceinit)
2968                 h->maxQsinceinit = h->Qdepth;
2969
2970         start_io(h);
2971 }
2972
2973 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2974         unsigned int msg_byte, unsigned int host_byte,
2975         unsigned int driver_byte)
2976 {
2977         /* inverse of macros in scsi.h */
2978         return (scsi_status_byte & 0xff) |
2979                 ((msg_byte & 0xff) << 8) |
2980                 ((host_byte & 0xff) << 16) |
2981                 ((driver_byte & 0xff) << 24);
2982 }
2983
2984 static inline int evaluate_target_status(ctlr_info_t *h,
2985                         CommandList_struct *cmd, int *retry_cmd)
2986 {
2987         unsigned char sense_key;
2988         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2989         int error_value;
2990
2991         *retry_cmd = 0;
2992         /* If we get in here, it means we got "target status", that is, scsi status */
2993         status_byte = cmd->err_info->ScsiStatus;
2994         driver_byte = DRIVER_OK;
2995         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2996
2997         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2998                 host_byte = DID_PASSTHROUGH;
2999         else
3000                 host_byte = DID_OK;
3001
3002         error_value = make_status_bytes(status_byte, msg_byte,
3003                 host_byte, driver_byte);
3004
3005         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3006                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3007                         dev_warn(&h->pdev->dev, "cmd %p "
3008                                "has SCSI Status 0x%x\n",
3009                                cmd, cmd->err_info->ScsiStatus);
3010                 return error_value;
3011         }
3012
3013         /* check the sense key */
3014         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3015         /* no status or recovered error */
3016         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3017             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3018                 error_value = 0;
3019
3020         if (check_for_unit_attention(h, cmd)) {
3021                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3022                 return 0;
3023         }
3024
3025         /* Not SG_IO or similar? */
3026         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3027                 if (error_value != 0)
3028                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3029                                " sense key = 0x%x\n", cmd, sense_key);
3030                 return error_value;
3031         }
3032
3033         /* SG_IO or similar, copy sense data back */
3034         if (cmd->rq->sense) {
3035                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3036                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3037                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3038                         cmd->rq->sense_len);
3039         } else
3040                 cmd->rq->sense_len = 0;
3041
3042         return error_value;
3043 }
3044
3045 /* checks the status of the job and calls complete buffers to mark all
3046  * buffers for the completed job. Note that this function does not need
3047  * to hold the hba/queue lock.
3048  */
3049 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3050                                     int timeout)
3051 {
3052         int retry_cmd = 0;
3053         struct request *rq = cmd->rq;
3054
3055         rq->errors = 0;
3056
3057         if (timeout)
3058                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3059
3060         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3061                 goto after_error_processing;
3062
3063         switch (cmd->err_info->CommandStatus) {
3064         case CMD_TARGET_STATUS:
3065                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3066                 break;
3067         case CMD_DATA_UNDERRUN:
3068                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3069                         dev_warn(&h->pdev->dev, "cmd %p has"
3070                                " completed with data underrun "
3071                                "reported\n", cmd);
3072                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3073                 }
3074                 break;
3075         case CMD_DATA_OVERRUN:
3076                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3077                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3078                                " completed with data overrun "
3079                                "reported\n", cmd);
3080                 break;
3081         case CMD_INVALID:
3082                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3083                        "reported invalid\n", cmd);
3084                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3085                         cmd->err_info->CommandStatus, DRIVER_OK,
3086                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3087                                 DID_PASSTHROUGH : DID_ERROR);
3088                 break;
3089         case CMD_PROTOCOL_ERR:
3090                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3091                        "protocol error\n", cmd);
3092                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3093                         cmd->err_info->CommandStatus, DRIVER_OK,
3094                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3095                                 DID_PASSTHROUGH : DID_ERROR);
3096                 break;
3097         case CMD_HARDWARE_ERR:
3098                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3099                        " hardware error\n", cmd);
3100                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3101                         cmd->err_info->CommandStatus, DRIVER_OK,
3102                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3103                                 DID_PASSTHROUGH : DID_ERROR);
3104                 break;
3105         case CMD_CONNECTION_LOST:
3106                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3107                        "connection lost\n", cmd);
3108                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3109                         cmd->err_info->CommandStatus, DRIVER_OK,
3110                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3111                                 DID_PASSTHROUGH : DID_ERROR);
3112                 break;
3113         case CMD_ABORTED:
3114                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3115                        "aborted\n", cmd);
3116                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3117                         cmd->err_info->CommandStatus, DRIVER_OK,
3118                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3119                                 DID_PASSTHROUGH : DID_ABORT);
3120                 break;
3121         case CMD_ABORT_FAILED:
3122                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3123                        "abort failed\n", cmd);
3124                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3125                         cmd->err_info->CommandStatus, DRIVER_OK,
3126                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3127                                 DID_PASSTHROUGH : DID_ERROR);
3128                 break;
3129         case CMD_UNSOLICITED_ABORT:
3130                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3131                        "abort %p\n", h->ctlr, cmd);
3132                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3133                         retry_cmd = 1;
3134                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3135                         cmd->retry_count++;
3136                 } else
3137                         dev_warn(&h->pdev->dev,
3138                                 "%p retried too many times\n", cmd);
3139                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3140                         cmd->err_info->CommandStatus, DRIVER_OK,
3141                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3142                                 DID_PASSTHROUGH : DID_ABORT);
3143                 break;
3144         case CMD_TIMEOUT:
3145                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3146                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3147                         cmd->err_info->CommandStatus, DRIVER_OK,
3148                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3149                                 DID_PASSTHROUGH : DID_ERROR);
3150                 break;
3151         default:
3152                 dev_warn(&h->pdev->dev, "cmd %p returned "
3153                        "unknown status %x\n", cmd,
3154                        cmd->err_info->CommandStatus);
3155                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3156                         cmd->err_info->CommandStatus, DRIVER_OK,
3157                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3158                                 DID_PASSTHROUGH : DID_ERROR);
3159         }
3160
3161 after_error_processing:
3162
3163         /* We need to return this command */
3164         if (retry_cmd) {
3165                 resend_cciss_cmd(h, cmd);
3166                 return;
3167         }
3168         cmd->rq->completion_data = cmd;
3169         blk_complete_request(cmd->rq);
3170 }
3171
3172 static inline u32 cciss_tag_contains_index(u32 tag)
3173 {
3174 #define DIRECT_LOOKUP_BIT 0x10
3175         return tag & DIRECT_LOOKUP_BIT;
3176 }
3177
3178 static inline u32 cciss_tag_to_index(u32 tag)
3179 {
3180 #define DIRECT_LOOKUP_SHIFT 5
3181         return tag >> DIRECT_LOOKUP_SHIFT;
3182 }
3183
3184 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3185 {
3186 #define CCISS_ERROR_BITS 0x03
3187         return tag & ~CCISS_ERROR_BITS;
3188 }
3189
3190 static inline void cciss_mark_tag_indexed(u32 *tag)
3191 {
3192         *tag |= DIRECT_LOOKUP_BIT;
3193 }
3194
3195 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3196 {
3197         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3198 }
3199
3200 /*
3201  * Get a request and submit it to the controller.
3202  */
3203 static void do_cciss_request(struct request_queue *q)
3204 {
3205         ctlr_info_t *h = q->queuedata;
3206         CommandList_struct *c;
3207         sector_t start_blk;
3208         int seg;
3209         struct request *creq;
3210         u64bit temp64;
3211         struct scatterlist *tmp_sg;
3212         SGDescriptor_struct *curr_sg;
3213         drive_info_struct *drv;
3214         int i, dir;
3215         int sg_index = 0;
3216         int chained = 0;
3217
3218         /* We call start_io here in case there is a command waiting on the
3219          * queue that has not been sent.
3220          */
3221         if (blk_queue_plugged(q))
3222                 goto startio;
3223
3224       queue:
3225         creq = blk_peek_request(q);
3226         if (!creq)
3227                 goto startio;
3228
3229         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3230
3231         c = cmd_alloc(h);
3232         if (!c)
3233                 goto full;
3234
3235         blk_start_request(creq);
3236
3237         tmp_sg = h->scatter_list[c->cmdindex];
3238         spin_unlock_irq(q->queue_lock);
3239
3240         c->cmd_type = CMD_RWREQ;
3241         c->rq = creq;
3242
3243         /* fill in the request */
3244         drv = creq->rq_disk->private_data;
3245         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3246         /* got command from pool, so use the command block index instead */
3247         /* for direct lookups. */
3248         /* The first 2 bits are reserved for controller error reporting. */
3249         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3250         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3251         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3252         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3253         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3254         c->Request.Type.Attribute = ATTR_SIMPLE;
3255         c->Request.Type.Direction =
3256             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3257         c->Request.Timeout = 0; /* Don't time out */
3258         c->Request.CDB[0] =
3259             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3260         start_blk = blk_rq_pos(creq);
3261         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3262                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3263         sg_init_table(tmp_sg, h->maxsgentries);
3264         seg = blk_rq_map_sg(q, creq, tmp_sg);
3265
3266         /* get the DMA records for the setup */
3267         if (c->Request.Type.Direction == XFER_READ)
3268                 dir = PCI_DMA_FROMDEVICE;
3269         else
3270                 dir = PCI_DMA_TODEVICE;
3271
3272         curr_sg = c->SG;
3273         sg_index = 0;
3274         chained = 0;
3275
3276         for (i = 0; i < seg; i++) {
3277                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3278                         !chained && ((seg - i) > 1)) {
3279                         /* Point to next chain block. */
3280                         curr_sg = h->cmd_sg_list[c->cmdindex];
3281                         sg_index = 0;
3282                         chained = 1;
3283                 }
3284                 curr_sg[sg_index].Len = tmp_sg[i].length;
3285                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3286                                                 tmp_sg[i].offset,
3287                                                 tmp_sg[i].length, dir);
3288                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3289                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3290                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3291                 ++sg_index;
3292         }
3293         if (chained)
3294                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3295                         (seg - (h->max_cmd_sgentries - 1)) *
3296                                 sizeof(SGDescriptor_struct));
3297
3298         /* track how many SG entries we are using */
3299         if (seg > h->maxSG)
3300                 h->maxSG = seg;
3301
3302         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3303                         "chained[%d]\n",
3304                         blk_rq_sectors(creq), seg, chained);
3305
3306         c->Header.SGTotal = seg + chained;
3307         if (seg <= h->max_cmd_sgentries)
3308                 c->Header.SGList = c->Header.SGTotal;
3309         else
3310                 c->Header.SGList = h->max_cmd_sgentries;
3311         set_performant_mode(h, c);
3312
3313         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3314                 if(h->cciss_read == CCISS_READ_10) {
3315                         c->Request.CDB[1] = 0;
3316                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3317                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3318                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3319                         c->Request.CDB[5] = start_blk & 0xff;
3320                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3321                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3322                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3323                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3324                 } else {
3325                         u32 upper32 = upper_32_bits(start_blk);
3326
3327                         c->Request.CDBLen = 16;
3328                         c->Request.CDB[1]= 0;
3329                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3330                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3331                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3332                         c->Request.CDB[5]= upper32 & 0xff;
3333                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3334                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3335                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3336                         c->Request.CDB[9]= start_blk & 0xff;
3337                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3338                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3339                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3340                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3341                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3342                 }
3343         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3344                 c->Request.CDBLen = creq->cmd_len;
3345                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3346         } else {
3347                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3348                         creq->cmd_type);
3349                 BUG();
3350         }
3351
3352         spin_lock_irq(q->queue_lock);
3353
3354         addQ(&h->reqQ, c);
3355         h->Qdepth++;
3356         if (h->Qdepth > h->maxQsinceinit)
3357                 h->maxQsinceinit = h->Qdepth;
3358
3359         goto queue;
3360 full:
3361         blk_stop_queue(q);
3362 startio:
3363         /* We will already have the driver lock here so not need
3364          * to lock it.
3365          */
3366         start_io(h);
3367 }
3368
3369 static inline unsigned long get_next_completion(ctlr_info_t *h)
3370 {
3371         return h->access.command_completed(h);
3372 }
3373
3374 static inline int interrupt_pending(ctlr_info_t *h)
3375 {
3376         return h->access.intr_pending(h);
3377 }
3378
3379 static inline long interrupt_not_for_us(ctlr_info_t *h)
3380 {
3381         return ((h->access.intr_pending(h) == 0) ||
3382                 (h->interrupts_enabled == 0));
3383 }
3384
3385 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3386                         u32 raw_tag)
3387 {
3388         if (unlikely(tag_index >= h->nr_cmds)) {
3389                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3390                 return 1;
3391         }
3392         return 0;
3393 }
3394
3395 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3396                                 u32 raw_tag)
3397 {
3398         removeQ(c);
3399         if (likely(c->cmd_type == CMD_RWREQ))
3400                 complete_command(h, c, 0);
3401         else if (c->cmd_type == CMD_IOCTL_PEND)
3402                 complete(c->waiting);
3403 #ifdef CONFIG_CISS_SCSI_TAPE
3404         else if (c->cmd_type == CMD_SCSI)
3405                 complete_scsi_command(c, 0, raw_tag);
3406 #endif
3407 }
3408
3409 static inline u32 next_command(ctlr_info_t *h)
3410 {
3411         u32 a;
3412
3413         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3414                 return h->access.command_completed(h);
3415
3416         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3417                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3418                 (h->reply_pool_head)++;
3419                 h->commands_outstanding--;
3420         } else {
3421                 a = FIFO_EMPTY;
3422         }
3423         /* Check for wraparound */
3424         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3425                 h->reply_pool_head = h->reply_pool;
3426                 h->reply_pool_wraparound ^= 1;
3427         }
3428         return a;
3429 }
3430
3431 /* process completion of an indexed ("direct lookup") command */
3432 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3433 {
3434         u32 tag_index;
3435         CommandList_struct *c;
3436
3437         tag_index = cciss_tag_to_index(raw_tag);
3438         if (bad_tag(h, tag_index, raw_tag))
3439                 return next_command(h);
3440         c = h->cmd_pool + tag_index;
3441         finish_cmd(h, c, raw_tag);
3442         return next_command(h);
3443 }
3444
3445 /* process completion of a non-indexed command */
3446 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3447 {
3448         u32 tag;
3449         CommandList_struct *c = NULL;
3450         __u32 busaddr_masked, tag_masked;
3451
3452         tag = cciss_tag_discard_error_bits(raw_tag);
3453         list_for_each_entry(c, &h->cmpQ, list) {
3454                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3455                 tag_masked = cciss_tag_discard_error_bits(tag);
3456                 if (busaddr_masked == tag_masked) {
3457                         finish_cmd(h, c, raw_tag);
3458                         return next_command(h);
3459                 }
3460         }
3461         bad_tag(h, h->nr_cmds + 1, raw_tag);
3462         return next_command(h);
3463 }
3464
3465 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3466 {
3467         ctlr_info_t *h = dev_id;
3468         unsigned long flags;
3469         u32 raw_tag;
3470
3471         if (interrupt_not_for_us(h))
3472                 return IRQ_NONE;
3473         spin_lock_irqsave(&h->lock, flags);
3474         while (interrupt_pending(h)) {
3475                 raw_tag = get_next_completion(h);
3476                 while (raw_tag != FIFO_EMPTY) {
3477                         if (cciss_tag_contains_index(raw_tag))
3478                                 raw_tag = process_indexed_cmd(h, raw_tag);
3479                         else
3480                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3481                 }
3482         }
3483         spin_unlock_irqrestore(&h->lock, flags);
3484         return IRQ_HANDLED;
3485 }
3486
3487 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3488  * check the interrupt pending register because it is not set.
3489  */
3490 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3491 {
3492         ctlr_info_t *h = dev_id;
3493         unsigned long flags;
3494         u32 raw_tag;
3495
3496         spin_lock_irqsave(&h->lock, flags);
3497         raw_tag = get_next_completion(h);
3498         while (raw_tag != FIFO_EMPTY) {
3499                 if (cciss_tag_contains_index(raw_tag))
3500                         raw_tag = process_indexed_cmd(h, raw_tag);
3501                 else
3502                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3503         }
3504         spin_unlock_irqrestore(&h->lock, flags);
3505         return IRQ_HANDLED;
3506 }
3507
3508 /**
3509  * add_to_scan_list() - add controller to rescan queue
3510  * @h:                Pointer to the controller.
3511  *
3512  * Adds the controller to the rescan queue if not already on the queue.
3513  *
3514  * returns 1 if added to the queue, 0 if skipped (could be on the
3515  * queue already, or the controller could be initializing or shutting
3516  * down).
3517  **/
3518 static int add_to_scan_list(struct ctlr_info *h)
3519 {
3520         struct ctlr_info *test_h;
3521         int found = 0;
3522         int ret = 0;
3523
3524         if (h->busy_initializing)
3525                 return 0;
3526
3527         if (!mutex_trylock(&h->busy_shutting_down))
3528                 return 0;
3529
3530         mutex_lock(&scan_mutex);
3531         list_for_each_entry(test_h, &scan_q, scan_list) {
3532                 if (test_h == h) {
3533                         found = 1;
3534                         break;
3535                 }
3536         }
3537         if (!found && !h->busy_scanning) {
3538                 INIT_COMPLETION(h->scan_wait);
3539                 list_add_tail(&h->scan_list, &scan_q);
3540                 ret = 1;
3541         }
3542         mutex_unlock(&scan_mutex);
3543         mutex_unlock(&h->busy_shutting_down);
3544
3545         return ret;
3546 }
3547
3548 /**
3549  * remove_from_scan_list() - remove controller from rescan queue
3550  * @h:                     Pointer to the controller.
3551  *
3552  * Removes the controller from the rescan queue if present. Blocks if
3553  * the controller is currently conducting a rescan.  The controller
3554  * can be in one of three states:
3555  * 1. Doesn't need a scan
3556  * 2. On the scan list, but not scanning yet (we remove it)
3557  * 3. Busy scanning (and not on the list). In this case we want to wait for
3558  *    the scan to complete to make sure the scanning thread for this
3559  *    controller is completely idle.
3560  **/
3561 static void remove_from_scan_list(struct ctlr_info *h)
3562 {
3563         struct ctlr_info *test_h, *tmp_h;
3564
3565         mutex_lock(&scan_mutex);
3566         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3567                 if (test_h == h) { /* state 2. */
3568                         list_del(&h->scan_list);
3569                         complete_all(&h->scan_wait);
3570                         mutex_unlock(&scan_mutex);
3571                         return;
3572                 }
3573         }
3574         if (h->busy_scanning) { /* state 3. */
3575                 mutex_unlock(&scan_mutex);
3576                 wait_for_completion(&h->scan_wait);
3577         } else { /* state 1, nothing to do. */
3578                 mutex_unlock(&scan_mutex);
3579         }
3580 }
3581
3582 /**
3583  * scan_thread() - kernel thread used to rescan controllers
3584  * @data:        Ignored.
3585  *
3586  * A kernel thread used scan for drive topology changes on
3587  * controllers. The thread processes only one controller at a time
3588  * using a queue.  Controllers are added to the queue using
3589  * add_to_scan_list() and removed from the queue either after done
3590  * processing or using remove_from_scan_list().
3591  *
3592  * returns 0.
3593  **/
3594 static int scan_thread(void *data)
3595 {
3596         struct ctlr_info *h;
3597
3598         while (1) {
3599                 set_current_state(TASK_INTERRUPTIBLE);
3600                 schedule();
3601                 if (kthread_should_stop())
3602                         break;
3603
3604                 while (1) {
3605                         mutex_lock(&scan_mutex);
3606                         if (list_empty(&scan_q)) {
3607                                 mutex_unlock(&scan_mutex);
3608                                 break;
3609                         }
3610
3611                         h = list_entry(scan_q.next,
3612                                        struct ctlr_info,
3613                                        scan_list);
3614                         list_del(&h->scan_list);
3615                         h->busy_scanning = 1;
3616                         mutex_unlock(&scan_mutex);
3617
3618                         rebuild_lun_table(h, 0, 0);
3619                         complete_all(&h->scan_wait);
3620                         mutex_lock(&scan_mutex);
3621                         h->busy_scanning = 0;
3622                         mutex_unlock(&scan_mutex);
3623                 }
3624         }
3625
3626         return 0;
3627 }
3628
3629 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3630 {
3631         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3632                 return 0;
3633
3634         switch (c->err_info->SenseInfo[12]) {
3635         case STATE_CHANGED:
3636                 dev_warn(&h->pdev->dev, "a state change "
3637                         "detected, command retried\n");
3638                 return 1;
3639         break;
3640         case LUN_FAILED:
3641                 dev_warn(&h->pdev->dev, "LUN failure "
3642                         "detected, action required\n");
3643                 return 1;
3644         break;
3645         case REPORT_LUNS_CHANGED:
3646                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3647         /*
3648          * Here, we could call add_to_scan_list and wake up the scan thread,
3649          * except that it's quite likely that we will get more than one
3650          * REPORT_LUNS_CHANGED condition in quick succession, which means
3651          * that those which occur after the first one will likely happen
3652          * *during* the scan_thread's rescan.  And the rescan code is not
3653          * robust enough to restart in the middle, undoing what it has already
3654          * done, and it's not clear that it's even possible to do this, since
3655          * part of what it does is notify the block layer, which starts
3656          * doing it's own i/o to read partition tables and so on, and the
3657          * driver doesn't have visibility to know what might need undoing.
3658          * In any event, if possible, it is horribly complicated to get right
3659          * so we just don't do it for now.
3660          *
3661          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3662          */
3663                 return 1;
3664         break;
3665         case POWER_OR_RESET:
3666                 dev_warn(&h->pdev->dev,
3667                         "a power on or device reset detected\n");
3668                 return 1;
3669         break;
3670         case UNIT_ATTENTION_CLEARED:
3671                 dev_warn(&h->pdev->dev,
3672                         "unit attention cleared by another initiator\n");
3673                 return 1;
3674         break;
3675         default:
3676                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3677                 return 1;
3678         }
3679 }
3680
3681 /*
3682  *  We cannot read the structure directly, for portability we must use
3683  *   the io functions.
3684  *   This is for debug only.
3685  */
3686 static void print_cfg_table(ctlr_info_t *h)
3687 {
3688         int i;
3689         char temp_name[17];
3690         CfgTable_struct *tb = h->cfgtable;
3691
3692         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3693         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3694         for (i = 0; i < 4; i++)
3695                 temp_name[i] = readb(&(tb->Signature[i]));
3696         temp_name[4] = '\0';
3697         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3698         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3699                 readl(&(tb->SpecValence)));
3700         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3701                readl(&(tb->TransportSupport)));
3702         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3703                readl(&(tb->TransportActive)));
3704         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3705                readl(&(tb->HostWrite.TransportRequest)));
3706         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3707                readl(&(tb->HostWrite.CoalIntDelay)));
3708         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3709                readl(&(tb->HostWrite.CoalIntCount)));
3710         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3711                readl(&(tb->CmdsOutMax)));
3712         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3713                 readl(&(tb->BusTypes)));
3714         for (i = 0; i < 16; i++)
3715                 temp_name[i] = readb(&(tb->ServerName[i]));
3716         temp_name[16] = '\0';
3717         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3718         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3719                 readl(&(tb->HeartBeat)));
3720 }
3721
3722 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3723 {
3724         int i, offset, mem_type, bar_type;
3725         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3726                 return 0;
3727         offset = 0;
3728         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3729                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3730                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3731                         offset += 4;
3732                 else {
3733                         mem_type = pci_resource_flags(pdev, i) &
3734                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3735                         switch (mem_type) {
3736                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3737                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3738                                 offset += 4;    /* 32 bit */
3739                                 break;
3740                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3741                                 offset += 8;
3742                                 break;
3743                         default:        /* reserved in PCI 2.2 */
3744                                 dev_warn(&pdev->dev,
3745                                        "Base address is invalid\n");
3746                                 return -1;
3747                                 break;
3748                         }
3749                 }
3750                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3751                         return i + 1;
3752         }
3753         return -1;
3754 }
3755
3756 /* Fill in bucket_map[], given nsgs (the max number of
3757  * scatter gather elements supported) and bucket[],
3758  * which is an array of 8 integers.  The bucket[] array
3759  * contains 8 different DMA transfer sizes (in 16
3760  * byte increments) which the controller uses to fetch
3761  * commands.  This function fills in bucket_map[], which
3762  * maps a given number of scatter gather elements to one of
3763  * the 8 DMA transfer sizes.  The point of it is to allow the
3764  * controller to only do as much DMA as needed to fetch the
3765  * command, with the DMA transfer size encoded in the lower
3766  * bits of the command address.
3767  */
3768 static void  calc_bucket_map(int bucket[], int num_buckets,
3769         int nsgs, int *bucket_map)
3770 {
3771         int i, j, b, size;
3772
3773         /* even a command with 0 SGs requires 4 blocks */
3774 #define MINIMUM_TRANSFER_BLOCKS 4
3775 #define NUM_BUCKETS 8
3776         /* Note, bucket_map must have nsgs+1 entries. */
3777         for (i = 0; i <= nsgs; i++) {
3778                 /* Compute size of a command with i SG entries */
3779                 size = i + MINIMUM_TRANSFER_BLOCKS;
3780                 b = num_buckets; /* Assume the biggest bucket */
3781                 /* Find the bucket that is just big enough */
3782                 for (j = 0; j < 8; j++) {
3783                         if (bucket[j] >= size) {
3784                                 b = j;
3785                                 break;
3786                         }
3787                 }
3788                 /* for a command with i SG entries, use bucket b. */
3789                 bucket_map[i] = b;
3790         }
3791 }
3792
3793 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3794 {
3795         int i;
3796
3797         /* under certain very rare conditions, this can take awhile.
3798          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3799          * as we enter this code.) */
3800         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3801                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3802                         break;
3803                 msleep(10);
3804         }
3805 }
3806
3807 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3808 {
3809         /* This is a bit complicated.  There are 8 registers on
3810          * the controller which we write to to tell it 8 different
3811          * sizes of commands which there may be.  It's a way of
3812          * reducing the DMA done to fetch each command.  Encoded into
3813          * each command's tag are 3 bits which communicate to the controller
3814          * which of the eight sizes that command fits within.  The size of
3815          * each command depends on how many scatter gather entries there are.
3816          * Each SG entry requires 16 bytes.  The eight registers are programmed
3817          * with the number of 16-byte blocks a command of that size requires.
3818          * The smallest command possible requires 5 such 16 byte blocks.
3819          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3820          * blocks.  Note, this only extends to the SG entries contained
3821          * within the command block, and does not extend to chained blocks
3822          * of SG elements.   bft[] contains the eight values we write to
3823          * the registers.  They are not evenly distributed, but have more
3824          * sizes for small commands, and fewer sizes for larger commands.
3825          */
3826         __u32 trans_offset;
3827         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3828                         /*
3829                          *  5 = 1 s/g entry or 4k
3830                          *  6 = 2 s/g entry or 8k
3831                          *  8 = 4 s/g entry or 16k
3832                          * 10 = 6 s/g entry or 24k
3833                          */
3834         unsigned long register_value;
3835         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3836
3837         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3838
3839         /* Controller spec: zero out this buffer. */
3840         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3841         h->reply_pool_head = h->reply_pool;
3842
3843         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3844         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3845                                 h->blockFetchTable);
3846         writel(bft[0], &h->transtable->BlockFetch0);
3847         writel(bft[1], &h->transtable->BlockFetch1);
3848         writel(bft[2], &h->transtable->BlockFetch2);
3849         writel(bft[3], &h->transtable->BlockFetch3);
3850         writel(bft[4], &h->transtable->BlockFetch4);
3851         writel(bft[5], &h->transtable->BlockFetch5);
3852         writel(bft[6], &h->transtable->BlockFetch6);
3853         writel(bft[7], &h->transtable->BlockFetch7);
3854
3855         /* size of controller ring buffer */
3856         writel(h->max_commands, &h->transtable->RepQSize);
3857         writel(1, &h->transtable->RepQCount);
3858         writel(0, &h->transtable->RepQCtrAddrLow32);
3859         writel(0, &h->transtable->RepQCtrAddrHigh32);
3860         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3861         writel(0, &h->transtable->RepQAddr0High32);
3862         writel(CFGTBL_Trans_Performant,
3863                         &(h->cfgtable->HostWrite.TransportRequest));
3864
3865         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3866         cciss_wait_for_mode_change_ack(h);
3867         register_value = readl(&(h->cfgtable->TransportActive));
3868         if (!(register_value & CFGTBL_Trans_Performant))
3869                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3870                                         " performant mode\n");
3871 }
3872
3873 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3874 {
3875         __u32 trans_support;
3876
3877         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3878         /* Attempt to put controller into performant mode if supported */
3879         /* Does board support performant mode? */
3880         trans_support = readl(&(h->cfgtable->TransportSupport));
3881         if (!(trans_support & PERFORMANT_MODE))
3882                 return;
3883
3884         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3885         /* Performant mode demands commands on a 32 byte boundary
3886          * pci_alloc_consistent aligns on page boundarys already.
3887          * Just need to check if divisible by 32
3888          */
3889         if ((sizeof(CommandList_struct) % 32) != 0) {
3890                 dev_warn(&h->pdev->dev, "%s %d %s\n",
3891                         "cciss info: command size[",
3892                         (int)sizeof(CommandList_struct),
3893                         "] not divisible by 32, no performant mode..\n");
3894                 return;
3895         }
3896
3897         /* Performant mode ring buffer and supporting data structures */
3898         h->reply_pool = (__u64 *)pci_alloc_consistent(
3899                 h->pdev, h->max_commands * sizeof(__u64),
3900                 &(h->reply_pool_dhandle));
3901
3902         /* Need a block fetch table for performant mode */
3903         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3904                 sizeof(__u32)), GFP_KERNEL);
3905
3906         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3907                 goto clean_up;
3908
3909         cciss_enter_performant_mode(h);
3910
3911         /* Change the access methods to the performant access methods */
3912         h->access = SA5_performant_access;
3913         h->transMethod = CFGTBL_Trans_Performant;
3914
3915         return;
3916 clean_up:
3917         kfree(h->blockFetchTable);
3918         if (h->reply_pool)
3919                 pci_free_consistent(h->pdev,
3920                                 h->max_commands * sizeof(__u64),
3921                                 h->reply_pool,
3922                                 h->reply_pool_dhandle);
3923         return;
3924
3925 } /* cciss_put_controller_into_performant_mode */
3926
3927 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3928  * controllers that are capable. If not, we use IO-APIC mode.
3929  */
3930
3931 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3932 {
3933 #ifdef CONFIG_PCI_MSI
3934         int err;
3935         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3936         {0, 2}, {0, 3}
3937         };
3938
3939         /* Some boards advertise MSI but don't really support it */
3940         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3941             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3942                 goto default_int_mode;
3943
3944         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3945                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3946                 if (!err) {
3947                         h->intr[0] = cciss_msix_entries[0].vector;
3948                         h->intr[1] = cciss_msix_entries[1].vector;
3949                         h->intr[2] = cciss_msix_entries[2].vector;
3950                         h->intr[3] = cciss_msix_entries[3].vector;
3951                         h->msix_vector = 1;
3952                         return;
3953                 }
3954                 if (err > 0) {
3955                         dev_warn(&h->pdev->dev,
3956                                 "only %d MSI-X vectors available\n", err);
3957                         goto default_int_mode;
3958                 } else {
3959                         dev_warn(&h->pdev->dev,
3960                                 "MSI-X init failed %d\n", err);
3961                         goto default_int_mode;
3962                 }
3963         }
3964         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3965                 if (!pci_enable_msi(h->pdev))
3966                         h->msi_vector = 1;
3967                 else
3968                         dev_warn(&h->pdev->dev, "MSI init failed\n");
3969         }
3970 default_int_mode:
3971 #endif                          /* CONFIG_PCI_MSI */
3972         /* if we get here we're going to use the default interrupt mode */
3973         h->intr[PERF_MODE_INT] = h->pdev->irq;
3974         return;
3975 }
3976
3977 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3978 {
3979         int i;
3980         u32 subsystem_vendor_id, subsystem_device_id;
3981
3982         subsystem_vendor_id = pdev->subsystem_vendor;
3983         subsystem_device_id = pdev->subsystem_device;
3984         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3985                         subsystem_vendor_id;
3986
3987         for (i = 0; i < ARRAY_SIZE(products); i++) {
3988                 /* Stand aside for hpsa driver on request */
3989                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3990                         return -ENODEV;
3991                 if (*board_id == products[i].board_id)
3992                         return i;
3993         }
3994         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
3995                 *board_id);
3996         return -ENODEV;
3997 }
3998
3999 static inline bool cciss_board_disabled(ctlr_info_t *h)
4000 {
4001         u16 command;
4002
4003         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4004         return ((command & PCI_COMMAND_MEMORY) == 0);
4005 }
4006
4007 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4008         unsigned long *memory_bar)
4009 {
4010         int i;
4011
4012         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4013                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4014                         /* addressing mode bits already removed */
4015                         *memory_bar = pci_resource_start(pdev, i);
4016                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4017                                 *memory_bar);
4018                         return 0;
4019                 }
4020         dev_warn(&pdev->dev, "no memory BAR found\n");
4021         return -ENODEV;
4022 }
4023
4024 static int __devinit cciss_wait_for_board_ready(ctlr_info_t *h)
4025 {
4026         int i;
4027         u32 scratchpad;
4028
4029         for (i = 0; i < CCISS_BOARD_READY_ITERATIONS; i++) {
4030                 scratchpad = readl(h->vaddr + SA5_SCRATCHPAD_OFFSET);
4031                 if (scratchpad == CCISS_FIRMWARE_READY)
4032                         return 0;
4033                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4034         }
4035         dev_warn(&h->pdev->dev, "board not ready, timed out.\n");
4036         return -ENODEV;
4037 }
4038
4039 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4040         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4041         u64 *cfg_offset)
4042 {
4043         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4044         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4045         *cfg_base_addr &= (u32) 0x0000ffff;
4046         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4047         if (*cfg_base_addr_index == -1) {
4048                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4049                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4050                 return -ENODEV;
4051         }
4052         return 0;
4053 }
4054
4055 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4056 {
4057         u64 cfg_offset;
4058         u32 cfg_base_addr;
4059         u64 cfg_base_addr_index;
4060         u32 trans_offset;
4061         int rc;
4062
4063         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4064                 &cfg_base_addr_index, &cfg_offset);
4065         if (rc)
4066                 return rc;
4067         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4068                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4069         if (!h->cfgtable)
4070                 return -ENOMEM;
4071         /* Find performant mode table. */
4072         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4073         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4074                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4075                                 sizeof(*h->transtable));
4076         if (!h->transtable)
4077                 return -ENOMEM;
4078         return 0;
4079 }
4080
4081 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4082 {
4083         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4084         if (h->max_commands < 16) {
4085                 dev_warn(&h->pdev->dev, "Controller reports "
4086                         "max supported commands of %d, an obvious lie. "
4087                         "Using 16.  Ensure that firmware is up to date.\n",
4088                         h->max_commands);
4089                 h->max_commands = 16;
4090         }
4091 }
4092
4093 /* Interrogate the hardware for some limits:
4094  * max commands, max SG elements without chaining, and with chaining,
4095  * SG chain block size, etc.
4096  */
4097 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4098 {
4099         cciss_get_max_perf_mode_cmds(h);
4100         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4101         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4102         /*
4103          * Limit in-command s/g elements to 32 save dma'able memory.
4104          * Howvever spec says if 0, use 31
4105          */
4106         h->max_cmd_sgentries = 31;
4107         if (h->maxsgentries > 512) {
4108                 h->max_cmd_sgentries = 32;
4109                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4110                 h->maxsgentries--; /* save one for chain pointer */
4111         } else {
4112                 h->maxsgentries = 31; /* default to traditional values */
4113                 h->chainsize = 0;
4114         }
4115 }
4116
4117 static inline bool CISS_signature_present(ctlr_info_t *h)
4118 {
4119         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4120             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4121             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4122             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4123                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4124                 return false;
4125         }
4126         return true;
4127 }
4128
4129 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4130 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4131 {
4132 #ifdef CONFIG_X86
4133         u32 prefetch;
4134
4135         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4136         prefetch |= 0x100;
4137         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4138 #endif
4139 }
4140
4141 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4142  * in a prefetch beyond physical memory.
4143  */
4144 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4145 {
4146         u32 dma_prefetch;
4147         __u32 dma_refetch;
4148
4149         if (h->board_id != 0x3225103C)
4150                 return;
4151         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4152         dma_prefetch |= 0x8000;
4153         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4154         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4155         dma_refetch |= 0x1;
4156         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4157 }
4158
4159 static int __devinit cciss_pci_init(ctlr_info_t *h)
4160 {
4161         int prod_index, err;
4162
4163         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4164         if (prod_index < 0)
4165                 return -ENODEV;
4166         h->product_name = products[prod_index].product_name;
4167         h->access = *(products[prod_index].access);
4168
4169         if (cciss_board_disabled(h)) {
4170                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4171                 return -ENODEV;
4172         }
4173         err = pci_enable_device(h->pdev);
4174         if (err) {
4175                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4176                 return err;
4177         }
4178
4179         err = pci_request_regions(h->pdev, "cciss");
4180         if (err) {
4181                 dev_warn(&h->pdev->dev,
4182                         "Cannot obtain PCI resources, aborting\n");
4183                 return err;
4184         }
4185
4186         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4187         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4188
4189 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4190  * else we use the IO-APIC interrupt assigned to us by system ROM.
4191  */
4192         cciss_interrupt_mode(h);
4193         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4194         if (err)
4195                 goto err_out_free_res;
4196         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4197         if (!h->vaddr) {
4198                 err = -ENOMEM;
4199                 goto err_out_free_res;
4200         }
4201         err = cciss_wait_for_board_ready(h);
4202         if (err)
4203                 goto err_out_free_res;
4204         err = cciss_find_cfgtables(h);
4205         if (err)
4206                 goto err_out_free_res;
4207         print_cfg_table(h);
4208         cciss_find_board_params(h);
4209
4210         if (!CISS_signature_present(h)) {
4211                 err = -ENODEV;
4212                 goto err_out_free_res;
4213         }
4214         cciss_enable_scsi_prefetch(h);
4215         cciss_p600_dma_prefetch_quirk(h);
4216         cciss_put_controller_into_performant_mode(h);
4217         return 0;
4218
4219 err_out_free_res:
4220         /*
4221          * Deliberately omit pci_disable_device(): it does something nasty to
4222          * Smart Array controllers that pci_enable_device does not undo
4223          */
4224         if (h->transtable)
4225                 iounmap(h->transtable);
4226         if (h->cfgtable)
4227                 iounmap(h->cfgtable);
4228         if (h->vaddr)
4229                 iounmap(h->vaddr);
4230         pci_release_regions(h->pdev);
4231         return err;
4232 }
4233
4234 /* Function to find the first free pointer into our hba[] array
4235  * Returns -1 if no free entries are left.
4236  */
4237 static int alloc_cciss_hba(struct pci_dev *pdev)
4238 {
4239         int i;
4240
4241         for (i = 0; i < MAX_CTLR; i++) {
4242                 if (!hba[i]) {
4243                         ctlr_info_t *h;
4244
4245                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4246                         if (!h)
4247                                 goto Enomem;
4248                         hba[i] = h;
4249                         return i;
4250                 }
4251         }
4252         dev_warn(&pdev->dev, "This driver supports a maximum"
4253                " of %d controllers.\n", MAX_CTLR);
4254         return -1;
4255 Enomem:
4256         dev_warn(&pdev->dev, "out of memory.\n");
4257         return -1;
4258 }
4259
4260 static void free_hba(ctlr_info_t *h)
4261 {
4262         int i;
4263
4264         hba[h->ctlr] = NULL;
4265         for (i = 0; i < h->highest_lun + 1; i++)
4266                 if (h->gendisk[i] != NULL)
4267                         put_disk(h->gendisk[i]);
4268         kfree(h);
4269 }
4270
4271 /* Send a message CDB to the firmware. */
4272 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4273 {
4274         typedef struct {
4275                 CommandListHeader_struct CommandHeader;
4276                 RequestBlock_struct Request;
4277                 ErrDescriptor_struct ErrorDescriptor;
4278         } Command;
4279         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4280         Command *cmd;
4281         dma_addr_t paddr64;
4282         uint32_t paddr32, tag;
4283         void __iomem *vaddr;
4284         int i, err;
4285
4286         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4287         if (vaddr == NULL)
4288                 return -ENOMEM;
4289
4290         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4291            CCISS commands, so they must be allocated from the lower 4GiB of
4292            memory. */
4293         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4294         if (err) {
4295                 iounmap(vaddr);
4296                 return -ENOMEM;
4297         }
4298
4299         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4300         if (cmd == NULL) {
4301                 iounmap(vaddr);
4302                 return -ENOMEM;
4303         }
4304
4305         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4306            although there's no guarantee, we assume that the address is at
4307            least 4-byte aligned (most likely, it's page-aligned). */
4308         paddr32 = paddr64;
4309
4310         cmd->CommandHeader.ReplyQueue = 0;
4311         cmd->CommandHeader.SGList = 0;
4312         cmd->CommandHeader.SGTotal = 0;
4313         cmd->CommandHeader.Tag.lower = paddr32;
4314         cmd->CommandHeader.Tag.upper = 0;
4315         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4316
4317         cmd->Request.CDBLen = 16;
4318         cmd->Request.Type.Type = TYPE_MSG;
4319         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4320         cmd->Request.Type.Direction = XFER_NONE;
4321         cmd->Request.Timeout = 0; /* Don't time out */
4322         cmd->Request.CDB[0] = opcode;
4323         cmd->Request.CDB[1] = type;
4324         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4325
4326         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4327         cmd->ErrorDescriptor.Addr.upper = 0;
4328         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4329
4330         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4331
4332         for (i = 0; i < 10; i++) {
4333                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4334                 if ((tag & ~3) == paddr32)
4335                         break;
4336                 schedule_timeout_uninterruptible(HZ);
4337         }
4338
4339         iounmap(vaddr);
4340
4341         /* we leak the DMA buffer here ... no choice since the controller could
4342            still complete the command. */
4343         if (i == 10) {
4344                 dev_err(&pdev->dev,
4345                         "controller message %02x:%02x timed out\n",
4346                         opcode, type);
4347                 return -ETIMEDOUT;
4348         }
4349
4350         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4351
4352         if (tag & 2) {
4353                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4354                         opcode, type);
4355                 return -EIO;
4356         }
4357
4358         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4359                 opcode, type);
4360         return 0;
4361 }
4362
4363 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4364 #define cciss_noop(p) cciss_message(p, 3, 0)
4365
4366 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4367 {
4368 /* the #defines are stolen from drivers/pci/msi.h. */
4369 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4370 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4371
4372         int pos;
4373         u16 control = 0;
4374
4375         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4376         if (pos) {
4377                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4378                 if (control & PCI_MSI_FLAGS_ENABLE) {
4379                         dev_info(&pdev->dev, "resetting MSI\n");
4380                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4381                 }
4382         }
4383
4384         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4385         if (pos) {
4386                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4387                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4388                         dev_info(&pdev->dev, "resetting MSI-X\n");
4389                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4390                 }
4391         }
4392
4393         return 0;
4394 }
4395
4396 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4397         void * __iomem vaddr, bool use_doorbell)
4398 {
4399         u16 pmcsr;
4400         int pos;
4401
4402         if (use_doorbell) {
4403                 /* For everything after the P600, the PCI power state method
4404                  * of resetting the controller doesn't work, so we have this
4405                  * other way using the doorbell register.
4406                  */
4407                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4408                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4409                 msleep(1000);
4410         } else { /* Try to do it the PCI power state way */
4411
4412                 /* Quoting from the Open CISS Specification: "The Power
4413                  * Management Control/Status Register (CSR) controls the power
4414                  * state of the device.  The normal operating state is D0,
4415                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4416                  * the controller, place the interface device in D3 then to D0,
4417                  * this causes a secondary PCI reset which will reset the
4418                  * controller." */
4419
4420                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4421                 if (pos == 0) {
4422                         dev_err(&pdev->dev,
4423                                 "cciss_controller_hard_reset: "
4424                                 "PCI PM not supported\n");
4425                         return -ENODEV;
4426                 }
4427                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4428                 /* enter the D3hot power management state */
4429                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4430                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4431                 pmcsr |= PCI_D3hot;
4432                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4433
4434                 msleep(500);
4435
4436                 /* enter the D0 power management state */
4437                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4438                 pmcsr |= PCI_D0;
4439                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4440
4441                 msleep(500);
4442         }
4443         return 0;
4444 }
4445
4446 /* This does a hard reset of the controller using PCI power management
4447  * states or using the doorbell register. */
4448 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4449 {
4450         u16 saved_config_space[32];
4451         u64 cfg_offset;
4452         u32 cfg_base_addr;
4453         u64 cfg_base_addr_index;
4454         void __iomem *vaddr;
4455         unsigned long paddr;
4456         u32 misc_fw_support, active_transport;
4457         int rc, i;
4458         CfgTable_struct __iomem *cfgtable;
4459         bool use_doorbell;
4460         u32 board_id;
4461
4462         /* For controllers as old a the p600, this is very nearly
4463          * the same thing as
4464          *
4465          * pci_save_state(pci_dev);
4466          * pci_set_power_state(pci_dev, PCI_D3hot);
4467          * pci_set_power_state(pci_dev, PCI_D0);
4468          * pci_restore_state(pci_dev);
4469          *
4470          * but we can't use these nice canned kernel routines on
4471          * kexec, because they also check the MSI/MSI-X state in PCI
4472          * configuration space and do the wrong thing when it is
4473          * set/cleared.  Also, the pci_save/restore_state functions
4474          * violate the ordering requirements for restoring the
4475          * configuration space from the CCISS document (see the
4476          * comment below).  So we roll our own ....
4477          *
4478          * For controllers newer than the P600, the pci power state
4479          * method of resetting doesn't work so we have another way
4480          * using the doorbell register.
4481          */
4482
4483         /* Exclude 640x boards.  These are two pci devices in one slot
4484          * which share a battery backed cache module.  One controls the
4485          * cache, the other accesses the cache through the one that controls
4486          * it.  If we reset the one controlling the cache, the other will
4487          * likely not be happy.  Just forbid resetting this conjoined mess.
4488          */
4489         cciss_lookup_board_id(pdev, &board_id);
4490         if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4491                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4492                                 "due to shared cache module.");
4493                 return -ENODEV;
4494         }
4495
4496         for (i = 0; i < 32; i++)
4497                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4498
4499         /* find the first memory BAR, so we can find the cfg table */
4500         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4501         if (rc)
4502                 return rc;
4503         vaddr = remap_pci_mem(paddr, 0x250);
4504         if (!vaddr)
4505                 return -ENOMEM;
4506
4507         /* find cfgtable in order to check if reset via doorbell is supported */
4508         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4509                                         &cfg_base_addr_index, &cfg_offset);
4510         if (rc)
4511                 goto unmap_vaddr;
4512         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4513                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4514         if (!cfgtable) {
4515                 rc = -ENOMEM;
4516                 goto unmap_vaddr;
4517         }
4518
4519         /* If reset via doorbell register is supported, use that. */
4520         misc_fw_support = readl(&cfgtable->misc_fw_support);
4521         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4522
4523         /* The doorbell reset seems to cause lockups on some Smart
4524          * Arrays (e.g. P410, P410i, maybe others).  Until this is
4525          * fixed or at least isolated, avoid the doorbell reset.
4526          */
4527         use_doorbell = 0;
4528
4529         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4530         if (rc)
4531                 goto unmap_cfgtable;
4532
4533         /* Restore the PCI configuration space.  The Open CISS
4534          * Specification says, "Restore the PCI Configuration
4535          * Registers, offsets 00h through 60h. It is important to
4536          * restore the command register, 16-bits at offset 04h,
4537          * last. Do not restore the configuration status register,
4538          * 16-bits at offset 06h."  Note that the offset is 2*i.
4539          */
4540         for (i = 0; i < 32; i++) {
4541                 if (i == 2 || i == 3)
4542                         continue;
4543                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4544         }
4545         wmb();
4546         pci_write_config_word(pdev, 4, saved_config_space[2]);
4547
4548         /* Some devices (notably the HP Smart Array 5i Controller)
4549            need a little pause here */
4550         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4551
4552         /* Controller should be in simple mode at this point.  If it's not,
4553          * It means we're on one of those controllers which doesn't support
4554          * the doorbell reset method and on which the PCI power management reset
4555          * method doesn't work (P800, for example.)
4556          * In those cases, don't try to proceed, as it generally doesn't work.
4557          */
4558         active_transport = readl(&cfgtable->TransportActive);
4559         if (active_transport & PERFORMANT_MODE) {
4560                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4561                         " Ignoring controller.\n");
4562                 rc = -ENODEV;
4563         }
4564
4565 unmap_cfgtable:
4566         iounmap(cfgtable);
4567
4568 unmap_vaddr:
4569         iounmap(vaddr);
4570         return rc;
4571 }
4572
4573 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4574 {
4575         int rc, i;
4576
4577         if (!reset_devices)
4578                 return 0;
4579
4580         /* Reset the controller with a PCI power-cycle or via doorbell */
4581         rc = cciss_kdump_hard_reset_controller(pdev);
4582
4583         /* -ENOTSUPP here means we cannot reset the controller
4584          * but it's already (and still) up and running in
4585          * "performant mode".  Or, it might be 640x, which can't reset
4586          * due to concerns about shared bbwc between 6402/6404 pair.
4587          */
4588         if (rc == -ENOTSUPP)
4589                 return 0; /* just try to do the kdump anyhow. */
4590         if (rc)
4591                 return -ENODEV;
4592         if (cciss_reset_msi(pdev))
4593                 return -ENODEV;
4594
4595         /* Now try to get the controller to respond to a no-op */
4596         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4597                 if (cciss_noop(pdev) == 0)
4598                         break;
4599                 else
4600                         dev_warn(&pdev->dev, "no-op failed%s\n",
4601                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4602                                         "; re-trying" : ""));
4603                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4604         }
4605         return 0;
4606 }
4607
4608 /*
4609  *  This is it.  Find all the controllers and register them.  I really hate
4610  *  stealing all these major device numbers.
4611  *  returns the number of block devices registered.
4612  */
4613 static int __devinit cciss_init_one(struct pci_dev *pdev,
4614                                     const struct pci_device_id *ent)
4615 {
4616         int i;
4617         int j = 0;
4618         int k = 0;
4619         int rc;
4620         int dac, return_code;
4621         InquiryData_struct *inq_buff;
4622         ctlr_info_t *h;
4623
4624         rc = cciss_init_reset_devices(pdev);
4625         if (rc)
4626                 return rc;
4627         i = alloc_cciss_hba(pdev);
4628         if (i < 0)
4629                 return -1;
4630
4631         h = hba[i];
4632         h->pdev = pdev;
4633         h->busy_initializing = 1;
4634         INIT_LIST_HEAD(&h->cmpQ);
4635         INIT_LIST_HEAD(&h->reqQ);
4636         mutex_init(&h->busy_shutting_down);
4637
4638         if (cciss_pci_init(h) != 0)
4639                 goto clean_no_release_regions;
4640
4641         sprintf(h->devname, "cciss%d", i);
4642         h->ctlr = i;
4643
4644         init_completion(&h->scan_wait);
4645
4646         if (cciss_create_hba_sysfs_entry(h))
4647                 goto clean0;
4648
4649         /* configure PCI DMA stuff */
4650         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4651                 dac = 1;
4652         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4653                 dac = 0;
4654         else {
4655                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4656                 goto clean1;
4657         }
4658
4659         /*
4660          * register with the major number, or get a dynamic major number
4661          * by passing 0 as argument.  This is done for greater than
4662          * 8 controller support.
4663          */
4664         if (i < MAX_CTLR_ORIG)
4665                 h->major = COMPAQ_CISS_MAJOR + i;
4666         rc = register_blkdev(h->major, h->devname);
4667         if (rc == -EBUSY || rc == -EINVAL) {
4668                 dev_err(&h->pdev->dev,
4669                        "Unable to get major number %d for %s "
4670                        "on hba %d\n", h->major, h->devname, i);
4671                 goto clean1;
4672         } else {
4673                 if (i >= MAX_CTLR_ORIG)
4674                         h->major = rc;
4675         }
4676
4677         /* make sure the board interrupts are off */
4678         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4679         if (h->msi_vector || h->msix_vector) {
4680                 if (request_irq(h->intr[PERF_MODE_INT],
4681                                 do_cciss_msix_intr,
4682                                 IRQF_DISABLED, h->devname, h)) {
4683                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4684                                h->intr[PERF_MODE_INT], h->devname);
4685                         goto clean2;
4686                 }
4687         } else {
4688                 if (request_irq(h->intr[PERF_MODE_INT], do_cciss_intx,
4689                                 IRQF_DISABLED, h->devname, h)) {
4690                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4691                                h->intr[PERF_MODE_INT], h->devname);
4692                         goto clean2;
4693                 }
4694         }
4695
4696         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4697                h->devname, pdev->device, pci_name(pdev),
4698                h->intr[PERF_MODE_INT], dac ? "" : " not");
4699
4700         h->cmd_pool_bits =
4701             kmalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4702                         * sizeof(unsigned long), GFP_KERNEL);
4703         h->cmd_pool = (CommandList_struct *)
4704             pci_alloc_consistent(h->pdev,
4705                     h->nr_cmds * sizeof(CommandList_struct),
4706                     &(h->cmd_pool_dhandle));
4707         h->errinfo_pool = (ErrorInfo_struct *)
4708             pci_alloc_consistent(h->pdev,
4709                     h->nr_cmds * sizeof(ErrorInfo_struct),
4710                     &(h->errinfo_pool_dhandle));
4711         if ((h->cmd_pool_bits == NULL)
4712             || (h->cmd_pool == NULL)
4713             || (h->errinfo_pool == NULL)) {
4714                 dev_err(&h->pdev->dev, "out of memory");
4715                 goto clean4;
4716         }
4717
4718         /* Need space for temp scatter list */
4719         h->scatter_list = kmalloc(h->max_commands *
4720                                                 sizeof(struct scatterlist *),
4721                                                 GFP_KERNEL);
4722         if (!h->scatter_list)
4723                 goto clean4;
4724
4725         for (k = 0; k < h->nr_cmds; k++) {
4726                 h->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4727                                                         h->maxsgentries,
4728                                                         GFP_KERNEL);
4729                 if (h->scatter_list[k] == NULL) {
4730                         dev_err(&h->pdev->dev,
4731                                 "could not allocate s/g lists\n");
4732                         goto clean4;
4733                 }
4734         }
4735         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4736                 h->chainsize, h->nr_cmds);
4737         if (!h->cmd_sg_list && h->chainsize > 0)
4738                 goto clean4;
4739
4740         spin_lock_init(&h->lock);
4741
4742         /* Initialize the pdev driver private data.
4743            have it point to h.  */
4744         pci_set_drvdata(pdev, h);
4745         /* command and error info recs zeroed out before
4746            they are used */
4747         memset(h->cmd_pool_bits, 0,
4748                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4749                         * sizeof(unsigned long));
4750
4751         h->num_luns = 0;
4752         h->highest_lun = -1;
4753         for (j = 0; j < CISS_MAX_LUN; j++) {
4754                 h->drv[j] = NULL;
4755                 h->gendisk[j] = NULL;
4756         }
4757
4758         cciss_scsi_setup(h);
4759
4760         /* Turn the interrupts on so we can service requests */
4761         h->access.set_intr_mask(h, CCISS_INTR_ON);
4762
4763         /* Get the firmware version */
4764         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4765         if (inq_buff == NULL) {
4766                 dev_err(&h->pdev->dev, "out of memory\n");
4767                 goto clean4;
4768         }
4769
4770         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4771                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4772         if (return_code == IO_OK) {
4773                 h->firm_ver[0] = inq_buff->data_byte[32];
4774                 h->firm_ver[1] = inq_buff->data_byte[33];
4775                 h->firm_ver[2] = inq_buff->data_byte[34];
4776                 h->firm_ver[3] = inq_buff->data_byte[35];
4777         } else {         /* send command failed */
4778                 dev_warn(&h->pdev->dev, "unable to determine firmware"
4779                         " version of controller\n");
4780         }
4781         kfree(inq_buff);
4782
4783         cciss_procinit(h);
4784
4785         h->cciss_max_sectors = 8192;
4786
4787         rebuild_lun_table(h, 1, 0);
4788         h->busy_initializing = 0;
4789         return 1;
4790
4791 clean4:
4792         kfree(h->cmd_pool_bits);
4793         /* Free up sg elements */
4794         for (k-- ; k >= 0; k--)
4795                 kfree(h->scatter_list[k]);
4796         kfree(h->scatter_list);
4797         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4798         if (h->cmd_pool)
4799                 pci_free_consistent(h->pdev,
4800                                     h->nr_cmds * sizeof(CommandList_struct),
4801                                     h->cmd_pool, h->cmd_pool_dhandle);
4802         if (h->errinfo_pool)
4803                 pci_free_consistent(h->pdev,
4804                                     h->nr_cmds * sizeof(ErrorInfo_struct),
4805                                     h->errinfo_pool,
4806                                     h->errinfo_pool_dhandle);
4807         free_irq(h->intr[PERF_MODE_INT], h);
4808 clean2:
4809         unregister_blkdev(h->major, h->devname);
4810 clean1:
4811         cciss_destroy_hba_sysfs_entry(h);
4812 clean0:
4813         pci_release_regions(pdev);
4814 clean_no_release_regions:
4815         h->busy_initializing = 0;
4816
4817         /*
4818          * Deliberately omit pci_disable_device(): it does something nasty to
4819          * Smart Array controllers that pci_enable_device does not undo
4820          */
4821         pci_set_drvdata(pdev, NULL);
4822         free_hba(h);
4823         return -1;
4824 }
4825
4826 static void cciss_shutdown(struct pci_dev *pdev)
4827 {
4828         ctlr_info_t *h;
4829         char *flush_buf;
4830         int return_code;
4831
4832         h = pci_get_drvdata(pdev);
4833         flush_buf = kzalloc(4, GFP_KERNEL);
4834         if (!flush_buf) {
4835                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
4836                 return;
4837         }
4838         /* write all data in the battery backed cache to disk */
4839         memset(flush_buf, 0, 4);
4840         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4841                 4, 0, CTLR_LUNID, TYPE_CMD);
4842         kfree(flush_buf);
4843         if (return_code != IO_OK)
4844                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
4845         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4846         free_irq(h->intr[PERF_MODE_INT], h);
4847 }
4848
4849 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4850 {
4851         ctlr_info_t *h;
4852         int i, j;
4853
4854         if (pci_get_drvdata(pdev) == NULL) {
4855                 dev_err(&pdev->dev, "Unable to remove device\n");
4856                 return;
4857         }
4858
4859         h = pci_get_drvdata(pdev);
4860         i = h->ctlr;
4861         if (hba[i] == NULL) {
4862                 dev_err(&pdev->dev, "device appears to already be removed\n");
4863                 return;
4864         }
4865
4866         mutex_lock(&h->busy_shutting_down);
4867
4868         remove_from_scan_list(h);
4869         remove_proc_entry(h->devname, proc_cciss);
4870         unregister_blkdev(h->major, h->devname);
4871
4872         /* remove it from the disk list */
4873         for (j = 0; j < CISS_MAX_LUN; j++) {
4874                 struct gendisk *disk = h->gendisk[j];
4875                 if (disk) {
4876                         struct request_queue *q = disk->queue;
4877
4878                         if (disk->flags & GENHD_FL_UP) {
4879                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
4880                                 del_gendisk(disk);
4881                         }
4882                         if (q)
4883                                 blk_cleanup_queue(q);
4884                 }
4885         }
4886
4887 #ifdef CONFIG_CISS_SCSI_TAPE
4888         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
4889 #endif
4890
4891         cciss_shutdown(pdev);
4892
4893 #ifdef CONFIG_PCI_MSI
4894         if (h->msix_vector)
4895                 pci_disable_msix(h->pdev);
4896         else if (h->msi_vector)
4897                 pci_disable_msi(h->pdev);
4898 #endif                          /* CONFIG_PCI_MSI */
4899
4900         iounmap(h->transtable);
4901         iounmap(h->cfgtable);
4902         iounmap(h->vaddr);
4903
4904         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(CommandList_struct),
4905                             h->cmd_pool, h->cmd_pool_dhandle);
4906         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(ErrorInfo_struct),
4907                             h->errinfo_pool, h->errinfo_pool_dhandle);
4908         kfree(h->cmd_pool_bits);
4909         /* Free up sg elements */
4910         for (j = 0; j < h->nr_cmds; j++)
4911                 kfree(h->scatter_list[j]);
4912         kfree(h->scatter_list);
4913         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4914         /*
4915          * Deliberately omit pci_disable_device(): it does something nasty to
4916          * Smart Array controllers that pci_enable_device does not undo
4917          */
4918         pci_release_regions(pdev);
4919         pci_set_drvdata(pdev, NULL);
4920         cciss_destroy_hba_sysfs_entry(h);
4921         mutex_unlock(&h->busy_shutting_down);
4922         free_hba(h);
4923 }
4924
4925 static struct pci_driver cciss_pci_driver = {
4926         .name = "cciss",
4927         .probe = cciss_init_one,
4928         .remove = __devexit_p(cciss_remove_one),
4929         .id_table = cciss_pci_device_id,        /* id_table */
4930         .shutdown = cciss_shutdown,
4931 };
4932
4933 /*
4934  *  This is it.  Register the PCI driver information for the cards we control
4935  *  the OS will call our registered routines when it finds one of our cards.
4936  */
4937 static int __init cciss_init(void)
4938 {
4939         int err;
4940
4941         /*
4942          * The hardware requires that commands are aligned on a 64-bit
4943          * boundary. Given that we use pci_alloc_consistent() to allocate an
4944          * array of them, the size must be a multiple of 8 bytes.
4945          */
4946         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4947         printk(KERN_INFO DRIVER_NAME "\n");
4948
4949         err = bus_register(&cciss_bus_type);
4950         if (err)
4951                 return err;
4952
4953         /* Start the scan thread */
4954         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4955         if (IS_ERR(cciss_scan_thread)) {
4956                 err = PTR_ERR(cciss_scan_thread);
4957                 goto err_bus_unregister;
4958         }
4959
4960         /* Register for our PCI devices */
4961         err = pci_register_driver(&cciss_pci_driver);
4962         if (err)
4963                 goto err_thread_stop;
4964
4965         return err;
4966
4967 err_thread_stop:
4968         kthread_stop(cciss_scan_thread);
4969 err_bus_unregister:
4970         bus_unregister(&cciss_bus_type);
4971
4972         return err;
4973 }
4974
4975 static void __exit cciss_cleanup(void)
4976 {
4977         int i;
4978
4979         pci_unregister_driver(&cciss_pci_driver);
4980         /* double check that all controller entrys have been removed */
4981         for (i = 0; i < MAX_CTLR; i++) {
4982                 if (hba[i] != NULL) {
4983                         dev_warn(&hba[i]->pdev->dev,
4984                                 "had to remove controller\n");
4985                         cciss_remove_one(hba[i]->pdev);
4986                 }
4987         }
4988         kthread_stop(cciss_scan_thread);
4989         remove_proc_entry("driver/cciss", NULL);
4990         bus_unregister(&cciss_bus_type);
4991 }
4992
4993 module_init(cciss_init);
4994 module_exit(cciss_cleanup);