drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/compat.h>
  43 #include <linux/mutex.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/io.h>
  46
  47 #include <linux/dma-mapping.h>
  48 #include <linux/blkdev.h>
  49 #include <linux/genhd.h>
  50 #include <linux/completion.h>
  51 #include <scsi/scsi.h>
  52 #include <scsi/sg.h>
  53 #include <scsi/scsi_ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/scatterlist.h>
  56 #include <linux/kthread.h>
  57
  58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  59 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  61
  62 /* Embedded module documentation macros - see modules.h */
  63 MODULE_AUTHOR("Hewlett-Packard Company");
  64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  65 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  66 MODULE_VERSION("3.6.26");
  67 MODULE_LICENSE("GPL");
  68
  69 static int cciss_allow_hpsa;
  70 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  71 MODULE_PARM_DESC(cciss_allow_hpsa,
  72         "Prevent cciss driver from accessing hardware known to be "
  73         " supported by the hpsa driver");
  74
  75 #include "cciss_cmd.h"
  76 #include "cciss.h"
  77 #include <linux/cciss_ioctl.h>
  78
  79 /* define the PCI info for the cards we can control */
  80 static const struct pci_device_id cciss_pci_device_id[] = {
  81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3250},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3251},
 110         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3252},
 111         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3253},
 112         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3254},
 113         {0,}
 114 };
 115
 116 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 117
 118 /*  board_id = Subsystem Device ID & Vendor ID
 119  *  product = Marketing Name for the board
 120  *  access = Address of the struct of function pointers
 121  */
 122 static struct board_type products[] = {
 123         {0x40700E11, "Smart Array 5300", &SA5_access},
 124         {0x40800E11, "Smart Array 5i", &SA5B_access},
 125         {0x40820E11, "Smart Array 532", &SA5B_access},
 126         {0x40830E11, "Smart Array 5312", &SA5B_access},
 127         {0x409A0E11, "Smart Array 641", &SA5_access},
 128         {0x409B0E11, "Smart Array 642", &SA5_access},
 129         {0x409C0E11, "Smart Array 6400", &SA5_access},
 130         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 131         {0x40910E11, "Smart Array 6i", &SA5_access},
 132         {0x3225103C, "Smart Array P600", &SA5_access},
 133         {0x3235103C, "Smart Array P400i", &SA5_access},
 134         {0x3211103C, "Smart Array E200i", &SA5_access},
 135         {0x3212103C, "Smart Array E200", &SA5_access},
 136         {0x3213103C, "Smart Array E200i", &SA5_access},
 137         {0x3214103C, "Smart Array E200i", &SA5_access},
 138         {0x3215103C, "Smart Array E200i", &SA5_access},
 139         {0x3237103C, "Smart Array E500", &SA5_access},
 140 /* controllers below this line are also supported by the hpsa driver. */
 141 #define HPSA_BOUNDARY 0x3223103C
 142         {0x3223103C, "Smart Array P800", &SA5_access},
 143         {0x3234103C, "Smart Array P400", &SA5_access},
 144         {0x323D103C, "Smart Array P700m", &SA5_access},
 145         {0x3241103C, "Smart Array P212", &SA5_access},
 146         {0x3243103C, "Smart Array P410", &SA5_access},
 147         {0x3245103C, "Smart Array P410i", &SA5_access},
 148         {0x3247103C, "Smart Array P411", &SA5_access},
 149         {0x3249103C, "Smart Array P812", &SA5_access},
 150         {0x324A103C, "Smart Array P712m", &SA5_access},
 151         {0x324B103C, "Smart Array P711m", &SA5_access},
 152         {0x3250103C, "Smart Array", &SA5_access},
 153         {0x3251103C, "Smart Array", &SA5_access},
 154         {0x3252103C, "Smart Array", &SA5_access},
 155         {0x3253103C, "Smart Array", &SA5_access},
 156         {0x3254103C, "Smart Array", &SA5_access},
 157 };
 158
 159 /* How long to wait (in milliseconds) for board to go into simple mode */
 160 #define MAX_CONFIG_WAIT 30000
 161 #define MAX_IOCTL_CONFIG_WAIT 1000
 162
 163 /*define how many times we will try a command because of bus resets */
 164 #define MAX_CMD_RETRIES 3
 165
 166 #define MAX_CTLR        32
 167
 168 /* Originally cciss driver only supports 8 major numbers */
 169 #define MAX_CTLR_ORIG   8
 170
 171 static ctlr_info_t *hba[MAX_CTLR];
 172
 173 static struct task_struct *cciss_scan_thread;
 174 static DEFINE_MUTEX(scan_mutex);
 175 static LIST_HEAD(scan_q);
 176
 177 static void do_cciss_request(struct request_queue *q);
 178 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 179 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 180 static int cciss_open(struct block_device *bdev, fmode_t mode);
 181 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 182 static int cciss_release(struct gendisk *disk, fmode_t mode);
 183 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 184                     unsigned int cmd, unsigned long arg);
 185 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 186                        unsigned int cmd, unsigned long arg);
 187 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 188
 189 static int cciss_revalidate(struct gendisk *disk);
 190 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 191 static int deregister_disk(ctlr_info_t *h, int drv_index,
 192                            int clear_all, int via_ioctl);
 193
 194 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 195                         sector_t *total_size, unsigned int *block_size);
 196 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 197                         sector_t *total_size, unsigned int *block_size);
 198 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 199                         sector_t total_size,
 200                         unsigned int block_size, InquiryData_struct *inq_buff,
 201                                    drive_info_struct *drv);
 202 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
 203 static void start_io(ctlr_info_t *h);
 204 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 205                         __u8 page_code, unsigned char scsi3addr[],
 206                         int cmd_type);
 207 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 208         int attempt_retry);
 209 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 210
 211 static int add_to_scan_list(struct ctlr_info *h);
 212 static int scan_thread(void *data);
 213 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 214 static void cciss_hba_release(struct device *dev);
 215 static void cciss_device_release(struct device *dev);
 216 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 217 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 218 static inline u32 next_command(ctlr_info_t *h);
 219 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
 220         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 221         u64 *cfg_offset);
 222 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 223         unsigned long *memory_bar);
 224
 225
 226 /* performant mode helper functions */
 227 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 228                                 int *bucket_map);
 229 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 230
 231 #ifdef CONFIG_PROC_FS
 232 static void cciss_procinit(ctlr_info_t *h);
 233 #else
 234 static void cciss_procinit(ctlr_info_t *h)
 235 {
 236 }
 237 #endif                          /* CONFIG_PROC_FS */
 238
 239 #ifdef CONFIG_COMPAT
 240 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 241                               unsigned, unsigned long);
 242 #endif
 243
 244 static const struct block_device_operations cciss_fops = {
 245         .owner = THIS_MODULE,
 246         .open = cciss_unlocked_open,
 247         .release = cciss_release,
 248         .ioctl = do_ioctl,
 249         .getgeo = cciss_getgeo,
 250 #ifdef CONFIG_COMPAT
 251         .compat_ioctl = cciss_compat_ioctl,
 252 #endif
 253         .revalidate_disk = cciss_revalidate,
 254 };
 255
 256 /* set_performant_mode: Modify the tag for cciss performant
 257  * set bit 0 for pull model, bits 3-1 for block fetch
 258  * register number
 259  */
 260 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 261 {
 262         if (likely(h->transMethod == CFGTBL_Trans_Performant))
 263                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 264 }
 265
 266 /*
 267  * Enqueuing and dequeuing functions for cmdlists.
 268  */
 269 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 270 {
 271         hlist_add_head(&c->list, list);
 272 }
 273
 274 static inline void removeQ(CommandList_struct *c)
 275 {
 276         /*
 277          * After kexec/dump some commands might still
 278          * be in flight, which the firmware will try
 279          * to complete. Resetting the firmware doesn't work
 280          * with old fw revisions, so we have to mark
 281          * them off as 'stale' to prevent the driver from
 282          * falling over.
 283          */
 284         if (WARN_ON(hlist_unhashed(&c->list))) {
 285                 c->cmd_type = CMD_MSG_STALE;
 286                 return;
 287         }
 288
 289         hlist_del_init(&c->list);
 290 }
 291
 292 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 293         CommandList_struct *c)
 294 {
 295         unsigned long flags;
 296         set_performant_mode(h, c);
 297         spin_lock_irqsave(&h->lock, flags);
 298         addQ(&h->reqQ, c);
 299         h->Qdepth++;
 300         start_io(h);
 301         spin_unlock_irqrestore(&h->lock, flags);
 302 }
 303
 304 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 305         int nr_cmds)
 306 {
 307         int i;
 308
 309         if (!cmd_sg_list)
 310                 return;
 311         for (i = 0; i < nr_cmds; i++) {
 312                 kfree(cmd_sg_list[i]);
 313                 cmd_sg_list[i] = NULL;
 314         }
 315         kfree(cmd_sg_list);
 316 }
 317
 318 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 319         ctlr_info_t *h, int chainsize, int nr_cmds)
 320 {
 321         int j;
 322         SGDescriptor_struct **cmd_sg_list;
 323
 324         if (chainsize <= 0)
 325                 return NULL;
 326
 327         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 328         if (!cmd_sg_list)
 329                 return NULL;
 330
 331         /* Build up chain blocks for each command */
 332         for (j = 0; j < nr_cmds; j++) {
 333                 /* Need a block of chainsized s/g elements. */
 334                 cmd_sg_list[j] = kmalloc((chainsize *
 335                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 336                 if (!cmd_sg_list[j]) {
 337                         dev_err(&h->pdev->dev, "Cannot get memory "
 338                                 "for s/g chains.\n");
 339                         goto clean;
 340                 }
 341         }
 342         return cmd_sg_list;
 343 clean:
 344         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 345         return NULL;
 346 }
 347
 348 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 349 {
 350         SGDescriptor_struct *chain_sg;
 351         u64bit temp64;
 352
 353         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 354                 return;
 355
 356         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 357         temp64.val32.lower = chain_sg->Addr.lower;
 358         temp64.val32.upper = chain_sg->Addr.upper;
 359         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 360 }
 361
 362 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 363         SGDescriptor_struct *chain_block, int len)
 364 {
 365         SGDescriptor_struct *chain_sg;
 366         u64bit temp64;
 367
 368         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 369         chain_sg->Ext = CCISS_SG_CHAIN;
 370         chain_sg->Len = len;
 371         temp64.val = pci_map_single(h->pdev, chain_block, len,
 372                                 PCI_DMA_TODEVICE);
 373         chain_sg->Addr.lower = temp64.val32.lower;
 374         chain_sg->Addr.upper = temp64.val32.upper;
 375 }
 376
 377 #include "cciss_scsi.c"         /* For SCSI tape support */
 378
 379 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 380         "UNKNOWN"
 381 };
 382 #define RAID_UNKNOWN (sizeof(raid_label) / sizeof(raid_label[0])-1)
 383
 384 #ifdef CONFIG_PROC_FS
 385
 386 /*
 387  * Report information about this controller.
 388  */
 389 #define ENG_GIG 1000000000
 390 #define ENG_GIG_FACTOR (ENG_GIG/512)
 391 #define ENGAGE_SCSI     "engage scsi"
 392
 393 static struct proc_dir_entry *proc_cciss;
 394
 395 static void cciss_seq_show_header(struct seq_file *seq)
 396 {
 397         ctlr_info_t *h = seq->private;
 398
 399         seq_printf(seq, "%s: HP %s Controller\n"
 400                 "Board ID: 0x%08lx\n"
 401                 "Firmware Version: %c%c%c%c\n"
 402                 "IRQ: %d\n"
 403                 "Logical drives: %d\n"
 404                 "Current Q depth: %d\n"
 405                 "Current # commands on controller: %d\n"
 406                 "Max Q depth since init: %d\n"
 407                 "Max # commands on controller since init: %d\n"
 408                 "Max SG entries since init: %d\n",
 409                 h->devname,
 410                 h->product_name,
 411                 (unsigned long)h->board_id,
 412                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 413                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 414                 h->num_luns,
 415                 h->Qdepth, h->commands_outstanding,
 416                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 417
 418 #ifdef CONFIG_CISS_SCSI_TAPE
 419         cciss_seq_tape_report(seq, h);
 420 #endif /* CONFIG_CISS_SCSI_TAPE */
 421 }
 422
 423 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 424 {
 425         ctlr_info_t *h = seq->private;
 426         unsigned long flags;
 427
 428         /* prevent displaying bogus info during configuration
 429          * or deconfiguration of a logical volume
 430          */
 431         spin_lock_irqsave(&h->lock, flags);
 432         if (h->busy_configuring) {
 433                 spin_unlock_irqrestore(&h->lock, flags);
 434                 return ERR_PTR(-EBUSY);
 435         }
 436         h->busy_configuring = 1;
 437         spin_unlock_irqrestore(&h->lock, flags);
 438
 439         if (*pos == 0)
 440                 cciss_seq_show_header(seq);
 441
 442         return pos;
 443 }
 444
 445 static int cciss_seq_show(struct seq_file *seq, void *v)
 446 {
 447         sector_t vol_sz, vol_sz_frac;
 448         ctlr_info_t *h = seq->private;
 449         unsigned ctlr = h->ctlr;
 450         loff_t *pos = v;
 451         drive_info_struct *drv = h->drv[*pos];
 452
 453         if (*pos > h->highest_lun)
 454                 return 0;
 455
 456         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 457                 return 0;
 458
 459         if (drv->heads == 0)
 460                 return 0;
 461
 462         vol_sz = drv->nr_blocks;
 463         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 464         vol_sz_frac *= 100;
 465         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 466
 467         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 468                 drv->raid_level = RAID_UNKNOWN;
 469         seq_printf(seq, "cciss/c%dd%d:"
 470                         "\t%4u.%02uGB\tRAID %s\n",
 471                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 472                         raid_label[drv->raid_level]);
 473         return 0;
 474 }
 475
 476 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 477 {
 478         ctlr_info_t *h = seq->private;
 479
 480         if (*pos > h->highest_lun)
 481                 return NULL;
 482         *pos += 1;
 483
 484         return pos;
 485 }
 486
 487 static void cciss_seq_stop(struct seq_file *seq, void *v)
 488 {
 489         ctlr_info_t *h = seq->private;
 490
 491         /* Only reset h->busy_configuring if we succeeded in setting
 492          * it during cciss_seq_start. */
 493         if (v == ERR_PTR(-EBUSY))
 494                 return;
 495
 496         h->busy_configuring = 0;
 497 }
 498
 499 static const struct seq_operations cciss_seq_ops = {
 500         .start = cciss_seq_start,
 501         .show  = cciss_seq_show,
 502         .next  = cciss_seq_next,
 503         .stop  = cciss_seq_stop,
 504 };
 505
 506 static int cciss_seq_open(struct inode *inode, struct file *file)
 507 {
 508         int ret = seq_open(file, &cciss_seq_ops);
 509         struct seq_file *seq = file->private_data;
 510
 511         if (!ret)
 512                 seq->private = PDE(inode)->data;
 513
 514         return ret;
 515 }
 516
 517 static ssize_t
 518 cciss_proc_write(struct file *file, const char __user *buf,
 519                  size_t length, loff_t *ppos)
 520 {
 521         int err;
 522         char *buffer;
 523
 524 #ifndef CONFIG_CISS_SCSI_TAPE
 525         return -EINVAL;
 526 #endif
 527
 528         if (!buf || length > PAGE_SIZE - 1)
 529                 return -EINVAL;
 530
 531         buffer = (char *)__get_free_page(GFP_KERNEL);
 532         if (!buffer)
 533                 return -ENOMEM;
 534
 535         err = -EFAULT;
 536         if (copy_from_user(buffer, buf, length))
 537                 goto out;
 538         buffer[length] = '\0';
 539
 540 #ifdef CONFIG_CISS_SCSI_TAPE
 541         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 542                 struct seq_file *seq = file->private_data;
 543                 ctlr_info_t *h = seq->private;
 544
 545                 err = cciss_engage_scsi(h);
 546                 if (err == 0)
 547                         err = length;
 548         } else
 549 #endif /* CONFIG_CISS_SCSI_TAPE */
 550                 err = -EINVAL;
 551         /* might be nice to have "disengage" too, but it's not
 552            safely possible. (only 1 module use count, lock issues.) */
 553
 554 out:
 555         free_page((unsigned long)buffer);
 556         return err;
 557 }
 558
 559 static const struct file_operations cciss_proc_fops = {
 560         .owner   = THIS_MODULE,
 561         .open    = cciss_seq_open,
 562         .read    = seq_read,
 563         .llseek  = seq_lseek,
 564         .release = seq_release,
 565         .write   = cciss_proc_write,
 566 };
 567
 568 static void __devinit cciss_procinit(ctlr_info_t *h)
 569 {
 570         struct proc_dir_entry *pde;
 571
 572         if (proc_cciss == NULL)
 573                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 574         if (!proc_cciss)
 575                 return;
 576         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 577                                         S_IROTH, proc_cciss,
 578                                         &cciss_proc_fops, h);
 579 }
 580 #endif                          /* CONFIG_PROC_FS */
 581
 582 #define MAX_PRODUCT_NAME_LEN 19
 583
 584 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 585 #define to_drv(n) container_of(n, drive_info_struct, dev)
 586
 587 static ssize_t host_store_rescan(struct device *dev,
 588                                  struct device_attribute *attr,
 589                                  const char *buf, size_t count)
 590 {
 591         struct ctlr_info *h = to_hba(dev);
 592
 593         add_to_scan_list(h);
 594         wake_up_process(cciss_scan_thread);
 595         wait_for_completion_interruptible(&h->scan_wait);
 596
 597         return count;
 598 }
 599 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 600
 601 static ssize_t dev_show_unique_id(struct device *dev,
 602                                  struct device_attribute *attr,
 603                                  char *buf)
 604 {
 605         drive_info_struct *drv = to_drv(dev);
 606         struct ctlr_info *h = to_hba(drv->dev.parent);
 607         __u8 sn[16];
 608         unsigned long flags;
 609         int ret = 0;
 610
 611         spin_lock_irqsave(&h->lock, flags);
 612         if (h->busy_configuring)
 613                 ret = -EBUSY;
 614         else
 615                 memcpy(sn, drv->serial_no, sizeof(sn));
 616         spin_unlock_irqrestore(&h->lock, flags);
 617
 618         if (ret)
 619                 return ret;
 620         else
 621                 return snprintf(buf, 16 * 2 + 2,
 622                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 623                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 624                                 sn[0], sn[1], sn[2], sn[3],
 625                                 sn[4], sn[5], sn[6], sn[7],
 626                                 sn[8], sn[9], sn[10], sn[11],
 627                                 sn[12], sn[13], sn[14], sn[15]);
 628 }
 629 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 630
 631 static ssize_t dev_show_vendor(struct device *dev,
 632                                struct device_attribute *attr,
 633                                char *buf)
 634 {
 635         drive_info_struct *drv = to_drv(dev);
 636         struct ctlr_info *h = to_hba(drv->dev.parent);
 637         char vendor[VENDOR_LEN + 1];
 638         unsigned long flags;
 639         int ret = 0;
 640
 641         spin_lock_irqsave(&h->lock, flags);
 642         if (h->busy_configuring)
 643                 ret = -EBUSY;
 644         else
 645                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 646         spin_unlock_irqrestore(&h->lock, flags);
 647
 648         if (ret)
 649                 return ret;
 650         else
 651                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 652 }
 653 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 654
 655 static ssize_t dev_show_model(struct device *dev,
 656                               struct device_attribute *attr,
 657                               char *buf)
 658 {
 659         drive_info_struct *drv = to_drv(dev);
 660         struct ctlr_info *h = to_hba(drv->dev.parent);
 661         char model[MODEL_LEN + 1];
 662         unsigned long flags;
 663         int ret = 0;
 664
 665         spin_lock_irqsave(&h->lock, flags);
 666         if (h->busy_configuring)
 667                 ret = -EBUSY;
 668         else
 669                 memcpy(model, drv->model, MODEL_LEN + 1);
 670         spin_unlock_irqrestore(&h->lock, flags);
 671
 672         if (ret)
 673                 return ret;
 674         else
 675                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 676 }
 677 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 678
 679 static ssize_t dev_show_rev(struct device *dev,
 680                             struct device_attribute *attr,
 681                             char *buf)
 682 {
 683         drive_info_struct *drv = to_drv(dev);
 684         struct ctlr_info *h = to_hba(drv->dev.parent);
 685         char rev[REV_LEN + 1];
 686         unsigned long flags;
 687         int ret = 0;
 688
 689         spin_lock_irqsave(&h->lock, flags);
 690         if (h->busy_configuring)
 691                 ret = -EBUSY;
 692         else
 693                 memcpy(rev, drv->rev, REV_LEN + 1);
 694         spin_unlock_irqrestore(&h->lock, flags);
 695
 696         if (ret)
 697                 return ret;
 698         else
 699                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 700 }
 701 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 702
 703 static ssize_t cciss_show_lunid(struct device *dev,
 704                                 struct device_attribute *attr, char *buf)
 705 {
 706         drive_info_struct *drv = to_drv(dev);
 707         struct ctlr_info *h = to_hba(drv->dev.parent);
 708         unsigned long flags;
 709         unsigned char lunid[8];
 710
 711         spin_lock_irqsave(&h->lock, flags);
 712         if (h->busy_configuring) {
 713                 spin_unlock_irqrestore(&h->lock, flags);
 714                 return -EBUSY;
 715         }
 716         if (!drv->heads) {
 717                 spin_unlock_irqrestore(&h->lock, flags);
 718                 return -ENOTTY;
 719         }
 720         memcpy(lunid, drv->LunID, sizeof(lunid));
 721         spin_unlock_irqrestore(&h->lock, flags);
 722         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 723                 lunid[0], lunid[1], lunid[2], lunid[3],
 724                 lunid[4], lunid[5], lunid[6], lunid[7]);
 725 }
 726 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 727
 728 static ssize_t cciss_show_raid_level(struct device *dev,
 729                                      struct device_attribute *attr, char *buf)
 730 {
 731         drive_info_struct *drv = to_drv(dev);
 732         struct ctlr_info *h = to_hba(drv->dev.parent);
 733         int raid;
 734         unsigned long flags;
 735
 736         spin_lock_irqsave(&h->lock, flags);
 737         if (h->busy_configuring) {
 738                 spin_unlock_irqrestore(&h->lock, flags);
 739                 return -EBUSY;
 740         }
 741         raid = drv->raid_level;
 742         spin_unlock_irqrestore(&h->lock, flags);
 743         if (raid < 0 || raid > RAID_UNKNOWN)
 744                 raid = RAID_UNKNOWN;
 745
 746         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 747                         raid_label[raid]);
 748 }
 749 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 750
 751 static ssize_t cciss_show_usage_count(struct device *dev,
 752                                       struct device_attribute *attr, char *buf)
 753 {
 754         drive_info_struct *drv = to_drv(dev);
 755         struct ctlr_info *h = to_hba(drv->dev.parent);
 756         unsigned long flags;
 757         int count;
 758
 759         spin_lock_irqsave(&h->lock, flags);
 760         if (h->busy_configuring) {
 761                 spin_unlock_irqrestore(&h->lock, flags);
 762                 return -EBUSY;
 763         }
 764         count = drv->usage_count;
 765         spin_unlock_irqrestore(&h->lock, flags);
 766         return snprintf(buf, 20, "%d\n", count);
 767 }
 768 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 769
 770 static struct attribute *cciss_host_attrs[] = {
 771         &dev_attr_rescan.attr,
 772         NULL
 773 };
 774
 775 static struct attribute_group cciss_host_attr_group = {
 776         .attrs = cciss_host_attrs,
 777 };
 778
 779 static const struct attribute_group *cciss_host_attr_groups[] = {
 780         &cciss_host_attr_group,
 781         NULL
 782 };
 783
 784 static struct device_type cciss_host_type = {
 785         .name           = "cciss_host",
 786         .groups         = cciss_host_attr_groups,
 787         .release        = cciss_hba_release,
 788 };
 789
 790 static struct attribute *cciss_dev_attrs[] = {
 791         &dev_attr_unique_id.attr,
 792         &dev_attr_model.attr,
 793         &dev_attr_vendor.attr,
 794         &dev_attr_rev.attr,
 795         &dev_attr_lunid.attr,
 796         &dev_attr_raid_level.attr,
 797         &dev_attr_usage_count.attr,
 798         NULL
 799 };
 800
 801 static struct attribute_group cciss_dev_attr_group = {
 802         .attrs = cciss_dev_attrs,
 803 };
 804
 805 static const struct attribute_group *cciss_dev_attr_groups[] = {
 806         &cciss_dev_attr_group,
 807         NULL
 808 };
 809
 810 static struct device_type cciss_dev_type = {
 811         .name           = "cciss_device",
 812         .groups         = cciss_dev_attr_groups,
 813         .release        = cciss_device_release,
 814 };
 815
 816 static struct bus_type cciss_bus_type = {
 817         .name           = "cciss",
 818 };
 819
 820 /*
 821  * cciss_hba_release is called when the reference count
 822  * of h->dev goes to zero.
 823  */
 824 static void cciss_hba_release(struct device *dev)
 825 {
 826         /*
 827          * nothing to do, but need this to avoid a warning
 828          * about not having a release handler from lib/kref.c.
 829          */
 830 }
 831
 832 /*
 833  * Initialize sysfs entry for each controller.  This sets up and registers
 834  * the 'cciss#' directory for each individual controller under
 835  * /sys/bus/pci/devices/<dev>/.
 836  */
 837 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 838 {
 839         device_initialize(&h->dev);
 840         h->dev.type = &cciss_host_type;
 841         h->dev.bus = &cciss_bus_type;
 842         dev_set_name(&h->dev, "%s", h->devname);
 843         h->dev.parent = &h->pdev->dev;
 844
 845         return device_add(&h->dev);
 846 }
 847
 848 /*
 849  * Remove sysfs entries for an hba.
 850  */
 851 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 852 {
 853         device_del(&h->dev);
 854         put_device(&h->dev); /* final put. */
 855 }
 856
 857 /* cciss_device_release is called when the reference count
 858  * of h->drv[x]dev goes to zero.
 859  */
 860 static void cciss_device_release(struct device *dev)
 861 {
 862         drive_info_struct *drv = to_drv(dev);
 863         kfree(drv);
 864 }
 865
 866 /*
 867  * Initialize sysfs for each logical drive.  This sets up and registers
 868  * the 'c#d#' directory for each individual logical drive under
 869  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 870  * /sys/block/cciss!c#d# to this entry.
 871  */
 872 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 873                                        int drv_index)
 874 {
 875         struct device *dev;
 876
 877         if (h->drv[drv_index]->device_initialized)
 878                 return 0;
 879
 880         dev = &h->drv[drv_index]->dev;
 881         device_initialize(dev);
 882         dev->type = &cciss_dev_type;
 883         dev->bus = &cciss_bus_type;
 884         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 885         dev->parent = &h->dev;
 886         h->drv[drv_index]->device_initialized = 1;
 887         return device_add(dev);
 888 }
 889
 890 /*
 891  * Remove sysfs entries for a logical drive.
 892  */
 893 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 894         int ctlr_exiting)
 895 {
 896         struct device *dev = &h->drv[drv_index]->dev;
 897
 898         /* special case for c*d0, we only destroy it on controller exit */
 899         if (drv_index == 0 && !ctlr_exiting)
 900                 return;
 901
 902         device_del(dev);
 903         put_device(dev); /* the "final" put. */
 904         h->drv[drv_index] = NULL;
 905 }
 906
 907 /*
 908  * For operations that cannot sleep, a command block is allocated at init,
 909  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 910  * which ones are free or in use.  For operations that can wait for kmalloc
 911  * to possible sleep, this routine can be called with get_from_pool set to 0.
 912  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 913  */
 914 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 915 {
 916         CommandList_struct *c;
 917         int i;
 918         u64bit temp64;
 919         dma_addr_t cmd_dma_handle, err_dma_handle;
 920
 921         if (!get_from_pool) {
 922                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 923                         sizeof(CommandList_struct), &cmd_dma_handle);
 924                 if (c == NULL)
 925                         return NULL;
 926                 memset(c, 0, sizeof(CommandList_struct));
 927
 928                 c->cmdindex = -1;
 929
 930                 c->err_info = (ErrorInfo_struct *)
 931                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 932                             &err_dma_handle);
 933
 934                 if (c->err_info == NULL) {
 935                         pci_free_consistent(h->pdev,
 936                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 937                         return NULL;
 938                 }
 939                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 940         } else {                /* get it out of the controllers pool */
 941
 942                 do {
 943                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 944                         if (i == h->nr_cmds)
 945                                 return NULL;
 946                 } while (test_and_set_bit
 947                          (i & (BITS_PER_LONG - 1),
 948                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 949 #ifdef CCISS_DEBUG
 950                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 951 #endif
 952                 c = h->cmd_pool + i;
 953                 memset(c, 0, sizeof(CommandList_struct));
 954                 cmd_dma_handle = h->cmd_pool_dhandle
 955                     + i * sizeof(CommandList_struct);
 956                 c->err_info = h->errinfo_pool + i;
 957                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 958                 err_dma_handle = h->errinfo_pool_dhandle
 959                     + i * sizeof(ErrorInfo_struct);
 960                 h->nr_allocs++;
 961
 962                 c->cmdindex = i;
 963         }
 964
 965         INIT_HLIST_NODE(&c->list);
 966         c->busaddr = (__u32) cmd_dma_handle;
 967         temp64.val = (__u64) err_dma_handle;
 968         c->ErrDesc.Addr.lower = temp64.val32.lower;
 969         c->ErrDesc.Addr.upper = temp64.val32.upper;
 970         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 971
 972         c->ctlr = h->ctlr;
 973         return c;
 974 }
 975
 976 /*
 977  * Frees a command block that was previously allocated with cmd_alloc().
 978  */
 979 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 980 {
 981         int i;
 982         u64bit temp64;
 983
 984         if (!got_from_pool) {
 985                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 986                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 987                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 988                                     c->err_info, (dma_addr_t) temp64.val);
 989                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 990                                     c, (dma_addr_t) c->busaddr);
 991         } else {
 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
 999 static inline ctlr_info_t *get_host(struct gendisk *disk)
1000 {
1001         return disk->queue->queuedata;
1002 }
1003
1004 static inline drive_info_struct *get_drv(struct gendisk *disk)
1005 {
1006         return disk->private_data;
1007 }
1008
1009 /*
1010  * Open.  Make sure the device is really there.
1011  */
1012 static int cciss_open(struct block_device *bdev, fmode_t mode)
1013 {
1014         ctlr_info_t *h = get_host(bdev->bd_disk);
1015         drive_info_struct *drv = get_drv(bdev->bd_disk);
1016
1017 #ifdef CCISS_DEBUG
1018         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
1019 #endif                          /* CCISS_DEBUG */
1020
1021         if (drv->busy_configuring)
1022                 return -EBUSY;
1023         /*
1024          * Root is allowed to open raw volume zero even if it's not configured
1025          * so array config can still work. Root is also allowed to open any
1026          * volume that has a LUN ID, so it can issue IOCTL to reread the
1027          * disk information.  I don't think I really like this
1028          * but I'm already using way to many device nodes to claim another one
1029          * for "raw controller".
1030          */
1031         if (drv->heads == 0) {
1032                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1033                         /* if not node 0 make sure it is a partition = 0 */
1034                         if (MINOR(bdev->bd_dev) & 0x0f) {
1035                                 return -ENXIO;
1036                                 /* if it is, make sure we have a LUN ID */
1037                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1038                                 sizeof(drv->LunID))) {
1039                                 return -ENXIO;
1040                         }
1041                 }
1042                 if (!capable(CAP_SYS_ADMIN))
1043                         return -EPERM;
1044         }
1045         drv->usage_count++;
1046         h->usage_count++;
1047         return 0;
1048 }
1049
1050 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1051 {
1052         int ret;
1053
1054         lock_kernel();
1055         ret = cciss_open(bdev, mode);
1056         unlock_kernel();
1057
1058         return ret;
1059 }
1060
1061 /*
1062  * Close.  Sync first.
1063  */
1064 static int cciss_release(struct gendisk *disk, fmode_t mode)
1065 {
1066         ctlr_info_t *h;
1067         drive_info_struct *drv;
1068
1069         lock_kernel();
1070         h = get_host(disk);
1071         drv = get_drv(disk);
1072
1073 #ifdef CCISS_DEBUG
1074         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
1075 #endif                          /* CCISS_DEBUG */
1076
1077         drv->usage_count--;
1078         h->usage_count--;
1079         unlock_kernel();
1080         return 0;
1081 }
1082
1083 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1084                     unsigned cmd, unsigned long arg)
1085 {
1086         int ret;
1087         lock_kernel();
1088         ret = cciss_ioctl(bdev, mode, cmd, arg);
1089         unlock_kernel();
1090         return ret;
1091 }
1092
1093 #ifdef CONFIG_COMPAT
1094
1095 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1096                                   unsigned cmd, unsigned long arg);
1097 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1098                                       unsigned cmd, unsigned long arg);
1099
1100 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1101                               unsigned cmd, unsigned long arg)
1102 {
1103         switch (cmd) {
1104         case CCISS_GETPCIINFO:
1105         case CCISS_GETINTINFO:
1106         case CCISS_SETINTINFO:
1107         case CCISS_GETNODENAME:
1108         case CCISS_SETNODENAME:
1109         case CCISS_GETHEARTBEAT:
1110         case CCISS_GETBUSTYPES:
1111         case CCISS_GETFIRMVER:
1112         case CCISS_GETDRIVVER:
1113         case CCISS_REVALIDVOLS:
1114         case CCISS_DEREGDISK:
1115         case CCISS_REGNEWDISK:
1116         case CCISS_REGNEWD:
1117         case CCISS_RESCANDISK:
1118         case CCISS_GETLUNINFO:
1119                 return do_ioctl(bdev, mode, cmd, arg);
1120
1121         case CCISS_PASSTHRU32:
1122                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1123         case CCISS_BIG_PASSTHRU32:
1124                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1125
1126         default:
1127                 return -ENOIOCTLCMD;
1128         }
1129 }
1130
1131 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1132                                   unsigned cmd, unsigned long arg)
1133 {
1134         IOCTL32_Command_struct __user *arg32 =
1135             (IOCTL32_Command_struct __user *) arg;
1136         IOCTL_Command_struct arg64;
1137         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1138         int err;
1139         u32 cp;
1140
1141         err = 0;
1142         err |=
1143             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1144                            sizeof(arg64.LUN_info));
1145         err |=
1146             copy_from_user(&arg64.Request, &arg32->Request,
1147                            sizeof(arg64.Request));
1148         err |=
1149             copy_from_user(&arg64.error_info, &arg32->error_info,
1150                            sizeof(arg64.error_info));
1151         err |= get_user(arg64.buf_size, &arg32->buf_size);
1152         err |= get_user(cp, &arg32->buf);
1153         arg64.buf = compat_ptr(cp);
1154         err |= copy_to_user(p, &arg64, sizeof(arg64));
1155
1156         if (err)
1157                 return -EFAULT;
1158
1159         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1160         if (err)
1161                 return err;
1162         err |=
1163             copy_in_user(&arg32->error_info, &p->error_info,
1164                          sizeof(arg32->error_info));
1165         if (err)
1166                 return -EFAULT;
1167         return err;
1168 }
1169
1170 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1171                                       unsigned cmd, unsigned long arg)
1172 {
1173         BIG_IOCTL32_Command_struct __user *arg32 =
1174             (BIG_IOCTL32_Command_struct __user *) arg;
1175         BIG_IOCTL_Command_struct arg64;
1176         BIG_IOCTL_Command_struct __user *p =
1177             compat_alloc_user_space(sizeof(arg64));
1178         int err;
1179         u32 cp;
1180
1181         err = 0;
1182         err |=
1183             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1184                            sizeof(arg64.LUN_info));
1185         err |=
1186             copy_from_user(&arg64.Request, &arg32->Request,
1187                            sizeof(arg64.Request));
1188         err |=
1189             copy_from_user(&arg64.error_info, &arg32->error_info,
1190                            sizeof(arg64.error_info));
1191         err |= get_user(arg64.buf_size, &arg32->buf_size);
1192         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1193         err |= get_user(cp, &arg32->buf);
1194         arg64.buf = compat_ptr(cp);
1195         err |= copy_to_user(p, &arg64, sizeof(arg64));
1196
1197         if (err)
1198                 return -EFAULT;
1199
1200         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1201         if (err)
1202                 return err;
1203         err |=
1204             copy_in_user(&arg32->error_info, &p->error_info,
1205                          sizeof(arg32->error_info));
1206         if (err)
1207                 return -EFAULT;
1208         return err;
1209 }
1210 #endif
1211
1212 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1213 {
1214         drive_info_struct *drv = get_drv(bdev->bd_disk);
1215
1216         if (!drv->cylinders)
1217                 return -ENXIO;
1218
1219         geo->heads = drv->heads;
1220         geo->sectors = drv->sectors;
1221         geo->cylinders = drv->cylinders;
1222         return 0;
1223 }
1224
1225 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1226 {
1227         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1228                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1229                 (void)check_for_unit_attention(h, c);
1230 }
1231 /*
1232  * ioctl
1233  */
1234 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1235                        unsigned int cmd, unsigned long arg)
1236 {
1237         struct gendisk *disk = bdev->bd_disk;
1238         ctlr_info_t *h = get_host(disk);
1239         drive_info_struct *drv = get_drv(disk);
1240         void __user *argp = (void __user *)arg;
1241
1242 #ifdef CCISS_DEBUG
1243         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1244 #endif                          /* CCISS_DEBUG */
1245
1246         switch (cmd) {
1247         case CCISS_GETPCIINFO:
1248                 {
1249                         cciss_pci_info_struct pciinfo;
1250
1251                         if (!arg)
1252                                 return -EINVAL;
1253                         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1254                         pciinfo.bus = h->pdev->bus->number;
1255                         pciinfo.dev_fn = h->pdev->devfn;
1256                         pciinfo.board_id = h->board_id;
1257                         if (copy_to_user
1258                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1259                                 return -EFAULT;
1260                         return 0;
1261                 }
1262         case CCISS_GETINTINFO:
1263                 {
1264                         cciss_coalint_struct intinfo;
1265                         if (!arg)
1266                                 return -EINVAL;
1267                         intinfo.delay =
1268                             readl(&h->cfgtable->HostWrite.CoalIntDelay);
1269                         intinfo.count =
1270                             readl(&h->cfgtable->HostWrite.CoalIntCount);
1271                         if (copy_to_user
1272                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1273                                 return -EFAULT;
1274                         return 0;
1275                 }
1276         case CCISS_SETINTINFO:
1277                 {
1278                         cciss_coalint_struct intinfo;
1279                         unsigned long flags;
1280                         int i;
1281
1282                         if (!arg)
1283                                 return -EINVAL;
1284                         if (!capable(CAP_SYS_ADMIN))
1285                                 return -EPERM;
1286                         if (copy_from_user
1287                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1288                                 return -EFAULT;
1289                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1290                         {
1291 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1292                                 return -EINVAL;
1293                         }
1294                         spin_lock_irqsave(&h->lock, flags);
1295                         /* Update the field, and then ring the doorbell */
1296                         writel(intinfo.delay,
1297                                &(h->cfgtable->HostWrite.CoalIntDelay));
1298                         writel(intinfo.count,
1299                                &(h->cfgtable->HostWrite.CoalIntCount));
1300                         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1301
1302                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1303                                 if (!(readl(h->vaddr + SA5_DOORBELL)
1304                                       & CFGTBL_ChangeReq))
1305                                         break;
1306                                 /* delay and try again */
1307                                 udelay(1000);
1308                         }
1309                         spin_unlock_irqrestore(&h->lock, flags);
1310                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1311                                 return -EAGAIN;
1312                         return 0;
1313                 }
1314         case CCISS_GETNODENAME:
1315                 {
1316                         NodeName_type NodeName;
1317                         int i;
1318
1319                         if (!arg)
1320                                 return -EINVAL;
1321                         for (i = 0; i < 16; i++)
1322                                 NodeName[i] =
1323                                     readb(&h->cfgtable->ServerName[i]);
1324                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1325                                 return -EFAULT;
1326                         return 0;
1327                 }
1328         case CCISS_SETNODENAME:
1329                 {
1330                         NodeName_type NodeName;
1331                         unsigned long flags;
1332                         int i;
1333
1334                         if (!arg)
1335                                 return -EINVAL;
1336                         if (!capable(CAP_SYS_ADMIN))
1337                                 return -EPERM;
1338
1339                         if (copy_from_user
1340                             (NodeName, argp, sizeof(NodeName_type)))
1341                                 return -EFAULT;
1342
1343                         spin_lock_irqsave(&h->lock, flags);
1344
1345                         /* Update the field, and then ring the doorbell */
1346                         for (i = 0; i < 16; i++)
1347                                 writeb(NodeName[i],
1348                                        &h->cfgtable->ServerName[i]);
1349
1350                         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1351
1352                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1353                                 if (!(readl(h->vaddr + SA5_DOORBELL)
1354                                       & CFGTBL_ChangeReq))
1355                                         break;
1356                                 /* delay and try again */
1357                                 udelay(1000);
1358                         }
1359                         spin_unlock_irqrestore(&h->lock, flags);
1360                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1361                                 return -EAGAIN;
1362                         return 0;
1363                 }
1364
1365         case CCISS_GETHEARTBEAT:
1366                 {
1367                         Heartbeat_type heartbeat;
1368
1369                         if (!arg)
1370                                 return -EINVAL;
1371                         heartbeat = readl(&h->cfgtable->HeartBeat);
1372                         if (copy_to_user
1373                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1374                                 return -EFAULT;
1375                         return 0;
1376                 }
1377         case CCISS_GETBUSTYPES:
1378                 {
1379                         BusTypes_type BusTypes;
1380
1381                         if (!arg)
1382                                 return -EINVAL;
1383                         BusTypes = readl(&h->cfgtable->BusTypes);
1384                         if (copy_to_user
1385                             (argp, &BusTypes, sizeof(BusTypes_type)))
1386                                 return -EFAULT;
1387                         return 0;
1388                 }
1389         case CCISS_GETFIRMVER:
1390                 {
1391                         FirmwareVer_type firmware;
1392
1393                         if (!arg)
1394                                 return -EINVAL;
1395                         memcpy(firmware, h->firm_ver, 4);
1396
1397                         if (copy_to_user
1398                             (argp, firmware, sizeof(FirmwareVer_type)))
1399                                 return -EFAULT;
1400                         return 0;
1401                 }
1402         case CCISS_GETDRIVVER:
1403                 {
1404                         DriverVer_type DriverVer = DRIVER_VERSION;
1405
1406                         if (!arg)
1407                                 return -EINVAL;
1408
1409                         if (copy_to_user
1410                             (argp, &DriverVer, sizeof(DriverVer_type)))
1411                                 return -EFAULT;
1412                         return 0;
1413                 }
1414
1415         case CCISS_DEREGDISK:
1416         case CCISS_REGNEWD:
1417         case CCISS_REVALIDVOLS:
1418                 return rebuild_lun_table(h, 0, 1);
1419
1420         case CCISS_GETLUNINFO:{
1421                         LogvolInfo_struct luninfo;
1422
1423                         memcpy(&luninfo.LunID, drv->LunID,
1424                                 sizeof(luninfo.LunID));
1425                         luninfo.num_opens = drv->usage_count;
1426                         luninfo.num_parts = 0;
1427                         if (copy_to_user(argp, &luninfo,
1428                                          sizeof(LogvolInfo_struct)))
1429                                 return -EFAULT;
1430                         return 0;
1431                 }
1432         case CCISS_PASSTHRU:
1433                 {
1434                         IOCTL_Command_struct iocommand;
1435                         CommandList_struct *c;
1436                         char *buff = NULL;
1437                         u64bit temp64;
1438                         DECLARE_COMPLETION_ONSTACK(wait);
1439
1440                         if (!arg)
1441                                 return -EINVAL;
1442
1443                         if (!capable(CAP_SYS_RAWIO))
1444                                 return -EPERM;
1445
1446                         if (copy_from_user
1447                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1448                                 return -EFAULT;
1449                         if ((iocommand.buf_size < 1) &&
1450                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1451                                 return -EINVAL;
1452                         }
1453 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1454                         /* Check kmalloc limits */
1455                         if (iocommand.buf_size > 128000)
1456                                 return -EINVAL;
1457 #endif
1458                         if (iocommand.buf_size > 0) {
1459                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1460                                 if (buff == NULL)
1461                                         return -EFAULT;
1462                         }
1463                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1464                                 /* Copy the data into the buffer we created */
1465                                 if (copy_from_user
1466                                     (buff, iocommand.buf, iocommand.buf_size)) {
1467                                         kfree(buff);
1468                                         return -EFAULT;
1469                                 }
1470                         } else {
1471                                 memset(buff, 0, iocommand.buf_size);
1472                         }
1473                         c = cmd_alloc(h, 0);
1474                         if (!c) {
1475                                 kfree(buff);
1476                                 return -ENOMEM;
1477                         }
1478                         /* Fill in the command type */
1479                         c->cmd_type = CMD_IOCTL_PEND;
1480                         /* Fill in Command Header */
1481                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1482                         if (iocommand.buf_size > 0) /* buffer to fill */
1483                         {
1484                                 c->Header.SGList = 1;
1485                                 c->Header.SGTotal = 1;
1486                         } else /* no buffers to fill */
1487                         {
1488                                 c->Header.SGList = 0;
1489                                 c->Header.SGTotal = 0;
1490                         }
1491                         c->Header.LUN = iocommand.LUN_info;
1492                         /* use the kernel address the cmd block for tag */
1493                         c->Header.Tag.lower = c->busaddr;
1494
1495                         /* Fill in Request block */
1496                         c->Request = iocommand.Request;
1497
1498                         /* Fill in the scatter gather information */
1499                         if (iocommand.buf_size > 0) {
1500                                 temp64.val = pci_map_single(h->pdev, buff,
1501                                         iocommand.buf_size,
1502                                         PCI_DMA_BIDIRECTIONAL);
1503                                 c->SG[0].Addr.lower = temp64.val32.lower;
1504                                 c->SG[0].Addr.upper = temp64.val32.upper;
1505                                 c->SG[0].Len = iocommand.buf_size;
1506                                 c->SG[0].Ext = 0;  /* we are not chaining */
1507                         }
1508                         c->waiting = &wait;
1509
1510                         enqueue_cmd_and_start_io(h, c);
1511                         wait_for_completion(&wait);
1512
1513                         /* unlock the buffers from DMA */
1514                         temp64.val32.lower = c->SG[0].Addr.lower;
1515                         temp64.val32.upper = c->SG[0].Addr.upper;
1516                         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val,
1517                                          iocommand.buf_size,
1518                                          PCI_DMA_BIDIRECTIONAL);
1519
1520                         check_ioctl_unit_attention(h, c);
1521
1522                         /* Copy the error information out */
1523                         iocommand.error_info = *(c->err_info);
1524                         if (copy_to_user
1525                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1526                                 kfree(buff);
1527                                 cmd_free(h, c, 0);
1528                                 return -EFAULT;
1529                         }
1530
1531                         if (iocommand.Request.Type.Direction == XFER_READ) {
1532                                 /* Copy the data out of the buffer we created */
1533                                 if (copy_to_user
1534                                     (iocommand.buf, buff, iocommand.buf_size)) {
1535                                         kfree(buff);
1536                                         cmd_free(h, c, 0);
1537                                         return -EFAULT;
1538                                 }
1539                         }
1540                         kfree(buff);
1541                         cmd_free(h, c, 0);
1542                         return 0;
1543                 }
1544         case CCISS_BIG_PASSTHRU:{
1545                         BIG_IOCTL_Command_struct *ioc;
1546                         CommandList_struct *c;
1547                         unsigned char **buff = NULL;
1548                         int *buff_size = NULL;
1549                         u64bit temp64;
1550                         BYTE sg_used = 0;
1551                         int status = 0;
1552                         int i;
1553                         DECLARE_COMPLETION_ONSTACK(wait);
1554                         __u32 left;
1555                         __u32 sz;
1556                         BYTE __user *data_ptr;
1557
1558                         if (!arg)
1559                                 return -EINVAL;
1560                         if (!capable(CAP_SYS_RAWIO))
1561                                 return -EPERM;
1562                         ioc = (BIG_IOCTL_Command_struct *)
1563                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1564                         if (!ioc) {
1565                                 status = -ENOMEM;
1566                                 goto cleanup1;
1567                         }
1568                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1569                                 status = -EFAULT;
1570                                 goto cleanup1;
1571                         }
1572                         if ((ioc->buf_size < 1) &&
1573                             (ioc->Request.Type.Direction != XFER_NONE)) {
1574                                 status = -EINVAL;
1575                                 goto cleanup1;
1576                         }
1577                         /* Check kmalloc limits  using all SGs */
1578                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1579                                 status = -EINVAL;
1580                                 goto cleanup1;
1581                         }
1582                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1583                                 status = -EINVAL;
1584                                 goto cleanup1;
1585                         }
1586                         buff =
1587                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1588                         if (!buff) {
1589                                 status = -ENOMEM;
1590                                 goto cleanup1;
1591                         }
1592                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1593                                                    GFP_KERNEL);
1594                         if (!buff_size) {
1595                                 status = -ENOMEM;
1596                                 goto cleanup1;
1597                         }
1598                         left = ioc->buf_size;
1599                         data_ptr = ioc->buf;
1600                         while (left) {
1601                                 sz = (left >
1602                                       ioc->malloc_size) ? ioc->
1603                                     malloc_size : left;
1604                                 buff_size[sg_used] = sz;
1605                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1606                                 if (buff[sg_used] == NULL) {
1607                                         status = -ENOMEM;
1608                                         goto cleanup1;
1609                                 }
1610                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1611                                         if (copy_from_user
1612                                             (buff[sg_used], data_ptr, sz)) {
1613                                                 status = -EFAULT;
1614                                                 goto cleanup1;
1615                                         }
1616                                 } else {
1617                                         memset(buff[sg_used], 0, sz);
1618                                 }
1619                                 left -= sz;
1620                                 data_ptr += sz;
1621                                 sg_used++;
1622                         }
1623                         c = cmd_alloc(h, 0);
1624                         if (!c) {
1625                                 status = -ENOMEM;
1626                                 goto cleanup1;
1627                         }
1628                         c->cmd_type = CMD_IOCTL_PEND;
1629                         c->Header.ReplyQueue = 0;
1630
1631                         if (ioc->buf_size > 0) {
1632                                 c->Header.SGList = sg_used;
1633                                 c->Header.SGTotal = sg_used;
1634                         } else {
1635                                 c->Header.SGList = 0;
1636                                 c->Header.SGTotal = 0;
1637                         }
1638                         c->Header.LUN = ioc->LUN_info;
1639                         c->Header.Tag.lower = c->busaddr;
1640
1641                         c->Request = ioc->Request;
1642                         if (ioc->buf_size > 0) {
1643                                 for (i = 0; i < sg_used; i++) {
1644                                         temp64.val =
1645                                             pci_map_single(h->pdev, buff[i],
1646                                                     buff_size[i],
1647                                                     PCI_DMA_BIDIRECTIONAL);
1648                                         c->SG[i].Addr.lower =
1649                                             temp64.val32.lower;
1650                                         c->SG[i].Addr.upper =
1651                                             temp64.val32.upper;
1652                                         c->SG[i].Len = buff_size[i];
1653                                         c->SG[i].Ext = 0;       /* we are not chaining */
1654                                 }
1655                         }
1656                         c->waiting = &wait;
1657                         enqueue_cmd_and_start_io(h, c);
1658                         wait_for_completion(&wait);
1659                         /* unlock the buffers from DMA */
1660                         for (i = 0; i < sg_used; i++) {
1661                                 temp64.val32.lower = c->SG[i].Addr.lower;
1662                                 temp64.val32.upper = c->SG[i].Addr.upper;
1663                                 pci_unmap_single(h->pdev,
1664                                         (dma_addr_t) temp64.val, buff_size[i],
1665                                         PCI_DMA_BIDIRECTIONAL);
1666                         }
1667                         check_ioctl_unit_attention(h, c);
1668                         /* Copy the error information out */
1669                         ioc->error_info = *(c->err_info);
1670                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1671                                 cmd_free(h, c, 0);
1672                                 status = -EFAULT;
1673                                 goto cleanup1;
1674                         }
1675                         if (ioc->Request.Type.Direction == XFER_READ) {
1676                                 /* Copy the data out of the buffer we created */
1677                                 BYTE __user *ptr = ioc->buf;
1678                                 for (i = 0; i < sg_used; i++) {
1679                                         if (copy_to_user
1680                                             (ptr, buff[i], buff_size[i])) {
1681                                                 cmd_free(h, c, 0);
1682                                                 status = -EFAULT;
1683                                                 goto cleanup1;
1684                                         }
1685                                         ptr += buff_size[i];
1686                                 }
1687                         }
1688                         cmd_free(h, c, 0);
1689                         status = 0;
1690                       cleanup1:
1691                         if (buff) {
1692                                 for (i = 0; i < sg_used; i++)
1693                                         kfree(buff[i]);
1694                                 kfree(buff);
1695                         }
1696                         kfree(buff_size);
1697                         kfree(ioc);
1698                         return status;
1699                 }
1700
1701         /* scsi_cmd_ioctl handles these, below, though some are not */
1702         /* very meaningful for cciss.  SG_IO is the main one people want. */
1703
1704         case SG_GET_VERSION_NUM:
1705         case SG_SET_TIMEOUT:
1706         case SG_GET_TIMEOUT:
1707         case SG_GET_RESERVED_SIZE:
1708         case SG_SET_RESERVED_SIZE:
1709         case SG_EMULATED_HOST:
1710         case SG_IO:
1711         case SCSI_IOCTL_SEND_COMMAND:
1712                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1713
1714         /* scsi_cmd_ioctl would normally handle these, below, but */
1715         /* they aren't a good fit for cciss, as CD-ROMs are */
1716         /* not supported, and we don't have any bus/target/lun */
1717         /* which we present to the kernel. */
1718
1719         case CDROM_SEND_PACKET:
1720         case CDROMCLOSETRAY:
1721         case CDROMEJECT:
1722         case SCSI_IOCTL_GET_IDLUN:
1723         case SCSI_IOCTL_GET_BUS_NUMBER:
1724         default:
1725                 return -ENOTTY;
1726         }
1727 }
1728
1729 static void cciss_check_queues(ctlr_info_t *h)
1730 {
1731         int start_queue = h->next_to_run;
1732         int i;
1733
1734         /* check to see if we have maxed out the number of commands that can
1735          * be placed on the queue.  If so then exit.  We do this check here
1736          * in case the interrupt we serviced was from an ioctl and did not
1737          * free any new commands.
1738          */
1739         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1740                 return;
1741
1742         /* We have room on the queue for more commands.  Now we need to queue
1743          * them up.  We will also keep track of the next queue to run so
1744          * that every queue gets a chance to be started first.
1745          */
1746         for (i = 0; i < h->highest_lun + 1; i++) {
1747                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1748                 /* make sure the disk has been added and the drive is real
1749                  * because this can be called from the middle of init_one.
1750                  */
1751                 if (!h->drv[curr_queue])
1752                         continue;
1753                 if (!(h->drv[curr_queue]->queue) ||
1754                         !(h->drv[curr_queue]->heads))
1755                         continue;
1756                 blk_start_queue(h->gendisk[curr_queue]->queue);
1757
1758                 /* check to see if we have maxed out the number of commands
1759                  * that can be placed on the queue.
1760                  */
1761                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1762                         if (curr_queue == start_queue) {
1763                                 h->next_to_run =
1764                                     (start_queue + 1) % (h->highest_lun + 1);
1765                                 break;
1766                         } else {
1767                                 h->next_to_run = curr_queue;
1768                                 break;
1769                         }
1770                 }
1771         }
1772 }
1773
1774 static void cciss_softirq_done(struct request *rq)
1775 {
1776         CommandList_struct *c = rq->completion_data;
1777         ctlr_info_t *h = hba[c->ctlr];
1778         SGDescriptor_struct *curr_sg = c->SG;
1779         u64bit temp64;
1780         unsigned long flags;
1781         int i, ddir;
1782         int sg_index = 0;
1783
1784         if (c->Request.Type.Direction == XFER_READ)
1785                 ddir = PCI_DMA_FROMDEVICE;
1786         else
1787                 ddir = PCI_DMA_TODEVICE;
1788
1789         /* command did not need to be retried */
1790         /* unmap the DMA mapping for all the scatter gather elements */
1791         for (i = 0; i < c->Header.SGList; i++) {
1792                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1793                         cciss_unmap_sg_chain_block(h, c);
1794                         /* Point to the next block */
1795                         curr_sg = h->cmd_sg_list[c->cmdindex];
1796                         sg_index = 0;
1797                 }
1798                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1799                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1800                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1801                                 ddir);
1802                 ++sg_index;
1803         }
1804
1805 #ifdef CCISS_DEBUG
1806         printk("Done with %p\n", rq);
1807 #endif                          /* CCISS_DEBUG */
1808
1809         /* set the residual count for pc requests */
1810         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1811                 rq->resid_len = c->err_info->ResidualCnt;
1812
1813         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1814
1815         spin_lock_irqsave(&h->lock, flags);
1816         cmd_free(h, c, 1);
1817         cciss_check_queues(h);
1818         spin_unlock_irqrestore(&h->lock, flags);
1819 }
1820
1821 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1822         unsigned char scsi3addr[], uint32_t log_unit)
1823 {
1824         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1825                 sizeof(h->drv[log_unit]->LunID));
1826 }
1827
1828 /* This function gets the SCSI vendor, model, and revision of a logical drive
1829  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1830  * they cannot be read.
1831  */
1832 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1833                                    char *vendor, char *model, char *rev)
1834 {
1835         int rc;
1836         InquiryData_struct *inq_buf;
1837         unsigned char scsi3addr[8];
1838
1839         *vendor = '\0';
1840         *model = '\0';
1841         *rev = '\0';
1842
1843         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1844         if (!inq_buf)
1845                 return;
1846
1847         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1848         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1849                         scsi3addr, TYPE_CMD);
1850         if (rc == IO_OK) {
1851                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1852                 vendor[VENDOR_LEN] = '\0';
1853                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1854                 model[MODEL_LEN] = '\0';
1855                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1856                 rev[REV_LEN] = '\0';
1857         }
1858
1859         kfree(inq_buf);
1860         return;
1861 }
1862
1863 /* This function gets the serial number of a logical drive via
1864  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1865  * number cannot be had, for whatever reason, 16 bytes of 0xff
1866  * are returned instead.
1867  */
1868 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1869                                 unsigned char *serial_no, int buflen)
1870 {
1871 #define PAGE_83_INQ_BYTES 64
1872         int rc;
1873         unsigned char *buf;
1874         unsigned char scsi3addr[8];
1875
1876         if (buflen > 16)
1877                 buflen = 16;
1878         memset(serial_no, 0xff, buflen);
1879         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1880         if (!buf)
1881                 return;
1882         memset(serial_no, 0, buflen);
1883         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1884         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1885                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1886         if (rc == IO_OK)
1887                 memcpy(serial_no, &buf[8], buflen);
1888         kfree(buf);
1889         return;
1890 }
1891
1892 /*
1893  * cciss_add_disk sets up the block device queue for a logical drive
1894  */
1895 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1896                                 int drv_index)
1897 {
1898         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1899         if (!disk->queue)
1900                 goto init_queue_failure;
1901         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1902         disk->major = h->major;
1903         disk->first_minor = drv_index << NWD_SHIFT;
1904         disk->fops = &cciss_fops;
1905         if (cciss_create_ld_sysfs_entry(h, drv_index))
1906                 goto cleanup_queue;
1907         disk->private_data = h->drv[drv_index];
1908         disk->driverfs_dev = &h->drv[drv_index]->dev;
1909
1910         /* Set up queue information */
1911         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1912
1913         /* This is a hardware imposed limit. */
1914         blk_queue_max_segments(disk->queue, h->maxsgentries);
1915
1916         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1917
1918         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1919
1920         disk->queue->queuedata = h;
1921
1922         blk_queue_logical_block_size(disk->queue,
1923                                      h->drv[drv_index]->block_size);
1924
1925         /* Make sure all queue data is written out before */
1926         /* setting h->drv[drv_index]->queue, as setting this */
1927         /* allows the interrupt handler to start the queue */
1928         wmb();
1929         h->drv[drv_index]->queue = disk->queue;
1930         add_disk(disk);
1931         return 0;
1932
1933 cleanup_queue:
1934         blk_cleanup_queue(disk->queue);
1935         disk->queue = NULL;
1936 init_queue_failure:
1937         return -1;
1938 }
1939
1940 /* This function will check the usage_count of the drive to be updated/added.
1941  * If the usage_count is zero and it is a heretofore unknown drive, or,
1942  * the drive's capacity, geometry, or serial number has changed,
1943  * then the drive information will be updated and the disk will be
1944  * re-registered with the kernel.  If these conditions don't hold,
1945  * then it will be left alone for the next reboot.  The exception to this
1946  * is disk 0 which will always be left registered with the kernel since it
1947  * is also the controller node.  Any changes to disk 0 will show up on
1948  * the next reboot.
1949  */
1950 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1951         int first_time, int via_ioctl)
1952 {
1953         struct gendisk *disk;
1954         InquiryData_struct *inq_buff = NULL;
1955         unsigned int block_size;
1956         sector_t total_size;
1957         unsigned long flags = 0;
1958         int ret = 0;
1959         drive_info_struct *drvinfo;
1960
1961         /* Get information about the disk and modify the driver structure */
1962         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1963         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1964         if (inq_buff == NULL || drvinfo == NULL)
1965                 goto mem_msg;
1966
1967         /* testing to see if 16-byte CDBs are already being used */
1968         if (h->cciss_read == CCISS_READ_16) {
1969                 cciss_read_capacity_16(h, drv_index,
1970                         &total_size, &block_size);
1971
1972         } else {
1973                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1974                 /* if read_capacity returns all F's this volume is >2TB */
1975                 /* in size so we switch to 16-byte CDB's for all */
1976                 /* read/write ops */
1977                 if (total_size == 0xFFFFFFFFULL) {
1978                         cciss_read_capacity_16(h, drv_index,
1979                         &total_size, &block_size);
1980                         h->cciss_read = CCISS_READ_16;
1981                         h->cciss_write = CCISS_WRITE_16;
1982                 } else {
1983                         h->cciss_read = CCISS_READ_10;
1984                         h->cciss_write = CCISS_WRITE_10;
1985                 }
1986         }
1987
1988         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1989                                inq_buff, drvinfo);
1990         drvinfo->block_size = block_size;
1991         drvinfo->nr_blocks = total_size + 1;
1992
1993         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1994                                 drvinfo->model, drvinfo->rev);
1995         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1996                         sizeof(drvinfo->serial_no));
1997         /* Save the lunid in case we deregister the disk, below. */
1998         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1999                 sizeof(drvinfo->LunID));
2000
2001         /* Is it the same disk we already know, and nothing's changed? */
2002         if (h->drv[drv_index]->raid_level != -1 &&
2003                 ((memcmp(drvinfo->serial_no,
2004                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2005                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2006                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2007                 drvinfo->heads == h->drv[drv_index]->heads &&
2008                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2009                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2010                         /* The disk is unchanged, nothing to update */
2011                         goto freeret;
2012
2013         /* If we get here it's not the same disk, or something's changed,
2014          * so we need to * deregister it, and re-register it, if it's not
2015          * in use.
2016          * If the disk already exists then deregister it before proceeding
2017          * (unless it's the first disk (for the controller node).
2018          */
2019         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2020                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
2021                 spin_lock_irqsave(&h->lock, flags);
2022                 h->drv[drv_index]->busy_configuring = 1;
2023                 spin_unlock_irqrestore(&h->lock, flags);
2024
2025                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2026                  * which keeps the interrupt handler from starting
2027                  * the queue.
2028                  */
2029                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2030         }
2031
2032         /* If the disk is in use return */
2033         if (ret)
2034                 goto freeret;
2035
2036         /* Save the new information from cciss_geometry_inquiry
2037          * and serial number inquiry.  If the disk was deregistered
2038          * above, then h->drv[drv_index] will be NULL.
2039          */
2040         if (h->drv[drv_index] == NULL) {
2041                 drvinfo->device_initialized = 0;
2042                 h->drv[drv_index] = drvinfo;
2043                 drvinfo = NULL; /* so it won't be freed below. */
2044         } else {
2045                 /* special case for cxd0 */
2046                 h->drv[drv_index]->block_size = drvinfo->block_size;
2047                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2048                 h->drv[drv_index]->heads = drvinfo->heads;
2049                 h->drv[drv_index]->sectors = drvinfo->sectors;
2050                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2051                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2052                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2053                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2054                         VENDOR_LEN + 1);
2055                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2056                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2057         }
2058
2059         ++h->num_luns;
2060         disk = h->gendisk[drv_index];
2061         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2062
2063         /* If it's not disk 0 (drv_index != 0)
2064          * or if it was disk 0, but there was previously
2065          * no actual corresponding configured logical drive
2066          * (raid_leve == -1) then we want to update the
2067          * logical drive's information.
2068          */
2069         if (drv_index || first_time) {
2070                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2071                         cciss_free_gendisk(h, drv_index);
2072                         cciss_free_drive_info(h, drv_index);
2073                         printk(KERN_WARNING "cciss:%d could not update "
2074                                 "disk %d\n", h->ctlr, drv_index);
2075                         --h->num_luns;
2076                 }
2077         }
2078
2079 freeret:
2080         kfree(inq_buff);
2081         kfree(drvinfo);
2082         return;
2083 mem_msg:
2084         printk(KERN_ERR "cciss: out of memory\n");
2085         goto freeret;
2086 }
2087
2088 /* This function will find the first index of the controllers drive array
2089  * that has a null drv pointer and allocate the drive info struct and
2090  * will return that index   This is where new drives will be added.
2091  * If the index to be returned is greater than the highest_lun index for
2092  * the controller then highest_lun is set * to this new index.
2093  * If there are no available indexes or if tha allocation fails, then -1
2094  * is returned.  * "controller_node" is used to know if this is a real
2095  * logical drive, or just the controller node, which determines if this
2096  * counts towards highest_lun.
2097  */
2098 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2099 {
2100         int i;
2101         drive_info_struct *drv;
2102
2103         /* Search for an empty slot for our drive info */
2104         for (i = 0; i < CISS_MAX_LUN; i++) {
2105
2106                 /* if not cxd0 case, and it's occupied, skip it. */
2107                 if (h->drv[i] && i != 0)
2108                         continue;
2109                 /*
2110                  * If it's cxd0 case, and drv is alloc'ed already, and a
2111                  * disk is configured there, skip it.
2112                  */
2113                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2114                         continue;
2115
2116                 /*
2117                  * We've found an empty slot.  Update highest_lun
2118                  * provided this isn't just the fake cxd0 controller node.
2119                  */
2120                 if (i > h->highest_lun && !controller_node)
2121                         h->highest_lun = i;
2122
2123                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2124                 if (i == 0 && h->drv[i] != NULL)
2125                         return i;
2126
2127                 /*
2128                  * Found an empty slot, not already alloc'ed.  Allocate it.
2129                  * Mark it with raid_level == -1, so we know it's new later on.
2130                  */
2131                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2132                 if (!drv)
2133                         return -1;
2134                 drv->raid_level = -1; /* so we know it's new */
2135                 h->drv[i] = drv;
2136                 return i;
2137         }
2138         return -1;
2139 }
2140
2141 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2142 {
2143         kfree(h->drv[drv_index]);
2144         h->drv[drv_index] = NULL;
2145 }
2146
2147 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2148 {
2149         put_disk(h->gendisk[drv_index]);
2150         h->gendisk[drv_index] = NULL;
2151 }
2152
2153 /* cciss_add_gendisk finds a free hba[]->drv structure
2154  * and allocates a gendisk if needed, and sets the lunid
2155  * in the drvinfo structure.   It returns the index into
2156  * the ->drv[] array, or -1 if none are free.
2157  * is_controller_node indicates whether highest_lun should
2158  * count this disk, or if it's only being added to provide
2159  * a means to talk to the controller in case no logical
2160  * drives have yet been configured.
2161  */
2162 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2163         int controller_node)
2164 {
2165         int drv_index;
2166
2167         drv_index = cciss_alloc_drive_info(h, controller_node);
2168         if (drv_index == -1)
2169                 return -1;
2170
2171         /*Check if the gendisk needs to be allocated */
2172         if (!h->gendisk[drv_index]) {
2173                 h->gendisk[drv_index] =
2174                         alloc_disk(1 << NWD_SHIFT);
2175                 if (!h->gendisk[drv_index]) {
2176                         printk(KERN_ERR "cciss%d: could not "
2177                                 "allocate a new disk %d\n",
2178                                 h->ctlr, drv_index);
2179                         goto err_free_drive_info;
2180                 }
2181         }
2182         memcpy(h->drv[drv_index]->LunID, lunid,
2183                 sizeof(h->drv[drv_index]->LunID));
2184         if (cciss_create_ld_sysfs_entry(h, drv_index))
2185                 goto err_free_disk;
2186         /* Don't need to mark this busy because nobody */
2187         /* else knows about this disk yet to contend */
2188         /* for access to it. */
2189         h->drv[drv_index]->busy_configuring = 0;
2190         wmb();
2191         return drv_index;
2192
2193 err_free_disk:
2194         cciss_free_gendisk(h, drv_index);
2195 err_free_drive_info:
2196         cciss_free_drive_info(h, drv_index);
2197         return -1;
2198 }
2199
2200 /* This is for the special case of a controller which
2201  * has no logical drives.  In this case, we still need
2202  * to register a disk so the controller can be accessed
2203  * by the Array Config Utility.
2204  */
2205 static void cciss_add_controller_node(ctlr_info_t *h)
2206 {
2207         struct gendisk *disk;
2208         int drv_index;
2209
2210         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2211                 return;
2212
2213         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2214         if (drv_index == -1)
2215                 goto error;
2216         h->drv[drv_index]->block_size = 512;
2217         h->drv[drv_index]->nr_blocks = 0;
2218         h->drv[drv_index]->heads = 0;
2219         h->drv[drv_index]->sectors = 0;
2220         h->drv[drv_index]->cylinders = 0;
2221         h->drv[drv_index]->raid_level = -1;
2222         memset(h->drv[drv_index]->serial_no, 0, 16);
2223         disk = h->gendisk[drv_index];
2224         if (cciss_add_disk(h, disk, drv_index) == 0)
2225                 return;
2226         cciss_free_gendisk(h, drv_index);
2227         cciss_free_drive_info(h, drv_index);
2228 error:
2229         printk(KERN_WARNING "cciss%d: could not "
2230                 "add disk 0.\n", h->ctlr);
2231         return;
2232 }
2233
2234 /* This function will add and remove logical drives from the Logical
2235  * drive array of the controller and maintain persistency of ordering
2236  * so that mount points are preserved until the next reboot.  This allows
2237  * for the removal of logical drives in the middle of the drive array
2238  * without a re-ordering of those drives.
2239  * INPUT
2240  * h            = The controller to perform the operations on
2241  */
2242 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2243         int via_ioctl)
2244 {
2245         int num_luns;
2246         ReportLunData_struct *ld_buff = NULL;
2247         int return_code;
2248         int listlength = 0;
2249         int i;
2250         int drv_found;
2251         int drv_index = 0;
2252         unsigned char lunid[8] = CTLR_LUNID;
2253         unsigned long flags;
2254
2255         if (!capable(CAP_SYS_RAWIO))
2256                 return -EPERM;
2257
2258         /* Set busy_configuring flag for this operation */
2259         spin_lock_irqsave(&h->lock, flags);
2260         if (h->busy_configuring) {
2261                 spin_unlock_irqrestore(&h->lock, flags);
2262                 return -EBUSY;
2263         }
2264         h->busy_configuring = 1;
2265         spin_unlock_irqrestore(&h->lock, flags);
2266
2267         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2268         if (ld_buff == NULL)
2269                 goto mem_msg;
2270
2271         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2272                                       sizeof(ReportLunData_struct),
2273                                       0, CTLR_LUNID, TYPE_CMD);
2274
2275         if (return_code == IO_OK)
2276                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2277         else {  /* reading number of logical volumes failed */
2278                 printk(KERN_WARNING "cciss: report logical volume"
2279                        " command failed\n");
2280                 listlength = 0;
2281                 goto freeret;
2282         }
2283
2284         num_luns = listlength / 8;      /* 8 bytes per entry */
2285         if (num_luns > CISS_MAX_LUN) {
2286                 num_luns = CISS_MAX_LUN;
2287                 printk(KERN_WARNING "cciss: more luns configured"
2288                        " on controller than can be handled by"
2289                        " this driver.\n");
2290         }
2291
2292         if (num_luns == 0)
2293                 cciss_add_controller_node(h);
2294
2295         /* Compare controller drive array to driver's drive array
2296          * to see if any drives are missing on the controller due
2297          * to action of Array Config Utility (user deletes drive)
2298          * and deregister logical drives which have disappeared.
2299          */
2300         for (i = 0; i <= h->highest_lun; i++) {
2301                 int j;
2302                 drv_found = 0;
2303
2304                 /* skip holes in the array from already deleted drives */
2305                 if (h->drv[i] == NULL)
2306                         continue;
2307
2308                 for (j = 0; j < num_luns; j++) {
2309                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2310                         if (memcmp(h->drv[i]->LunID, lunid,
2311                                 sizeof(lunid)) == 0) {
2312                                 drv_found = 1;
2313                                 break;
2314                         }
2315                 }
2316                 if (!drv_found) {
2317                         /* Deregister it from the OS, it's gone. */
2318                         spin_lock_irqsave(&h->lock, flags);
2319                         h->drv[i]->busy_configuring = 1;
2320                         spin_unlock_irqrestore(&h->lock, flags);
2321                         return_code = deregister_disk(h, i, 1, via_ioctl);
2322                         if (h->drv[i] != NULL)
2323                                 h->drv[i]->busy_configuring = 0;
2324                 }
2325         }
2326
2327         /* Compare controller drive array to driver's drive array.
2328          * Check for updates in the drive information and any new drives
2329          * on the controller due to ACU adding logical drives, or changing
2330          * a logical drive's size, etc.  Reregister any new/changed drives
2331          */
2332         for (i = 0; i < num_luns; i++) {
2333                 int j;
2334
2335                 drv_found = 0;
2336
2337                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2338                 /* Find if the LUN is already in the drive array
2339                  * of the driver.  If so then update its info
2340                  * if not in use.  If it does not exist then find
2341                  * the first free index and add it.
2342                  */
2343                 for (j = 0; j <= h->highest_lun; j++) {
2344                         if (h->drv[j] != NULL &&
2345                                 memcmp(h->drv[j]->LunID, lunid,
2346                                         sizeof(h->drv[j]->LunID)) == 0) {
2347                                 drv_index = j;
2348                                 drv_found = 1;
2349                                 break;
2350                         }
2351                 }
2352
2353                 /* check if the drive was found already in the array */
2354                 if (!drv_found) {
2355                         drv_index = cciss_add_gendisk(h, lunid, 0);
2356                         if (drv_index == -1)
2357                                 goto freeret;
2358                 }
2359                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2360         }               /* end for */
2361
2362 freeret:
2363         kfree(ld_buff);
2364         h->busy_configuring = 0;
2365         /* We return -1 here to tell the ACU that we have registered/updated
2366          * all of the drives that we can and to keep it from calling us
2367          * additional times.
2368          */
2369         return -1;
2370 mem_msg:
2371         printk(KERN_ERR "cciss: out of memory\n");
2372         h->busy_configuring = 0;
2373         goto freeret;
2374 }
2375
2376 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2377 {
2378         /* zero out the disk size info */
2379         drive_info->nr_blocks = 0;
2380         drive_info->block_size = 0;
2381         drive_info->heads = 0;
2382         drive_info->sectors = 0;
2383         drive_info->cylinders = 0;
2384         drive_info->raid_level = -1;
2385         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2386         memset(drive_info->model, 0, sizeof(drive_info->model));
2387         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2388         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2389         /*
2390          * don't clear the LUNID though, we need to remember which
2391          * one this one is.
2392          */
2393 }
2394
2395 /* This function will deregister the disk and it's queue from the
2396  * kernel.  It must be called with the controller lock held and the
2397  * drv structures busy_configuring flag set.  It's parameters are:
2398  *
2399  * disk = This is the disk to be deregistered
2400  * drv  = This is the drive_info_struct associated with the disk to be
2401  *        deregistered.  It contains information about the disk used
2402  *        by the driver.
2403  * clear_all = This flag determines whether or not the disk information
2404  *             is going to be completely cleared out and the highest_lun
2405  *             reset.  Sometimes we want to clear out information about
2406  *             the disk in preparation for re-adding it.  In this case
2407  *             the highest_lun should be left unchanged and the LunID
2408  *             should not be cleared.
2409  * via_ioctl
2410  *    This indicates whether we've reached this path via ioctl.
2411  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2412  *    If this path is reached via ioctl(), then the max_usage_count will
2413  *    be 1, as the process calling ioctl() has got to have the device open.
2414  *    If we get here via sysfs, then the max usage count will be zero.
2415 */
2416 static int deregister_disk(ctlr_info_t *h, int drv_index,
2417                            int clear_all, int via_ioctl)
2418 {
2419         int i;
2420         struct gendisk *disk;
2421         drive_info_struct *drv;
2422         int recalculate_highest_lun;
2423
2424         if (!capable(CAP_SYS_RAWIO))
2425                 return -EPERM;
2426
2427         drv = h->drv[drv_index];
2428         disk = h->gendisk[drv_index];
2429
2430         /* make sure logical volume is NOT is use */
2431         if (clear_all || (h->gendisk[0] == disk)) {
2432                 if (drv->usage_count > via_ioctl)
2433                         return -EBUSY;
2434         } else if (drv->usage_count > 0)
2435                 return -EBUSY;
2436
2437         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2438
2439         /* invalidate the devices and deregister the disk.  If it is disk
2440          * zero do not deregister it but just zero out it's values.  This
2441          * allows us to delete disk zero but keep the controller registered.
2442          */
2443         if (h->gendisk[0] != disk) {
2444                 struct request_queue *q = disk->queue;
2445                 if (disk->flags & GENHD_FL_UP) {
2446                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2447                         del_gendisk(disk);
2448                 }
2449                 if (q)
2450                         blk_cleanup_queue(q);
2451                 /* If clear_all is set then we are deleting the logical
2452                  * drive, not just refreshing its info.  For drives
2453                  * other than disk 0 we will call put_disk.  We do not
2454                  * do this for disk 0 as we need it to be able to
2455                  * configure the controller.
2456                  */
2457                 if (clear_all){
2458                         /* This isn't pretty, but we need to find the
2459                          * disk in our array and NULL our the pointer.
2460                          * This is so that we will call alloc_disk if
2461                          * this index is used again later.
2462                          */
2463                         for (i=0; i < CISS_MAX_LUN; i++){
2464                                 if (h->gendisk[i] == disk) {
2465                                         h->gendisk[i] = NULL;
2466                                         break;
2467                                 }
2468                         }
2469                         put_disk(disk);
2470                 }
2471         } else {
2472                 set_capacity(disk, 0);
2473                 cciss_clear_drive_info(drv);
2474         }
2475
2476         --h->num_luns;
2477
2478         /* if it was the last disk, find the new hightest lun */
2479         if (clear_all && recalculate_highest_lun) {
2480                 int newhighest = -1;
2481                 for (i = 0; i <= h->highest_lun; i++) {
2482                         /* if the disk has size > 0, it is available */
2483                         if (h->drv[i] && h->drv[i]->heads)
2484                                 newhighest = i;
2485                 }
2486                 h->highest_lun = newhighest;
2487         }
2488         return 0;
2489 }
2490
2491 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2492                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2493                 int cmd_type)
2494 {
2495         u64bit buff_dma_handle;
2496         int status = IO_OK;
2497
2498         c->cmd_type = CMD_IOCTL_PEND;
2499         c->Header.ReplyQueue = 0;
2500         if (buff != NULL) {
2501                 c->Header.SGList = 1;
2502                 c->Header.SGTotal = 1;
2503         } else {
2504                 c->Header.SGList = 0;
2505                 c->Header.SGTotal = 0;
2506         }
2507         c->Header.Tag.lower = c->busaddr;
2508         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2509
2510         c->Request.Type.Type = cmd_type;
2511         if (cmd_type == TYPE_CMD) {
2512                 switch (cmd) {
2513                 case CISS_INQUIRY:
2514                         /* are we trying to read a vital product page */
2515                         if (page_code != 0) {
2516                                 c->Request.CDB[1] = 0x01;
2517                                 c->Request.CDB[2] = page_code;
2518                         }
2519                         c->Request.CDBLen = 6;
2520                         c->Request.Type.Attribute = ATTR_SIMPLE;
2521                         c->Request.Type.Direction = XFER_READ;
2522                         c->Request.Timeout = 0;
2523                         c->Request.CDB[0] = CISS_INQUIRY;
2524                         c->Request.CDB[4] = size & 0xFF;
2525                         break;
2526                 case CISS_REPORT_LOG:
2527                 case CISS_REPORT_PHYS:
2528                         /* Talking to controller so It's a physical command
2529                            mode = 00 target = 0.  Nothing to write.
2530                          */
2531                         c->Request.CDBLen = 12;
2532                         c->Request.Type.Attribute = ATTR_SIMPLE;
2533                         c->Request.Type.Direction = XFER_READ;
2534                         c->Request.Timeout = 0;
2535                         c->Request.CDB[0] = cmd;
2536                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2537                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2538                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2539                         c->Request.CDB[9] = size & 0xFF;
2540                         break;
2541
2542                 case CCISS_READ_CAPACITY:
2543                         c->Request.CDBLen = 10;
2544                         c->Request.Type.Attribute = ATTR_SIMPLE;
2545                         c->Request.Type.Direction = XFER_READ;
2546                         c->Request.Timeout = 0;
2547                         c->Request.CDB[0] = cmd;
2548                         break;
2549                 case CCISS_READ_CAPACITY_16:
2550                         c->Request.CDBLen = 16;
2551                         c->Request.Type.Attribute = ATTR_SIMPLE;
2552                         c->Request.Type.Direction = XFER_READ;
2553                         c->Request.Timeout = 0;
2554                         c->Request.CDB[0] = cmd;
2555                         c->Request.CDB[1] = 0x10;
2556                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2557                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2558                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2559                         c->Request.CDB[13] = size & 0xFF;
2560                         c->Request.Timeout = 0;
2561                         c->Request.CDB[0] = cmd;
2562                         break;
2563                 case CCISS_CACHE_FLUSH:
2564                         c->Request.CDBLen = 12;
2565                         c->Request.Type.Attribute = ATTR_SIMPLE;
2566                         c->Request.Type.Direction = XFER_WRITE;
2567                         c->Request.Timeout = 0;
2568                         c->Request.CDB[0] = BMIC_WRITE;
2569                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2570                         break;
2571                 case TEST_UNIT_READY:
2572                         c->Request.CDBLen = 6;
2573                         c->Request.Type.Attribute = ATTR_SIMPLE;
2574                         c->Request.Type.Direction = XFER_NONE;
2575                         c->Request.Timeout = 0;
2576                         break;
2577                 default:
2578                         printk(KERN_WARNING
2579                                "cciss%d:  Unknown Command 0x%c\n",
2580                                 h->ctlr, cmd);
2581                         return IO_ERROR;
2582                 }
2583         } else if (cmd_type == TYPE_MSG) {
2584                 switch (cmd) {
2585                 case 0: /* ABORT message */
2586                         c->Request.CDBLen = 12;
2587                         c->Request.Type.Attribute = ATTR_SIMPLE;
2588                         c->Request.Type.Direction = XFER_WRITE;
2589                         c->Request.Timeout = 0;
2590                         c->Request.CDB[0] = cmd;        /* abort */
2591                         c->Request.CDB[1] = 0;  /* abort a command */
2592                         /* buff contains the tag of the command to abort */
2593                         memcpy(&c->Request.CDB[4], buff, 8);
2594                         break;
2595                 case 1: /* RESET message */
2596                         c->Request.CDBLen = 16;
2597                         c->Request.Type.Attribute = ATTR_SIMPLE;
2598                         c->Request.Type.Direction = XFER_NONE;
2599                         c->Request.Timeout = 0;
2600                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2601                         c->Request.CDB[0] = cmd;        /* reset */
2602                         c->Request.CDB[1] = 0x03;       /* reset a target */
2603                         break;
2604                 case 3: /* No-Op message */
2605                         c->Request.CDBLen = 1;
2606                         c->Request.Type.Attribute = ATTR_SIMPLE;
2607                         c->Request.Type.Direction = XFER_WRITE;
2608                         c->Request.Timeout = 0;
2609                         c->Request.CDB[0] = cmd;
2610                         break;
2611                 default:
2612                         printk(KERN_WARNING
2613                                "cciss%d: unknown message type %d\n",
2614                                 h->ctlr, cmd);
2615                         return IO_ERROR;
2616                 }
2617         } else {
2618                 printk(KERN_WARNING
2619                        "cciss%d: unknown command type %d\n", h->ctlr, cmd_type);
2620                 return IO_ERROR;
2621         }
2622         /* Fill in the scatter gather information */
2623         if (size > 0) {
2624                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2625                                                              buff, size,
2626                                                              PCI_DMA_BIDIRECTIONAL);
2627                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2628                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2629                 c->SG[0].Len = size;
2630                 c->SG[0].Ext = 0;       /* we are not chaining */
2631         }
2632         return status;
2633 }
2634
2635 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2636 {
2637         switch (c->err_info->ScsiStatus) {
2638         case SAM_STAT_GOOD:
2639                 return IO_OK;
2640         case SAM_STAT_CHECK_CONDITION:
2641                 switch (0xf & c->err_info->SenseInfo[2]) {
2642                 case 0: return IO_OK; /* no sense */
2643                 case 1: return IO_OK; /* recovered error */
2644                 default:
2645                         if (check_for_unit_attention(h, c))
2646                                 return IO_NEEDS_RETRY;
2647                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2648                                 "check condition, sense key = 0x%02x\n",
2649                                 h->ctlr, c->Request.CDB[0],
2650                                 c->err_info->SenseInfo[2]);
2651                 }
2652                 break;
2653         default:
2654                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2655                         "scsi status = 0x%02x\n", h->ctlr,
2656                         c->Request.CDB[0], c->err_info->ScsiStatus);
2657                 break;
2658         }
2659         return IO_ERROR;
2660 }
2661
2662 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2663 {
2664         int return_status = IO_OK;
2665
2666         if (c->err_info->CommandStatus == CMD_SUCCESS)
2667                 return IO_OK;
2668
2669         switch (c->err_info->CommandStatus) {
2670         case CMD_TARGET_STATUS:
2671                 return_status = check_target_status(h, c);
2672                 break;
2673         case CMD_DATA_UNDERRUN:
2674         case CMD_DATA_OVERRUN:
2675                 /* expected for inquiry and report lun commands */
2676                 break;
2677         case CMD_INVALID:
2678                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2679                        "reported invalid\n", c->Request.CDB[0]);
2680                 return_status = IO_ERROR;
2681                 break;
2682         case CMD_PROTOCOL_ERR:
2683                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2684                        "protocol error \n", c->Request.CDB[0]);
2685                 return_status = IO_ERROR;
2686                 break;
2687         case CMD_HARDWARE_ERR:
2688                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2689                        " hardware error\n", c->Request.CDB[0]);
2690                 return_status = IO_ERROR;
2691                 break;
2692         case CMD_CONNECTION_LOST:
2693                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2694                        "connection lost\n", c->Request.CDB[0]);
2695                 return_status = IO_ERROR;
2696                 break;
2697         case CMD_ABORTED:
2698                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2699                        "aborted\n", c->Request.CDB[0]);
2700                 return_status = IO_ERROR;
2701                 break;
2702         case CMD_ABORT_FAILED:
2703                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2704                        "abort failed\n", c->Request.CDB[0]);
2705                 return_status = IO_ERROR;
2706                 break;
2707         case CMD_UNSOLICITED_ABORT:
2708                 printk(KERN_WARNING
2709                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2710                         c->Request.CDB[0]);
2711                 return_status = IO_NEEDS_RETRY;
2712                 break;
2713         default:
2714                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2715                        "unknown status %x\n", c->Request.CDB[0],
2716                        c->err_info->CommandStatus);
2717                 return_status = IO_ERROR;
2718         }
2719         return return_status;
2720 }
2721
2722 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2723         int attempt_retry)
2724 {
2725         DECLARE_COMPLETION_ONSTACK(wait);
2726         u64bit buff_dma_handle;
2727         int return_status = IO_OK;
2728
2729 resend_cmd2:
2730         c->waiting = &wait;
2731         enqueue_cmd_and_start_io(h, c);
2732
2733         wait_for_completion(&wait);
2734
2735         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2736                 goto command_done;
2737
2738         return_status = process_sendcmd_error(h, c);
2739
2740         if (return_status == IO_NEEDS_RETRY &&
2741                 c->retry_count < MAX_CMD_RETRIES) {
2742                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2743                         c->Request.CDB[0]);
2744                 c->retry_count++;
2745                 /* erase the old error information */
2746                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2747                 return_status = IO_OK;
2748                 INIT_COMPLETION(wait);
2749                 goto resend_cmd2;
2750         }
2751
2752 command_done:
2753         /* unlock the buffers from DMA */
2754         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2755         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2756         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2757                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2758         return return_status;
2759 }
2760
2761 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2762                            __u8 page_code, unsigned char scsi3addr[],
2763                         int cmd_type)
2764 {
2765         CommandList_struct *c;
2766         int return_status;
2767
2768         c = cmd_alloc(h, 0);
2769         if (!c)
2770                 return -ENOMEM;
2771         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2772                 scsi3addr, cmd_type);
2773         if (return_status == IO_OK)
2774                 return_status = sendcmd_withirq_core(h, c, 1);
2775
2776         cmd_free(h, c, 0);
2777         return return_status;
2778 }
2779
2780 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2781                                    sector_t total_size,
2782                                    unsigned int block_size,
2783                                    InquiryData_struct *inq_buff,
2784                                    drive_info_struct *drv)
2785 {
2786         int return_code;
2787         unsigned long t;
2788         unsigned char scsi3addr[8];
2789
2790         memset(inq_buff, 0, sizeof(InquiryData_struct));
2791         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2792         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2793                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2794         if (return_code == IO_OK) {
2795                 if (inq_buff->data_byte[8] == 0xFF) {
2796                         printk(KERN_WARNING
2797                                "cciss: reading geometry failed, volume "
2798                                "does not support reading geometry\n");
2799                         drv->heads = 255;
2800                         drv->sectors = 32;      /* Sectors per track */
2801                         drv->cylinders = total_size + 1;
2802                         drv->raid_level = RAID_UNKNOWN;
2803                 } else {
2804                         drv->heads = inq_buff->data_byte[6];
2805                         drv->sectors = inq_buff->data_byte[7];
2806                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2807                         drv->cylinders += inq_buff->data_byte[5];
2808                         drv->raid_level = inq_buff->data_byte[8];
2809                 }
2810                 drv->block_size = block_size;
2811                 drv->nr_blocks = total_size + 1;
2812                 t = drv->heads * drv->sectors;
2813                 if (t > 1) {
2814                         sector_t real_size = total_size + 1;
2815                         unsigned long rem = sector_div(real_size, t);
2816                         if (rem)
2817                                 real_size++;
2818                         drv->cylinders = real_size;
2819                 }
2820         } else {                /* Get geometry failed */
2821                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2822         }
2823 }
2824
2825 static void
2826 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2827                     unsigned int *block_size)
2828 {
2829         ReadCapdata_struct *buf;
2830         int return_code;
2831         unsigned char scsi3addr[8];
2832
2833         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2834         if (!buf) {
2835                 printk(KERN_WARNING "cciss: out of memory\n");
2836                 return;
2837         }
2838
2839         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2840         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2841                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2842         if (return_code == IO_OK) {
2843                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2844                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2845         } else {                /* read capacity command failed */
2846                 printk(KERN_WARNING "cciss: read capacity failed\n");
2847                 *total_size = 0;
2848                 *block_size = BLOCK_SIZE;
2849         }
2850         kfree(buf);
2851 }
2852
2853 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2854         sector_t *total_size, unsigned int *block_size)
2855 {
2856         ReadCapdata_struct_16 *buf;
2857         int return_code;
2858         unsigned char scsi3addr[8];
2859
2860         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2861         if (!buf) {
2862                 printk(KERN_WARNING "cciss: out of memory\n");
2863                 return;
2864         }
2865
2866         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2867         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2868                 buf, sizeof(ReadCapdata_struct_16),
2869                         0, scsi3addr, TYPE_CMD);
2870         if (return_code == IO_OK) {
2871                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2872                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2873         } else {                /* read capacity command failed */
2874                 printk(KERN_WARNING "cciss: read capacity failed\n");
2875                 *total_size = 0;
2876                 *block_size = BLOCK_SIZE;
2877         }
2878         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2879                (unsigned long long)*total_size+1, *block_size);
2880         kfree(buf);
2881 }
2882
2883 static int cciss_revalidate(struct gendisk *disk)
2884 {
2885         ctlr_info_t *h = get_host(disk);
2886         drive_info_struct *drv = get_drv(disk);
2887         int logvol;
2888         int FOUND = 0;
2889         unsigned int block_size;
2890         sector_t total_size;
2891         InquiryData_struct *inq_buff = NULL;
2892
2893         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2894                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2895                         sizeof(drv->LunID)) == 0) {
2896                         FOUND = 1;
2897                         break;
2898                 }
2899         }
2900
2901         if (!FOUND)
2902                 return 1;
2903
2904         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2905         if (inq_buff == NULL) {
2906                 printk(KERN_WARNING "cciss: out of memory\n");
2907                 return 1;
2908         }
2909         if (h->cciss_read == CCISS_READ_10) {
2910                 cciss_read_capacity(h, logvol,
2911                                         &total_size, &block_size);
2912         } else {
2913                 cciss_read_capacity_16(h, logvol,
2914                                         &total_size, &block_size);
2915         }
2916         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2917                                inq_buff, drv);
2918
2919         blk_queue_logical_block_size(drv->queue, drv->block_size);
2920         set_capacity(disk, drv->nr_blocks);
2921
2922         kfree(inq_buff);
2923         return 0;
2924 }
2925
2926 /*
2927  * Map (physical) PCI mem into (virtual) kernel space
2928  */
2929 static void __iomem *remap_pci_mem(ulong base, ulong size)
2930 {
2931         ulong page_base = ((ulong) base) & PAGE_MASK;
2932         ulong page_offs = ((ulong) base) - page_base;
2933         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2934
2935         return page_remapped ? (page_remapped + page_offs) : NULL;
2936 }
2937
2938 /*
2939  * Takes jobs of the Q and sends them to the hardware, then puts it on
2940  * the Q to wait for completion.
2941  */
2942 static void start_io(ctlr_info_t *h)
2943 {
2944         CommandList_struct *c;
2945
2946         while (!hlist_empty(&h->reqQ)) {
2947                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2948                 /* can't do anything if fifo is full */
2949                 if ((h->access.fifo_full(h))) {
2950                         printk(KERN_WARNING "cciss: fifo full\n");
2951                         break;
2952                 }
2953
2954                 /* Get the first entry from the Request Q */
2955                 removeQ(c);
2956                 h->Qdepth--;
2957
2958                 /* Tell the controller execute command */
2959                 h->access.submit_command(h, c);
2960
2961                 /* Put job onto the completed Q */
2962                 addQ(&h->cmpQ, c);
2963         }
2964 }
2965
2966 /* Assumes that h->lock is held. */
2967 /* Zeros out the error record and then resends the command back */
2968 /* to the controller */
2969 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2970 {
2971         /* erase the old error information */
2972         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2973
2974         /* add it to software queue and then send it to the controller */
2975         addQ(&h->reqQ, c);
2976         h->Qdepth++;
2977         if (h->Qdepth > h->maxQsinceinit)
2978                 h->maxQsinceinit = h->Qdepth;
2979
2980         start_io(h);
2981 }
2982
2983 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2984         unsigned int msg_byte, unsigned int host_byte,
2985         unsigned int driver_byte)
2986 {
2987         /* inverse of macros in scsi.h */
2988         return (scsi_status_byte & 0xff) |
2989                 ((msg_byte & 0xff) << 8) |
2990                 ((host_byte & 0xff) << 16) |
2991                 ((driver_byte & 0xff) << 24);
2992 }
2993
2994 static inline int evaluate_target_status(ctlr_info_t *h,
2995                         CommandList_struct *cmd, int *retry_cmd)
2996 {
2997         unsigned char sense_key;
2998         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2999         int error_value;
3000
3001         *retry_cmd = 0;
3002         /* If we get in here, it means we got "target status", that is, scsi status */
3003         status_byte = cmd->err_info->ScsiStatus;
3004         driver_byte = DRIVER_OK;
3005         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3006
3007         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3008                 host_byte = DID_PASSTHROUGH;
3009         else
3010                 host_byte = DID_OK;
3011
3012         error_value = make_status_bytes(status_byte, msg_byte,
3013                 host_byte, driver_byte);
3014
3015         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3016                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3017                         printk(KERN_WARNING "cciss: cmd %p "
3018                                "has SCSI Status 0x%x\n",
3019                                cmd, cmd->err_info->ScsiStatus);
3020                 return error_value;
3021         }
3022
3023         /* check the sense key */
3024         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3025         /* no status or recovered error */
3026         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3027             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3028                 error_value = 0;
3029
3030         if (check_for_unit_attention(h, cmd)) {
3031                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3032                 return 0;
3033         }
3034
3035         /* Not SG_IO or similar? */
3036         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3037                 if (error_value != 0)
3038                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3039                                " sense key = 0x%x\n", cmd, sense_key);
3040                 return error_value;
3041         }
3042
3043         /* SG_IO or similar, copy sense data back */
3044         if (cmd->rq->sense) {
3045                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3046                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3047                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3048                         cmd->rq->sense_len);
3049         } else
3050                 cmd->rq->sense_len = 0;
3051
3052         return error_value;
3053 }
3054
3055 /* checks the status of the job and calls complete buffers to mark all
3056  * buffers for the completed job. Note that this function does not need
3057  * to hold the hba/queue lock.
3058  */
3059 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3060                                     int timeout)
3061 {
3062         int retry_cmd = 0;
3063         struct request *rq = cmd->rq;
3064
3065         rq->errors = 0;
3066
3067         if (timeout)
3068                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3069
3070         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3071                 goto after_error_processing;
3072
3073         switch (cmd->err_info->CommandStatus) {
3074         case CMD_TARGET_STATUS:
3075                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3076                 break;
3077         case CMD_DATA_UNDERRUN:
3078                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3079                         printk(KERN_WARNING "cciss: cmd %p has"
3080                                " completed with data underrun "
3081                                "reported\n", cmd);
3082                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3083                 }
3084                 break;
3085         case CMD_DATA_OVERRUN:
3086                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3087                         printk(KERN_WARNING "cciss: cmd %p has"
3088                                " completed with data overrun "
3089                                "reported\n", cmd);
3090                 break;
3091         case CMD_INVALID:
3092                 printk(KERN_WARNING "cciss: cmd %p is "
3093                        "reported invalid\n", cmd);
3094                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3095                         cmd->err_info->CommandStatus, DRIVER_OK,
3096                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3097                                 DID_PASSTHROUGH : DID_ERROR);
3098                 break;
3099         case CMD_PROTOCOL_ERR:
3100                 printk(KERN_WARNING "cciss: cmd %p has "
3101                        "protocol error \n", cmd);
3102                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3103                         cmd->err_info->CommandStatus, DRIVER_OK,
3104                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3105                                 DID_PASSTHROUGH : DID_ERROR);
3106                 break;
3107         case CMD_HARDWARE_ERR:
3108                 printk(KERN_WARNING "cciss: cmd %p had "
3109                        " hardware error\n", cmd);
3110                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3111                         cmd->err_info->CommandStatus, DRIVER_OK,
3112                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3113                                 DID_PASSTHROUGH : DID_ERROR);
3114                 break;
3115         case CMD_CONNECTION_LOST:
3116                 printk(KERN_WARNING "cciss: cmd %p had "
3117                        "connection lost\n", cmd);
3118                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3119                         cmd->err_info->CommandStatus, DRIVER_OK,
3120                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3121                                 DID_PASSTHROUGH : DID_ERROR);
3122                 break;
3123         case CMD_ABORTED:
3124                 printk(KERN_WARNING "cciss: cmd %p was "
3125                        "aborted\n", cmd);
3126                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3127                         cmd->err_info->CommandStatus, DRIVER_OK,
3128                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3129                                 DID_PASSTHROUGH : DID_ABORT);
3130                 break;
3131         case CMD_ABORT_FAILED:
3132                 printk(KERN_WARNING "cciss: cmd %p reports "
3133                        "abort failed\n", cmd);
3134                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3135                         cmd->err_info->CommandStatus, DRIVER_OK,
3136                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3137                                 DID_PASSTHROUGH : DID_ERROR);
3138                 break;
3139         case CMD_UNSOLICITED_ABORT:
3140                 printk(KERN_WARNING "cciss%d: unsolicited "
3141                        "abort %p\n", h->ctlr, cmd);
3142                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3143                         retry_cmd = 1;
3144                         printk(KERN_WARNING
3145                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3146                         cmd->retry_count++;
3147                 } else
3148                         printk(KERN_WARNING
3149                                "cciss%d: %p retried too "
3150                                "many times\n", h->ctlr, cmd);
3151                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3152                         cmd->err_info->CommandStatus, DRIVER_OK,
3153                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3154                                 DID_PASSTHROUGH : DID_ABORT);
3155                 break;
3156         case CMD_TIMEOUT:
3157                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3158                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3159                         cmd->err_info->CommandStatus, DRIVER_OK,
3160                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3161                                 DID_PASSTHROUGH : DID_ERROR);
3162                 break;
3163         default:
3164                 printk(KERN_WARNING "cciss: cmd %p returned "
3165                        "unknown status %x\n", cmd,
3166                        cmd->err_info->CommandStatus);
3167                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3168                         cmd->err_info->CommandStatus, DRIVER_OK,
3169                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3170                                 DID_PASSTHROUGH : DID_ERROR);
3171         }
3172
3173 after_error_processing:
3174
3175         /* We need to return this command */
3176         if (retry_cmd) {
3177                 resend_cciss_cmd(h, cmd);
3178                 return;
3179         }
3180         cmd->rq->completion_data = cmd;
3181         blk_complete_request(cmd->rq);
3182 }
3183
3184 static inline u32 cciss_tag_contains_index(u32 tag)
3185 {
3186 #define DIRECT_LOOKUP_BIT 0x10
3187         return tag & DIRECT_LOOKUP_BIT;
3188 }
3189
3190 static inline u32 cciss_tag_to_index(u32 tag)
3191 {
3192 #define DIRECT_LOOKUP_SHIFT 5
3193         return tag >> DIRECT_LOOKUP_SHIFT;
3194 }
3195
3196 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3197 {
3198 #define CCISS_ERROR_BITS 0x03
3199         return tag & ~CCISS_ERROR_BITS;
3200 }
3201
3202 static inline void cciss_mark_tag_indexed(u32 *tag)
3203 {
3204         *tag |= DIRECT_LOOKUP_BIT;
3205 }
3206
3207 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3208 {
3209         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3210 }
3211
3212 /*
3213  * Get a request and submit it to the controller.
3214  */
3215 static void do_cciss_request(struct request_queue *q)
3216 {
3217         ctlr_info_t *h = q->queuedata;
3218         CommandList_struct *c;
3219         sector_t start_blk;
3220         int seg;
3221         struct request *creq;
3222         u64bit temp64;
3223         struct scatterlist *tmp_sg;
3224         SGDescriptor_struct *curr_sg;
3225         drive_info_struct *drv;
3226         int i, dir;
3227         int sg_index = 0;
3228         int chained = 0;
3229
3230         /* We call start_io here in case there is a command waiting on the
3231          * queue that has not been sent.
3232          */
3233         if (blk_queue_plugged(q))
3234                 goto startio;
3235
3236       queue:
3237         creq = blk_peek_request(q);
3238         if (!creq)
3239                 goto startio;
3240
3241         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3242
3243         if ((c = cmd_alloc(h, 1)) == NULL)
3244                 goto full;
3245
3246         blk_start_request(creq);
3247
3248         tmp_sg = h->scatter_list[c->cmdindex];
3249         spin_unlock_irq(q->queue_lock);
3250
3251         c->cmd_type = CMD_RWREQ;
3252         c->rq = creq;
3253
3254         /* fill in the request */
3255         drv = creq->rq_disk->private_data;
3256         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3257         /* got command from pool, so use the command block index instead */
3258         /* for direct lookups. */
3259         /* The first 2 bits are reserved for controller error reporting. */
3260         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3261         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3262         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3263         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3264         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3265         c->Request.Type.Attribute = ATTR_SIMPLE;
3266         c->Request.Type.Direction =
3267             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3268         c->Request.Timeout = 0; /* Don't time out */
3269         c->Request.CDB[0] =
3270             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3271         start_blk = blk_rq_pos(creq);
3272 #ifdef CCISS_DEBUG
3273         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3274                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3275 #endif                          /* CCISS_DEBUG */
3276
3277         sg_init_table(tmp_sg, h->maxsgentries);
3278         seg = blk_rq_map_sg(q, creq, tmp_sg);
3279
3280         /* get the DMA records for the setup */
3281         if (c->Request.Type.Direction == XFER_READ)
3282                 dir = PCI_DMA_FROMDEVICE;
3283         else
3284                 dir = PCI_DMA_TODEVICE;
3285
3286         curr_sg = c->SG;
3287         sg_index = 0;
3288         chained = 0;
3289
3290         for (i = 0; i < seg; i++) {
3291                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3292                         !chained && ((seg - i) > 1)) {
3293                         /* Point to next chain block. */
3294                         curr_sg = h->cmd_sg_list[c->cmdindex];
3295                         sg_index = 0;
3296                         chained = 1;
3297                 }
3298                 curr_sg[sg_index].Len = tmp_sg[i].length;
3299                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3300                                                 tmp_sg[i].offset,
3301                                                 tmp_sg[i].length, dir);
3302                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3303                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3304                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3305                 ++sg_index;
3306         }
3307         if (chained)
3308                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3309                         (seg - (h->max_cmd_sgentries - 1)) *
3310                                 sizeof(SGDescriptor_struct));
3311
3312         /* track how many SG entries we are using */
3313         if (seg > h->maxSG)
3314                 h->maxSG = seg;
3315
3316 #ifdef CCISS_DEBUG
3317         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3318                         "chained[%d]\n",
3319                         blk_rq_sectors(creq), seg, chained);
3320 #endif                          /* CCISS_DEBUG */
3321
3322         c->Header.SGTotal = seg + chained;
3323         if (seg <= h->max_cmd_sgentries)
3324                 c->Header.SGList = c->Header.SGTotal;
3325         else
3326                 c->Header.SGList = h->max_cmd_sgentries;
3327         set_performant_mode(h, c);
3328
3329         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3330                 if(h->cciss_read == CCISS_READ_10) {
3331                         c->Request.CDB[1] = 0;
3332                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3333                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3334                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3335                         c->Request.CDB[5] = start_blk & 0xff;
3336                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3337                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3338                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3339                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3340                 } else {
3341                         u32 upper32 = upper_32_bits(start_blk);
3342
3343                         c->Request.CDBLen = 16;
3344                         c->Request.CDB[1]= 0;
3345                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3346                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3347                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3348                         c->Request.CDB[5]= upper32 & 0xff;
3349                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3350                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3351                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3352                         c->Request.CDB[9]= start_blk & 0xff;
3353                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3354                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3355                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3356                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3357                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3358                 }
3359         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3360                 c->Request.CDBLen = creq->cmd_len;
3361                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3362         } else {
3363                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3364                 BUG();
3365         }
3366
3367         spin_lock_irq(q->queue_lock);
3368
3369         addQ(&h->reqQ, c);
3370         h->Qdepth++;
3371         if (h->Qdepth > h->maxQsinceinit)
3372                 h->maxQsinceinit = h->Qdepth;
3373
3374         goto queue;
3375 full:
3376         blk_stop_queue(q);
3377 startio:
3378         /* We will already have the driver lock here so not need
3379          * to lock it.
3380          */
3381         start_io(h);
3382 }
3383
3384 static inline unsigned long get_next_completion(ctlr_info_t *h)
3385 {
3386         return h->access.command_completed(h);
3387 }
3388
3389 static inline int interrupt_pending(ctlr_info_t *h)
3390 {
3391         return h->access.intr_pending(h);
3392 }
3393
3394 static inline long interrupt_not_for_us(ctlr_info_t *h)
3395 {
3396         return !(h->msi_vector || h->msix_vector) &&
3397                 ((h->access.intr_pending(h) == 0) ||
3398                 (h->interrupts_enabled == 0));
3399 }
3400
3401 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3402                         u32 raw_tag)
3403 {
3404         if (unlikely(tag_index >= h->nr_cmds)) {
3405                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3406                 return 1;
3407         }
3408         return 0;
3409 }
3410
3411 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3412                                 u32 raw_tag)
3413 {
3414         removeQ(c);
3415         if (likely(c->cmd_type == CMD_RWREQ))
3416                 complete_command(h, c, 0);
3417         else if (c->cmd_type == CMD_IOCTL_PEND)
3418                 complete(c->waiting);
3419 #ifdef CONFIG_CISS_SCSI_TAPE
3420         else if (c->cmd_type == CMD_SCSI)
3421                 complete_scsi_command(c, 0, raw_tag);
3422 #endif
3423 }
3424
3425 static inline u32 next_command(ctlr_info_t *h)
3426 {
3427         u32 a;
3428
3429         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3430                 return h->access.command_completed(h);
3431
3432         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3433                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3434                 (h->reply_pool_head)++;
3435                 h->commands_outstanding--;
3436         } else {
3437                 a = FIFO_EMPTY;
3438         }
3439         /* Check for wraparound */
3440         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3441                 h->reply_pool_head = h->reply_pool;
3442                 h->reply_pool_wraparound ^= 1;
3443         }
3444         return a;
3445 }
3446
3447 /* process completion of an indexed ("direct lookup") command */
3448 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3449 {
3450         u32 tag_index;
3451         CommandList_struct *c;
3452
3453         tag_index = cciss_tag_to_index(raw_tag);
3454         if (bad_tag(h, tag_index, raw_tag))
3455                 return next_command(h);
3456         c = h->cmd_pool + tag_index;
3457         finish_cmd(h, c, raw_tag);
3458         return next_command(h);
3459 }
3460
3461 /* process completion of a non-indexed command */
3462 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3463 {
3464         u32 tag;
3465         CommandList_struct *c = NULL;
3466         struct hlist_node *tmp;
3467         __u32 busaddr_masked, tag_masked;
3468
3469         tag = cciss_tag_discard_error_bits(raw_tag);
3470         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3471                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3472                 tag_masked = cciss_tag_discard_error_bits(tag);
3473                 if (busaddr_masked == tag_masked) {
3474                         finish_cmd(h, c, raw_tag);
3475                         return next_command(h);
3476                 }
3477         }
3478         bad_tag(h, h->nr_cmds + 1, raw_tag);
3479         return next_command(h);
3480 }
3481
3482 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3483 {
3484         ctlr_info_t *h = dev_id;
3485         unsigned long flags;
3486         u32 raw_tag;
3487
3488         if (interrupt_not_for_us(h))
3489                 return IRQ_NONE;
3490         /*
3491          * If there are completed commands in the completion queue,
3492          * we had better do something about it.
3493          */
3494         spin_lock_irqsave(&h->lock, flags);
3495         while (interrupt_pending(h)) {
3496                 raw_tag = get_next_completion(h);
3497                 while (raw_tag != FIFO_EMPTY) {
3498                         if (cciss_tag_contains_index(raw_tag))
3499                                 raw_tag = process_indexed_cmd(h, raw_tag);
3500                         else
3501                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3502                 }
3503         }
3504
3505         spin_unlock_irqrestore(&h->lock, flags);
3506         return IRQ_HANDLED;
3507 }
3508
3509 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3510  * check the interrupt pending register because it is not set.
3511  */
3512 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3513 {
3514         ctlr_info_t *h = dev_id;
3515         unsigned long flags;
3516         u32 raw_tag;
3517
3518         if (interrupt_not_for_us(h))
3519                 return IRQ_NONE;
3520         /*
3521          * If there are completed commands in the completion queue,
3522          * we had better do something about it.
3523          */
3524         spin_lock_irqsave(&h->lock, flags);
3525         raw_tag = get_next_completion(h);
3526         while (raw_tag != FIFO_EMPTY) {
3527                 if (cciss_tag_contains_index(raw_tag))
3528                         raw_tag = process_indexed_cmd(h, raw_tag);
3529                 else
3530                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3531         }
3532
3533         spin_unlock_irqrestore(&h->lock, flags);
3534         return IRQ_HANDLED;
3535 }
3536
3537 /**
3538  * add_to_scan_list() - add controller to rescan queue
3539  * @h:                Pointer to the controller.
3540  *
3541  * Adds the controller to the rescan queue if not already on the queue.
3542  *
3543  * returns 1 if added to the queue, 0 if skipped (could be on the
3544  * queue already, or the controller could be initializing or shutting
3545  * down).
3546  **/
3547 static int add_to_scan_list(struct ctlr_info *h)
3548 {
3549         struct ctlr_info *test_h;
3550         int found = 0;
3551         int ret = 0;
3552
3553         if (h->busy_initializing)
3554                 return 0;
3555
3556         if (!mutex_trylock(&h->busy_shutting_down))
3557                 return 0;
3558
3559         mutex_lock(&scan_mutex);
3560         list_for_each_entry(test_h, &scan_q, scan_list) {
3561                 if (test_h == h) {
3562                         found = 1;
3563                         break;
3564                 }
3565         }
3566         if (!found && !h->busy_scanning) {
3567                 INIT_COMPLETION(h->scan_wait);
3568                 list_add_tail(&h->scan_list, &scan_q);
3569                 ret = 1;
3570         }
3571         mutex_unlock(&scan_mutex);
3572         mutex_unlock(&h->busy_shutting_down);
3573
3574         return ret;
3575 }
3576
3577 /**
3578  * remove_from_scan_list() - remove controller from rescan queue
3579  * @h:                     Pointer to the controller.
3580  *
3581  * Removes the controller from the rescan queue if present. Blocks if
3582  * the controller is currently conducting a rescan.  The controller
3583  * can be in one of three states:
3584  * 1. Doesn't need a scan
3585  * 2. On the scan list, but not scanning yet (we remove it)
3586  * 3. Busy scanning (and not on the list). In this case we want to wait for
3587  *    the scan to complete to make sure the scanning thread for this
3588  *    controller is completely idle.
3589  **/
3590 static void remove_from_scan_list(struct ctlr_info *h)
3591 {
3592         struct ctlr_info *test_h, *tmp_h;
3593
3594         mutex_lock(&scan_mutex);
3595         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3596                 if (test_h == h) { /* state 2. */
3597                         list_del(&h->scan_list);
3598                         complete_all(&h->scan_wait);
3599                         mutex_unlock(&scan_mutex);
3600                         return;
3601                 }
3602         }
3603         if (h->busy_scanning) { /* state 3. */
3604                 mutex_unlock(&scan_mutex);
3605                 wait_for_completion(&h->scan_wait);
3606         } else { /* state 1, nothing to do. */
3607                 mutex_unlock(&scan_mutex);
3608         }
3609 }
3610
3611 /**
3612  * scan_thread() - kernel thread used to rescan controllers
3613  * @data:        Ignored.
3614  *
3615  * A kernel thread used scan for drive topology changes on
3616  * controllers. The thread processes only one controller at a time
3617  * using a queue.  Controllers are added to the queue using
3618  * add_to_scan_list() and removed from the queue either after done
3619  * processing or using remove_from_scan_list().
3620  *
3621  * returns 0.
3622  **/
3623 static int scan_thread(void *data)
3624 {
3625         struct ctlr_info *h;
3626
3627         while (1) {
3628                 set_current_state(TASK_INTERRUPTIBLE);
3629                 schedule();
3630                 if (kthread_should_stop())
3631                         break;
3632
3633                 while (1) {
3634                         mutex_lock(&scan_mutex);
3635                         if (list_empty(&scan_q)) {
3636                                 mutex_unlock(&scan_mutex);
3637                                 break;
3638                         }
3639
3640                         h = list_entry(scan_q.next,
3641                                        struct ctlr_info,
3642                                        scan_list);
3643                         list_del(&h->scan_list);
3644                         h->busy_scanning = 1;
3645                         mutex_unlock(&scan_mutex);
3646
3647                         rebuild_lun_table(h, 0, 0);
3648                         complete_all(&h->scan_wait);
3649                         mutex_lock(&scan_mutex);
3650                         h->busy_scanning = 0;
3651                         mutex_unlock(&scan_mutex);
3652                 }
3653         }
3654
3655         return 0;
3656 }
3657
3658 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3659 {
3660         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3661                 return 0;
3662
3663         switch (c->err_info->SenseInfo[12]) {
3664         case STATE_CHANGED:
3665                 printk(KERN_WARNING "cciss%d: a state change "
3666                         "detected, command retried\n", h->ctlr);
3667                 return 1;
3668         break;
3669         case LUN_FAILED:
3670                 printk(KERN_WARNING "cciss%d: LUN failure "
3671                         "detected, action required\n", h->ctlr);
3672                 return 1;
3673         break;
3674         case REPORT_LUNS_CHANGED:
3675                 printk(KERN_WARNING "cciss%d: report LUN data "
3676                         "changed\n", h->ctlr);
3677         /*
3678          * Here, we could call add_to_scan_list and wake up the scan thread,
3679          * except that it's quite likely that we will get more than one
3680          * REPORT_LUNS_CHANGED condition in quick succession, which means
3681          * that those which occur after the first one will likely happen
3682          * *during* the scan_thread's rescan.  And the rescan code is not
3683          * robust enough to restart in the middle, undoing what it has already
3684          * done, and it's not clear that it's even possible to do this, since
3685          * part of what it does is notify the block layer, which starts
3686          * doing it's own i/o to read partition tables and so on, and the
3687          * driver doesn't have visibility to know what might need undoing.
3688          * In any event, if possible, it is horribly complicated to get right
3689          * so we just don't do it for now.
3690          *
3691          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3692          */
3693                 return 1;
3694         break;
3695         case POWER_OR_RESET:
3696                 printk(KERN_WARNING "cciss%d: a power on "
3697                         "or device reset detected\n", h->ctlr);
3698                 return 1;
3699         break;
3700         case UNIT_ATTENTION_CLEARED:
3701                 printk(KERN_WARNING "cciss%d: unit attention "
3702                     "cleared by another initiator\n", h->ctlr);
3703                 return 1;
3704         break;
3705         default:
3706                 printk(KERN_WARNING "cciss%d: unknown "
3707                         "unit attention detected\n", h->ctlr);
3708                                 return 1;
3709         }
3710 }
3711
3712 /*
3713  *  We cannot read the structure directly, for portability we must use
3714  *   the io functions.
3715  *   This is for debug only.
3716  */
3717 static void print_cfg_table(CfgTable_struct *tb)
3718 {
3719 #ifdef CCISS_DEBUG
3720         int i;
3721         char temp_name[17];
3722
3723         printk("Controller Configuration information\n");
3724         printk("------------------------------------\n");
3725         for (i = 0; i < 4; i++)
3726                 temp_name[i] = readb(&(tb->Signature[i]));
3727         temp_name[4] = '\0';
3728         printk("   Signature = %s\n", temp_name);
3729         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3730         printk("   Transport methods supported = 0x%x\n",
3731                readl(&(tb->TransportSupport)));
3732         printk("   Transport methods active = 0x%x\n",
3733                readl(&(tb->TransportActive)));
3734         printk("   Requested transport Method = 0x%x\n",
3735                readl(&(tb->HostWrite.TransportRequest)));
3736         printk("   Coalesce Interrupt Delay = 0x%x\n",
3737                readl(&(tb->HostWrite.CoalIntDelay)));
3738         printk("   Coalesce Interrupt Count = 0x%x\n",
3739                readl(&(tb->HostWrite.CoalIntCount)));
3740         printk("   Max outstanding commands = 0x%d\n",
3741                readl(&(tb->CmdsOutMax)));
3742         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3743         for (i = 0; i < 16; i++)
3744                 temp_name[i] = readb(&(tb->ServerName[i]));
3745         temp_name[16] = '\0';
3746         printk("   Server Name = %s\n", temp_name);
3747         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3748 #endif                          /* CCISS_DEBUG */
3749 }
3750
3751 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3752 {
3753         int i, offset, mem_type, bar_type;
3754         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3755                 return 0;
3756         offset = 0;
3757         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3758                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3759                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3760                         offset += 4;
3761                 else {
3762                         mem_type = pci_resource_flags(pdev, i) &
3763                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3764                         switch (mem_type) {
3765                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3766                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3767                                 offset += 4;    /* 32 bit */
3768                                 break;
3769                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3770                                 offset += 8;
3771                                 break;
3772                         default:        /* reserved in PCI 2.2 */
3773                                 printk(KERN_WARNING
3774                                        "Base address is invalid\n");
3775                                 return -1;
3776                                 break;
3777                         }
3778                 }
3779                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3780                         return i + 1;
3781         }
3782         return -1;
3783 }
3784
3785 /* Fill in bucket_map[], given nsgs (the max number of
3786  * scatter gather elements supported) and bucket[],
3787  * which is an array of 8 integers.  The bucket[] array
3788  * contains 8 different DMA transfer sizes (in 16
3789  * byte increments) which the controller uses to fetch
3790  * commands.  This function fills in bucket_map[], which
3791  * maps a given number of scatter gather elements to one of
3792  * the 8 DMA transfer sizes.  The point of it is to allow the
3793  * controller to only do as much DMA as needed to fetch the
3794  * command, with the DMA transfer size encoded in the lower
3795  * bits of the command address.
3796  */
3797 static void  calc_bucket_map(int bucket[], int num_buckets,
3798         int nsgs, int *bucket_map)
3799 {
3800         int i, j, b, size;
3801
3802         /* even a command with 0 SGs requires 4 blocks */
3803 #define MINIMUM_TRANSFER_BLOCKS 4
3804 #define NUM_BUCKETS 8
3805         /* Note, bucket_map must have nsgs+1 entries. */
3806         for (i = 0; i <= nsgs; i++) {
3807                 /* Compute size of a command with i SG entries */
3808                 size = i + MINIMUM_TRANSFER_BLOCKS;
3809                 b = num_buckets; /* Assume the biggest bucket */
3810                 /* Find the bucket that is just big enough */
3811                 for (j = 0; j < 8; j++) {
3812                         if (bucket[j] >= size) {
3813                                 b = j;
3814                                 break;
3815                         }
3816                 }
3817                 /* for a command with i SG entries, use bucket b. */
3818                 bucket_map[i] = b;
3819         }
3820 }
3821
3822 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3823 {
3824         int i;
3825
3826         /* under certain very rare conditions, this can take awhile.
3827          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3828          * as we enter this code.) */
3829         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3830                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3831                         break;
3832                 msleep(10);
3833         }
3834 }
3835
3836 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3837 {
3838         /* This is a bit complicated.  There are 8 registers on
3839          * the controller which we write to to tell it 8 different
3840          * sizes of commands which there may be.  It's a way of
3841          * reducing the DMA done to fetch each command.  Encoded into
3842          * each command's tag are 3 bits which communicate to the controller
3843          * which of the eight sizes that command fits within.  The size of
3844          * each command depends on how many scatter gather entries there are.
3845          * Each SG entry requires 16 bytes.  The eight registers are programmed
3846          * with the number of 16-byte blocks a command of that size requires.
3847          * The smallest command possible requires 5 such 16 byte blocks.
3848          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3849          * blocks.  Note, this only extends to the SG entries contained
3850          * within the command block, and does not extend to chained blocks
3851          * of SG elements.   bft[] contains the eight values we write to
3852          * the registers.  They are not evenly distributed, but have more
3853          * sizes for small commands, and fewer sizes for larger commands.
3854          */
3855         __u32 trans_offset;
3856         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3857                         /*
3858                          *  5 = 1 s/g entry or 4k
3859                          *  6 = 2 s/g entry or 8k
3860                          *  8 = 4 s/g entry or 16k
3861                          * 10 = 6 s/g entry or 24k
3862                          */
3863         unsigned long register_value;
3864         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3865
3866         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3867
3868         /* Controller spec: zero out this buffer. */
3869         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3870         h->reply_pool_head = h->reply_pool;
3871
3872         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3873         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3874                                 h->blockFetchTable);
3875         writel(bft[0], &h->transtable->BlockFetch0);
3876         writel(bft[1], &h->transtable->BlockFetch1);
3877         writel(bft[2], &h->transtable->BlockFetch2);
3878         writel(bft[3], &h->transtable->BlockFetch3);
3879         writel(bft[4], &h->transtable->BlockFetch4);
3880         writel(bft[5], &h->transtable->BlockFetch5);
3881         writel(bft[6], &h->transtable->BlockFetch6);
3882         writel(bft[7], &h->transtable->BlockFetch7);
3883
3884         /* size of controller ring buffer */
3885         writel(h->max_commands, &h->transtable->RepQSize);
3886         writel(1, &h->transtable->RepQCount);
3887         writel(0, &h->transtable->RepQCtrAddrLow32);
3888         writel(0, &h->transtable->RepQCtrAddrHigh32);
3889         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3890         writel(0, &h->transtable->RepQAddr0High32);
3891         writel(CFGTBL_Trans_Performant,
3892                         &(h->cfgtable->HostWrite.TransportRequest));
3893
3894         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3895         cciss_wait_for_mode_change_ack(h);
3896         register_value = readl(&(h->cfgtable->TransportActive));
3897         if (!(register_value & CFGTBL_Trans_Performant))
3898                 printk(KERN_WARNING "cciss: unable to get board into"
3899                                         " performant mode\n");
3900 }
3901
3902 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3903 {
3904         __u32 trans_support;
3905
3906         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3907         /* Attempt to put controller into performant mode if supported */
3908         /* Does board support performant mode? */
3909         trans_support = readl(&(h->cfgtable->TransportSupport));
3910         if (!(trans_support & PERFORMANT_MODE))
3911                 return;
3912
3913         printk(KERN_WARNING "cciss%d: Placing controller into "
3914                                 "performant mode\n", h->ctlr);
3915         /* Performant mode demands commands on a 32 byte boundary
3916          * pci_alloc_consistent aligns on page boundarys already.
3917          * Just need to check if divisible by 32
3918          */
3919         if ((sizeof(CommandList_struct) % 32) != 0) {
3920                 printk(KERN_WARNING "%s %d %s\n",
3921                         "cciss info: command size[",
3922                         (int)sizeof(CommandList_struct),
3923                         "] not divisible by 32, no performant mode..\n");
3924                 return;
3925         }
3926
3927         /* Performant mode ring buffer and supporting data structures */
3928         h->reply_pool = (__u64 *)pci_alloc_consistent(
3929                 h->pdev, h->max_commands * sizeof(__u64),
3930                 &(h->reply_pool_dhandle));
3931
3932         /* Need a block fetch table for performant mode */
3933         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3934                 sizeof(__u32)), GFP_KERNEL);
3935
3936         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3937                 goto clean_up;
3938
3939         cciss_enter_performant_mode(h);
3940
3941         /* Change the access methods to the performant access methods */
3942         h->access = SA5_performant_access;
3943         h->transMethod = CFGTBL_Trans_Performant;
3944
3945         return;
3946 clean_up:
3947         kfree(h->blockFetchTable);
3948         if (h->reply_pool)
3949                 pci_free_consistent(h->pdev,
3950                                 h->max_commands * sizeof(__u64),
3951                                 h->reply_pool,
3952                                 h->reply_pool_dhandle);
3953         return;
3954
3955 } /* cciss_put_controller_into_performant_mode */
3956
3957 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3958  * controllers that are capable. If not, we use IO-APIC mode.
3959  */
3960
3961 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3962 {
3963 #ifdef CONFIG_PCI_MSI
3964         int err;
3965         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3966         {0, 2}, {0, 3}
3967         };
3968
3969         /* Some boards advertise MSI but don't really support it */
3970         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3971             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3972                 goto default_int_mode;
3973
3974         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3975                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3976                 if (!err) {
3977                         h->intr[0] = cciss_msix_entries[0].vector;
3978                         h->intr[1] = cciss_msix_entries[1].vector;
3979                         h->intr[2] = cciss_msix_entries[2].vector;
3980                         h->intr[3] = cciss_msix_entries[3].vector;
3981                         h->msix_vector = 1;
3982                         return;
3983                 }
3984                 if (err > 0) {
3985                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3986                                "available\n", err);
3987                         goto default_int_mode;
3988                 } else {
3989                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3990                                err);
3991                         goto default_int_mode;
3992                 }
3993         }
3994         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3995                 if (!pci_enable_msi(h->pdev))
3996                         h->msi_vector = 1;
3997                 else
3998                         printk(KERN_WARNING "cciss: MSI init failed\n");
3999         }
4000 default_int_mode:
4001 #endif                          /* CONFIG_PCI_MSI */
4002         /* if we get here we're going to use the default interrupt mode */
4003         h->intr[PERF_MODE_INT] = h->pdev->irq;
4004         return;
4005 }
4006
4007 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
4008 {
4009         int i;
4010         u32 subsystem_vendor_id, subsystem_device_id;
4011
4012         subsystem_vendor_id = pdev->subsystem_vendor;
4013         subsystem_device_id = pdev->subsystem_device;
4014         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
4015                         subsystem_vendor_id;
4016
4017         for (i = 0; i < ARRAY_SIZE(products); i++) {
4018                 /* Stand aside for hpsa driver on request */
4019                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
4020                         return -ENODEV;
4021                 if (*board_id == products[i].board_id)
4022                         return i;
4023         }
4024         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4025                 *board_id);
4026         return -ENODEV;
4027 }
4028
4029 static inline bool cciss_board_disabled(ctlr_info_t *h)
4030 {
4031         u16 command;
4032
4033         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4034         return ((command & PCI_COMMAND_MEMORY) == 0);
4035 }
4036
4037 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4038         unsigned long *memory_bar)
4039 {
4040         int i;
4041
4042         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4043                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4044                         /* addressing mode bits already removed */
4045                         *memory_bar = pci_resource_start(pdev, i);
4046                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4047                                 *memory_bar);
4048                         return 0;
4049                 }
4050         dev_warn(&pdev->dev, "no memory BAR found\n");
4051         return -ENODEV;
4052 }
4053
4054 static int __devinit cciss_wait_for_board_ready(ctlr_info_t *h)
4055 {
4056         int i;
4057         u32 scratchpad;
4058
4059         for (i = 0; i < CCISS_BOARD_READY_ITERATIONS; i++) {
4060                 scratchpad = readl(h->vaddr + SA5_SCRATCHPAD_OFFSET);
4061                 if (scratchpad == CCISS_FIRMWARE_READY)
4062                         return 0;
4063                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4064         }
4065         dev_warn(&h->pdev->dev, "board not ready, timed out.\n");
4066         return -ENODEV;
4067 }
4068
4069 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4070         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4071         u64 *cfg_offset)
4072 {
4073         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4074         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4075         *cfg_base_addr &= (u32) 0x0000ffff;
4076         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4077         if (*cfg_base_addr_index == -1) {
4078                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4079                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4080                 return -ENODEV;
4081         }
4082         return 0;
4083 }
4084
4085 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4086 {
4087         u64 cfg_offset;
4088         u32 cfg_base_addr;
4089         u64 cfg_base_addr_index;
4090         u32 trans_offset;
4091         int rc;
4092
4093         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4094                 &cfg_base_addr_index, &cfg_offset);
4095         if (rc)
4096                 return rc;
4097         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4098                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4099         if (!h->cfgtable)
4100                 return -ENOMEM;
4101         /* Find performant mode table. */
4102         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4103         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4104                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4105                                 sizeof(*h->transtable));
4106         if (!h->transtable)
4107                 return -ENOMEM;
4108         return 0;
4109 }
4110
4111 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4112 {
4113         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4114         if (h->max_commands < 16) {
4115                 dev_warn(&h->pdev->dev, "Controller reports "
4116                         "max supported commands of %d, an obvious lie. "
4117                         "Using 16.  Ensure that firmware is up to date.\n",
4118                         h->max_commands);
4119                 h->max_commands = 16;
4120         }
4121 }
4122
4123 /* Interrogate the hardware for some limits:
4124  * max commands, max SG elements without chaining, and with chaining,
4125  * SG chain block size, etc.
4126  */
4127 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4128 {
4129         cciss_get_max_perf_mode_cmds(h);
4130         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4131         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4132         /*
4133          * Limit in-command s/g elements to 32 save dma'able memory.
4134          * Howvever spec says if 0, use 31
4135          */
4136         h->max_cmd_sgentries = 31;
4137         if (h->maxsgentries > 512) {
4138                 h->max_cmd_sgentries = 32;
4139                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4140                 h->maxsgentries--; /* save one for chain pointer */
4141         } else {
4142                 h->maxsgentries = 31; /* default to traditional values */
4143                 h->chainsize = 0;
4144         }
4145 }
4146
4147 static inline bool CISS_signature_present(ctlr_info_t *h)
4148 {
4149         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4150             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4151             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4152             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4153                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4154                 return false;
4155         }
4156         return true;
4157 }
4158
4159 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4160 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4161 {
4162 #ifdef CONFIG_X86
4163         u32 prefetch;
4164
4165         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4166         prefetch |= 0x100;
4167         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4168 #endif
4169 }
4170
4171 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4172  * in a prefetch beyond physical memory.
4173  */
4174 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4175 {
4176         u32 dma_prefetch;
4177         __u32 dma_refetch;
4178
4179         if (h->board_id != 0x3225103C)
4180                 return;
4181         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4182         dma_prefetch |= 0x8000;
4183         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4184         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4185         dma_refetch |= 0x1;
4186         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4187 }
4188
4189 static int __devinit cciss_pci_init(ctlr_info_t *h)
4190 {
4191         int prod_index, err;
4192
4193         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4194         if (prod_index < 0)
4195                 return -ENODEV;
4196         h->product_name = products[prod_index].product_name;
4197         h->access = *(products[prod_index].access);
4198
4199         if (cciss_board_disabled(h)) {
4200                 printk(KERN_WARNING
4201                        "cciss: controller appears to be disabled\n");
4202                 return -ENODEV;
4203         }
4204         err = pci_enable_device(h->pdev);
4205         if (err) {
4206                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
4207                 return err;
4208         }
4209
4210         err = pci_request_regions(h->pdev, "cciss");
4211         if (err) {
4212                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
4213                        "aborting\n");
4214                 return err;
4215         }
4216
4217 #ifdef CCISS_DEBUG
4218         printk(KERN_INFO "command = %x\n", command);
4219         printk(KERN_INFO "irq = %x\n", h->pdev->irq);
4220         printk(KERN_INFO "board_id = %x\n", h->board_id);
4221 #endif                          /* CCISS_DEBUG */
4222
4223 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4224  * else we use the IO-APIC interrupt assigned to us by system ROM.
4225  */
4226         cciss_interrupt_mode(h);
4227         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4228         if (err)
4229                 goto err_out_free_res;
4230         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4231         if (!h->vaddr) {
4232                 err = -ENOMEM;
4233                 goto err_out_free_res;
4234         }
4235         err = cciss_wait_for_board_ready(h);
4236         if (err)
4237                 goto err_out_free_res;
4238         err = cciss_find_cfgtables(h);
4239         if (err)
4240                 goto err_out_free_res;
4241         print_cfg_table(h->cfgtable);
4242         cciss_find_board_params(h);
4243
4244         if (!CISS_signature_present(h)) {
4245                 err = -ENODEV;
4246                 goto err_out_free_res;
4247         }
4248         cciss_enable_scsi_prefetch(h);
4249         cciss_p600_dma_prefetch_quirk(h);
4250         cciss_put_controller_into_performant_mode(h);
4251         return 0;
4252
4253 err_out_free_res:
4254         /*
4255          * Deliberately omit pci_disable_device(): it does something nasty to
4256          * Smart Array controllers that pci_enable_device does not undo
4257          */
4258         if (h->transtable)
4259                 iounmap(h->transtable);
4260         if (h->cfgtable)
4261                 iounmap(h->cfgtable);
4262         if (h->vaddr)
4263                 iounmap(h->vaddr);
4264         pci_release_regions(h->pdev);
4265         return err;
4266 }
4267
4268 /* Function to find the first free pointer into our hba[] array
4269  * Returns -1 if no free entries are left.
4270  */
4271 static int alloc_cciss_hba(void)
4272 {
4273         int i;
4274
4275         for (i = 0; i < MAX_CTLR; i++) {
4276                 if (!hba[i]) {
4277                         ctlr_info_t *h;
4278
4279                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4280                         if (!h)
4281                                 goto Enomem;
4282                         hba[i] = h;
4283                         return i;
4284                 }
4285         }
4286         printk(KERN_WARNING "cciss: This driver supports a maximum"
4287                " of %d controllers.\n", MAX_CTLR);
4288         return -1;
4289 Enomem:
4290         printk(KERN_ERR "cciss: out of memory.\n");
4291         return -1;
4292 }
4293
4294 static void free_hba(ctlr_info_t *h)
4295 {
4296         int i;
4297
4298         hba[h->ctlr] = NULL;
4299         for (i = 0; i < h->highest_lun + 1; i++)
4300                 if (h->gendisk[i] != NULL)
4301                         put_disk(h->gendisk[i]);
4302         kfree(h);
4303 }
4304
4305 /* Send a message CDB to the firmware. */
4306 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4307 {
4308         typedef struct {
4309                 CommandListHeader_struct CommandHeader;
4310                 RequestBlock_struct Request;
4311                 ErrDescriptor_struct ErrorDescriptor;
4312         } Command;
4313         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4314         Command *cmd;
4315         dma_addr_t paddr64;
4316         uint32_t paddr32, tag;
4317         void __iomem *vaddr;
4318         int i, err;
4319
4320         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4321         if (vaddr == NULL)
4322                 return -ENOMEM;
4323
4324         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4325            CCISS commands, so they must be allocated from the lower 4GiB of
4326            memory. */
4327         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4328         if (err) {
4329                 iounmap(vaddr);
4330                 return -ENOMEM;
4331         }
4332
4333         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4334         if (cmd == NULL) {
4335                 iounmap(vaddr);
4336                 return -ENOMEM;
4337         }
4338
4339         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4340            although there's no guarantee, we assume that the address is at
4341            least 4-byte aligned (most likely, it's page-aligned). */
4342         paddr32 = paddr64;
4343
4344         cmd->CommandHeader.ReplyQueue = 0;
4345         cmd->CommandHeader.SGList = 0;
4346         cmd->CommandHeader.SGTotal = 0;
4347         cmd->CommandHeader.Tag.lower = paddr32;
4348         cmd->CommandHeader.Tag.upper = 0;
4349         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4350
4351         cmd->Request.CDBLen = 16;
4352         cmd->Request.Type.Type = TYPE_MSG;
4353         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4354         cmd->Request.Type.Direction = XFER_NONE;
4355         cmd->Request.Timeout = 0; /* Don't time out */
4356         cmd->Request.CDB[0] = opcode;
4357         cmd->Request.CDB[1] = type;
4358         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4359
4360         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4361         cmd->ErrorDescriptor.Addr.upper = 0;
4362         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4363
4364         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4365
4366         for (i = 0; i < 10; i++) {
4367                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4368                 if ((tag & ~3) == paddr32)
4369                         break;
4370                 schedule_timeout_uninterruptible(HZ);
4371         }
4372
4373         iounmap(vaddr);
4374
4375         /* we leak the DMA buffer here ... no choice since the controller could
4376            still complete the command. */
4377         if (i == 10) {
4378                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4379                         opcode, type);
4380                 return -ETIMEDOUT;
4381         }
4382
4383         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4384
4385         if (tag & 2) {
4386                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4387                         opcode, type);
4388                 return -EIO;
4389         }
4390
4391         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4392                 opcode, type);
4393         return 0;
4394 }
4395
4396 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4397 #define cciss_noop(p) cciss_message(p, 3, 0)
4398
4399 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4400 {
4401 /* the #defines are stolen from drivers/pci/msi.h. */
4402 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4403 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4404
4405         int pos;
4406         u16 control = 0;
4407
4408         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4409         if (pos) {
4410                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4411                 if (control & PCI_MSI_FLAGS_ENABLE) {
4412                         printk(KERN_INFO "cciss: resetting MSI\n");
4413                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4414                 }
4415         }
4416
4417         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4418         if (pos) {
4419                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4420                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4421                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4422                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4423                 }
4424         }
4425
4426         return 0;
4427 }
4428
4429 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4430         void * __iomem vaddr, bool use_doorbell)
4431 {
4432         u16 pmcsr;
4433         int pos;
4434
4435         if (use_doorbell) {
4436                 /* For everything after the P600, the PCI power state method
4437                  * of resetting the controller doesn't work, so we have this
4438                  * other way using the doorbell register.
4439                  */
4440                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4441                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4442                 msleep(1000);
4443         } else { /* Try to do it the PCI power state way */
4444
4445                 /* Quoting from the Open CISS Specification: "The Power
4446                  * Management Control/Status Register (CSR) controls the power
4447                  * state of the device.  The normal operating state is D0,
4448                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4449                  * the controller, place the interface device in D3 then to D0,
4450                  * this causes a secondary PCI reset which will reset the
4451                  * controller." */
4452
4453                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4454                 if (pos == 0) {
4455                         dev_err(&pdev->dev,
4456                                 "cciss_controller_hard_reset: "
4457                                 "PCI PM not supported\n");
4458                         return -ENODEV;
4459                 }
4460                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4461                 /* enter the D3hot power management state */
4462                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4463                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4464                 pmcsr |= PCI_D3hot;
4465                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4466
4467                 msleep(500);
4468
4469                 /* enter the D0 power management state */
4470                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4471                 pmcsr |= PCI_D0;
4472                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4473
4474                 msleep(500);
4475         }
4476         return 0;
4477 }
4478
4479 /* This does a hard reset of the controller using PCI power management
4480  * states or using the doorbell register. */
4481 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4482 {
4483         u16 saved_config_space[32];
4484         u64 cfg_offset;
4485         u32 cfg_base_addr;
4486         u64 cfg_base_addr_index;
4487         void __iomem *vaddr;
4488         unsigned long paddr;
4489         u32 misc_fw_support, active_transport;
4490         int rc, i;
4491         CfgTable_struct __iomem *cfgtable;
4492         bool use_doorbell;
4493         u32 board_id;
4494
4495         /* For controllers as old a the p600, this is very nearly
4496          * the same thing as
4497          *
4498          * pci_save_state(pci_dev);
4499          * pci_set_power_state(pci_dev, PCI_D3hot);
4500          * pci_set_power_state(pci_dev, PCI_D0);
4501          * pci_restore_state(pci_dev);
4502          *
4503          * but we can't use these nice canned kernel routines on
4504          * kexec, because they also check the MSI/MSI-X state in PCI
4505          * configuration space and do the wrong thing when it is
4506          * set/cleared.  Also, the pci_save/restore_state functions
4507          * violate the ordering requirements for restoring the
4508          * configuration space from the CCISS document (see the
4509          * comment below).  So we roll our own ....
4510          *
4511          * For controllers newer than the P600, the pci power state
4512          * method of resetting doesn't work so we have another way
4513          * using the doorbell register.
4514          */
4515
4516         /* Exclude 640x boards.  These are two pci devices in one slot
4517          * which share a battery backed cache module.  One controls the
4518          * cache, the other accesses the cache through the one that controls
4519          * it.  If we reset the one controlling the cache, the other will
4520          * likely not be happy.  Just forbid resetting this conjoined mess.
4521          */
4522         cciss_lookup_board_id(pdev, &board_id);
4523         if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4524                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4525                                 "due to shared cache module.");
4526                 return -ENODEV;
4527         }
4528
4529         for (i = 0; i < 32; i++)
4530                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4531
4532         /* find the first memory BAR, so we can find the cfg table */
4533         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4534         if (rc)
4535                 return rc;
4536         vaddr = remap_pci_mem(paddr, 0x250);
4537         if (!vaddr)
4538                 return -ENOMEM;
4539
4540         /* find cfgtable in order to check if reset via doorbell is supported */
4541         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4542                                         &cfg_base_addr_index, &cfg_offset);
4543         if (rc)
4544                 goto unmap_vaddr;
4545         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4546                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4547         if (!cfgtable) {
4548                 rc = -ENOMEM;
4549                 goto unmap_vaddr;
4550         }
4551
4552         /* If reset via doorbell register is supported, use that. */
4553         misc_fw_support = readl(&cfgtable->misc_fw_support);
4554         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4555
4556         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4557         if (rc)
4558                 goto unmap_cfgtable;
4559
4560         /* Restore the PCI configuration space.  The Open CISS
4561          * Specification says, "Restore the PCI Configuration
4562          * Registers, offsets 00h through 60h. It is important to
4563          * restore the command register, 16-bits at offset 04h,
4564          * last. Do not restore the configuration status register,
4565          * 16-bits at offset 06h."  Note that the offset is 2*i.
4566          */
4567         for (i = 0; i < 32; i++) {
4568                 if (i == 2 || i == 3)
4569                         continue;
4570                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4571         }
4572         wmb();
4573         pci_write_config_word(pdev, 4, saved_config_space[2]);
4574
4575         /* Some devices (notably the HP Smart Array 5i Controller)
4576            need a little pause here */
4577         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4578
4579         /* Controller should be in simple mode at this point.  If it's not,
4580          * It means we're on one of those controllers which doesn't support
4581          * the doorbell reset method and on which the PCI power management reset
4582          * method doesn't work (P800, for example.)
4583          * In those cases, don't try to proceed, as it generally doesn't work.
4584          */
4585         active_transport = readl(&cfgtable->TransportActive);
4586         if (active_transport & PERFORMANT_MODE) {
4587                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4588                         " Ignoring controller.\n");
4589                 rc = -ENODEV;
4590         }
4591
4592 unmap_cfgtable:
4593         iounmap(cfgtable);
4594
4595 unmap_vaddr:
4596         iounmap(vaddr);
4597         return rc;
4598 }
4599
4600 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4601 {
4602         int rc, i;
4603
4604         if (!reset_devices)
4605                 return 0;
4606
4607         /* Reset the controller with a PCI power-cycle or via doorbell */
4608         rc = cciss_kdump_hard_reset_controller(pdev);
4609
4610         /* -ENOTSUPP here means we cannot reset the controller
4611          * but it's already (and still) up and running in
4612          * "performant mode".  Or, it might be 640x, which can't reset
4613          * due to concerns about shared bbwc between 6402/6404 pair.
4614          */
4615         if (rc == -ENOTSUPP)
4616                 return 0; /* just try to do the kdump anyhow. */
4617         if (rc)
4618                 return -ENODEV;
4619         if (cciss_reset_msi(pdev))
4620                 return -ENODEV;
4621
4622         /* Now try to get the controller to respond to a no-op */
4623         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4624                 if (cciss_noop(pdev) == 0)
4625                         break;
4626                 else
4627                         dev_warn(&pdev->dev, "no-op failed%s\n",
4628                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4629                                         "; re-trying" : ""));
4630                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4631         }
4632         return 0;
4633 }
4634
4635 /*
4636  *  This is it.  Find all the controllers and register them.  I really hate
4637  *  stealing all these major device numbers.
4638  *  returns the number of block devices registered.
4639  */
4640 static int __devinit cciss_init_one(struct pci_dev *pdev,
4641                                     const struct pci_device_id *ent)
4642 {
4643         int i;
4644         int j = 0;
4645         int k = 0;
4646         int rc;
4647         int dac, return_code;
4648         InquiryData_struct *inq_buff;
4649         ctlr_info_t *h;
4650
4651         rc = cciss_init_reset_devices(pdev);
4652         if (rc)
4653                 return rc;
4654         i = alloc_cciss_hba();
4655         if (i < 0)
4656                 return -1;
4657
4658         h = hba[i];
4659         h->pdev = pdev;
4660         h->busy_initializing = 1;
4661         INIT_HLIST_HEAD(&h->cmpQ);
4662         INIT_HLIST_HEAD(&h->reqQ);
4663         mutex_init(&h->busy_shutting_down);
4664
4665         if (cciss_pci_init(h) != 0)
4666                 goto clean_no_release_regions;
4667
4668         sprintf(h->devname, "cciss%d", i);
4669         h->ctlr = i;
4670
4671         init_completion(&h->scan_wait);
4672
4673         if (cciss_create_hba_sysfs_entry(h))
4674                 goto clean0;
4675
4676         /* configure PCI DMA stuff */
4677         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4678                 dac = 1;
4679         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4680                 dac = 0;
4681         else {
4682                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4683                 goto clean1;
4684         }
4685
4686         /*
4687          * register with the major number, or get a dynamic major number
4688          * by passing 0 as argument.  This is done for greater than
4689          * 8 controller support.
4690          */
4691         if (i < MAX_CTLR_ORIG)
4692                 h->major = COMPAQ_CISS_MAJOR + i;
4693         rc = register_blkdev(h->major, h->devname);
4694         if (rc == -EBUSY || rc == -EINVAL) {
4695                 printk(KERN_ERR
4696                        "cciss:  Unable to get major number %d for %s "
4697                        "on hba %d\n", h->major, h->devname, i);
4698                 goto clean1;
4699         } else {
4700                 if (i >= MAX_CTLR_ORIG)
4701                         h->major = rc;
4702         }
4703
4704         /* make sure the board interrupts are off */
4705         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4706         if (h->msi_vector || h->msix_vector) {
4707                 if (request_irq(h->intr[PERF_MODE_INT],
4708                                 do_cciss_msix_intr,
4709                                 IRQF_DISABLED, h->devname, h)) {
4710                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4711                                h->intr[PERF_MODE_INT], h->devname);
4712                         goto clean2;
4713                 }
4714         } else {
4715                 if (request_irq(h->intr[PERF_MODE_INT], do_cciss_intx,
4716                                 IRQF_DISABLED, h->devname, h)) {
4717                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4718                                h->intr[PERF_MODE_INT], h->devname);
4719                         goto clean2;
4720                 }
4721         }
4722
4723         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4724                h->devname, pdev->device, pci_name(pdev),
4725                h->intr[PERF_MODE_INT], dac ? "" : " not");
4726
4727         h->cmd_pool_bits =
4728             kmalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4729                         * sizeof(unsigned long), GFP_KERNEL);
4730         h->cmd_pool = (CommandList_struct *)
4731             pci_alloc_consistent(h->pdev,
4732                     h->nr_cmds * sizeof(CommandList_struct),
4733                     &(h->cmd_pool_dhandle));
4734         h->errinfo_pool = (ErrorInfo_struct *)
4735             pci_alloc_consistent(h->pdev,
4736                     h->nr_cmds * sizeof(ErrorInfo_struct),
4737                     &(h->errinfo_pool_dhandle));
4738         if ((h->cmd_pool_bits == NULL)
4739             || (h->cmd_pool == NULL)
4740             || (h->errinfo_pool == NULL)) {
4741                 printk(KERN_ERR "cciss: out of memory");
4742                 goto clean4;
4743         }
4744
4745         /* Need space for temp scatter list */
4746         h->scatter_list = kmalloc(h->max_commands *
4747                                                 sizeof(struct scatterlist *),
4748                                                 GFP_KERNEL);
4749         for (k = 0; k < h->nr_cmds; k++) {
4750                 h->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4751                                                         h->maxsgentries,
4752                                                         GFP_KERNEL);
4753                 if (h->scatter_list[k] == NULL) {
4754                         printk(KERN_ERR "cciss%d: could not allocate "
4755                                 "s/g lists\n", i);
4756                         goto clean4;
4757                 }
4758         }
4759         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4760                 h->chainsize, h->nr_cmds);
4761         if (!h->cmd_sg_list && h->chainsize > 0)
4762                 goto clean4;
4763
4764         spin_lock_init(&h->lock);
4765
4766         /* Initialize the pdev driver private data.
4767            have it point to h.  */
4768         pci_set_drvdata(pdev, h);
4769         /* command and error info recs zeroed out before
4770            they are used */
4771         memset(h->cmd_pool_bits, 0,
4772                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4773                         * sizeof(unsigned long));
4774
4775         h->num_luns = 0;
4776         h->highest_lun = -1;
4777         for (j = 0; j < CISS_MAX_LUN; j++) {
4778                 h->drv[j] = NULL;
4779                 h->gendisk[j] = NULL;
4780         }
4781
4782         cciss_scsi_setup(h);
4783
4784         /* Turn the interrupts on so we can service requests */
4785         h->access.set_intr_mask(h, CCISS_INTR_ON);
4786
4787         /* Get the firmware version */
4788         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4789         if (inq_buff == NULL) {
4790                 printk(KERN_ERR "cciss: out of memory\n");
4791                 goto clean4;
4792         }
4793
4794         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4795                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4796         if (return_code == IO_OK) {
4797                 h->firm_ver[0] = inq_buff->data_byte[32];
4798                 h->firm_ver[1] = inq_buff->data_byte[33];
4799                 h->firm_ver[2] = inq_buff->data_byte[34];
4800                 h->firm_ver[3] = inq_buff->data_byte[35];
4801         } else {         /* send command failed */
4802                 printk(KERN_WARNING "cciss: unable to determine firmware"
4803                         " version of controller\n");
4804         }
4805         kfree(inq_buff);
4806
4807         cciss_procinit(h);
4808
4809         h->cciss_max_sectors = 8192;
4810
4811         rebuild_lun_table(h, 1, 0);
4812         h->busy_initializing = 0;
4813         return 1;
4814
4815 clean4:
4816         kfree(h->cmd_pool_bits);
4817         /* Free up sg elements */
4818         for (k = 0; k < h->nr_cmds; k++)
4819                 kfree(h->scatter_list[k]);
4820         kfree(h->scatter_list);
4821         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4822         if (h->cmd_pool)
4823                 pci_free_consistent(h->pdev,
4824                                     h->nr_cmds * sizeof(CommandList_struct),
4825                                     h->cmd_pool, h->cmd_pool_dhandle);
4826         if (h->errinfo_pool)
4827                 pci_free_consistent(h->pdev,
4828                                     h->nr_cmds * sizeof(ErrorInfo_struct),
4829                                     h->errinfo_pool,
4830                                     h->errinfo_pool_dhandle);
4831         free_irq(h->intr[PERF_MODE_INT], h);
4832 clean2:
4833         unregister_blkdev(h->major, h->devname);
4834 clean1:
4835         cciss_destroy_hba_sysfs_entry(h);
4836 clean0:
4837         pci_release_regions(pdev);
4838 clean_no_release_regions:
4839         h->busy_initializing = 0;
4840
4841         /*
4842          * Deliberately omit pci_disable_device(): it does something nasty to
4843          * Smart Array controllers that pci_enable_device does not undo
4844          */
4845         pci_set_drvdata(pdev, NULL);
4846         free_hba(h);
4847         return -1;
4848 }
4849
4850 static void cciss_shutdown(struct pci_dev *pdev)
4851 {
4852         ctlr_info_t *h;
4853         char *flush_buf;
4854         int return_code;
4855
4856         h = pci_get_drvdata(pdev);
4857         flush_buf = kzalloc(4, GFP_KERNEL);
4858         if (!flush_buf) {
4859                 printk(KERN_WARNING
4860                         "cciss:%d cache not flushed, out of memory.\n",
4861                         h->ctlr);
4862                 return;
4863         }
4864         /* write all data in the battery backed cache to disk */
4865         memset(flush_buf, 0, 4);
4866         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4867                 4, 0, CTLR_LUNID, TYPE_CMD);
4868         kfree(flush_buf);
4869         if (return_code != IO_OK)
4870                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4871                         h->ctlr);
4872         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4873         free_irq(h->intr[PERF_MODE_INT], h);
4874 }
4875
4876 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4877 {
4878         ctlr_info_t *h;
4879         int i, j;
4880
4881         if (pci_get_drvdata(pdev) == NULL) {
4882                 printk(KERN_ERR "cciss: Unable to remove device \n");
4883                 return;
4884         }
4885
4886         h = pci_get_drvdata(pdev);
4887         i = h->ctlr;
4888         if (hba[i] == NULL) {
4889                 printk(KERN_ERR "cciss: device appears to "
4890                        "already be removed\n");
4891                 return;
4892         }
4893
4894         mutex_lock(&h->busy_shutting_down);
4895
4896         remove_from_scan_list(h);
4897         remove_proc_entry(h->devname, proc_cciss);
4898         unregister_blkdev(h->major, h->devname);
4899
4900         /* remove it from the disk list */
4901         for (j = 0; j < CISS_MAX_LUN; j++) {
4902                 struct gendisk *disk = h->gendisk[j];
4903                 if (disk) {
4904                         struct request_queue *q = disk->queue;
4905
4906                         if (disk->flags & GENHD_FL_UP) {
4907                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
4908                                 del_gendisk(disk);
4909                         }
4910                         if (q)
4911                                 blk_cleanup_queue(q);
4912                 }
4913         }
4914
4915 #ifdef CONFIG_CISS_SCSI_TAPE
4916         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
4917 #endif
4918
4919         cciss_shutdown(pdev);
4920
4921 #ifdef CONFIG_PCI_MSI
4922         if (h->msix_vector)
4923                 pci_disable_msix(h->pdev);
4924         else if (h->msi_vector)
4925                 pci_disable_msi(h->pdev);
4926 #endif                          /* CONFIG_PCI_MSI */
4927
4928         iounmap(h->transtable);
4929         iounmap(h->cfgtable);
4930         iounmap(h->vaddr);
4931
4932         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(CommandList_struct),
4933                             h->cmd_pool, h->cmd_pool_dhandle);
4934         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(ErrorInfo_struct),
4935                             h->errinfo_pool, h->errinfo_pool_dhandle);
4936         kfree(h->cmd_pool_bits);
4937         /* Free up sg elements */
4938         for (j = 0; j < h->nr_cmds; j++)
4939                 kfree(h->scatter_list[j]);
4940         kfree(h->scatter_list);
4941         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4942         /*
4943          * Deliberately omit pci_disable_device(): it does something nasty to
4944          * Smart Array controllers that pci_enable_device does not undo
4945          */
4946         pci_release_regions(pdev);
4947         pci_set_drvdata(pdev, NULL);
4948         cciss_destroy_hba_sysfs_entry(h);
4949         mutex_unlock(&h->busy_shutting_down);
4950         free_hba(h);
4951 }
4952
4953 static struct pci_driver cciss_pci_driver = {
4954         .name = "cciss",
4955         .probe = cciss_init_one,
4956         .remove = __devexit_p(cciss_remove_one),
4957         .id_table = cciss_pci_device_id,        /* id_table */
4958         .shutdown = cciss_shutdown,
4959 };
4960
4961 /*
4962  *  This is it.  Register the PCI driver information for the cards we control
4963  *  the OS will call our registered routines when it finds one of our cards.
4964  */
4965 static int __init cciss_init(void)
4966 {
4967         int err;
4968
4969         /*
4970          * The hardware requires that commands are aligned on a 64-bit
4971          * boundary. Given that we use pci_alloc_consistent() to allocate an
4972          * array of them, the size must be a multiple of 8 bytes.
4973          */
4974         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4975         printk(KERN_INFO DRIVER_NAME "\n");
4976
4977         err = bus_register(&cciss_bus_type);
4978         if (err)
4979                 return err;
4980
4981         /* Start the scan thread */
4982         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4983         if (IS_ERR(cciss_scan_thread)) {
4984                 err = PTR_ERR(cciss_scan_thread);
4985                 goto err_bus_unregister;
4986         }
4987
4988         /* Register for our PCI devices */
4989         err = pci_register_driver(&cciss_pci_driver);
4990         if (err)
4991                 goto err_thread_stop;
4992
4993         return err;
4994
4995 err_thread_stop:
4996         kthread_stop(cciss_scan_thread);
4997 err_bus_unregister:
4998         bus_unregister(&cciss_bus_type);
4999
5000         return err;
5001 }
5002
5003 static void __exit cciss_cleanup(void)
5004 {
5005         int i;
5006
5007         pci_unregister_driver(&cciss_pci_driver);
5008         /* double check that all controller entrys have been removed */
5009         for (i = 0; i < MAX_CTLR; i++) {
5010                 if (hba[i] != NULL) {
5011                         printk(KERN_WARNING "cciss: had to remove"
5012                                " controller %d\n", i);
5013                         cciss_remove_one(hba[i]->pdev);
5014                 }
5015         }
5016         kthread_stop(cciss_scan_thread);
5017         remove_proc_entry("driver/cciss", NULL);
5018         bus_unregister(&cciss_bus_type);
5019 }
5020
5021 module_init(cciss_init);
5022 module_exit(cciss_cleanup);