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(int ctlr, int logvol,
 195                         sector_t *total_size, unsigned int *block_size);
 196 static void cciss_read_capacity_16(int ctlr, int logvol,
 197                         sector_t *total_size, unsigned int *block_size);
 198 static void cciss_geometry_inquiry(int ctlr, 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(__u8 cmd, int ctlr, 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(int i);
 233 #else
 234 static void cciss_procinit(int i)
 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->ctlr);
 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 ctlr = h->ctlr;
 427         unsigned long flags;
 428
 429         /* prevent displaying bogus info during configuration
 430          * or deconfiguration of a logical volume
 431          */
 432         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 433         if (h->busy_configuring) {
 434                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 435                 return ERR_PTR(-EBUSY);
 436         }
 437         h->busy_configuring = 1;
 438         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 439
 440         if (*pos == 0)
 441                 cciss_seq_show_header(seq);
 442
 443         return pos;
 444 }
 445
 446 static int cciss_seq_show(struct seq_file *seq, void *v)
 447 {
 448         sector_t vol_sz, vol_sz_frac;
 449         ctlr_info_t *h = seq->private;
 450         unsigned ctlr = h->ctlr;
 451         loff_t *pos = v;
 452         drive_info_struct *drv = h->drv[*pos];
 453
 454         if (*pos > h->highest_lun)
 455                 return 0;
 456
 457         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 458                 return 0;
 459
 460         if (drv->heads == 0)
 461                 return 0;
 462
 463         vol_sz = drv->nr_blocks;
 464         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 465         vol_sz_frac *= 100;
 466         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 467
 468         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 469                 drv->raid_level = RAID_UNKNOWN;
 470         seq_printf(seq, "cciss/c%dd%d:"
 471                         "\t%4u.%02uGB\tRAID %s\n",
 472                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 473                         raid_label[drv->raid_level]);
 474         return 0;
 475 }
 476
 477 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 478 {
 479         ctlr_info_t *h = seq->private;
 480
 481         if (*pos > h->highest_lun)
 482                 return NULL;
 483         *pos += 1;
 484
 485         return pos;
 486 }
 487
 488 static void cciss_seq_stop(struct seq_file *seq, void *v)
 489 {
 490         ctlr_info_t *h = seq->private;
 491
 492         /* Only reset h->busy_configuring if we succeeded in setting
 493          * it during cciss_seq_start. */
 494         if (v == ERR_PTR(-EBUSY))
 495                 return;
 496
 497         h->busy_configuring = 0;
 498 }
 499
 500 static const struct seq_operations cciss_seq_ops = {
 501         .start = cciss_seq_start,
 502         .show  = cciss_seq_show,
 503         .next  = cciss_seq_next,
 504         .stop  = cciss_seq_stop,
 505 };
 506
 507 static int cciss_seq_open(struct inode *inode, struct file *file)
 508 {
 509         int ret = seq_open(file, &cciss_seq_ops);
 510         struct seq_file *seq = file->private_data;
 511
 512         if (!ret)
 513                 seq->private = PDE(inode)->data;
 514
 515         return ret;
 516 }
 517
 518 static ssize_t
 519 cciss_proc_write(struct file *file, const char __user *buf,
 520                  size_t length, loff_t *ppos)
 521 {
 522         int err;
 523         char *buffer;
 524
 525 #ifndef CONFIG_CISS_SCSI_TAPE
 526         return -EINVAL;
 527 #endif
 528
 529         if (!buf || length > PAGE_SIZE - 1)
 530                 return -EINVAL;
 531
 532         buffer = (char *)__get_free_page(GFP_KERNEL);
 533         if (!buffer)
 534                 return -ENOMEM;
 535
 536         err = -EFAULT;
 537         if (copy_from_user(buffer, buf, length))
 538                 goto out;
 539         buffer[length] = '\0';
 540
 541 #ifdef CONFIG_CISS_SCSI_TAPE
 542         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 543                 struct seq_file *seq = file->private_data;
 544                 ctlr_info_t *h = seq->private;
 545
 546                 err = cciss_engage_scsi(h->ctlr);
 547                 if (err == 0)
 548                         err = length;
 549         } else
 550 #endif /* CONFIG_CISS_SCSI_TAPE */
 551                 err = -EINVAL;
 552         /* might be nice to have "disengage" too, but it's not
 553            safely possible. (only 1 module use count, lock issues.) */
 554
 555 out:
 556         free_page((unsigned long)buffer);
 557         return err;
 558 }
 559
 560 static const struct file_operations cciss_proc_fops = {
 561         .owner   = THIS_MODULE,
 562         .open    = cciss_seq_open,
 563         .read    = seq_read,
 564         .llseek  = seq_lseek,
 565         .release = seq_release,
 566         .write   = cciss_proc_write,
 567 };
 568
 569 static void __devinit cciss_procinit(int i)
 570 {
 571         struct proc_dir_entry *pde;
 572
 573         if (proc_cciss == NULL)
 574                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 575         if (!proc_cciss)
 576                 return;
 577         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 578                                         S_IROTH, proc_cciss,
 579                                         &cciss_proc_fops, hba[i]);
 580 }
 581 #endif                          /* CONFIG_PROC_FS */
 582
 583 #define MAX_PRODUCT_NAME_LEN 19
 584
 585 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 586 #define to_drv(n) container_of(n, drive_info_struct, dev)
 587
 588 static ssize_t host_store_rescan(struct device *dev,
 589                                  struct device_attribute *attr,
 590                                  const char *buf, size_t count)
 591 {
 592         struct ctlr_info *h = to_hba(dev);
 593
 594         add_to_scan_list(h);
 595         wake_up_process(cciss_scan_thread);
 596         wait_for_completion_interruptible(&h->scan_wait);
 597
 598         return count;
 599 }
 600 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 601
 602 static ssize_t dev_show_unique_id(struct device *dev,
 603                                  struct device_attribute *attr,
 604                                  char *buf)
 605 {
 606         drive_info_struct *drv = to_drv(dev);
 607         struct ctlr_info *h = to_hba(drv->dev.parent);
 608         __u8 sn[16];
 609         unsigned long flags;
 610         int ret = 0;
 611
 612         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 613         if (h->busy_configuring)
 614                 ret = -EBUSY;
 615         else
 616                 memcpy(sn, drv->serial_no, sizeof(sn));
 617         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 618
 619         if (ret)
 620                 return ret;
 621         else
 622                 return snprintf(buf, 16 * 2 + 2,
 623                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 624                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 625                                 sn[0], sn[1], sn[2], sn[3],
 626                                 sn[4], sn[5], sn[6], sn[7],
 627                                 sn[8], sn[9], sn[10], sn[11],
 628                                 sn[12], sn[13], sn[14], sn[15]);
 629 }
 630 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 631
 632 static ssize_t dev_show_vendor(struct device *dev,
 633                                struct device_attribute *attr,
 634                                char *buf)
 635 {
 636         drive_info_struct *drv = to_drv(dev);
 637         struct ctlr_info *h = to_hba(drv->dev.parent);
 638         char vendor[VENDOR_LEN + 1];
 639         unsigned long flags;
 640         int ret = 0;
 641
 642         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 643         if (h->busy_configuring)
 644                 ret = -EBUSY;
 645         else
 646                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 647         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 648
 649         if (ret)
 650                 return ret;
 651         else
 652                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 653 }
 654 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 655
 656 static ssize_t dev_show_model(struct device *dev,
 657                               struct device_attribute *attr,
 658                               char *buf)
 659 {
 660         drive_info_struct *drv = to_drv(dev);
 661         struct ctlr_info *h = to_hba(drv->dev.parent);
 662         char model[MODEL_LEN + 1];
 663         unsigned long flags;
 664         int ret = 0;
 665
 666         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 667         if (h->busy_configuring)
 668                 ret = -EBUSY;
 669         else
 670                 memcpy(model, drv->model, MODEL_LEN + 1);
 671         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 672
 673         if (ret)
 674                 return ret;
 675         else
 676                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 677 }
 678 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 679
 680 static ssize_t dev_show_rev(struct device *dev,
 681                             struct device_attribute *attr,
 682                             char *buf)
 683 {
 684         drive_info_struct *drv = to_drv(dev);
 685         struct ctlr_info *h = to_hba(drv->dev.parent);
 686         char rev[REV_LEN + 1];
 687         unsigned long flags;
 688         int ret = 0;
 689
 690         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 691         if (h->busy_configuring)
 692                 ret = -EBUSY;
 693         else
 694                 memcpy(rev, drv->rev, REV_LEN + 1);
 695         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 696
 697         if (ret)
 698                 return ret;
 699         else
 700                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 701 }
 702 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 703
 704 static ssize_t cciss_show_lunid(struct device *dev,
 705                                 struct device_attribute *attr, char *buf)
 706 {
 707         drive_info_struct *drv = to_drv(dev);
 708         struct ctlr_info *h = to_hba(drv->dev.parent);
 709         unsigned long flags;
 710         unsigned char lunid[8];
 711
 712         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 713         if (h->busy_configuring) {
 714                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 715                 return -EBUSY;
 716         }
 717         if (!drv->heads) {
 718                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 719                 return -ENOTTY;
 720         }
 721         memcpy(lunid, drv->LunID, sizeof(lunid));
 722         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 723         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 724                 lunid[0], lunid[1], lunid[2], lunid[3],
 725                 lunid[4], lunid[5], lunid[6], lunid[7]);
 726 }
 727 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 728
 729 static ssize_t cciss_show_raid_level(struct device *dev,
 730                                      struct device_attribute *attr, char *buf)
 731 {
 732         drive_info_struct *drv = to_drv(dev);
 733         struct ctlr_info *h = to_hba(drv->dev.parent);
 734         int raid;
 735         unsigned long flags;
 736
 737         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 738         if (h->busy_configuring) {
 739                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 740                 return -EBUSY;
 741         }
 742         raid = drv->raid_level;
 743         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 744         if (raid < 0 || raid > RAID_UNKNOWN)
 745                 raid = RAID_UNKNOWN;
 746
 747         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 748                         raid_label[raid]);
 749 }
 750 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 751
 752 static ssize_t cciss_show_usage_count(struct device *dev,
 753                                       struct device_attribute *attr, char *buf)
 754 {
 755         drive_info_struct *drv = to_drv(dev);
 756         struct ctlr_info *h = to_hba(drv->dev.parent);
 757         unsigned long flags;
 758         int count;
 759
 760         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 761         if (h->busy_configuring) {
 762                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 763                 return -EBUSY;
 764         }
 765         count = drv->usage_count;
 766         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 767         return snprintf(buf, 20, "%d\n", count);
 768 }
 769 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 770
 771 static struct attribute *cciss_host_attrs[] = {
 772         &dev_attr_rescan.attr,
 773         NULL
 774 };
 775
 776 static struct attribute_group cciss_host_attr_group = {
 777         .attrs = cciss_host_attrs,
 778 };
 779
 780 static const struct attribute_group *cciss_host_attr_groups[] = {
 781         &cciss_host_attr_group,
 782         NULL
 783 };
 784
 785 static struct device_type cciss_host_type = {
 786         .name           = "cciss_host",
 787         .groups         = cciss_host_attr_groups,
 788         .release        = cciss_hba_release,
 789 };
 790
 791 static struct attribute *cciss_dev_attrs[] = {
 792         &dev_attr_unique_id.attr,
 793         &dev_attr_model.attr,
 794         &dev_attr_vendor.attr,
 795         &dev_attr_rev.attr,
 796         &dev_attr_lunid.attr,
 797         &dev_attr_raid_level.attr,
 798         &dev_attr_usage_count.attr,
 799         NULL
 800 };
 801
 802 static struct attribute_group cciss_dev_attr_group = {
 803         .attrs = cciss_dev_attrs,
 804 };
 805
 806 static const struct attribute_group *cciss_dev_attr_groups[] = {
 807         &cciss_dev_attr_group,
 808         NULL
 809 };
 810
 811 static struct device_type cciss_dev_type = {
 812         .name           = "cciss_device",
 813         .groups         = cciss_dev_attr_groups,
 814         .release        = cciss_device_release,
 815 };
 816
 817 static struct bus_type cciss_bus_type = {
 818         .name           = "cciss",
 819 };
 820
 821 /*
 822  * cciss_hba_release is called when the reference count
 823  * of h->dev goes to zero.
 824  */
 825 static void cciss_hba_release(struct device *dev)
 826 {
 827         /*
 828          * nothing to do, but need this to avoid a warning
 829          * about not having a release handler from lib/kref.c.
 830          */
 831 }
 832
 833 /*
 834  * Initialize sysfs entry for each controller.  This sets up and registers
 835  * the 'cciss#' directory for each individual controller under
 836  * /sys/bus/pci/devices/<dev>/.
 837  */
 838 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 839 {
 840         device_initialize(&h->dev);
 841         h->dev.type = &cciss_host_type;
 842         h->dev.bus = &cciss_bus_type;
 843         dev_set_name(&h->dev, "%s", h->devname);
 844         h->dev.parent = &h->pdev->dev;
 845
 846         return device_add(&h->dev);
 847 }
 848
 849 /*
 850  * Remove sysfs entries for an hba.
 851  */
 852 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 853 {
 854         device_del(&h->dev);
 855         put_device(&h->dev); /* final put. */
 856 }
 857
 858 /* cciss_device_release is called when the reference count
 859  * of h->drv[x]dev goes to zero.
 860  */
 861 static void cciss_device_release(struct device *dev)
 862 {
 863         drive_info_struct *drv = to_drv(dev);
 864         kfree(drv);
 865 }
 866
 867 /*
 868  * Initialize sysfs for each logical drive.  This sets up and registers
 869  * the 'c#d#' directory for each individual logical drive under
 870  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 871  * /sys/block/cciss!c#d# to this entry.
 872  */
 873 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 874                                        int drv_index)
 875 {
 876         struct device *dev;
 877
 878         if (h->drv[drv_index]->device_initialized)
 879                 return 0;
 880
 881         dev = &h->drv[drv_index]->dev;
 882         device_initialize(dev);
 883         dev->type = &cciss_dev_type;
 884         dev->bus = &cciss_bus_type;
 885         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 886         dev->parent = &h->dev;
 887         h->drv[drv_index]->device_initialized = 1;
 888         return device_add(dev);
 889 }
 890
 891 /*
 892  * Remove sysfs entries for a logical drive.
 893  */
 894 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 895         int ctlr_exiting)
 896 {
 897         struct device *dev = &h->drv[drv_index]->dev;
 898
 899         /* special case for c*d0, we only destroy it on controller exit */
 900         if (drv_index == 0 && !ctlr_exiting)
 901                 return;
 902
 903         device_del(dev);
 904         put_device(dev); /* the "final" put. */
 905         h->drv[drv_index] = NULL;
 906 }
 907
 908 /*
 909  * For operations that cannot sleep, a command block is allocated at init,
 910  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 911  * which ones are free or in use.  For operations that can wait for kmalloc
 912  * to possible sleep, this routine can be called with get_from_pool set to 0.
 913  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 914  */
 915 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 916 {
 917         CommandList_struct *c;
 918         int i;
 919         u64bit temp64;
 920         dma_addr_t cmd_dma_handle, err_dma_handle;
 921
 922         if (!get_from_pool) {
 923                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 924                         sizeof(CommandList_struct), &cmd_dma_handle);
 925                 if (c == NULL)
 926                         return NULL;
 927                 memset(c, 0, sizeof(CommandList_struct));
 928
 929                 c->cmdindex = -1;
 930
 931                 c->err_info = (ErrorInfo_struct *)
 932                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 933                             &err_dma_handle);
 934
 935                 if (c->err_info == NULL) {
 936                         pci_free_consistent(h->pdev,
 937                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 938                         return NULL;
 939                 }
 940                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 941         } else {                /* get it out of the controllers pool */
 942
 943                 do {
 944                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 945                         if (i == h->nr_cmds)
 946                                 return NULL;
 947                 } while (test_and_set_bit
 948                          (i & (BITS_PER_LONG - 1),
 949                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 950 #ifdef CCISS_DEBUG
 951                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 952 #endif
 953                 c = h->cmd_pool + i;
 954                 memset(c, 0, sizeof(CommandList_struct));
 955                 cmd_dma_handle = h->cmd_pool_dhandle
 956                     + i * sizeof(CommandList_struct);
 957                 c->err_info = h->errinfo_pool + i;
 958                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 959                 err_dma_handle = h->errinfo_pool_dhandle
 960                     + i * sizeof(ErrorInfo_struct);
 961                 h->nr_allocs++;
 962
 963                 c->cmdindex = i;
 964         }
 965
 966         INIT_HLIST_NODE(&c->list);
 967         c->busaddr = (__u32) cmd_dma_handle;
 968         temp64.val = (__u64) err_dma_handle;
 969         c->ErrDesc.Addr.lower = temp64.val32.lower;
 970         c->ErrDesc.Addr.upper = temp64.val32.upper;
 971         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 972
 973         c->ctlr = h->ctlr;
 974         return c;
 975 }
 976
 977 /*
 978  * Frees a command block that was previously allocated with cmd_alloc().
 979  */
 980 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 981 {
 982         int i;
 983         u64bit temp64;
 984
 985         if (!got_from_pool) {
 986                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 987                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 988                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 989                                     c->err_info, (dma_addr_t) temp64.val);
 990                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 991                                     c, (dma_addr_t) c->busaddr);
 992         } else {
 993                 i = c - h->cmd_pool;
 994                 clear_bit(i & (BITS_PER_LONG - 1),
 995                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 996                 h->nr_frees++;
 997         }
 998 }
 999
1000 static inline ctlr_info_t *get_host(struct gendisk *disk)
1001 {
1002         return disk->queue->queuedata;
1003 }
1004
1005 static inline drive_info_struct *get_drv(struct gendisk *disk)
1006 {
1007         return disk->private_data;
1008 }
1009
1010 /*
1011  * Open.  Make sure the device is really there.
1012  */
1013 static int cciss_open(struct block_device *bdev, fmode_t mode)
1014 {
1015         ctlr_info_t *host = get_host(bdev->bd_disk);
1016         drive_info_struct *drv = get_drv(bdev->bd_disk);
1017
1018 #ifdef CCISS_DEBUG
1019         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
1020 #endif                          /* CCISS_DEBUG */
1021
1022         if (drv->busy_configuring)
1023                 return -EBUSY;
1024         /*
1025          * Root is allowed to open raw volume zero even if it's not configured
1026          * so array config can still work. Root is also allowed to open any
1027          * volume that has a LUN ID, so it can issue IOCTL to reread the
1028          * disk information.  I don't think I really like this
1029          * but I'm already using way to many device nodes to claim another one
1030          * for "raw controller".
1031          */
1032         if (drv->heads == 0) {
1033                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1034                         /* if not node 0 make sure it is a partition = 0 */
1035                         if (MINOR(bdev->bd_dev) & 0x0f) {
1036                                 return -ENXIO;
1037                                 /* if it is, make sure we have a LUN ID */
1038                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1039                                 sizeof(drv->LunID))) {
1040                                 return -ENXIO;
1041                         }
1042                 }
1043                 if (!capable(CAP_SYS_ADMIN))
1044                         return -EPERM;
1045         }
1046         drv->usage_count++;
1047         host->usage_count++;
1048         return 0;
1049 }
1050
1051 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1052 {
1053         int ret;
1054
1055         lock_kernel();
1056         ret = cciss_open(bdev, mode);
1057         unlock_kernel();
1058
1059         return ret;
1060 }
1061
1062 /*
1063  * Close.  Sync first.
1064  */
1065 static int cciss_release(struct gendisk *disk, fmode_t mode)
1066 {
1067         ctlr_info_t *host;
1068         drive_info_struct *drv;
1069
1070         lock_kernel();
1071         host = get_host(disk);
1072         drv = get_drv(disk);
1073
1074 #ifdef CCISS_DEBUG
1075         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
1076 #endif                          /* CCISS_DEBUG */
1077
1078         drv->usage_count--;
1079         host->usage_count--;
1080         unlock_kernel();
1081         return 0;
1082 }
1083
1084 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1085                     unsigned cmd, unsigned long arg)
1086 {
1087         int ret;
1088         lock_kernel();
1089         ret = cciss_ioctl(bdev, mode, cmd, arg);
1090         unlock_kernel();
1091         return ret;
1092 }
1093
1094 #ifdef CONFIG_COMPAT
1095
1096 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1097                                   unsigned cmd, unsigned long arg);
1098 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1099                                       unsigned cmd, unsigned long arg);
1100
1101 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1102                               unsigned cmd, unsigned long arg)
1103 {
1104         switch (cmd) {
1105         case CCISS_GETPCIINFO:
1106         case CCISS_GETINTINFO:
1107         case CCISS_SETINTINFO:
1108         case CCISS_GETNODENAME:
1109         case CCISS_SETNODENAME:
1110         case CCISS_GETHEARTBEAT:
1111         case CCISS_GETBUSTYPES:
1112         case CCISS_GETFIRMVER:
1113         case CCISS_GETDRIVVER:
1114         case CCISS_REVALIDVOLS:
1115         case CCISS_DEREGDISK:
1116         case CCISS_REGNEWDISK:
1117         case CCISS_REGNEWD:
1118         case CCISS_RESCANDISK:
1119         case CCISS_GETLUNINFO:
1120                 return do_ioctl(bdev, mode, cmd, arg);
1121
1122         case CCISS_PASSTHRU32:
1123                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1124         case CCISS_BIG_PASSTHRU32:
1125                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1126
1127         default:
1128                 return -ENOIOCTLCMD;
1129         }
1130 }
1131
1132 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1133                                   unsigned cmd, unsigned long arg)
1134 {
1135         IOCTL32_Command_struct __user *arg32 =
1136             (IOCTL32_Command_struct __user *) arg;
1137         IOCTL_Command_struct arg64;
1138         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1139         int err;
1140         u32 cp;
1141
1142         err = 0;
1143         err |=
1144             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1145                            sizeof(arg64.LUN_info));
1146         err |=
1147             copy_from_user(&arg64.Request, &arg32->Request,
1148                            sizeof(arg64.Request));
1149         err |=
1150             copy_from_user(&arg64.error_info, &arg32->error_info,
1151                            sizeof(arg64.error_info));
1152         err |= get_user(arg64.buf_size, &arg32->buf_size);
1153         err |= get_user(cp, &arg32->buf);
1154         arg64.buf = compat_ptr(cp);
1155         err |= copy_to_user(p, &arg64, sizeof(arg64));
1156
1157         if (err)
1158                 return -EFAULT;
1159
1160         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1161         if (err)
1162                 return err;
1163         err |=
1164             copy_in_user(&arg32->error_info, &p->error_info,
1165                          sizeof(arg32->error_info));
1166         if (err)
1167                 return -EFAULT;
1168         return err;
1169 }
1170
1171 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1172                                       unsigned cmd, unsigned long arg)
1173 {
1174         BIG_IOCTL32_Command_struct __user *arg32 =
1175             (BIG_IOCTL32_Command_struct __user *) arg;
1176         BIG_IOCTL_Command_struct arg64;
1177         BIG_IOCTL_Command_struct __user *p =
1178             compat_alloc_user_space(sizeof(arg64));
1179         int err;
1180         u32 cp;
1181
1182         err = 0;
1183         err |=
1184             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1185                            sizeof(arg64.LUN_info));
1186         err |=
1187             copy_from_user(&arg64.Request, &arg32->Request,
1188                            sizeof(arg64.Request));
1189         err |=
1190             copy_from_user(&arg64.error_info, &arg32->error_info,
1191                            sizeof(arg64.error_info));
1192         err |= get_user(arg64.buf_size, &arg32->buf_size);
1193         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1194         err |= get_user(cp, &arg32->buf);
1195         arg64.buf = compat_ptr(cp);
1196         err |= copy_to_user(p, &arg64, sizeof(arg64));
1197
1198         if (err)
1199                 return -EFAULT;
1200
1201         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1202         if (err)
1203                 return err;
1204         err |=
1205             copy_in_user(&arg32->error_info, &p->error_info,
1206                          sizeof(arg32->error_info));
1207         if (err)
1208                 return -EFAULT;
1209         return err;
1210 }
1211 #endif
1212
1213 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1214 {
1215         drive_info_struct *drv = get_drv(bdev->bd_disk);
1216
1217         if (!drv->cylinders)
1218                 return -ENXIO;
1219
1220         geo->heads = drv->heads;
1221         geo->sectors = drv->sectors;
1222         geo->cylinders = drv->cylinders;
1223         return 0;
1224 }
1225
1226 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1227 {
1228         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1229                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1230                 (void)check_for_unit_attention(host, c);
1231 }
1232 /*
1233  * ioctl
1234  */
1235 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1236                        unsigned int cmd, unsigned long arg)
1237 {
1238         struct gendisk *disk = bdev->bd_disk;
1239         ctlr_info_t *host = get_host(disk);
1240         drive_info_struct *drv = get_drv(disk);
1241         int ctlr = host->ctlr;
1242         void __user *argp = (void __user *)arg;
1243
1244 #ifdef CCISS_DEBUG
1245         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1246 #endif                          /* CCISS_DEBUG */
1247
1248         switch (cmd) {
1249         case CCISS_GETPCIINFO:
1250                 {
1251                         cciss_pci_info_struct pciinfo;
1252
1253                         if (!arg)
1254                                 return -EINVAL;
1255                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1256                         pciinfo.bus = host->pdev->bus->number;
1257                         pciinfo.dev_fn = host->pdev->devfn;
1258                         pciinfo.board_id = host->board_id;
1259                         if (copy_to_user
1260                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1261                                 return -EFAULT;
1262                         return 0;
1263                 }
1264         case CCISS_GETINTINFO:
1265                 {
1266                         cciss_coalint_struct intinfo;
1267                         if (!arg)
1268                                 return -EINVAL;
1269                         intinfo.delay =
1270                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1271                         intinfo.count =
1272                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1273                         if (copy_to_user
1274                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1275                                 return -EFAULT;
1276                         return 0;
1277                 }
1278         case CCISS_SETINTINFO:
1279                 {
1280                         cciss_coalint_struct intinfo;
1281                         unsigned long flags;
1282                         int i;
1283
1284                         if (!arg)
1285                                 return -EINVAL;
1286                         if (!capable(CAP_SYS_ADMIN))
1287                                 return -EPERM;
1288                         if (copy_from_user
1289                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1290                                 return -EFAULT;
1291                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1292                         {
1293 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1294                                 return -EINVAL;
1295                         }
1296                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1297                         /* Update the field, and then ring the doorbell */
1298                         writel(intinfo.delay,
1299                                &(host->cfgtable->HostWrite.CoalIntDelay));
1300                         writel(intinfo.count,
1301                                &(host->cfgtable->HostWrite.CoalIntCount));
1302                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1303
1304                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1305                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1306                                       & CFGTBL_ChangeReq))
1307                                         break;
1308                                 /* delay and try again */
1309                                 udelay(1000);
1310                         }
1311                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1312                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1313                                 return -EAGAIN;
1314                         return 0;
1315                 }
1316         case CCISS_GETNODENAME:
1317                 {
1318                         NodeName_type NodeName;
1319                         int i;
1320
1321                         if (!arg)
1322                                 return -EINVAL;
1323                         for (i = 0; i < 16; i++)
1324                                 NodeName[i] =
1325                                     readb(&host->cfgtable->ServerName[i]);
1326                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1327                                 return -EFAULT;
1328                         return 0;
1329                 }
1330         case CCISS_SETNODENAME:
1331                 {
1332                         NodeName_type NodeName;
1333                         unsigned long flags;
1334                         int i;
1335
1336                         if (!arg)
1337                                 return -EINVAL;
1338                         if (!capable(CAP_SYS_ADMIN))
1339                                 return -EPERM;
1340
1341                         if (copy_from_user
1342                             (NodeName, argp, sizeof(NodeName_type)))
1343                                 return -EFAULT;
1344
1345                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1346
1347                         /* Update the field, and then ring the doorbell */
1348                         for (i = 0; i < 16; i++)
1349                                 writeb(NodeName[i],
1350                                        &host->cfgtable->ServerName[i]);
1351
1352                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1353
1354                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1355                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1356                                       & CFGTBL_ChangeReq))
1357                                         break;
1358                                 /* delay and try again */
1359                                 udelay(1000);
1360                         }
1361                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1362                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1363                                 return -EAGAIN;
1364                         return 0;
1365                 }
1366
1367         case CCISS_GETHEARTBEAT:
1368                 {
1369                         Heartbeat_type heartbeat;
1370
1371                         if (!arg)
1372                                 return -EINVAL;
1373                         heartbeat = readl(&host->cfgtable->HeartBeat);
1374                         if (copy_to_user
1375                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1376                                 return -EFAULT;
1377                         return 0;
1378                 }
1379         case CCISS_GETBUSTYPES:
1380                 {
1381                         BusTypes_type BusTypes;
1382
1383                         if (!arg)
1384                                 return -EINVAL;
1385                         BusTypes = readl(&host->cfgtable->BusTypes);
1386                         if (copy_to_user
1387                             (argp, &BusTypes, sizeof(BusTypes_type)))
1388                                 return -EFAULT;
1389                         return 0;
1390                 }
1391         case CCISS_GETFIRMVER:
1392                 {
1393                         FirmwareVer_type firmware;
1394
1395                         if (!arg)
1396                                 return -EINVAL;
1397                         memcpy(firmware, host->firm_ver, 4);
1398
1399                         if (copy_to_user
1400                             (argp, firmware, sizeof(FirmwareVer_type)))
1401                                 return -EFAULT;
1402                         return 0;
1403                 }
1404         case CCISS_GETDRIVVER:
1405                 {
1406                         DriverVer_type DriverVer = DRIVER_VERSION;
1407
1408                         if (!arg)
1409                                 return -EINVAL;
1410
1411                         if (copy_to_user
1412                             (argp, &DriverVer, sizeof(DriverVer_type)))
1413                                 return -EFAULT;
1414                         return 0;
1415                 }
1416
1417         case CCISS_DEREGDISK:
1418         case CCISS_REGNEWD:
1419         case CCISS_REVALIDVOLS:
1420                 return rebuild_lun_table(host, 0, 1);
1421
1422         case CCISS_GETLUNINFO:{
1423                         LogvolInfo_struct luninfo;
1424
1425                         memcpy(&luninfo.LunID, drv->LunID,
1426                                 sizeof(luninfo.LunID));
1427                         luninfo.num_opens = drv->usage_count;
1428                         luninfo.num_parts = 0;
1429                         if (copy_to_user(argp, &luninfo,
1430                                          sizeof(LogvolInfo_struct)))
1431                                 return -EFAULT;
1432                         return 0;
1433                 }
1434         case CCISS_PASSTHRU:
1435                 {
1436                         IOCTL_Command_struct iocommand;
1437                         CommandList_struct *c;
1438                         char *buff = NULL;
1439                         u64bit temp64;
1440                         DECLARE_COMPLETION_ONSTACK(wait);
1441
1442                         if (!arg)
1443                                 return -EINVAL;
1444
1445                         if (!capable(CAP_SYS_RAWIO))
1446                                 return -EPERM;
1447
1448                         if (copy_from_user
1449                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1450                                 return -EFAULT;
1451                         if ((iocommand.buf_size < 1) &&
1452                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1453                                 return -EINVAL;
1454                         }
1455 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1456                         /* Check kmalloc limits */
1457                         if (iocommand.buf_size > 128000)
1458                                 return -EINVAL;
1459 #endif
1460                         if (iocommand.buf_size > 0) {
1461                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1462                                 if (buff == NULL)
1463                                         return -EFAULT;
1464                         }
1465                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1466                                 /* Copy the data into the buffer we created */
1467                                 if (copy_from_user
1468                                     (buff, iocommand.buf, iocommand.buf_size)) {
1469                                         kfree(buff);
1470                                         return -EFAULT;
1471                                 }
1472                         } else {
1473                                 memset(buff, 0, iocommand.buf_size);
1474                         }
1475                         if ((c = cmd_alloc(host, 0)) == NULL) {
1476                                 kfree(buff);
1477                                 return -ENOMEM;
1478                         }
1479                         /* Fill in the command type */
1480                         c->cmd_type = CMD_IOCTL_PEND;
1481                         /* Fill in Command Header */
1482                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1483                         if (iocommand.buf_size > 0) /* buffer to fill */
1484                         {
1485                                 c->Header.SGList = 1;
1486                                 c->Header.SGTotal = 1;
1487                         } else /* no buffers to fill */
1488                         {
1489                                 c->Header.SGList = 0;
1490                                 c->Header.SGTotal = 0;
1491                         }
1492                         c->Header.LUN = iocommand.LUN_info;
1493                         /* use the kernel address the cmd block for tag */
1494                         c->Header.Tag.lower = c->busaddr;
1495
1496                         /* Fill in Request block */
1497                         c->Request = iocommand.Request;
1498
1499                         /* Fill in the scatter gather information */
1500                         if (iocommand.buf_size > 0) {
1501                                 temp64.val = pci_map_single(host->pdev, buff,
1502                                         iocommand.buf_size,
1503                                         PCI_DMA_BIDIRECTIONAL);
1504                                 c->SG[0].Addr.lower = temp64.val32.lower;
1505                                 c->SG[0].Addr.upper = temp64.val32.upper;
1506                                 c->SG[0].Len = iocommand.buf_size;
1507                                 c->SG[0].Ext = 0;  /* we are not chaining */
1508                         }
1509                         c->waiting = &wait;
1510
1511                         enqueue_cmd_and_start_io(host, c);
1512                         wait_for_completion(&wait);
1513
1514                         /* unlock the buffers from DMA */
1515                         temp64.val32.lower = c->SG[0].Addr.lower;
1516                         temp64.val32.upper = c->SG[0].Addr.upper;
1517                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1518                                          iocommand.buf_size,
1519                                          PCI_DMA_BIDIRECTIONAL);
1520
1521                         check_ioctl_unit_attention(host, c);
1522
1523                         /* Copy the error information out */
1524                         iocommand.error_info = *(c->err_info);
1525                         if (copy_to_user
1526                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1527                                 kfree(buff);
1528                                 cmd_free(host, c, 0);
1529                                 return -EFAULT;
1530                         }
1531
1532                         if (iocommand.Request.Type.Direction == XFER_READ) {
1533                                 /* Copy the data out of the buffer we created */
1534                                 if (copy_to_user
1535                                     (iocommand.buf, buff, iocommand.buf_size)) {
1536                                         kfree(buff);
1537                                         cmd_free(host, c, 0);
1538                                         return -EFAULT;
1539                                 }
1540                         }
1541                         kfree(buff);
1542                         cmd_free(host, c, 0);
1543                         return 0;
1544                 }
1545         case CCISS_BIG_PASSTHRU:{
1546                         BIG_IOCTL_Command_struct *ioc;
1547                         CommandList_struct *c;
1548                         unsigned char **buff = NULL;
1549                         int *buff_size = NULL;
1550                         u64bit temp64;
1551                         BYTE sg_used = 0;
1552                         int status = 0;
1553                         int i;
1554                         DECLARE_COMPLETION_ONSTACK(wait);
1555                         __u32 left;
1556                         __u32 sz;
1557                         BYTE __user *data_ptr;
1558
1559                         if (!arg)
1560                                 return -EINVAL;
1561                         if (!capable(CAP_SYS_RAWIO))
1562                                 return -EPERM;
1563                         ioc = (BIG_IOCTL_Command_struct *)
1564                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1565                         if (!ioc) {
1566                                 status = -ENOMEM;
1567                                 goto cleanup1;
1568                         }
1569                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1570                                 status = -EFAULT;
1571                                 goto cleanup1;
1572                         }
1573                         if ((ioc->buf_size < 1) &&
1574                             (ioc->Request.Type.Direction != XFER_NONE)) {
1575                                 status = -EINVAL;
1576                                 goto cleanup1;
1577                         }
1578                         /* Check kmalloc limits  using all SGs */
1579                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1580                                 status = -EINVAL;
1581                                 goto cleanup1;
1582                         }
1583                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1584                                 status = -EINVAL;
1585                                 goto cleanup1;
1586                         }
1587                         buff =
1588                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1589                         if (!buff) {
1590                                 status = -ENOMEM;
1591                                 goto cleanup1;
1592                         }
1593                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1594                                                    GFP_KERNEL);
1595                         if (!buff_size) {
1596                                 status = -ENOMEM;
1597                                 goto cleanup1;
1598                         }
1599                         left = ioc->buf_size;
1600                         data_ptr = ioc->buf;
1601                         while (left) {
1602                                 sz = (left >
1603                                       ioc->malloc_size) ? ioc->
1604                                     malloc_size : left;
1605                                 buff_size[sg_used] = sz;
1606                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1607                                 if (buff[sg_used] == NULL) {
1608                                         status = -ENOMEM;
1609                                         goto cleanup1;
1610                                 }
1611                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1612                                         if (copy_from_user
1613                                             (buff[sg_used], data_ptr, sz)) {
1614                                                 status = -EFAULT;
1615                                                 goto cleanup1;
1616                                         }
1617                                 } else {
1618                                         memset(buff[sg_used], 0, sz);
1619                                 }
1620                                 left -= sz;
1621                                 data_ptr += sz;
1622                                 sg_used++;
1623                         }
1624                         if ((c = cmd_alloc(host, 0)) == NULL) {
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(host->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(host, 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(host->pdev,
1664                                         (dma_addr_t) temp64.val, buff_size[i],
1665                                         PCI_DMA_BIDIRECTIONAL);
1666                         }
1667                         check_ioctl_unit_attention(host, 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(host, 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(host, c, 0);
1682                                                 status = -EFAULT;
1683                                                 goto cleanup1;
1684                                         }
1685                                         ptr += buff_size[i];
1686                                 }
1687                         }
1688                         cmd_free(host, 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 *cmd = rq->completion_data;
1777         ctlr_info_t *h = hba[cmd->ctlr];
1778         SGDescriptor_struct *curr_sg = cmd->SG;
1779         u64bit temp64;
1780         unsigned long flags;
1781         int i, ddir;
1782         int sg_index = 0;
1783
1784         if (cmd->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 < cmd->Header.SGList; i++) {
1792                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1793                         cciss_unmap_sg_chain_block(h, cmd);
1794                         /* Point to the next block */
1795                         curr_sg = h->cmd_sg_list[cmd->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 = cmd->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, cmd, 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(int ctlr, 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(hba[ctlr], scsi3addr, logvol);
1848         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, 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(int ctlr, 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(hba[ctlr], scsi3addr, logvol);
1884         rc = sendcmd_withirq(CISS_INQUIRY, ctlr, 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(int ctlr, int drv_index, int first_time,
1951         int via_ioctl)
1952 {
1953         ctlr_info_t *h = hba[ctlr];
1954         struct gendisk *disk;
1955         InquiryData_struct *inq_buff = NULL;
1956         unsigned int block_size;
1957         sector_t total_size;
1958         unsigned long flags = 0;
1959         int ret = 0;
1960         drive_info_struct *drvinfo;
1961
1962         /* Get information about the disk and modify the driver structure */
1963         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1964         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1965         if (inq_buff == NULL || drvinfo == NULL)
1966                 goto mem_msg;
1967
1968         /* testing to see if 16-byte CDBs are already being used */
1969         if (h->cciss_read == CCISS_READ_16) {
1970                 cciss_read_capacity_16(h->ctlr, drv_index,
1971                         &total_size, &block_size);
1972
1973         } else {
1974                 cciss_read_capacity(ctlr, drv_index, &total_size, &block_size);
1975                 /* if read_capacity returns all F's this volume is >2TB */
1976                 /* in size so we switch to 16-byte CDB's for all */
1977                 /* read/write ops */
1978                 if (total_size == 0xFFFFFFFFULL) {
1979                         cciss_read_capacity_16(ctlr, drv_index,
1980                         &total_size, &block_size);
1981                         h->cciss_read = CCISS_READ_16;
1982                         h->cciss_write = CCISS_WRITE_16;
1983                 } else {
1984                         h->cciss_read = CCISS_READ_10;
1985                         h->cciss_write = CCISS_WRITE_10;
1986                 }
1987         }
1988
1989         cciss_geometry_inquiry(ctlr, drv_index, total_size, block_size,
1990                                inq_buff, drvinfo);
1991         drvinfo->block_size = block_size;
1992         drvinfo->nr_blocks = total_size + 1;
1993
1994         cciss_get_device_descr(ctlr, drv_index, drvinfo->vendor,
1995                                 drvinfo->model, drvinfo->rev);
1996         cciss_get_serial_no(ctlr, drv_index, drvinfo->serial_no,
1997                         sizeof(drvinfo->serial_no));
1998         /* Save the lunid in case we deregister the disk, below. */
1999         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
2000                 sizeof(drvinfo->LunID));
2001
2002         /* Is it the same disk we already know, and nothing's changed? */
2003         if (h->drv[drv_index]->raid_level != -1 &&
2004                 ((memcmp(drvinfo->serial_no,
2005                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2006                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2007                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2008                 drvinfo->heads == h->drv[drv_index]->heads &&
2009                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2010                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2011                         /* The disk is unchanged, nothing to update */
2012                         goto freeret;
2013
2014         /* If we get here it's not the same disk, or something's changed,
2015          * so we need to * deregister it, and re-register it, if it's not
2016          * in use.
2017          * If the disk already exists then deregister it before proceeding
2018          * (unless it's the first disk (for the controller node).
2019          */
2020         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2021                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
2022                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2023                 h->drv[drv_index]->busy_configuring = 1;
2024                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2025
2026                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2027                  * which keeps the interrupt handler from starting
2028                  * the queue.
2029                  */
2030                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2031         }
2032
2033         /* If the disk is in use return */
2034         if (ret)
2035                 goto freeret;
2036
2037         /* Save the new information from cciss_geometry_inquiry
2038          * and serial number inquiry.  If the disk was deregistered
2039          * above, then h->drv[drv_index] will be NULL.
2040          */
2041         if (h->drv[drv_index] == NULL) {
2042                 drvinfo->device_initialized = 0;
2043                 h->drv[drv_index] = drvinfo;
2044                 drvinfo = NULL; /* so it won't be freed below. */
2045         } else {
2046                 /* special case for cxd0 */
2047                 h->drv[drv_index]->block_size = drvinfo->block_size;
2048                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2049                 h->drv[drv_index]->heads = drvinfo->heads;
2050                 h->drv[drv_index]->sectors = drvinfo->sectors;
2051                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2052                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2053                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2054                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2055                         VENDOR_LEN + 1);
2056                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2057                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2058         }
2059
2060         ++h->num_luns;
2061         disk = h->gendisk[drv_index];
2062         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2063
2064         /* If it's not disk 0 (drv_index != 0)
2065          * or if it was disk 0, but there was previously
2066          * no actual corresponding configured logical drive
2067          * (raid_leve == -1) then we want to update the
2068          * logical drive's information.
2069          */
2070         if (drv_index || first_time) {
2071                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2072                         cciss_free_gendisk(h, drv_index);
2073                         cciss_free_drive_info(h, drv_index);
2074                         printk(KERN_WARNING "cciss:%d could not update "
2075                                 "disk %d\n", h->ctlr, drv_index);
2076                         --h->num_luns;
2077                 }
2078         }
2079
2080 freeret:
2081         kfree(inq_buff);
2082         kfree(drvinfo);
2083         return;
2084 mem_msg:
2085         printk(KERN_ERR "cciss: out of memory\n");
2086         goto freeret;
2087 }
2088
2089 /* This function will find the first index of the controllers drive array
2090  * that has a null drv pointer and allocate the drive info struct and
2091  * will return that index   This is where new drives will be added.
2092  * If the index to be returned is greater than the highest_lun index for
2093  * the controller then highest_lun is set * to this new index.
2094  * If there are no available indexes or if tha allocation fails, then -1
2095  * is returned.  * "controller_node" is used to know if this is a real
2096  * logical drive, or just the controller node, which determines if this
2097  * counts towards highest_lun.
2098  */
2099 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2100 {
2101         int i;
2102         drive_info_struct *drv;
2103
2104         /* Search for an empty slot for our drive info */
2105         for (i = 0; i < CISS_MAX_LUN; i++) {
2106
2107                 /* if not cxd0 case, and it's occupied, skip it. */
2108                 if (h->drv[i] && i != 0)
2109                         continue;
2110                 /*
2111                  * If it's cxd0 case, and drv is alloc'ed already, and a
2112                  * disk is configured there, skip it.
2113                  */
2114                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2115                         continue;
2116
2117                 /*
2118                  * We've found an empty slot.  Update highest_lun
2119                  * provided this isn't just the fake cxd0 controller node.
2120                  */
2121                 if (i > h->highest_lun && !controller_node)
2122                         h->highest_lun = i;
2123
2124                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2125                 if (i == 0 && h->drv[i] != NULL)
2126                         return i;
2127
2128                 /*
2129                  * Found an empty slot, not already alloc'ed.  Allocate it.
2130                  * Mark it with raid_level == -1, so we know it's new later on.
2131                  */
2132                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2133                 if (!drv)
2134                         return -1;
2135                 drv->raid_level = -1; /* so we know it's new */
2136                 h->drv[i] = drv;
2137                 return i;
2138         }
2139         return -1;
2140 }
2141
2142 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2143 {
2144         kfree(h->drv[drv_index]);
2145         h->drv[drv_index] = NULL;
2146 }
2147
2148 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2149 {
2150         put_disk(h->gendisk[drv_index]);
2151         h->gendisk[drv_index] = NULL;
2152 }
2153
2154 /* cciss_add_gendisk finds a free hba[]->drv structure
2155  * and allocates a gendisk if needed, and sets the lunid
2156  * in the drvinfo structure.   It returns the index into
2157  * the ->drv[] array, or -1 if none are free.
2158  * is_controller_node indicates whether highest_lun should
2159  * count this disk, or if it's only being added to provide
2160  * a means to talk to the controller in case no logical
2161  * drives have yet been configured.
2162  */
2163 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2164         int controller_node)
2165 {
2166         int drv_index;
2167
2168         drv_index = cciss_alloc_drive_info(h, controller_node);
2169         if (drv_index == -1)
2170                 return -1;
2171
2172         /*Check if the gendisk needs to be allocated */
2173         if (!h->gendisk[drv_index]) {
2174                 h->gendisk[drv_index] =
2175                         alloc_disk(1 << NWD_SHIFT);
2176                 if (!h->gendisk[drv_index]) {
2177                         printk(KERN_ERR "cciss%d: could not "
2178                                 "allocate a new disk %d\n",
2179                                 h->ctlr, drv_index);
2180                         goto err_free_drive_info;
2181                 }
2182         }
2183         memcpy(h->drv[drv_index]->LunID, lunid,
2184                 sizeof(h->drv[drv_index]->LunID));
2185         if (cciss_create_ld_sysfs_entry(h, drv_index))
2186                 goto err_free_disk;
2187         /* Don't need to mark this busy because nobody */
2188         /* else knows about this disk yet to contend */
2189         /* for access to it. */
2190         h->drv[drv_index]->busy_configuring = 0;
2191         wmb();
2192         return drv_index;
2193
2194 err_free_disk:
2195         cciss_free_gendisk(h, drv_index);
2196 err_free_drive_info:
2197         cciss_free_drive_info(h, drv_index);
2198         return -1;
2199 }
2200
2201 /* This is for the special case of a controller which
2202  * has no logical drives.  In this case, we still need
2203  * to register a disk so the controller can be accessed
2204  * by the Array Config Utility.
2205  */
2206 static void cciss_add_controller_node(ctlr_info_t *h)
2207 {
2208         struct gendisk *disk;
2209         int drv_index;
2210
2211         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2212                 return;
2213
2214         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2215         if (drv_index == -1)
2216                 goto error;
2217         h->drv[drv_index]->block_size = 512;
2218         h->drv[drv_index]->nr_blocks = 0;
2219         h->drv[drv_index]->heads = 0;
2220         h->drv[drv_index]->sectors = 0;
2221         h->drv[drv_index]->cylinders = 0;
2222         h->drv[drv_index]->raid_level = -1;
2223         memset(h->drv[drv_index]->serial_no, 0, 16);
2224         disk = h->gendisk[drv_index];
2225         if (cciss_add_disk(h, disk, drv_index) == 0)
2226                 return;
2227         cciss_free_gendisk(h, drv_index);
2228         cciss_free_drive_info(h, drv_index);
2229 error:
2230         printk(KERN_WARNING "cciss%d: could not "
2231                 "add disk 0.\n", h->ctlr);
2232         return;
2233 }
2234
2235 /* This function will add and remove logical drives from the Logical
2236  * drive array of the controller and maintain persistency of ordering
2237  * so that mount points are preserved until the next reboot.  This allows
2238  * for the removal of logical drives in the middle of the drive array
2239  * without a re-ordering of those drives.
2240  * INPUT
2241  * h            = The controller to perform the operations on
2242  */
2243 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2244         int via_ioctl)
2245 {
2246         int ctlr = h->ctlr;
2247         int num_luns;
2248         ReportLunData_struct *ld_buff = NULL;
2249         int return_code;
2250         int listlength = 0;
2251         int i;
2252         int drv_found;
2253         int drv_index = 0;
2254         unsigned char lunid[8] = CTLR_LUNID;
2255         unsigned long flags;
2256
2257         if (!capable(CAP_SYS_RAWIO))
2258                 return -EPERM;
2259
2260         /* Set busy_configuring flag for this operation */
2261         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2262         if (h->busy_configuring) {
2263                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2264                 return -EBUSY;
2265         }
2266         h->busy_configuring = 1;
2267         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2268
2269         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2270         if (ld_buff == NULL)
2271                 goto mem_msg;
2272
2273         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2274                                       sizeof(ReportLunData_struct),
2275                                       0, CTLR_LUNID, TYPE_CMD);
2276
2277         if (return_code == IO_OK)
2278                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2279         else {  /* reading number of logical volumes failed */
2280                 printk(KERN_WARNING "cciss: report logical volume"
2281                        " command failed\n");
2282                 listlength = 0;
2283                 goto freeret;
2284         }
2285
2286         num_luns = listlength / 8;      /* 8 bytes per entry */
2287         if (num_luns > CISS_MAX_LUN) {
2288                 num_luns = CISS_MAX_LUN;
2289                 printk(KERN_WARNING "cciss: more luns configured"
2290                        " on controller than can be handled by"
2291                        " this driver.\n");
2292         }
2293
2294         if (num_luns == 0)
2295                 cciss_add_controller_node(h);
2296
2297         /* Compare controller drive array to driver's drive array
2298          * to see if any drives are missing on the controller due
2299          * to action of Array Config Utility (user deletes drive)
2300          * and deregister logical drives which have disappeared.
2301          */
2302         for (i = 0; i <= h->highest_lun; i++) {
2303                 int j;
2304                 drv_found = 0;
2305
2306                 /* skip holes in the array from already deleted drives */
2307                 if (h->drv[i] == NULL)
2308                         continue;
2309
2310                 for (j = 0; j < num_luns; j++) {
2311                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2312                         if (memcmp(h->drv[i]->LunID, lunid,
2313                                 sizeof(lunid)) == 0) {
2314                                 drv_found = 1;
2315                                 break;
2316                         }
2317                 }
2318                 if (!drv_found) {
2319                         /* Deregister it from the OS, it's gone. */
2320                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2321                         h->drv[i]->busy_configuring = 1;
2322                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2323                         return_code = deregister_disk(h, i, 1, via_ioctl);
2324                         if (h->drv[i] != NULL)
2325                                 h->drv[i]->busy_configuring = 0;
2326                 }
2327         }
2328
2329         /* Compare controller drive array to driver's drive array.
2330          * Check for updates in the drive information and any new drives
2331          * on the controller due to ACU adding logical drives, or changing
2332          * a logical drive's size, etc.  Reregister any new/changed drives
2333          */
2334         for (i = 0; i < num_luns; i++) {
2335                 int j;
2336
2337                 drv_found = 0;
2338
2339                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2340                 /* Find if the LUN is already in the drive array
2341                  * of the driver.  If so then update its info
2342                  * if not in use.  If it does not exist then find
2343                  * the first free index and add it.
2344                  */
2345                 for (j = 0; j <= h->highest_lun; j++) {
2346                         if (h->drv[j] != NULL &&
2347                                 memcmp(h->drv[j]->LunID, lunid,
2348                                         sizeof(h->drv[j]->LunID)) == 0) {
2349                                 drv_index = j;
2350                                 drv_found = 1;
2351                                 break;
2352                         }
2353                 }
2354
2355                 /* check if the drive was found already in the array */
2356                 if (!drv_found) {
2357                         drv_index = cciss_add_gendisk(h, lunid, 0);
2358                         if (drv_index == -1)
2359                                 goto freeret;
2360                 }
2361                 cciss_update_drive_info(ctlr, drv_index, first_time,
2362                         via_ioctl);
2363         }               /* end for */
2364
2365 freeret:
2366         kfree(ld_buff);
2367         h->busy_configuring = 0;
2368         /* We return -1 here to tell the ACU that we have registered/updated
2369          * all of the drives that we can and to keep it from calling us
2370          * additional times.
2371          */
2372         return -1;
2373 mem_msg:
2374         printk(KERN_ERR "cciss: out of memory\n");
2375         h->busy_configuring = 0;
2376         goto freeret;
2377 }
2378
2379 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2380 {
2381         /* zero out the disk size info */
2382         drive_info->nr_blocks = 0;
2383         drive_info->block_size = 0;
2384         drive_info->heads = 0;
2385         drive_info->sectors = 0;
2386         drive_info->cylinders = 0;
2387         drive_info->raid_level = -1;
2388         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2389         memset(drive_info->model, 0, sizeof(drive_info->model));
2390         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2391         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2392         /*
2393          * don't clear the LUNID though, we need to remember which
2394          * one this one is.
2395          */
2396 }
2397
2398 /* This function will deregister the disk and it's queue from the
2399  * kernel.  It must be called with the controller lock held and the
2400  * drv structures busy_configuring flag set.  It's parameters are:
2401  *
2402  * disk = This is the disk to be deregistered
2403  * drv  = This is the drive_info_struct associated with the disk to be
2404  *        deregistered.  It contains information about the disk used
2405  *        by the driver.
2406  * clear_all = This flag determines whether or not the disk information
2407  *             is going to be completely cleared out and the highest_lun
2408  *             reset.  Sometimes we want to clear out information about
2409  *             the disk in preparation for re-adding it.  In this case
2410  *             the highest_lun should be left unchanged and the LunID
2411  *             should not be cleared.
2412  * via_ioctl
2413  *    This indicates whether we've reached this path via ioctl.
2414  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2415  *    If this path is reached via ioctl(), then the max_usage_count will
2416  *    be 1, as the process calling ioctl() has got to have the device open.
2417  *    If we get here via sysfs, then the max usage count will be zero.
2418 */
2419 static int deregister_disk(ctlr_info_t *h, int drv_index,
2420                            int clear_all, int via_ioctl)
2421 {
2422         int i;
2423         struct gendisk *disk;
2424         drive_info_struct *drv;
2425         int recalculate_highest_lun;
2426
2427         if (!capable(CAP_SYS_RAWIO))
2428                 return -EPERM;
2429
2430         drv = h->drv[drv_index];
2431         disk = h->gendisk[drv_index];
2432
2433         /* make sure logical volume is NOT is use */
2434         if (clear_all || (h->gendisk[0] == disk)) {
2435                 if (drv->usage_count > via_ioctl)
2436                         return -EBUSY;
2437         } else if (drv->usage_count > 0)
2438                 return -EBUSY;
2439
2440         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2441
2442         /* invalidate the devices and deregister the disk.  If it is disk
2443          * zero do not deregister it but just zero out it's values.  This
2444          * allows us to delete disk zero but keep the controller registered.
2445          */
2446         if (h->gendisk[0] != disk) {
2447                 struct request_queue *q = disk->queue;
2448                 if (disk->flags & GENHD_FL_UP) {
2449                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2450                         del_gendisk(disk);
2451                 }
2452                 if (q)
2453                         blk_cleanup_queue(q);
2454                 /* If clear_all is set then we are deleting the logical
2455                  * drive, not just refreshing its info.  For drives
2456                  * other than disk 0 we will call put_disk.  We do not
2457                  * do this for disk 0 as we need it to be able to
2458                  * configure the controller.
2459                  */
2460                 if (clear_all){
2461                         /* This isn't pretty, but we need to find the
2462                          * disk in our array and NULL our the pointer.
2463                          * This is so that we will call alloc_disk if
2464                          * this index is used again later.
2465                          */
2466                         for (i=0; i < CISS_MAX_LUN; i++){
2467                                 if (h->gendisk[i] == disk) {
2468                                         h->gendisk[i] = NULL;
2469                                         break;
2470                                 }
2471                         }
2472                         put_disk(disk);
2473                 }
2474         } else {
2475                 set_capacity(disk, 0);
2476                 cciss_clear_drive_info(drv);
2477         }
2478
2479         --h->num_luns;
2480
2481         /* if it was the last disk, find the new hightest lun */
2482         if (clear_all && recalculate_highest_lun) {
2483                 int newhighest = -1;
2484                 for (i = 0; i <= h->highest_lun; i++) {
2485                         /* if the disk has size > 0, it is available */
2486                         if (h->drv[i] && h->drv[i]->heads)
2487                                 newhighest = i;
2488                 }
2489                 h->highest_lun = newhighest;
2490         }
2491         return 0;
2492 }
2493
2494 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2495                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2496                 int cmd_type)
2497 {
2498         ctlr_info_t *h = hba[ctlr];
2499         u64bit buff_dma_handle;
2500         int status = IO_OK;
2501
2502         c->cmd_type = CMD_IOCTL_PEND;
2503         c->Header.ReplyQueue = 0;
2504         if (buff != NULL) {
2505                 c->Header.SGList = 1;
2506                 c->Header.SGTotal = 1;
2507         } else {
2508                 c->Header.SGList = 0;
2509                 c->Header.SGTotal = 0;
2510         }
2511         c->Header.Tag.lower = c->busaddr;
2512         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2513
2514         c->Request.Type.Type = cmd_type;
2515         if (cmd_type == TYPE_CMD) {
2516                 switch (cmd) {
2517                 case CISS_INQUIRY:
2518                         /* are we trying to read a vital product page */
2519                         if (page_code != 0) {
2520                                 c->Request.CDB[1] = 0x01;
2521                                 c->Request.CDB[2] = page_code;
2522                         }
2523                         c->Request.CDBLen = 6;
2524                         c->Request.Type.Attribute = ATTR_SIMPLE;
2525                         c->Request.Type.Direction = XFER_READ;
2526                         c->Request.Timeout = 0;
2527                         c->Request.CDB[0] = CISS_INQUIRY;
2528                         c->Request.CDB[4] = size & 0xFF;
2529                         break;
2530                 case CISS_REPORT_LOG:
2531                 case CISS_REPORT_PHYS:
2532                         /* Talking to controller so It's a physical command
2533                            mode = 00 target = 0.  Nothing to write.
2534                          */
2535                         c->Request.CDBLen = 12;
2536                         c->Request.Type.Attribute = ATTR_SIMPLE;
2537                         c->Request.Type.Direction = XFER_READ;
2538                         c->Request.Timeout = 0;
2539                         c->Request.CDB[0] = cmd;
2540                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2541                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2542                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2543                         c->Request.CDB[9] = size & 0xFF;
2544                         break;
2545
2546                 case CCISS_READ_CAPACITY:
2547                         c->Request.CDBLen = 10;
2548                         c->Request.Type.Attribute = ATTR_SIMPLE;
2549                         c->Request.Type.Direction = XFER_READ;
2550                         c->Request.Timeout = 0;
2551                         c->Request.CDB[0] = cmd;
2552                         break;
2553                 case CCISS_READ_CAPACITY_16:
2554                         c->Request.CDBLen = 16;
2555                         c->Request.Type.Attribute = ATTR_SIMPLE;
2556                         c->Request.Type.Direction = XFER_READ;
2557                         c->Request.Timeout = 0;
2558                         c->Request.CDB[0] = cmd;
2559                         c->Request.CDB[1] = 0x10;
2560                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2561                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2562                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2563                         c->Request.CDB[13] = size & 0xFF;
2564                         c->Request.Timeout = 0;
2565                         c->Request.CDB[0] = cmd;
2566                         break;
2567                 case CCISS_CACHE_FLUSH:
2568                         c->Request.CDBLen = 12;
2569                         c->Request.Type.Attribute = ATTR_SIMPLE;
2570                         c->Request.Type.Direction = XFER_WRITE;
2571                         c->Request.Timeout = 0;
2572                         c->Request.CDB[0] = BMIC_WRITE;
2573                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2574                         break;
2575                 case TEST_UNIT_READY:
2576                         c->Request.CDBLen = 6;
2577                         c->Request.Type.Attribute = ATTR_SIMPLE;
2578                         c->Request.Type.Direction = XFER_NONE;
2579                         c->Request.Timeout = 0;
2580                         break;
2581                 default:
2582                         printk(KERN_WARNING
2583                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2584                         return IO_ERROR;
2585                 }
2586         } else if (cmd_type == TYPE_MSG) {
2587                 switch (cmd) {
2588                 case 0: /* ABORT message */
2589                         c->Request.CDBLen = 12;
2590                         c->Request.Type.Attribute = ATTR_SIMPLE;
2591                         c->Request.Type.Direction = XFER_WRITE;
2592                         c->Request.Timeout = 0;
2593                         c->Request.CDB[0] = cmd;        /* abort */
2594                         c->Request.CDB[1] = 0;  /* abort a command */
2595                         /* buff contains the tag of the command to abort */
2596                         memcpy(&c->Request.CDB[4], buff, 8);
2597                         break;
2598                 case 1: /* RESET message */
2599                         c->Request.CDBLen = 16;
2600                         c->Request.Type.Attribute = ATTR_SIMPLE;
2601                         c->Request.Type.Direction = XFER_NONE;
2602                         c->Request.Timeout = 0;
2603                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2604                         c->Request.CDB[0] = cmd;        /* reset */
2605                         c->Request.CDB[1] = 0x03;       /* reset a target */
2606                         break;
2607                 case 3: /* No-Op message */
2608                         c->Request.CDBLen = 1;
2609                         c->Request.Type.Attribute = ATTR_SIMPLE;
2610                         c->Request.Type.Direction = XFER_WRITE;
2611                         c->Request.Timeout = 0;
2612                         c->Request.CDB[0] = cmd;
2613                         break;
2614                 default:
2615                         printk(KERN_WARNING
2616                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2617                         return IO_ERROR;
2618                 }
2619         } else {
2620                 printk(KERN_WARNING
2621                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2622                 return IO_ERROR;
2623         }
2624         /* Fill in the scatter gather information */
2625         if (size > 0) {
2626                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2627                                                              buff, size,
2628                                                              PCI_DMA_BIDIRECTIONAL);
2629                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2630                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2631                 c->SG[0].Len = size;
2632                 c->SG[0].Ext = 0;       /* we are not chaining */
2633         }
2634         return status;
2635 }
2636
2637 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2638 {
2639         switch (c->err_info->ScsiStatus) {
2640         case SAM_STAT_GOOD:
2641                 return IO_OK;
2642         case SAM_STAT_CHECK_CONDITION:
2643                 switch (0xf & c->err_info->SenseInfo[2]) {
2644                 case 0: return IO_OK; /* no sense */
2645                 case 1: return IO_OK; /* recovered error */
2646                 default:
2647                         if (check_for_unit_attention(h, c))
2648                                 return IO_NEEDS_RETRY;
2649                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2650                                 "check condition, sense key = 0x%02x\n",
2651                                 h->ctlr, c->Request.CDB[0],
2652                                 c->err_info->SenseInfo[2]);
2653                 }
2654                 break;
2655         default:
2656                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2657                         "scsi status = 0x%02x\n", h->ctlr,
2658                         c->Request.CDB[0], c->err_info->ScsiStatus);
2659                 break;
2660         }
2661         return IO_ERROR;
2662 }
2663
2664 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2665 {
2666         int return_status = IO_OK;
2667
2668         if (c->err_info->CommandStatus == CMD_SUCCESS)
2669                 return IO_OK;
2670
2671         switch (c->err_info->CommandStatus) {
2672         case CMD_TARGET_STATUS:
2673                 return_status = check_target_status(h, c);
2674                 break;
2675         case CMD_DATA_UNDERRUN:
2676         case CMD_DATA_OVERRUN:
2677                 /* expected for inquiry and report lun commands */
2678                 break;
2679         case CMD_INVALID:
2680                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2681                        "reported invalid\n", c->Request.CDB[0]);
2682                 return_status = IO_ERROR;
2683                 break;
2684         case CMD_PROTOCOL_ERR:
2685                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2686                        "protocol error \n", c->Request.CDB[0]);
2687                 return_status = IO_ERROR;
2688                 break;
2689         case CMD_HARDWARE_ERR:
2690                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2691                        " hardware error\n", c->Request.CDB[0]);
2692                 return_status = IO_ERROR;
2693                 break;
2694         case CMD_CONNECTION_LOST:
2695                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2696                        "connection lost\n", c->Request.CDB[0]);
2697                 return_status = IO_ERROR;
2698                 break;
2699         case CMD_ABORTED:
2700                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2701                        "aborted\n", c->Request.CDB[0]);
2702                 return_status = IO_ERROR;
2703                 break;
2704         case CMD_ABORT_FAILED:
2705                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2706                        "abort failed\n", c->Request.CDB[0]);
2707                 return_status = IO_ERROR;
2708                 break;
2709         case CMD_UNSOLICITED_ABORT:
2710                 printk(KERN_WARNING
2711                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2712                         c->Request.CDB[0]);
2713                 return_status = IO_NEEDS_RETRY;
2714                 break;
2715         default:
2716                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2717                        "unknown status %x\n", c->Request.CDB[0],
2718                        c->err_info->CommandStatus);
2719                 return_status = IO_ERROR;
2720         }
2721         return return_status;
2722 }
2723
2724 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2725         int attempt_retry)
2726 {
2727         DECLARE_COMPLETION_ONSTACK(wait);
2728         u64bit buff_dma_handle;
2729         int return_status = IO_OK;
2730
2731 resend_cmd2:
2732         c->waiting = &wait;
2733         enqueue_cmd_and_start_io(h, c);
2734
2735         wait_for_completion(&wait);
2736
2737         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2738                 goto command_done;
2739
2740         return_status = process_sendcmd_error(h, c);
2741
2742         if (return_status == IO_NEEDS_RETRY &&
2743                 c->retry_count < MAX_CMD_RETRIES) {
2744                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2745                         c->Request.CDB[0]);
2746                 c->retry_count++;
2747                 /* erase the old error information */
2748                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2749                 return_status = IO_OK;
2750                 INIT_COMPLETION(wait);
2751                 goto resend_cmd2;
2752         }
2753
2754 command_done:
2755         /* unlock the buffers from DMA */
2756         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2757         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2758         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2759                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2760         return return_status;
2761 }
2762
2763 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2764                            __u8 page_code, unsigned char scsi3addr[],
2765                         int cmd_type)
2766 {
2767         ctlr_info_t *h = hba[ctlr];
2768         CommandList_struct *c;
2769         int return_status;
2770
2771         c = cmd_alloc(h, 0);
2772         if (!c)
2773                 return -ENOMEM;
2774         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2775                 scsi3addr, cmd_type);
2776         if (return_status == IO_OK)
2777                 return_status = sendcmd_withirq_core(h, c, 1);
2778
2779         cmd_free(h, c, 0);
2780         return return_status;
2781 }
2782
2783 static void cciss_geometry_inquiry(int ctlr, int logvol,
2784                                    sector_t total_size,
2785                                    unsigned int block_size,
2786                                    InquiryData_struct *inq_buff,
2787                                    drive_info_struct *drv)
2788 {
2789         int return_code;
2790         unsigned long t;
2791         unsigned char scsi3addr[8];
2792
2793         memset(inq_buff, 0, sizeof(InquiryData_struct));
2794         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2795         return_code = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buff,
2796                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2797         if (return_code == IO_OK) {
2798                 if (inq_buff->data_byte[8] == 0xFF) {
2799                         printk(KERN_WARNING
2800                                "cciss: reading geometry failed, volume "
2801                                "does not support reading geometry\n");
2802                         drv->heads = 255;
2803                         drv->sectors = 32;      /* Sectors per track */
2804                         drv->cylinders = total_size + 1;
2805                         drv->raid_level = RAID_UNKNOWN;
2806                 } else {
2807                         drv->heads = inq_buff->data_byte[6];
2808                         drv->sectors = inq_buff->data_byte[7];
2809                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2810                         drv->cylinders += inq_buff->data_byte[5];
2811                         drv->raid_level = inq_buff->data_byte[8];
2812                 }
2813                 drv->block_size = block_size;
2814                 drv->nr_blocks = total_size + 1;
2815                 t = drv->heads * drv->sectors;
2816                 if (t > 1) {
2817                         sector_t real_size = total_size + 1;
2818                         unsigned long rem = sector_div(real_size, t);
2819                         if (rem)
2820                                 real_size++;
2821                         drv->cylinders = real_size;
2822                 }
2823         } else {                /* Get geometry failed */
2824                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2825         }
2826 }
2827
2828 static void
2829 cciss_read_capacity(int ctlr, int logvol, sector_t *total_size,
2830                     unsigned int *block_size)
2831 {
2832         ReadCapdata_struct *buf;
2833         int return_code;
2834         unsigned char scsi3addr[8];
2835
2836         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2837         if (!buf) {
2838                 printk(KERN_WARNING "cciss: out of memory\n");
2839                 return;
2840         }
2841
2842         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2843         return_code = sendcmd_withirq(CCISS_READ_CAPACITY, ctlr, buf,
2844                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2845         if (return_code == IO_OK) {
2846                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2847                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2848         } else {                /* read capacity command failed */
2849                 printk(KERN_WARNING "cciss: read capacity failed\n");
2850                 *total_size = 0;
2851                 *block_size = BLOCK_SIZE;
2852         }
2853         kfree(buf);
2854 }
2855
2856 static void cciss_read_capacity_16(int ctlr, int logvol,
2857         sector_t *total_size, unsigned int *block_size)
2858 {
2859         ReadCapdata_struct_16 *buf;
2860         int return_code;
2861         unsigned char scsi3addr[8];
2862
2863         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2864         if (!buf) {
2865                 printk(KERN_WARNING "cciss: out of memory\n");
2866                 return;
2867         }
2868
2869         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2870         return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2871                 ctlr, buf, sizeof(ReadCapdata_struct_16),
2872                         0, scsi3addr, TYPE_CMD);
2873         if (return_code == IO_OK) {
2874                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2875                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2876         } else {                /* read capacity command failed */
2877                 printk(KERN_WARNING "cciss: read capacity failed\n");
2878                 *total_size = 0;
2879                 *block_size = BLOCK_SIZE;
2880         }
2881         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2882                (unsigned long long)*total_size+1, *block_size);
2883         kfree(buf);
2884 }
2885
2886 static int cciss_revalidate(struct gendisk *disk)
2887 {
2888         ctlr_info_t *h = get_host(disk);
2889         drive_info_struct *drv = get_drv(disk);
2890         int logvol;
2891         int FOUND = 0;
2892         unsigned int block_size;
2893         sector_t total_size;
2894         InquiryData_struct *inq_buff = NULL;
2895
2896         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2897                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2898                         sizeof(drv->LunID)) == 0) {
2899                         FOUND = 1;
2900                         break;
2901                 }
2902         }
2903
2904         if (!FOUND)
2905                 return 1;
2906
2907         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2908         if (inq_buff == NULL) {
2909                 printk(KERN_WARNING "cciss: out of memory\n");
2910                 return 1;
2911         }
2912         if (h->cciss_read == CCISS_READ_10) {
2913                 cciss_read_capacity(h->ctlr, logvol,
2914                                         &total_size, &block_size);
2915         } else {
2916                 cciss_read_capacity_16(h->ctlr, logvol,
2917                                         &total_size, &block_size);
2918         }
2919         cciss_geometry_inquiry(h->ctlr, logvol, total_size, block_size,
2920                                inq_buff, drv);
2921
2922         blk_queue_logical_block_size(drv->queue, drv->block_size);
2923         set_capacity(disk, drv->nr_blocks);
2924
2925         kfree(inq_buff);
2926         return 0;
2927 }
2928
2929 /*
2930  * Map (physical) PCI mem into (virtual) kernel space
2931  */
2932 static void __iomem *remap_pci_mem(ulong base, ulong size)
2933 {
2934         ulong page_base = ((ulong) base) & PAGE_MASK;
2935         ulong page_offs = ((ulong) base) - page_base;
2936         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2937
2938         return page_remapped ? (page_remapped + page_offs) : NULL;
2939 }
2940
2941 /*
2942  * Takes jobs of the Q and sends them to the hardware, then puts it on
2943  * the Q to wait for completion.
2944  */
2945 static void start_io(ctlr_info_t *h)
2946 {
2947         CommandList_struct *c;
2948
2949         while (!hlist_empty(&h->reqQ)) {
2950                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2951                 /* can't do anything if fifo is full */
2952                 if ((h->access.fifo_full(h))) {
2953                         printk(KERN_WARNING "cciss: fifo full\n");
2954                         break;
2955                 }
2956
2957                 /* Get the first entry from the Request Q */
2958                 removeQ(c);
2959                 h->Qdepth--;
2960
2961                 /* Tell the controller execute command */
2962                 h->access.submit_command(h, c);
2963
2964                 /* Put job onto the completed Q */
2965                 addQ(&h->cmpQ, c);
2966         }
2967 }
2968
2969 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2970 /* Zeros out the error record and then resends the command back */
2971 /* to the controller */
2972 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2973 {
2974         /* erase the old error information */
2975         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2976
2977         /* add it to software queue and then send it to the controller */
2978         addQ(&h->reqQ, c);
2979         h->Qdepth++;
2980         if (h->Qdepth > h->maxQsinceinit)
2981                 h->maxQsinceinit = h->Qdepth;
2982
2983         start_io(h);
2984 }
2985
2986 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2987         unsigned int msg_byte, unsigned int host_byte,
2988         unsigned int driver_byte)
2989 {
2990         /* inverse of macros in scsi.h */
2991         return (scsi_status_byte & 0xff) |
2992                 ((msg_byte & 0xff) << 8) |
2993                 ((host_byte & 0xff) << 16) |
2994                 ((driver_byte & 0xff) << 24);
2995 }
2996
2997 static inline int evaluate_target_status(ctlr_info_t *h,
2998                         CommandList_struct *cmd, int *retry_cmd)
2999 {
3000         unsigned char sense_key;
3001         unsigned char status_byte, msg_byte, host_byte, driver_byte;
3002         int error_value;
3003
3004         *retry_cmd = 0;
3005         /* If we get in here, it means we got "target status", that is, scsi status */
3006         status_byte = cmd->err_info->ScsiStatus;
3007         driver_byte = DRIVER_OK;
3008         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
3009
3010         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
3011                 host_byte = DID_PASSTHROUGH;
3012         else
3013                 host_byte = DID_OK;
3014
3015         error_value = make_status_bytes(status_byte, msg_byte,
3016                 host_byte, driver_byte);
3017
3018         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3019                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3020                         printk(KERN_WARNING "cciss: cmd %p "
3021                                "has SCSI Status 0x%x\n",
3022                                cmd, cmd->err_info->ScsiStatus);
3023                 return error_value;
3024         }
3025
3026         /* check the sense key */
3027         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3028         /* no status or recovered error */
3029         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3030             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3031                 error_value = 0;
3032
3033         if (check_for_unit_attention(h, cmd)) {
3034                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3035                 return 0;
3036         }
3037
3038         /* Not SG_IO or similar? */
3039         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3040                 if (error_value != 0)
3041                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3042                                " sense key = 0x%x\n", cmd, sense_key);
3043                 return error_value;
3044         }
3045
3046         /* SG_IO or similar, copy sense data back */
3047         if (cmd->rq->sense) {
3048                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3049                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3050                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3051                         cmd->rq->sense_len);
3052         } else
3053                 cmd->rq->sense_len = 0;
3054
3055         return error_value;
3056 }
3057
3058 /* checks the status of the job and calls complete buffers to mark all
3059  * buffers for the completed job. Note that this function does not need
3060  * to hold the hba/queue lock.
3061  */
3062 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3063                                     int timeout)
3064 {
3065         int retry_cmd = 0;
3066         struct request *rq = cmd->rq;
3067
3068         rq->errors = 0;
3069
3070         if (timeout)
3071                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3072
3073         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3074                 goto after_error_processing;
3075
3076         switch (cmd->err_info->CommandStatus) {
3077         case CMD_TARGET_STATUS:
3078                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3079                 break;
3080         case CMD_DATA_UNDERRUN:
3081                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3082                         printk(KERN_WARNING "cciss: cmd %p has"
3083                                " completed with data underrun "
3084                                "reported\n", cmd);
3085                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3086                 }
3087                 break;
3088         case CMD_DATA_OVERRUN:
3089                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3090                         printk(KERN_WARNING "cciss: cmd %p has"
3091                                " completed with data overrun "
3092                                "reported\n", cmd);
3093                 break;
3094         case CMD_INVALID:
3095                 printk(KERN_WARNING "cciss: cmd %p is "
3096                        "reported invalid\n", cmd);
3097                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3098                         cmd->err_info->CommandStatus, DRIVER_OK,
3099                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3100                                 DID_PASSTHROUGH : DID_ERROR);
3101                 break;
3102         case CMD_PROTOCOL_ERR:
3103                 printk(KERN_WARNING "cciss: cmd %p has "
3104                        "protocol error \n", cmd);
3105                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3106                         cmd->err_info->CommandStatus, DRIVER_OK,
3107                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3108                                 DID_PASSTHROUGH : DID_ERROR);
3109                 break;
3110         case CMD_HARDWARE_ERR:
3111                 printk(KERN_WARNING "cciss: cmd %p had "
3112                        " hardware error\n", cmd);
3113                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3114                         cmd->err_info->CommandStatus, DRIVER_OK,
3115                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3116                                 DID_PASSTHROUGH : DID_ERROR);
3117                 break;
3118         case CMD_CONNECTION_LOST:
3119                 printk(KERN_WARNING "cciss: cmd %p had "
3120                        "connection lost\n", cmd);
3121                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3122                         cmd->err_info->CommandStatus, DRIVER_OK,
3123                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3124                                 DID_PASSTHROUGH : DID_ERROR);
3125                 break;
3126         case CMD_ABORTED:
3127                 printk(KERN_WARNING "cciss: cmd %p was "
3128                        "aborted\n", cmd);
3129                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3130                         cmd->err_info->CommandStatus, DRIVER_OK,
3131                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3132                                 DID_PASSTHROUGH : DID_ABORT);
3133                 break;
3134         case CMD_ABORT_FAILED:
3135                 printk(KERN_WARNING "cciss: cmd %p reports "
3136                        "abort failed\n", cmd);
3137                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3138                         cmd->err_info->CommandStatus, DRIVER_OK,
3139                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3140                                 DID_PASSTHROUGH : DID_ERROR);
3141                 break;
3142         case CMD_UNSOLICITED_ABORT:
3143                 printk(KERN_WARNING "cciss%d: unsolicited "
3144                        "abort %p\n", h->ctlr, cmd);
3145                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3146                         retry_cmd = 1;
3147                         printk(KERN_WARNING
3148                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3149                         cmd->retry_count++;
3150                 } else
3151                         printk(KERN_WARNING
3152                                "cciss%d: %p retried too "
3153                                "many times\n", h->ctlr, cmd);
3154                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3155                         cmd->err_info->CommandStatus, DRIVER_OK,
3156                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3157                                 DID_PASSTHROUGH : DID_ABORT);
3158                 break;
3159         case CMD_TIMEOUT:
3160                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3161                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3162                         cmd->err_info->CommandStatus, DRIVER_OK,
3163                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3164                                 DID_PASSTHROUGH : DID_ERROR);
3165                 break;
3166         default:
3167                 printk(KERN_WARNING "cciss: cmd %p returned "
3168                        "unknown status %x\n", cmd,
3169                        cmd->err_info->CommandStatus);
3170                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3171                         cmd->err_info->CommandStatus, DRIVER_OK,
3172                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3173                                 DID_PASSTHROUGH : DID_ERROR);
3174         }
3175
3176 after_error_processing:
3177
3178         /* We need to return this command */
3179         if (retry_cmd) {
3180                 resend_cciss_cmd(h, cmd);
3181                 return;
3182         }
3183         cmd->rq->completion_data = cmd;
3184         blk_complete_request(cmd->rq);
3185 }
3186
3187 static inline u32 cciss_tag_contains_index(u32 tag)
3188 {
3189 #define DIRECT_LOOKUP_BIT 0x10
3190         return tag & DIRECT_LOOKUP_BIT;
3191 }
3192
3193 static inline u32 cciss_tag_to_index(u32 tag)
3194 {
3195 #define DIRECT_LOOKUP_SHIFT 5
3196         return tag >> DIRECT_LOOKUP_SHIFT;
3197 }
3198
3199 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3200 {
3201 #define CCISS_ERROR_BITS 0x03
3202         return tag & ~CCISS_ERROR_BITS;
3203 }
3204
3205 static inline void cciss_mark_tag_indexed(u32 *tag)
3206 {
3207         *tag |= DIRECT_LOOKUP_BIT;
3208 }
3209
3210 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3211 {
3212         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3213 }
3214
3215 /*
3216  * Get a request and submit it to the controller.
3217  */
3218 static void do_cciss_request(struct request_queue *q)
3219 {
3220         ctlr_info_t *h = q->queuedata;
3221         CommandList_struct *c;
3222         sector_t start_blk;
3223         int seg;
3224         struct request *creq;
3225         u64bit temp64;
3226         struct scatterlist *tmp_sg;
3227         SGDescriptor_struct *curr_sg;
3228         drive_info_struct *drv;
3229         int i, dir;
3230         int sg_index = 0;
3231         int chained = 0;
3232
3233         /* We call start_io here in case there is a command waiting on the
3234          * queue that has not been sent.
3235          */
3236         if (blk_queue_plugged(q))
3237                 goto startio;
3238
3239       queue:
3240         creq = blk_peek_request(q);
3241         if (!creq)
3242                 goto startio;
3243
3244         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3245
3246         if ((c = cmd_alloc(h, 1)) == NULL)
3247                 goto full;
3248
3249         blk_start_request(creq);
3250
3251         tmp_sg = h->scatter_list[c->cmdindex];
3252         spin_unlock_irq(q->queue_lock);
3253
3254         c->cmd_type = CMD_RWREQ;
3255         c->rq = creq;
3256
3257         /* fill in the request */
3258         drv = creq->rq_disk->private_data;
3259         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3260         /* got command from pool, so use the command block index instead */
3261         /* for direct lookups. */
3262         /* The first 2 bits are reserved for controller error reporting. */
3263         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3264         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3265         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3266         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3267         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3268         c->Request.Type.Attribute = ATTR_SIMPLE;
3269         c->Request.Type.Direction =
3270             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3271         c->Request.Timeout = 0; /* Don't time out */
3272         c->Request.CDB[0] =
3273             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3274         start_blk = blk_rq_pos(creq);
3275 #ifdef CCISS_DEBUG
3276         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3277                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3278 #endif                          /* CCISS_DEBUG */
3279
3280         sg_init_table(tmp_sg, h->maxsgentries);
3281         seg = blk_rq_map_sg(q, creq, tmp_sg);
3282
3283         /* get the DMA records for the setup */
3284         if (c->Request.Type.Direction == XFER_READ)
3285                 dir = PCI_DMA_FROMDEVICE;
3286         else
3287                 dir = PCI_DMA_TODEVICE;
3288
3289         curr_sg = c->SG;
3290         sg_index = 0;
3291         chained = 0;
3292
3293         for (i = 0; i < seg; i++) {
3294                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3295                         !chained && ((seg - i) > 1)) {
3296                         /* Point to next chain block. */
3297                         curr_sg = h->cmd_sg_list[c->cmdindex];
3298                         sg_index = 0;
3299                         chained = 1;
3300                 }
3301                 curr_sg[sg_index].Len = tmp_sg[i].length;
3302                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3303                                                 tmp_sg[i].offset,
3304                                                 tmp_sg[i].length, dir);
3305                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3306                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3307                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3308                 ++sg_index;
3309         }
3310         if (chained)
3311                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3312                         (seg - (h->max_cmd_sgentries - 1)) *
3313                                 sizeof(SGDescriptor_struct));
3314
3315         /* track how many SG entries we are using */
3316         if (seg > h->maxSG)
3317                 h->maxSG = seg;
3318
3319 #ifdef CCISS_DEBUG
3320         printk(KERN_DEBUG "cciss: Submitting %ld sectors in %d segments "
3321                         "chained[%d]\n",
3322                         blk_rq_sectors(creq), seg, chained);
3323 #endif                          /* CCISS_DEBUG */
3324
3325         c->Header.SGTotal = seg + chained;
3326         if (seg <= h->max_cmd_sgentries)
3327                 c->Header.SGList = c->Header.SGTotal;
3328         else
3329                 c->Header.SGList = h->max_cmd_sgentries;
3330         set_performant_mode(h, c);
3331
3332         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3333                 if(h->cciss_read == CCISS_READ_10) {
3334                         c->Request.CDB[1] = 0;
3335                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3336                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3337                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3338                         c->Request.CDB[5] = start_blk & 0xff;
3339                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3340                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3341                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3342                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3343                 } else {
3344                         u32 upper32 = upper_32_bits(start_blk);
3345
3346                         c->Request.CDBLen = 16;
3347                         c->Request.CDB[1]= 0;
3348                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3349                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3350                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3351                         c->Request.CDB[5]= upper32 & 0xff;
3352                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3353                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3354                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3355                         c->Request.CDB[9]= start_blk & 0xff;
3356                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3357                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3358                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3359                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3360                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3361                 }
3362         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3363                 c->Request.CDBLen = creq->cmd_len;
3364                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3365         } else {
3366                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3367                 BUG();
3368         }
3369
3370         spin_lock_irq(q->queue_lock);
3371
3372         addQ(&h->reqQ, c);
3373         h->Qdepth++;
3374         if (h->Qdepth > h->maxQsinceinit)
3375                 h->maxQsinceinit = h->Qdepth;
3376
3377         goto queue;
3378 full:
3379         blk_stop_queue(q);
3380 startio:
3381         /* We will already have the driver lock here so not need
3382          * to lock it.
3383          */
3384         start_io(h);
3385 }
3386
3387 static inline unsigned long get_next_completion(ctlr_info_t *h)
3388 {
3389         return h->access.command_completed(h);
3390 }
3391
3392 static inline int interrupt_pending(ctlr_info_t *h)
3393 {
3394         return h->access.intr_pending(h);
3395 }
3396
3397 static inline long interrupt_not_for_us(ctlr_info_t *h)
3398 {
3399         return !(h->msi_vector || h->msix_vector) &&
3400                 ((h->access.intr_pending(h) == 0) ||
3401                 (h->interrupts_enabled == 0));
3402 }
3403
3404 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3405                         u32 raw_tag)
3406 {
3407         if (unlikely(tag_index >= h->nr_cmds)) {
3408                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3409                 return 1;
3410         }
3411         return 0;
3412 }
3413
3414 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3415                                 u32 raw_tag)
3416 {
3417         removeQ(c);
3418         if (likely(c->cmd_type == CMD_RWREQ))
3419                 complete_command(h, c, 0);
3420         else if (c->cmd_type == CMD_IOCTL_PEND)
3421                 complete(c->waiting);
3422 #ifdef CONFIG_CISS_SCSI_TAPE
3423         else if (c->cmd_type == CMD_SCSI)
3424                 complete_scsi_command(c, 0, raw_tag);
3425 #endif
3426 }
3427
3428 static inline u32 next_command(ctlr_info_t *h)
3429 {
3430         u32 a;
3431
3432         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3433                 return h->access.command_completed(h);
3434
3435         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3436                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3437                 (h->reply_pool_head)++;
3438                 h->commands_outstanding--;
3439         } else {
3440                 a = FIFO_EMPTY;
3441         }
3442         /* Check for wraparound */
3443         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3444                 h->reply_pool_head = h->reply_pool;
3445                 h->reply_pool_wraparound ^= 1;
3446         }
3447         return a;
3448 }
3449
3450 /* process completion of an indexed ("direct lookup") command */
3451 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3452 {
3453         u32 tag_index;
3454         CommandList_struct *c;
3455
3456         tag_index = cciss_tag_to_index(raw_tag);
3457         if (bad_tag(h, tag_index, raw_tag))
3458                 return next_command(h);
3459         c = h->cmd_pool + tag_index;
3460         finish_cmd(h, c, raw_tag);
3461         return next_command(h);
3462 }
3463
3464 /* process completion of a non-indexed command */
3465 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3466 {
3467         u32 tag;
3468         CommandList_struct *c = NULL;
3469         struct hlist_node *tmp;
3470         __u32 busaddr_masked, tag_masked;
3471
3472         tag = cciss_tag_discard_error_bits(raw_tag);
3473         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3474                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3475                 tag_masked = cciss_tag_discard_error_bits(tag);
3476                 if (busaddr_masked == tag_masked) {
3477                         finish_cmd(h, c, raw_tag);
3478                         return next_command(h);
3479                 }
3480         }
3481         bad_tag(h, h->nr_cmds + 1, raw_tag);
3482         return next_command(h);
3483 }
3484
3485 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3486 {
3487         ctlr_info_t *h = dev_id;
3488         unsigned long flags;
3489         u32 raw_tag;
3490
3491         if (interrupt_not_for_us(h))
3492                 return IRQ_NONE;
3493         /*
3494          * If there are completed commands in the completion queue,
3495          * we had better do something about it.
3496          */
3497         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3498         while (interrupt_pending(h)) {
3499                 raw_tag = get_next_completion(h);
3500                 while (raw_tag != FIFO_EMPTY) {
3501                         if (cciss_tag_contains_index(raw_tag))
3502                                 raw_tag = process_indexed_cmd(h, raw_tag);
3503                         else
3504                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3505                 }
3506         }
3507
3508         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3509         return IRQ_HANDLED;
3510 }
3511
3512 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3513  * check the interrupt pending register because it is not set.
3514  */
3515 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3516 {
3517         ctlr_info_t *h = dev_id;
3518         unsigned long flags;
3519         u32 raw_tag;
3520
3521         if (interrupt_not_for_us(h))
3522                 return IRQ_NONE;
3523         /*
3524          * If there are completed commands in the completion queue,
3525          * we had better do something about it.
3526          */
3527         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3528         raw_tag = get_next_completion(h);
3529         while (raw_tag != FIFO_EMPTY) {
3530                 if (cciss_tag_contains_index(raw_tag))
3531                         raw_tag = process_indexed_cmd(h, raw_tag);
3532                 else
3533                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3534         }
3535
3536         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3537         return IRQ_HANDLED;
3538 }
3539
3540 /**
3541  * add_to_scan_list() - add controller to rescan queue
3542  * @h:                Pointer to the controller.
3543  *
3544  * Adds the controller to the rescan queue if not already on the queue.
3545  *
3546  * returns 1 if added to the queue, 0 if skipped (could be on the
3547  * queue already, or the controller could be initializing or shutting
3548  * down).
3549  **/
3550 static int add_to_scan_list(struct ctlr_info *h)
3551 {
3552         struct ctlr_info *test_h;
3553         int found = 0;
3554         int ret = 0;
3555
3556         if (h->busy_initializing)
3557                 return 0;
3558
3559         if (!mutex_trylock(&h->busy_shutting_down))
3560                 return 0;
3561
3562         mutex_lock(&scan_mutex);
3563         list_for_each_entry(test_h, &scan_q, scan_list) {
3564                 if (test_h == h) {
3565                         found = 1;
3566                         break;
3567                 }
3568         }
3569         if (!found && !h->busy_scanning) {
3570                 INIT_COMPLETION(h->scan_wait);
3571                 list_add_tail(&h->scan_list, &scan_q);
3572                 ret = 1;
3573         }
3574         mutex_unlock(&scan_mutex);
3575         mutex_unlock(&h->busy_shutting_down);
3576
3577         return ret;
3578 }
3579
3580 /**
3581  * remove_from_scan_list() - remove controller from rescan queue
3582  * @h:                     Pointer to the controller.
3583  *
3584  * Removes the controller from the rescan queue if present. Blocks if
3585  * the controller is currently conducting a rescan.  The controller
3586  * can be in one of three states:
3587  * 1. Doesn't need a scan
3588  * 2. On the scan list, but not scanning yet (we remove it)
3589  * 3. Busy scanning (and not on the list). In this case we want to wait for
3590  *    the scan to complete to make sure the scanning thread for this
3591  *    controller is completely idle.
3592  **/
3593 static void remove_from_scan_list(struct ctlr_info *h)
3594 {
3595         struct ctlr_info *test_h, *tmp_h;
3596
3597         mutex_lock(&scan_mutex);
3598         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3599                 if (test_h == h) { /* state 2. */
3600                         list_del(&h->scan_list);
3601                         complete_all(&h->scan_wait);
3602                         mutex_unlock(&scan_mutex);
3603                         return;
3604                 }
3605         }
3606         if (h->busy_scanning) { /* state 3. */
3607                 mutex_unlock(&scan_mutex);
3608                 wait_for_completion(&h->scan_wait);
3609         } else { /* state 1, nothing to do. */
3610                 mutex_unlock(&scan_mutex);
3611         }
3612 }
3613
3614 /**
3615  * scan_thread() - kernel thread used to rescan controllers
3616  * @data:        Ignored.
3617  *
3618  * A kernel thread used scan for drive topology changes on
3619  * controllers. The thread processes only one controller at a time
3620  * using a queue.  Controllers are added to the queue using
3621  * add_to_scan_list() and removed from the queue either after done
3622  * processing or using remove_from_scan_list().
3623  *
3624  * returns 0.
3625  **/
3626 static int scan_thread(void *data)
3627 {
3628         struct ctlr_info *h;
3629
3630         while (1) {
3631                 set_current_state(TASK_INTERRUPTIBLE);
3632                 schedule();
3633                 if (kthread_should_stop())
3634                         break;
3635
3636                 while (1) {
3637                         mutex_lock(&scan_mutex);
3638                         if (list_empty(&scan_q)) {
3639                                 mutex_unlock(&scan_mutex);
3640                                 break;
3641                         }
3642
3643                         h = list_entry(scan_q.next,
3644                                        struct ctlr_info,
3645                                        scan_list);
3646                         list_del(&h->scan_list);
3647                         h->busy_scanning = 1;
3648                         mutex_unlock(&scan_mutex);
3649
3650                         rebuild_lun_table(h, 0, 0);
3651                         complete_all(&h->scan_wait);
3652                         mutex_lock(&scan_mutex);
3653                         h->busy_scanning = 0;
3654                         mutex_unlock(&scan_mutex);
3655                 }
3656         }
3657
3658         return 0;
3659 }
3660
3661 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3662 {
3663         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3664                 return 0;
3665
3666         switch (c->err_info->SenseInfo[12]) {
3667         case STATE_CHANGED:
3668                 printk(KERN_WARNING "cciss%d: a state change "
3669                         "detected, command retried\n", h->ctlr);
3670                 return 1;
3671         break;
3672         case LUN_FAILED:
3673                 printk(KERN_WARNING "cciss%d: LUN failure "
3674                         "detected, action required\n", h->ctlr);
3675                 return 1;
3676         break;
3677         case REPORT_LUNS_CHANGED:
3678                 printk(KERN_WARNING "cciss%d: report LUN data "
3679                         "changed\n", h->ctlr);
3680         /*
3681          * Here, we could call add_to_scan_list and wake up the scan thread,
3682          * except that it's quite likely that we will get more than one
3683          * REPORT_LUNS_CHANGED condition in quick succession, which means
3684          * that those which occur after the first one will likely happen
3685          * *during* the scan_thread's rescan.  And the rescan code is not
3686          * robust enough to restart in the middle, undoing what it has already
3687          * done, and it's not clear that it's even possible to do this, since
3688          * part of what it does is notify the block layer, which starts
3689          * doing it's own i/o to read partition tables and so on, and the
3690          * driver doesn't have visibility to know what might need undoing.
3691          * In any event, if possible, it is horribly complicated to get right
3692          * so we just don't do it for now.
3693          *
3694          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3695          */
3696                 return 1;
3697         break;
3698         case POWER_OR_RESET:
3699                 printk(KERN_WARNING "cciss%d: a power on "
3700                         "or device reset detected\n", h->ctlr);
3701                 return 1;
3702         break;
3703         case UNIT_ATTENTION_CLEARED:
3704                 printk(KERN_WARNING "cciss%d: unit attention "
3705                     "cleared by another initiator\n", h->ctlr);
3706                 return 1;
3707         break;
3708         default:
3709                 printk(KERN_WARNING "cciss%d: unknown "
3710                         "unit attention detected\n", h->ctlr);
3711                                 return 1;
3712         }
3713 }
3714
3715 /*
3716  *  We cannot read the structure directly, for portability we must use
3717  *   the io functions.
3718  *   This is for debug only.
3719  */
3720 static void print_cfg_table(CfgTable_struct *tb)
3721 {
3722 #ifdef CCISS_DEBUG
3723         int i;
3724         char temp_name[17];
3725
3726         printk("Controller Configuration information\n");
3727         printk("------------------------------------\n");
3728         for (i = 0; i < 4; i++)
3729                 temp_name[i] = readb(&(tb->Signature[i]));
3730         temp_name[4] = '\0';
3731         printk("   Signature = %s\n", temp_name);
3732         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3733         printk("   Transport methods supported = 0x%x\n",
3734                readl(&(tb->TransportSupport)));
3735         printk("   Transport methods active = 0x%x\n",
3736                readl(&(tb->TransportActive)));
3737         printk("   Requested transport Method = 0x%x\n",
3738                readl(&(tb->HostWrite.TransportRequest)));
3739         printk("   Coalesce Interrupt Delay = 0x%x\n",
3740                readl(&(tb->HostWrite.CoalIntDelay)));
3741         printk("   Coalesce Interrupt Count = 0x%x\n",
3742                readl(&(tb->HostWrite.CoalIntCount)));
3743         printk("   Max outstanding commands = 0x%d\n",
3744                readl(&(tb->CmdsOutMax)));
3745         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3746         for (i = 0; i < 16; i++)
3747                 temp_name[i] = readb(&(tb->ServerName[i]));
3748         temp_name[16] = '\0';
3749         printk("   Server Name = %s\n", temp_name);
3750         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3751 #endif                          /* CCISS_DEBUG */
3752 }
3753
3754 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3755 {
3756         int i, offset, mem_type, bar_type;
3757         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3758                 return 0;
3759         offset = 0;
3760         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3761                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3762                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3763                         offset += 4;
3764                 else {
3765                         mem_type = pci_resource_flags(pdev, i) &
3766                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3767                         switch (mem_type) {
3768                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3769                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3770                                 offset += 4;    /* 32 bit */
3771                                 break;
3772                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3773                                 offset += 8;
3774                                 break;
3775                         default:        /* reserved in PCI 2.2 */
3776                                 printk(KERN_WARNING
3777                                        "Base address is invalid\n");
3778                                 return -1;
3779                                 break;
3780                         }
3781                 }
3782                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3783                         return i + 1;
3784         }
3785         return -1;
3786 }
3787
3788 /* Fill in bucket_map[], given nsgs (the max number of
3789  * scatter gather elements supported) and bucket[],
3790  * which is an array of 8 integers.  The bucket[] array
3791  * contains 8 different DMA transfer sizes (in 16
3792  * byte increments) which the controller uses to fetch
3793  * commands.  This function fills in bucket_map[], which
3794  * maps a given number of scatter gather elements to one of
3795  * the 8 DMA transfer sizes.  The point of it is to allow the
3796  * controller to only do as much DMA as needed to fetch the
3797  * command, with the DMA transfer size encoded in the lower
3798  * bits of the command address.
3799  */
3800 static void  calc_bucket_map(int bucket[], int num_buckets,
3801         int nsgs, int *bucket_map)
3802 {
3803         int i, j, b, size;
3804
3805         /* even a command with 0 SGs requires 4 blocks */
3806 #define MINIMUM_TRANSFER_BLOCKS 4
3807 #define NUM_BUCKETS 8
3808         /* Note, bucket_map must have nsgs+1 entries. */
3809         for (i = 0; i <= nsgs; i++) {
3810                 /* Compute size of a command with i SG entries */
3811                 size = i + MINIMUM_TRANSFER_BLOCKS;
3812                 b = num_buckets; /* Assume the biggest bucket */
3813                 /* Find the bucket that is just big enough */
3814                 for (j = 0; j < 8; j++) {
3815                         if (bucket[j] >= size) {
3816                                 b = j;
3817                                 break;
3818                         }
3819                 }
3820                 /* for a command with i SG entries, use bucket b. */
3821                 bucket_map[i] = b;
3822         }
3823 }
3824
3825 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3826 {
3827         int i;
3828
3829         /* under certain very rare conditions, this can take awhile.
3830          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3831          * as we enter this code.) */
3832         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3833                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3834                         break;
3835                 msleep(10);
3836         }
3837 }
3838
3839 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3840 {
3841         /* This is a bit complicated.  There are 8 registers on
3842          * the controller which we write to to tell it 8 different
3843          * sizes of commands which there may be.  It's a way of
3844          * reducing the DMA done to fetch each command.  Encoded into
3845          * each command's tag are 3 bits which communicate to the controller
3846          * which of the eight sizes that command fits within.  The size of
3847          * each command depends on how many scatter gather entries there are.
3848          * Each SG entry requires 16 bytes.  The eight registers are programmed
3849          * with the number of 16-byte blocks a command of that size requires.
3850          * The smallest command possible requires 5 such 16 byte blocks.
3851          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3852          * blocks.  Note, this only extends to the SG entries contained
3853          * within the command block, and does not extend to chained blocks
3854          * of SG elements.   bft[] contains the eight values we write to
3855          * the registers.  They are not evenly distributed, but have more
3856          * sizes for small commands, and fewer sizes for larger commands.
3857          */
3858         __u32 trans_offset;
3859         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3860                         /*
3861                          *  5 = 1 s/g entry or 4k
3862                          *  6 = 2 s/g entry or 8k
3863                          *  8 = 4 s/g entry or 16k
3864                          * 10 = 6 s/g entry or 24k
3865                          */
3866         unsigned long register_value;
3867         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3868
3869         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3870
3871         /* Controller spec: zero out this buffer. */
3872         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3873         h->reply_pool_head = h->reply_pool;
3874
3875         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3876         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3877                                 h->blockFetchTable);
3878         writel(bft[0], &h->transtable->BlockFetch0);
3879         writel(bft[1], &h->transtable->BlockFetch1);
3880         writel(bft[2], &h->transtable->BlockFetch2);
3881         writel(bft[3], &h->transtable->BlockFetch3);
3882         writel(bft[4], &h->transtable->BlockFetch4);
3883         writel(bft[5], &h->transtable->BlockFetch5);
3884         writel(bft[6], &h->transtable->BlockFetch6);
3885         writel(bft[7], &h->transtable->BlockFetch7);
3886
3887         /* size of controller ring buffer */
3888         writel(h->max_commands, &h->transtable->RepQSize);
3889         writel(1, &h->transtable->RepQCount);
3890         writel(0, &h->transtable->RepQCtrAddrLow32);
3891         writel(0, &h->transtable->RepQCtrAddrHigh32);
3892         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3893         writel(0, &h->transtable->RepQAddr0High32);
3894         writel(CFGTBL_Trans_Performant,
3895                         &(h->cfgtable->HostWrite.TransportRequest));
3896
3897         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3898         cciss_wait_for_mode_change_ack(h);
3899         register_value = readl(&(h->cfgtable->TransportActive));
3900         if (!(register_value & CFGTBL_Trans_Performant))
3901                 printk(KERN_WARNING "cciss: unable to get board into"
3902                                         " performant mode\n");
3903 }
3904
3905 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3906 {
3907         __u32 trans_support;
3908
3909         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3910         /* Attempt to put controller into performant mode if supported */
3911         /* Does board support performant mode? */
3912         trans_support = readl(&(h->cfgtable->TransportSupport));
3913         if (!(trans_support & PERFORMANT_MODE))
3914                 return;
3915
3916         printk(KERN_WARNING "cciss%d: Placing controller into "
3917                                 "performant mode\n", h->ctlr);
3918         /* Performant mode demands commands on a 32 byte boundary
3919          * pci_alloc_consistent aligns on page boundarys already.
3920          * Just need to check if divisible by 32
3921          */
3922         if ((sizeof(CommandList_struct) % 32) != 0) {
3923                 printk(KERN_WARNING "%s %d %s\n",
3924                         "cciss info: command size[",
3925                         (int)sizeof(CommandList_struct),
3926                         "] not divisible by 32, no performant mode..\n");
3927                 return;
3928         }
3929
3930         /* Performant mode ring buffer and supporting data structures */
3931         h->reply_pool = (__u64 *)pci_alloc_consistent(
3932                 h->pdev, h->max_commands * sizeof(__u64),
3933                 &(h->reply_pool_dhandle));
3934
3935         /* Need a block fetch table for performant mode */
3936         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3937                 sizeof(__u32)), GFP_KERNEL);
3938
3939         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3940                 goto clean_up;
3941
3942         cciss_enter_performant_mode(h);
3943
3944         /* Change the access methods to the performant access methods */
3945         h->access = SA5_performant_access;
3946         h->transMethod = CFGTBL_Trans_Performant;
3947
3948         return;
3949 clean_up:
3950         kfree(h->blockFetchTable);
3951         if (h->reply_pool)
3952                 pci_free_consistent(h->pdev,
3953                                 h->max_commands * sizeof(__u64),
3954                                 h->reply_pool,
3955                                 h->reply_pool_dhandle);
3956         return;
3957
3958 } /* cciss_put_controller_into_performant_mode */
3959
3960 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3961  * controllers that are capable. If not, we use IO-APIC mode.
3962  */
3963
3964 static void __devinit cciss_interrupt_mode(ctlr_info_t *c)
3965 {
3966 #ifdef CONFIG_PCI_MSI
3967         int err;
3968         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3969         {0, 2}, {0, 3}
3970         };
3971
3972         /* Some boards advertise MSI but don't really support it */
3973         if ((c->board_id == 0x40700E11) || (c->board_id == 0x40800E11) ||
3974             (c->board_id == 0x40820E11) || (c->board_id == 0x40830E11))
3975                 goto default_int_mode;
3976
3977         if (pci_find_capability(c->pdev, PCI_CAP_ID_MSIX)) {
3978                 err = pci_enable_msix(c->pdev, cciss_msix_entries, 4);
3979                 if (!err) {
3980                         c->intr[0] = cciss_msix_entries[0].vector;
3981                         c->intr[1] = cciss_msix_entries[1].vector;
3982                         c->intr[2] = cciss_msix_entries[2].vector;
3983                         c->intr[3] = cciss_msix_entries[3].vector;
3984                         c->msix_vector = 1;
3985                         return;
3986                 }
3987                 if (err > 0) {
3988                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3989                                "available\n", err);
3990                         goto default_int_mode;
3991                 } else {
3992                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3993                                err);
3994                         goto default_int_mode;
3995                 }
3996         }
3997         if (pci_find_capability(c->pdev, PCI_CAP_ID_MSI)) {
3998                 if (!pci_enable_msi(c->pdev)) {
3999                         c->msi_vector = 1;
4000                 } else {
4001                         printk(KERN_WARNING "cciss: MSI init failed\n");
4002                 }
4003         }
4004 default_int_mode:
4005 #endif                          /* CONFIG_PCI_MSI */
4006         /* if we get here we're going to use the default interrupt mode */
4007         c->intr[PERF_MODE_INT] = c->pdev->irq;
4008         return;
4009 }
4010
4011 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
4012 {
4013         int i;
4014         u32 subsystem_vendor_id, subsystem_device_id;
4015
4016         subsystem_vendor_id = pdev->subsystem_vendor;
4017         subsystem_device_id = pdev->subsystem_device;
4018         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
4019                         subsystem_vendor_id;
4020
4021         for (i = 0; i < ARRAY_SIZE(products); i++) {
4022                 /* Stand aside for hpsa driver on request */
4023                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
4024                         return -ENODEV;
4025                 if (*board_id == products[i].board_id)
4026                         return i;
4027         }
4028         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
4029                 *board_id);
4030         return -ENODEV;
4031 }
4032
4033 static inline bool cciss_board_disabled(ctlr_info_t *h)
4034 {
4035         u16 command;
4036
4037         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4038         return ((command & PCI_COMMAND_MEMORY) == 0);
4039 }
4040
4041 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4042         unsigned long *memory_bar)
4043 {
4044         int i;
4045
4046         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4047                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4048                         /* addressing mode bits already removed */
4049                         *memory_bar = pci_resource_start(pdev, i);
4050                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4051                                 *memory_bar);
4052                         return 0;
4053                 }
4054         dev_warn(&pdev->dev, "no memory BAR found\n");
4055         return -ENODEV;
4056 }
4057
4058 static int __devinit cciss_wait_for_board_ready(ctlr_info_t *h)
4059 {
4060         int i;
4061         u32 scratchpad;
4062
4063         for (i = 0; i < CCISS_BOARD_READY_ITERATIONS; i++) {
4064                 scratchpad = readl(h->vaddr + SA5_SCRATCHPAD_OFFSET);
4065                 if (scratchpad == CCISS_FIRMWARE_READY)
4066                         return 0;
4067                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4068         }
4069         dev_warn(&h->pdev->dev, "board not ready, timed out.\n");
4070         return -ENODEV;
4071 }
4072
4073 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4074         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4075         u64 *cfg_offset)
4076 {
4077         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4078         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4079         *cfg_base_addr &= (u32) 0x0000ffff;
4080         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4081         if (*cfg_base_addr_index == -1) {
4082                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4083                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4084                 return -ENODEV;
4085         }
4086         return 0;
4087 }
4088
4089 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4090 {
4091         u64 cfg_offset;
4092         u32 cfg_base_addr;
4093         u64 cfg_base_addr_index;
4094         u32 trans_offset;
4095         int rc;
4096
4097         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4098                 &cfg_base_addr_index, &cfg_offset);
4099         if (rc)
4100                 return rc;
4101         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4102                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4103         if (!h->cfgtable)
4104                 return -ENOMEM;
4105         /* Find performant mode table. */
4106         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4107         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4108                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4109                                 sizeof(*h->transtable));
4110         if (!h->transtable)
4111                 return -ENOMEM;
4112         return 0;
4113 }
4114
4115 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4116 {
4117         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4118         if (h->max_commands < 16) {
4119                 dev_warn(&h->pdev->dev, "Controller reports "
4120                         "max supported commands of %d, an obvious lie. "
4121                         "Using 16.  Ensure that firmware is up to date.\n",
4122                         h->max_commands);
4123                 h->max_commands = 16;
4124         }
4125 }
4126
4127 /* Interrogate the hardware for some limits:
4128  * max commands, max SG elements without chaining, and with chaining,
4129  * SG chain block size, etc.
4130  */
4131 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4132 {
4133         cciss_get_max_perf_mode_cmds(h);
4134         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4135         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4136         /*
4137          * Limit in-command s/g elements to 32 save dma'able memory.
4138          * Howvever spec says if 0, use 31
4139          */
4140         h->max_cmd_sgentries = 31;
4141         if (h->maxsgentries > 512) {
4142                 h->max_cmd_sgentries = 32;
4143                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4144                 h->maxsgentries--; /* save one for chain pointer */
4145         } else {
4146                 h->maxsgentries = 31; /* default to traditional values */
4147                 h->chainsize = 0;
4148         }
4149 }
4150
4151 static inline bool CISS_signature_present(ctlr_info_t *h)
4152 {
4153         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4154             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4155             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4156             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4157                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4158                 return false;
4159         }
4160         return true;
4161 }
4162
4163 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4164 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4165 {
4166 #ifdef CONFIG_X86
4167         u32 prefetch;
4168
4169         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4170         prefetch |= 0x100;
4171         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4172 #endif
4173 }
4174
4175 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4176  * in a prefetch beyond physical memory.
4177  */
4178 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4179 {
4180         u32 dma_prefetch;
4181         __u32 dma_refetch;
4182
4183         if (h->board_id != 0x3225103C)
4184                 return;
4185         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4186         dma_prefetch |= 0x8000;
4187         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4188         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4189         dma_refetch |= 0x1;
4190         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4191 }
4192
4193 static int __devinit cciss_pci_init(ctlr_info_t *c)
4194 {
4195         int prod_index, err;
4196
4197         prod_index = cciss_lookup_board_id(c->pdev, &c->board_id);
4198         if (prod_index < 0)
4199                 return -ENODEV;
4200         c->product_name = products[prod_index].product_name;
4201         c->access = *(products[prod_index].access);
4202
4203         if (cciss_board_disabled(c)) {
4204                 printk(KERN_WARNING
4205                        "cciss: controller appears to be disabled\n");
4206                 return -ENODEV;
4207         }
4208         err = pci_enable_device(c->pdev);
4209         if (err) {
4210                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
4211                 return err;
4212         }
4213
4214         err = pci_request_regions(c->pdev, "cciss");
4215         if (err) {
4216                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
4217                        "aborting\n");
4218                 return err;
4219         }
4220
4221 #ifdef CCISS_DEBUG
4222         printk(KERN_INFO "command = %x\n", command);
4223         printk(KERN_INFO "irq = %x\n", c->pdev->irq);
4224         printk(KERN_INFO "board_id = %x\n", c->board_id);
4225 #endif                          /* CCISS_DEBUG */
4226
4227 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4228  * else we use the IO-APIC interrupt assigned to us by system ROM.
4229  */
4230         cciss_interrupt_mode(c);
4231         err = cciss_pci_find_memory_BAR(c->pdev, &c->paddr);
4232         if (err)
4233                 goto err_out_free_res;
4234         c->vaddr = remap_pci_mem(c->paddr, 0x250);
4235         if (!c->vaddr) {
4236                 err = -ENOMEM;
4237                 goto err_out_free_res;
4238         }
4239         err = cciss_wait_for_board_ready(c);
4240         if (err)
4241                 goto err_out_free_res;
4242         err = cciss_find_cfgtables(c);
4243         if (err)
4244                 goto err_out_free_res;
4245         print_cfg_table(c->cfgtable);
4246         cciss_find_board_params(c);
4247
4248         if (!CISS_signature_present(c)) {
4249                 err = -ENODEV;
4250                 goto err_out_free_res;
4251         }
4252         cciss_enable_scsi_prefetch(c);
4253         cciss_p600_dma_prefetch_quirk(c);
4254         cciss_put_controller_into_performant_mode(c);
4255         return 0;
4256
4257 err_out_free_res:
4258         /*
4259          * Deliberately omit pci_disable_device(): it does something nasty to
4260          * Smart Array controllers that pci_enable_device does not undo
4261          */
4262         if (c->transtable)
4263                 iounmap(c->transtable);
4264         if (c->cfgtable)
4265                 iounmap(c->cfgtable);
4266         if (c->vaddr)
4267                 iounmap(c->vaddr);
4268         pci_release_regions(c->pdev);
4269         return err;
4270 }
4271
4272 /* Function to find the first free pointer into our hba[] array
4273  * Returns -1 if no free entries are left.
4274  */
4275 static int alloc_cciss_hba(void)
4276 {
4277         int i;
4278
4279         for (i = 0; i < MAX_CTLR; i++) {
4280                 if (!hba[i]) {
4281                         ctlr_info_t *p;
4282
4283                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4284                         if (!p)
4285                                 goto Enomem;
4286                         hba[i] = p;
4287                         return i;
4288                 }
4289         }
4290         printk(KERN_WARNING "cciss: This driver supports a maximum"
4291                " of %d controllers.\n", MAX_CTLR);
4292         return -1;
4293 Enomem:
4294         printk(KERN_ERR "cciss: out of memory.\n");
4295         return -1;
4296 }
4297
4298 static void free_hba(int n)
4299 {
4300         ctlr_info_t *h = hba[n];
4301         int i;
4302
4303         hba[n] = NULL;
4304         for (i = 0; i < h->highest_lun + 1; i++)
4305                 if (h->gendisk[i] != NULL)
4306                         put_disk(h->gendisk[i]);
4307         kfree(h);
4308 }
4309
4310 /* Send a message CDB to the firmware. */
4311 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4312 {
4313         typedef struct {
4314                 CommandListHeader_struct CommandHeader;
4315                 RequestBlock_struct Request;
4316                 ErrDescriptor_struct ErrorDescriptor;
4317         } Command;
4318         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4319         Command *cmd;
4320         dma_addr_t paddr64;
4321         uint32_t paddr32, tag;
4322         void __iomem *vaddr;
4323         int i, err;
4324
4325         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4326         if (vaddr == NULL)
4327                 return -ENOMEM;
4328
4329         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4330            CCISS commands, so they must be allocated from the lower 4GiB of
4331            memory. */
4332         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4333         if (err) {
4334                 iounmap(vaddr);
4335                 return -ENOMEM;
4336         }
4337
4338         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4339         if (cmd == NULL) {
4340                 iounmap(vaddr);
4341                 return -ENOMEM;
4342         }
4343
4344         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4345            although there's no guarantee, we assume that the address is at
4346            least 4-byte aligned (most likely, it's page-aligned). */
4347         paddr32 = paddr64;
4348
4349         cmd->CommandHeader.ReplyQueue = 0;
4350         cmd->CommandHeader.SGList = 0;
4351         cmd->CommandHeader.SGTotal = 0;
4352         cmd->CommandHeader.Tag.lower = paddr32;
4353         cmd->CommandHeader.Tag.upper = 0;
4354         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4355
4356         cmd->Request.CDBLen = 16;
4357         cmd->Request.Type.Type = TYPE_MSG;
4358         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4359         cmd->Request.Type.Direction = XFER_NONE;
4360         cmd->Request.Timeout = 0; /* Don't time out */
4361         cmd->Request.CDB[0] = opcode;
4362         cmd->Request.CDB[1] = type;
4363         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4364
4365         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4366         cmd->ErrorDescriptor.Addr.upper = 0;
4367         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4368
4369         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4370
4371         for (i = 0; i < 10; i++) {
4372                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4373                 if ((tag & ~3) == paddr32)
4374                         break;
4375                 schedule_timeout_uninterruptible(HZ);
4376         }
4377
4378         iounmap(vaddr);
4379
4380         /* we leak the DMA buffer here ... no choice since the controller could
4381            still complete the command. */
4382         if (i == 10) {
4383                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4384                         opcode, type);
4385                 return -ETIMEDOUT;
4386         }
4387
4388         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4389
4390         if (tag & 2) {
4391                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4392                         opcode, type);
4393                 return -EIO;
4394         }
4395
4396         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4397                 opcode, type);
4398         return 0;
4399 }
4400
4401 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4402 #define cciss_noop(p) cciss_message(p, 3, 0)
4403
4404 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4405 {
4406 /* the #defines are stolen from drivers/pci/msi.h. */
4407 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4408 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4409
4410         int pos;
4411         u16 control = 0;
4412
4413         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4414         if (pos) {
4415                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4416                 if (control & PCI_MSI_FLAGS_ENABLE) {
4417                         printk(KERN_INFO "cciss: resetting MSI\n");
4418                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4419                 }
4420         }
4421
4422         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4423         if (pos) {
4424                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4425                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4426                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4427                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4428                 }
4429         }
4430
4431         return 0;
4432 }
4433
4434 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4435         void * __iomem vaddr, bool use_doorbell)
4436 {
4437         u16 pmcsr;
4438         int pos;
4439
4440         if (use_doorbell) {
4441                 /* For everything after the P600, the PCI power state method
4442                  * of resetting the controller doesn't work, so we have this
4443                  * other way using the doorbell register.
4444                  */
4445                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4446                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4447                 msleep(1000);
4448         } else { /* Try to do it the PCI power state way */
4449
4450                 /* Quoting from the Open CISS Specification: "The Power
4451                  * Management Control/Status Register (CSR) controls the power
4452                  * state of the device.  The normal operating state is D0,
4453                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4454                  * the controller, place the interface device in D3 then to D0,
4455                  * this causes a secondary PCI reset which will reset the
4456                  * controller." */
4457
4458                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4459                 if (pos == 0) {
4460                         dev_err(&pdev->dev,
4461                                 "cciss_controller_hard_reset: "
4462                                 "PCI PM not supported\n");
4463                         return -ENODEV;
4464                 }
4465                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4466                 /* enter the D3hot power management state */
4467                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4468                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4469                 pmcsr |= PCI_D3hot;
4470                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4471
4472                 msleep(500);
4473
4474                 /* enter the D0 power management state */
4475                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4476                 pmcsr |= PCI_D0;
4477                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4478
4479                 msleep(500);
4480         }
4481         return 0;
4482 }
4483
4484 /* This does a hard reset of the controller using PCI power management
4485  * states or using the doorbell register. */
4486 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4487 {
4488         u16 saved_config_space[32];
4489         u64 cfg_offset;
4490         u32 cfg_base_addr;
4491         u64 cfg_base_addr_index;
4492         void __iomem *vaddr;
4493         unsigned long paddr;
4494         u32 misc_fw_support, active_transport;
4495         int rc, i;
4496         CfgTable_struct __iomem *cfgtable;
4497         bool use_doorbell;
4498
4499         /* For controllers as old a the p600, this is very nearly
4500          * the same thing as
4501          *
4502          * pci_save_state(pci_dev);
4503          * pci_set_power_state(pci_dev, PCI_D3hot);
4504          * pci_set_power_state(pci_dev, PCI_D0);
4505          * pci_restore_state(pci_dev);
4506          *
4507          * but we can't use these nice canned kernel routines on
4508          * kexec, because they also check the MSI/MSI-X state in PCI
4509          * configuration space and do the wrong thing when it is
4510          * set/cleared.  Also, the pci_save/restore_state functions
4511          * violate the ordering requirements for restoring the
4512          * configuration space from the CCISS document (see the
4513          * comment below).  So we roll our own ....
4514          *
4515          * For controllers newer than the P600, the pci power state
4516          * method of resetting doesn't work so we have another way
4517          * using the doorbell register.
4518          */
4519
4520         for (i = 0; i < 32; i++)
4521                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4522
4523         /* find the first memory BAR, so we can find the cfg table */
4524         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4525         if (rc)
4526                 return rc;
4527         vaddr = remap_pci_mem(paddr, 0x250);
4528         if (!vaddr)
4529                 return -ENOMEM;
4530
4531         /* find cfgtable in order to check if reset via doorbell is supported */
4532         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4533                                         &cfg_base_addr_index, &cfg_offset);
4534         if (rc)
4535                 goto unmap_vaddr;
4536         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4537                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4538         if (!cfgtable) {
4539                 rc = -ENOMEM;
4540                 goto unmap_vaddr;
4541         }
4542
4543         /* If reset via doorbell register is supported, use that. */
4544         misc_fw_support = readl(&cfgtable->misc_fw_support);
4545         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4546
4547         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4548         if (rc)
4549                 goto unmap_cfgtable;
4550
4551         /* Restore the PCI configuration space.  The Open CISS
4552          * Specification says, "Restore the PCI Configuration
4553          * Registers, offsets 00h through 60h. It is important to
4554          * restore the command register, 16-bits at offset 04h,
4555          * last. Do not restore the configuration status register,
4556          * 16-bits at offset 06h."  Note that the offset is 2*i.
4557          */
4558         for (i = 0; i < 32; i++) {
4559                 if (i == 2 || i == 3)
4560                         continue;
4561                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4562         }
4563         wmb();
4564         pci_write_config_word(pdev, 4, saved_config_space[2]);
4565
4566         /* Some devices (notably the HP Smart Array 5i Controller)
4567            need a little pause here */
4568         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4569
4570         /* Controller should be in simple mode at this point.  If it's not,
4571          * It means we're on one of those controllers which doesn't support
4572          * the doorbell reset method and on which the PCI power management reset
4573          * method doesn't work (P800, for example.)
4574          * In those cases, don't try to proceed, as it generally doesn't work.
4575          */
4576         active_transport = readl(&cfgtable->TransportActive);
4577         if (active_transport & PERFORMANT_MODE) {
4578                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4579                         " Ignoring controller.\n");
4580                 rc = -ENODEV;
4581         }
4582
4583 unmap_cfgtable:
4584         iounmap(cfgtable);
4585
4586 unmap_vaddr:
4587         iounmap(vaddr);
4588         return rc;
4589 }
4590
4591 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4592 {
4593         int rc, i;
4594
4595         if (!reset_devices)
4596                 return 0;
4597
4598         /* Reset the controller with a PCI power-cycle or via doorbell */
4599         rc = cciss_kdump_hard_reset_controller(pdev);
4600
4601         /* -ENOTSUPP here means we cannot reset the controller
4602          * but it's already (and still) up and running in
4603          * "performant mode".
4604          */
4605         if (rc == -ENOTSUPP)
4606                 return 0; /* just try to do the kdump anyhow. */
4607         if (rc)
4608                 return -ENODEV;
4609         if (cciss_reset_msi(pdev))
4610                 return -ENODEV;
4611
4612         /* Now try to get the controller to respond to a no-op */
4613         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4614                 if (cciss_noop(pdev) == 0)
4615                         break;
4616                 else
4617                         dev_warn(&pdev->dev, "no-op failed%s\n",
4618                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4619                                         "; re-trying" : ""));
4620                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4621         }
4622         return 0;
4623 }
4624
4625 /*
4626  *  This is it.  Find all the controllers and register them.  I really hate
4627  *  stealing all these major device numbers.
4628  *  returns the number of block devices registered.
4629  */
4630 static int __devinit cciss_init_one(struct pci_dev *pdev,
4631                                     const struct pci_device_id *ent)
4632 {
4633         int i;
4634         int j = 0;
4635         int k = 0;
4636         int rc;
4637         int dac, return_code;
4638         InquiryData_struct *inq_buff;
4639
4640         rc = cciss_init_reset_devices(pdev);
4641         if (rc)
4642                 return rc;
4643         i = alloc_cciss_hba();
4644         if (i < 0)
4645                 return -1;
4646
4647         hba[i]->pdev = pdev;
4648         hba[i]->busy_initializing = 1;
4649         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4650         INIT_HLIST_HEAD(&hba[i]->reqQ);
4651         mutex_init(&hba[i]->busy_shutting_down);
4652
4653         if (cciss_pci_init(hba[i]) != 0)
4654                 goto clean_no_release_regions;
4655
4656         sprintf(hba[i]->devname, "cciss%d", i);
4657         hba[i]->ctlr = i;
4658
4659         init_completion(&hba[i]->scan_wait);
4660
4661         if (cciss_create_hba_sysfs_entry(hba[i]))
4662                 goto clean0;
4663
4664         /* configure PCI DMA stuff */
4665         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4666                 dac = 1;
4667         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4668                 dac = 0;
4669         else {
4670                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4671                 goto clean1;
4672         }
4673
4674         /*
4675          * register with the major number, or get a dynamic major number
4676          * by passing 0 as argument.  This is done for greater than
4677          * 8 controller support.
4678          */
4679         if (i < MAX_CTLR_ORIG)
4680                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4681         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4682         if (rc == -EBUSY || rc == -EINVAL) {
4683                 printk(KERN_ERR
4684                        "cciss:  Unable to get major number %d for %s "
4685                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4686                 goto clean1;
4687         } else {
4688                 if (i >= MAX_CTLR_ORIG)
4689                         hba[i]->major = rc;
4690         }
4691
4692         /* make sure the board interrupts are off */
4693         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4694         if (hba[i]->msi_vector || hba[i]->msix_vector) {
4695                 if (request_irq(hba[i]->intr[PERF_MODE_INT],
4696                                 do_cciss_msix_intr,
4697                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4698                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4699                                hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4700                         goto clean2;
4701                 }
4702         } else {
4703                 if (request_irq(hba[i]->intr[PERF_MODE_INT], do_cciss_intx,
4704                                 IRQF_DISABLED, hba[i]->devname, hba[i])) {
4705                         printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4706                                hba[i]->intr[PERF_MODE_INT], hba[i]->devname);
4707                         goto clean2;
4708                 }
4709         }
4710
4711         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4712                hba[i]->devname, pdev->device, pci_name(pdev),
4713                hba[i]->intr[PERF_MODE_INT], dac ? "" : " not");
4714
4715         hba[i]->cmd_pool_bits =
4716             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4717                         * sizeof(unsigned long), GFP_KERNEL);
4718         hba[i]->cmd_pool = (CommandList_struct *)
4719             pci_alloc_consistent(hba[i]->pdev,
4720                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4721                     &(hba[i]->cmd_pool_dhandle));
4722         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4723             pci_alloc_consistent(hba[i]->pdev,
4724                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4725                     &(hba[i]->errinfo_pool_dhandle));
4726         if ((hba[i]->cmd_pool_bits == NULL)
4727             || (hba[i]->cmd_pool == NULL)
4728             || (hba[i]->errinfo_pool == NULL)) {
4729                 printk(KERN_ERR "cciss: out of memory");
4730                 goto clean4;
4731         }
4732
4733         /* Need space for temp scatter list */
4734         hba[i]->scatter_list = kmalloc(hba[i]->max_commands *
4735                                                 sizeof(struct scatterlist *),
4736                                                 GFP_KERNEL);
4737         for (k = 0; k < hba[i]->nr_cmds; k++) {
4738                 hba[i]->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4739                                                         hba[i]->maxsgentries,
4740                                                         GFP_KERNEL);
4741                 if (hba[i]->scatter_list[k] == NULL) {
4742                         printk(KERN_ERR "cciss%d: could not allocate "
4743                                 "s/g lists\n", i);
4744                         goto clean4;
4745                 }
4746         }
4747         hba[i]->cmd_sg_list = cciss_allocate_sg_chain_blocks(hba[i],
4748                 hba[i]->chainsize, hba[i]->nr_cmds);
4749         if (!hba[i]->cmd_sg_list && hba[i]->chainsize > 0)
4750                 goto clean4;
4751
4752         spin_lock_init(&hba[i]->lock);
4753
4754         /* Initialize the pdev driver private data.
4755            have it point to hba[i].  */
4756         pci_set_drvdata(pdev, hba[i]);
4757         /* command and error info recs zeroed out before
4758            they are used */
4759         memset(hba[i]->cmd_pool_bits, 0,
4760                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4761                         * sizeof(unsigned long));
4762
4763         hba[i]->num_luns = 0;
4764         hba[i]->highest_lun = -1;
4765         for (j = 0; j < CISS_MAX_LUN; j++) {
4766                 hba[i]->drv[j] = NULL;
4767                 hba[i]->gendisk[j] = NULL;
4768         }
4769
4770         cciss_scsi_setup(i);
4771
4772         /* Turn the interrupts on so we can service requests */
4773         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4774
4775         /* Get the firmware version */
4776         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4777         if (inq_buff == NULL) {
4778                 printk(KERN_ERR "cciss: out of memory\n");
4779                 goto clean4;
4780         }
4781
4782         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4783                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4784         if (return_code == IO_OK) {
4785                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4786                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4787                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4788                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4789         } else {         /* send command failed */
4790                 printk(KERN_WARNING "cciss: unable to determine firmware"
4791                         " version of controller\n");
4792         }
4793         kfree(inq_buff);
4794
4795         cciss_procinit(i);
4796
4797         hba[i]->cciss_max_sectors = 8192;
4798
4799         rebuild_lun_table(hba[i], 1, 0);
4800         hba[i]->busy_initializing = 0;
4801         return 1;
4802
4803 clean4:
4804         kfree(hba[i]->cmd_pool_bits);
4805         /* Free up sg elements */
4806         for (k = 0; k < hba[i]->nr_cmds; k++)
4807                 kfree(hba[i]->scatter_list[k]);
4808         kfree(hba[i]->scatter_list);
4809         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4810         if (hba[i]->cmd_pool)
4811                 pci_free_consistent(hba[i]->pdev,
4812                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4813                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4814         if (hba[i]->errinfo_pool)
4815                 pci_free_consistent(hba[i]->pdev,
4816                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4817                                     hba[i]->errinfo_pool,
4818                                     hba[i]->errinfo_pool_dhandle);
4819         free_irq(hba[i]->intr[PERF_MODE_INT], hba[i]);
4820 clean2:
4821         unregister_blkdev(hba[i]->major, hba[i]->devname);
4822 clean1:
4823         cciss_destroy_hba_sysfs_entry(hba[i]);
4824 clean0:
4825         pci_release_regions(pdev);
4826 clean_no_release_regions:
4827         hba[i]->busy_initializing = 0;
4828
4829         /*
4830          * Deliberately omit pci_disable_device(): it does something nasty to
4831          * Smart Array controllers that pci_enable_device does not undo
4832          */
4833         pci_set_drvdata(pdev, NULL);
4834         free_hba(i);
4835         return -1;
4836 }
4837
4838 static void cciss_shutdown(struct pci_dev *pdev)
4839 {
4840         ctlr_info_t *h;
4841         char *flush_buf;
4842         int return_code;
4843
4844         h = pci_get_drvdata(pdev);
4845         flush_buf = kzalloc(4, GFP_KERNEL);
4846         if (!flush_buf) {
4847                 printk(KERN_WARNING
4848                         "cciss:%d cache not flushed, out of memory.\n",
4849                         h->ctlr);
4850                 return;
4851         }
4852         /* write all data in the battery backed cache to disk */
4853         memset(flush_buf, 0, 4);
4854         return_code = sendcmd_withirq(CCISS_CACHE_FLUSH, h->ctlr, flush_buf,
4855                 4, 0, CTLR_LUNID, TYPE_CMD);
4856         kfree(flush_buf);
4857         if (return_code != IO_OK)
4858                 printk(KERN_WARNING "cciss%d: Error flushing cache\n",
4859                         h->ctlr);
4860         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4861         free_irq(h->intr[PERF_MODE_INT], h);
4862 }
4863
4864 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4865 {
4866         ctlr_info_t *tmp_ptr;
4867         int i, j;
4868
4869         if (pci_get_drvdata(pdev) == NULL) {
4870                 printk(KERN_ERR "cciss: Unable to remove device \n");
4871                 return;
4872         }
4873
4874         tmp_ptr = pci_get_drvdata(pdev);
4875         i = tmp_ptr->ctlr;
4876         if (hba[i] == NULL) {
4877                 printk(KERN_ERR "cciss: device appears to "
4878                        "already be removed \n");
4879                 return;
4880         }
4881
4882         mutex_lock(&hba[i]->busy_shutting_down);
4883
4884         remove_from_scan_list(hba[i]);
4885         remove_proc_entry(hba[i]->devname, proc_cciss);
4886         unregister_blkdev(hba[i]->major, hba[i]->devname);
4887
4888         /* remove it from the disk list */
4889         for (j = 0; j < CISS_MAX_LUN; j++) {
4890                 struct gendisk *disk = hba[i]->gendisk[j];
4891                 if (disk) {
4892                         struct request_queue *q = disk->queue;
4893
4894                         if (disk->flags & GENHD_FL_UP) {
4895                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4896                                 del_gendisk(disk);
4897                         }
4898                         if (q)
4899                                 blk_cleanup_queue(q);
4900                 }
4901         }
4902
4903 #ifdef CONFIG_CISS_SCSI_TAPE
4904         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4905 #endif
4906
4907         cciss_shutdown(pdev);
4908
4909 #ifdef CONFIG_PCI_MSI
4910         if (hba[i]->msix_vector)
4911                 pci_disable_msix(hba[i]->pdev);
4912         else if (hba[i]->msi_vector)
4913                 pci_disable_msi(hba[i]->pdev);
4914 #endif                          /* CONFIG_PCI_MSI */
4915
4916         iounmap(hba[i]->transtable);
4917         iounmap(hba[i]->cfgtable);
4918         iounmap(hba[i]->vaddr);
4919
4920         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4921                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4922         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4923                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4924         kfree(hba[i]->cmd_pool_bits);
4925         /* Free up sg elements */
4926         for (j = 0; j < hba[i]->nr_cmds; j++)
4927                 kfree(hba[i]->scatter_list[j]);
4928         kfree(hba[i]->scatter_list);
4929         cciss_free_sg_chain_blocks(hba[i]->cmd_sg_list, hba[i]->nr_cmds);
4930         /*
4931          * Deliberately omit pci_disable_device(): it does something nasty to
4932          * Smart Array controllers that pci_enable_device does not undo
4933          */
4934         pci_release_regions(pdev);
4935         pci_set_drvdata(pdev, NULL);
4936         cciss_destroy_hba_sysfs_entry(hba[i]);
4937         mutex_unlock(&hba[i]->busy_shutting_down);
4938         free_hba(i);
4939 }
4940
4941 static struct pci_driver cciss_pci_driver = {
4942         .name = "cciss",
4943         .probe = cciss_init_one,
4944         .remove = __devexit_p(cciss_remove_one),
4945         .id_table = cciss_pci_device_id,        /* id_table */
4946         .shutdown = cciss_shutdown,
4947 };
4948
4949 /*
4950  *  This is it.  Register the PCI driver information for the cards we control
4951  *  the OS will call our registered routines when it finds one of our cards.
4952  */
4953 static int __init cciss_init(void)
4954 {
4955         int err;
4956
4957         /*
4958          * The hardware requires that commands are aligned on a 64-bit
4959          * boundary. Given that we use pci_alloc_consistent() to allocate an
4960          * array of them, the size must be a multiple of 8 bytes.
4961          */
4962         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4963         printk(KERN_INFO DRIVER_NAME "\n");
4964
4965         err = bus_register(&cciss_bus_type);
4966         if (err)
4967                 return err;
4968
4969         /* Start the scan thread */
4970         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4971         if (IS_ERR(cciss_scan_thread)) {
4972                 err = PTR_ERR(cciss_scan_thread);
4973                 goto err_bus_unregister;
4974         }
4975
4976         /* Register for our PCI devices */
4977         err = pci_register_driver(&cciss_pci_driver);
4978         if (err)
4979                 goto err_thread_stop;
4980
4981         return err;
4982
4983 err_thread_stop:
4984         kthread_stop(cciss_scan_thread);
4985 err_bus_unregister:
4986         bus_unregister(&cciss_bus_type);
4987
4988         return err;
4989 }
4990
4991 static void __exit cciss_cleanup(void)
4992 {
4993         int i;
4994
4995         pci_unregister_driver(&cciss_pci_driver);
4996         /* double check that all controller entrys have been removed */
4997         for (i = 0; i < MAX_CTLR; i++) {
4998                 if (hba[i] != NULL) {
4999                         printk(KERN_WARNING "cciss: had to remove"
5000                                " controller %d\n", i);
5001                         cciss_remove_one(hba[i]->pdev);
5002                 }
5003         }
5004         kthread_stop(cciss_scan_thread);
5005         remove_proc_entry("driver/cciss", NULL);
5006         bus_unregister(&cciss_bus_type);
5007 }
5008
5009 module_init(cciss_init);
5010 module_exit(cciss_cleanup);