drivers/block/cciss.c

   1 /*
   2  *    Disk Array driver for HP Smart Array controllers.
   3  *    (C) Copyright 2000, 2007 Hewlett-Packard Development Company, L.P.
   4  *
   5  *    This program is free software; you can redistribute it and/or modify
   6  *    it under the terms of the GNU General Public License as published by
   7  *    the Free Software Foundation; version 2 of the License.
   8  *
   9  *    This program is distributed in the hope that it will be useful,
  10  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  *    MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12  *    General Public License for more details.
  13  *
  14  *    You should have received a copy of the GNU General Public License
  15  *    along with this program; if not, write to the Free Software
  16  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
  17  *    02111-1307, USA.
  18  *
  19  *    Questions/Comments/Bugfixes to iss_storagedev@hp.com
  20  *
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/types.h>
  26 #include <linux/pci.h>
  27 #include <linux/kernel.h>
  28 #include <linux/slab.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/delay.h>
  31 #include <linux/major.h>
  32 #include <linux/fs.h>
  33 #include <linux/bio.h>
  34 #include <linux/blkpg.h>
  35 #include <linux/timer.h>
  36 #include <linux/proc_fs.h>
  37 #include <linux/seq_file.h>
  38 #include <linux/init.h>
  39 #include <linux/jiffies.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/compat.h>
  43 #include <linux/mutex.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/io.h>
  46
  47 #include <linux/dma-mapping.h>
  48 #include <linux/blkdev.h>
  49 #include <linux/genhd.h>
  50 #include <linux/completion.h>
  51 #include <scsi/scsi.h>
  52 #include <scsi/sg.h>
  53 #include <scsi/scsi_ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/scatterlist.h>
  56 #include <linux/kthread.h>
  57
  58 #define CCISS_DRIVER_VERSION(maj,min,submin) ((maj<<16)|(min<<8)|(submin))
  59 #define DRIVER_NAME "HP CISS Driver (v 3.6.26)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 26)
  61
  62 /* Embedded module documentation macros - see modules.h */
  63 MODULE_AUTHOR("Hewlett-Packard Company");
  64 MODULE_DESCRIPTION("Driver for HP Smart Array Controllers");
  65 MODULE_SUPPORTED_DEVICE("HP Smart Array Controllers");
  66 MODULE_VERSION("3.6.26");
  67 MODULE_LICENSE("GPL");
  68
  69 static int cciss_allow_hpsa;
  70 module_param(cciss_allow_hpsa, int, S_IRUGO|S_IWUSR);
  71 MODULE_PARM_DESC(cciss_allow_hpsa,
  72         "Prevent cciss driver from accessing hardware known to be "
  73         " supported by the hpsa driver");
  74
  75 #include "cciss_cmd.h"
  76 #include "cciss.h"
  77 #include <linux/cciss_ioctl.h>
  78
  79 /* define the PCI info for the cards we can control */
  80 static const struct pci_device_id cciss_pci_device_id[] = {
  81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  86         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  87         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  88         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  89         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
 104         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
 105         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
 106         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 107         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 108         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3250},
 109         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3251},
 110         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3252},
 111         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3253},
 112         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3254},
 113         {0,}
 114 };
 115
 116 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 117
 118 /*  board_id = Subsystem Device ID & Vendor ID
 119  *  product = Marketing Name for the board
 120  *  access = Address of the struct of function pointers
 121  */
 122 static struct board_type products[] = {
 123         {0x40700E11, "Smart Array 5300", &SA5_access},
 124         {0x40800E11, "Smart Array 5i", &SA5B_access},
 125         {0x40820E11, "Smart Array 532", &SA5B_access},
 126         {0x40830E11, "Smart Array 5312", &SA5B_access},
 127         {0x409A0E11, "Smart Array 641", &SA5_access},
 128         {0x409B0E11, "Smart Array 642", &SA5_access},
 129         {0x409C0E11, "Smart Array 6400", &SA5_access},
 130         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 131         {0x40910E11, "Smart Array 6i", &SA5_access},
 132         {0x3225103C, "Smart Array P600", &SA5_access},
 133         {0x3235103C, "Smart Array P400i", &SA5_access},
 134         {0x3211103C, "Smart Array E200i", &SA5_access},
 135         {0x3212103C, "Smart Array E200", &SA5_access},
 136         {0x3213103C, "Smart Array E200i", &SA5_access},
 137         {0x3214103C, "Smart Array E200i", &SA5_access},
 138         {0x3215103C, "Smart Array E200i", &SA5_access},
 139         {0x3237103C, "Smart Array E500", &SA5_access},
 140 /* controllers below this line are also supported by the hpsa driver. */
 141 #define HPSA_BOUNDARY 0x3223103C
 142         {0x3223103C, "Smart Array P800", &SA5_access},
 143         {0x3234103C, "Smart Array P400", &SA5_access},
 144         {0x323D103C, "Smart Array P700m", &SA5_access},
 145         {0x3241103C, "Smart Array P212", &SA5_access},
 146         {0x3243103C, "Smart Array P410", &SA5_access},
 147         {0x3245103C, "Smart Array P410i", &SA5_access},
 148         {0x3247103C, "Smart Array P411", &SA5_access},
 149         {0x3249103C, "Smart Array P812", &SA5_access},
 150         {0x324A103C, "Smart Array P712m", &SA5_access},
 151         {0x324B103C, "Smart Array P711m", &SA5_access},
 152         {0x3250103C, "Smart Array", &SA5_access},
 153         {0x3251103C, "Smart Array", &SA5_access},
 154         {0x3252103C, "Smart Array", &SA5_access},
 155         {0x3253103C, "Smart Array", &SA5_access},
 156         {0x3254103C, "Smart Array", &SA5_access},
 157 };
 158
 159 /* How long to wait (in milliseconds) for board to go into simple mode */
 160 #define MAX_CONFIG_WAIT 30000
 161 #define MAX_IOCTL_CONFIG_WAIT 1000
 162
 163 /*define how many times we will try a command because of bus resets */
 164 #define MAX_CMD_RETRIES 3
 165
 166 #define MAX_CTLR        32
 167
 168 /* Originally cciss driver only supports 8 major numbers */
 169 #define MAX_CTLR_ORIG   8
 170
 171 static ctlr_info_t *hba[MAX_CTLR];
 172
 173 static struct task_struct *cciss_scan_thread;
 174 static DEFINE_MUTEX(scan_mutex);
 175 static LIST_HEAD(scan_q);
 176
 177 static void do_cciss_request(struct request_queue *q);
 178 static irqreturn_t do_cciss_intx(int irq, void *dev_id);
 179 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id);
 180 static int cciss_open(struct block_device *bdev, fmode_t mode);
 181 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode);
 182 static int cciss_release(struct gendisk *disk, fmode_t mode);
 183 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 184                     unsigned int cmd, unsigned long arg);
 185 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 186                        unsigned int cmd, unsigned long arg);
 187 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 188
 189 static int cciss_revalidate(struct gendisk *disk);
 190 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 191 static int deregister_disk(ctlr_info_t *h, int drv_index,
 192                            int clear_all, int via_ioctl);
 193
 194 static void cciss_read_capacity(ctlr_info_t *h, int logvol,
 195                         sector_t *total_size, unsigned int *block_size);
 196 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
 197                         sector_t *total_size, unsigned int *block_size);
 198 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
 199                         sector_t total_size,
 200                         unsigned int block_size, InquiryData_struct *inq_buff,
 201                                    drive_info_struct *drv);
 202 static void __devinit cciss_interrupt_mode(ctlr_info_t *);
 203 static void start_io(ctlr_info_t *h);
 204 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
 205                         __u8 page_code, unsigned char scsi3addr[],
 206                         int cmd_type);
 207 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 208         int attempt_retry);
 209 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 210
 211 static int add_to_scan_list(struct ctlr_info *h);
 212 static int scan_thread(void *data);
 213 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 214 static void cciss_hba_release(struct device *dev);
 215 static void cciss_device_release(struct device *dev);
 216 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 217 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index);
 218 static inline u32 next_command(ctlr_info_t *h);
 219 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
 220         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
 221         u64 *cfg_offset);
 222 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
 223         unsigned long *memory_bar);
 224
 225
 226 /* performant mode helper functions */
 227 static void  calc_bucket_map(int *bucket, int num_buckets, int nsgs,
 228                                 int *bucket_map);
 229 static void cciss_put_controller_into_performant_mode(ctlr_info_t *h);
 230
 231 #ifdef CONFIG_PROC_FS
 232 static void cciss_procinit(ctlr_info_t *h);
 233 #else
 234 static void cciss_procinit(ctlr_info_t *h)
 235 {
 236 }
 237 #endif                          /* CONFIG_PROC_FS */
 238
 239 #ifdef CONFIG_COMPAT
 240 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 241                               unsigned, unsigned long);
 242 #endif
 243
 244 static const struct block_device_operations cciss_fops = {
 245         .owner = THIS_MODULE,
 246         .open = cciss_unlocked_open,
 247         .release = cciss_release,
 248         .ioctl = do_ioctl,
 249         .getgeo = cciss_getgeo,
 250 #ifdef CONFIG_COMPAT
 251         .compat_ioctl = cciss_compat_ioctl,
 252 #endif
 253         .revalidate_disk = cciss_revalidate,
 254 };
 255
 256 /* set_performant_mode: Modify the tag for cciss performant
 257  * set bit 0 for pull model, bits 3-1 for block fetch
 258  * register number
 259  */
 260 static void set_performant_mode(ctlr_info_t *h, CommandList_struct *c)
 261 {
 262         if (likely(h->transMethod == CFGTBL_Trans_Performant))
 263                 c->busaddr |= 1 | (h->blockFetchTable[c->Header.SGList] << 1);
 264 }
 265
 266 /*
 267  * Enqueuing and dequeuing functions for cmdlists.
 268  */
 269 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 270 {
 271         hlist_add_head(&c->list, list);
 272 }
 273
 274 static inline void removeQ(CommandList_struct *c)
 275 {
 276         /*
 277          * After kexec/dump some commands might still
 278          * be in flight, which the firmware will try
 279          * to complete. Resetting the firmware doesn't work
 280          * with old fw revisions, so we have to mark
 281          * them off as 'stale' to prevent the driver from
 282          * falling over.
 283          */
 284         if (WARN_ON(hlist_unhashed(&c->list))) {
 285                 c->cmd_type = CMD_MSG_STALE;
 286                 return;
 287         }
 288
 289         hlist_del_init(&c->list);
 290 }
 291
 292 static void enqueue_cmd_and_start_io(ctlr_info_t *h,
 293         CommandList_struct *c)
 294 {
 295         unsigned long flags;
 296         set_performant_mode(h, c);
 297         spin_lock_irqsave(&h->lock, flags);
 298         addQ(&h->reqQ, c);
 299         h->Qdepth++;
 300         start_io(h);
 301         spin_unlock_irqrestore(&h->lock, flags);
 302 }
 303
 304 static void cciss_free_sg_chain_blocks(SGDescriptor_struct **cmd_sg_list,
 305         int nr_cmds)
 306 {
 307         int i;
 308
 309         if (!cmd_sg_list)
 310                 return;
 311         for (i = 0; i < nr_cmds; i++) {
 312                 kfree(cmd_sg_list[i]);
 313                 cmd_sg_list[i] = NULL;
 314         }
 315         kfree(cmd_sg_list);
 316 }
 317
 318 static SGDescriptor_struct **cciss_allocate_sg_chain_blocks(
 319         ctlr_info_t *h, int chainsize, int nr_cmds)
 320 {
 321         int j;
 322         SGDescriptor_struct **cmd_sg_list;
 323
 324         if (chainsize <= 0)
 325                 return NULL;
 326
 327         cmd_sg_list = kmalloc(sizeof(*cmd_sg_list) * nr_cmds, GFP_KERNEL);
 328         if (!cmd_sg_list)
 329                 return NULL;
 330
 331         /* Build up chain blocks for each command */
 332         for (j = 0; j < nr_cmds; j++) {
 333                 /* Need a block of chainsized s/g elements. */
 334                 cmd_sg_list[j] = kmalloc((chainsize *
 335                         sizeof(*cmd_sg_list[j])), GFP_KERNEL);
 336                 if (!cmd_sg_list[j]) {
 337                         dev_err(&h->pdev->dev, "Cannot get memory "
 338                                 "for s/g chains.\n");
 339                         goto clean;
 340                 }
 341         }
 342         return cmd_sg_list;
 343 clean:
 344         cciss_free_sg_chain_blocks(cmd_sg_list, nr_cmds);
 345         return NULL;
 346 }
 347
 348 static void cciss_unmap_sg_chain_block(ctlr_info_t *h, CommandList_struct *c)
 349 {
 350         SGDescriptor_struct *chain_sg;
 351         u64bit temp64;
 352
 353         if (c->Header.SGTotal <= h->max_cmd_sgentries)
 354                 return;
 355
 356         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 357         temp64.val32.lower = chain_sg->Addr.lower;
 358         temp64.val32.upper = chain_sg->Addr.upper;
 359         pci_unmap_single(h->pdev, temp64.val, chain_sg->Len, PCI_DMA_TODEVICE);
 360 }
 361
 362 static void cciss_map_sg_chain_block(ctlr_info_t *h, CommandList_struct *c,
 363         SGDescriptor_struct *chain_block, int len)
 364 {
 365         SGDescriptor_struct *chain_sg;
 366         u64bit temp64;
 367
 368         chain_sg = &c->SG[h->max_cmd_sgentries - 1];
 369         chain_sg->Ext = CCISS_SG_CHAIN;
 370         chain_sg->Len = len;
 371         temp64.val = pci_map_single(h->pdev, chain_block, len,
 372                                 PCI_DMA_TODEVICE);
 373         chain_sg->Addr.lower = temp64.val32.lower;
 374         chain_sg->Addr.upper = temp64.val32.upper;
 375 }
 376
 377 #include "cciss_scsi.c"         /* For SCSI tape support */
 378
 379 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 380         "UNKNOWN"
 381 };
 382 #define RAID_UNKNOWN (ARRAY_SIZE(raid_label)-1)
 383
 384 #ifdef CONFIG_PROC_FS
 385
 386 /*
 387  * Report information about this controller.
 388  */
 389 #define ENG_GIG 1000000000
 390 #define ENG_GIG_FACTOR (ENG_GIG/512)
 391 #define ENGAGE_SCSI     "engage scsi"
 392
 393 static struct proc_dir_entry *proc_cciss;
 394
 395 static void cciss_seq_show_header(struct seq_file *seq)
 396 {
 397         ctlr_info_t *h = seq->private;
 398
 399         seq_printf(seq, "%s: HP %s Controller\n"
 400                 "Board ID: 0x%08lx\n"
 401                 "Firmware Version: %c%c%c%c\n"
 402                 "IRQ: %d\n"
 403                 "Logical drives: %d\n"
 404                 "Current Q depth: %d\n"
 405                 "Current # commands on controller: %d\n"
 406                 "Max Q depth since init: %d\n"
 407                 "Max # commands on controller since init: %d\n"
 408                 "Max SG entries since init: %d\n",
 409                 h->devname,
 410                 h->product_name,
 411                 (unsigned long)h->board_id,
 412                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 413                 h->firm_ver[3], (unsigned int)h->intr[PERF_MODE_INT],
 414                 h->num_luns,
 415                 h->Qdepth, h->commands_outstanding,
 416                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 417
 418 #ifdef CONFIG_CISS_SCSI_TAPE
 419         cciss_seq_tape_report(seq, h);
 420 #endif /* CONFIG_CISS_SCSI_TAPE */
 421 }
 422
 423 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 424 {
 425         ctlr_info_t *h = seq->private;
 426         unsigned long flags;
 427
 428         /* prevent displaying bogus info during configuration
 429          * or deconfiguration of a logical volume
 430          */
 431         spin_lock_irqsave(&h->lock, flags);
 432         if (h->busy_configuring) {
 433                 spin_unlock_irqrestore(&h->lock, flags);
 434                 return ERR_PTR(-EBUSY);
 435         }
 436         h->busy_configuring = 1;
 437         spin_unlock_irqrestore(&h->lock, flags);
 438
 439         if (*pos == 0)
 440                 cciss_seq_show_header(seq);
 441
 442         return pos;
 443 }
 444
 445 static int cciss_seq_show(struct seq_file *seq, void *v)
 446 {
 447         sector_t vol_sz, vol_sz_frac;
 448         ctlr_info_t *h = seq->private;
 449         unsigned ctlr = h->ctlr;
 450         loff_t *pos = v;
 451         drive_info_struct *drv = h->drv[*pos];
 452
 453         if (*pos > h->highest_lun)
 454                 return 0;
 455
 456         if (drv == NULL) /* it's possible for h->drv[] to have holes. */
 457                 return 0;
 458
 459         if (drv->heads == 0)
 460                 return 0;
 461
 462         vol_sz = drv->nr_blocks;
 463         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 464         vol_sz_frac *= 100;
 465         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 466
 467         if (drv->raid_level < 0 || drv->raid_level > RAID_UNKNOWN)
 468                 drv->raid_level = RAID_UNKNOWN;
 469         seq_printf(seq, "cciss/c%dd%d:"
 470                         "\t%4u.%02uGB\tRAID %s\n",
 471                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 472                         raid_label[drv->raid_level]);
 473         return 0;
 474 }
 475
 476 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 477 {
 478         ctlr_info_t *h = seq->private;
 479
 480         if (*pos > h->highest_lun)
 481                 return NULL;
 482         *pos += 1;
 483
 484         return pos;
 485 }
 486
 487 static void cciss_seq_stop(struct seq_file *seq, void *v)
 488 {
 489         ctlr_info_t *h = seq->private;
 490
 491         /* Only reset h->busy_configuring if we succeeded in setting
 492          * it during cciss_seq_start. */
 493         if (v == ERR_PTR(-EBUSY))
 494                 return;
 495
 496         h->busy_configuring = 0;
 497 }
 498
 499 static const struct seq_operations cciss_seq_ops = {
 500         .start = cciss_seq_start,
 501         .show  = cciss_seq_show,
 502         .next  = cciss_seq_next,
 503         .stop  = cciss_seq_stop,
 504 };
 505
 506 static int cciss_seq_open(struct inode *inode, struct file *file)
 507 {
 508         int ret = seq_open(file, &cciss_seq_ops);
 509         struct seq_file *seq = file->private_data;
 510
 511         if (!ret)
 512                 seq->private = PDE(inode)->data;
 513
 514         return ret;
 515 }
 516
 517 static ssize_t
 518 cciss_proc_write(struct file *file, const char __user *buf,
 519                  size_t length, loff_t *ppos)
 520 {
 521         int err;
 522         char *buffer;
 523
 524 #ifndef CONFIG_CISS_SCSI_TAPE
 525         return -EINVAL;
 526 #endif
 527
 528         if (!buf || length > PAGE_SIZE - 1)
 529                 return -EINVAL;
 530
 531         buffer = (char *)__get_free_page(GFP_KERNEL);
 532         if (!buffer)
 533                 return -ENOMEM;
 534
 535         err = -EFAULT;
 536         if (copy_from_user(buffer, buf, length))
 537                 goto out;
 538         buffer[length] = '\0';
 539
 540 #ifdef CONFIG_CISS_SCSI_TAPE
 541         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 542                 struct seq_file *seq = file->private_data;
 543                 ctlr_info_t *h = seq->private;
 544
 545                 err = cciss_engage_scsi(h);
 546                 if (err == 0)
 547                         err = length;
 548         } else
 549 #endif /* CONFIG_CISS_SCSI_TAPE */
 550                 err = -EINVAL;
 551         /* might be nice to have "disengage" too, but it's not
 552            safely possible. (only 1 module use count, lock issues.) */
 553
 554 out:
 555         free_page((unsigned long)buffer);
 556         return err;
 557 }
 558
 559 static const struct file_operations cciss_proc_fops = {
 560         .owner   = THIS_MODULE,
 561         .open    = cciss_seq_open,
 562         .read    = seq_read,
 563         .llseek  = seq_lseek,
 564         .release = seq_release,
 565         .write   = cciss_proc_write,
 566 };
 567
 568 static void __devinit cciss_procinit(ctlr_info_t *h)
 569 {
 570         struct proc_dir_entry *pde;
 571
 572         if (proc_cciss == NULL)
 573                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 574         if (!proc_cciss)
 575                 return;
 576         pde = proc_create_data(h->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 577                                         S_IROTH, proc_cciss,
 578                                         &cciss_proc_fops, h);
 579 }
 580 #endif                          /* CONFIG_PROC_FS */
 581
 582 #define MAX_PRODUCT_NAME_LEN 19
 583
 584 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 585 #define to_drv(n) container_of(n, drive_info_struct, dev)
 586
 587 static ssize_t host_store_rescan(struct device *dev,
 588                                  struct device_attribute *attr,
 589                                  const char *buf, size_t count)
 590 {
 591         struct ctlr_info *h = to_hba(dev);
 592
 593         add_to_scan_list(h);
 594         wake_up_process(cciss_scan_thread);
 595         wait_for_completion_interruptible(&h->scan_wait);
 596
 597         return count;
 598 }
 599 static DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 600
 601 static ssize_t dev_show_unique_id(struct device *dev,
 602                                  struct device_attribute *attr,
 603                                  char *buf)
 604 {
 605         drive_info_struct *drv = to_drv(dev);
 606         struct ctlr_info *h = to_hba(drv->dev.parent);
 607         __u8 sn[16];
 608         unsigned long flags;
 609         int ret = 0;
 610
 611         spin_lock_irqsave(&h->lock, flags);
 612         if (h->busy_configuring)
 613                 ret = -EBUSY;
 614         else
 615                 memcpy(sn, drv->serial_no, sizeof(sn));
 616         spin_unlock_irqrestore(&h->lock, flags);
 617
 618         if (ret)
 619                 return ret;
 620         else
 621                 return snprintf(buf, 16 * 2 + 2,
 622                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 623                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 624                                 sn[0], sn[1], sn[2], sn[3],
 625                                 sn[4], sn[5], sn[6], sn[7],
 626                                 sn[8], sn[9], sn[10], sn[11],
 627                                 sn[12], sn[13], sn[14], sn[15]);
 628 }
 629 static DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 630
 631 static ssize_t dev_show_vendor(struct device *dev,
 632                                struct device_attribute *attr,
 633                                char *buf)
 634 {
 635         drive_info_struct *drv = to_drv(dev);
 636         struct ctlr_info *h = to_hba(drv->dev.parent);
 637         char vendor[VENDOR_LEN + 1];
 638         unsigned long flags;
 639         int ret = 0;
 640
 641         spin_lock_irqsave(&h->lock, flags);
 642         if (h->busy_configuring)
 643                 ret = -EBUSY;
 644         else
 645                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 646         spin_unlock_irqrestore(&h->lock, flags);
 647
 648         if (ret)
 649                 return ret;
 650         else
 651                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 652 }
 653 static DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 654
 655 static ssize_t dev_show_model(struct device *dev,
 656                               struct device_attribute *attr,
 657                               char *buf)
 658 {
 659         drive_info_struct *drv = to_drv(dev);
 660         struct ctlr_info *h = to_hba(drv->dev.parent);
 661         char model[MODEL_LEN + 1];
 662         unsigned long flags;
 663         int ret = 0;
 664
 665         spin_lock_irqsave(&h->lock, flags);
 666         if (h->busy_configuring)
 667                 ret = -EBUSY;
 668         else
 669                 memcpy(model, drv->model, MODEL_LEN + 1);
 670         spin_unlock_irqrestore(&h->lock, flags);
 671
 672         if (ret)
 673                 return ret;
 674         else
 675                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 676 }
 677 static DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 678
 679 static ssize_t dev_show_rev(struct device *dev,
 680                             struct device_attribute *attr,
 681                             char *buf)
 682 {
 683         drive_info_struct *drv = to_drv(dev);
 684         struct ctlr_info *h = to_hba(drv->dev.parent);
 685         char rev[REV_LEN + 1];
 686         unsigned long flags;
 687         int ret = 0;
 688
 689         spin_lock_irqsave(&h->lock, flags);
 690         if (h->busy_configuring)
 691                 ret = -EBUSY;
 692         else
 693                 memcpy(rev, drv->rev, REV_LEN + 1);
 694         spin_unlock_irqrestore(&h->lock, flags);
 695
 696         if (ret)
 697                 return ret;
 698         else
 699                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 700 }
 701 static DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 702
 703 static ssize_t cciss_show_lunid(struct device *dev,
 704                                 struct device_attribute *attr, char *buf)
 705 {
 706         drive_info_struct *drv = to_drv(dev);
 707         struct ctlr_info *h = to_hba(drv->dev.parent);
 708         unsigned long flags;
 709         unsigned char lunid[8];
 710
 711         spin_lock_irqsave(&h->lock, flags);
 712         if (h->busy_configuring) {
 713                 spin_unlock_irqrestore(&h->lock, flags);
 714                 return -EBUSY;
 715         }
 716         if (!drv->heads) {
 717                 spin_unlock_irqrestore(&h->lock, flags);
 718                 return -ENOTTY;
 719         }
 720         memcpy(lunid, drv->LunID, sizeof(lunid));
 721         spin_unlock_irqrestore(&h->lock, flags);
 722         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 723                 lunid[0], lunid[1], lunid[2], lunid[3],
 724                 lunid[4], lunid[5], lunid[6], lunid[7]);
 725 }
 726 static DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 727
 728 static ssize_t cciss_show_raid_level(struct device *dev,
 729                                      struct device_attribute *attr, char *buf)
 730 {
 731         drive_info_struct *drv = to_drv(dev);
 732         struct ctlr_info *h = to_hba(drv->dev.parent);
 733         int raid;
 734         unsigned long flags;
 735
 736         spin_lock_irqsave(&h->lock, flags);
 737         if (h->busy_configuring) {
 738                 spin_unlock_irqrestore(&h->lock, flags);
 739                 return -EBUSY;
 740         }
 741         raid = drv->raid_level;
 742         spin_unlock_irqrestore(&h->lock, flags);
 743         if (raid < 0 || raid > RAID_UNKNOWN)
 744                 raid = RAID_UNKNOWN;
 745
 746         return snprintf(buf, strlen(raid_label[raid]) + 7, "RAID %s\n",
 747                         raid_label[raid]);
 748 }
 749 static DEVICE_ATTR(raid_level, S_IRUGO, cciss_show_raid_level, NULL);
 750
 751 static ssize_t cciss_show_usage_count(struct device *dev,
 752                                       struct device_attribute *attr, char *buf)
 753 {
 754         drive_info_struct *drv = to_drv(dev);
 755         struct ctlr_info *h = to_hba(drv->dev.parent);
 756         unsigned long flags;
 757         int count;
 758
 759         spin_lock_irqsave(&h->lock, flags);
 760         if (h->busy_configuring) {
 761                 spin_unlock_irqrestore(&h->lock, flags);
 762                 return -EBUSY;
 763         }
 764         count = drv->usage_count;
 765         spin_unlock_irqrestore(&h->lock, flags);
 766         return snprintf(buf, 20, "%d\n", count);
 767 }
 768 static DEVICE_ATTR(usage_count, S_IRUGO, cciss_show_usage_count, NULL);
 769
 770 static struct attribute *cciss_host_attrs[] = {
 771         &dev_attr_rescan.attr,
 772         NULL
 773 };
 774
 775 static struct attribute_group cciss_host_attr_group = {
 776         .attrs = cciss_host_attrs,
 777 };
 778
 779 static const struct attribute_group *cciss_host_attr_groups[] = {
 780         &cciss_host_attr_group,
 781         NULL
 782 };
 783
 784 static struct device_type cciss_host_type = {
 785         .name           = "cciss_host",
 786         .groups         = cciss_host_attr_groups,
 787         .release        = cciss_hba_release,
 788 };
 789
 790 static struct attribute *cciss_dev_attrs[] = {
 791         &dev_attr_unique_id.attr,
 792         &dev_attr_model.attr,
 793         &dev_attr_vendor.attr,
 794         &dev_attr_rev.attr,
 795         &dev_attr_lunid.attr,
 796         &dev_attr_raid_level.attr,
 797         &dev_attr_usage_count.attr,
 798         NULL
 799 };
 800
 801 static struct attribute_group cciss_dev_attr_group = {
 802         .attrs = cciss_dev_attrs,
 803 };
 804
 805 static const struct attribute_group *cciss_dev_attr_groups[] = {
 806         &cciss_dev_attr_group,
 807         NULL
 808 };
 809
 810 static struct device_type cciss_dev_type = {
 811         .name           = "cciss_device",
 812         .groups         = cciss_dev_attr_groups,
 813         .release        = cciss_device_release,
 814 };
 815
 816 static struct bus_type cciss_bus_type = {
 817         .name           = "cciss",
 818 };
 819
 820 /*
 821  * cciss_hba_release is called when the reference count
 822  * of h->dev goes to zero.
 823  */
 824 static void cciss_hba_release(struct device *dev)
 825 {
 826         /*
 827          * nothing to do, but need this to avoid a warning
 828          * about not having a release handler from lib/kref.c.
 829          */
 830 }
 831
 832 /*
 833  * Initialize sysfs entry for each controller.  This sets up and registers
 834  * the 'cciss#' directory for each individual controller under
 835  * /sys/bus/pci/devices/<dev>/.
 836  */
 837 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 838 {
 839         device_initialize(&h->dev);
 840         h->dev.type = &cciss_host_type;
 841         h->dev.bus = &cciss_bus_type;
 842         dev_set_name(&h->dev, "%s", h->devname);
 843         h->dev.parent = &h->pdev->dev;
 844
 845         return device_add(&h->dev);
 846 }
 847
 848 /*
 849  * Remove sysfs entries for an hba.
 850  */
 851 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 852 {
 853         device_del(&h->dev);
 854         put_device(&h->dev); /* final put. */
 855 }
 856
 857 /* cciss_device_release is called when the reference count
 858  * of h->drv[x]dev goes to zero.
 859  */
 860 static void cciss_device_release(struct device *dev)
 861 {
 862         drive_info_struct *drv = to_drv(dev);
 863         kfree(drv);
 864 }
 865
 866 /*
 867  * Initialize sysfs for each logical drive.  This sets up and registers
 868  * the 'c#d#' directory for each individual logical drive under
 869  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 870  * /sys/block/cciss!c#d# to this entry.
 871  */
 872 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 873                                        int drv_index)
 874 {
 875         struct device *dev;
 876
 877         if (h->drv[drv_index]->device_initialized)
 878                 return 0;
 879
 880         dev = &h->drv[drv_index]->dev;
 881         device_initialize(dev);
 882         dev->type = &cciss_dev_type;
 883         dev->bus = &cciss_bus_type;
 884         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 885         dev->parent = &h->dev;
 886         h->drv[drv_index]->device_initialized = 1;
 887         return device_add(dev);
 888 }
 889
 890 /*
 891  * Remove sysfs entries for a logical drive.
 892  */
 893 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 894         int ctlr_exiting)
 895 {
 896         struct device *dev = &h->drv[drv_index]->dev;
 897
 898         /* special case for c*d0, we only destroy it on controller exit */
 899         if (drv_index == 0 && !ctlr_exiting)
 900                 return;
 901
 902         device_del(dev);
 903         put_device(dev); /* the "final" put. */
 904         h->drv[drv_index] = NULL;
 905 }
 906
 907 /*
 908  * For operations that cannot sleep, a command block is allocated at init,
 909  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 910  * which ones are free or in use.
 911  */
 912 static CommandList_struct *cmd_alloc(ctlr_info_t *h)
 913 {
 914         CommandList_struct *c;
 915         int i;
 916         u64bit temp64;
 917         dma_addr_t cmd_dma_handle, err_dma_handle;
 918
 919         do {
 920                 i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 921                 if (i == h->nr_cmds)
 922                         return NULL;
 923         } while (test_and_set_bit(i & (BITS_PER_LONG - 1),
 924                   h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 925         c = h->cmd_pool + i;
 926         memset(c, 0, sizeof(CommandList_struct));
 927         cmd_dma_handle = h->cmd_pool_dhandle + i * sizeof(CommandList_struct);
 928         c->err_info = h->errinfo_pool + i;
 929         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 930         err_dma_handle = h->errinfo_pool_dhandle
 931             + i * sizeof(ErrorInfo_struct);
 932         h->nr_allocs++;
 933
 934         c->cmdindex = i;
 935
 936         INIT_HLIST_NODE(&c->list);
 937         c->busaddr = (__u32) cmd_dma_handle;
 938         temp64.val = (__u64) err_dma_handle;
 939         c->ErrDesc.Addr.lower = temp64.val32.lower;
 940         c->ErrDesc.Addr.upper = temp64.val32.upper;
 941         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 942
 943         c->ctlr = h->ctlr;
 944         return c;
 945 }
 946
 947 /* allocate a command using pci_alloc_consistent, used for ioctls,
 948  * etc., not for the main i/o path.
 949  */
 950 static CommandList_struct *cmd_special_alloc(ctlr_info_t *h)
 951 {
 952         CommandList_struct *c;
 953         u64bit temp64;
 954         dma_addr_t cmd_dma_handle, err_dma_handle;
 955
 956         c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 957                 sizeof(CommandList_struct), &cmd_dma_handle);
 958         if (c == NULL)
 959                 return NULL;
 960         memset(c, 0, sizeof(CommandList_struct));
 961
 962         c->cmdindex = -1;
 963
 964         c->err_info = (ErrorInfo_struct *)
 965             pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 966                     &err_dma_handle);
 967
 968         if (c->err_info == NULL) {
 969                 pci_free_consistent(h->pdev,
 970                         sizeof(CommandList_struct), c, cmd_dma_handle);
 971                 return NULL;
 972         }
 973         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 974
 975         INIT_HLIST_NODE(&c->list);
 976         c->busaddr = (__u32) cmd_dma_handle;
 977         temp64.val = (__u64) err_dma_handle;
 978         c->ErrDesc.Addr.lower = temp64.val32.lower;
 979         c->ErrDesc.Addr.upper = temp64.val32.upper;
 980         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 981
 982         c->ctlr = h->ctlr;
 983         return c;
 984 }
 985
 986 static void cmd_free(ctlr_info_t *h, CommandList_struct *c)
 987 {
 988         int i;
 989
 990         i = c - h->cmd_pool;
 991         clear_bit(i & (BITS_PER_LONG - 1),
 992                   h->cmd_pool_bits + (i / BITS_PER_LONG));
 993         h->nr_frees++;
 994 }
 995
 996 static void cmd_special_free(ctlr_info_t *h, CommandList_struct *c)
 997 {
 998         u64bit temp64;
 999
1000         temp64.val32.lower = c->ErrDesc.Addr.lower;
1001         temp64.val32.upper = c->ErrDesc.Addr.upper;
1002         pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
1003                             c->err_info, (dma_addr_t) temp64.val);
1004         pci_free_consistent(h->pdev, sizeof(CommandList_struct),
1005                             c, (dma_addr_t) c->busaddr);
1006 }
1007
1008 static inline ctlr_info_t *get_host(struct gendisk *disk)
1009 {
1010         return disk->queue->queuedata;
1011 }
1012
1013 static inline drive_info_struct *get_drv(struct gendisk *disk)
1014 {
1015         return disk->private_data;
1016 }
1017
1018 /*
1019  * Open.  Make sure the device is really there.
1020  */
1021 static int cciss_open(struct block_device *bdev, fmode_t mode)
1022 {
1023         ctlr_info_t *h = get_host(bdev->bd_disk);
1024         drive_info_struct *drv = get_drv(bdev->bd_disk);
1025
1026         dev_dbg(&h->pdev->dev, "cciss_open %s\n", bdev->bd_disk->disk_name);
1027         if (drv->busy_configuring)
1028                 return -EBUSY;
1029         /*
1030          * Root is allowed to open raw volume zero even if it's not configured
1031          * so array config can still work. Root is also allowed to open any
1032          * volume that has a LUN ID, so it can issue IOCTL to reread the
1033          * disk information.  I don't think I really like this
1034          * but I'm already using way to many device nodes to claim another one
1035          * for "raw controller".
1036          */
1037         if (drv->heads == 0) {
1038                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
1039                         /* if not node 0 make sure it is a partition = 0 */
1040                         if (MINOR(bdev->bd_dev) & 0x0f) {
1041                                 return -ENXIO;
1042                                 /* if it is, make sure we have a LUN ID */
1043                         } else if (memcmp(drv->LunID, CTLR_LUNID,
1044                                 sizeof(drv->LunID))) {
1045                                 return -ENXIO;
1046                         }
1047                 }
1048                 if (!capable(CAP_SYS_ADMIN))
1049                         return -EPERM;
1050         }
1051         drv->usage_count++;
1052         h->usage_count++;
1053         return 0;
1054 }
1055
1056 static int cciss_unlocked_open(struct block_device *bdev, fmode_t mode)
1057 {
1058         int ret;
1059
1060         lock_kernel();
1061         ret = cciss_open(bdev, mode);
1062         unlock_kernel();
1063
1064         return ret;
1065 }
1066
1067 /*
1068  * Close.  Sync first.
1069  */
1070 static int cciss_release(struct gendisk *disk, fmode_t mode)
1071 {
1072         ctlr_info_t *h;
1073         drive_info_struct *drv;
1074
1075         lock_kernel();
1076         h = get_host(disk);
1077         drv = get_drv(disk);
1078         dev_dbg(&h->pdev->dev, "cciss_release %s\n", disk->disk_name);
1079         drv->usage_count--;
1080         h->usage_count--;
1081         unlock_kernel();
1082         return 0;
1083 }
1084
1085 static int do_ioctl(struct block_device *bdev, fmode_t mode,
1086                     unsigned cmd, unsigned long arg)
1087 {
1088         int ret;
1089         lock_kernel();
1090         ret = cciss_ioctl(bdev, mode, cmd, arg);
1091         unlock_kernel();
1092         return ret;
1093 }
1094
1095 #ifdef CONFIG_COMPAT
1096
1097 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1098                                   unsigned cmd, unsigned long arg);
1099 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1100                                       unsigned cmd, unsigned long arg);
1101
1102 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
1103                               unsigned cmd, unsigned long arg)
1104 {
1105         switch (cmd) {
1106         case CCISS_GETPCIINFO:
1107         case CCISS_GETINTINFO:
1108         case CCISS_SETINTINFO:
1109         case CCISS_GETNODENAME:
1110         case CCISS_SETNODENAME:
1111         case CCISS_GETHEARTBEAT:
1112         case CCISS_GETBUSTYPES:
1113         case CCISS_GETFIRMVER:
1114         case CCISS_GETDRIVVER:
1115         case CCISS_REVALIDVOLS:
1116         case CCISS_DEREGDISK:
1117         case CCISS_REGNEWDISK:
1118         case CCISS_REGNEWD:
1119         case CCISS_RESCANDISK:
1120         case CCISS_GETLUNINFO:
1121                 return do_ioctl(bdev, mode, cmd, arg);
1122
1123         case CCISS_PASSTHRU32:
1124                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
1125         case CCISS_BIG_PASSTHRU32:
1126                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
1127
1128         default:
1129                 return -ENOIOCTLCMD;
1130         }
1131 }
1132
1133 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
1134                                   unsigned cmd, unsigned long arg)
1135 {
1136         IOCTL32_Command_struct __user *arg32 =
1137             (IOCTL32_Command_struct __user *) arg;
1138         IOCTL_Command_struct arg64;
1139         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
1140         int err;
1141         u32 cp;
1142
1143         err = 0;
1144         err |=
1145             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1146                            sizeof(arg64.LUN_info));
1147         err |=
1148             copy_from_user(&arg64.Request, &arg32->Request,
1149                            sizeof(arg64.Request));
1150         err |=
1151             copy_from_user(&arg64.error_info, &arg32->error_info,
1152                            sizeof(arg64.error_info));
1153         err |= get_user(arg64.buf_size, &arg32->buf_size);
1154         err |= get_user(cp, &arg32->buf);
1155         arg64.buf = compat_ptr(cp);
1156         err |= copy_to_user(p, &arg64, sizeof(arg64));
1157
1158         if (err)
1159                 return -EFAULT;
1160
1161         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
1162         if (err)
1163                 return err;
1164         err |=
1165             copy_in_user(&arg32->error_info, &p->error_info,
1166                          sizeof(arg32->error_info));
1167         if (err)
1168                 return -EFAULT;
1169         return err;
1170 }
1171
1172 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
1173                                       unsigned cmd, unsigned long arg)
1174 {
1175         BIG_IOCTL32_Command_struct __user *arg32 =
1176             (BIG_IOCTL32_Command_struct __user *) arg;
1177         BIG_IOCTL_Command_struct arg64;
1178         BIG_IOCTL_Command_struct __user *p =
1179             compat_alloc_user_space(sizeof(arg64));
1180         int err;
1181         u32 cp;
1182
1183         err = 0;
1184         err |=
1185             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1186                            sizeof(arg64.LUN_info));
1187         err |=
1188             copy_from_user(&arg64.Request, &arg32->Request,
1189                            sizeof(arg64.Request));
1190         err |=
1191             copy_from_user(&arg64.error_info, &arg32->error_info,
1192                            sizeof(arg64.error_info));
1193         err |= get_user(arg64.buf_size, &arg32->buf_size);
1194         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1195         err |= get_user(cp, &arg32->buf);
1196         arg64.buf = compat_ptr(cp);
1197         err |= copy_to_user(p, &arg64, sizeof(arg64));
1198
1199         if (err)
1200                 return -EFAULT;
1201
1202         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1203         if (err)
1204                 return err;
1205         err |=
1206             copy_in_user(&arg32->error_info, &p->error_info,
1207                          sizeof(arg32->error_info));
1208         if (err)
1209                 return -EFAULT;
1210         return err;
1211 }
1212 #endif
1213
1214 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1215 {
1216         drive_info_struct *drv = get_drv(bdev->bd_disk);
1217
1218         if (!drv->cylinders)
1219                 return -ENXIO;
1220
1221         geo->heads = drv->heads;
1222         geo->sectors = drv->sectors;
1223         geo->cylinders = drv->cylinders;
1224         return 0;
1225 }
1226
1227 static void check_ioctl_unit_attention(ctlr_info_t *h, CommandList_struct *c)
1228 {
1229         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1230                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1231                 (void)check_for_unit_attention(h, c);
1232 }
1233 /*
1234  * ioctl
1235  */
1236 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1237                        unsigned int cmd, unsigned long arg)
1238 {
1239         struct gendisk *disk = bdev->bd_disk;
1240         ctlr_info_t *h = get_host(disk);
1241         drive_info_struct *drv = get_drv(disk);
1242         void __user *argp = (void __user *)arg;
1243
1244         dev_dbg(&h->pdev->dev, "cciss_ioctl: Called with cmd=%x %lx\n",
1245                 cmd, arg);
1246         switch (cmd) {
1247         case CCISS_GETPCIINFO:
1248                 {
1249                         cciss_pci_info_struct pciinfo;
1250
1251                         if (!arg)
1252                                 return -EINVAL;
1253                         pciinfo.domain = pci_domain_nr(h->pdev->bus);
1254                         pciinfo.bus = h->pdev->bus->number;
1255                         pciinfo.dev_fn = h->pdev->devfn;
1256                         pciinfo.board_id = h->board_id;
1257                         if (copy_to_user
1258                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1259                                 return -EFAULT;
1260                         return 0;
1261                 }
1262         case CCISS_GETINTINFO:
1263                 {
1264                         cciss_coalint_struct intinfo;
1265                         if (!arg)
1266                                 return -EINVAL;
1267                         intinfo.delay =
1268                             readl(&h->cfgtable->HostWrite.CoalIntDelay);
1269                         intinfo.count =
1270                             readl(&h->cfgtable->HostWrite.CoalIntCount);
1271                         if (copy_to_user
1272                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1273                                 return -EFAULT;
1274                         return 0;
1275                 }
1276         case CCISS_SETINTINFO:
1277                 {
1278                         cciss_coalint_struct intinfo;
1279                         unsigned long flags;
1280                         int i;
1281
1282                         if (!arg)
1283                                 return -EINVAL;
1284                         if (!capable(CAP_SYS_ADMIN))
1285                                 return -EPERM;
1286                         if (copy_from_user
1287                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1288                                 return -EFAULT;
1289                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1290                                 return -EINVAL;
1291                         spin_lock_irqsave(&h->lock, flags);
1292                         /* Update the field, and then ring the doorbell */
1293                         writel(intinfo.delay,
1294                                &(h->cfgtable->HostWrite.CoalIntDelay));
1295                         writel(intinfo.count,
1296                                &(h->cfgtable->HostWrite.CoalIntCount));
1297                         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1298
1299                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1300                                 if (!(readl(h->vaddr + SA5_DOORBELL)
1301                                       & CFGTBL_ChangeReq))
1302                                         break;
1303                                 /* delay and try again */
1304                                 udelay(1000);
1305                         }
1306                         spin_unlock_irqrestore(&h->lock, flags);
1307                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1308                                 return -EAGAIN;
1309                         return 0;
1310                 }
1311         case CCISS_GETNODENAME:
1312                 {
1313                         NodeName_type NodeName;
1314                         int i;
1315
1316                         if (!arg)
1317                                 return -EINVAL;
1318                         for (i = 0; i < 16; i++)
1319                                 NodeName[i] =
1320                                     readb(&h->cfgtable->ServerName[i]);
1321                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1322                                 return -EFAULT;
1323                         return 0;
1324                 }
1325         case CCISS_SETNODENAME:
1326                 {
1327                         NodeName_type NodeName;
1328                         unsigned long flags;
1329                         int i;
1330
1331                         if (!arg)
1332                                 return -EINVAL;
1333                         if (!capable(CAP_SYS_ADMIN))
1334                                 return -EPERM;
1335
1336                         if (copy_from_user
1337                             (NodeName, argp, sizeof(NodeName_type)))
1338                                 return -EFAULT;
1339
1340                         spin_lock_irqsave(&h->lock, flags);
1341
1342                         /* Update the field, and then ring the doorbell */
1343                         for (i = 0; i < 16; i++)
1344                                 writeb(NodeName[i],
1345                                        &h->cfgtable->ServerName[i]);
1346
1347                         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
1348
1349                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1350                                 if (!(readl(h->vaddr + SA5_DOORBELL)
1351                                       & CFGTBL_ChangeReq))
1352                                         break;
1353                                 /* delay and try again */
1354                                 udelay(1000);
1355                         }
1356                         spin_unlock_irqrestore(&h->lock, flags);
1357                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1358                                 return -EAGAIN;
1359                         return 0;
1360                 }
1361
1362         case CCISS_GETHEARTBEAT:
1363                 {
1364                         Heartbeat_type heartbeat;
1365
1366                         if (!arg)
1367                                 return -EINVAL;
1368                         heartbeat = readl(&h->cfgtable->HeartBeat);
1369                         if (copy_to_user
1370                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1371                                 return -EFAULT;
1372                         return 0;
1373                 }
1374         case CCISS_GETBUSTYPES:
1375                 {
1376                         BusTypes_type BusTypes;
1377
1378                         if (!arg)
1379                                 return -EINVAL;
1380                         BusTypes = readl(&h->cfgtable->BusTypes);
1381                         if (copy_to_user
1382                             (argp, &BusTypes, sizeof(BusTypes_type)))
1383                                 return -EFAULT;
1384                         return 0;
1385                 }
1386         case CCISS_GETFIRMVER:
1387                 {
1388                         FirmwareVer_type firmware;
1389
1390                         if (!arg)
1391                                 return -EINVAL;
1392                         memcpy(firmware, h->firm_ver, 4);
1393
1394                         if (copy_to_user
1395                             (argp, firmware, sizeof(FirmwareVer_type)))
1396                                 return -EFAULT;
1397                         return 0;
1398                 }
1399         case CCISS_GETDRIVVER:
1400                 {
1401                         DriverVer_type DriverVer = DRIVER_VERSION;
1402
1403                         if (!arg)
1404                                 return -EINVAL;
1405
1406                         if (copy_to_user
1407                             (argp, &DriverVer, sizeof(DriverVer_type)))
1408                                 return -EFAULT;
1409                         return 0;
1410                 }
1411
1412         case CCISS_DEREGDISK:
1413         case CCISS_REGNEWD:
1414         case CCISS_REVALIDVOLS:
1415                 return rebuild_lun_table(h, 0, 1);
1416
1417         case CCISS_GETLUNINFO:{
1418                         LogvolInfo_struct luninfo;
1419
1420                         memcpy(&luninfo.LunID, drv->LunID,
1421                                 sizeof(luninfo.LunID));
1422                         luninfo.num_opens = drv->usage_count;
1423                         luninfo.num_parts = 0;
1424                         if (copy_to_user(argp, &luninfo,
1425                                          sizeof(LogvolInfo_struct)))
1426                                 return -EFAULT;
1427                         return 0;
1428                 }
1429         case CCISS_PASSTHRU:
1430                 {
1431                         IOCTL_Command_struct iocommand;
1432                         CommandList_struct *c;
1433                         char *buff = NULL;
1434                         u64bit temp64;
1435                         DECLARE_COMPLETION_ONSTACK(wait);
1436
1437                         if (!arg)
1438                                 return -EINVAL;
1439
1440                         if (!capable(CAP_SYS_RAWIO))
1441                                 return -EPERM;
1442
1443                         if (copy_from_user
1444                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1445                                 return -EFAULT;
1446                         if ((iocommand.buf_size < 1) &&
1447                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1448                                 return -EINVAL;
1449                         }
1450 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1451                         /* Check kmalloc limits */
1452                         if (iocommand.buf_size > 128000)
1453                                 return -EINVAL;
1454 #endif
1455                         if (iocommand.buf_size > 0) {
1456                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1457                                 if (buff == NULL)
1458                                         return -EFAULT;
1459                         }
1460                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1461                                 /* Copy the data into the buffer we created */
1462                                 if (copy_from_user
1463                                     (buff, iocommand.buf, iocommand.buf_size)) {
1464                                         kfree(buff);
1465                                         return -EFAULT;
1466                                 }
1467                         } else {
1468                                 memset(buff, 0, iocommand.buf_size);
1469                         }
1470                         c = cmd_special_alloc(h);
1471                         if (!c) {
1472                                 kfree(buff);
1473                                 return -ENOMEM;
1474                         }
1475                         /* Fill in the command type */
1476                         c->cmd_type = CMD_IOCTL_PEND;
1477                         /* Fill in Command Header */
1478                         c->Header.ReplyQueue = 0;   /* unused in simple mode */
1479                         if (iocommand.buf_size > 0) /* buffer to fill */
1480                         {
1481                                 c->Header.SGList = 1;
1482                                 c->Header.SGTotal = 1;
1483                         } else /* no buffers to fill */
1484                         {
1485                                 c->Header.SGList = 0;
1486                                 c->Header.SGTotal = 0;
1487                         }
1488                         c->Header.LUN = iocommand.LUN_info;
1489                         /* use the kernel address the cmd block for tag */
1490                         c->Header.Tag.lower = c->busaddr;
1491
1492                         /* Fill in Request block */
1493                         c->Request = iocommand.Request;
1494
1495                         /* Fill in the scatter gather information */
1496                         if (iocommand.buf_size > 0) {
1497                                 temp64.val = pci_map_single(h->pdev, buff,
1498                                         iocommand.buf_size,
1499                                         PCI_DMA_BIDIRECTIONAL);
1500                                 c->SG[0].Addr.lower = temp64.val32.lower;
1501                                 c->SG[0].Addr.upper = temp64.val32.upper;
1502                                 c->SG[0].Len = iocommand.buf_size;
1503                                 c->SG[0].Ext = 0;  /* we are not chaining */
1504                         }
1505                         c->waiting = &wait;
1506
1507                         enqueue_cmd_and_start_io(h, c);
1508                         wait_for_completion(&wait);
1509
1510                         /* unlock the buffers from DMA */
1511                         temp64.val32.lower = c->SG[0].Addr.lower;
1512                         temp64.val32.upper = c->SG[0].Addr.upper;
1513                         pci_unmap_single(h->pdev, (dma_addr_t) temp64.val,
1514                                          iocommand.buf_size,
1515                                          PCI_DMA_BIDIRECTIONAL);
1516
1517                         check_ioctl_unit_attention(h, c);
1518
1519                         /* Copy the error information out */
1520                         iocommand.error_info = *(c->err_info);
1521                         if (copy_to_user
1522                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1523                                 kfree(buff);
1524                                 cmd_special_free(h, c);
1525                                 return -EFAULT;
1526                         }
1527
1528                         if (iocommand.Request.Type.Direction == XFER_READ) {
1529                                 /* Copy the data out of the buffer we created */
1530                                 if (copy_to_user
1531                                     (iocommand.buf, buff, iocommand.buf_size)) {
1532                                         kfree(buff);
1533                                         cmd_special_free(h, c);
1534                                         return -EFAULT;
1535                                 }
1536                         }
1537                         kfree(buff);
1538                         cmd_special_free(h, c);
1539                         return 0;
1540                 }
1541         case CCISS_BIG_PASSTHRU:{
1542                         BIG_IOCTL_Command_struct *ioc;
1543                         CommandList_struct *c;
1544                         unsigned char **buff = NULL;
1545                         int *buff_size = NULL;
1546                         u64bit temp64;
1547                         BYTE sg_used = 0;
1548                         int status = 0;
1549                         int i;
1550                         DECLARE_COMPLETION_ONSTACK(wait);
1551                         __u32 left;
1552                         __u32 sz;
1553                         BYTE __user *data_ptr;
1554
1555                         if (!arg)
1556                                 return -EINVAL;
1557                         if (!capable(CAP_SYS_RAWIO))
1558                                 return -EPERM;
1559                         ioc = (BIG_IOCTL_Command_struct *)
1560                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1561                         if (!ioc) {
1562                                 status = -ENOMEM;
1563                                 goto cleanup1;
1564                         }
1565                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1566                                 status = -EFAULT;
1567                                 goto cleanup1;
1568                         }
1569                         if ((ioc->buf_size < 1) &&
1570                             (ioc->Request.Type.Direction != XFER_NONE)) {
1571                                 status = -EINVAL;
1572                                 goto cleanup1;
1573                         }
1574                         /* Check kmalloc limits  using all SGs */
1575                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1576                                 status = -EINVAL;
1577                                 goto cleanup1;
1578                         }
1579                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1580                                 status = -EINVAL;
1581                                 goto cleanup1;
1582                         }
1583                         buff =
1584                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1585                         if (!buff) {
1586                                 status = -ENOMEM;
1587                                 goto cleanup1;
1588                         }
1589                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1590                                                    GFP_KERNEL);
1591                         if (!buff_size) {
1592                                 status = -ENOMEM;
1593                                 goto cleanup1;
1594                         }
1595                         left = ioc->buf_size;
1596                         data_ptr = ioc->buf;
1597                         while (left) {
1598                                 sz = (left >
1599                                       ioc->malloc_size) ? ioc->
1600                                     malloc_size : left;
1601                                 buff_size[sg_used] = sz;
1602                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1603                                 if (buff[sg_used] == NULL) {
1604                                         status = -ENOMEM;
1605                                         goto cleanup1;
1606                                 }
1607                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1608                                         if (copy_from_user
1609                                             (buff[sg_used], data_ptr, sz)) {
1610                                                 status = -EFAULT;
1611                                                 goto cleanup1;
1612                                         }
1613                                 } else {
1614                                         memset(buff[sg_used], 0, sz);
1615                                 }
1616                                 left -= sz;
1617                                 data_ptr += sz;
1618                                 sg_used++;
1619                         }
1620                         c = cmd_special_alloc(h);
1621                         if (!c) {
1622                                 status = -ENOMEM;
1623                                 goto cleanup1;
1624                         }
1625                         c->cmd_type = CMD_IOCTL_PEND;
1626                         c->Header.ReplyQueue = 0;
1627
1628                         if (ioc->buf_size > 0) {
1629                                 c->Header.SGList = sg_used;
1630                                 c->Header.SGTotal = sg_used;
1631                         } else {
1632                                 c->Header.SGList = 0;
1633                                 c->Header.SGTotal = 0;
1634                         }
1635                         c->Header.LUN = ioc->LUN_info;
1636                         c->Header.Tag.lower = c->busaddr;
1637
1638                         c->Request = ioc->Request;
1639                         if (ioc->buf_size > 0) {
1640                                 for (i = 0; i < sg_used; i++) {
1641                                         temp64.val =
1642                                             pci_map_single(h->pdev, buff[i],
1643                                                     buff_size[i],
1644                                                     PCI_DMA_BIDIRECTIONAL);
1645                                         c->SG[i].Addr.lower =
1646                                             temp64.val32.lower;
1647                                         c->SG[i].Addr.upper =
1648                                             temp64.val32.upper;
1649                                         c->SG[i].Len = buff_size[i];
1650                                         c->SG[i].Ext = 0;       /* we are not chaining */
1651                                 }
1652                         }
1653                         c->waiting = &wait;
1654                         enqueue_cmd_and_start_io(h, c);
1655                         wait_for_completion(&wait);
1656                         /* unlock the buffers from DMA */
1657                         for (i = 0; i < sg_used; i++) {
1658                                 temp64.val32.lower = c->SG[i].Addr.lower;
1659                                 temp64.val32.upper = c->SG[i].Addr.upper;
1660                                 pci_unmap_single(h->pdev,
1661                                         (dma_addr_t) temp64.val, buff_size[i],
1662                                         PCI_DMA_BIDIRECTIONAL);
1663                         }
1664                         check_ioctl_unit_attention(h, c);
1665                         /* Copy the error information out */
1666                         ioc->error_info = *(c->err_info);
1667                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1668                                 cmd_special_free(h, c);
1669                                 status = -EFAULT;
1670                                 goto cleanup1;
1671                         }
1672                         if (ioc->Request.Type.Direction == XFER_READ) {
1673                                 /* Copy the data out of the buffer we created */
1674                                 BYTE __user *ptr = ioc->buf;
1675                                 for (i = 0; i < sg_used; i++) {
1676                                         if (copy_to_user
1677                                             (ptr, buff[i], buff_size[i])) {
1678                                                 cmd_special_free(h, c);
1679                                                 status = -EFAULT;
1680                                                 goto cleanup1;
1681                                         }
1682                                         ptr += buff_size[i];
1683                                 }
1684                         }
1685                         cmd_special_free(h, c);
1686                         status = 0;
1687                       cleanup1:
1688                         if (buff) {
1689                                 for (i = 0; i < sg_used; i++)
1690                                         kfree(buff[i]);
1691                                 kfree(buff);
1692                         }
1693                         kfree(buff_size);
1694                         kfree(ioc);
1695                         return status;
1696                 }
1697
1698         /* scsi_cmd_ioctl handles these, below, though some are not */
1699         /* very meaningful for cciss.  SG_IO is the main one people want. */
1700
1701         case SG_GET_VERSION_NUM:
1702         case SG_SET_TIMEOUT:
1703         case SG_GET_TIMEOUT:
1704         case SG_GET_RESERVED_SIZE:
1705         case SG_SET_RESERVED_SIZE:
1706         case SG_EMULATED_HOST:
1707         case SG_IO:
1708         case SCSI_IOCTL_SEND_COMMAND:
1709                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1710
1711         /* scsi_cmd_ioctl would normally handle these, below, but */
1712         /* they aren't a good fit for cciss, as CD-ROMs are */
1713         /* not supported, and we don't have any bus/target/lun */
1714         /* which we present to the kernel. */
1715
1716         case CDROM_SEND_PACKET:
1717         case CDROMCLOSETRAY:
1718         case CDROMEJECT:
1719         case SCSI_IOCTL_GET_IDLUN:
1720         case SCSI_IOCTL_GET_BUS_NUMBER:
1721         default:
1722                 return -ENOTTY;
1723         }
1724 }
1725
1726 static void cciss_check_queues(ctlr_info_t *h)
1727 {
1728         int start_queue = h->next_to_run;
1729         int i;
1730
1731         /* check to see if we have maxed out the number of commands that can
1732          * be placed on the queue.  If so then exit.  We do this check here
1733          * in case the interrupt we serviced was from an ioctl and did not
1734          * free any new commands.
1735          */
1736         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1737                 return;
1738
1739         /* We have room on the queue for more commands.  Now we need to queue
1740          * them up.  We will also keep track of the next queue to run so
1741          * that every queue gets a chance to be started first.
1742          */
1743         for (i = 0; i < h->highest_lun + 1; i++) {
1744                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1745                 /* make sure the disk has been added and the drive is real
1746                  * because this can be called from the middle of init_one.
1747                  */
1748                 if (!h->drv[curr_queue])
1749                         continue;
1750                 if (!(h->drv[curr_queue]->queue) ||
1751                         !(h->drv[curr_queue]->heads))
1752                         continue;
1753                 blk_start_queue(h->gendisk[curr_queue]->queue);
1754
1755                 /* check to see if we have maxed out the number of commands
1756                  * that can be placed on the queue.
1757                  */
1758                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1759                         if (curr_queue == start_queue) {
1760                                 h->next_to_run =
1761                                     (start_queue + 1) % (h->highest_lun + 1);
1762                                 break;
1763                         } else {
1764                                 h->next_to_run = curr_queue;
1765                                 break;
1766                         }
1767                 }
1768         }
1769 }
1770
1771 static void cciss_softirq_done(struct request *rq)
1772 {
1773         CommandList_struct *c = rq->completion_data;
1774         ctlr_info_t *h = hba[c->ctlr];
1775         SGDescriptor_struct *curr_sg = c->SG;
1776         u64bit temp64;
1777         unsigned long flags;
1778         int i, ddir;
1779         int sg_index = 0;
1780
1781         if (c->Request.Type.Direction == XFER_READ)
1782                 ddir = PCI_DMA_FROMDEVICE;
1783         else
1784                 ddir = PCI_DMA_TODEVICE;
1785
1786         /* command did not need to be retried */
1787         /* unmap the DMA mapping for all the scatter gather elements */
1788         for (i = 0; i < c->Header.SGList; i++) {
1789                 if (curr_sg[sg_index].Ext == CCISS_SG_CHAIN) {
1790                         cciss_unmap_sg_chain_block(h, c);
1791                         /* Point to the next block */
1792                         curr_sg = h->cmd_sg_list[c->cmdindex];
1793                         sg_index = 0;
1794                 }
1795                 temp64.val32.lower = curr_sg[sg_index].Addr.lower;
1796                 temp64.val32.upper = curr_sg[sg_index].Addr.upper;
1797                 pci_unmap_page(h->pdev, temp64.val, curr_sg[sg_index].Len,
1798                                 ddir);
1799                 ++sg_index;
1800         }
1801
1802         dev_dbg(&h->pdev->dev, "Done with %p\n", rq);
1803
1804         /* set the residual count for pc requests */
1805         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1806                 rq->resid_len = c->err_info->ResidualCnt;
1807
1808         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1809
1810         spin_lock_irqsave(&h->lock, flags);
1811         cmd_free(h, c);
1812         cciss_check_queues(h);
1813         spin_unlock_irqrestore(&h->lock, flags);
1814 }
1815
1816 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1817         unsigned char scsi3addr[], uint32_t log_unit)
1818 {
1819         memcpy(scsi3addr, h->drv[log_unit]->LunID,
1820                 sizeof(h->drv[log_unit]->LunID));
1821 }
1822
1823 /* This function gets the SCSI vendor, model, and revision of a logical drive
1824  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1825  * they cannot be read.
1826  */
1827 static void cciss_get_device_descr(ctlr_info_t *h, int logvol,
1828                                    char *vendor, char *model, char *rev)
1829 {
1830         int rc;
1831         InquiryData_struct *inq_buf;
1832         unsigned char scsi3addr[8];
1833
1834         *vendor = '\0';
1835         *model = '\0';
1836         *rev = '\0';
1837
1838         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1839         if (!inq_buf)
1840                 return;
1841
1842         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1843         rc = sendcmd_withirq(h, CISS_INQUIRY, inq_buf, sizeof(*inq_buf), 0,
1844                         scsi3addr, TYPE_CMD);
1845         if (rc == IO_OK) {
1846                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1847                 vendor[VENDOR_LEN] = '\0';
1848                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1849                 model[MODEL_LEN] = '\0';
1850                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1851                 rev[REV_LEN] = '\0';
1852         }
1853
1854         kfree(inq_buf);
1855         return;
1856 }
1857
1858 /* This function gets the serial number of a logical drive via
1859  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1860  * number cannot be had, for whatever reason, 16 bytes of 0xff
1861  * are returned instead.
1862  */
1863 static void cciss_get_serial_no(ctlr_info_t *h, int logvol,
1864                                 unsigned char *serial_no, int buflen)
1865 {
1866 #define PAGE_83_INQ_BYTES 64
1867         int rc;
1868         unsigned char *buf;
1869         unsigned char scsi3addr[8];
1870
1871         if (buflen > 16)
1872                 buflen = 16;
1873         memset(serial_no, 0xff, buflen);
1874         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1875         if (!buf)
1876                 return;
1877         memset(serial_no, 0, buflen);
1878         log_unit_to_scsi3addr(h, scsi3addr, logvol);
1879         rc = sendcmd_withirq(h, CISS_INQUIRY, buf,
1880                 PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1881         if (rc == IO_OK)
1882                 memcpy(serial_no, &buf[8], buflen);
1883         kfree(buf);
1884         return;
1885 }
1886
1887 /*
1888  * cciss_add_disk sets up the block device queue for a logical drive
1889  */
1890 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1891                                 int drv_index)
1892 {
1893         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1894         if (!disk->queue)
1895                 goto init_queue_failure;
1896         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1897         disk->major = h->major;
1898         disk->first_minor = drv_index << NWD_SHIFT;
1899         disk->fops = &cciss_fops;
1900         if (cciss_create_ld_sysfs_entry(h, drv_index))
1901                 goto cleanup_queue;
1902         disk->private_data = h->drv[drv_index];
1903         disk->driverfs_dev = &h->drv[drv_index]->dev;
1904
1905         /* Set up queue information */
1906         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1907
1908         /* This is a hardware imposed limit. */
1909         blk_queue_max_segments(disk->queue, h->maxsgentries);
1910
1911         blk_queue_max_hw_sectors(disk->queue, h->cciss_max_sectors);
1912
1913         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1914
1915         disk->queue->queuedata = h;
1916
1917         blk_queue_logical_block_size(disk->queue,
1918                                      h->drv[drv_index]->block_size);
1919
1920         /* Make sure all queue data is written out before */
1921         /* setting h->drv[drv_index]->queue, as setting this */
1922         /* allows the interrupt handler to start the queue */
1923         wmb();
1924         h->drv[drv_index]->queue = disk->queue;
1925         add_disk(disk);
1926         return 0;
1927
1928 cleanup_queue:
1929         blk_cleanup_queue(disk->queue);
1930         disk->queue = NULL;
1931 init_queue_failure:
1932         return -1;
1933 }
1934
1935 /* This function will check the usage_count of the drive to be updated/added.
1936  * If the usage_count is zero and it is a heretofore unknown drive, or,
1937  * the drive's capacity, geometry, or serial number has changed,
1938  * then the drive information will be updated and the disk will be
1939  * re-registered with the kernel.  If these conditions don't hold,
1940  * then it will be left alone for the next reboot.  The exception to this
1941  * is disk 0 which will always be left registered with the kernel since it
1942  * is also the controller node.  Any changes to disk 0 will show up on
1943  * the next reboot.
1944  */
1945 static void cciss_update_drive_info(ctlr_info_t *h, int drv_index,
1946         int first_time, int via_ioctl)
1947 {
1948         struct gendisk *disk;
1949         InquiryData_struct *inq_buff = NULL;
1950         unsigned int block_size;
1951         sector_t total_size;
1952         unsigned long flags = 0;
1953         int ret = 0;
1954         drive_info_struct *drvinfo;
1955
1956         /* Get information about the disk and modify the driver structure */
1957         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1958         drvinfo = kzalloc(sizeof(*drvinfo), GFP_KERNEL);
1959         if (inq_buff == NULL || drvinfo == NULL)
1960                 goto mem_msg;
1961
1962         /* testing to see if 16-byte CDBs are already being used */
1963         if (h->cciss_read == CCISS_READ_16) {
1964                 cciss_read_capacity_16(h, drv_index,
1965                         &total_size, &block_size);
1966
1967         } else {
1968                 cciss_read_capacity(h, drv_index, &total_size, &block_size);
1969                 /* if read_capacity returns all F's this volume is >2TB */
1970                 /* in size so we switch to 16-byte CDB's for all */
1971                 /* read/write ops */
1972                 if (total_size == 0xFFFFFFFFULL) {
1973                         cciss_read_capacity_16(h, drv_index,
1974                         &total_size, &block_size);
1975                         h->cciss_read = CCISS_READ_16;
1976                         h->cciss_write = CCISS_WRITE_16;
1977                 } else {
1978                         h->cciss_read = CCISS_READ_10;
1979                         h->cciss_write = CCISS_WRITE_10;
1980                 }
1981         }
1982
1983         cciss_geometry_inquiry(h, drv_index, total_size, block_size,
1984                                inq_buff, drvinfo);
1985         drvinfo->block_size = block_size;
1986         drvinfo->nr_blocks = total_size + 1;
1987
1988         cciss_get_device_descr(h, drv_index, drvinfo->vendor,
1989                                 drvinfo->model, drvinfo->rev);
1990         cciss_get_serial_no(h, drv_index, drvinfo->serial_no,
1991                         sizeof(drvinfo->serial_no));
1992         /* Save the lunid in case we deregister the disk, below. */
1993         memcpy(drvinfo->LunID, h->drv[drv_index]->LunID,
1994                 sizeof(drvinfo->LunID));
1995
1996         /* Is it the same disk we already know, and nothing's changed? */
1997         if (h->drv[drv_index]->raid_level != -1 &&
1998                 ((memcmp(drvinfo->serial_no,
1999                                 h->drv[drv_index]->serial_no, 16) == 0) &&
2000                 drvinfo->block_size == h->drv[drv_index]->block_size &&
2001                 drvinfo->nr_blocks == h->drv[drv_index]->nr_blocks &&
2002                 drvinfo->heads == h->drv[drv_index]->heads &&
2003                 drvinfo->sectors == h->drv[drv_index]->sectors &&
2004                 drvinfo->cylinders == h->drv[drv_index]->cylinders))
2005                         /* The disk is unchanged, nothing to update */
2006                         goto freeret;
2007
2008         /* If we get here it's not the same disk, or something's changed,
2009          * so we need to * deregister it, and re-register it, if it's not
2010          * in use.
2011          * If the disk already exists then deregister it before proceeding
2012          * (unless it's the first disk (for the controller node).
2013          */
2014         if (h->drv[drv_index]->raid_level != -1 && drv_index != 0) {
2015                 dev_warn(&h->pdev->dev, "disk %d has changed.\n", drv_index);
2016                 spin_lock_irqsave(&h->lock, flags);
2017                 h->drv[drv_index]->busy_configuring = 1;
2018                 spin_unlock_irqrestore(&h->lock, flags);
2019
2020                 /* deregister_disk sets h->drv[drv_index]->queue = NULL
2021                  * which keeps the interrupt handler from starting
2022                  * the queue.
2023                  */
2024                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
2025         }
2026
2027         /* If the disk is in use return */
2028         if (ret)
2029                 goto freeret;
2030
2031         /* Save the new information from cciss_geometry_inquiry
2032          * and serial number inquiry.  If the disk was deregistered
2033          * above, then h->drv[drv_index] will be NULL.
2034          */
2035         if (h->drv[drv_index] == NULL) {
2036                 drvinfo->device_initialized = 0;
2037                 h->drv[drv_index] = drvinfo;
2038                 drvinfo = NULL; /* so it won't be freed below. */
2039         } else {
2040                 /* special case for cxd0 */
2041                 h->drv[drv_index]->block_size = drvinfo->block_size;
2042                 h->drv[drv_index]->nr_blocks = drvinfo->nr_blocks;
2043                 h->drv[drv_index]->heads = drvinfo->heads;
2044                 h->drv[drv_index]->sectors = drvinfo->sectors;
2045                 h->drv[drv_index]->cylinders = drvinfo->cylinders;
2046                 h->drv[drv_index]->raid_level = drvinfo->raid_level;
2047                 memcpy(h->drv[drv_index]->serial_no, drvinfo->serial_no, 16);
2048                 memcpy(h->drv[drv_index]->vendor, drvinfo->vendor,
2049                         VENDOR_LEN + 1);
2050                 memcpy(h->drv[drv_index]->model, drvinfo->model, MODEL_LEN + 1);
2051                 memcpy(h->drv[drv_index]->rev, drvinfo->rev, REV_LEN + 1);
2052         }
2053
2054         ++h->num_luns;
2055         disk = h->gendisk[drv_index];
2056         set_capacity(disk, h->drv[drv_index]->nr_blocks);
2057
2058         /* If it's not disk 0 (drv_index != 0)
2059          * or if it was disk 0, but there was previously
2060          * no actual corresponding configured logical drive
2061          * (raid_leve == -1) then we want to update the
2062          * logical drive's information.
2063          */
2064         if (drv_index || first_time) {
2065                 if (cciss_add_disk(h, disk, drv_index) != 0) {
2066                         cciss_free_gendisk(h, drv_index);
2067                         cciss_free_drive_info(h, drv_index);
2068                         dev_warn(&h->pdev->dev, "could not update disk %d\n",
2069                                 drv_index);
2070                         --h->num_luns;
2071                 }
2072         }
2073
2074 freeret:
2075         kfree(inq_buff);
2076         kfree(drvinfo);
2077         return;
2078 mem_msg:
2079         dev_err(&h->pdev->dev, "out of memory\n");
2080         goto freeret;
2081 }
2082
2083 /* This function will find the first index of the controllers drive array
2084  * that has a null drv pointer and allocate the drive info struct and
2085  * will return that index   This is where new drives will be added.
2086  * If the index to be returned is greater than the highest_lun index for
2087  * the controller then highest_lun is set * to this new index.
2088  * If there are no available indexes or if tha allocation fails, then -1
2089  * is returned.  * "controller_node" is used to know if this is a real
2090  * logical drive, or just the controller node, which determines if this
2091  * counts towards highest_lun.
2092  */
2093 static int cciss_alloc_drive_info(ctlr_info_t *h, int controller_node)
2094 {
2095         int i;
2096         drive_info_struct *drv;
2097
2098         /* Search for an empty slot for our drive info */
2099         for (i = 0; i < CISS_MAX_LUN; i++) {
2100
2101                 /* if not cxd0 case, and it's occupied, skip it. */
2102                 if (h->drv[i] && i != 0)
2103                         continue;
2104                 /*
2105                  * If it's cxd0 case, and drv is alloc'ed already, and a
2106                  * disk is configured there, skip it.
2107                  */
2108                 if (i == 0 && h->drv[i] && h->drv[i]->raid_level != -1)
2109                         continue;
2110
2111                 /*
2112                  * We've found an empty slot.  Update highest_lun
2113                  * provided this isn't just the fake cxd0 controller node.
2114                  */
2115                 if (i > h->highest_lun && !controller_node)
2116                         h->highest_lun = i;
2117
2118                 /* If adding a real disk at cxd0, and it's already alloc'ed */
2119                 if (i == 0 && h->drv[i] != NULL)
2120                         return i;
2121
2122                 /*
2123                  * Found an empty slot, not already alloc'ed.  Allocate it.
2124                  * Mark it with raid_level == -1, so we know it's new later on.
2125                  */
2126                 drv = kzalloc(sizeof(*drv), GFP_KERNEL);
2127                 if (!drv)
2128                         return -1;
2129                 drv->raid_level = -1; /* so we know it's new */
2130                 h->drv[i] = drv;
2131                 return i;
2132         }
2133         return -1;
2134 }
2135
2136 static void cciss_free_drive_info(ctlr_info_t *h, int drv_index)
2137 {
2138         kfree(h->drv[drv_index]);
2139         h->drv[drv_index] = NULL;
2140 }
2141
2142 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
2143 {
2144         put_disk(h->gendisk[drv_index]);
2145         h->gendisk[drv_index] = NULL;
2146 }
2147
2148 /* cciss_add_gendisk finds a free hba[]->drv structure
2149  * and allocates a gendisk if needed, and sets the lunid
2150  * in the drvinfo structure.   It returns the index into
2151  * the ->drv[] array, or -1 if none are free.
2152  * is_controller_node indicates whether highest_lun should
2153  * count this disk, or if it's only being added to provide
2154  * a means to talk to the controller in case no logical
2155  * drives have yet been configured.
2156  */
2157 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
2158         int controller_node)
2159 {
2160         int drv_index;
2161
2162         drv_index = cciss_alloc_drive_info(h, controller_node);
2163         if (drv_index == -1)
2164                 return -1;
2165
2166         /*Check if the gendisk needs to be allocated */
2167         if (!h->gendisk[drv_index]) {
2168                 h->gendisk[drv_index] =
2169                         alloc_disk(1 << NWD_SHIFT);
2170                 if (!h->gendisk[drv_index]) {
2171                         dev_err(&h->pdev->dev,
2172                                 "could not allocate a new disk %d\n",
2173                                 drv_index);
2174                         goto err_free_drive_info;
2175                 }
2176         }
2177         memcpy(h->drv[drv_index]->LunID, lunid,
2178                 sizeof(h->drv[drv_index]->LunID));
2179         if (cciss_create_ld_sysfs_entry(h, drv_index))
2180                 goto err_free_disk;
2181         /* Don't need to mark this busy because nobody */
2182         /* else knows about this disk yet to contend */
2183         /* for access to it. */
2184         h->drv[drv_index]->busy_configuring = 0;
2185         wmb();
2186         return drv_index;
2187
2188 err_free_disk:
2189         cciss_free_gendisk(h, drv_index);
2190 err_free_drive_info:
2191         cciss_free_drive_info(h, drv_index);
2192         return -1;
2193 }
2194
2195 /* This is for the special case of a controller which
2196  * has no logical drives.  In this case, we still need
2197  * to register a disk so the controller can be accessed
2198  * by the Array Config Utility.
2199  */
2200 static void cciss_add_controller_node(ctlr_info_t *h)
2201 {
2202         struct gendisk *disk;
2203         int drv_index;
2204
2205         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2206                 return;
2207
2208         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2209         if (drv_index == -1)
2210                 goto error;
2211         h->drv[drv_index]->block_size = 512;
2212         h->drv[drv_index]->nr_blocks = 0;
2213         h->drv[drv_index]->heads = 0;
2214         h->drv[drv_index]->sectors = 0;
2215         h->drv[drv_index]->cylinders = 0;
2216         h->drv[drv_index]->raid_level = -1;
2217         memset(h->drv[drv_index]->serial_no, 0, 16);
2218         disk = h->gendisk[drv_index];
2219         if (cciss_add_disk(h, disk, drv_index) == 0)
2220                 return;
2221         cciss_free_gendisk(h, drv_index);
2222         cciss_free_drive_info(h, drv_index);
2223 error:
2224         dev_warn(&h->pdev->dev, "could not add disk 0.\n");
2225         return;
2226 }
2227
2228 /* This function will add and remove logical drives from the Logical
2229  * drive array of the controller and maintain persistency of ordering
2230  * so that mount points are preserved until the next reboot.  This allows
2231  * for the removal of logical drives in the middle of the drive array
2232  * without a re-ordering of those drives.
2233  * INPUT
2234  * h            = The controller to perform the operations on
2235  */
2236 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2237         int via_ioctl)
2238 {
2239         int num_luns;
2240         ReportLunData_struct *ld_buff = NULL;
2241         int return_code;
2242         int listlength = 0;
2243         int i;
2244         int drv_found;
2245         int drv_index = 0;
2246         unsigned char lunid[8] = CTLR_LUNID;
2247         unsigned long flags;
2248
2249         if (!capable(CAP_SYS_RAWIO))
2250                 return -EPERM;
2251
2252         /* Set busy_configuring flag for this operation */
2253         spin_lock_irqsave(&h->lock, flags);
2254         if (h->busy_configuring) {
2255                 spin_unlock_irqrestore(&h->lock, flags);
2256                 return -EBUSY;
2257         }
2258         h->busy_configuring = 1;
2259         spin_unlock_irqrestore(&h->lock, flags);
2260
2261         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2262         if (ld_buff == NULL)
2263                 goto mem_msg;
2264
2265         return_code = sendcmd_withirq(h, CISS_REPORT_LOG, ld_buff,
2266                                       sizeof(ReportLunData_struct),
2267                                       0, CTLR_LUNID, TYPE_CMD);
2268
2269         if (return_code == IO_OK)
2270                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2271         else {  /* reading number of logical volumes failed */
2272                 dev_warn(&h->pdev->dev,
2273                         "report logical volume command failed\n");
2274                 listlength = 0;
2275                 goto freeret;
2276         }
2277
2278         num_luns = listlength / 8;      /* 8 bytes per entry */
2279         if (num_luns > CISS_MAX_LUN) {
2280                 num_luns = CISS_MAX_LUN;
2281                 dev_warn(&h->pdev->dev, "more luns configured"
2282                        " on controller than can be handled by"
2283                        " this driver.\n");
2284         }
2285
2286         if (num_luns == 0)
2287                 cciss_add_controller_node(h);
2288
2289         /* Compare controller drive array to driver's drive array
2290          * to see if any drives are missing on the controller due
2291          * to action of Array Config Utility (user deletes drive)
2292          * and deregister logical drives which have disappeared.
2293          */
2294         for (i = 0; i <= h->highest_lun; i++) {
2295                 int j;
2296                 drv_found = 0;
2297
2298                 /* skip holes in the array from already deleted drives */
2299                 if (h->drv[i] == NULL)
2300                         continue;
2301
2302                 for (j = 0; j < num_luns; j++) {
2303                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2304                         if (memcmp(h->drv[i]->LunID, lunid,
2305                                 sizeof(lunid)) == 0) {
2306                                 drv_found = 1;
2307                                 break;
2308                         }
2309                 }
2310                 if (!drv_found) {
2311                         /* Deregister it from the OS, it's gone. */
2312                         spin_lock_irqsave(&h->lock, flags);
2313                         h->drv[i]->busy_configuring = 1;
2314                         spin_unlock_irqrestore(&h->lock, flags);
2315                         return_code = deregister_disk(h, i, 1, via_ioctl);
2316                         if (h->drv[i] != NULL)
2317                                 h->drv[i]->busy_configuring = 0;
2318                 }
2319         }
2320
2321         /* Compare controller drive array to driver's drive array.
2322          * Check for updates in the drive information and any new drives
2323          * on the controller due to ACU adding logical drives, or changing
2324          * a logical drive's size, etc.  Reregister any new/changed drives
2325          */
2326         for (i = 0; i < num_luns; i++) {
2327                 int j;
2328
2329                 drv_found = 0;
2330
2331                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2332                 /* Find if the LUN is already in the drive array
2333                  * of the driver.  If so then update its info
2334                  * if not in use.  If it does not exist then find
2335                  * the first free index and add it.
2336                  */
2337                 for (j = 0; j <= h->highest_lun; j++) {
2338                         if (h->drv[j] != NULL &&
2339                                 memcmp(h->drv[j]->LunID, lunid,
2340                                         sizeof(h->drv[j]->LunID)) == 0) {
2341                                 drv_index = j;
2342                                 drv_found = 1;
2343                                 break;
2344                         }
2345                 }
2346
2347                 /* check if the drive was found already in the array */
2348                 if (!drv_found) {
2349                         drv_index = cciss_add_gendisk(h, lunid, 0);
2350                         if (drv_index == -1)
2351                                 goto freeret;
2352                 }
2353                 cciss_update_drive_info(h, drv_index, first_time, via_ioctl);
2354         }               /* end for */
2355
2356 freeret:
2357         kfree(ld_buff);
2358         h->busy_configuring = 0;
2359         /* We return -1 here to tell the ACU that we have registered/updated
2360          * all of the drives that we can and to keep it from calling us
2361          * additional times.
2362          */
2363         return -1;
2364 mem_msg:
2365         dev_err(&h->pdev->dev, "out of memory\n");
2366         h->busy_configuring = 0;
2367         goto freeret;
2368 }
2369
2370 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2371 {
2372         /* zero out the disk size info */
2373         drive_info->nr_blocks = 0;
2374         drive_info->block_size = 0;
2375         drive_info->heads = 0;
2376         drive_info->sectors = 0;
2377         drive_info->cylinders = 0;
2378         drive_info->raid_level = -1;
2379         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2380         memset(drive_info->model, 0, sizeof(drive_info->model));
2381         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2382         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2383         /*
2384          * don't clear the LUNID though, we need to remember which
2385          * one this one is.
2386          */
2387 }
2388
2389 /* This function will deregister the disk and it's queue from the
2390  * kernel.  It must be called with the controller lock held and the
2391  * drv structures busy_configuring flag set.  It's parameters are:
2392  *
2393  * disk = This is the disk to be deregistered
2394  * drv  = This is the drive_info_struct associated with the disk to be
2395  *        deregistered.  It contains information about the disk used
2396  *        by the driver.
2397  * clear_all = This flag determines whether or not the disk information
2398  *             is going to be completely cleared out and the highest_lun
2399  *             reset.  Sometimes we want to clear out information about
2400  *             the disk in preparation for re-adding it.  In this case
2401  *             the highest_lun should be left unchanged and the LunID
2402  *             should not be cleared.
2403  * via_ioctl
2404  *    This indicates whether we've reached this path via ioctl.
2405  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2406  *    If this path is reached via ioctl(), then the max_usage_count will
2407  *    be 1, as the process calling ioctl() has got to have the device open.
2408  *    If we get here via sysfs, then the max usage count will be zero.
2409 */
2410 static int deregister_disk(ctlr_info_t *h, int drv_index,
2411                            int clear_all, int via_ioctl)
2412 {
2413         int i;
2414         struct gendisk *disk;
2415         drive_info_struct *drv;
2416         int recalculate_highest_lun;
2417
2418         if (!capable(CAP_SYS_RAWIO))
2419                 return -EPERM;
2420
2421         drv = h->drv[drv_index];
2422         disk = h->gendisk[drv_index];
2423
2424         /* make sure logical volume is NOT is use */
2425         if (clear_all || (h->gendisk[0] == disk)) {
2426                 if (drv->usage_count > via_ioctl)
2427                         return -EBUSY;
2428         } else if (drv->usage_count > 0)
2429                 return -EBUSY;
2430
2431         recalculate_highest_lun = (drv == h->drv[h->highest_lun]);
2432
2433         /* invalidate the devices and deregister the disk.  If it is disk
2434          * zero do not deregister it but just zero out it's values.  This
2435          * allows us to delete disk zero but keep the controller registered.
2436          */
2437         if (h->gendisk[0] != disk) {
2438                 struct request_queue *q = disk->queue;
2439                 if (disk->flags & GENHD_FL_UP) {
2440                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2441                         del_gendisk(disk);
2442                 }
2443                 if (q)
2444                         blk_cleanup_queue(q);
2445                 /* If clear_all is set then we are deleting the logical
2446                  * drive, not just refreshing its info.  For drives
2447                  * other than disk 0 we will call put_disk.  We do not
2448                  * do this for disk 0 as we need it to be able to
2449                  * configure the controller.
2450                  */
2451                 if (clear_all){
2452                         /* This isn't pretty, but we need to find the
2453                          * disk in our array and NULL our the pointer.
2454                          * This is so that we will call alloc_disk if
2455                          * this index is used again later.
2456                          */
2457                         for (i=0; i < CISS_MAX_LUN; i++){
2458                                 if (h->gendisk[i] == disk) {
2459                                         h->gendisk[i] = NULL;
2460                                         break;
2461                                 }
2462                         }
2463                         put_disk(disk);
2464                 }
2465         } else {
2466                 set_capacity(disk, 0);
2467                 cciss_clear_drive_info(drv);
2468         }
2469
2470         --h->num_luns;
2471
2472         /* if it was the last disk, find the new hightest lun */
2473         if (clear_all && recalculate_highest_lun) {
2474                 int newhighest = -1;
2475                 for (i = 0; i <= h->highest_lun; i++) {
2476                         /* if the disk has size > 0, it is available */
2477                         if (h->drv[i] && h->drv[i]->heads)
2478                                 newhighest = i;
2479                 }
2480                 h->highest_lun = newhighest;
2481         }
2482         return 0;
2483 }
2484
2485 static int fill_cmd(ctlr_info_t *h, CommandList_struct *c, __u8 cmd, void *buff,
2486                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2487                 int cmd_type)
2488 {
2489         u64bit buff_dma_handle;
2490         int status = IO_OK;
2491
2492         c->cmd_type = CMD_IOCTL_PEND;
2493         c->Header.ReplyQueue = 0;
2494         if (buff != NULL) {
2495                 c->Header.SGList = 1;
2496                 c->Header.SGTotal = 1;
2497         } else {
2498                 c->Header.SGList = 0;
2499                 c->Header.SGTotal = 0;
2500         }
2501         c->Header.Tag.lower = c->busaddr;
2502         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2503
2504         c->Request.Type.Type = cmd_type;
2505         if (cmd_type == TYPE_CMD) {
2506                 switch (cmd) {
2507                 case CISS_INQUIRY:
2508                         /* are we trying to read a vital product page */
2509                         if (page_code != 0) {
2510                                 c->Request.CDB[1] = 0x01;
2511                                 c->Request.CDB[2] = page_code;
2512                         }
2513                         c->Request.CDBLen = 6;
2514                         c->Request.Type.Attribute = ATTR_SIMPLE;
2515                         c->Request.Type.Direction = XFER_READ;
2516                         c->Request.Timeout = 0;
2517                         c->Request.CDB[0] = CISS_INQUIRY;
2518                         c->Request.CDB[4] = size & 0xFF;
2519                         break;
2520                 case CISS_REPORT_LOG:
2521                 case CISS_REPORT_PHYS:
2522                         /* Talking to controller so It's a physical command
2523                            mode = 00 target = 0.  Nothing to write.
2524                          */
2525                         c->Request.CDBLen = 12;
2526                         c->Request.Type.Attribute = ATTR_SIMPLE;
2527                         c->Request.Type.Direction = XFER_READ;
2528                         c->Request.Timeout = 0;
2529                         c->Request.CDB[0] = cmd;
2530                         c->Request.CDB[6] = (size >> 24) & 0xFF; /* MSB */
2531                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2532                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2533                         c->Request.CDB[9] = size & 0xFF;
2534                         break;
2535
2536                 case CCISS_READ_CAPACITY:
2537                         c->Request.CDBLen = 10;
2538                         c->Request.Type.Attribute = ATTR_SIMPLE;
2539                         c->Request.Type.Direction = XFER_READ;
2540                         c->Request.Timeout = 0;
2541                         c->Request.CDB[0] = cmd;
2542                         break;
2543                 case CCISS_READ_CAPACITY_16:
2544                         c->Request.CDBLen = 16;
2545                         c->Request.Type.Attribute = ATTR_SIMPLE;
2546                         c->Request.Type.Direction = XFER_READ;
2547                         c->Request.Timeout = 0;
2548                         c->Request.CDB[0] = cmd;
2549                         c->Request.CDB[1] = 0x10;
2550                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2551                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2552                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2553                         c->Request.CDB[13] = size & 0xFF;
2554                         c->Request.Timeout = 0;
2555                         c->Request.CDB[0] = cmd;
2556                         break;
2557                 case CCISS_CACHE_FLUSH:
2558                         c->Request.CDBLen = 12;
2559                         c->Request.Type.Attribute = ATTR_SIMPLE;
2560                         c->Request.Type.Direction = XFER_WRITE;
2561                         c->Request.Timeout = 0;
2562                         c->Request.CDB[0] = BMIC_WRITE;
2563                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2564                         break;
2565                 case TEST_UNIT_READY:
2566                         c->Request.CDBLen = 6;
2567                         c->Request.Type.Attribute = ATTR_SIMPLE;
2568                         c->Request.Type.Direction = XFER_NONE;
2569                         c->Request.Timeout = 0;
2570                         break;
2571                 default:
2572                         dev_warn(&h->pdev->dev, "Unknown Command 0x%c\n", cmd);
2573                         return IO_ERROR;
2574                 }
2575         } else if (cmd_type == TYPE_MSG) {
2576                 switch (cmd) {
2577                 case 0: /* ABORT message */
2578                         c->Request.CDBLen = 12;
2579                         c->Request.Type.Attribute = ATTR_SIMPLE;
2580                         c->Request.Type.Direction = XFER_WRITE;
2581                         c->Request.Timeout = 0;
2582                         c->Request.CDB[0] = cmd;        /* abort */
2583                         c->Request.CDB[1] = 0;  /* abort a command */
2584                         /* buff contains the tag of the command to abort */
2585                         memcpy(&c->Request.CDB[4], buff, 8);
2586                         break;
2587                 case 1: /* RESET message */
2588                         c->Request.CDBLen = 16;
2589                         c->Request.Type.Attribute = ATTR_SIMPLE;
2590                         c->Request.Type.Direction = XFER_NONE;
2591                         c->Request.Timeout = 0;
2592                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2593                         c->Request.CDB[0] = cmd;        /* reset */
2594                         c->Request.CDB[1] = 0x03;       /* reset a target */
2595                         break;
2596                 case 3: /* No-Op message */
2597                         c->Request.CDBLen = 1;
2598                         c->Request.Type.Attribute = ATTR_SIMPLE;
2599                         c->Request.Type.Direction = XFER_WRITE;
2600                         c->Request.Timeout = 0;
2601                         c->Request.CDB[0] = cmd;
2602                         break;
2603                 default:
2604                         dev_warn(&h->pdev->dev,
2605                                 "unknown message type %d\n", cmd);
2606                         return IO_ERROR;
2607                 }
2608         } else {
2609                 dev_warn(&h->pdev->dev, "unknown command type %d\n", cmd_type);
2610                 return IO_ERROR;
2611         }
2612         /* Fill in the scatter gather information */
2613         if (size > 0) {
2614                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2615                                                              buff, size,
2616                                                              PCI_DMA_BIDIRECTIONAL);
2617                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2618                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2619                 c->SG[0].Len = size;
2620                 c->SG[0].Ext = 0;       /* we are not chaining */
2621         }
2622         return status;
2623 }
2624
2625 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2626 {
2627         switch (c->err_info->ScsiStatus) {
2628         case SAM_STAT_GOOD:
2629                 return IO_OK;
2630         case SAM_STAT_CHECK_CONDITION:
2631                 switch (0xf & c->err_info->SenseInfo[2]) {
2632                 case 0: return IO_OK; /* no sense */
2633                 case 1: return IO_OK; /* recovered error */
2634                 default:
2635                         if (check_for_unit_attention(h, c))
2636                                 return IO_NEEDS_RETRY;
2637                         dev_warn(&h->pdev->dev, "cmd 0x%02x "
2638                                 "check condition, sense key = 0x%02x\n",
2639                                 c->Request.CDB[0], c->err_info->SenseInfo[2]);
2640                 }
2641                 break;
2642         default:
2643                 dev_warn(&h->pdev->dev, "cmd 0x%02x"
2644                         "scsi status = 0x%02x\n",
2645                         c->Request.CDB[0], c->err_info->ScsiStatus);
2646                 break;
2647         }
2648         return IO_ERROR;
2649 }
2650
2651 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2652 {
2653         int return_status = IO_OK;
2654
2655         if (c->err_info->CommandStatus == CMD_SUCCESS)
2656                 return IO_OK;
2657
2658         switch (c->err_info->CommandStatus) {
2659         case CMD_TARGET_STATUS:
2660                 return_status = check_target_status(h, c);
2661                 break;
2662         case CMD_DATA_UNDERRUN:
2663         case CMD_DATA_OVERRUN:
2664                 /* expected for inquiry and report lun commands */
2665                 break;
2666         case CMD_INVALID:
2667                 dev_warn(&h->pdev->dev, "cmd 0x%02x is "
2668                        "reported invalid\n", c->Request.CDB[0]);
2669                 return_status = IO_ERROR;
2670                 break;
2671         case CMD_PROTOCOL_ERR:
2672                 dev_warn(&h->pdev->dev, "cmd 0x%02x has "
2673                        "protocol error\n", c->Request.CDB[0]);
2674                 return_status = IO_ERROR;
2675                 break;
2676         case CMD_HARDWARE_ERR:
2677                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2678                        " hardware error\n", c->Request.CDB[0]);
2679                 return_status = IO_ERROR;
2680                 break;
2681         case CMD_CONNECTION_LOST:
2682                 dev_warn(&h->pdev->dev, "cmd 0x%02x had "
2683                        "connection lost\n", c->Request.CDB[0]);
2684                 return_status = IO_ERROR;
2685                 break;
2686         case CMD_ABORTED:
2687                 dev_warn(&h->pdev->dev, "cmd 0x%02x was "
2688                        "aborted\n", c->Request.CDB[0]);
2689                 return_status = IO_ERROR;
2690                 break;
2691         case CMD_ABORT_FAILED:
2692                 dev_warn(&h->pdev->dev, "cmd 0x%02x reports "
2693                        "abort failed\n", c->Request.CDB[0]);
2694                 return_status = IO_ERROR;
2695                 break;
2696         case CMD_UNSOLICITED_ABORT:
2697                 dev_warn(&h->pdev->dev, "unsolicited abort 0x%02x\n",
2698                         c->Request.CDB[0]);
2699                 return_status = IO_NEEDS_RETRY;
2700                 break;
2701         default:
2702                 dev_warn(&h->pdev->dev, "cmd 0x%02x returned "
2703                        "unknown status %x\n", c->Request.CDB[0],
2704                        c->err_info->CommandStatus);
2705                 return_status = IO_ERROR;
2706         }
2707         return return_status;
2708 }
2709
2710 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2711         int attempt_retry)
2712 {
2713         DECLARE_COMPLETION_ONSTACK(wait);
2714         u64bit buff_dma_handle;
2715         int return_status = IO_OK;
2716
2717 resend_cmd2:
2718         c->waiting = &wait;
2719         enqueue_cmd_and_start_io(h, c);
2720
2721         wait_for_completion(&wait);
2722
2723         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2724                 goto command_done;
2725
2726         return_status = process_sendcmd_error(h, c);
2727
2728         if (return_status == IO_NEEDS_RETRY &&
2729                 c->retry_count < MAX_CMD_RETRIES) {
2730                 dev_warn(&h->pdev->dev, "retrying 0x%02x\n",
2731                         c->Request.CDB[0]);
2732                 c->retry_count++;
2733                 /* erase the old error information */
2734                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2735                 return_status = IO_OK;
2736                 INIT_COMPLETION(wait);
2737                 goto resend_cmd2;
2738         }
2739
2740 command_done:
2741         /* unlock the buffers from DMA */
2742         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2743         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2744         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2745                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2746         return return_status;
2747 }
2748
2749 static int sendcmd_withirq(ctlr_info_t *h, __u8 cmd, void *buff, size_t size,
2750                            __u8 page_code, unsigned char scsi3addr[],
2751                         int cmd_type)
2752 {
2753         CommandList_struct *c;
2754         int return_status;
2755
2756         c = cmd_special_alloc(h);
2757         if (!c)
2758                 return -ENOMEM;
2759         return_status = fill_cmd(h, c, cmd, buff, size, page_code,
2760                 scsi3addr, cmd_type);
2761         if (return_status == IO_OK)
2762                 return_status = sendcmd_withirq_core(h, c, 1);
2763
2764         cmd_special_free(h, c);
2765         return return_status;
2766 }
2767
2768 static void cciss_geometry_inquiry(ctlr_info_t *h, int logvol,
2769                                    sector_t total_size,
2770                                    unsigned int block_size,
2771                                    InquiryData_struct *inq_buff,
2772                                    drive_info_struct *drv)
2773 {
2774         int return_code;
2775         unsigned long t;
2776         unsigned char scsi3addr[8];
2777
2778         memset(inq_buff, 0, sizeof(InquiryData_struct));
2779         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2780         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
2781                         sizeof(*inq_buff), 0xC1, scsi3addr, TYPE_CMD);
2782         if (return_code == IO_OK) {
2783                 if (inq_buff->data_byte[8] == 0xFF) {
2784                         dev_warn(&h->pdev->dev,
2785                                "reading geometry failed, volume "
2786                                "does not support reading geometry\n");
2787                         drv->heads = 255;
2788                         drv->sectors = 32;      /* Sectors per track */
2789                         drv->cylinders = total_size + 1;
2790                         drv->raid_level = RAID_UNKNOWN;
2791                 } else {
2792                         drv->heads = inq_buff->data_byte[6];
2793                         drv->sectors = inq_buff->data_byte[7];
2794                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2795                         drv->cylinders += inq_buff->data_byte[5];
2796                         drv->raid_level = inq_buff->data_byte[8];
2797                 }
2798                 drv->block_size = block_size;
2799                 drv->nr_blocks = total_size + 1;
2800                 t = drv->heads * drv->sectors;
2801                 if (t > 1) {
2802                         sector_t real_size = total_size + 1;
2803                         unsigned long rem = sector_div(real_size, t);
2804                         if (rem)
2805                                 real_size++;
2806                         drv->cylinders = real_size;
2807                 }
2808         } else {                /* Get geometry failed */
2809                 dev_warn(&h->pdev->dev, "reading geometry failed\n");
2810         }
2811 }
2812
2813 static void
2814 cciss_read_capacity(ctlr_info_t *h, int logvol, sector_t *total_size,
2815                     unsigned int *block_size)
2816 {
2817         ReadCapdata_struct *buf;
2818         int return_code;
2819         unsigned char scsi3addr[8];
2820
2821         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2822         if (!buf) {
2823                 dev_warn(&h->pdev->dev, "out of memory\n");
2824                 return;
2825         }
2826
2827         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2828         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY, buf,
2829                 sizeof(ReadCapdata_struct), 0, scsi3addr, TYPE_CMD);
2830         if (return_code == IO_OK) {
2831                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2832                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2833         } else {                /* read capacity command failed */
2834                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2835                 *total_size = 0;
2836                 *block_size = BLOCK_SIZE;
2837         }
2838         kfree(buf);
2839 }
2840
2841 static void cciss_read_capacity_16(ctlr_info_t *h, int logvol,
2842         sector_t *total_size, unsigned int *block_size)
2843 {
2844         ReadCapdata_struct_16 *buf;
2845         int return_code;
2846         unsigned char scsi3addr[8];
2847
2848         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2849         if (!buf) {
2850                 dev_warn(&h->pdev->dev, "out of memory\n");
2851                 return;
2852         }
2853
2854         log_unit_to_scsi3addr(h, scsi3addr, logvol);
2855         return_code = sendcmd_withirq(h, CCISS_READ_CAPACITY_16,
2856                 buf, sizeof(ReadCapdata_struct_16),
2857                         0, scsi3addr, TYPE_CMD);
2858         if (return_code == IO_OK) {
2859                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2860                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2861         } else {                /* read capacity command failed */
2862                 dev_warn(&h->pdev->dev, "read capacity failed\n");
2863                 *total_size = 0;
2864                 *block_size = BLOCK_SIZE;
2865         }
2866         dev_info(&h->pdev->dev, "      blocks= %llu block_size= %d\n",
2867                (unsigned long long)*total_size+1, *block_size);
2868         kfree(buf);
2869 }
2870
2871 static int cciss_revalidate(struct gendisk *disk)
2872 {
2873         ctlr_info_t *h = get_host(disk);
2874         drive_info_struct *drv = get_drv(disk);
2875         int logvol;
2876         int FOUND = 0;
2877         unsigned int block_size;
2878         sector_t total_size;
2879         InquiryData_struct *inq_buff = NULL;
2880
2881         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2882                 if (memcmp(h->drv[logvol]->LunID, drv->LunID,
2883                         sizeof(drv->LunID)) == 0) {
2884                         FOUND = 1;
2885                         break;
2886                 }
2887         }
2888
2889         if (!FOUND)
2890                 return 1;
2891
2892         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2893         if (inq_buff == NULL) {
2894                 dev_warn(&h->pdev->dev, "out of memory\n");
2895                 return 1;
2896         }
2897         if (h->cciss_read == CCISS_READ_10) {
2898                 cciss_read_capacity(h, logvol,
2899                                         &total_size, &block_size);
2900         } else {
2901                 cciss_read_capacity_16(h, logvol,
2902                                         &total_size, &block_size);
2903         }
2904         cciss_geometry_inquiry(h, logvol, total_size, block_size,
2905                                inq_buff, drv);
2906
2907         blk_queue_logical_block_size(drv->queue, drv->block_size);
2908         set_capacity(disk, drv->nr_blocks);
2909
2910         kfree(inq_buff);
2911         return 0;
2912 }
2913
2914 /*
2915  * Map (physical) PCI mem into (virtual) kernel space
2916  */
2917 static void __iomem *remap_pci_mem(ulong base, ulong size)
2918 {
2919         ulong page_base = ((ulong) base) & PAGE_MASK;
2920         ulong page_offs = ((ulong) base) - page_base;
2921         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2922
2923         return page_remapped ? (page_remapped + page_offs) : NULL;
2924 }
2925
2926 /*
2927  * Takes jobs of the Q and sends them to the hardware, then puts it on
2928  * the Q to wait for completion.
2929  */
2930 static void start_io(ctlr_info_t *h)
2931 {
2932         CommandList_struct *c;
2933
2934         while (!hlist_empty(&h->reqQ)) {
2935                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2936                 /* can't do anything if fifo is full */
2937                 if ((h->access.fifo_full(h))) {
2938                         dev_warn(&h->pdev->dev, "fifo full\n");
2939                         break;
2940                 }
2941
2942                 /* Get the first entry from the Request Q */
2943                 removeQ(c);
2944                 h->Qdepth--;
2945
2946                 /* Tell the controller execute command */
2947                 h->access.submit_command(h, c);
2948
2949                 /* Put job onto the completed Q */
2950                 addQ(&h->cmpQ, c);
2951         }
2952 }
2953
2954 /* Assumes that h->lock is held. */
2955 /* Zeros out the error record and then resends the command back */
2956 /* to the controller */
2957 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2958 {
2959         /* erase the old error information */
2960         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2961
2962         /* add it to software queue and then send it to the controller */
2963         addQ(&h->reqQ, c);
2964         h->Qdepth++;
2965         if (h->Qdepth > h->maxQsinceinit)
2966                 h->maxQsinceinit = h->Qdepth;
2967
2968         start_io(h);
2969 }
2970
2971 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2972         unsigned int msg_byte, unsigned int host_byte,
2973         unsigned int driver_byte)
2974 {
2975         /* inverse of macros in scsi.h */
2976         return (scsi_status_byte & 0xff) |
2977                 ((msg_byte & 0xff) << 8) |
2978                 ((host_byte & 0xff) << 16) |
2979                 ((driver_byte & 0xff) << 24);
2980 }
2981
2982 static inline int evaluate_target_status(ctlr_info_t *h,
2983                         CommandList_struct *cmd, int *retry_cmd)
2984 {
2985         unsigned char sense_key;
2986         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2987         int error_value;
2988
2989         *retry_cmd = 0;
2990         /* If we get in here, it means we got "target status", that is, scsi status */
2991         status_byte = cmd->err_info->ScsiStatus;
2992         driver_byte = DRIVER_OK;
2993         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2994
2995         if (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC)
2996                 host_byte = DID_PASSTHROUGH;
2997         else
2998                 host_byte = DID_OK;
2999
3000         error_value = make_status_bytes(status_byte, msg_byte,
3001                 host_byte, driver_byte);
3002
3003         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
3004                 if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC)
3005                         dev_warn(&h->pdev->dev, "cmd %p "
3006                                "has SCSI Status 0x%x\n",
3007                                cmd, cmd->err_info->ScsiStatus);
3008                 return error_value;
3009         }
3010
3011         /* check the sense key */
3012         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3013         /* no status or recovered error */
3014         if (((sense_key == 0x0) || (sense_key == 0x1)) &&
3015             (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC))
3016                 error_value = 0;
3017
3018         if (check_for_unit_attention(h, cmd)) {
3019                 *retry_cmd = !(cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC);
3020                 return 0;
3021         }
3022
3023         /* Not SG_IO or similar? */
3024         if (cmd->rq->cmd_type != REQ_TYPE_BLOCK_PC) {
3025                 if (error_value != 0)
3026                         dev_warn(&h->pdev->dev, "cmd %p has CHECK CONDITION"
3027                                " sense key = 0x%x\n", cmd, sense_key);
3028                 return error_value;
3029         }
3030
3031         /* SG_IO or similar, copy sense data back */
3032         if (cmd->rq->sense) {
3033                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3034                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3035                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3036                         cmd->rq->sense_len);
3037         } else
3038                 cmd->rq->sense_len = 0;
3039
3040         return error_value;
3041 }
3042
3043 /* checks the status of the job and calls complete buffers to mark all
3044  * buffers for the completed job. Note that this function does not need
3045  * to hold the hba/queue lock.
3046  */
3047 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3048                                     int timeout)
3049 {
3050         int retry_cmd = 0;
3051         struct request *rq = cmd->rq;
3052
3053         rq->errors = 0;
3054
3055         if (timeout)
3056                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3057
3058         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3059                 goto after_error_processing;
3060
3061         switch (cmd->err_info->CommandStatus) {
3062         case CMD_TARGET_STATUS:
3063                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3064                 break;
3065         case CMD_DATA_UNDERRUN:
3066                 if (cmd->rq->cmd_type == REQ_TYPE_FS) {
3067                         dev_warn(&h->pdev->dev, "cmd %p has"
3068                                " completed with data underrun "
3069                                "reported\n", cmd);
3070                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3071                 }
3072                 break;
3073         case CMD_DATA_OVERRUN:
3074                 if (cmd->rq->cmd_type == REQ_TYPE_FS)
3075                         dev_warn(&h->pdev->dev, "cciss: cmd %p has"
3076                                " completed with data overrun "
3077                                "reported\n", cmd);
3078                 break;
3079         case CMD_INVALID:
3080                 dev_warn(&h->pdev->dev, "cciss: cmd %p is "
3081                        "reported invalid\n", cmd);
3082                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3083                         cmd->err_info->CommandStatus, DRIVER_OK,
3084                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3085                                 DID_PASSTHROUGH : DID_ERROR);
3086                 break;
3087         case CMD_PROTOCOL_ERR:
3088                 dev_warn(&h->pdev->dev, "cciss: cmd %p has "
3089                        "protocol error\n", cmd);
3090                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3091                         cmd->err_info->CommandStatus, DRIVER_OK,
3092                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3093                                 DID_PASSTHROUGH : DID_ERROR);
3094                 break;
3095         case CMD_HARDWARE_ERR:
3096                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3097                        " hardware error\n", cmd);
3098                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3099                         cmd->err_info->CommandStatus, DRIVER_OK,
3100                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3101                                 DID_PASSTHROUGH : DID_ERROR);
3102                 break;
3103         case CMD_CONNECTION_LOST:
3104                 dev_warn(&h->pdev->dev, "cciss: cmd %p had "
3105                        "connection lost\n", cmd);
3106                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3107                         cmd->err_info->CommandStatus, DRIVER_OK,
3108                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3109                                 DID_PASSTHROUGH : DID_ERROR);
3110                 break;
3111         case CMD_ABORTED:
3112                 dev_warn(&h->pdev->dev, "cciss: cmd %p was "
3113                        "aborted\n", cmd);
3114                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3115                         cmd->err_info->CommandStatus, DRIVER_OK,
3116                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3117                                 DID_PASSTHROUGH : DID_ABORT);
3118                 break;
3119         case CMD_ABORT_FAILED:
3120                 dev_warn(&h->pdev->dev, "cciss: cmd %p reports "
3121                        "abort failed\n", cmd);
3122                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3123                         cmd->err_info->CommandStatus, DRIVER_OK,
3124                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3125                                 DID_PASSTHROUGH : DID_ERROR);
3126                 break;
3127         case CMD_UNSOLICITED_ABORT:
3128                 dev_warn(&h->pdev->dev, "cciss%d: unsolicited "
3129                        "abort %p\n", h->ctlr, cmd);
3130                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3131                         retry_cmd = 1;
3132                         dev_warn(&h->pdev->dev, "retrying %p\n", cmd);
3133                         cmd->retry_count++;
3134                 } else
3135                         dev_warn(&h->pdev->dev,
3136                                 "%p retried too many times\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_ABORT);
3141                 break;
3142         case CMD_TIMEOUT:
3143                 dev_warn(&h->pdev->dev, "cmd %p timedout\n", cmd);
3144                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3145                         cmd->err_info->CommandStatus, DRIVER_OK,
3146                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3147                                 DID_PASSTHROUGH : DID_ERROR);
3148                 break;
3149         default:
3150                 dev_warn(&h->pdev->dev, "cmd %p returned "
3151                        "unknown status %x\n", cmd,
3152                        cmd->err_info->CommandStatus);
3153                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3154                         cmd->err_info->CommandStatus, DRIVER_OK,
3155                         (cmd->rq->cmd_type == REQ_TYPE_BLOCK_PC) ?
3156                                 DID_PASSTHROUGH : DID_ERROR);
3157         }
3158
3159 after_error_processing:
3160
3161         /* We need to return this command */
3162         if (retry_cmd) {
3163                 resend_cciss_cmd(h, cmd);
3164                 return;
3165         }
3166         cmd->rq->completion_data = cmd;
3167         blk_complete_request(cmd->rq);
3168 }
3169
3170 static inline u32 cciss_tag_contains_index(u32 tag)
3171 {
3172 #define DIRECT_LOOKUP_BIT 0x10
3173         return tag & DIRECT_LOOKUP_BIT;
3174 }
3175
3176 static inline u32 cciss_tag_to_index(u32 tag)
3177 {
3178 #define DIRECT_LOOKUP_SHIFT 5
3179         return tag >> DIRECT_LOOKUP_SHIFT;
3180 }
3181
3182 static inline u32 cciss_tag_discard_error_bits(u32 tag)
3183 {
3184 #define CCISS_ERROR_BITS 0x03
3185         return tag & ~CCISS_ERROR_BITS;
3186 }
3187
3188 static inline void cciss_mark_tag_indexed(u32 *tag)
3189 {
3190         *tag |= DIRECT_LOOKUP_BIT;
3191 }
3192
3193 static inline void cciss_set_tag_index(u32 *tag, u32 index)
3194 {
3195         *tag |= (index << DIRECT_LOOKUP_SHIFT);
3196 }
3197
3198 /*
3199  * Get a request and submit it to the controller.
3200  */
3201 static void do_cciss_request(struct request_queue *q)
3202 {
3203         ctlr_info_t *h = q->queuedata;
3204         CommandList_struct *c;
3205         sector_t start_blk;
3206         int seg;
3207         struct request *creq;
3208         u64bit temp64;
3209         struct scatterlist *tmp_sg;
3210         SGDescriptor_struct *curr_sg;
3211         drive_info_struct *drv;
3212         int i, dir;
3213         int sg_index = 0;
3214         int chained = 0;
3215
3216         /* We call start_io here in case there is a command waiting on the
3217          * queue that has not been sent.
3218          */
3219         if (blk_queue_plugged(q))
3220                 goto startio;
3221
3222       queue:
3223         creq = blk_peek_request(q);
3224         if (!creq)
3225                 goto startio;
3226
3227         BUG_ON(creq->nr_phys_segments > h->maxsgentries);
3228
3229         c = cmd_alloc(h);
3230         if (!c)
3231                 goto full;
3232
3233         blk_start_request(creq);
3234
3235         tmp_sg = h->scatter_list[c->cmdindex];
3236         spin_unlock_irq(q->queue_lock);
3237
3238         c->cmd_type = CMD_RWREQ;
3239         c->rq = creq;
3240
3241         /* fill in the request */
3242         drv = creq->rq_disk->private_data;
3243         c->Header.ReplyQueue = 0;       /* unused in simple mode */
3244         /* got command from pool, so use the command block index instead */
3245         /* for direct lookups. */
3246         /* The first 2 bits are reserved for controller error reporting. */
3247         cciss_set_tag_index(&c->Header.Tag.lower, c->cmdindex);
3248         cciss_mark_tag_indexed(&c->Header.Tag.lower);
3249         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3250         c->Request.CDBLen = 10; /* 12 byte commands not in FW yet; */
3251         c->Request.Type.Type = TYPE_CMD;        /* It is a command. */
3252         c->Request.Type.Attribute = ATTR_SIMPLE;
3253         c->Request.Type.Direction =
3254             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3255         c->Request.Timeout = 0; /* Don't time out */
3256         c->Request.CDB[0] =
3257             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3258         start_blk = blk_rq_pos(creq);
3259         dev_dbg(&h->pdev->dev, "sector =%d nr_sectors=%d\n",
3260                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3261         sg_init_table(tmp_sg, h->maxsgentries);
3262         seg = blk_rq_map_sg(q, creq, tmp_sg);
3263
3264         /* get the DMA records for the setup */
3265         if (c->Request.Type.Direction == XFER_READ)
3266                 dir = PCI_DMA_FROMDEVICE;
3267         else
3268                 dir = PCI_DMA_TODEVICE;
3269
3270         curr_sg = c->SG;
3271         sg_index = 0;
3272         chained = 0;
3273
3274         for (i = 0; i < seg; i++) {
3275                 if (((sg_index+1) == (h->max_cmd_sgentries)) &&
3276                         !chained && ((seg - i) > 1)) {
3277                         /* Point to next chain block. */
3278                         curr_sg = h->cmd_sg_list[c->cmdindex];
3279                         sg_index = 0;
3280                         chained = 1;
3281                 }
3282                 curr_sg[sg_index].Len = tmp_sg[i].length;
3283                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3284                                                 tmp_sg[i].offset,
3285                                                 tmp_sg[i].length, dir);
3286                 curr_sg[sg_index].Addr.lower = temp64.val32.lower;
3287                 curr_sg[sg_index].Addr.upper = temp64.val32.upper;
3288                 curr_sg[sg_index].Ext = 0;  /* we are not chaining */
3289                 ++sg_index;
3290         }
3291         if (chained)
3292                 cciss_map_sg_chain_block(h, c, h->cmd_sg_list[c->cmdindex],
3293                         (seg - (h->max_cmd_sgentries - 1)) *
3294                                 sizeof(SGDescriptor_struct));
3295
3296         /* track how many SG entries we are using */
3297         if (seg > h->maxSG)
3298                 h->maxSG = seg;
3299
3300         dev_dbg(&h->pdev->dev, "Submitting %u sectors in %d segments "
3301                         "chained[%d]\n",
3302                         blk_rq_sectors(creq), seg, chained);
3303
3304         c->Header.SGTotal = seg + chained;
3305         if (seg <= h->max_cmd_sgentries)
3306                 c->Header.SGList = c->Header.SGTotal;
3307         else
3308                 c->Header.SGList = h->max_cmd_sgentries;
3309         set_performant_mode(h, c);
3310
3311         if (likely(creq->cmd_type == REQ_TYPE_FS)) {
3312                 if(h->cciss_read == CCISS_READ_10) {
3313                         c->Request.CDB[1] = 0;
3314                         c->Request.CDB[2] = (start_blk >> 24) & 0xff; /* MSB */
3315                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3316                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3317                         c->Request.CDB[5] = start_blk & 0xff;
3318                         c->Request.CDB[6] = 0; /* (sect >> 24) & 0xff; MSB */
3319                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3320                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3321                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3322                 } else {
3323                         u32 upper32 = upper_32_bits(start_blk);
3324
3325                         c->Request.CDBLen = 16;
3326                         c->Request.CDB[1]= 0;
3327                         c->Request.CDB[2]= (upper32 >> 24) & 0xff; /* MSB */
3328                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3329                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3330                         c->Request.CDB[5]= upper32 & 0xff;
3331                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3332                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3333                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3334                         c->Request.CDB[9]= start_blk & 0xff;
3335                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3336                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3337                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3338                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3339                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3340                 }
3341         } else if (creq->cmd_type == REQ_TYPE_BLOCK_PC) {
3342                 c->Request.CDBLen = creq->cmd_len;
3343                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3344         } else {
3345                 dev_warn(&h->pdev->dev, "bad request type %d\n",
3346                         creq->cmd_type);
3347                 BUG();
3348         }
3349
3350         spin_lock_irq(q->queue_lock);
3351
3352         addQ(&h->reqQ, c);
3353         h->Qdepth++;
3354         if (h->Qdepth > h->maxQsinceinit)
3355                 h->maxQsinceinit = h->Qdepth;
3356
3357         goto queue;
3358 full:
3359         blk_stop_queue(q);
3360 startio:
3361         /* We will already have the driver lock here so not need
3362          * to lock it.
3363          */
3364         start_io(h);
3365 }
3366
3367 static inline unsigned long get_next_completion(ctlr_info_t *h)
3368 {
3369         return h->access.command_completed(h);
3370 }
3371
3372 static inline int interrupt_pending(ctlr_info_t *h)
3373 {
3374         return h->access.intr_pending(h);
3375 }
3376
3377 static inline long interrupt_not_for_us(ctlr_info_t *h)
3378 {
3379         return ((h->access.intr_pending(h) == 0) ||
3380                 (h->interrupts_enabled == 0));
3381 }
3382
3383 static inline int bad_tag(ctlr_info_t *h, u32 tag_index,
3384                         u32 raw_tag)
3385 {
3386         if (unlikely(tag_index >= h->nr_cmds)) {
3387                 dev_warn(&h->pdev->dev, "bad tag 0x%08x ignored.\n", raw_tag);
3388                 return 1;
3389         }
3390         return 0;
3391 }
3392
3393 static inline void finish_cmd(ctlr_info_t *h, CommandList_struct *c,
3394                                 u32 raw_tag)
3395 {
3396         removeQ(c);
3397         if (likely(c->cmd_type == CMD_RWREQ))
3398                 complete_command(h, c, 0);
3399         else if (c->cmd_type == CMD_IOCTL_PEND)
3400                 complete(c->waiting);
3401 #ifdef CONFIG_CISS_SCSI_TAPE
3402         else if (c->cmd_type == CMD_SCSI)
3403                 complete_scsi_command(c, 0, raw_tag);
3404 #endif
3405 }
3406
3407 static inline u32 next_command(ctlr_info_t *h)
3408 {
3409         u32 a;
3410
3411         if (unlikely(h->transMethod != CFGTBL_Trans_Performant))
3412                 return h->access.command_completed(h);
3413
3414         if ((*(h->reply_pool_head) & 1) == (h->reply_pool_wraparound)) {
3415                 a = *(h->reply_pool_head); /* Next cmd in ring buffer */
3416                 (h->reply_pool_head)++;
3417                 h->commands_outstanding--;
3418         } else {
3419                 a = FIFO_EMPTY;
3420         }
3421         /* Check for wraparound */
3422         if (h->reply_pool_head == (h->reply_pool + h->max_commands)) {
3423                 h->reply_pool_head = h->reply_pool;
3424                 h->reply_pool_wraparound ^= 1;
3425         }
3426         return a;
3427 }
3428
3429 /* process completion of an indexed ("direct lookup") command */
3430 static inline u32 process_indexed_cmd(ctlr_info_t *h, u32 raw_tag)
3431 {
3432         u32 tag_index;
3433         CommandList_struct *c;
3434
3435         tag_index = cciss_tag_to_index(raw_tag);
3436         if (bad_tag(h, tag_index, raw_tag))
3437                 return next_command(h);
3438         c = h->cmd_pool + tag_index;
3439         finish_cmd(h, c, raw_tag);
3440         return next_command(h);
3441 }
3442
3443 /* process completion of a non-indexed command */
3444 static inline u32 process_nonindexed_cmd(ctlr_info_t *h, u32 raw_tag)
3445 {
3446         u32 tag;
3447         CommandList_struct *c = NULL;
3448         struct hlist_node *tmp;
3449         __u32 busaddr_masked, tag_masked;
3450
3451         tag = cciss_tag_discard_error_bits(raw_tag);
3452         hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3453                 busaddr_masked = cciss_tag_discard_error_bits(c->busaddr);
3454                 tag_masked = cciss_tag_discard_error_bits(tag);
3455                 if (busaddr_masked == tag_masked) {
3456                         finish_cmd(h, c, raw_tag);
3457                         return next_command(h);
3458                 }
3459         }
3460         bad_tag(h, h->nr_cmds + 1, raw_tag);
3461         return next_command(h);
3462 }
3463
3464 static irqreturn_t do_cciss_intx(int irq, void *dev_id)
3465 {
3466         ctlr_info_t *h = dev_id;
3467         unsigned long flags;
3468         u32 raw_tag;
3469
3470         if (interrupt_not_for_us(h))
3471                 return IRQ_NONE;
3472         spin_lock_irqsave(&h->lock, flags);
3473         while (interrupt_pending(h)) {
3474                 raw_tag = get_next_completion(h);
3475                 while (raw_tag != FIFO_EMPTY) {
3476                         if (cciss_tag_contains_index(raw_tag))
3477                                 raw_tag = process_indexed_cmd(h, raw_tag);
3478                         else
3479                                 raw_tag = process_nonindexed_cmd(h, raw_tag);
3480                 }
3481         }
3482         spin_unlock_irqrestore(&h->lock, flags);
3483         return IRQ_HANDLED;
3484 }
3485
3486 /* Add a second interrupt handler for MSI/MSI-X mode. In this mode we never
3487  * check the interrupt pending register because it is not set.
3488  */
3489 static irqreturn_t do_cciss_msix_intr(int irq, void *dev_id)
3490 {
3491         ctlr_info_t *h = dev_id;
3492         unsigned long flags;
3493         u32 raw_tag;
3494
3495         spin_lock_irqsave(&h->lock, flags);
3496         raw_tag = get_next_completion(h);
3497         while (raw_tag != FIFO_EMPTY) {
3498                 if (cciss_tag_contains_index(raw_tag))
3499                         raw_tag = process_indexed_cmd(h, raw_tag);
3500                 else
3501                         raw_tag = process_nonindexed_cmd(h, raw_tag);
3502         }
3503         spin_unlock_irqrestore(&h->lock, flags);
3504         return IRQ_HANDLED;
3505 }
3506
3507 /**
3508  * add_to_scan_list() - add controller to rescan queue
3509  * @h:                Pointer to the controller.
3510  *
3511  * Adds the controller to the rescan queue if not already on the queue.
3512  *
3513  * returns 1 if added to the queue, 0 if skipped (could be on the
3514  * queue already, or the controller could be initializing or shutting
3515  * down).
3516  **/
3517 static int add_to_scan_list(struct ctlr_info *h)
3518 {
3519         struct ctlr_info *test_h;
3520         int found = 0;
3521         int ret = 0;
3522
3523         if (h->busy_initializing)
3524                 return 0;
3525
3526         if (!mutex_trylock(&h->busy_shutting_down))
3527                 return 0;
3528
3529         mutex_lock(&scan_mutex);
3530         list_for_each_entry(test_h, &scan_q, scan_list) {
3531                 if (test_h == h) {
3532                         found = 1;
3533                         break;
3534                 }
3535         }
3536         if (!found && !h->busy_scanning) {
3537                 INIT_COMPLETION(h->scan_wait);
3538                 list_add_tail(&h->scan_list, &scan_q);
3539                 ret = 1;
3540         }
3541         mutex_unlock(&scan_mutex);
3542         mutex_unlock(&h->busy_shutting_down);
3543
3544         return ret;
3545 }
3546
3547 /**
3548  * remove_from_scan_list() - remove controller from rescan queue
3549  * @h:                     Pointer to the controller.
3550  *
3551  * Removes the controller from the rescan queue if present. Blocks if
3552  * the controller is currently conducting a rescan.  The controller
3553  * can be in one of three states:
3554  * 1. Doesn't need a scan
3555  * 2. On the scan list, but not scanning yet (we remove it)
3556  * 3. Busy scanning (and not on the list). In this case we want to wait for
3557  *    the scan to complete to make sure the scanning thread for this
3558  *    controller is completely idle.
3559  **/
3560 static void remove_from_scan_list(struct ctlr_info *h)
3561 {
3562         struct ctlr_info *test_h, *tmp_h;
3563
3564         mutex_lock(&scan_mutex);
3565         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3566                 if (test_h == h) { /* state 2. */
3567                         list_del(&h->scan_list);
3568                         complete_all(&h->scan_wait);
3569                         mutex_unlock(&scan_mutex);
3570                         return;
3571                 }
3572         }
3573         if (h->busy_scanning) { /* state 3. */
3574                 mutex_unlock(&scan_mutex);
3575                 wait_for_completion(&h->scan_wait);
3576         } else { /* state 1, nothing to do. */
3577                 mutex_unlock(&scan_mutex);
3578         }
3579 }
3580
3581 /**
3582  * scan_thread() - kernel thread used to rescan controllers
3583  * @data:        Ignored.
3584  *
3585  * A kernel thread used scan for drive topology changes on
3586  * controllers. The thread processes only one controller at a time
3587  * using a queue.  Controllers are added to the queue using
3588  * add_to_scan_list() and removed from the queue either after done
3589  * processing or using remove_from_scan_list().
3590  *
3591  * returns 0.
3592  **/
3593 static int scan_thread(void *data)
3594 {
3595         struct ctlr_info *h;
3596
3597         while (1) {
3598                 set_current_state(TASK_INTERRUPTIBLE);
3599                 schedule();
3600                 if (kthread_should_stop())
3601                         break;
3602
3603                 while (1) {
3604                         mutex_lock(&scan_mutex);
3605                         if (list_empty(&scan_q)) {
3606                                 mutex_unlock(&scan_mutex);
3607                                 break;
3608                         }
3609
3610                         h = list_entry(scan_q.next,
3611                                        struct ctlr_info,
3612                                        scan_list);
3613                         list_del(&h->scan_list);
3614                         h->busy_scanning = 1;
3615                         mutex_unlock(&scan_mutex);
3616
3617                         rebuild_lun_table(h, 0, 0);
3618                         complete_all(&h->scan_wait);
3619                         mutex_lock(&scan_mutex);
3620                         h->busy_scanning = 0;
3621                         mutex_unlock(&scan_mutex);
3622                 }
3623         }
3624
3625         return 0;
3626 }
3627
3628 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3629 {
3630         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3631                 return 0;
3632
3633         switch (c->err_info->SenseInfo[12]) {
3634         case STATE_CHANGED:
3635                 dev_warn(&h->pdev->dev, "a state change "
3636                         "detected, command retried\n");
3637                 return 1;
3638         break;
3639         case LUN_FAILED:
3640                 dev_warn(&h->pdev->dev, "LUN failure "
3641                         "detected, action required\n");
3642                 return 1;
3643         break;
3644         case REPORT_LUNS_CHANGED:
3645                 dev_warn(&h->pdev->dev, "report LUN data changed\n");
3646         /*
3647          * Here, we could call add_to_scan_list and wake up the scan thread,
3648          * except that it's quite likely that we will get more than one
3649          * REPORT_LUNS_CHANGED condition in quick succession, which means
3650          * that those which occur after the first one will likely happen
3651          * *during* the scan_thread's rescan.  And the rescan code is not
3652          * robust enough to restart in the middle, undoing what it has already
3653          * done, and it's not clear that it's even possible to do this, since
3654          * part of what it does is notify the block layer, which starts
3655          * doing it's own i/o to read partition tables and so on, and the
3656          * driver doesn't have visibility to know what might need undoing.
3657          * In any event, if possible, it is horribly complicated to get right
3658          * so we just don't do it for now.
3659          *
3660          * Note: this REPORT_LUNS_CHANGED condition only occurs on the MSA2012.
3661          */
3662                 return 1;
3663         break;
3664         case POWER_OR_RESET:
3665                 dev_warn(&h->pdev->dev,
3666                         "a power on or device reset detected\n");
3667                 return 1;
3668         break;
3669         case UNIT_ATTENTION_CLEARED:
3670                 dev_warn(&h->pdev->dev,
3671                         "unit attention cleared by another initiator\n");
3672                 return 1;
3673         break;
3674         default:
3675                 dev_warn(&h->pdev->dev, "unknown unit attention detected\n");
3676                 return 1;
3677         }
3678 }
3679
3680 /*
3681  *  We cannot read the structure directly, for portability we must use
3682  *   the io functions.
3683  *   This is for debug only.
3684  */
3685 static void print_cfg_table(ctlr_info_t *h)
3686 {
3687         int i;
3688         char temp_name[17];
3689         CfgTable_struct *tb = h->cfgtable;
3690
3691         dev_dbg(&h->pdev->dev, "Controller Configuration information\n");
3692         dev_dbg(&h->pdev->dev, "------------------------------------\n");
3693         for (i = 0; i < 4; i++)
3694                 temp_name[i] = readb(&(tb->Signature[i]));
3695         temp_name[4] = '\0';
3696         dev_dbg(&h->pdev->dev, "   Signature = %s\n", temp_name);
3697         dev_dbg(&h->pdev->dev, "   Spec Number = %d\n",
3698                 readl(&(tb->SpecValence)));
3699         dev_dbg(&h->pdev->dev, "   Transport methods supported = 0x%x\n",
3700                readl(&(tb->TransportSupport)));
3701         dev_dbg(&h->pdev->dev, "   Transport methods active = 0x%x\n",
3702                readl(&(tb->TransportActive)));
3703         dev_dbg(&h->pdev->dev, "   Requested transport Method = 0x%x\n",
3704                readl(&(tb->HostWrite.TransportRequest)));
3705         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Delay = 0x%x\n",
3706                readl(&(tb->HostWrite.CoalIntDelay)));
3707         dev_dbg(&h->pdev->dev, "   Coalesce Interrupt Count = 0x%x\n",
3708                readl(&(tb->HostWrite.CoalIntCount)));
3709         dev_dbg(&h->pdev->dev, "   Max outstanding commands = 0x%d\n",
3710                readl(&(tb->CmdsOutMax)));
3711         dev_dbg(&h->pdev->dev, "   Bus Types = 0x%x\n",
3712                 readl(&(tb->BusTypes)));
3713         for (i = 0; i < 16; i++)
3714                 temp_name[i] = readb(&(tb->ServerName[i]));
3715         temp_name[16] = '\0';
3716         dev_dbg(&h->pdev->dev, "   Server Name = %s\n", temp_name);
3717         dev_dbg(&h->pdev->dev, "   Heartbeat Counter = 0x%x\n\n\n",
3718                 readl(&(tb->HeartBeat)));
3719 }
3720
3721 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3722 {
3723         int i, offset, mem_type, bar_type;
3724         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3725                 return 0;
3726         offset = 0;
3727         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3728                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3729                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3730                         offset += 4;
3731                 else {
3732                         mem_type = pci_resource_flags(pdev, i) &
3733                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3734                         switch (mem_type) {
3735                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3736                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3737                                 offset += 4;    /* 32 bit */
3738                                 break;
3739                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3740                                 offset += 8;
3741                                 break;
3742                         default:        /* reserved in PCI 2.2 */
3743                                 dev_warn(&pdev->dev,
3744                                        "Base address is invalid\n");
3745                                 return -1;
3746                                 break;
3747                         }
3748                 }
3749                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3750                         return i + 1;
3751         }
3752         return -1;
3753 }
3754
3755 /* Fill in bucket_map[], given nsgs (the max number of
3756  * scatter gather elements supported) and bucket[],
3757  * which is an array of 8 integers.  The bucket[] array
3758  * contains 8 different DMA transfer sizes (in 16
3759  * byte increments) which the controller uses to fetch
3760  * commands.  This function fills in bucket_map[], which
3761  * maps a given number of scatter gather elements to one of
3762  * the 8 DMA transfer sizes.  The point of it is to allow the
3763  * controller to only do as much DMA as needed to fetch the
3764  * command, with the DMA transfer size encoded in the lower
3765  * bits of the command address.
3766  */
3767 static void  calc_bucket_map(int bucket[], int num_buckets,
3768         int nsgs, int *bucket_map)
3769 {
3770         int i, j, b, size;
3771
3772         /* even a command with 0 SGs requires 4 blocks */
3773 #define MINIMUM_TRANSFER_BLOCKS 4
3774 #define NUM_BUCKETS 8
3775         /* Note, bucket_map must have nsgs+1 entries. */
3776         for (i = 0; i <= nsgs; i++) {
3777                 /* Compute size of a command with i SG entries */
3778                 size = i + MINIMUM_TRANSFER_BLOCKS;
3779                 b = num_buckets; /* Assume the biggest bucket */
3780                 /* Find the bucket that is just big enough */
3781                 for (j = 0; j < 8; j++) {
3782                         if (bucket[j] >= size) {
3783                                 b = j;
3784                                 break;
3785                         }
3786                 }
3787                 /* for a command with i SG entries, use bucket b. */
3788                 bucket_map[i] = b;
3789         }
3790 }
3791
3792 static void __devinit cciss_wait_for_mode_change_ack(ctlr_info_t *h)
3793 {
3794         int i;
3795
3796         /* under certain very rare conditions, this can take awhile.
3797          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3798          * as we enter this code.) */
3799         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3800                 if (!(readl(h->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3801                         break;
3802                 msleep(10);
3803         }
3804 }
3805
3806 static __devinit void cciss_enter_performant_mode(ctlr_info_t *h)
3807 {
3808         /* This is a bit complicated.  There are 8 registers on
3809          * the controller which we write to to tell it 8 different
3810          * sizes of commands which there may be.  It's a way of
3811          * reducing the DMA done to fetch each command.  Encoded into
3812          * each command's tag are 3 bits which communicate to the controller
3813          * which of the eight sizes that command fits within.  The size of
3814          * each command depends on how many scatter gather entries there are.
3815          * Each SG entry requires 16 bytes.  The eight registers are programmed
3816          * with the number of 16-byte blocks a command of that size requires.
3817          * The smallest command possible requires 5 such 16 byte blocks.
3818          * the largest command possible requires MAXSGENTRIES + 4 16-byte
3819          * blocks.  Note, this only extends to the SG entries contained
3820          * within the command block, and does not extend to chained blocks
3821          * of SG elements.   bft[] contains the eight values we write to
3822          * the registers.  They are not evenly distributed, but have more
3823          * sizes for small commands, and fewer sizes for larger commands.
3824          */
3825         __u32 trans_offset;
3826         int bft[8] = { 5, 6, 8, 10, 12, 20, 28, MAXSGENTRIES + 4};
3827                         /*
3828                          *  5 = 1 s/g entry or 4k
3829                          *  6 = 2 s/g entry or 8k
3830                          *  8 = 4 s/g entry or 16k
3831                          * 10 = 6 s/g entry or 24k
3832                          */
3833         unsigned long register_value;
3834         BUILD_BUG_ON(28 > MAXSGENTRIES + 4);
3835
3836         h->reply_pool_wraparound = 1; /* spec: init to 1 */
3837
3838         /* Controller spec: zero out this buffer. */
3839         memset(h->reply_pool, 0, h->max_commands * sizeof(__u64));
3840         h->reply_pool_head = h->reply_pool;
3841
3842         trans_offset = readl(&(h->cfgtable->TransMethodOffset));
3843         calc_bucket_map(bft, ARRAY_SIZE(bft), h->maxsgentries,
3844                                 h->blockFetchTable);
3845         writel(bft[0], &h->transtable->BlockFetch0);
3846         writel(bft[1], &h->transtable->BlockFetch1);
3847         writel(bft[2], &h->transtable->BlockFetch2);
3848         writel(bft[3], &h->transtable->BlockFetch3);
3849         writel(bft[4], &h->transtable->BlockFetch4);
3850         writel(bft[5], &h->transtable->BlockFetch5);
3851         writel(bft[6], &h->transtable->BlockFetch6);
3852         writel(bft[7], &h->transtable->BlockFetch7);
3853
3854         /* size of controller ring buffer */
3855         writel(h->max_commands, &h->transtable->RepQSize);
3856         writel(1, &h->transtable->RepQCount);
3857         writel(0, &h->transtable->RepQCtrAddrLow32);
3858         writel(0, &h->transtable->RepQCtrAddrHigh32);
3859         writel(h->reply_pool_dhandle, &h->transtable->RepQAddr0Low32);
3860         writel(0, &h->transtable->RepQAddr0High32);
3861         writel(CFGTBL_Trans_Performant,
3862                         &(h->cfgtable->HostWrite.TransportRequest));
3863
3864         writel(CFGTBL_ChangeReq, h->vaddr + SA5_DOORBELL);
3865         cciss_wait_for_mode_change_ack(h);
3866         register_value = readl(&(h->cfgtable->TransportActive));
3867         if (!(register_value & CFGTBL_Trans_Performant))
3868                 dev_warn(&h->pdev->dev, "cciss: unable to get board into"
3869                                         " performant mode\n");
3870 }
3871
3872 static void __devinit cciss_put_controller_into_performant_mode(ctlr_info_t *h)
3873 {
3874         __u32 trans_support;
3875
3876         dev_dbg(&h->pdev->dev, "Trying to put board into Performant mode\n");
3877         /* Attempt to put controller into performant mode if supported */
3878         /* Does board support performant mode? */
3879         trans_support = readl(&(h->cfgtable->TransportSupport));
3880         if (!(trans_support & PERFORMANT_MODE))
3881                 return;
3882
3883         dev_dbg(&h->pdev->dev, "Placing controller into performant mode\n");
3884         /* Performant mode demands commands on a 32 byte boundary
3885          * pci_alloc_consistent aligns on page boundarys already.
3886          * Just need to check if divisible by 32
3887          */
3888         if ((sizeof(CommandList_struct) % 32) != 0) {
3889                 dev_warn(&h->pdev->dev, "%s %d %s\n",
3890                         "cciss info: command size[",
3891                         (int)sizeof(CommandList_struct),
3892                         "] not divisible by 32, no performant mode..\n");
3893                 return;
3894         }
3895
3896         /* Performant mode ring buffer and supporting data structures */
3897         h->reply_pool = (__u64 *)pci_alloc_consistent(
3898                 h->pdev, h->max_commands * sizeof(__u64),
3899                 &(h->reply_pool_dhandle));
3900
3901         /* Need a block fetch table for performant mode */
3902         h->blockFetchTable = kmalloc(((h->maxsgentries+1) *
3903                 sizeof(__u32)), GFP_KERNEL);
3904
3905         if ((h->reply_pool == NULL) || (h->blockFetchTable == NULL))
3906                 goto clean_up;
3907
3908         cciss_enter_performant_mode(h);
3909
3910         /* Change the access methods to the performant access methods */
3911         h->access = SA5_performant_access;
3912         h->transMethod = CFGTBL_Trans_Performant;
3913
3914         return;
3915 clean_up:
3916         kfree(h->blockFetchTable);
3917         if (h->reply_pool)
3918                 pci_free_consistent(h->pdev,
3919                                 h->max_commands * sizeof(__u64),
3920                                 h->reply_pool,
3921                                 h->reply_pool_dhandle);
3922         return;
3923
3924 } /* cciss_put_controller_into_performant_mode */
3925
3926 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3927  * controllers that are capable. If not, we use IO-APIC mode.
3928  */
3929
3930 static void __devinit cciss_interrupt_mode(ctlr_info_t *h)
3931 {
3932 #ifdef CONFIG_PCI_MSI
3933         int err;
3934         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3935         {0, 2}, {0, 3}
3936         };
3937
3938         /* Some boards advertise MSI but don't really support it */
3939         if ((h->board_id == 0x40700E11) || (h->board_id == 0x40800E11) ||
3940             (h->board_id == 0x40820E11) || (h->board_id == 0x40830E11))
3941                 goto default_int_mode;
3942
3943         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSIX)) {
3944                 err = pci_enable_msix(h->pdev, cciss_msix_entries, 4);
3945                 if (!err) {
3946                         h->intr[0] = cciss_msix_entries[0].vector;
3947                         h->intr[1] = cciss_msix_entries[1].vector;
3948                         h->intr[2] = cciss_msix_entries[2].vector;
3949                         h->intr[3] = cciss_msix_entries[3].vector;
3950                         h->msix_vector = 1;
3951                         return;
3952                 }
3953                 if (err > 0) {
3954                         dev_warn(&h->pdev->dev,
3955                                 "only %d MSI-X vectors available\n", err);
3956                         goto default_int_mode;
3957                 } else {
3958                         dev_warn(&h->pdev->dev,
3959                                 "MSI-X init failed %d\n", err);
3960                         goto default_int_mode;
3961                 }
3962         }
3963         if (pci_find_capability(h->pdev, PCI_CAP_ID_MSI)) {
3964                 if (!pci_enable_msi(h->pdev))
3965                         h->msi_vector = 1;
3966                 else
3967                         dev_warn(&h->pdev->dev, "MSI init failed\n");
3968         }
3969 default_int_mode:
3970 #endif                          /* CONFIG_PCI_MSI */
3971         /* if we get here we're going to use the default interrupt mode */
3972         h->intr[PERF_MODE_INT] = h->pdev->irq;
3973         return;
3974 }
3975
3976 static int __devinit cciss_lookup_board_id(struct pci_dev *pdev, u32 *board_id)
3977 {
3978         int i;
3979         u32 subsystem_vendor_id, subsystem_device_id;
3980
3981         subsystem_vendor_id = pdev->subsystem_vendor;
3982         subsystem_device_id = pdev->subsystem_device;
3983         *board_id = ((subsystem_device_id << 16) & 0xffff0000) |
3984                         subsystem_vendor_id;
3985
3986         for (i = 0; i < ARRAY_SIZE(products); i++) {
3987                 /* Stand aside for hpsa driver on request */
3988                 if (cciss_allow_hpsa && products[i].board_id == HPSA_BOUNDARY)
3989                         return -ENODEV;
3990                 if (*board_id == products[i].board_id)
3991                         return i;
3992         }
3993         dev_warn(&pdev->dev, "unrecognized board ID: 0x%08x, ignoring.\n",
3994                 *board_id);
3995         return -ENODEV;
3996 }
3997
3998 static inline bool cciss_board_disabled(ctlr_info_t *h)
3999 {
4000         u16 command;
4001
4002         (void) pci_read_config_word(h->pdev, PCI_COMMAND, &command);
4003         return ((command & PCI_COMMAND_MEMORY) == 0);
4004 }
4005
4006 static int __devinit cciss_pci_find_memory_BAR(struct pci_dev *pdev,
4007         unsigned long *memory_bar)
4008 {
4009         int i;
4010
4011         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++)
4012                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
4013                         /* addressing mode bits already removed */
4014                         *memory_bar = pci_resource_start(pdev, i);
4015                         dev_dbg(&pdev->dev, "memory BAR = %lx\n",
4016                                 *memory_bar);
4017                         return 0;
4018                 }
4019         dev_warn(&pdev->dev, "no memory BAR found\n");
4020         return -ENODEV;
4021 }
4022
4023 static int __devinit cciss_wait_for_board_ready(ctlr_info_t *h)
4024 {
4025         int i;
4026         u32 scratchpad;
4027
4028         for (i = 0; i < CCISS_BOARD_READY_ITERATIONS; i++) {
4029                 scratchpad = readl(h->vaddr + SA5_SCRATCHPAD_OFFSET);
4030                 if (scratchpad == CCISS_FIRMWARE_READY)
4031                         return 0;
4032                 msleep(CCISS_BOARD_READY_POLL_INTERVAL_MSECS);
4033         }
4034         dev_warn(&h->pdev->dev, "board not ready, timed out.\n");
4035         return -ENODEV;
4036 }
4037
4038 static int __devinit cciss_find_cfg_addrs(struct pci_dev *pdev,
4039         void __iomem *vaddr, u32 *cfg_base_addr, u64 *cfg_base_addr_index,
4040         u64 *cfg_offset)
4041 {
4042         *cfg_base_addr = readl(vaddr + SA5_CTCFG_OFFSET);
4043         *cfg_offset = readl(vaddr + SA5_CTMEM_OFFSET);
4044         *cfg_base_addr &= (u32) 0x0000ffff;
4045         *cfg_base_addr_index = find_PCI_BAR_index(pdev, *cfg_base_addr);
4046         if (*cfg_base_addr_index == -1) {
4047                 dev_warn(&pdev->dev, "cannot find cfg_base_addr_index, "
4048                         "*cfg_base_addr = 0x%08x\n", *cfg_base_addr);
4049                 return -ENODEV;
4050         }
4051         return 0;
4052 }
4053
4054 static int __devinit cciss_find_cfgtables(ctlr_info_t *h)
4055 {
4056         u64 cfg_offset;
4057         u32 cfg_base_addr;
4058         u64 cfg_base_addr_index;
4059         u32 trans_offset;
4060         int rc;
4061
4062         rc = cciss_find_cfg_addrs(h->pdev, h->vaddr, &cfg_base_addr,
4063                 &cfg_base_addr_index, &cfg_offset);
4064         if (rc)
4065                 return rc;
4066         h->cfgtable = remap_pci_mem(pci_resource_start(h->pdev,
4067                 cfg_base_addr_index) + cfg_offset, sizeof(h->cfgtable));
4068         if (!h->cfgtable)
4069                 return -ENOMEM;
4070         /* Find performant mode table. */
4071         trans_offset = readl(&h->cfgtable->TransMethodOffset);
4072         h->transtable = remap_pci_mem(pci_resource_start(h->pdev,
4073                                 cfg_base_addr_index)+cfg_offset+trans_offset,
4074                                 sizeof(*h->transtable));
4075         if (!h->transtable)
4076                 return -ENOMEM;
4077         return 0;
4078 }
4079
4080 static void __devinit cciss_get_max_perf_mode_cmds(struct ctlr_info *h)
4081 {
4082         h->max_commands = readl(&(h->cfgtable->MaxPerformantModeCommands));
4083         if (h->max_commands < 16) {
4084                 dev_warn(&h->pdev->dev, "Controller reports "
4085                         "max supported commands of %d, an obvious lie. "
4086                         "Using 16.  Ensure that firmware is up to date.\n",
4087                         h->max_commands);
4088                 h->max_commands = 16;
4089         }
4090 }
4091
4092 /* Interrogate the hardware for some limits:
4093  * max commands, max SG elements without chaining, and with chaining,
4094  * SG chain block size, etc.
4095  */
4096 static void __devinit cciss_find_board_params(ctlr_info_t *h)
4097 {
4098         cciss_get_max_perf_mode_cmds(h);
4099         h->nr_cmds = h->max_commands - 4; /* Allow room for some ioctls */
4100         h->maxsgentries = readl(&(h->cfgtable->MaxSGElements));
4101         /*
4102          * Limit in-command s/g elements to 32 save dma'able memory.
4103          * Howvever spec says if 0, use 31
4104          */
4105         h->max_cmd_sgentries = 31;
4106         if (h->maxsgentries > 512) {
4107                 h->max_cmd_sgentries = 32;
4108                 h->chainsize = h->maxsgentries - h->max_cmd_sgentries + 1;
4109                 h->maxsgentries--; /* save one for chain pointer */
4110         } else {
4111                 h->maxsgentries = 31; /* default to traditional values */
4112                 h->chainsize = 0;
4113         }
4114 }
4115
4116 static inline bool CISS_signature_present(ctlr_info_t *h)
4117 {
4118         if ((readb(&h->cfgtable->Signature[0]) != 'C') ||
4119             (readb(&h->cfgtable->Signature[1]) != 'I') ||
4120             (readb(&h->cfgtable->Signature[2]) != 'S') ||
4121             (readb(&h->cfgtable->Signature[3]) != 'S')) {
4122                 dev_warn(&h->pdev->dev, "not a valid CISS config table\n");
4123                 return false;
4124         }
4125         return true;
4126 }
4127
4128 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
4129 static inline void cciss_enable_scsi_prefetch(ctlr_info_t *h)
4130 {
4131 #ifdef CONFIG_X86
4132         u32 prefetch;
4133
4134         prefetch = readl(&(h->cfgtable->SCSI_Prefetch));
4135         prefetch |= 0x100;
4136         writel(prefetch, &(h->cfgtable->SCSI_Prefetch));
4137 #endif
4138 }
4139
4140 /* Disable DMA prefetch for the P600.  Otherwise an ASIC bug may result
4141  * in a prefetch beyond physical memory.
4142  */
4143 static inline void cciss_p600_dma_prefetch_quirk(ctlr_info_t *h)
4144 {
4145         u32 dma_prefetch;
4146         __u32 dma_refetch;
4147
4148         if (h->board_id != 0x3225103C)
4149                 return;
4150         dma_prefetch = readl(h->vaddr + I2O_DMA1_CFG);
4151         dma_prefetch |= 0x8000;
4152         writel(dma_prefetch, h->vaddr + I2O_DMA1_CFG);
4153         pci_read_config_dword(h->pdev, PCI_COMMAND_PARITY, &dma_refetch);
4154         dma_refetch |= 0x1;
4155         pci_write_config_dword(h->pdev, PCI_COMMAND_PARITY, dma_refetch);
4156 }
4157
4158 static int __devinit cciss_pci_init(ctlr_info_t *h)
4159 {
4160         int prod_index, err;
4161
4162         prod_index = cciss_lookup_board_id(h->pdev, &h->board_id);
4163         if (prod_index < 0)
4164                 return -ENODEV;
4165         h->product_name = products[prod_index].product_name;
4166         h->access = *(products[prod_index].access);
4167
4168         if (cciss_board_disabled(h)) {
4169                 dev_warn(&h->pdev->dev, "controller appears to be disabled\n");
4170                 return -ENODEV;
4171         }
4172         err = pci_enable_device(h->pdev);
4173         if (err) {
4174                 dev_warn(&h->pdev->dev, "Unable to Enable PCI device\n");
4175                 return err;
4176         }
4177
4178         err = pci_request_regions(h->pdev, "cciss");
4179         if (err) {
4180                 dev_warn(&h->pdev->dev,
4181                         "Cannot obtain PCI resources, aborting\n");
4182                 return err;
4183         }
4184
4185         dev_dbg(&h->pdev->dev, "irq = %x\n", h->pdev->irq);
4186         dev_dbg(&h->pdev->dev, "board_id = %x\n", h->board_id);
4187
4188 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
4189  * else we use the IO-APIC interrupt assigned to us by system ROM.
4190  */
4191         cciss_interrupt_mode(h);
4192         err = cciss_pci_find_memory_BAR(h->pdev, &h->paddr);
4193         if (err)
4194                 goto err_out_free_res;
4195         h->vaddr = remap_pci_mem(h->paddr, 0x250);
4196         if (!h->vaddr) {
4197                 err = -ENOMEM;
4198                 goto err_out_free_res;
4199         }
4200         err = cciss_wait_for_board_ready(h);
4201         if (err)
4202                 goto err_out_free_res;
4203         err = cciss_find_cfgtables(h);
4204         if (err)
4205                 goto err_out_free_res;
4206         print_cfg_table(h);
4207         cciss_find_board_params(h);
4208
4209         if (!CISS_signature_present(h)) {
4210                 err = -ENODEV;
4211                 goto err_out_free_res;
4212         }
4213         cciss_enable_scsi_prefetch(h);
4214         cciss_p600_dma_prefetch_quirk(h);
4215         cciss_put_controller_into_performant_mode(h);
4216         return 0;
4217
4218 err_out_free_res:
4219         /*
4220          * Deliberately omit pci_disable_device(): it does something nasty to
4221          * Smart Array controllers that pci_enable_device does not undo
4222          */
4223         if (h->transtable)
4224                 iounmap(h->transtable);
4225         if (h->cfgtable)
4226                 iounmap(h->cfgtable);
4227         if (h->vaddr)
4228                 iounmap(h->vaddr);
4229         pci_release_regions(h->pdev);
4230         return err;
4231 }
4232
4233 /* Function to find the first free pointer into our hba[] array
4234  * Returns -1 if no free entries are left.
4235  */
4236 static int alloc_cciss_hba(struct pci_dev *pdev)
4237 {
4238         int i;
4239
4240         for (i = 0; i < MAX_CTLR; i++) {
4241                 if (!hba[i]) {
4242                         ctlr_info_t *h;
4243
4244                         h = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
4245                         if (!h)
4246                                 goto Enomem;
4247                         hba[i] = h;
4248                         return i;
4249                 }
4250         }
4251         dev_warn(&pdev->dev, "This driver supports a maximum"
4252                " of %d controllers.\n", MAX_CTLR);
4253         return -1;
4254 Enomem:
4255         dev_warn(&pdev->dev, "out of memory.\n");
4256         return -1;
4257 }
4258
4259 static void free_hba(ctlr_info_t *h)
4260 {
4261         int i;
4262
4263         hba[h->ctlr] = NULL;
4264         for (i = 0; i < h->highest_lun + 1; i++)
4265                 if (h->gendisk[i] != NULL)
4266                         put_disk(h->gendisk[i]);
4267         kfree(h);
4268 }
4269
4270 /* Send a message CDB to the firmware. */
4271 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
4272 {
4273         typedef struct {
4274                 CommandListHeader_struct CommandHeader;
4275                 RequestBlock_struct Request;
4276                 ErrDescriptor_struct ErrorDescriptor;
4277         } Command;
4278         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
4279         Command *cmd;
4280         dma_addr_t paddr64;
4281         uint32_t paddr32, tag;
4282         void __iomem *vaddr;
4283         int i, err;
4284
4285         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
4286         if (vaddr == NULL)
4287                 return -ENOMEM;
4288
4289         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
4290            CCISS commands, so they must be allocated from the lower 4GiB of
4291            memory. */
4292         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
4293         if (err) {
4294                 iounmap(vaddr);
4295                 return -ENOMEM;
4296         }
4297
4298         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
4299         if (cmd == NULL) {
4300                 iounmap(vaddr);
4301                 return -ENOMEM;
4302         }
4303
4304         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
4305            although there's no guarantee, we assume that the address is at
4306            least 4-byte aligned (most likely, it's page-aligned). */
4307         paddr32 = paddr64;
4308
4309         cmd->CommandHeader.ReplyQueue = 0;
4310         cmd->CommandHeader.SGList = 0;
4311         cmd->CommandHeader.SGTotal = 0;
4312         cmd->CommandHeader.Tag.lower = paddr32;
4313         cmd->CommandHeader.Tag.upper = 0;
4314         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
4315
4316         cmd->Request.CDBLen = 16;
4317         cmd->Request.Type.Type = TYPE_MSG;
4318         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
4319         cmd->Request.Type.Direction = XFER_NONE;
4320         cmd->Request.Timeout = 0; /* Don't time out */
4321         cmd->Request.CDB[0] = opcode;
4322         cmd->Request.CDB[1] = type;
4323         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
4324
4325         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
4326         cmd->ErrorDescriptor.Addr.upper = 0;
4327         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
4328
4329         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
4330
4331         for (i = 0; i < 10; i++) {
4332                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
4333                 if ((tag & ~3) == paddr32)
4334                         break;
4335                 schedule_timeout_uninterruptible(HZ);
4336         }
4337
4338         iounmap(vaddr);
4339
4340         /* we leak the DMA buffer here ... no choice since the controller could
4341            still complete the command. */
4342         if (i == 10) {
4343                 dev_err(&pdev->dev,
4344                         "controller message %02x:%02x timed out\n",
4345                         opcode, type);
4346                 return -ETIMEDOUT;
4347         }
4348
4349         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4350
4351         if (tag & 2) {
4352                 dev_err(&pdev->dev, "controller message %02x:%02x failed\n",
4353                         opcode, type);
4354                 return -EIO;
4355         }
4356
4357         dev_info(&pdev->dev, "controller message %02x:%02x succeeded\n",
4358                 opcode, type);
4359         return 0;
4360 }
4361
4362 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4363 #define cciss_noop(p) cciss_message(p, 3, 0)
4364
4365 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4366 {
4367 /* the #defines are stolen from drivers/pci/msi.h. */
4368 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4369 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4370
4371         int pos;
4372         u16 control = 0;
4373
4374         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4375         if (pos) {
4376                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4377                 if (control & PCI_MSI_FLAGS_ENABLE) {
4378                         dev_info(&pdev->dev, "resetting MSI\n");
4379                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4380                 }
4381         }
4382
4383         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4384         if (pos) {
4385                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4386                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4387                         dev_info(&pdev->dev, "resetting MSI-X\n");
4388                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4389                 }
4390         }
4391
4392         return 0;
4393 }
4394
4395 static int cciss_controller_hard_reset(struct pci_dev *pdev,
4396         void * __iomem vaddr, bool use_doorbell)
4397 {
4398         u16 pmcsr;
4399         int pos;
4400
4401         if (use_doorbell) {
4402                 /* For everything after the P600, the PCI power state method
4403                  * of resetting the controller doesn't work, so we have this
4404                  * other way using the doorbell register.
4405                  */
4406                 dev_info(&pdev->dev, "using doorbell to reset controller\n");
4407                 writel(DOORBELL_CTLR_RESET, vaddr + SA5_DOORBELL);
4408                 msleep(1000);
4409         } else { /* Try to do it the PCI power state way */
4410
4411                 /* Quoting from the Open CISS Specification: "The Power
4412                  * Management Control/Status Register (CSR) controls the power
4413                  * state of the device.  The normal operating state is D0,
4414                  * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4415                  * the controller, place the interface device in D3 then to D0,
4416                  * this causes a secondary PCI reset which will reset the
4417                  * controller." */
4418
4419                 pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4420                 if (pos == 0) {
4421                         dev_err(&pdev->dev,
4422                                 "cciss_controller_hard_reset: "
4423                                 "PCI PM not supported\n");
4424                         return -ENODEV;
4425                 }
4426                 dev_info(&pdev->dev, "using PCI PM to reset controller\n");
4427                 /* enter the D3hot power management state */
4428                 pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4429                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4430                 pmcsr |= PCI_D3hot;
4431                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4432
4433                 msleep(500);
4434
4435                 /* enter the D0 power management state */
4436                 pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4437                 pmcsr |= PCI_D0;
4438                 pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4439
4440                 msleep(500);
4441         }
4442         return 0;
4443 }
4444
4445 /* This does a hard reset of the controller using PCI power management
4446  * states or using the doorbell register. */
4447 static __devinit int cciss_kdump_hard_reset_controller(struct pci_dev *pdev)
4448 {
4449         u16 saved_config_space[32];
4450         u64 cfg_offset;
4451         u32 cfg_base_addr;
4452         u64 cfg_base_addr_index;
4453         void __iomem *vaddr;
4454         unsigned long paddr;
4455         u32 misc_fw_support, active_transport;
4456         int rc, i;
4457         CfgTable_struct __iomem *cfgtable;
4458         bool use_doorbell;
4459         u32 board_id;
4460
4461         /* For controllers as old a the p600, this is very nearly
4462          * the same thing as
4463          *
4464          * pci_save_state(pci_dev);
4465          * pci_set_power_state(pci_dev, PCI_D3hot);
4466          * pci_set_power_state(pci_dev, PCI_D0);
4467          * pci_restore_state(pci_dev);
4468          *
4469          * but we can't use these nice canned kernel routines on
4470          * kexec, because they also check the MSI/MSI-X state in PCI
4471          * configuration space and do the wrong thing when it is
4472          * set/cleared.  Also, the pci_save/restore_state functions
4473          * violate the ordering requirements for restoring the
4474          * configuration space from the CCISS document (see the
4475          * comment below).  So we roll our own ....
4476          *
4477          * For controllers newer than the P600, the pci power state
4478          * method of resetting doesn't work so we have another way
4479          * using the doorbell register.
4480          */
4481
4482         /* Exclude 640x boards.  These are two pci devices in one slot
4483          * which share a battery backed cache module.  One controls the
4484          * cache, the other accesses the cache through the one that controls
4485          * it.  If we reset the one controlling the cache, the other will
4486          * likely not be happy.  Just forbid resetting this conjoined mess.
4487          */
4488         cciss_lookup_board_id(pdev, &board_id);
4489         if (board_id == 0x409C0E11 || board_id == 0x409D0E11) {
4490                 dev_warn(&pdev->dev, "Cannot reset Smart Array 640x "
4491                                 "due to shared cache module.");
4492                 return -ENODEV;
4493         }
4494
4495         for (i = 0; i < 32; i++)
4496                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4497
4498         /* find the first memory BAR, so we can find the cfg table */
4499         rc = cciss_pci_find_memory_BAR(pdev, &paddr);
4500         if (rc)
4501                 return rc;
4502         vaddr = remap_pci_mem(paddr, 0x250);
4503         if (!vaddr)
4504                 return -ENOMEM;
4505
4506         /* find cfgtable in order to check if reset via doorbell is supported */
4507         rc = cciss_find_cfg_addrs(pdev, vaddr, &cfg_base_addr,
4508                                         &cfg_base_addr_index, &cfg_offset);
4509         if (rc)
4510                 goto unmap_vaddr;
4511         cfgtable = remap_pci_mem(pci_resource_start(pdev,
4512                        cfg_base_addr_index) + cfg_offset, sizeof(*cfgtable));
4513         if (!cfgtable) {
4514                 rc = -ENOMEM;
4515                 goto unmap_vaddr;
4516         }
4517
4518         /* If reset via doorbell register is supported, use that. */
4519         misc_fw_support = readl(&cfgtable->misc_fw_support);
4520         use_doorbell = misc_fw_support & MISC_FW_DOORBELL_RESET;
4521
4522         rc = cciss_controller_hard_reset(pdev, vaddr, use_doorbell);
4523         if (rc)
4524                 goto unmap_cfgtable;
4525
4526         /* Restore the PCI configuration space.  The Open CISS
4527          * Specification says, "Restore the PCI Configuration
4528          * Registers, offsets 00h through 60h. It is important to
4529          * restore the command register, 16-bits at offset 04h,
4530          * last. Do not restore the configuration status register,
4531          * 16-bits at offset 06h."  Note that the offset is 2*i.
4532          */
4533         for (i = 0; i < 32; i++) {
4534                 if (i == 2 || i == 3)
4535                         continue;
4536                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4537         }
4538         wmb();
4539         pci_write_config_word(pdev, 4, saved_config_space[2]);
4540
4541         /* Some devices (notably the HP Smart Array 5i Controller)
4542            need a little pause here */
4543         msleep(CCISS_POST_RESET_PAUSE_MSECS);
4544
4545         /* Controller should be in simple mode at this point.  If it's not,
4546          * It means we're on one of those controllers which doesn't support
4547          * the doorbell reset method and on which the PCI power management reset
4548          * method doesn't work (P800, for example.)
4549          * In those cases, don't try to proceed, as it generally doesn't work.
4550          */
4551         active_transport = readl(&cfgtable->TransportActive);
4552         if (active_transport & PERFORMANT_MODE) {
4553                 dev_warn(&pdev->dev, "Unable to successfully reset controller,"
4554                         " Ignoring controller.\n");
4555                 rc = -ENODEV;
4556         }
4557
4558 unmap_cfgtable:
4559         iounmap(cfgtable);
4560
4561 unmap_vaddr:
4562         iounmap(vaddr);
4563         return rc;
4564 }
4565
4566 static __devinit int cciss_init_reset_devices(struct pci_dev *pdev)
4567 {
4568         int rc, i;
4569
4570         if (!reset_devices)
4571                 return 0;
4572
4573         /* Reset the controller with a PCI power-cycle or via doorbell */
4574         rc = cciss_kdump_hard_reset_controller(pdev);
4575
4576         /* -ENOTSUPP here means we cannot reset the controller
4577          * but it's already (and still) up and running in
4578          * "performant mode".  Or, it might be 640x, which can't reset
4579          * due to concerns about shared bbwc between 6402/6404 pair.
4580          */
4581         if (rc == -ENOTSUPP)
4582                 return 0; /* just try to do the kdump anyhow. */
4583         if (rc)
4584                 return -ENODEV;
4585         if (cciss_reset_msi(pdev))
4586                 return -ENODEV;
4587
4588         /* Now try to get the controller to respond to a no-op */
4589         for (i = 0; i < CCISS_POST_RESET_NOOP_RETRIES; i++) {
4590                 if (cciss_noop(pdev) == 0)
4591                         break;
4592                 else
4593                         dev_warn(&pdev->dev, "no-op failed%s\n",
4594                                 (i < CCISS_POST_RESET_NOOP_RETRIES - 1 ?
4595                                         "; re-trying" : ""));
4596                 msleep(CCISS_POST_RESET_NOOP_INTERVAL_MSECS);
4597         }
4598         return 0;
4599 }
4600
4601 /*
4602  *  This is it.  Find all the controllers and register them.  I really hate
4603  *  stealing all these major device numbers.
4604  *  returns the number of block devices registered.
4605  */
4606 static int __devinit cciss_init_one(struct pci_dev *pdev,
4607                                     const struct pci_device_id *ent)
4608 {
4609         int i;
4610         int j = 0;
4611         int k = 0;
4612         int rc;
4613         int dac, return_code;
4614         InquiryData_struct *inq_buff;
4615         ctlr_info_t *h;
4616
4617         rc = cciss_init_reset_devices(pdev);
4618         if (rc)
4619                 return rc;
4620         i = alloc_cciss_hba(pdev);
4621         if (i < 0)
4622                 return -1;
4623
4624         h = hba[i];
4625         h->pdev = pdev;
4626         h->busy_initializing = 1;
4627         INIT_HLIST_HEAD(&h->cmpQ);
4628         INIT_HLIST_HEAD(&h->reqQ);
4629         mutex_init(&h->busy_shutting_down);
4630
4631         if (cciss_pci_init(h) != 0)
4632                 goto clean_no_release_regions;
4633
4634         sprintf(h->devname, "cciss%d", i);
4635         h->ctlr = i;
4636
4637         init_completion(&h->scan_wait);
4638
4639         if (cciss_create_hba_sysfs_entry(h))
4640                 goto clean0;
4641
4642         /* configure PCI DMA stuff */
4643         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4644                 dac = 1;
4645         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4646                 dac = 0;
4647         else {
4648                 dev_err(&h->pdev->dev, "no suitable DMA available\n");
4649                 goto clean1;
4650         }
4651
4652         /*
4653          * register with the major number, or get a dynamic major number
4654          * by passing 0 as argument.  This is done for greater than
4655          * 8 controller support.
4656          */
4657         if (i < MAX_CTLR_ORIG)
4658                 h->major = COMPAQ_CISS_MAJOR + i;
4659         rc = register_blkdev(h->major, h->devname);
4660         if (rc == -EBUSY || rc == -EINVAL) {
4661                 dev_err(&h->pdev->dev,
4662                        "Unable to get major number %d for %s "
4663                        "on hba %d\n", h->major, h->devname, i);
4664                 goto clean1;
4665         } else {
4666                 if (i >= MAX_CTLR_ORIG)
4667                         h->major = rc;
4668         }
4669
4670         /* make sure the board interrupts are off */
4671         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4672         if (h->msi_vector || h->msix_vector) {
4673                 if (request_irq(h->intr[PERF_MODE_INT],
4674                                 do_cciss_msix_intr,
4675                                 IRQF_DISABLED, h->devname, h)) {
4676                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4677                                h->intr[PERF_MODE_INT], h->devname);
4678                         goto clean2;
4679                 }
4680         } else {
4681                 if (request_irq(h->intr[PERF_MODE_INT], do_cciss_intx,
4682                                 IRQF_DISABLED, h->devname, h)) {
4683                         dev_err(&h->pdev->dev, "Unable to get irq %d for %s\n",
4684                                h->intr[PERF_MODE_INT], h->devname);
4685                         goto clean2;
4686                 }
4687         }
4688
4689         dev_info(&h->pdev->dev, "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4690                h->devname, pdev->device, pci_name(pdev),
4691                h->intr[PERF_MODE_INT], dac ? "" : " not");
4692
4693         h->cmd_pool_bits =
4694             kmalloc(DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4695                         * sizeof(unsigned long), GFP_KERNEL);
4696         h->cmd_pool = (CommandList_struct *)
4697             pci_alloc_consistent(h->pdev,
4698                     h->nr_cmds * sizeof(CommandList_struct),
4699                     &(h->cmd_pool_dhandle));
4700         h->errinfo_pool = (ErrorInfo_struct *)
4701             pci_alloc_consistent(h->pdev,
4702                     h->nr_cmds * sizeof(ErrorInfo_struct),
4703                     &(h->errinfo_pool_dhandle));
4704         if ((h->cmd_pool_bits == NULL)
4705             || (h->cmd_pool == NULL)
4706             || (h->errinfo_pool == NULL)) {
4707                 dev_err(&h->pdev->dev, "out of memory");
4708                 goto clean4;
4709         }
4710
4711         /* Need space for temp scatter list */
4712         h->scatter_list = kmalloc(h->max_commands *
4713                                                 sizeof(struct scatterlist *),
4714                                                 GFP_KERNEL);
4715         for (k = 0; k < h->nr_cmds; k++) {
4716                 h->scatter_list[k] = kmalloc(sizeof(struct scatterlist) *
4717                                                         h->maxsgentries,
4718                                                         GFP_KERNEL);
4719                 if (h->scatter_list[k] == NULL) {
4720                         dev_err(&h->pdev->dev,
4721                                 "could not allocate s/g lists\n");
4722                         goto clean4;
4723                 }
4724         }
4725         h->cmd_sg_list = cciss_allocate_sg_chain_blocks(h,
4726                 h->chainsize, h->nr_cmds);
4727         if (!h->cmd_sg_list && h->chainsize > 0)
4728                 goto clean4;
4729
4730         spin_lock_init(&h->lock);
4731
4732         /* Initialize the pdev driver private data.
4733            have it point to h.  */
4734         pci_set_drvdata(pdev, h);
4735         /* command and error info recs zeroed out before
4736            they are used */
4737         memset(h->cmd_pool_bits, 0,
4738                DIV_ROUND_UP(h->nr_cmds, BITS_PER_LONG)
4739                         * sizeof(unsigned long));
4740
4741         h->num_luns = 0;
4742         h->highest_lun = -1;
4743         for (j = 0; j < CISS_MAX_LUN; j++) {
4744                 h->drv[j] = NULL;
4745                 h->gendisk[j] = NULL;
4746         }
4747
4748         cciss_scsi_setup(h);
4749
4750         /* Turn the interrupts on so we can service requests */
4751         h->access.set_intr_mask(h, CCISS_INTR_ON);
4752
4753         /* Get the firmware version */
4754         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4755         if (inq_buff == NULL) {
4756                 dev_err(&h->pdev->dev, "out of memory\n");
4757                 goto clean4;
4758         }
4759
4760         return_code = sendcmd_withirq(h, CISS_INQUIRY, inq_buff,
4761                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4762         if (return_code == IO_OK) {
4763                 h->firm_ver[0] = inq_buff->data_byte[32];
4764                 h->firm_ver[1] = inq_buff->data_byte[33];
4765                 h->firm_ver[2] = inq_buff->data_byte[34];
4766                 h->firm_ver[3] = inq_buff->data_byte[35];
4767         } else {         /* send command failed */
4768                 dev_warn(&h->pdev->dev, "unable to determine firmware"
4769                         " version of controller\n");
4770         }
4771         kfree(inq_buff);
4772
4773         cciss_procinit(h);
4774
4775         h->cciss_max_sectors = 8192;
4776
4777         rebuild_lun_table(h, 1, 0);
4778         h->busy_initializing = 0;
4779         return 1;
4780
4781 clean4:
4782         kfree(h->cmd_pool_bits);
4783         /* Free up sg elements */
4784         for (k = 0; k < h->nr_cmds; k++)
4785                 kfree(h->scatter_list[k]);
4786         kfree(h->scatter_list);
4787         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4788         if (h->cmd_pool)
4789                 pci_free_consistent(h->pdev,
4790                                     h->nr_cmds * sizeof(CommandList_struct),
4791                                     h->cmd_pool, h->cmd_pool_dhandle);
4792         if (h->errinfo_pool)
4793                 pci_free_consistent(h->pdev,
4794                                     h->nr_cmds * sizeof(ErrorInfo_struct),
4795                                     h->errinfo_pool,
4796                                     h->errinfo_pool_dhandle);
4797         free_irq(h->intr[PERF_MODE_INT], h);
4798 clean2:
4799         unregister_blkdev(h->major, h->devname);
4800 clean1:
4801         cciss_destroy_hba_sysfs_entry(h);
4802 clean0:
4803         pci_release_regions(pdev);
4804 clean_no_release_regions:
4805         h->busy_initializing = 0;
4806
4807         /*
4808          * Deliberately omit pci_disable_device(): it does something nasty to
4809          * Smart Array controllers that pci_enable_device does not undo
4810          */
4811         pci_set_drvdata(pdev, NULL);
4812         free_hba(h);
4813         return -1;
4814 }
4815
4816 static void cciss_shutdown(struct pci_dev *pdev)
4817 {
4818         ctlr_info_t *h;
4819         char *flush_buf;
4820         int return_code;
4821
4822         h = pci_get_drvdata(pdev);
4823         flush_buf = kzalloc(4, GFP_KERNEL);
4824         if (!flush_buf) {
4825                 dev_warn(&h->pdev->dev, "cache not flushed, out of memory.\n");
4826                 return;
4827         }
4828         /* write all data in the battery backed cache to disk */
4829         memset(flush_buf, 0, 4);
4830         return_code = sendcmd_withirq(h, CCISS_CACHE_FLUSH, flush_buf,
4831                 4, 0, CTLR_LUNID, TYPE_CMD);
4832         kfree(flush_buf);
4833         if (return_code != IO_OK)
4834                 dev_warn(&h->pdev->dev, "Error flushing cache\n");
4835         h->access.set_intr_mask(h, CCISS_INTR_OFF);
4836         free_irq(h->intr[PERF_MODE_INT], h);
4837 }
4838
4839 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4840 {
4841         ctlr_info_t *h;
4842         int i, j;
4843
4844         if (pci_get_drvdata(pdev) == NULL) {
4845                 dev_err(&pdev->dev, "Unable to remove device\n");
4846                 return;
4847         }
4848
4849         h = pci_get_drvdata(pdev);
4850         i = h->ctlr;
4851         if (hba[i] == NULL) {
4852                 dev_err(&pdev->dev, "device appears to already be removed\n");
4853                 return;
4854         }
4855
4856         mutex_lock(&h->busy_shutting_down);
4857
4858         remove_from_scan_list(h);
4859         remove_proc_entry(h->devname, proc_cciss);
4860         unregister_blkdev(h->major, h->devname);
4861
4862         /* remove it from the disk list */
4863         for (j = 0; j < CISS_MAX_LUN; j++) {
4864                 struct gendisk *disk = h->gendisk[j];
4865                 if (disk) {
4866                         struct request_queue *q = disk->queue;
4867
4868                         if (disk->flags & GENHD_FL_UP) {
4869                                 cciss_destroy_ld_sysfs_entry(h, j, 1);
4870                                 del_gendisk(disk);
4871                         }
4872                         if (q)
4873                                 blk_cleanup_queue(q);
4874                 }
4875         }
4876
4877 #ifdef CONFIG_CISS_SCSI_TAPE
4878         cciss_unregister_scsi(h);       /* unhook from SCSI subsystem */
4879 #endif
4880
4881         cciss_shutdown(pdev);
4882
4883 #ifdef CONFIG_PCI_MSI
4884         if (h->msix_vector)
4885                 pci_disable_msix(h->pdev);
4886         else if (h->msi_vector)
4887                 pci_disable_msi(h->pdev);
4888 #endif                          /* CONFIG_PCI_MSI */
4889
4890         iounmap(h->transtable);
4891         iounmap(h->cfgtable);
4892         iounmap(h->vaddr);
4893
4894         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(CommandList_struct),
4895                             h->cmd_pool, h->cmd_pool_dhandle);
4896         pci_free_consistent(h->pdev, h->nr_cmds * sizeof(ErrorInfo_struct),
4897                             h->errinfo_pool, h->errinfo_pool_dhandle);
4898         kfree(h->cmd_pool_bits);
4899         /* Free up sg elements */
4900         for (j = 0; j < h->nr_cmds; j++)
4901                 kfree(h->scatter_list[j]);
4902         kfree(h->scatter_list);
4903         cciss_free_sg_chain_blocks(h->cmd_sg_list, h->nr_cmds);
4904         /*
4905          * Deliberately omit pci_disable_device(): it does something nasty to
4906          * Smart Array controllers that pci_enable_device does not undo
4907          */
4908         pci_release_regions(pdev);
4909         pci_set_drvdata(pdev, NULL);
4910         cciss_destroy_hba_sysfs_entry(h);
4911         mutex_unlock(&h->busy_shutting_down);
4912         free_hba(h);
4913 }
4914
4915 static struct pci_driver cciss_pci_driver = {
4916         .name = "cciss",
4917         .probe = cciss_init_one,
4918         .remove = __devexit_p(cciss_remove_one),
4919         .id_table = cciss_pci_device_id,        /* id_table */
4920         .shutdown = cciss_shutdown,
4921 };
4922
4923 /*
4924  *  This is it.  Register the PCI driver information for the cards we control
4925  *  the OS will call our registered routines when it finds one of our cards.
4926  */
4927 static int __init cciss_init(void)
4928 {
4929         int err;
4930
4931         /*
4932          * The hardware requires that commands are aligned on a 64-bit
4933          * boundary. Given that we use pci_alloc_consistent() to allocate an
4934          * array of them, the size must be a multiple of 8 bytes.
4935          */
4936         BUILD_BUG_ON(sizeof(CommandList_struct) % COMMANDLIST_ALIGNMENT);
4937         printk(KERN_INFO DRIVER_NAME "\n");
4938
4939         err = bus_register(&cciss_bus_type);
4940         if (err)
4941                 return err;
4942
4943         /* Start the scan thread */
4944         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4945         if (IS_ERR(cciss_scan_thread)) {
4946                 err = PTR_ERR(cciss_scan_thread);
4947                 goto err_bus_unregister;
4948         }
4949
4950         /* Register for our PCI devices */
4951         err = pci_register_driver(&cciss_pci_driver);
4952         if (err)
4953                 goto err_thread_stop;
4954
4955         return err;
4956
4957 err_thread_stop:
4958         kthread_stop(cciss_scan_thread);
4959 err_bus_unregister:
4960         bus_unregister(&cciss_bus_type);
4961
4962         return err;
4963 }
4964
4965 static void __exit cciss_cleanup(void)
4966 {
4967         int i;
4968
4969         pci_unregister_driver(&cciss_pci_driver);
4970         /* double check that all controller entrys have been removed */
4971         for (i = 0; i < MAX_CTLR; i++) {
4972                 if (hba[i] != NULL) {
4973                         dev_warn(&hba[i]->pdev->dev,
4974                                 "had to remove controller\n");
4975                         cciss_remove_one(hba[i]->pdev);
4976                 }
4977         }
4978         kthread_stop(cciss_scan_thread);
4979         remove_proc_entry("driver/cciss", NULL);
4980         bus_unregister(&cciss_bus_type);
4981 }
4982
4983 module_init(cciss_init);
4984 module_exit(cciss_cleanup);