SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 3.6.20)"
  60 #define DRIVER_VERSION CCISS_DRIVER_VERSION(3, 6, 20)
  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 SA5i SA5i+ SA532 SA5300 SA5312 SA641 SA642 SA6400"
  66                         " SA6i P600 P800 P400 P400i E200 E200i E500 P700m"
  67                         " Smart Array G2 Series SAS/SATA Controllers");
  68 MODULE_VERSION("3.6.20");
  69 MODULE_LICENSE("GPL");
  70
  71 #include "cciss_cmd.h"
  72 #include "cciss.h"
  73 #include <linux/cciss_ioctl.h>
  74
  75 /* define the PCI info for the cards we can control */
  76 static const struct pci_device_id cciss_pci_device_id[] = {
  77         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISS,  0x0E11, 0x4070},
  78         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4080},
  79         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4082},
  80         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSB, 0x0E11, 0x4083},
  81         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x4091},
  82         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409A},
  83         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409B},
  84         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409C},
  85         {PCI_VENDOR_ID_COMPAQ, PCI_DEVICE_ID_COMPAQ_CISSC, 0x0E11, 0x409D},
  86         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSA,     0x103C, 0x3225},
  87         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3223},
  88         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3234},
  89         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3235},
  90         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3211},
  91         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3212},
  92         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3213},
  93         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3214},
  94         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSD,     0x103C, 0x3215},
  95         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x3237},
  96         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSC,     0x103C, 0x323D},
  97         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3241},
  98         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3243},
  99         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3245},
 100         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3247},
 101         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x3249},
 102         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324A},
 103         {PCI_VENDOR_ID_HP,     PCI_DEVICE_ID_HP_CISSE,     0x103C, 0x324B},
 104         {PCI_VENDOR_ID_HP,     PCI_ANY_ID,      PCI_ANY_ID, PCI_ANY_ID,
 105                 PCI_CLASS_STORAGE_RAID << 8, 0xffff << 8, 0},
 106         {0,}
 107 };
 108
 109 MODULE_DEVICE_TABLE(pci, cciss_pci_device_id);
 110
 111 /*  board_id = Subsystem Device ID & Vendor ID
 112  *  product = Marketing Name for the board
 113  *  access = Address of the struct of function pointers
 114  */
 115 static struct board_type products[] = {
 116         {0x40700E11, "Smart Array 5300", &SA5_access},
 117         {0x40800E11, "Smart Array 5i", &SA5B_access},
 118         {0x40820E11, "Smart Array 532", &SA5B_access},
 119         {0x40830E11, "Smart Array 5312", &SA5B_access},
 120         {0x409A0E11, "Smart Array 641", &SA5_access},
 121         {0x409B0E11, "Smart Array 642", &SA5_access},
 122         {0x409C0E11, "Smart Array 6400", &SA5_access},
 123         {0x409D0E11, "Smart Array 6400 EM", &SA5_access},
 124         {0x40910E11, "Smart Array 6i", &SA5_access},
 125         {0x3225103C, "Smart Array P600", &SA5_access},
 126         {0x3223103C, "Smart Array P800", &SA5_access},
 127         {0x3234103C, "Smart Array P400", &SA5_access},
 128         {0x3235103C, "Smart Array P400i", &SA5_access},
 129         {0x3211103C, "Smart Array E200i", &SA5_access},
 130         {0x3212103C, "Smart Array E200", &SA5_access},
 131         {0x3213103C, "Smart Array E200i", &SA5_access},
 132         {0x3214103C, "Smart Array E200i", &SA5_access},
 133         {0x3215103C, "Smart Array E200i", &SA5_access},
 134         {0x3237103C, "Smart Array E500", &SA5_access},
 135         {0x323D103C, "Smart Array P700m", &SA5_access},
 136         {0x3241103C, "Smart Array P212", &SA5_access},
 137         {0x3243103C, "Smart Array P410", &SA5_access},
 138         {0x3245103C, "Smart Array P410i", &SA5_access},
 139         {0x3247103C, "Smart Array P411", &SA5_access},
 140         {0x3249103C, "Smart Array P812", &SA5_access},
 141         {0x324A103C, "Smart Array P712m", &SA5_access},
 142         {0x324B103C, "Smart Array P711m", &SA5_access},
 143         {0xFFFF103C, "Unknown Smart Array", &SA5_access},
 144 };
 145
 146 /* How long to wait (in milliseconds) for board to go into simple mode */
 147 #define MAX_CONFIG_WAIT 30000
 148 #define MAX_IOCTL_CONFIG_WAIT 1000
 149
 150 /*define how many times we will try a command because of bus resets */
 151 #define MAX_CMD_RETRIES 3
 152
 153 #define MAX_CTLR        32
 154
 155 /* Originally cciss driver only supports 8 major numbers */
 156 #define MAX_CTLR_ORIG   8
 157
 158 static ctlr_info_t *hba[MAX_CTLR];
 159
 160 static struct task_struct *cciss_scan_thread;
 161 static DEFINE_MUTEX(scan_mutex);
 162 static LIST_HEAD(scan_q);
 163
 164 static void do_cciss_request(struct request_queue *q);
 165 static irqreturn_t do_cciss_intr(int irq, void *dev_id);
 166 static int cciss_open(struct block_device *bdev, fmode_t mode);
 167 static int cciss_release(struct gendisk *disk, fmode_t mode);
 168 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
 169                        unsigned int cmd, unsigned long arg);
 170 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 171
 172 static int cciss_revalidate(struct gendisk *disk);
 173 static int rebuild_lun_table(ctlr_info_t *h, int first_time, int via_ioctl);
 174 static int deregister_disk(ctlr_info_t *h, int drv_index,
 175                            int clear_all, int via_ioctl);
 176
 177 static void cciss_read_capacity(int ctlr, int logvol, int withirq,
 178                         sector_t *total_size, unsigned int *block_size);
 179 static void cciss_read_capacity_16(int ctlr, int logvol, int withirq,
 180                         sector_t *total_size, unsigned int *block_size);
 181 static void cciss_geometry_inquiry(int ctlr, int logvol,
 182                         int withirq, sector_t total_size,
 183                         unsigned int block_size, InquiryData_struct *inq_buff,
 184                                    drive_info_struct *drv);
 185 static void __devinit cciss_interrupt_mode(ctlr_info_t *, struct pci_dev *,
 186                                            __u32);
 187 static void start_io(ctlr_info_t *h);
 188 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
 189                    __u8 page_code, unsigned char *scsi3addr, int cmd_type);
 190 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
 191                         __u8 page_code, unsigned char scsi3addr[],
 192                         int cmd_type);
 193 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
 194         int attempt_retry);
 195 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c);
 196
 197 static void fail_all_cmds(unsigned long ctlr);
 198 static int add_to_scan_list(struct ctlr_info *h);
 199 static int scan_thread(void *data);
 200 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c);
 201 static void cciss_hba_release(struct device *dev);
 202 static void cciss_device_release(struct device *dev);
 203 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index);
 204
 205 #ifdef CONFIG_PROC_FS
 206 static void cciss_procinit(int i);
 207 #else
 208 static void cciss_procinit(int i)
 209 {
 210 }
 211 #endif                          /* CONFIG_PROC_FS */
 212
 213 #ifdef CONFIG_COMPAT
 214 static int cciss_compat_ioctl(struct block_device *, fmode_t,
 215                               unsigned, unsigned long);
 216 #endif
 217
 218 static const struct block_device_operations cciss_fops = {
 219         .owner = THIS_MODULE,
 220         .open = cciss_open,
 221         .release = cciss_release,
 222         .locked_ioctl = cciss_ioctl,
 223         .getgeo = cciss_getgeo,
 224 #ifdef CONFIG_COMPAT
 225         .compat_ioctl = cciss_compat_ioctl,
 226 #endif
 227         .revalidate_disk = cciss_revalidate,
 228 };
 229
 230 /*
 231  * Enqueuing and dequeuing functions for cmdlists.
 232  */
 233 static inline void addQ(struct hlist_head *list, CommandList_struct *c)
 234 {
 235         hlist_add_head(&c->list, list);
 236 }
 237
 238 static inline void removeQ(CommandList_struct *c)
 239 {
 240         /*
 241          * After kexec/dump some commands might still
 242          * be in flight, which the firmware will try
 243          * to complete. Resetting the firmware doesn't work
 244          * with old fw revisions, so we have to mark
 245          * them off as 'stale' to prevent the driver from
 246          * falling over.
 247          */
 248         if (WARN_ON(hlist_unhashed(&c->list))) {
 249                 c->cmd_type = CMD_MSG_STALE;
 250                 return;
 251         }
 252
 253         hlist_del_init(&c->list);
 254 }
 255
 256 #include "cciss_scsi.c"         /* For SCSI tape support */
 257
 258 #define RAID_UNKNOWN 6
 259
 260 #ifdef CONFIG_PROC_FS
 261
 262 /*
 263  * Report information about this controller.
 264  */
 265 #define ENG_GIG 1000000000
 266 #define ENG_GIG_FACTOR (ENG_GIG/512)
 267 #define ENGAGE_SCSI     "engage scsi"
 268 static const char *raid_label[] = { "0", "4", "1(1+0)", "5", "5+1", "ADG",
 269         "UNKNOWN"
 270 };
 271
 272 static struct proc_dir_entry *proc_cciss;
 273
 274 static void cciss_seq_show_header(struct seq_file *seq)
 275 {
 276         ctlr_info_t *h = seq->private;
 277
 278         seq_printf(seq, "%s: HP %s Controller\n"
 279                 "Board ID: 0x%08lx\n"
 280                 "Firmware Version: %c%c%c%c\n"
 281                 "IRQ: %d\n"
 282                 "Logical drives: %d\n"
 283                 "Current Q depth: %d\n"
 284                 "Current # commands on controller: %d\n"
 285                 "Max Q depth since init: %d\n"
 286                 "Max # commands on controller since init: %d\n"
 287                 "Max SG entries since init: %d\n",
 288                 h->devname,
 289                 h->product_name,
 290                 (unsigned long)h->board_id,
 291                 h->firm_ver[0], h->firm_ver[1], h->firm_ver[2],
 292                 h->firm_ver[3], (unsigned int)h->intr[SIMPLE_MODE_INT],
 293                 h->num_luns,
 294                 h->Qdepth, h->commands_outstanding,
 295                 h->maxQsinceinit, h->max_outstanding, h->maxSG);
 296
 297 #ifdef CONFIG_CISS_SCSI_TAPE
 298         cciss_seq_tape_report(seq, h->ctlr);
 299 #endif /* CONFIG_CISS_SCSI_TAPE */
 300 }
 301
 302 static void *cciss_seq_start(struct seq_file *seq, loff_t *pos)
 303 {
 304         ctlr_info_t *h = seq->private;
 305         unsigned ctlr = h->ctlr;
 306         unsigned long flags;
 307
 308         /* prevent displaying bogus info during configuration
 309          * or deconfiguration of a logical volume
 310          */
 311         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
 312         if (h->busy_configuring) {
 313                 spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 314                 return ERR_PTR(-EBUSY);
 315         }
 316         h->busy_configuring = 1;
 317         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
 318
 319         if (*pos == 0)
 320                 cciss_seq_show_header(seq);
 321
 322         return pos;
 323 }
 324
 325 static int cciss_seq_show(struct seq_file *seq, void *v)
 326 {
 327         sector_t vol_sz, vol_sz_frac;
 328         ctlr_info_t *h = seq->private;
 329         unsigned ctlr = h->ctlr;
 330         loff_t *pos = v;
 331         drive_info_struct *drv = &h->drv[*pos];
 332
 333         if (*pos > h->highest_lun)
 334                 return 0;
 335
 336         if (drv->heads == 0)
 337                 return 0;
 338
 339         vol_sz = drv->nr_blocks;
 340         vol_sz_frac = sector_div(vol_sz, ENG_GIG_FACTOR);
 341         vol_sz_frac *= 100;
 342         sector_div(vol_sz_frac, ENG_GIG_FACTOR);
 343
 344         if (drv->raid_level > 5)
 345                 drv->raid_level = RAID_UNKNOWN;
 346         seq_printf(seq, "cciss/c%dd%d:"
 347                         "\t%4u.%02uGB\tRAID %s\n",
 348                         ctlr, (int) *pos, (int)vol_sz, (int)vol_sz_frac,
 349                         raid_label[drv->raid_level]);
 350         return 0;
 351 }
 352
 353 static void *cciss_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 354 {
 355         ctlr_info_t *h = seq->private;
 356
 357         if (*pos > h->highest_lun)
 358                 return NULL;
 359         *pos += 1;
 360
 361         return pos;
 362 }
 363
 364 static void cciss_seq_stop(struct seq_file *seq, void *v)
 365 {
 366         ctlr_info_t *h = seq->private;
 367
 368         /* Only reset h->busy_configuring if we succeeded in setting
 369          * it during cciss_seq_start. */
 370         if (v == ERR_PTR(-EBUSY))
 371                 return;
 372
 373         h->busy_configuring = 0;
 374 }
 375
 376 static const struct seq_operations cciss_seq_ops = {
 377         .start = cciss_seq_start,
 378         .show  = cciss_seq_show,
 379         .next  = cciss_seq_next,
 380         .stop  = cciss_seq_stop,
 381 };
 382
 383 static int cciss_seq_open(struct inode *inode, struct file *file)
 384 {
 385         int ret = seq_open(file, &cciss_seq_ops);
 386         struct seq_file *seq = file->private_data;
 387
 388         if (!ret)
 389                 seq->private = PDE(inode)->data;
 390
 391         return ret;
 392 }
 393
 394 static ssize_t
 395 cciss_proc_write(struct file *file, const char __user *buf,
 396                  size_t length, loff_t *ppos)
 397 {
 398         int err;
 399         char *buffer;
 400
 401 #ifndef CONFIG_CISS_SCSI_TAPE
 402         return -EINVAL;
 403 #endif
 404
 405         if (!buf || length > PAGE_SIZE - 1)
 406                 return -EINVAL;
 407
 408         buffer = (char *)__get_free_page(GFP_KERNEL);
 409         if (!buffer)
 410                 return -ENOMEM;
 411
 412         err = -EFAULT;
 413         if (copy_from_user(buffer, buf, length))
 414                 goto out;
 415         buffer[length] = '\0';
 416
 417 #ifdef CONFIG_CISS_SCSI_TAPE
 418         if (strncmp(ENGAGE_SCSI, buffer, sizeof ENGAGE_SCSI - 1) == 0) {
 419                 struct seq_file *seq = file->private_data;
 420                 ctlr_info_t *h = seq->private;
 421                 int rc;
 422
 423                 rc = cciss_engage_scsi(h->ctlr);
 424                 if (rc != 0)
 425                         err = -rc;
 426                 else
 427                         err = length;
 428         } else
 429 #endif /* CONFIG_CISS_SCSI_TAPE */
 430                 err = -EINVAL;
 431         /* might be nice to have "disengage" too, but it's not
 432            safely possible. (only 1 module use count, lock issues.) */
 433
 434 out:
 435         free_page((unsigned long)buffer);
 436         return err;
 437 }
 438
 439 static struct file_operations cciss_proc_fops = {
 440         .owner   = THIS_MODULE,
 441         .open    = cciss_seq_open,
 442         .read    = seq_read,
 443         .llseek  = seq_lseek,
 444         .release = seq_release,
 445         .write   = cciss_proc_write,
 446 };
 447
 448 static void __devinit cciss_procinit(int i)
 449 {
 450         struct proc_dir_entry *pde;
 451
 452         if (proc_cciss == NULL)
 453                 proc_cciss = proc_mkdir("driver/cciss", NULL);
 454         if (!proc_cciss)
 455                 return;
 456         pde = proc_create_data(hba[i]->devname, S_IWUSR | S_IRUSR | S_IRGRP |
 457                                         S_IROTH, proc_cciss,
 458                                         &cciss_proc_fops, hba[i]);
 459 }
 460 #endif                          /* CONFIG_PROC_FS */
 461
 462 #define MAX_PRODUCT_NAME_LEN 19
 463
 464 #define to_hba(n) container_of(n, struct ctlr_info, dev)
 465
 466 static ssize_t host_store_rescan(struct device *dev,
 467                                  struct device_attribute *attr,
 468                                  const char *buf, size_t count)
 469 {
 470         struct ctlr_info *h = to_hba(dev);
 471
 472         add_to_scan_list(h);
 473         wake_up_process(cciss_scan_thread);
 474         wait_for_completion_interruptible(&h->scan_wait);
 475
 476         return count;
 477 }
 478 DEVICE_ATTR(rescan, S_IWUSR, NULL, host_store_rescan);
 479
 480 static ssize_t dev_show_unique_id(struct device *dev,
 481                                  struct device_attribute *attr,
 482                                  char *buf)
 483 {
 484         drive_info_struct *drv = dev_get_drvdata(dev);
 485         struct ctlr_info *h = to_hba(drv->dev->parent);
 486         __u8 sn[16];
 487         unsigned long flags;
 488         int ret = 0;
 489
 490         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 491         if (h->busy_configuring)
 492                 ret = -EBUSY;
 493         else
 494                 memcpy(sn, drv->serial_no, sizeof(sn));
 495         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 496
 497         if (ret)
 498                 return ret;
 499         else
 500                 return snprintf(buf, 16 * 2 + 2,
 501                                 "%02X%02X%02X%02X%02X%02X%02X%02X"
 502                                 "%02X%02X%02X%02X%02X%02X%02X%02X\n",
 503                                 sn[0], sn[1], sn[2], sn[3],
 504                                 sn[4], sn[5], sn[6], sn[7],
 505                                 sn[8], sn[9], sn[10], sn[11],
 506                                 sn[12], sn[13], sn[14], sn[15]);
 507 }
 508 DEVICE_ATTR(unique_id, S_IRUGO, dev_show_unique_id, NULL);
 509
 510 static ssize_t dev_show_vendor(struct device *dev,
 511                                struct device_attribute *attr,
 512                                char *buf)
 513 {
 514         drive_info_struct *drv = dev_get_drvdata(dev);
 515         struct ctlr_info *h = to_hba(drv->dev->parent);
 516         char vendor[VENDOR_LEN + 1];
 517         unsigned long flags;
 518         int ret = 0;
 519
 520         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 521         if (h->busy_configuring)
 522                 ret = -EBUSY;
 523         else
 524                 memcpy(vendor, drv->vendor, VENDOR_LEN + 1);
 525         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 526
 527         if (ret)
 528                 return ret;
 529         else
 530                 return snprintf(buf, sizeof(vendor) + 1, "%s\n", drv->vendor);
 531 }
 532 DEVICE_ATTR(vendor, S_IRUGO, dev_show_vendor, NULL);
 533
 534 static ssize_t dev_show_model(struct device *dev,
 535                               struct device_attribute *attr,
 536                               char *buf)
 537 {
 538         drive_info_struct *drv = dev_get_drvdata(dev);
 539         struct ctlr_info *h = to_hba(drv->dev->parent);
 540         char model[MODEL_LEN + 1];
 541         unsigned long flags;
 542         int ret = 0;
 543
 544         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 545         if (h->busy_configuring)
 546                 ret = -EBUSY;
 547         else
 548                 memcpy(model, drv->model, MODEL_LEN + 1);
 549         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 550
 551         if (ret)
 552                 return ret;
 553         else
 554                 return snprintf(buf, sizeof(model) + 1, "%s\n", drv->model);
 555 }
 556 DEVICE_ATTR(model, S_IRUGO, dev_show_model, NULL);
 557
 558 static ssize_t dev_show_rev(struct device *dev,
 559                             struct device_attribute *attr,
 560                             char *buf)
 561 {
 562         drive_info_struct *drv = dev_get_drvdata(dev);
 563         struct ctlr_info *h = to_hba(drv->dev->parent);
 564         char rev[REV_LEN + 1];
 565         unsigned long flags;
 566         int ret = 0;
 567
 568         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 569         if (h->busy_configuring)
 570                 ret = -EBUSY;
 571         else
 572                 memcpy(rev, drv->rev, REV_LEN + 1);
 573         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 574
 575         if (ret)
 576                 return ret;
 577         else
 578                 return snprintf(buf, sizeof(rev) + 1, "%s\n", drv->rev);
 579 }
 580 DEVICE_ATTR(rev, S_IRUGO, dev_show_rev, NULL);
 581
 582 static ssize_t cciss_show_lunid(struct device *dev,
 583                                 struct device_attribute *attr, char *buf)
 584 {
 585         drive_info_struct *drv = dev_get_drvdata(dev);
 586         struct ctlr_info *h = to_hba(drv->dev->parent);
 587         unsigned long flags;
 588         unsigned char lunid[8];
 589
 590         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
 591         if (h->busy_configuring) {
 592                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 593                 return -EBUSY;
 594         }
 595         if (!drv->heads) {
 596                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 597                 return -ENOTTY;
 598         }
 599         memcpy(lunid, drv->LunID, sizeof(lunid));
 600         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
 601         return snprintf(buf, 20, "0x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 602                 lunid[0], lunid[1], lunid[2], lunid[3],
 603                 lunid[4], lunid[5], lunid[6], lunid[7]);
 604 }
 605 DEVICE_ATTR(lunid, S_IRUGO, cciss_show_lunid, NULL);
 606
 607 static struct attribute *cciss_host_attrs[] = {
 608         &dev_attr_rescan.attr,
 609         NULL
 610 };
 611
 612 static struct attribute_group cciss_host_attr_group = {
 613         .attrs = cciss_host_attrs,
 614 };
 615
 616 static struct attribute_group *cciss_host_attr_groups[] = {
 617         &cciss_host_attr_group,
 618         NULL
 619 };
 620
 621 static struct device_type cciss_host_type = {
 622         .name           = "cciss_host",
 623         .groups         = cciss_host_attr_groups,
 624         .release        = cciss_hba_release,
 625 };
 626
 627 static struct attribute *cciss_dev_attrs[] = {
 628         &dev_attr_unique_id.attr,
 629         &dev_attr_model.attr,
 630         &dev_attr_vendor.attr,
 631         &dev_attr_rev.attr,
 632         &dev_attr_lunid.attr,
 633         NULL
 634 };
 635
 636 static struct attribute_group cciss_dev_attr_group = {
 637         .attrs = cciss_dev_attrs,
 638 };
 639
 640 static const struct attribute_group *cciss_dev_attr_groups[] = {
 641         &cciss_dev_attr_group,
 642         NULL
 643 };
 644
 645 static struct device_type cciss_dev_type = {
 646         .name           = "cciss_device",
 647         .groups         = cciss_dev_attr_groups,
 648         .release        = cciss_device_release,
 649 };
 650
 651 static struct bus_type cciss_bus_type = {
 652         .name           = "cciss",
 653 };
 654
 655 /*
 656  * cciss_hba_release is called when the reference count
 657  * of h->dev goes to zero.
 658  */
 659 static void cciss_hba_release(struct device *dev)
 660 {
 661         /*
 662          * nothing to do, but need this to avoid a warning
 663          * about not having a release handler from lib/kref.c.
 664          */
 665 }
 666
 667 /*
 668  * Initialize sysfs entry for each controller.  This sets up and registers
 669  * the 'cciss#' directory for each individual controller under
 670  * /sys/bus/pci/devices/<dev>/.
 671  */
 672 static int cciss_create_hba_sysfs_entry(struct ctlr_info *h)
 673 {
 674         device_initialize(&h->dev);
 675         h->dev.type = &cciss_host_type;
 676         h->dev.bus = &cciss_bus_type;
 677         dev_set_name(&h->dev, "%s", h->devname);
 678         h->dev.parent = &h->pdev->dev;
 679
 680         return device_add(&h->dev);
 681 }
 682
 683 /*
 684  * Remove sysfs entries for an hba.
 685  */
 686 static void cciss_destroy_hba_sysfs_entry(struct ctlr_info *h)
 687 {
 688         device_del(&h->dev);
 689         put_device(&h->dev); /* final put. */
 690 }
 691
 692 /* cciss_device_release is called when the reference count
 693  * of h->drv[x].dev goes to zero.
 694  */
 695 static void cciss_device_release(struct device *dev)
 696 {
 697         kfree(dev);
 698 }
 699
 700 /*
 701  * Initialize sysfs for each logical drive.  This sets up and registers
 702  * the 'c#d#' directory for each individual logical drive under
 703  * /sys/bus/pci/devices/<dev/ccis#/. We also create a link from
 704  * /sys/block/cciss!c#d# to this entry.
 705  */
 706 static long cciss_create_ld_sysfs_entry(struct ctlr_info *h,
 707                                        int drv_index)
 708 {
 709         struct device *dev;
 710
 711         /* Special case for c*d0, we only create it once. */
 712         if (drv_index == 0 && h->drv[drv_index].dev != NULL)
 713                 return 0;
 714
 715         dev = kzalloc(sizeof(*dev), GFP_KERNEL);
 716         if (!dev)
 717                 return -ENOMEM;
 718         device_initialize(dev);
 719         dev->type = &cciss_dev_type;
 720         dev->bus = &cciss_bus_type;
 721         dev_set_name(dev, "c%dd%d", h->ctlr, drv_index);
 722         dev->parent = &h->dev;
 723         h->drv[drv_index].dev = dev;
 724         dev_set_drvdata(dev, &h->drv[drv_index]);
 725         return device_add(dev);
 726 }
 727
 728 /*
 729  * Remove sysfs entries for a logical drive.
 730  */
 731 static void cciss_destroy_ld_sysfs_entry(struct ctlr_info *h, int drv_index,
 732         int ctlr_exiting)
 733 {
 734         struct device *dev = h->drv[drv_index].dev;
 735
 736         /* special case for c*d0, we only destroy it on controller exit */
 737         if (drv_index == 0 && !ctlr_exiting)
 738                 return;
 739
 740         device_del(dev);
 741         put_device(dev); /* the "final" put. */
 742         h->drv[drv_index].dev = NULL;
 743 }
 744
 745 /*
 746  * For operations that cannot sleep, a command block is allocated at init,
 747  * and managed by cmd_alloc() and cmd_free() using a simple bitmap to track
 748  * which ones are free or in use.  For operations that can wait for kmalloc
 749  * to possible sleep, this routine can be called with get_from_pool set to 0.
 750  * cmd_free() MUST be called with a got_from_pool set to 0 if cmd_alloc was.
 751  */
 752 static CommandList_struct *cmd_alloc(ctlr_info_t *h, int get_from_pool)
 753 {
 754         CommandList_struct *c;
 755         int i;
 756         u64bit temp64;
 757         dma_addr_t cmd_dma_handle, err_dma_handle;
 758
 759         if (!get_from_pool) {
 760                 c = (CommandList_struct *) pci_alloc_consistent(h->pdev,
 761                         sizeof(CommandList_struct), &cmd_dma_handle);
 762                 if (c == NULL)
 763                         return NULL;
 764                 memset(c, 0, sizeof(CommandList_struct));
 765
 766                 c->cmdindex = -1;
 767
 768                 c->err_info = (ErrorInfo_struct *)
 769                     pci_alloc_consistent(h->pdev, sizeof(ErrorInfo_struct),
 770                             &err_dma_handle);
 771
 772                 if (c->err_info == NULL) {
 773                         pci_free_consistent(h->pdev,
 774                                 sizeof(CommandList_struct), c, cmd_dma_handle);
 775                         return NULL;
 776                 }
 777                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 778         } else {                /* get it out of the controllers pool */
 779
 780                 do {
 781                         i = find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds);
 782                         if (i == h->nr_cmds)
 783                                 return NULL;
 784                 } while (test_and_set_bit
 785                          (i & (BITS_PER_LONG - 1),
 786                           h->cmd_pool_bits + (i / BITS_PER_LONG)) != 0);
 787 #ifdef CCISS_DEBUG
 788                 printk(KERN_DEBUG "cciss: using command buffer %d\n", i);
 789 #endif
 790                 c = h->cmd_pool + i;
 791                 memset(c, 0, sizeof(CommandList_struct));
 792                 cmd_dma_handle = h->cmd_pool_dhandle
 793                     + i * sizeof(CommandList_struct);
 794                 c->err_info = h->errinfo_pool + i;
 795                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
 796                 err_dma_handle = h->errinfo_pool_dhandle
 797                     + i * sizeof(ErrorInfo_struct);
 798                 h->nr_allocs++;
 799
 800                 c->cmdindex = i;
 801         }
 802
 803         INIT_HLIST_NODE(&c->list);
 804         c->busaddr = (__u32) cmd_dma_handle;
 805         temp64.val = (__u64) err_dma_handle;
 806         c->ErrDesc.Addr.lower = temp64.val32.lower;
 807         c->ErrDesc.Addr.upper = temp64.val32.upper;
 808         c->ErrDesc.Len = sizeof(ErrorInfo_struct);
 809
 810         c->ctlr = h->ctlr;
 811         return c;
 812 }
 813
 814 /*
 815  * Frees a command block that was previously allocated with cmd_alloc().
 816  */
 817 static void cmd_free(ctlr_info_t *h, CommandList_struct *c, int got_from_pool)
 818 {
 819         int i;
 820         u64bit temp64;
 821
 822         if (!got_from_pool) {
 823                 temp64.val32.lower = c->ErrDesc.Addr.lower;
 824                 temp64.val32.upper = c->ErrDesc.Addr.upper;
 825                 pci_free_consistent(h->pdev, sizeof(ErrorInfo_struct),
 826                                     c->err_info, (dma_addr_t) temp64.val);
 827                 pci_free_consistent(h->pdev, sizeof(CommandList_struct),
 828                                     c, (dma_addr_t) c->busaddr);
 829         } else {
 830                 i = c - h->cmd_pool;
 831                 clear_bit(i & (BITS_PER_LONG - 1),
 832                           h->cmd_pool_bits + (i / BITS_PER_LONG));
 833                 h->nr_frees++;
 834         }
 835 }
 836
 837 static inline ctlr_info_t *get_host(struct gendisk *disk)
 838 {
 839         return disk->queue->queuedata;
 840 }
 841
 842 static inline drive_info_struct *get_drv(struct gendisk *disk)
 843 {
 844         return disk->private_data;
 845 }
 846
 847 /*
 848  * Open.  Make sure the device is really there.
 849  */
 850 static int cciss_open(struct block_device *bdev, fmode_t mode)
 851 {
 852         ctlr_info_t *host = get_host(bdev->bd_disk);
 853         drive_info_struct *drv = get_drv(bdev->bd_disk);
 854
 855 #ifdef CCISS_DEBUG
 856         printk(KERN_DEBUG "cciss_open %s\n", bdev->bd_disk->disk_name);
 857 #endif                          /* CCISS_DEBUG */
 858
 859         if (drv->busy_configuring)
 860                 return -EBUSY;
 861         /*
 862          * Root is allowed to open raw volume zero even if it's not configured
 863          * so array config can still work. Root is also allowed to open any
 864          * volume that has a LUN ID, so it can issue IOCTL to reread the
 865          * disk information.  I don't think I really like this
 866          * but I'm already using way to many device nodes to claim another one
 867          * for "raw controller".
 868          */
 869         if (drv->heads == 0) {
 870                 if (MINOR(bdev->bd_dev) != 0) { /* not node 0? */
 871                         /* if not node 0 make sure it is a partition = 0 */
 872                         if (MINOR(bdev->bd_dev) & 0x0f) {
 873                                 return -ENXIO;
 874                                 /* if it is, make sure we have a LUN ID */
 875                         } else if (memcmp(drv->LunID, CTLR_LUNID,
 876                                 sizeof(drv->LunID))) {
 877                                 return -ENXIO;
 878                         }
 879                 }
 880                 if (!capable(CAP_SYS_ADMIN))
 881                         return -EPERM;
 882         }
 883         drv->usage_count++;
 884         host->usage_count++;
 885         return 0;
 886 }
 887
 888 /*
 889  * Close.  Sync first.
 890  */
 891 static int cciss_release(struct gendisk *disk, fmode_t mode)
 892 {
 893         ctlr_info_t *host = get_host(disk);
 894         drive_info_struct *drv = get_drv(disk);
 895
 896 #ifdef CCISS_DEBUG
 897         printk(KERN_DEBUG "cciss_release %s\n", disk->disk_name);
 898 #endif                          /* CCISS_DEBUG */
 899
 900         drv->usage_count--;
 901         host->usage_count--;
 902         return 0;
 903 }
 904
 905 #ifdef CONFIG_COMPAT
 906
 907 static int do_ioctl(struct block_device *bdev, fmode_t mode,
 908                     unsigned cmd, unsigned long arg)
 909 {
 910         int ret;
 911         lock_kernel();
 912         ret = cciss_ioctl(bdev, mode, cmd, arg);
 913         unlock_kernel();
 914         return ret;
 915 }
 916
 917 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 918                                   unsigned cmd, unsigned long arg);
 919 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
 920                                       unsigned cmd, unsigned long arg);
 921
 922 static int cciss_compat_ioctl(struct block_device *bdev, fmode_t mode,
 923                               unsigned cmd, unsigned long arg)
 924 {
 925         switch (cmd) {
 926         case CCISS_GETPCIINFO:
 927         case CCISS_GETINTINFO:
 928         case CCISS_SETINTINFO:
 929         case CCISS_GETNODENAME:
 930         case CCISS_SETNODENAME:
 931         case CCISS_GETHEARTBEAT:
 932         case CCISS_GETBUSTYPES:
 933         case CCISS_GETFIRMVER:
 934         case CCISS_GETDRIVVER:
 935         case CCISS_REVALIDVOLS:
 936         case CCISS_DEREGDISK:
 937         case CCISS_REGNEWDISK:
 938         case CCISS_REGNEWD:
 939         case CCISS_RESCANDISK:
 940         case CCISS_GETLUNINFO:
 941                 return do_ioctl(bdev, mode, cmd, arg);
 942
 943         case CCISS_PASSTHRU32:
 944                 return cciss_ioctl32_passthru(bdev, mode, cmd, arg);
 945         case CCISS_BIG_PASSTHRU32:
 946                 return cciss_ioctl32_big_passthru(bdev, mode, cmd, arg);
 947
 948         default:
 949                 return -ENOIOCTLCMD;
 950         }
 951 }
 952
 953 static int cciss_ioctl32_passthru(struct block_device *bdev, fmode_t mode,
 954                                   unsigned cmd, unsigned long arg)
 955 {
 956         IOCTL32_Command_struct __user *arg32 =
 957             (IOCTL32_Command_struct __user *) arg;
 958         IOCTL_Command_struct arg64;
 959         IOCTL_Command_struct __user *p = compat_alloc_user_space(sizeof(arg64));
 960         int err;
 961         u32 cp;
 962
 963         err = 0;
 964         err |=
 965             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
 966                            sizeof(arg64.LUN_info));
 967         err |=
 968             copy_from_user(&arg64.Request, &arg32->Request,
 969                            sizeof(arg64.Request));
 970         err |=
 971             copy_from_user(&arg64.error_info, &arg32->error_info,
 972                            sizeof(arg64.error_info));
 973         err |= get_user(arg64.buf_size, &arg32->buf_size);
 974         err |= get_user(cp, &arg32->buf);
 975         arg64.buf = compat_ptr(cp);
 976         err |= copy_to_user(p, &arg64, sizeof(arg64));
 977
 978         if (err)
 979                 return -EFAULT;
 980
 981         err = do_ioctl(bdev, mode, CCISS_PASSTHRU, (unsigned long)p);
 982         if (err)
 983                 return err;
 984         err |=
 985             copy_in_user(&arg32->error_info, &p->error_info,
 986                          sizeof(arg32->error_info));
 987         if (err)
 988                 return -EFAULT;
 989         return err;
 990 }
 991
 992 static int cciss_ioctl32_big_passthru(struct block_device *bdev, fmode_t mode,
 993                                       unsigned cmd, unsigned long arg)
 994 {
 995         BIG_IOCTL32_Command_struct __user *arg32 =
 996             (BIG_IOCTL32_Command_struct __user *) arg;
 997         BIG_IOCTL_Command_struct arg64;
 998         BIG_IOCTL_Command_struct __user *p =
 999             compat_alloc_user_space(sizeof(arg64));
1000         int err;
1001         u32 cp;
1002
1003         err = 0;
1004         err |=
1005             copy_from_user(&arg64.LUN_info, &arg32->LUN_info,
1006                            sizeof(arg64.LUN_info));
1007         err |=
1008             copy_from_user(&arg64.Request, &arg32->Request,
1009                            sizeof(arg64.Request));
1010         err |=
1011             copy_from_user(&arg64.error_info, &arg32->error_info,
1012                            sizeof(arg64.error_info));
1013         err |= get_user(arg64.buf_size, &arg32->buf_size);
1014         err |= get_user(arg64.malloc_size, &arg32->malloc_size);
1015         err |= get_user(cp, &arg32->buf);
1016         arg64.buf = compat_ptr(cp);
1017         err |= copy_to_user(p, &arg64, sizeof(arg64));
1018
1019         if (err)
1020                 return -EFAULT;
1021
1022         err = do_ioctl(bdev, mode, CCISS_BIG_PASSTHRU, (unsigned long)p);
1023         if (err)
1024                 return err;
1025         err |=
1026             copy_in_user(&arg32->error_info, &p->error_info,
1027                          sizeof(arg32->error_info));
1028         if (err)
1029                 return -EFAULT;
1030         return err;
1031 }
1032 #endif
1033
1034 static int cciss_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1035 {
1036         drive_info_struct *drv = get_drv(bdev->bd_disk);
1037
1038         if (!drv->cylinders)
1039                 return -ENXIO;
1040
1041         geo->heads = drv->heads;
1042         geo->sectors = drv->sectors;
1043         geo->cylinders = drv->cylinders;
1044         return 0;
1045 }
1046
1047 static void check_ioctl_unit_attention(ctlr_info_t *host, CommandList_struct *c)
1048 {
1049         if (c->err_info->CommandStatus == CMD_TARGET_STATUS &&
1050                         c->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION)
1051                 (void)check_for_unit_attention(host, c);
1052 }
1053 /*
1054  * ioctl
1055  */
1056 static int cciss_ioctl(struct block_device *bdev, fmode_t mode,
1057                        unsigned int cmd, unsigned long arg)
1058 {
1059         struct gendisk *disk = bdev->bd_disk;
1060         ctlr_info_t *host = get_host(disk);
1061         drive_info_struct *drv = get_drv(disk);
1062         int ctlr = host->ctlr;
1063         void __user *argp = (void __user *)arg;
1064
1065 #ifdef CCISS_DEBUG
1066         printk(KERN_DEBUG "cciss_ioctl: Called with cmd=%x %lx\n", cmd, arg);
1067 #endif                          /* CCISS_DEBUG */
1068
1069         switch (cmd) {
1070         case CCISS_GETPCIINFO:
1071                 {
1072                         cciss_pci_info_struct pciinfo;
1073
1074                         if (!arg)
1075                                 return -EINVAL;
1076                         pciinfo.domain = pci_domain_nr(host->pdev->bus);
1077                         pciinfo.bus = host->pdev->bus->number;
1078                         pciinfo.dev_fn = host->pdev->devfn;
1079                         pciinfo.board_id = host->board_id;
1080                         if (copy_to_user
1081                             (argp, &pciinfo, sizeof(cciss_pci_info_struct)))
1082                                 return -EFAULT;
1083                         return 0;
1084                 }
1085         case CCISS_GETINTINFO:
1086                 {
1087                         cciss_coalint_struct intinfo;
1088                         if (!arg)
1089                                 return -EINVAL;
1090                         intinfo.delay =
1091                             readl(&host->cfgtable->HostWrite.CoalIntDelay);
1092                         intinfo.count =
1093                             readl(&host->cfgtable->HostWrite.CoalIntCount);
1094                         if (copy_to_user
1095                             (argp, &intinfo, sizeof(cciss_coalint_struct)))
1096                                 return -EFAULT;
1097                         return 0;
1098                 }
1099         case CCISS_SETINTINFO:
1100                 {
1101                         cciss_coalint_struct intinfo;
1102                         unsigned long flags;
1103                         int i;
1104
1105                         if (!arg)
1106                                 return -EINVAL;
1107                         if (!capable(CAP_SYS_ADMIN))
1108                                 return -EPERM;
1109                         if (copy_from_user
1110                             (&intinfo, argp, sizeof(cciss_coalint_struct)))
1111                                 return -EFAULT;
1112                         if ((intinfo.delay == 0) && (intinfo.count == 0))
1113                         {
1114 //                      printk("cciss_ioctl: delay and count cannot be 0\n");
1115                                 return -EINVAL;
1116                         }
1117                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1118                         /* Update the field, and then ring the doorbell */
1119                         writel(intinfo.delay,
1120                                &(host->cfgtable->HostWrite.CoalIntDelay));
1121                         writel(intinfo.count,
1122                                &(host->cfgtable->HostWrite.CoalIntCount));
1123                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1124
1125                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1126                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1127                                       & CFGTBL_ChangeReq))
1128                                         break;
1129                                 /* delay and try again */
1130                                 udelay(1000);
1131                         }
1132                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1133                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1134                                 return -EAGAIN;
1135                         return 0;
1136                 }
1137         case CCISS_GETNODENAME:
1138                 {
1139                         NodeName_type NodeName;
1140                         int i;
1141
1142                         if (!arg)
1143                                 return -EINVAL;
1144                         for (i = 0; i < 16; i++)
1145                                 NodeName[i] =
1146                                     readb(&host->cfgtable->ServerName[i]);
1147                         if (copy_to_user(argp, NodeName, sizeof(NodeName_type)))
1148                                 return -EFAULT;
1149                         return 0;
1150                 }
1151         case CCISS_SETNODENAME:
1152                 {
1153                         NodeName_type NodeName;
1154                         unsigned long flags;
1155                         int i;
1156
1157                         if (!arg)
1158                                 return -EINVAL;
1159                         if (!capable(CAP_SYS_ADMIN))
1160                                 return -EPERM;
1161
1162                         if (copy_from_user
1163                             (NodeName, argp, sizeof(NodeName_type)))
1164                                 return -EFAULT;
1165
1166                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1167
1168                         /* Update the field, and then ring the doorbell */
1169                         for (i = 0; i < 16; i++)
1170                                 writeb(NodeName[i],
1171                                        &host->cfgtable->ServerName[i]);
1172
1173                         writel(CFGTBL_ChangeReq, host->vaddr + SA5_DOORBELL);
1174
1175                         for (i = 0; i < MAX_IOCTL_CONFIG_WAIT; i++) {
1176                                 if (!(readl(host->vaddr + SA5_DOORBELL)
1177                                       & CFGTBL_ChangeReq))
1178                                         break;
1179                                 /* delay and try again */
1180                                 udelay(1000);
1181                         }
1182                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1183                         if (i >= MAX_IOCTL_CONFIG_WAIT)
1184                                 return -EAGAIN;
1185                         return 0;
1186                 }
1187
1188         case CCISS_GETHEARTBEAT:
1189                 {
1190                         Heartbeat_type heartbeat;
1191
1192                         if (!arg)
1193                                 return -EINVAL;
1194                         heartbeat = readl(&host->cfgtable->HeartBeat);
1195                         if (copy_to_user
1196                             (argp, &heartbeat, sizeof(Heartbeat_type)))
1197                                 return -EFAULT;
1198                         return 0;
1199                 }
1200         case CCISS_GETBUSTYPES:
1201                 {
1202                         BusTypes_type BusTypes;
1203
1204                         if (!arg)
1205                                 return -EINVAL;
1206                         BusTypes = readl(&host->cfgtable->BusTypes);
1207                         if (copy_to_user
1208                             (argp, &BusTypes, sizeof(BusTypes_type)))
1209                                 return -EFAULT;
1210                         return 0;
1211                 }
1212         case CCISS_GETFIRMVER:
1213                 {
1214                         FirmwareVer_type firmware;
1215
1216                         if (!arg)
1217                                 return -EINVAL;
1218                         memcpy(firmware, host->firm_ver, 4);
1219
1220                         if (copy_to_user
1221                             (argp, firmware, sizeof(FirmwareVer_type)))
1222                                 return -EFAULT;
1223                         return 0;
1224                 }
1225         case CCISS_GETDRIVVER:
1226                 {
1227                         DriverVer_type DriverVer = DRIVER_VERSION;
1228
1229                         if (!arg)
1230                                 return -EINVAL;
1231
1232                         if (copy_to_user
1233                             (argp, &DriverVer, sizeof(DriverVer_type)))
1234                                 return -EFAULT;
1235                         return 0;
1236                 }
1237
1238         case CCISS_DEREGDISK:
1239         case CCISS_REGNEWD:
1240         case CCISS_REVALIDVOLS:
1241                 return rebuild_lun_table(host, 0, 1);
1242
1243         case CCISS_GETLUNINFO:{
1244                         LogvolInfo_struct luninfo;
1245
1246                         memcpy(&luninfo.LunID, drv->LunID,
1247                                 sizeof(luninfo.LunID));
1248                         luninfo.num_opens = drv->usage_count;
1249                         luninfo.num_parts = 0;
1250                         if (copy_to_user(argp, &luninfo,
1251                                          sizeof(LogvolInfo_struct)))
1252                                 return -EFAULT;
1253                         return 0;
1254                 }
1255         case CCISS_PASSTHRU:
1256                 {
1257                         IOCTL_Command_struct iocommand;
1258                         CommandList_struct *c;
1259                         char *buff = NULL;
1260                         u64bit temp64;
1261                         unsigned long flags;
1262                         DECLARE_COMPLETION_ONSTACK(wait);
1263
1264                         if (!arg)
1265                                 return -EINVAL;
1266
1267                         if (!capable(CAP_SYS_RAWIO))
1268                                 return -EPERM;
1269
1270                         if (copy_from_user
1271                             (&iocommand, argp, sizeof(IOCTL_Command_struct)))
1272                                 return -EFAULT;
1273                         if ((iocommand.buf_size < 1) &&
1274                             (iocommand.Request.Type.Direction != XFER_NONE)) {
1275                                 return -EINVAL;
1276                         }
1277 #if 0                           /* 'buf_size' member is 16-bits, and always smaller than kmalloc limit */
1278                         /* Check kmalloc limits */
1279                         if (iocommand.buf_size > 128000)
1280                                 return -EINVAL;
1281 #endif
1282                         if (iocommand.buf_size > 0) {
1283                                 buff = kmalloc(iocommand.buf_size, GFP_KERNEL);
1284                                 if (buff == NULL)
1285                                         return -EFAULT;
1286                         }
1287                         if (iocommand.Request.Type.Direction == XFER_WRITE) {
1288                                 /* Copy the data into the buffer we created */
1289                                 if (copy_from_user
1290                                     (buff, iocommand.buf, iocommand.buf_size)) {
1291                                         kfree(buff);
1292                                         return -EFAULT;
1293                                 }
1294                         } else {
1295                                 memset(buff, 0, iocommand.buf_size);
1296                         }
1297                         if ((c = cmd_alloc(host, 0)) == NULL) {
1298                                 kfree(buff);
1299                                 return -ENOMEM;
1300                         }
1301                         // Fill in the command type
1302                         c->cmd_type = CMD_IOCTL_PEND;
1303                         // Fill in Command Header
1304                         c->Header.ReplyQueue = 0;       // unused in simple mode
1305                         if (iocommand.buf_size > 0)     // buffer to fill
1306                         {
1307                                 c->Header.SGList = 1;
1308                                 c->Header.SGTotal = 1;
1309                         } else  // no buffers to fill
1310                         {
1311                                 c->Header.SGList = 0;
1312                                 c->Header.SGTotal = 0;
1313                         }
1314                         c->Header.LUN = iocommand.LUN_info;
1315                         c->Header.Tag.lower = c->busaddr;       // use the kernel address the cmd block for tag
1316
1317                         // Fill in Request block
1318                         c->Request = iocommand.Request;
1319
1320                         // Fill in the scatter gather information
1321                         if (iocommand.buf_size > 0) {
1322                                 temp64.val = pci_map_single(host->pdev, buff,
1323                                         iocommand.buf_size,
1324                                         PCI_DMA_BIDIRECTIONAL);
1325                                 c->SG[0].Addr.lower = temp64.val32.lower;
1326                                 c->SG[0].Addr.upper = temp64.val32.upper;
1327                                 c->SG[0].Len = iocommand.buf_size;
1328                                 c->SG[0].Ext = 0;       // we are not chaining
1329                         }
1330                         c->waiting = &wait;
1331
1332                         /* Put the request on the tail of the request queue */
1333                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1334                         addQ(&host->reqQ, c);
1335                         host->Qdepth++;
1336                         start_io(host);
1337                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1338
1339                         wait_for_completion(&wait);
1340
1341                         /* unlock the buffers from DMA */
1342                         temp64.val32.lower = c->SG[0].Addr.lower;
1343                         temp64.val32.upper = c->SG[0].Addr.upper;
1344                         pci_unmap_single(host->pdev, (dma_addr_t) temp64.val,
1345                                          iocommand.buf_size,
1346                                          PCI_DMA_BIDIRECTIONAL);
1347
1348                         check_ioctl_unit_attention(host, c);
1349
1350                         /* Copy the error information out */
1351                         iocommand.error_info = *(c->err_info);
1352                         if (copy_to_user
1353                             (argp, &iocommand, sizeof(IOCTL_Command_struct))) {
1354                                 kfree(buff);
1355                                 cmd_free(host, c, 0);
1356                                 return -EFAULT;
1357                         }
1358
1359                         if (iocommand.Request.Type.Direction == XFER_READ) {
1360                                 /* Copy the data out of the buffer we created */
1361                                 if (copy_to_user
1362                                     (iocommand.buf, buff, iocommand.buf_size)) {
1363                                         kfree(buff);
1364                                         cmd_free(host, c, 0);
1365                                         return -EFAULT;
1366                                 }
1367                         }
1368                         kfree(buff);
1369                         cmd_free(host, c, 0);
1370                         return 0;
1371                 }
1372         case CCISS_BIG_PASSTHRU:{
1373                         BIG_IOCTL_Command_struct *ioc;
1374                         CommandList_struct *c;
1375                         unsigned char **buff = NULL;
1376                         int *buff_size = NULL;
1377                         u64bit temp64;
1378                         unsigned long flags;
1379                         BYTE sg_used = 0;
1380                         int status = 0;
1381                         int i;
1382                         DECLARE_COMPLETION_ONSTACK(wait);
1383                         __u32 left;
1384                         __u32 sz;
1385                         BYTE __user *data_ptr;
1386
1387                         if (!arg)
1388                                 return -EINVAL;
1389                         if (!capable(CAP_SYS_RAWIO))
1390                                 return -EPERM;
1391                         ioc = (BIG_IOCTL_Command_struct *)
1392                             kmalloc(sizeof(*ioc), GFP_KERNEL);
1393                         if (!ioc) {
1394                                 status = -ENOMEM;
1395                                 goto cleanup1;
1396                         }
1397                         if (copy_from_user(ioc, argp, sizeof(*ioc))) {
1398                                 status = -EFAULT;
1399                                 goto cleanup1;
1400                         }
1401                         if ((ioc->buf_size < 1) &&
1402                             (ioc->Request.Type.Direction != XFER_NONE)) {
1403                                 status = -EINVAL;
1404                                 goto cleanup1;
1405                         }
1406                         /* Check kmalloc limits  using all SGs */
1407                         if (ioc->malloc_size > MAX_KMALLOC_SIZE) {
1408                                 status = -EINVAL;
1409                                 goto cleanup1;
1410                         }
1411                         if (ioc->buf_size > ioc->malloc_size * MAXSGENTRIES) {
1412                                 status = -EINVAL;
1413                                 goto cleanup1;
1414                         }
1415                         buff =
1416                             kzalloc(MAXSGENTRIES * sizeof(char *), GFP_KERNEL);
1417                         if (!buff) {
1418                                 status = -ENOMEM;
1419                                 goto cleanup1;
1420                         }
1421                         buff_size = kmalloc(MAXSGENTRIES * sizeof(int),
1422                                                    GFP_KERNEL);
1423                         if (!buff_size) {
1424                                 status = -ENOMEM;
1425                                 goto cleanup1;
1426                         }
1427                         left = ioc->buf_size;
1428                         data_ptr = ioc->buf;
1429                         while (left) {
1430                                 sz = (left >
1431                                       ioc->malloc_size) ? ioc->
1432                                     malloc_size : left;
1433                                 buff_size[sg_used] = sz;
1434                                 buff[sg_used] = kmalloc(sz, GFP_KERNEL);
1435                                 if (buff[sg_used] == NULL) {
1436                                         status = -ENOMEM;
1437                                         goto cleanup1;
1438                                 }
1439                                 if (ioc->Request.Type.Direction == XFER_WRITE) {
1440                                         if (copy_from_user
1441                                             (buff[sg_used], data_ptr, sz)) {
1442                                                 status = -EFAULT;
1443                                                 goto cleanup1;
1444                                         }
1445                                 } else {
1446                                         memset(buff[sg_used], 0, sz);
1447                                 }
1448                                 left -= sz;
1449                                 data_ptr += sz;
1450                                 sg_used++;
1451                         }
1452                         if ((c = cmd_alloc(host, 0)) == NULL) {
1453                                 status = -ENOMEM;
1454                                 goto cleanup1;
1455                         }
1456                         c->cmd_type = CMD_IOCTL_PEND;
1457                         c->Header.ReplyQueue = 0;
1458
1459                         if (ioc->buf_size > 0) {
1460                                 c->Header.SGList = sg_used;
1461                                 c->Header.SGTotal = sg_used;
1462                         } else {
1463                                 c->Header.SGList = 0;
1464                                 c->Header.SGTotal = 0;
1465                         }
1466                         c->Header.LUN = ioc->LUN_info;
1467                         c->Header.Tag.lower = c->busaddr;
1468
1469                         c->Request = ioc->Request;
1470                         if (ioc->buf_size > 0) {
1471                                 int i;
1472                                 for (i = 0; i < sg_used; i++) {
1473                                         temp64.val =
1474                                             pci_map_single(host->pdev, buff[i],
1475                                                     buff_size[i],
1476                                                     PCI_DMA_BIDIRECTIONAL);
1477                                         c->SG[i].Addr.lower =
1478                                             temp64.val32.lower;
1479                                         c->SG[i].Addr.upper =
1480                                             temp64.val32.upper;
1481                                         c->SG[i].Len = buff_size[i];
1482                                         c->SG[i].Ext = 0;       /* we are not chaining */
1483                                 }
1484                         }
1485                         c->waiting = &wait;
1486                         /* Put the request on the tail of the request queue */
1487                         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
1488                         addQ(&host->reqQ, c);
1489                         host->Qdepth++;
1490                         start_io(host);
1491                         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
1492                         wait_for_completion(&wait);
1493                         /* unlock the buffers from DMA */
1494                         for (i = 0; i < sg_used; i++) {
1495                                 temp64.val32.lower = c->SG[i].Addr.lower;
1496                                 temp64.val32.upper = c->SG[i].Addr.upper;
1497                                 pci_unmap_single(host->pdev,
1498                                         (dma_addr_t) temp64.val, buff_size[i],
1499                                         PCI_DMA_BIDIRECTIONAL);
1500                         }
1501                         check_ioctl_unit_attention(host, c);
1502                         /* Copy the error information out */
1503                         ioc->error_info = *(c->err_info);
1504                         if (copy_to_user(argp, ioc, sizeof(*ioc))) {
1505                                 cmd_free(host, c, 0);
1506                                 status = -EFAULT;
1507                                 goto cleanup1;
1508                         }
1509                         if (ioc->Request.Type.Direction == XFER_READ) {
1510                                 /* Copy the data out of the buffer we created */
1511                                 BYTE __user *ptr = ioc->buf;
1512                                 for (i = 0; i < sg_used; i++) {
1513                                         if (copy_to_user
1514                                             (ptr, buff[i], buff_size[i])) {
1515                                                 cmd_free(host, c, 0);
1516                                                 status = -EFAULT;
1517                                                 goto cleanup1;
1518                                         }
1519                                         ptr += buff_size[i];
1520                                 }
1521                         }
1522                         cmd_free(host, c, 0);
1523                         status = 0;
1524                       cleanup1:
1525                         if (buff) {
1526                                 for (i = 0; i < sg_used; i++)
1527                                         kfree(buff[i]);
1528                                 kfree(buff);
1529                         }
1530                         kfree(buff_size);
1531                         kfree(ioc);
1532                         return status;
1533                 }
1534
1535         /* scsi_cmd_ioctl handles these, below, though some are not */
1536         /* very meaningful for cciss.  SG_IO is the main one people want. */
1537
1538         case SG_GET_VERSION_NUM:
1539         case SG_SET_TIMEOUT:
1540         case SG_GET_TIMEOUT:
1541         case SG_GET_RESERVED_SIZE:
1542         case SG_SET_RESERVED_SIZE:
1543         case SG_EMULATED_HOST:
1544         case SG_IO:
1545         case SCSI_IOCTL_SEND_COMMAND:
1546                 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
1547
1548         /* scsi_cmd_ioctl would normally handle these, below, but */
1549         /* they aren't a good fit for cciss, as CD-ROMs are */
1550         /* not supported, and we don't have any bus/target/lun */
1551         /* which we present to the kernel. */
1552
1553         case CDROM_SEND_PACKET:
1554         case CDROMCLOSETRAY:
1555         case CDROMEJECT:
1556         case SCSI_IOCTL_GET_IDLUN:
1557         case SCSI_IOCTL_GET_BUS_NUMBER:
1558         default:
1559                 return -ENOTTY;
1560         }
1561 }
1562
1563 static void cciss_check_queues(ctlr_info_t *h)
1564 {
1565         int start_queue = h->next_to_run;
1566         int i;
1567
1568         /* check to see if we have maxed out the number of commands that can
1569          * be placed on the queue.  If so then exit.  We do this check here
1570          * in case the interrupt we serviced was from an ioctl and did not
1571          * free any new commands.
1572          */
1573         if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds)
1574                 return;
1575
1576         /* We have room on the queue for more commands.  Now we need to queue
1577          * them up.  We will also keep track of the next queue to run so
1578          * that every queue gets a chance to be started first.
1579          */
1580         for (i = 0; i < h->highest_lun + 1; i++) {
1581                 int curr_queue = (start_queue + i) % (h->highest_lun + 1);
1582                 /* make sure the disk has been added and the drive is real
1583                  * because this can be called from the middle of init_one.
1584                  */
1585                 if (!(h->drv[curr_queue].queue) || !(h->drv[curr_queue].heads))
1586                         continue;
1587                 blk_start_queue(h->gendisk[curr_queue]->queue);
1588
1589                 /* check to see if we have maxed out the number of commands
1590                  * that can be placed on the queue.
1591                  */
1592                 if ((find_first_zero_bit(h->cmd_pool_bits, h->nr_cmds)) == h->nr_cmds) {
1593                         if (curr_queue == start_queue) {
1594                                 h->next_to_run =
1595                                     (start_queue + 1) % (h->highest_lun + 1);
1596                                 break;
1597                         } else {
1598                                 h->next_to_run = curr_queue;
1599                                 break;
1600                         }
1601                 }
1602         }
1603 }
1604
1605 static void cciss_softirq_done(struct request *rq)
1606 {
1607         CommandList_struct *cmd = rq->completion_data;
1608         ctlr_info_t *h = hba[cmd->ctlr];
1609         unsigned long flags;
1610         u64bit temp64;
1611         int i, ddir;
1612
1613         if (cmd->Request.Type.Direction == XFER_READ)
1614                 ddir = PCI_DMA_FROMDEVICE;
1615         else
1616                 ddir = PCI_DMA_TODEVICE;
1617
1618         /* command did not need to be retried */
1619         /* unmap the DMA mapping for all the scatter gather elements */
1620         for (i = 0; i < cmd->Header.SGList; i++) {
1621                 temp64.val32.lower = cmd->SG[i].Addr.lower;
1622                 temp64.val32.upper = cmd->SG[i].Addr.upper;
1623                 pci_unmap_page(h->pdev, temp64.val, cmd->SG[i].Len, ddir);
1624         }
1625
1626 #ifdef CCISS_DEBUG
1627         printk("Done with %p\n", rq);
1628 #endif                          /* CCISS_DEBUG */
1629
1630         /* set the residual count for pc requests */
1631         if (blk_pc_request(rq))
1632                 rq->resid_len = cmd->err_info->ResidualCnt;
1633
1634         blk_end_request_all(rq, (rq->errors == 0) ? 0 : -EIO);
1635
1636         spin_lock_irqsave(&h->lock, flags);
1637         cmd_free(h, cmd, 1);
1638         cciss_check_queues(h);
1639         spin_unlock_irqrestore(&h->lock, flags);
1640 }
1641
1642 static inline void log_unit_to_scsi3addr(ctlr_info_t *h,
1643         unsigned char scsi3addr[], uint32_t log_unit)
1644 {
1645         memcpy(scsi3addr, h->drv[log_unit].LunID,
1646                 sizeof(h->drv[log_unit].LunID));
1647 }
1648
1649 /* This function gets the SCSI vendor, model, and revision of a logical drive
1650  * via the inquiry page 0.  Model, vendor, and rev are set to empty strings if
1651  * they cannot be read.
1652  */
1653 static void cciss_get_device_descr(int ctlr, int logvol, int withirq,
1654                                    char *vendor, char *model, char *rev)
1655 {
1656         int rc;
1657         InquiryData_struct *inq_buf;
1658         unsigned char scsi3addr[8];
1659
1660         *vendor = '\0';
1661         *model = '\0';
1662         *rev = '\0';
1663
1664         inq_buf = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1665         if (!inq_buf)
1666                 return;
1667
1668         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1669         if (withirq)
1670                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, inq_buf,
1671                              sizeof(InquiryData_struct), 0,
1672                                 scsi3addr, TYPE_CMD);
1673         else
1674                 rc = sendcmd(CISS_INQUIRY, ctlr, inq_buf,
1675                              sizeof(InquiryData_struct), 0,
1676                                 scsi3addr, TYPE_CMD);
1677         if (rc == IO_OK) {
1678                 memcpy(vendor, &inq_buf->data_byte[8], VENDOR_LEN);
1679                 vendor[VENDOR_LEN] = '\0';
1680                 memcpy(model, &inq_buf->data_byte[16], MODEL_LEN);
1681                 model[MODEL_LEN] = '\0';
1682                 memcpy(rev, &inq_buf->data_byte[32], REV_LEN);
1683                 rev[REV_LEN] = '\0';
1684         }
1685
1686         kfree(inq_buf);
1687         return;
1688 }
1689
1690 /* This function gets the serial number of a logical drive via
1691  * inquiry page 0x83.  Serial no. is 16 bytes.  If the serial
1692  * number cannot be had, for whatever reason, 16 bytes of 0xff
1693  * are returned instead.
1694  */
1695 static void cciss_get_serial_no(int ctlr, int logvol, int withirq,
1696                                 unsigned char *serial_no, int buflen)
1697 {
1698 #define PAGE_83_INQ_BYTES 64
1699         int rc;
1700         unsigned char *buf;
1701         unsigned char scsi3addr[8];
1702
1703         if (buflen > 16)
1704                 buflen = 16;
1705         memset(serial_no, 0xff, buflen);
1706         buf = kzalloc(PAGE_83_INQ_BYTES, GFP_KERNEL);
1707         if (!buf)
1708                 return;
1709         memset(serial_no, 0, buflen);
1710         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
1711         if (withirq)
1712                 rc = sendcmd_withirq(CISS_INQUIRY, ctlr, buf,
1713                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1714         else
1715                 rc = sendcmd(CISS_INQUIRY, ctlr, buf,
1716                         PAGE_83_INQ_BYTES, 0x83, scsi3addr, TYPE_CMD);
1717         if (rc == IO_OK)
1718                 memcpy(serial_no, &buf[8], buflen);
1719         kfree(buf);
1720         return;
1721 }
1722
1723 /*
1724  * cciss_add_disk sets up the block device queue for a logical drive
1725  */
1726 static int cciss_add_disk(ctlr_info_t *h, struct gendisk *disk,
1727                                 int drv_index)
1728 {
1729         disk->queue = blk_init_queue(do_cciss_request, &h->lock);
1730         if (!disk->queue)
1731                 goto init_queue_failure;
1732         sprintf(disk->disk_name, "cciss/c%dd%d", h->ctlr, drv_index);
1733         disk->major = h->major;
1734         disk->first_minor = drv_index << NWD_SHIFT;
1735         disk->fops = &cciss_fops;
1736         if (h->drv[drv_index].dev == NULL) {
1737                 if (cciss_create_ld_sysfs_entry(h, drv_index))
1738                         goto cleanup_queue;
1739         }
1740         disk->private_data = &h->drv[drv_index];
1741         disk->driverfs_dev = h->drv[drv_index].dev;
1742
1743         /* Set up queue information */
1744         blk_queue_bounce_limit(disk->queue, h->pdev->dma_mask);
1745
1746         /* This is a hardware imposed limit. */
1747         blk_queue_max_hw_segments(disk->queue, MAXSGENTRIES);
1748
1749         /* This is a limit in the driver and could be eliminated. */
1750         blk_queue_max_phys_segments(disk->queue, MAXSGENTRIES);
1751
1752         blk_queue_max_sectors(disk->queue, h->cciss_max_sectors);
1753
1754         blk_queue_softirq_done(disk->queue, cciss_softirq_done);
1755
1756         disk->queue->queuedata = h;
1757
1758         blk_queue_logical_block_size(disk->queue,
1759                                      h->drv[drv_index].block_size);
1760
1761         /* Make sure all queue data is written out before */
1762         /* setting h->drv[drv_index].queue, as setting this */
1763         /* allows the interrupt handler to start the queue */
1764         wmb();
1765         h->drv[drv_index].queue = disk->queue;
1766         add_disk(disk);
1767         return 0;
1768
1769 cleanup_queue:
1770         blk_cleanup_queue(disk->queue);
1771         disk->queue = NULL;
1772 init_queue_failure:
1773         return -1;
1774 }
1775
1776 /* This function will check the usage_count of the drive to be updated/added.
1777  * If the usage_count is zero and it is a heretofore unknown drive, or,
1778  * the drive's capacity, geometry, or serial number has changed,
1779  * then the drive information will be updated and the disk will be
1780  * re-registered with the kernel.  If these conditions don't hold,
1781  * then it will be left alone for the next reboot.  The exception to this
1782  * is disk 0 which will always be left registered with the kernel since it
1783  * is also the controller node.  Any changes to disk 0 will show up on
1784  * the next reboot.
1785  */
1786 static void cciss_update_drive_info(int ctlr, int drv_index, int first_time,
1787         int via_ioctl)
1788 {
1789         ctlr_info_t *h = hba[ctlr];
1790         struct gendisk *disk;
1791         InquiryData_struct *inq_buff = NULL;
1792         unsigned int block_size;
1793         sector_t total_size;
1794         unsigned long flags = 0;
1795         int ret = 0;
1796         drive_info_struct *drvinfo;
1797
1798         /* Get information about the disk and modify the driver structure */
1799         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
1800         drvinfo = kmalloc(sizeof(*drvinfo), GFP_KERNEL);
1801         if (inq_buff == NULL || drvinfo == NULL)
1802                 goto mem_msg;
1803
1804         /* testing to see if 16-byte CDBs are already being used */
1805         if (h->cciss_read == CCISS_READ_16) {
1806                 cciss_read_capacity_16(h->ctlr, drv_index, 1,
1807                         &total_size, &block_size);
1808
1809         } else {
1810                 cciss_read_capacity(ctlr, drv_index, 1,
1811                                     &total_size, &block_size);
1812
1813                 /* if read_capacity returns all F's this volume is >2TB */
1814                 /* in size so we switch to 16-byte CDB's for all */
1815                 /* read/write ops */
1816                 if (total_size == 0xFFFFFFFFULL) {
1817                         cciss_read_capacity_16(ctlr, drv_index, 1,
1818                         &total_size, &block_size);
1819                         h->cciss_read = CCISS_READ_16;
1820                         h->cciss_write = CCISS_WRITE_16;
1821                 } else {
1822                         h->cciss_read = CCISS_READ_10;
1823                         h->cciss_write = CCISS_WRITE_10;
1824                 }
1825         }
1826
1827         cciss_geometry_inquiry(ctlr, drv_index, 1, total_size, block_size,
1828                                inq_buff, drvinfo);
1829         drvinfo->block_size = block_size;
1830         drvinfo->nr_blocks = total_size + 1;
1831
1832         cciss_get_device_descr(ctlr, drv_index, 1, drvinfo->vendor,
1833                                 drvinfo->model, drvinfo->rev);
1834         cciss_get_serial_no(ctlr, drv_index, 1, drvinfo->serial_no,
1835                         sizeof(drvinfo->serial_no));
1836
1837         /* Is it the same disk we already know, and nothing's changed? */
1838         if (h->drv[drv_index].raid_level != -1 &&
1839                 ((memcmp(drvinfo->serial_no,
1840                                 h->drv[drv_index].serial_no, 16) == 0) &&
1841                 drvinfo->block_size == h->drv[drv_index].block_size &&
1842                 drvinfo->nr_blocks == h->drv[drv_index].nr_blocks &&
1843                 drvinfo->heads == h->drv[drv_index].heads &&
1844                 drvinfo->sectors == h->drv[drv_index].sectors &&
1845                 drvinfo->cylinders == h->drv[drv_index].cylinders))
1846                         /* The disk is unchanged, nothing to update */
1847                         goto freeret;
1848
1849         /* If we get here it's not the same disk, or something's changed,
1850          * so we need to * deregister it, and re-register it, if it's not
1851          * in use.
1852          * If the disk already exists then deregister it before proceeding
1853          * (unless it's the first disk (for the controller node).
1854          */
1855         if (h->drv[drv_index].raid_level != -1 && drv_index != 0) {
1856                 printk(KERN_WARNING "disk %d has changed.\n", drv_index);
1857                 spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
1858                 h->drv[drv_index].busy_configuring = 1;
1859                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
1860
1861                 /* deregister_disk sets h->drv[drv_index].queue = NULL
1862                  * which keeps the interrupt handler from starting
1863                  * the queue.
1864                  */
1865                 ret = deregister_disk(h, drv_index, 0, via_ioctl);
1866                 h->drv[drv_index].busy_configuring = 0;
1867         }
1868
1869         /* If the disk is in use return */
1870         if (ret)
1871                 goto freeret;
1872
1873         /* Save the new information from cciss_geometry_inquiry
1874          * and serial number inquiry.
1875          */
1876         h->drv[drv_index].block_size = drvinfo->block_size;
1877         h->drv[drv_index].nr_blocks = drvinfo->nr_blocks;
1878         h->drv[drv_index].heads = drvinfo->heads;
1879         h->drv[drv_index].sectors = drvinfo->sectors;
1880         h->drv[drv_index].cylinders = drvinfo->cylinders;
1881         h->drv[drv_index].raid_level = drvinfo->raid_level;
1882         memcpy(h->drv[drv_index].serial_no, drvinfo->serial_no, 16);
1883         memcpy(h->drv[drv_index].vendor, drvinfo->vendor, VENDOR_LEN + 1);
1884         memcpy(h->drv[drv_index].model, drvinfo->model, MODEL_LEN + 1);
1885         memcpy(h->drv[drv_index].rev, drvinfo->rev, REV_LEN + 1);
1886
1887         ++h->num_luns;
1888         disk = h->gendisk[drv_index];
1889         set_capacity(disk, h->drv[drv_index].nr_blocks);
1890
1891         /* If it's not disk 0 (drv_index != 0)
1892          * or if it was disk 0, but there was previously
1893          * no actual corresponding configured logical drive
1894          * (raid_leve == -1) then we want to update the
1895          * logical drive's information.
1896          */
1897         if (drv_index || first_time) {
1898                 if (cciss_add_disk(h, disk, drv_index) != 0) {
1899                         cciss_free_gendisk(h, drv_index);
1900                         printk(KERN_WARNING "cciss:%d could not update "
1901                                 "disk %d\n", h->ctlr, drv_index);
1902                         --h->num_luns;
1903                 }
1904         }
1905
1906 freeret:
1907         kfree(inq_buff);
1908         kfree(drvinfo);
1909         return;
1910 mem_msg:
1911         printk(KERN_ERR "cciss: out of memory\n");
1912         goto freeret;
1913 }
1914
1915 /* This function will find the first index of the controllers drive array
1916  * that has a -1 for the raid_level and will return that index.  This is
1917  * where new drives will be added.  If the index to be returned is greater
1918  * than the highest_lun index for the controller then highest_lun is set
1919  * to this new index.  If there are no available indexes then -1 is returned.
1920  * "controller_node" is used to know if this is a real logical drive, or just
1921  * the controller node, which determines if this counts towards highest_lun.
1922  */
1923 static int cciss_find_free_drive_index(int ctlr, int controller_node)
1924 {
1925         int i;
1926
1927         for (i = 0; i < CISS_MAX_LUN; i++) {
1928                 if (hba[ctlr]->drv[i].raid_level == -1) {
1929                         if (i > hba[ctlr]->highest_lun)
1930                                 if (!controller_node)
1931                                         hba[ctlr]->highest_lun = i;
1932                         return i;
1933                 }
1934         }
1935         return -1;
1936 }
1937
1938 static void cciss_free_gendisk(ctlr_info_t *h, int drv_index)
1939 {
1940         put_disk(h->gendisk[drv_index]);
1941         h->gendisk[drv_index] = NULL;
1942 }
1943
1944 /* cciss_add_gendisk finds a free hba[]->drv structure
1945  * and allocates a gendisk if needed, and sets the lunid
1946  * in the drvinfo structure.   It returns the index into
1947  * the ->drv[] array, or -1 if none are free.
1948  * is_controller_node indicates whether highest_lun should
1949  * count this disk, or if it's only being added to provide
1950  * a means to talk to the controller in case no logical
1951  * drives have yet been configured.
1952  */
1953 static int cciss_add_gendisk(ctlr_info_t *h, unsigned char lunid[],
1954         int controller_node)
1955 {
1956         int drv_index;
1957
1958         drv_index = cciss_find_free_drive_index(h->ctlr, controller_node);
1959         if (drv_index == -1)
1960                 return -1;
1961
1962         /*Check if the gendisk needs to be allocated */
1963         if (!h->gendisk[drv_index]) {
1964                 h->gendisk[drv_index] =
1965                         alloc_disk(1 << NWD_SHIFT);
1966                 if (!h->gendisk[drv_index]) {
1967                         printk(KERN_ERR "cciss%d: could not "
1968                                 "allocate a new disk %d\n",
1969                                 h->ctlr, drv_index);
1970                         return -1;
1971                 }
1972         }
1973         memcpy(h->drv[drv_index].LunID, lunid,
1974                 sizeof(h->drv[drv_index].LunID));
1975         if (h->drv[drv_index].dev == NULL) {
1976                 if (cciss_create_ld_sysfs_entry(h, drv_index))
1977                         goto err_free_disk;
1978         }
1979         /* Don't need to mark this busy because nobody */
1980         /* else knows about this disk yet to contend */
1981         /* for access to it. */
1982         h->drv[drv_index].busy_configuring = 0;
1983         wmb();
1984         return drv_index;
1985
1986 err_free_disk:
1987         cciss_free_gendisk(h, drv_index);
1988         return -1;
1989 }
1990
1991 /* This is for the special case of a controller which
1992  * has no logical drives.  In this case, we still need
1993  * to register a disk so the controller can be accessed
1994  * by the Array Config Utility.
1995  */
1996 static void cciss_add_controller_node(ctlr_info_t *h)
1997 {
1998         struct gendisk *disk;
1999         int drv_index;
2000
2001         if (h->gendisk[0] != NULL) /* already did this? Then bail. */
2002                 return;
2003
2004         drv_index = cciss_add_gendisk(h, CTLR_LUNID, 1);
2005         if (drv_index == -1)
2006                 goto error;
2007         h->drv[drv_index].block_size = 512;
2008         h->drv[drv_index].nr_blocks = 0;
2009         h->drv[drv_index].heads = 0;
2010         h->drv[drv_index].sectors = 0;
2011         h->drv[drv_index].cylinders = 0;
2012         h->drv[drv_index].raid_level = -1;
2013         memset(h->drv[drv_index].serial_no, 0, 16);
2014         disk = h->gendisk[drv_index];
2015         if (cciss_add_disk(h, disk, drv_index) == 0)
2016                 return;
2017         cciss_free_gendisk(h, drv_index);
2018 error:
2019         printk(KERN_WARNING "cciss%d: could not "
2020                 "add disk 0.\n", h->ctlr);
2021         return;
2022 }
2023
2024 /* This function will add and remove logical drives from the Logical
2025  * drive array of the controller and maintain persistency of ordering
2026  * so that mount points are preserved until the next reboot.  This allows
2027  * for the removal of logical drives in the middle of the drive array
2028  * without a re-ordering of those drives.
2029  * INPUT
2030  * h            = The controller to perform the operations on
2031  */
2032 static int rebuild_lun_table(ctlr_info_t *h, int first_time,
2033         int via_ioctl)
2034 {
2035         int ctlr = h->ctlr;
2036         int num_luns;
2037         ReportLunData_struct *ld_buff = NULL;
2038         int return_code;
2039         int listlength = 0;
2040         int i;
2041         int drv_found;
2042         int drv_index = 0;
2043         unsigned char lunid[8] = CTLR_LUNID;
2044         unsigned long flags;
2045
2046         if (!capable(CAP_SYS_RAWIO))
2047                 return -EPERM;
2048
2049         /* Set busy_configuring flag for this operation */
2050         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2051         if (h->busy_configuring) {
2052                 spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2053                 return -EBUSY;
2054         }
2055         h->busy_configuring = 1;
2056         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2057
2058         ld_buff = kzalloc(sizeof(ReportLunData_struct), GFP_KERNEL);
2059         if (ld_buff == NULL)
2060                 goto mem_msg;
2061
2062         return_code = sendcmd_withirq(CISS_REPORT_LOG, ctlr, ld_buff,
2063                                       sizeof(ReportLunData_struct),
2064                                       0, CTLR_LUNID, TYPE_CMD);
2065
2066         if (return_code == IO_OK)
2067                 listlength = be32_to_cpu(*(__be32 *) ld_buff->LUNListLength);
2068         else {  /* reading number of logical volumes failed */
2069                 printk(KERN_WARNING "cciss: report logical volume"
2070                        " command failed\n");
2071                 listlength = 0;
2072                 goto freeret;
2073         }
2074
2075         num_luns = listlength / 8;      /* 8 bytes per entry */
2076         if (num_luns > CISS_MAX_LUN) {
2077                 num_luns = CISS_MAX_LUN;
2078                 printk(KERN_WARNING "cciss: more luns configured"
2079                        " on controller than can be handled by"
2080                        " this driver.\n");
2081         }
2082
2083         if (num_luns == 0)
2084                 cciss_add_controller_node(h);
2085
2086         /* Compare controller drive array to driver's drive array
2087          * to see if any drives are missing on the controller due
2088          * to action of Array Config Utility (user deletes drive)
2089          * and deregister logical drives which have disappeared.
2090          */
2091         for (i = 0; i <= h->highest_lun; i++) {
2092                 int j;
2093                 drv_found = 0;
2094
2095                 /* skip holes in the array from already deleted drives */
2096                 if (h->drv[i].raid_level == -1)
2097                         continue;
2098
2099                 for (j = 0; j < num_luns; j++) {
2100                         memcpy(lunid, &ld_buff->LUN[j][0], sizeof(lunid));
2101                         if (memcmp(h->drv[i].LunID, lunid,
2102                                 sizeof(lunid)) == 0) {
2103                                 drv_found = 1;
2104                                 break;
2105                         }
2106                 }
2107                 if (!drv_found) {
2108                         /* Deregister it from the OS, it's gone. */
2109                         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2110                         h->drv[i].busy_configuring = 1;
2111                         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2112                         return_code = deregister_disk(h, i, 1, via_ioctl);
2113                         h->drv[i].busy_configuring = 0;
2114                 }
2115         }
2116
2117         /* Compare controller drive array to driver's drive array.
2118          * Check for updates in the drive information and any new drives
2119          * on the controller due to ACU adding logical drives, or changing
2120          * a logical drive's size, etc.  Reregister any new/changed drives
2121          */
2122         for (i = 0; i < num_luns; i++) {
2123                 int j;
2124
2125                 drv_found = 0;
2126
2127                 memcpy(lunid, &ld_buff->LUN[i][0], sizeof(lunid));
2128                 /* Find if the LUN is already in the drive array
2129                  * of the driver.  If so then update its info
2130                  * if not in use.  If it does not exist then find
2131                  * the first free index and add it.
2132                  */
2133                 for (j = 0; j <= h->highest_lun; j++) {
2134                         if (h->drv[j].raid_level != -1 &&
2135                                 memcmp(h->drv[j].LunID, lunid,
2136                                         sizeof(h->drv[j].LunID)) == 0) {
2137                                 drv_index = j;
2138                                 drv_found = 1;
2139                                 break;
2140                         }
2141                 }
2142
2143                 /* check if the drive was found already in the array */
2144                 if (!drv_found) {
2145                         drv_index = cciss_add_gendisk(h, lunid, 0);
2146                         if (drv_index == -1)
2147                                 goto freeret;
2148                 }
2149                 cciss_update_drive_info(ctlr, drv_index, first_time,
2150                         via_ioctl);
2151         }               /* end for */
2152
2153 freeret:
2154         kfree(ld_buff);
2155         h->busy_configuring = 0;
2156         /* We return -1 here to tell the ACU that we have registered/updated
2157          * all of the drives that we can and to keep it from calling us
2158          * additional times.
2159          */
2160         return -1;
2161 mem_msg:
2162         printk(KERN_ERR "cciss: out of memory\n");
2163         h->busy_configuring = 0;
2164         goto freeret;
2165 }
2166
2167 static void cciss_clear_drive_info(drive_info_struct *drive_info)
2168 {
2169         /* zero out the disk size info */
2170         drive_info->nr_blocks = 0;
2171         drive_info->block_size = 0;
2172         drive_info->heads = 0;
2173         drive_info->sectors = 0;
2174         drive_info->cylinders = 0;
2175         drive_info->raid_level = -1;
2176         memset(drive_info->serial_no, 0, sizeof(drive_info->serial_no));
2177         memset(drive_info->model, 0, sizeof(drive_info->model));
2178         memset(drive_info->rev, 0, sizeof(drive_info->rev));
2179         memset(drive_info->vendor, 0, sizeof(drive_info->vendor));
2180         /*
2181          * don't clear the LUNID though, we need to remember which
2182          * one this one is.
2183          */
2184 }
2185
2186 /* This function will deregister the disk and it's queue from the
2187  * kernel.  It must be called with the controller lock held and the
2188  * drv structures busy_configuring flag set.  It's parameters are:
2189  *
2190  * disk = This is the disk to be deregistered
2191  * drv  = This is the drive_info_struct associated with the disk to be
2192  *        deregistered.  It contains information about the disk used
2193  *        by the driver.
2194  * clear_all = This flag determines whether or not the disk information
2195  *             is going to be completely cleared out and the highest_lun
2196  *             reset.  Sometimes we want to clear out information about
2197  *             the disk in preparation for re-adding it.  In this case
2198  *             the highest_lun should be left unchanged and the LunID
2199  *             should not be cleared.
2200  * via_ioctl
2201  *    This indicates whether we've reached this path via ioctl.
2202  *    This affects the maximum usage count allowed for c0d0 to be messed with.
2203  *    If this path is reached via ioctl(), then the max_usage_count will
2204  *    be 1, as the process calling ioctl() has got to have the device open.
2205  *    If we get here via sysfs, then the max usage count will be zero.
2206 */
2207 static int deregister_disk(ctlr_info_t *h, int drv_index,
2208                            int clear_all, int via_ioctl)
2209 {
2210         int i;
2211         struct gendisk *disk;
2212         drive_info_struct *drv;
2213
2214         if (!capable(CAP_SYS_RAWIO))
2215                 return -EPERM;
2216
2217         drv = &h->drv[drv_index];
2218         disk = h->gendisk[drv_index];
2219
2220         /* make sure logical volume is NOT is use */
2221         if (clear_all || (h->gendisk[0] == disk)) {
2222                 if (drv->usage_count > via_ioctl)
2223                         return -EBUSY;
2224         } else if (drv->usage_count > 0)
2225                 return -EBUSY;
2226
2227         /* invalidate the devices and deregister the disk.  If it is disk
2228          * zero do not deregister it but just zero out it's values.  This
2229          * allows us to delete disk zero but keep the controller registered.
2230          */
2231         if (h->gendisk[0] != disk) {
2232                 struct request_queue *q = disk->queue;
2233                 if (disk->flags & GENHD_FL_UP) {
2234                         cciss_destroy_ld_sysfs_entry(h, drv_index, 0);
2235                         del_gendisk(disk);
2236                 }
2237                 if (q) {
2238                         blk_cleanup_queue(q);
2239                         /* Set drv->queue to NULL so that we do not try
2240                          * to call blk_start_queue on this queue in the
2241                          * interrupt handler
2242                          */
2243                         drv->queue = NULL;
2244                 }
2245                 /* If clear_all is set then we are deleting the logical
2246                  * drive, not just refreshing its info.  For drives
2247                  * other than disk 0 we will call put_disk.  We do not
2248                  * do this for disk 0 as we need it to be able to
2249                  * configure the controller.
2250                  */
2251                 if (clear_all){
2252                         /* This isn't pretty, but we need to find the
2253                          * disk in our array and NULL our the pointer.
2254                          * This is so that we will call alloc_disk if
2255                          * this index is used again later.
2256                          */
2257                         for (i=0; i < CISS_MAX_LUN; i++){
2258                                 if (h->gendisk[i] == disk) {
2259                                         h->gendisk[i] = NULL;
2260                                         break;
2261                                 }
2262                         }
2263                         put_disk(disk);
2264                 }
2265         } else {
2266                 set_capacity(disk, 0);
2267         }
2268
2269         --h->num_luns;
2270         cciss_clear_drive_info(drv);
2271
2272         if (clear_all) {
2273                 /* check to see if it was the last disk */
2274                 if (drv == h->drv + h->highest_lun) {
2275                         /* if so, find the new hightest lun */
2276                         int i, newhighest = -1;
2277                         for (i = 0; i <= h->highest_lun; i++) {
2278                                 /* if the disk has size > 0, it is available */
2279                                 if (h->drv[i].heads)
2280                                         newhighest = i;
2281                         }
2282                         h->highest_lun = newhighest;
2283                 }
2284                 memset(drv->LunID, 0, sizeof(drv->LunID));
2285         }
2286         return 0;
2287 }
2288
2289 static int fill_cmd(CommandList_struct *c, __u8 cmd, int ctlr, void *buff,
2290                 size_t size, __u8 page_code, unsigned char *scsi3addr,
2291                 int cmd_type)
2292 {
2293         ctlr_info_t *h = hba[ctlr];
2294         u64bit buff_dma_handle;
2295         int status = IO_OK;
2296
2297         c->cmd_type = CMD_IOCTL_PEND;
2298         c->Header.ReplyQueue = 0;
2299         if (buff != NULL) {
2300                 c->Header.SGList = 1;
2301                 c->Header.SGTotal = 1;
2302         } else {
2303                 c->Header.SGList = 0;
2304                 c->Header.SGTotal = 0;
2305         }
2306         c->Header.Tag.lower = c->busaddr;
2307         memcpy(c->Header.LUN.LunAddrBytes, scsi3addr, 8);
2308
2309         c->Request.Type.Type = cmd_type;
2310         if (cmd_type == TYPE_CMD) {
2311                 switch (cmd) {
2312                 case CISS_INQUIRY:
2313                         /* are we trying to read a vital product page */
2314                         if (page_code != 0) {
2315                                 c->Request.CDB[1] = 0x01;
2316                                 c->Request.CDB[2] = page_code;
2317                         }
2318                         c->Request.CDBLen = 6;
2319                         c->Request.Type.Attribute = ATTR_SIMPLE;
2320                         c->Request.Type.Direction = XFER_READ;
2321                         c->Request.Timeout = 0;
2322                         c->Request.CDB[0] = CISS_INQUIRY;
2323                         c->Request.CDB[4] = size & 0xFF;
2324                         break;
2325                 case CISS_REPORT_LOG:
2326                 case CISS_REPORT_PHYS:
2327                         /* Talking to controller so It's a physical command
2328                            mode = 00 target = 0.  Nothing to write.
2329                          */
2330                         c->Request.CDBLen = 12;
2331                         c->Request.Type.Attribute = ATTR_SIMPLE;
2332                         c->Request.Type.Direction = XFER_READ;
2333                         c->Request.Timeout = 0;
2334                         c->Request.CDB[0] = cmd;
2335                         c->Request.CDB[6] = (size >> 24) & 0xFF;        //MSB
2336                         c->Request.CDB[7] = (size >> 16) & 0xFF;
2337                         c->Request.CDB[8] = (size >> 8) & 0xFF;
2338                         c->Request.CDB[9] = size & 0xFF;
2339                         break;
2340
2341                 case CCISS_READ_CAPACITY:
2342                         c->Request.CDBLen = 10;
2343                         c->Request.Type.Attribute = ATTR_SIMPLE;
2344                         c->Request.Type.Direction = XFER_READ;
2345                         c->Request.Timeout = 0;
2346                         c->Request.CDB[0] = cmd;
2347                         break;
2348                 case CCISS_READ_CAPACITY_16:
2349                         c->Request.CDBLen = 16;
2350                         c->Request.Type.Attribute = ATTR_SIMPLE;
2351                         c->Request.Type.Direction = XFER_READ;
2352                         c->Request.Timeout = 0;
2353                         c->Request.CDB[0] = cmd;
2354                         c->Request.CDB[1] = 0x10;
2355                         c->Request.CDB[10] = (size >> 24) & 0xFF;
2356                         c->Request.CDB[11] = (size >> 16) & 0xFF;
2357                         c->Request.CDB[12] = (size >> 8) & 0xFF;
2358                         c->Request.CDB[13] = size & 0xFF;
2359                         c->Request.Timeout = 0;
2360                         c->Request.CDB[0] = cmd;
2361                         break;
2362                 case CCISS_CACHE_FLUSH:
2363                         c->Request.CDBLen = 12;
2364                         c->Request.Type.Attribute = ATTR_SIMPLE;
2365                         c->Request.Type.Direction = XFER_WRITE;
2366                         c->Request.Timeout = 0;
2367                         c->Request.CDB[0] = BMIC_WRITE;
2368                         c->Request.CDB[6] = BMIC_CACHE_FLUSH;
2369                         break;
2370                 case TEST_UNIT_READY:
2371                         c->Request.CDBLen = 6;
2372                         c->Request.Type.Attribute = ATTR_SIMPLE;
2373                         c->Request.Type.Direction = XFER_NONE;
2374                         c->Request.Timeout = 0;
2375                         break;
2376                 default:
2377                         printk(KERN_WARNING
2378                                "cciss%d:  Unknown Command 0x%c\n", ctlr, cmd);
2379                         return IO_ERROR;
2380                 }
2381         } else if (cmd_type == TYPE_MSG) {
2382                 switch (cmd) {
2383                 case 0: /* ABORT message */
2384                         c->Request.CDBLen = 12;
2385                         c->Request.Type.Attribute = ATTR_SIMPLE;
2386                         c->Request.Type.Direction = XFER_WRITE;
2387                         c->Request.Timeout = 0;
2388                         c->Request.CDB[0] = cmd;        /* abort */
2389                         c->Request.CDB[1] = 0;  /* abort a command */
2390                         /* buff contains the tag of the command to abort */
2391                         memcpy(&c->Request.CDB[4], buff, 8);
2392                         break;
2393                 case 1: /* RESET message */
2394                         c->Request.CDBLen = 16;
2395                         c->Request.Type.Attribute = ATTR_SIMPLE;
2396                         c->Request.Type.Direction = XFER_NONE;
2397                         c->Request.Timeout = 0;
2398                         memset(&c->Request.CDB[0], 0, sizeof(c->Request.CDB));
2399                         c->Request.CDB[0] = cmd;        /* reset */
2400                         c->Request.CDB[1] = 0x03;       /* reset a target */
2401                         break;
2402                 case 3: /* No-Op message */
2403                         c->Request.CDBLen = 1;
2404                         c->Request.Type.Attribute = ATTR_SIMPLE;
2405                         c->Request.Type.Direction = XFER_WRITE;
2406                         c->Request.Timeout = 0;
2407                         c->Request.CDB[0] = cmd;
2408                         break;
2409                 default:
2410                         printk(KERN_WARNING
2411                                "cciss%d: unknown message type %d\n", ctlr, cmd);
2412                         return IO_ERROR;
2413                 }
2414         } else {
2415                 printk(KERN_WARNING
2416                        "cciss%d: unknown command type %d\n", ctlr, cmd_type);
2417                 return IO_ERROR;
2418         }
2419         /* Fill in the scatter gather information */
2420         if (size > 0) {
2421                 buff_dma_handle.val = (__u64) pci_map_single(h->pdev,
2422                                                              buff, size,
2423                                                              PCI_DMA_BIDIRECTIONAL);
2424                 c->SG[0].Addr.lower = buff_dma_handle.val32.lower;
2425                 c->SG[0].Addr.upper = buff_dma_handle.val32.upper;
2426                 c->SG[0].Len = size;
2427                 c->SG[0].Ext = 0;       /* we are not chaining */
2428         }
2429         return status;
2430 }
2431
2432 static int check_target_status(ctlr_info_t *h, CommandList_struct *c)
2433 {
2434         switch (c->err_info->ScsiStatus) {
2435         case SAM_STAT_GOOD:
2436                 return IO_OK;
2437         case SAM_STAT_CHECK_CONDITION:
2438                 switch (0xf & c->err_info->SenseInfo[2]) {
2439                 case 0: return IO_OK; /* no sense */
2440                 case 1: return IO_OK; /* recovered error */
2441                 default:
2442                         printk(KERN_WARNING "cciss%d: cmd 0x%02x "
2443                                 "check condition, sense key = 0x%02x\n",
2444                                 h->ctlr, c->Request.CDB[0],
2445                                 c->err_info->SenseInfo[2]);
2446                 }
2447                 break;
2448         default:
2449                 printk(KERN_WARNING "cciss%d: cmd 0x%02x"
2450                         "scsi status = 0x%02x\n", h->ctlr,
2451                         c->Request.CDB[0], c->err_info->ScsiStatus);
2452                 break;
2453         }
2454         return IO_ERROR;
2455 }
2456
2457 static int process_sendcmd_error(ctlr_info_t *h, CommandList_struct *c)
2458 {
2459         int return_status = IO_OK;
2460
2461         if (c->err_info->CommandStatus == CMD_SUCCESS)
2462                 return IO_OK;
2463
2464         switch (c->err_info->CommandStatus) {
2465         case CMD_TARGET_STATUS:
2466                 return_status = check_target_status(h, c);
2467                 break;
2468         case CMD_DATA_UNDERRUN:
2469         case CMD_DATA_OVERRUN:
2470                 /* expected for inquiry and report lun commands */
2471                 break;
2472         case CMD_INVALID:
2473                 printk(KERN_WARNING "cciss: cmd 0x%02x is "
2474                        "reported invalid\n", c->Request.CDB[0]);
2475                 return_status = IO_ERROR;
2476                 break;
2477         case CMD_PROTOCOL_ERR:
2478                 printk(KERN_WARNING "cciss: cmd 0x%02x has "
2479                        "protocol error \n", c->Request.CDB[0]);
2480                 return_status = IO_ERROR;
2481                 break;
2482         case CMD_HARDWARE_ERR:
2483                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2484                        " hardware error\n", c->Request.CDB[0]);
2485                 return_status = IO_ERROR;
2486                 break;
2487         case CMD_CONNECTION_LOST:
2488                 printk(KERN_WARNING "cciss: cmd 0x%02x had "
2489                        "connection lost\n", c->Request.CDB[0]);
2490                 return_status = IO_ERROR;
2491                 break;
2492         case CMD_ABORTED:
2493                 printk(KERN_WARNING "cciss: cmd 0x%02x was "
2494                        "aborted\n", c->Request.CDB[0]);
2495                 return_status = IO_ERROR;
2496                 break;
2497         case CMD_ABORT_FAILED:
2498                 printk(KERN_WARNING "cciss: cmd 0x%02x reports "
2499                        "abort failed\n", c->Request.CDB[0]);
2500                 return_status = IO_ERROR;
2501                 break;
2502         case CMD_UNSOLICITED_ABORT:
2503                 printk(KERN_WARNING
2504                        "cciss%d: unsolicited abort 0x%02x\n", h->ctlr,
2505                         c->Request.CDB[0]);
2506                 return_status = IO_NEEDS_RETRY;
2507                 break;
2508         default:
2509                 printk(KERN_WARNING "cciss: cmd 0x%02x returned "
2510                        "unknown status %x\n", c->Request.CDB[0],
2511                        c->err_info->CommandStatus);
2512                 return_status = IO_ERROR;
2513         }
2514         return return_status;
2515 }
2516
2517 static int sendcmd_withirq_core(ctlr_info_t *h, CommandList_struct *c,
2518         int attempt_retry)
2519 {
2520         DECLARE_COMPLETION_ONSTACK(wait);
2521         u64bit buff_dma_handle;
2522         unsigned long flags;
2523         int return_status = IO_OK;
2524
2525 resend_cmd2:
2526         c->waiting = &wait;
2527         /* Put the request on the tail of the queue and send it */
2528         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
2529         addQ(&h->reqQ, c);
2530         h->Qdepth++;
2531         start_io(h);
2532         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
2533
2534         wait_for_completion(&wait);
2535
2536         if (c->err_info->CommandStatus == 0 || !attempt_retry)
2537                 goto command_done;
2538
2539         return_status = process_sendcmd_error(h, c);
2540
2541         if (return_status == IO_NEEDS_RETRY &&
2542                 c->retry_count < MAX_CMD_RETRIES) {
2543                 printk(KERN_WARNING "cciss%d: retrying 0x%02x\n", h->ctlr,
2544                         c->Request.CDB[0]);
2545                 c->retry_count++;
2546                 /* erase the old error information */
2547                 memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2548                 return_status = IO_OK;
2549                 INIT_COMPLETION(wait);
2550                 goto resend_cmd2;
2551         }
2552
2553 command_done:
2554         /* unlock the buffers from DMA */
2555         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2556         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2557         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2558                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2559         return return_status;
2560 }
2561
2562 static int sendcmd_withirq(__u8 cmd, int ctlr, void *buff, size_t size,
2563                            __u8 page_code, unsigned char scsi3addr[],
2564                         int cmd_type)
2565 {
2566         ctlr_info_t *h = hba[ctlr];
2567         CommandList_struct *c;
2568         int return_status;
2569
2570         c = cmd_alloc(h, 0);
2571         if (!c)
2572                 return -ENOMEM;
2573         return_status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2574                 scsi3addr, cmd_type);
2575         if (return_status == IO_OK)
2576                 return_status = sendcmd_withirq_core(h, c, 1);
2577
2578         cmd_free(h, c, 0);
2579         return return_status;
2580 }
2581
2582 static void cciss_geometry_inquiry(int ctlr, int logvol,
2583                                    int withirq, sector_t total_size,
2584                                    unsigned int block_size,
2585                                    InquiryData_struct *inq_buff,
2586                                    drive_info_struct *drv)
2587 {
2588         int return_code;
2589         unsigned long t;
2590         unsigned char scsi3addr[8];
2591
2592         memset(inq_buff, 0, sizeof(InquiryData_struct));
2593         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2594         if (withirq)
2595                 return_code = sendcmd_withirq(CISS_INQUIRY, ctlr,
2596                                               inq_buff, sizeof(*inq_buff),
2597                                               0xC1, scsi3addr, TYPE_CMD);
2598         else
2599                 return_code = sendcmd(CISS_INQUIRY, ctlr, inq_buff,
2600                                       sizeof(*inq_buff), 0xC1, scsi3addr,
2601                                       TYPE_CMD);
2602         if (return_code == IO_OK) {
2603                 if (inq_buff->data_byte[8] == 0xFF) {
2604                         printk(KERN_WARNING
2605                                "cciss: reading geometry failed, volume "
2606                                "does not support reading geometry\n");
2607                         drv->heads = 255;
2608                         drv->sectors = 32;      // Sectors per track
2609                         drv->cylinders = total_size + 1;
2610                         drv->raid_level = RAID_UNKNOWN;
2611                 } else {
2612                         drv->heads = inq_buff->data_byte[6];
2613                         drv->sectors = inq_buff->data_byte[7];
2614                         drv->cylinders = (inq_buff->data_byte[4] & 0xff) << 8;
2615                         drv->cylinders += inq_buff->data_byte[5];
2616                         drv->raid_level = inq_buff->data_byte[8];
2617                 }
2618                 drv->block_size = block_size;
2619                 drv->nr_blocks = total_size + 1;
2620                 t = drv->heads * drv->sectors;
2621                 if (t > 1) {
2622                         sector_t real_size = total_size + 1;
2623                         unsigned long rem = sector_div(real_size, t);
2624                         if (rem)
2625                                 real_size++;
2626                         drv->cylinders = real_size;
2627                 }
2628         } else {                /* Get geometry failed */
2629                 printk(KERN_WARNING "cciss: reading geometry failed\n");
2630         }
2631 }
2632
2633 static void
2634 cciss_read_capacity(int ctlr, int logvol, int withirq, sector_t *total_size,
2635                     unsigned int *block_size)
2636 {
2637         ReadCapdata_struct *buf;
2638         int return_code;
2639         unsigned char scsi3addr[8];
2640
2641         buf = kzalloc(sizeof(ReadCapdata_struct), GFP_KERNEL);
2642         if (!buf) {
2643                 printk(KERN_WARNING "cciss: out of memory\n");
2644                 return;
2645         }
2646
2647         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2648         if (withirq)
2649                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY,
2650                                 ctlr, buf, sizeof(ReadCapdata_struct),
2651                                         0, scsi3addr, TYPE_CMD);
2652         else
2653                 return_code = sendcmd(CCISS_READ_CAPACITY,
2654                                 ctlr, buf, sizeof(ReadCapdata_struct),
2655                                         0, scsi3addr, TYPE_CMD);
2656         if (return_code == IO_OK) {
2657                 *total_size = be32_to_cpu(*(__be32 *) buf->total_size);
2658                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2659         } else {                /* read capacity command failed */
2660                 printk(KERN_WARNING "cciss: read capacity failed\n");
2661                 *total_size = 0;
2662                 *block_size = BLOCK_SIZE;
2663         }
2664         kfree(buf);
2665 }
2666
2667 static void
2668 cciss_read_capacity_16(int ctlr, int logvol, int withirq, sector_t *total_size,                                 unsigned int *block_size)
2669 {
2670         ReadCapdata_struct_16 *buf;
2671         int return_code;
2672         unsigned char scsi3addr[8];
2673
2674         buf = kzalloc(sizeof(ReadCapdata_struct_16), GFP_KERNEL);
2675         if (!buf) {
2676                 printk(KERN_WARNING "cciss: out of memory\n");
2677                 return;
2678         }
2679
2680         log_unit_to_scsi3addr(hba[ctlr], scsi3addr, logvol);
2681         if (withirq) {
2682                 return_code = sendcmd_withirq(CCISS_READ_CAPACITY_16,
2683                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2684                                 0, scsi3addr, TYPE_CMD);
2685         }
2686         else {
2687                 return_code = sendcmd(CCISS_READ_CAPACITY_16,
2688                         ctlr, buf, sizeof(ReadCapdata_struct_16),
2689                                 0, scsi3addr, TYPE_CMD);
2690         }
2691         if (return_code == IO_OK) {
2692                 *total_size = be64_to_cpu(*(__be64 *) buf->total_size);
2693                 *block_size = be32_to_cpu(*(__be32 *) buf->block_size);
2694         } else {                /* read capacity command failed */
2695                 printk(KERN_WARNING "cciss: read capacity failed\n");
2696                 *total_size = 0;
2697                 *block_size = BLOCK_SIZE;
2698         }
2699         printk(KERN_INFO "      blocks= %llu block_size= %d\n",
2700                (unsigned long long)*total_size+1, *block_size);
2701         kfree(buf);
2702 }
2703
2704 static int cciss_revalidate(struct gendisk *disk)
2705 {
2706         ctlr_info_t *h = get_host(disk);
2707         drive_info_struct *drv = get_drv(disk);
2708         int logvol;
2709         int FOUND = 0;
2710         unsigned int block_size;
2711         sector_t total_size;
2712         InquiryData_struct *inq_buff = NULL;
2713
2714         for (logvol = 0; logvol < CISS_MAX_LUN; logvol++) {
2715                 if (memcmp(h->drv[logvol].LunID, drv->LunID,
2716                         sizeof(drv->LunID)) == 0) {
2717                         FOUND = 1;
2718                         break;
2719                 }
2720         }
2721
2722         if (!FOUND)
2723                 return 1;
2724
2725         inq_buff = kmalloc(sizeof(InquiryData_struct), GFP_KERNEL);
2726         if (inq_buff == NULL) {
2727                 printk(KERN_WARNING "cciss: out of memory\n");
2728                 return 1;
2729         }
2730         if (h->cciss_read == CCISS_READ_10) {
2731                 cciss_read_capacity(h->ctlr, logvol, 1,
2732                                         &total_size, &block_size);
2733         } else {
2734                 cciss_read_capacity_16(h->ctlr, logvol, 1,
2735                                         &total_size, &block_size);
2736         }
2737         cciss_geometry_inquiry(h->ctlr, logvol, 1, total_size, block_size,
2738                                inq_buff, drv);
2739
2740         blk_queue_logical_block_size(drv->queue, drv->block_size);
2741         set_capacity(disk, drv->nr_blocks);
2742
2743         kfree(inq_buff);
2744         return 0;
2745 }
2746
2747 /*
2748  *   Wait polling for a command to complete.
2749  *   The memory mapped FIFO is polled for the completion.
2750  *   Used only at init time, interrupts from the HBA are disabled.
2751  */
2752 static unsigned long pollcomplete(int ctlr)
2753 {
2754         unsigned long done;
2755         int i;
2756
2757         /* Wait (up to 20 seconds) for a command to complete */
2758
2759         for (i = 20 * HZ; i > 0; i--) {
2760                 done = hba[ctlr]->access.command_completed(hba[ctlr]);
2761                 if (done == FIFO_EMPTY)
2762                         schedule_timeout_uninterruptible(1);
2763                 else
2764                         return done;
2765         }
2766         /* Invalid address to tell caller we ran out of time */
2767         return 1;
2768 }
2769
2770 /* Send command c to controller h and poll for it to complete.
2771  * Turns interrupts off on the board.  Used at driver init time
2772  * and during SCSI error recovery.
2773  */
2774 static int sendcmd_core(ctlr_info_t *h, CommandList_struct *c)
2775 {
2776         int i;
2777         unsigned long complete;
2778         int status = IO_ERROR;
2779         u64bit buff_dma_handle;
2780
2781 resend_cmd1:
2782
2783         /* Disable interrupt on the board. */
2784         h->access.set_intr_mask(h, CCISS_INTR_OFF);
2785
2786         /* Make sure there is room in the command FIFO */
2787         /* Actually it should be completely empty at this time */
2788         /* unless we are in here doing error handling for the scsi */
2789         /* tape side of the driver. */
2790         for (i = 200000; i > 0; i--) {
2791                 /* if fifo isn't full go */
2792                 if (!(h->access.fifo_full(h)))
2793                         break;
2794                 udelay(10);
2795                 printk(KERN_WARNING "cciss cciss%d: SendCmd FIFO full,"
2796                        " waiting!\n", h->ctlr);
2797         }
2798         h->access.submit_command(h, c); /* Send the cmd */
2799         do {
2800                 complete = pollcomplete(h->ctlr);
2801
2802 #ifdef CCISS_DEBUG
2803                 printk(KERN_DEBUG "cciss: command completed\n");
2804 #endif                          /* CCISS_DEBUG */
2805
2806                 if (complete == 1) {
2807                         printk(KERN_WARNING
2808                                "cciss cciss%d: SendCmd Timeout out, "
2809                                "No command list address returned!\n", h->ctlr);
2810                         status = IO_ERROR;
2811                         break;
2812                 }
2813
2814                 /* Make sure it's the command we're expecting. */
2815                 if ((complete & ~CISS_ERROR_BIT) != c->busaddr) {
2816                         printk(KERN_WARNING "cciss%d: Unexpected command "
2817                                 "completion.\n", h->ctlr);
2818                         continue;
2819                 }
2820
2821                 /* It is our command.  If no error, we're done. */
2822                 if (!(complete & CISS_ERROR_BIT)) {
2823                         status = IO_OK;
2824                         break;
2825                 }
2826
2827                 /* There is an error... */
2828
2829                 /* if data overrun or underun on Report command ignore it */
2830                 if (((c->Request.CDB[0] == CISS_REPORT_LOG) ||
2831                      (c->Request.CDB[0] == CISS_REPORT_PHYS) ||
2832                      (c->Request.CDB[0] == CISS_INQUIRY)) &&
2833                         ((c->err_info->CommandStatus == CMD_DATA_OVERRUN) ||
2834                          (c->err_info->CommandStatus == CMD_DATA_UNDERRUN))) {
2835                         complete = c->busaddr;
2836                         status = IO_OK;
2837                         break;
2838                 }
2839
2840                 if (c->err_info->CommandStatus == CMD_UNSOLICITED_ABORT) {
2841                         printk(KERN_WARNING "cciss%d: unsolicited abort %p\n",
2842                                 h->ctlr, c);
2843                         if (c->retry_count < MAX_CMD_RETRIES) {
2844                                 printk(KERN_WARNING "cciss%d: retrying %p\n",
2845                                    h->ctlr, c);
2846                                 c->retry_count++;
2847                                 /* erase the old error information */
2848                                 memset(c->err_info, 0, sizeof(c->err_info));
2849                                 goto resend_cmd1;
2850                         }
2851                         printk(KERN_WARNING "cciss%d: retried %p too many "
2852                                 "times\n", h->ctlr, c);
2853                         status = IO_ERROR;
2854                         break;
2855                 }
2856
2857                 if (c->err_info->CommandStatus == CMD_UNABORTABLE) {
2858                         printk(KERN_WARNING "cciss%d: command could not be "
2859                                 "aborted.\n", h->ctlr);
2860                         status = IO_ERROR;
2861                         break;
2862                 }
2863
2864                 if (c->err_info->CommandStatus == CMD_TARGET_STATUS) {
2865                         status = check_target_status(h, c);
2866                         break;
2867                 }
2868
2869                 printk(KERN_WARNING "cciss%d: sendcmd error\n", h->ctlr);
2870                 printk(KERN_WARNING "cmd = 0x%02x, CommandStatus = 0x%02x\n",
2871                         c->Request.CDB[0], c->err_info->CommandStatus);
2872                 status = IO_ERROR;
2873                 break;
2874
2875         } while (1);
2876
2877         /* unlock the data buffer from DMA */
2878         buff_dma_handle.val32.lower = c->SG[0].Addr.lower;
2879         buff_dma_handle.val32.upper = c->SG[0].Addr.upper;
2880         pci_unmap_single(h->pdev, (dma_addr_t) buff_dma_handle.val,
2881                          c->SG[0].Len, PCI_DMA_BIDIRECTIONAL);
2882         return status;
2883 }
2884
2885 /*
2886  * Send a command to the controller, and wait for it to complete.
2887  * Used at init time, and during SCSI error recovery.
2888  */
2889 static int sendcmd(__u8 cmd, int ctlr, void *buff, size_t size,
2890         __u8 page_code, unsigned char *scsi3addr, int cmd_type)
2891 {
2892         CommandList_struct *c;
2893         int status;
2894
2895         c = cmd_alloc(hba[ctlr], 1);
2896         if (!c) {
2897                 printk(KERN_WARNING "cciss: unable to get memory");
2898                 return IO_ERROR;
2899         }
2900         status = fill_cmd(c, cmd, ctlr, buff, size, page_code,
2901                 scsi3addr, cmd_type);
2902         if (status == IO_OK)
2903                 status = sendcmd_core(hba[ctlr], c);
2904         cmd_free(hba[ctlr], c, 1);
2905         return status;
2906 }
2907
2908 /*
2909  * Map (physical) PCI mem into (virtual) kernel space
2910  */
2911 static void __iomem *remap_pci_mem(ulong base, ulong size)
2912 {
2913         ulong page_base = ((ulong) base) & PAGE_MASK;
2914         ulong page_offs = ((ulong) base) - page_base;
2915         void __iomem *page_remapped = ioremap(page_base, page_offs + size);
2916
2917         return page_remapped ? (page_remapped + page_offs) : NULL;
2918 }
2919
2920 /*
2921  * Takes jobs of the Q and sends them to the hardware, then puts it on
2922  * the Q to wait for completion.
2923  */
2924 static void start_io(ctlr_info_t *h)
2925 {
2926         CommandList_struct *c;
2927
2928         while (!hlist_empty(&h->reqQ)) {
2929                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
2930                 /* can't do anything if fifo is full */
2931                 if ((h->access.fifo_full(h))) {
2932                         printk(KERN_WARNING "cciss: fifo full\n");
2933                         break;
2934                 }
2935
2936                 /* Get the first entry from the Request Q */
2937                 removeQ(c);
2938                 h->Qdepth--;
2939
2940                 /* Tell the controller execute command */
2941                 h->access.submit_command(h, c);
2942
2943                 /* Put job onto the completed Q */
2944                 addQ(&h->cmpQ, c);
2945         }
2946 }
2947
2948 /* Assumes that CCISS_LOCK(h->ctlr) is held. */
2949 /* Zeros out the error record and then resends the command back */
2950 /* to the controller */
2951 static inline void resend_cciss_cmd(ctlr_info_t *h, CommandList_struct *c)
2952 {
2953         /* erase the old error information */
2954         memset(c->err_info, 0, sizeof(ErrorInfo_struct));
2955
2956         /* add it to software queue and then send it to the controller */
2957         addQ(&h->reqQ, c);
2958         h->Qdepth++;
2959         if (h->Qdepth > h->maxQsinceinit)
2960                 h->maxQsinceinit = h->Qdepth;
2961
2962         start_io(h);
2963 }
2964
2965 static inline unsigned int make_status_bytes(unsigned int scsi_status_byte,
2966         unsigned int msg_byte, unsigned int host_byte,
2967         unsigned int driver_byte)
2968 {
2969         /* inverse of macros in scsi.h */
2970         return (scsi_status_byte & 0xff) |
2971                 ((msg_byte & 0xff) << 8) |
2972                 ((host_byte & 0xff) << 16) |
2973                 ((driver_byte & 0xff) << 24);
2974 }
2975
2976 static inline int evaluate_target_status(ctlr_info_t *h,
2977                         CommandList_struct *cmd, int *retry_cmd)
2978 {
2979         unsigned char sense_key;
2980         unsigned char status_byte, msg_byte, host_byte, driver_byte;
2981         int error_value;
2982
2983         *retry_cmd = 0;
2984         /* If we get in here, it means we got "target status", that is, scsi status */
2985         status_byte = cmd->err_info->ScsiStatus;
2986         driver_byte = DRIVER_OK;
2987         msg_byte = cmd->err_info->CommandStatus; /* correct?  seems too device specific */
2988
2989         if (blk_pc_request(cmd->rq))
2990                 host_byte = DID_PASSTHROUGH;
2991         else
2992                 host_byte = DID_OK;
2993
2994         error_value = make_status_bytes(status_byte, msg_byte,
2995                 host_byte, driver_byte);
2996
2997         if (cmd->err_info->ScsiStatus != SAM_STAT_CHECK_CONDITION) {
2998                 if (!blk_pc_request(cmd->rq))
2999                         printk(KERN_WARNING "cciss: cmd %p "
3000                                "has SCSI Status 0x%x\n",
3001                                cmd, cmd->err_info->ScsiStatus);
3002                 return error_value;
3003         }
3004
3005         /* check the sense key */
3006         sense_key = 0xf & cmd->err_info->SenseInfo[2];
3007         /* no status or recovered error */
3008         if (((sense_key == 0x0) || (sense_key == 0x1)) && !blk_pc_request(cmd->rq))
3009                 error_value = 0;
3010
3011         if (check_for_unit_attention(h, cmd)) {
3012                 *retry_cmd = !blk_pc_request(cmd->rq);
3013                 return 0;
3014         }
3015
3016         if (!blk_pc_request(cmd->rq)) { /* Not SG_IO or similar? */
3017                 if (error_value != 0)
3018                         printk(KERN_WARNING "cciss: cmd %p has CHECK CONDITION"
3019                                " sense key = 0x%x\n", cmd, sense_key);
3020                 return error_value;
3021         }
3022
3023         /* SG_IO or similar, copy sense data back */
3024         if (cmd->rq->sense) {
3025                 if (cmd->rq->sense_len > cmd->err_info->SenseLen)
3026                         cmd->rq->sense_len = cmd->err_info->SenseLen;
3027                 memcpy(cmd->rq->sense, cmd->err_info->SenseInfo,
3028                         cmd->rq->sense_len);
3029         } else
3030                 cmd->rq->sense_len = 0;
3031
3032         return error_value;
3033 }
3034
3035 /* checks the status of the job and calls complete buffers to mark all
3036  * buffers for the completed job. Note that this function does not need
3037  * to hold the hba/queue lock.
3038  */
3039 static inline void complete_command(ctlr_info_t *h, CommandList_struct *cmd,
3040                                     int timeout)
3041 {
3042         int retry_cmd = 0;
3043         struct request *rq = cmd->rq;
3044
3045         rq->errors = 0;
3046
3047         if (timeout)
3048                 rq->errors = make_status_bytes(0, 0, 0, DRIVER_TIMEOUT);
3049
3050         if (cmd->err_info->CommandStatus == 0)  /* no error has occurred */
3051                 goto after_error_processing;
3052
3053         switch (cmd->err_info->CommandStatus) {
3054         case CMD_TARGET_STATUS:
3055                 rq->errors = evaluate_target_status(h, cmd, &retry_cmd);
3056                 break;
3057         case CMD_DATA_UNDERRUN:
3058                 if (blk_fs_request(cmd->rq)) {
3059                         printk(KERN_WARNING "cciss: cmd %p has"
3060                                " completed with data underrun "
3061                                "reported\n", cmd);
3062                         cmd->rq->resid_len = cmd->err_info->ResidualCnt;
3063                 }
3064                 break;
3065         case CMD_DATA_OVERRUN:
3066                 if (blk_fs_request(cmd->rq))
3067                         printk(KERN_WARNING "cciss: cmd %p has"
3068                                " completed with data overrun "
3069                                "reported\n", cmd);
3070                 break;
3071         case CMD_INVALID:
3072                 printk(KERN_WARNING "cciss: cmd %p is "
3073                        "reported invalid\n", cmd);
3074                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3075                         cmd->err_info->CommandStatus, DRIVER_OK,
3076                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3077                 break;
3078         case CMD_PROTOCOL_ERR:
3079                 printk(KERN_WARNING "cciss: cmd %p has "
3080                        "protocol error \n", cmd);
3081                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3082                         cmd->err_info->CommandStatus, DRIVER_OK,
3083                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3084                 break;
3085         case CMD_HARDWARE_ERR:
3086                 printk(KERN_WARNING "cciss: cmd %p had "
3087                        " hardware error\n", cmd);
3088                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3089                         cmd->err_info->CommandStatus, DRIVER_OK,
3090                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3091                 break;
3092         case CMD_CONNECTION_LOST:
3093                 printk(KERN_WARNING "cciss: cmd %p had "
3094                        "connection lost\n", cmd);
3095                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3096                         cmd->err_info->CommandStatus, DRIVER_OK,
3097                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3098                 break;
3099         case CMD_ABORTED:
3100                 printk(KERN_WARNING "cciss: cmd %p was "
3101                        "aborted\n", cmd);
3102                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3103                         cmd->err_info->CommandStatus, DRIVER_OK,
3104                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3105                 break;
3106         case CMD_ABORT_FAILED:
3107                 printk(KERN_WARNING "cciss: cmd %p reports "
3108                        "abort failed\n", cmd);
3109                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3110                         cmd->err_info->CommandStatus, DRIVER_OK,
3111                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3112                 break;
3113         case CMD_UNSOLICITED_ABORT:
3114                 printk(KERN_WARNING "cciss%d: unsolicited "
3115                        "abort %p\n", h->ctlr, cmd);
3116                 if (cmd->retry_count < MAX_CMD_RETRIES) {
3117                         retry_cmd = 1;
3118                         printk(KERN_WARNING
3119                                "cciss%d: retrying %p\n", h->ctlr, cmd);
3120                         cmd->retry_count++;
3121                 } else
3122                         printk(KERN_WARNING
3123                                "cciss%d: %p retried too "
3124                                "many times\n", h->ctlr, cmd);
3125                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3126                         cmd->err_info->CommandStatus, DRIVER_OK,
3127                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ABORT);
3128                 break;
3129         case CMD_TIMEOUT:
3130                 printk(KERN_WARNING "cciss: cmd %p timedout\n", cmd);
3131                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3132                         cmd->err_info->CommandStatus, DRIVER_OK,
3133                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3134                 break;
3135         default:
3136                 printk(KERN_WARNING "cciss: cmd %p returned "
3137                        "unknown status %x\n", cmd,
3138                        cmd->err_info->CommandStatus);
3139                 rq->errors = make_status_bytes(SAM_STAT_GOOD,
3140                         cmd->err_info->CommandStatus, DRIVER_OK,
3141                         blk_pc_request(cmd->rq) ? DID_PASSTHROUGH : DID_ERROR);
3142         }
3143
3144 after_error_processing:
3145
3146         /* We need to return this command */
3147         if (retry_cmd) {
3148                 resend_cciss_cmd(h, cmd);
3149                 return;
3150         }
3151         cmd->rq->completion_data = cmd;
3152         blk_complete_request(cmd->rq);
3153 }
3154
3155 /*
3156  * Get a request and submit it to the controller.
3157  */
3158 static void do_cciss_request(struct request_queue *q)
3159 {
3160         ctlr_info_t *h = q->queuedata;
3161         CommandList_struct *c;
3162         sector_t start_blk;
3163         int seg;
3164         struct request *creq;
3165         u64bit temp64;
3166         struct scatterlist tmp_sg[MAXSGENTRIES];
3167         drive_info_struct *drv;
3168         int i, dir;
3169
3170         /* We call start_io here in case there is a command waiting on the
3171          * queue that has not been sent.
3172          */
3173         if (blk_queue_plugged(q))
3174                 goto startio;
3175
3176       queue:
3177         creq = blk_peek_request(q);
3178         if (!creq)
3179                 goto startio;
3180
3181         BUG_ON(creq->nr_phys_segments > MAXSGENTRIES);
3182
3183         if ((c = cmd_alloc(h, 1)) == NULL)
3184                 goto full;
3185
3186         blk_start_request(creq);
3187
3188         spin_unlock_irq(q->queue_lock);
3189
3190         c->cmd_type = CMD_RWREQ;
3191         c->rq = creq;
3192
3193         /* fill in the request */
3194         drv = creq->rq_disk->private_data;
3195         c->Header.ReplyQueue = 0;       // unused in simple mode
3196         /* got command from pool, so use the command block index instead */
3197         /* for direct lookups. */
3198         /* The first 2 bits are reserved for controller error reporting. */
3199         c->Header.Tag.lower = (c->cmdindex << 3);
3200         c->Header.Tag.lower |= 0x04;    /* flag for direct lookup. */
3201         memcpy(&c->Header.LUN, drv->LunID, sizeof(drv->LunID));
3202         c->Request.CDBLen = 10; // 12 byte commands not in FW yet;
3203         c->Request.Type.Type = TYPE_CMD;        // It is a command.
3204         c->Request.Type.Attribute = ATTR_SIMPLE;
3205         c->Request.Type.Direction =
3206             (rq_data_dir(creq) == READ) ? XFER_READ : XFER_WRITE;
3207         c->Request.Timeout = 0; // Don't time out
3208         c->Request.CDB[0] =
3209             (rq_data_dir(creq) == READ) ? h->cciss_read : h->cciss_write;
3210         start_blk = blk_rq_pos(creq);
3211 #ifdef CCISS_DEBUG
3212         printk(KERN_DEBUG "ciss: sector =%d nr_sectors=%d\n",
3213                (int)blk_rq_pos(creq), (int)blk_rq_sectors(creq));
3214 #endif                          /* CCISS_DEBUG */
3215
3216         sg_init_table(tmp_sg, MAXSGENTRIES);
3217         seg = blk_rq_map_sg(q, creq, tmp_sg);
3218
3219         /* get the DMA records for the setup */
3220         if (c->Request.Type.Direction == XFER_READ)
3221                 dir = PCI_DMA_FROMDEVICE;
3222         else
3223                 dir = PCI_DMA_TODEVICE;
3224
3225         for (i = 0; i < seg; i++) {
3226                 c->SG[i].Len = tmp_sg[i].length;
3227                 temp64.val = (__u64) pci_map_page(h->pdev, sg_page(&tmp_sg[i]),
3228                                                   tmp_sg[i].offset,
3229                                                   tmp_sg[i].length, dir);
3230                 c->SG[i].Addr.lower = temp64.val32.lower;
3231                 c->SG[i].Addr.upper = temp64.val32.upper;
3232                 c->SG[i].Ext = 0;       // we are not chaining
3233         }
3234         /* track how many SG entries we are using */
3235         if (seg > h->maxSG)
3236                 h->maxSG = seg;
3237
3238 #ifdef CCISS_DEBUG
3239         printk(KERN_DEBUG "cciss: Submitting %u sectors in %d segments\n",
3240                blk_rq_sectors(creq), seg);
3241 #endif                          /* CCISS_DEBUG */
3242
3243         c->Header.SGList = c->Header.SGTotal = seg;
3244         if (likely(blk_fs_request(creq))) {
3245                 if(h->cciss_read == CCISS_READ_10) {
3246                         c->Request.CDB[1] = 0;
3247                         c->Request.CDB[2] = (start_blk >> 24) & 0xff;   //MSB
3248                         c->Request.CDB[3] = (start_blk >> 16) & 0xff;
3249                         c->Request.CDB[4] = (start_blk >> 8) & 0xff;
3250                         c->Request.CDB[5] = start_blk & 0xff;
3251                         c->Request.CDB[6] = 0;  // (sect >> 24) & 0xff; MSB
3252                         c->Request.CDB[7] = (blk_rq_sectors(creq) >> 8) & 0xff;
3253                         c->Request.CDB[8] = blk_rq_sectors(creq) & 0xff;
3254                         c->Request.CDB[9] = c->Request.CDB[11] = c->Request.CDB[12] = 0;
3255                 } else {
3256                         u32 upper32 = upper_32_bits(start_blk);
3257
3258                         c->Request.CDBLen = 16;
3259                         c->Request.CDB[1]= 0;
3260                         c->Request.CDB[2]= (upper32 >> 24) & 0xff;      //MSB
3261                         c->Request.CDB[3]= (upper32 >> 16) & 0xff;
3262                         c->Request.CDB[4]= (upper32 >>  8) & 0xff;
3263                         c->Request.CDB[5]= upper32 & 0xff;
3264                         c->Request.CDB[6]= (start_blk >> 24) & 0xff;
3265                         c->Request.CDB[7]= (start_blk >> 16) & 0xff;
3266                         c->Request.CDB[8]= (start_blk >>  8) & 0xff;
3267                         c->Request.CDB[9]= start_blk & 0xff;
3268                         c->Request.CDB[10]= (blk_rq_sectors(creq) >> 24) & 0xff;
3269                         c->Request.CDB[11]= (blk_rq_sectors(creq) >> 16) & 0xff;
3270                         c->Request.CDB[12]= (blk_rq_sectors(creq) >>  8) & 0xff;
3271                         c->Request.CDB[13]= blk_rq_sectors(creq) & 0xff;
3272                         c->Request.CDB[14] = c->Request.CDB[15] = 0;
3273                 }
3274         } else if (blk_pc_request(creq)) {
3275                 c->Request.CDBLen = creq->cmd_len;
3276                 memcpy(c->Request.CDB, creq->cmd, BLK_MAX_CDB);
3277         } else {
3278                 printk(KERN_WARNING "cciss%d: bad request type %d\n", h->ctlr, creq->cmd_type);
3279                 BUG();
3280         }
3281
3282         spin_lock_irq(q->queue_lock);
3283
3284         addQ(&h->reqQ, c);
3285         h->Qdepth++;
3286         if (h->Qdepth > h->maxQsinceinit)
3287                 h->maxQsinceinit = h->Qdepth;
3288
3289         goto queue;
3290 full:
3291         blk_stop_queue(q);
3292 startio:
3293         /* We will already have the driver lock here so not need
3294          * to lock it.
3295          */
3296         start_io(h);
3297 }
3298
3299 static inline unsigned long get_next_completion(ctlr_info_t *h)
3300 {
3301         return h->access.command_completed(h);
3302 }
3303
3304 static inline int interrupt_pending(ctlr_info_t *h)
3305 {
3306         return h->access.intr_pending(h);
3307 }
3308
3309 static inline long interrupt_not_for_us(ctlr_info_t *h)
3310 {
3311         return (((h->access.intr_pending(h) == 0) ||
3312                  (h->interrupts_enabled == 0)));
3313 }
3314
3315 static irqreturn_t do_cciss_intr(int irq, void *dev_id)
3316 {
3317         ctlr_info_t *h = dev_id;
3318         CommandList_struct *c;
3319         unsigned long flags;
3320         __u32 a, a1, a2;
3321
3322         if (interrupt_not_for_us(h))
3323                 return IRQ_NONE;
3324         /*
3325          * If there are completed commands in the completion queue,
3326          * we had better do something about it.
3327          */
3328         spin_lock_irqsave(CCISS_LOCK(h->ctlr), flags);
3329         while (interrupt_pending(h)) {
3330                 while ((a = get_next_completion(h)) != FIFO_EMPTY) {
3331                         a1 = a;
3332                         if ((a & 0x04)) {
3333                                 a2 = (a >> 3);
3334                                 if (a2 >= h->nr_cmds) {
3335                                         printk(KERN_WARNING
3336                                                "cciss: controller cciss%d failed, stopping.\n",
3337                                                h->ctlr);
3338                                         fail_all_cmds(h->ctlr);
3339                                         return IRQ_HANDLED;
3340                                 }
3341
3342                                 c = h->cmd_pool + a2;
3343                                 a = c->busaddr;
3344
3345                         } else {
3346                                 struct hlist_node *tmp;
3347
3348                                 a &= ~3;
3349                                 c = NULL;
3350                                 hlist_for_each_entry(c, tmp, &h->cmpQ, list) {
3351                                         if (c->busaddr == a)
3352                                                 break;
3353                                 }
3354                         }
3355                         /*
3356                          * If we've found the command, take it off the
3357                          * completion Q and free it
3358                          */
3359                         if (c && c->busaddr == a) {
3360                                 removeQ(c);
3361                                 if (c->cmd_type == CMD_RWREQ) {
3362                                         complete_command(h, c, 0);
3363                                 } else if (c->cmd_type == CMD_IOCTL_PEND) {
3364                                         complete(c->waiting);
3365                                 }
3366 #                               ifdef CONFIG_CISS_SCSI_TAPE
3367                                 else if (c->cmd_type == CMD_SCSI)
3368                                         complete_scsi_command(c, 0, a1);
3369 #                               endif
3370                                 continue;
3371                         }
3372                 }
3373         }
3374
3375         spin_unlock_irqrestore(CCISS_LOCK(h->ctlr), flags);
3376         return IRQ_HANDLED;
3377 }
3378
3379 /**
3380  * add_to_scan_list() - add controller to rescan queue
3381  * @h:                Pointer to the controller.
3382  *
3383  * Adds the controller to the rescan queue if not already on the queue.
3384  *
3385  * returns 1 if added to the queue, 0 if skipped (could be on the
3386  * queue already, or the controller could be initializing or shutting
3387  * down).
3388  **/
3389 static int add_to_scan_list(struct ctlr_info *h)
3390 {
3391         struct ctlr_info *test_h;
3392         int found = 0;
3393         int ret = 0;
3394
3395         if (h->busy_initializing)
3396                 return 0;
3397
3398         if (!mutex_trylock(&h->busy_shutting_down))
3399                 return 0;
3400
3401         mutex_lock(&scan_mutex);
3402         list_for_each_entry(test_h, &scan_q, scan_list) {
3403                 if (test_h == h) {
3404                         found = 1;
3405                         break;
3406                 }
3407         }
3408         if (!found && !h->busy_scanning) {
3409                 INIT_COMPLETION(h->scan_wait);
3410                 list_add_tail(&h->scan_list, &scan_q);
3411                 ret = 1;
3412         }
3413         mutex_unlock(&scan_mutex);
3414         mutex_unlock(&h->busy_shutting_down);
3415
3416         return ret;
3417 }
3418
3419 /**
3420  * remove_from_scan_list() - remove controller from rescan queue
3421  * @h:                     Pointer to the controller.
3422  *
3423  * Removes the controller from the rescan queue if present. Blocks if
3424  * the controller is currently conducting a rescan.
3425  **/
3426 static void remove_from_scan_list(struct ctlr_info *h)
3427 {
3428         struct ctlr_info *test_h, *tmp_h;
3429         int scanning = 0;
3430
3431         mutex_lock(&scan_mutex);
3432         list_for_each_entry_safe(test_h, tmp_h, &scan_q, scan_list) {
3433                 if (test_h == h) {
3434                         list_del(&h->scan_list);
3435                         complete_all(&h->scan_wait);
3436                         mutex_unlock(&scan_mutex);
3437                         return;
3438                 }
3439         }
3440         if (&h->busy_scanning)
3441                 scanning = 0;
3442         mutex_unlock(&scan_mutex);
3443
3444         if (scanning)
3445                 wait_for_completion(&h->scan_wait);
3446 }
3447
3448 /**
3449  * scan_thread() - kernel thread used to rescan controllers
3450  * @data:        Ignored.
3451  *
3452  * A kernel thread used scan for drive topology changes on
3453  * controllers. The thread processes only one controller at a time
3454  * using a queue.  Controllers are added to the queue using
3455  * add_to_scan_list() and removed from the queue either after done
3456  * processing or using remove_from_scan_list().
3457  *
3458  * returns 0.
3459  **/
3460 static int scan_thread(void *data)
3461 {
3462         struct ctlr_info *h;
3463
3464         while (1) {
3465                 set_current_state(TASK_INTERRUPTIBLE);
3466                 schedule();
3467                 if (kthread_should_stop())
3468                         break;
3469
3470                 while (1) {
3471                         mutex_lock(&scan_mutex);
3472                         if (list_empty(&scan_q)) {
3473                                 mutex_unlock(&scan_mutex);
3474                                 break;
3475                         }
3476
3477                         h = list_entry(scan_q.next,
3478                                        struct ctlr_info,
3479                                        scan_list);
3480                         list_del(&h->scan_list);
3481                         h->busy_scanning = 1;
3482                         mutex_unlock(&scan_mutex);
3483
3484                         if (h) {
3485                                 rebuild_lun_table(h, 0, 0);
3486                                 complete_all(&h->scan_wait);
3487                                 mutex_lock(&scan_mutex);
3488                                 h->busy_scanning = 0;
3489                                 mutex_unlock(&scan_mutex);
3490                         }
3491                 }
3492         }
3493
3494         return 0;
3495 }
3496
3497 static int check_for_unit_attention(ctlr_info_t *h, CommandList_struct *c)
3498 {
3499         if (c->err_info->SenseInfo[2] != UNIT_ATTENTION)
3500                 return 0;
3501
3502         switch (c->err_info->SenseInfo[12]) {
3503         case STATE_CHANGED:
3504                 printk(KERN_WARNING "cciss%d: a state change "
3505                         "detected, command retried\n", h->ctlr);
3506                 return 1;
3507         break;
3508         case LUN_FAILED:
3509                 printk(KERN_WARNING "cciss%d: LUN failure "
3510                         "detected, action required\n", h->ctlr);
3511                 return 1;
3512         break;
3513         case REPORT_LUNS_CHANGED:
3514                 printk(KERN_WARNING "cciss%d: report LUN data "
3515                         "changed\n", h->ctlr);
3516                 add_to_scan_list(h);
3517                 wake_up_process(cciss_scan_thread);
3518                 return 1;
3519         break;
3520         case POWER_OR_RESET:
3521                 printk(KERN_WARNING "cciss%d: a power on "
3522                         "or device reset detected\n", h->ctlr);
3523                 return 1;
3524         break;
3525         case UNIT_ATTENTION_CLEARED:
3526                 printk(KERN_WARNING "cciss%d: unit attention "
3527                     "cleared by another initiator\n", h->ctlr);
3528                 return 1;
3529         break;
3530         default:
3531                 printk(KERN_WARNING "cciss%d: unknown "
3532                         "unit attention detected\n", h->ctlr);
3533                                 return 1;
3534         }
3535 }
3536
3537 /*
3538  *  We cannot read the structure directly, for portability we must use
3539  *   the io functions.
3540  *   This is for debug only.
3541  */
3542 #ifdef CCISS_DEBUG
3543 static void print_cfg_table(CfgTable_struct *tb)
3544 {
3545         int i;
3546         char temp_name[17];
3547
3548         printk("Controller Configuration information\n");
3549         printk("------------------------------------\n");
3550         for (i = 0; i < 4; i++)
3551                 temp_name[i] = readb(&(tb->Signature[i]));
3552         temp_name[4] = '\0';
3553         printk("   Signature = %s\n", temp_name);
3554         printk("   Spec Number = %d\n", readl(&(tb->SpecValence)));
3555         printk("   Transport methods supported = 0x%x\n",
3556                readl(&(tb->TransportSupport)));
3557         printk("   Transport methods active = 0x%x\n",
3558                readl(&(tb->TransportActive)));
3559         printk("   Requested transport Method = 0x%x\n",
3560                readl(&(tb->HostWrite.TransportRequest)));
3561         printk("   Coalesce Interrupt Delay = 0x%x\n",
3562                readl(&(tb->HostWrite.CoalIntDelay)));
3563         printk("   Coalesce Interrupt Count = 0x%x\n",
3564                readl(&(tb->HostWrite.CoalIntCount)));
3565         printk("   Max outstanding commands = 0x%d\n",
3566                readl(&(tb->CmdsOutMax)));
3567         printk("   Bus Types = 0x%x\n", readl(&(tb->BusTypes)));
3568         for (i = 0; i < 16; i++)
3569                 temp_name[i] = readb(&(tb->ServerName[i]));
3570         temp_name[16] = '\0';
3571         printk("   Server Name = %s\n", temp_name);
3572         printk("   Heartbeat Counter = 0x%x\n\n\n", readl(&(tb->HeartBeat)));
3573 }
3574 #endif                          /* CCISS_DEBUG */
3575
3576 static int find_PCI_BAR_index(struct pci_dev *pdev, unsigned long pci_bar_addr)
3577 {
3578         int i, offset, mem_type, bar_type;
3579         if (pci_bar_addr == PCI_BASE_ADDRESS_0) /* looking for BAR zero? */
3580                 return 0;
3581         offset = 0;
3582         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3583                 bar_type = pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE;
3584                 if (bar_type == PCI_BASE_ADDRESS_SPACE_IO)
3585                         offset += 4;
3586                 else {
3587                         mem_type = pci_resource_flags(pdev, i) &
3588                             PCI_BASE_ADDRESS_MEM_TYPE_MASK;
3589                         switch (mem_type) {
3590                         case PCI_BASE_ADDRESS_MEM_TYPE_32:
3591                         case PCI_BASE_ADDRESS_MEM_TYPE_1M:
3592                                 offset += 4;    /* 32 bit */
3593                                 break;
3594                         case PCI_BASE_ADDRESS_MEM_TYPE_64:
3595                                 offset += 8;
3596                                 break;
3597                         default:        /* reserved in PCI 2.2 */
3598                                 printk(KERN_WARNING
3599                                        "Base address is invalid\n");
3600                                 return -1;
3601                                 break;
3602                         }
3603                 }
3604                 if (offset == pci_bar_addr - PCI_BASE_ADDRESS_0)
3605                         return i + 1;
3606         }
3607         return -1;
3608 }
3609
3610 /* If MSI/MSI-X is supported by the kernel we will try to enable it on
3611  * controllers that are capable. If not, we use IO-APIC mode.
3612  */
3613
3614 static void __devinit cciss_interrupt_mode(ctlr_info_t *c,
3615                                            struct pci_dev *pdev, __u32 board_id)
3616 {
3617 #ifdef CONFIG_PCI_MSI
3618         int err;
3619         struct msix_entry cciss_msix_entries[4] = { {0, 0}, {0, 1},
3620         {0, 2}, {0, 3}
3621         };
3622
3623         /* Some boards advertise MSI but don't really support it */
3624         if ((board_id == 0x40700E11) ||
3625             (board_id == 0x40800E11) ||
3626             (board_id == 0x40820E11) || (board_id == 0x40830E11))
3627                 goto default_int_mode;
3628
3629         if (pci_find_capability(pdev, PCI_CAP_ID_MSIX)) {
3630                 err = pci_enable_msix(pdev, cciss_msix_entries, 4);
3631                 if (!err) {
3632                         c->intr[0] = cciss_msix_entries[0].vector;
3633                         c->intr[1] = cciss_msix_entries[1].vector;
3634                         c->intr[2] = cciss_msix_entries[2].vector;
3635                         c->intr[3] = cciss_msix_entries[3].vector;
3636                         c->msix_vector = 1;
3637                         return;
3638                 }
3639                 if (err > 0) {
3640                         printk(KERN_WARNING "cciss: only %d MSI-X vectors "
3641                                "available\n", err);
3642                         goto default_int_mode;
3643                 } else {
3644                         printk(KERN_WARNING "cciss: MSI-X init failed %d\n",
3645                                err);
3646                         goto default_int_mode;
3647                 }
3648         }
3649         if (pci_find_capability(pdev, PCI_CAP_ID_MSI)) {
3650                 if (!pci_enable_msi(pdev)) {
3651                         c->msi_vector = 1;
3652                 } else {
3653                         printk(KERN_WARNING "cciss: MSI init failed\n");
3654                 }
3655         }
3656 default_int_mode:
3657 #endif                          /* CONFIG_PCI_MSI */
3658         /* if we get here we're going to use the default interrupt mode */
3659         c->intr[SIMPLE_MODE_INT] = pdev->irq;
3660         return;
3661 }
3662
3663 static int __devinit cciss_pci_init(ctlr_info_t *c, struct pci_dev *pdev)
3664 {
3665         ushort subsystem_vendor_id, subsystem_device_id, command;
3666         __u32 board_id, scratchpad = 0;
3667         __u64 cfg_offset;
3668         __u32 cfg_base_addr;
3669         __u64 cfg_base_addr_index;
3670         int i, err;
3671
3672         /* check to see if controller has been disabled */
3673         /* BEFORE trying to enable it */
3674         (void)pci_read_config_word(pdev, PCI_COMMAND, &command);
3675         if (!(command & 0x02)) {
3676                 printk(KERN_WARNING
3677                        "cciss: controller appears to be disabled\n");
3678                 return -ENODEV;
3679         }
3680
3681         err = pci_enable_device(pdev);
3682         if (err) {
3683                 printk(KERN_ERR "cciss: Unable to Enable PCI device\n");
3684                 return err;
3685         }
3686
3687         err = pci_request_regions(pdev, "cciss");
3688         if (err) {
3689                 printk(KERN_ERR "cciss: Cannot obtain PCI resources, "
3690                        "aborting\n");
3691                 return err;
3692         }
3693
3694         subsystem_vendor_id = pdev->subsystem_vendor;
3695         subsystem_device_id = pdev->subsystem_device;
3696         board_id = (((__u32) (subsystem_device_id << 16) & 0xffff0000) |
3697                     subsystem_vendor_id);
3698
3699 #ifdef CCISS_DEBUG
3700         printk("command = %x\n", command);
3701         printk("irq = %x\n", pdev->irq);
3702         printk("board_id = %x\n", board_id);
3703 #endif                          /* CCISS_DEBUG */
3704
3705 /* If the kernel supports MSI/MSI-X we will try to enable that functionality,
3706  * else we use the IO-APIC interrupt assigned to us by system ROM.
3707  */
3708         cciss_interrupt_mode(c, pdev, board_id);
3709
3710         /* find the memory BAR */
3711         for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
3712                 if (pci_resource_flags(pdev, i) & IORESOURCE_MEM)
3713                         break;
3714         }
3715         if (i == DEVICE_COUNT_RESOURCE) {
3716                 printk(KERN_WARNING "cciss: No memory BAR found\n");
3717                 err = -ENODEV;
3718                 goto err_out_free_res;
3719         }
3720
3721         c->paddr = pci_resource_start(pdev, i); /* addressing mode bits
3722                                                  * already removed
3723                                                  */
3724
3725 #ifdef CCISS_DEBUG
3726         printk("address 0 = %lx\n", c->paddr);
3727 #endif                          /* CCISS_DEBUG */
3728         c->vaddr = remap_pci_mem(c->paddr, 0x250);
3729
3730         /* Wait for the board to become ready.  (PCI hotplug needs this.)
3731          * We poll for up to 120 secs, once per 100ms. */
3732         for (i = 0; i < 1200; i++) {
3733                 scratchpad = readl(c->vaddr + SA5_SCRATCHPAD_OFFSET);
3734                 if (scratchpad == CCISS_FIRMWARE_READY)
3735                         break;
3736                 set_current_state(TASK_INTERRUPTIBLE);
3737                 schedule_timeout(msecs_to_jiffies(100));        /* wait 100ms */
3738         }
3739         if (scratchpad != CCISS_FIRMWARE_READY) {
3740                 printk(KERN_WARNING "cciss: Board not ready.  Timed out.\n");
3741                 err = -ENODEV;
3742                 goto err_out_free_res;
3743         }
3744
3745         /* get the address index number */
3746         cfg_base_addr = readl(c->vaddr + SA5_CTCFG_OFFSET);
3747         cfg_base_addr &= (__u32) 0x0000ffff;
3748 #ifdef CCISS_DEBUG
3749         printk("cfg base address = %x\n", cfg_base_addr);
3750 #endif                          /* CCISS_DEBUG */
3751         cfg_base_addr_index = find_PCI_BAR_index(pdev, cfg_base_addr);
3752 #ifdef CCISS_DEBUG
3753         printk("cfg base address index = %llx\n",
3754                 (unsigned long long)cfg_base_addr_index);
3755 #endif                          /* CCISS_DEBUG */
3756         if (cfg_base_addr_index == -1) {
3757                 printk(KERN_WARNING "cciss: Cannot find cfg_base_addr_index\n");
3758                 err = -ENODEV;
3759                 goto err_out_free_res;
3760         }
3761
3762         cfg_offset = readl(c->vaddr + SA5_CTMEM_OFFSET);
3763 #ifdef CCISS_DEBUG
3764         printk("cfg offset = %llx\n", (unsigned long long)cfg_offset);
3765 #endif                          /* CCISS_DEBUG */
3766         c->cfgtable = remap_pci_mem(pci_resource_start(pdev,
3767                                                        cfg_base_addr_index) +
3768                                     cfg_offset, sizeof(CfgTable_struct));
3769         c->board_id = board_id;
3770
3771 #ifdef CCISS_DEBUG
3772         print_cfg_table(c->cfgtable);
3773 #endif                          /* CCISS_DEBUG */
3774
3775         /* Some controllers support Zero Memory Raid (ZMR).
3776          * When configured in ZMR mode the number of supported
3777          * commands drops to 64. So instead of just setting an
3778          * arbitrary value we make the driver a little smarter.
3779          * We read the config table to tell us how many commands
3780          * are supported on the controller then subtract 4 to
3781          * leave a little room for ioctl calls.
3782          */
3783         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3784         for (i = 0; i < ARRAY_SIZE(products); i++) {
3785                 if (board_id == products[i].board_id) {
3786                         c->product_name = products[i].product_name;
3787                         c->access = *(products[i].access);
3788                         c->nr_cmds = c->max_commands - 4;
3789                         break;
3790                 }
3791         }
3792         if ((readb(&c->cfgtable->Signature[0]) != 'C') ||
3793             (readb(&c->cfgtable->Signature[1]) != 'I') ||
3794             (readb(&c->cfgtable->Signature[2]) != 'S') ||
3795             (readb(&c->cfgtable->Signature[3]) != 'S')) {
3796                 printk("Does not appear to be a valid CISS config table\n");
3797                 err = -ENODEV;
3798                 goto err_out_free_res;
3799         }
3800         /* We didn't find the controller in our list. We know the
3801          * signature is valid. If it's an HP device let's try to
3802          * bind to the device and fire it up. Otherwise we bail.
3803          */
3804         if (i == ARRAY_SIZE(products)) {
3805                 if (subsystem_vendor_id == PCI_VENDOR_ID_HP) {
3806                         c->product_name = products[i-1].product_name;
3807                         c->access = *(products[i-1].access);
3808                         c->nr_cmds = c->max_commands - 4;
3809                         printk(KERN_WARNING "cciss: This is an unknown "
3810                                 "Smart Array controller.\n"
3811                                 "cciss: Please update to the latest driver "
3812                                 "available from www.hp.com.\n");
3813                 } else {
3814                         printk(KERN_WARNING "cciss: Sorry, I don't know how"
3815                                 " to access the Smart Array controller %08lx\n"
3816                                         , (unsigned long)board_id);
3817                         err = -ENODEV;
3818                         goto err_out_free_res;
3819                 }
3820         }
3821 #ifdef CONFIG_X86
3822         {
3823                 /* Need to enable prefetch in the SCSI core for 6400 in x86 */
3824                 __u32 prefetch;
3825                 prefetch = readl(&(c->cfgtable->SCSI_Prefetch));
3826                 prefetch |= 0x100;
3827                 writel(prefetch, &(c->cfgtable->SCSI_Prefetch));
3828         }
3829 #endif
3830
3831         /* Disabling DMA prefetch and refetch for the P600.
3832          * An ASIC bug may result in accesses to invalid memory addresses.
3833          * We've disabled prefetch for some time now. Testing with XEN
3834          * kernels revealed a bug in the refetch if dom0 resides on a P600.
3835          */
3836         if(board_id == 0x3225103C) {
3837                 __u32 dma_prefetch;
3838                 __u32 dma_refetch;
3839                 dma_prefetch = readl(c->vaddr + I2O_DMA1_CFG);
3840                 dma_prefetch |= 0x8000;
3841                 writel(dma_prefetch, c->vaddr + I2O_DMA1_CFG);
3842                 pci_read_config_dword(pdev, PCI_COMMAND_PARITY, &dma_refetch);
3843                 dma_refetch |= 0x1;
3844                 pci_write_config_dword(pdev, PCI_COMMAND_PARITY, dma_refetch);
3845         }
3846
3847 #ifdef CCISS_DEBUG
3848         printk("Trying to put board into Simple mode\n");
3849 #endif                          /* CCISS_DEBUG */
3850         c->max_commands = readl(&(c->cfgtable->CmdsOutMax));
3851         /* Update the field, and then ring the doorbell */
3852         writel(CFGTBL_Trans_Simple, &(c->cfgtable->HostWrite.TransportRequest));
3853         writel(CFGTBL_ChangeReq, c->vaddr + SA5_DOORBELL);
3854
3855         /* under certain very rare conditions, this can take awhile.
3856          * (e.g.: hot replace a failed 144GB drive in a RAID 5 set right
3857          * as we enter this code.) */
3858         for (i = 0; i < MAX_CONFIG_WAIT; i++) {
3859                 if (!(readl(c->vaddr + SA5_DOORBELL) & CFGTBL_ChangeReq))
3860                         break;
3861                 /* delay and try again */
3862                 set_current_state(TASK_INTERRUPTIBLE);
3863                 schedule_timeout(msecs_to_jiffies(1));
3864         }
3865
3866 #ifdef CCISS_DEBUG
3867         printk(KERN_DEBUG "I counter got to %d %x\n", i,
3868                readl(c->vaddr + SA5_DOORBELL));
3869 #endif                          /* CCISS_DEBUG */
3870 #ifdef CCISS_DEBUG
3871         print_cfg_table(c->cfgtable);
3872 #endif                          /* CCISS_DEBUG */
3873
3874         if (!(readl(&(c->cfgtable->TransportActive)) & CFGTBL_Trans_Simple)) {
3875                 printk(KERN_WARNING "cciss: unable to get board into"
3876                        " simple mode\n");
3877                 err = -ENODEV;
3878                 goto err_out_free_res;
3879         }
3880         return 0;
3881
3882 err_out_free_res:
3883         /*
3884          * Deliberately omit pci_disable_device(): it does something nasty to
3885          * Smart Array controllers that pci_enable_device does not undo
3886          */
3887         pci_release_regions(pdev);
3888         return err;
3889 }
3890
3891 /* Function to find the first free pointer into our hba[] array
3892  * Returns -1 if no free entries are left.
3893  */
3894 static int alloc_cciss_hba(void)
3895 {
3896         int i;
3897
3898         for (i = 0; i < MAX_CTLR; i++) {
3899                 if (!hba[i]) {
3900                         ctlr_info_t *p;
3901
3902                         p = kzalloc(sizeof(ctlr_info_t), GFP_KERNEL);
3903                         if (!p)
3904                                 goto Enomem;
3905                         hba[i] = p;
3906                         return i;
3907                 }
3908         }
3909         printk(KERN_WARNING "cciss: This driver supports a maximum"
3910                " of %d controllers.\n", MAX_CTLR);
3911         return -1;
3912 Enomem:
3913         printk(KERN_ERR "cciss: out of memory.\n");
3914         return -1;
3915 }
3916
3917 static void free_hba(int n)
3918 {
3919         ctlr_info_t *h = hba[n];
3920         int i;
3921
3922         hba[n] = NULL;
3923         for (i = 0; i < h->highest_lun + 1; i++)
3924                 if (h->gendisk[i] != NULL)
3925                         put_disk(h->gendisk[i]);
3926         kfree(h);
3927 }
3928
3929 /* Send a message CDB to the firmware. */
3930 static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type)
3931 {
3932         typedef struct {
3933                 CommandListHeader_struct CommandHeader;
3934                 RequestBlock_struct Request;
3935                 ErrDescriptor_struct ErrorDescriptor;
3936         } Command;
3937         static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct);
3938         Command *cmd;
3939         dma_addr_t paddr64;
3940         uint32_t paddr32, tag;
3941         void __iomem *vaddr;
3942         int i, err;
3943
3944         vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0));
3945         if (vaddr == NULL)
3946                 return -ENOMEM;
3947
3948         /* The Inbound Post Queue only accepts 32-bit physical addresses for the
3949            CCISS commands, so they must be allocated from the lower 4GiB of
3950            memory. */
3951         err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
3952         if (err) {
3953                 iounmap(vaddr);
3954                 return -ENOMEM;
3955         }
3956
3957         cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64);
3958         if (cmd == NULL) {
3959                 iounmap(vaddr);
3960                 return -ENOMEM;
3961         }
3962
3963         /* This must fit, because of the 32-bit consistent DMA mask.  Also,
3964            although there's no guarantee, we assume that the address is at
3965            least 4-byte aligned (most likely, it's page-aligned). */
3966         paddr32 = paddr64;
3967
3968         cmd->CommandHeader.ReplyQueue = 0;
3969         cmd->CommandHeader.SGList = 0;
3970         cmd->CommandHeader.SGTotal = 0;
3971         cmd->CommandHeader.Tag.lower = paddr32;
3972         cmd->CommandHeader.Tag.upper = 0;
3973         memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8);
3974
3975         cmd->Request.CDBLen = 16;
3976         cmd->Request.Type.Type = TYPE_MSG;
3977         cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE;
3978         cmd->Request.Type.Direction = XFER_NONE;
3979         cmd->Request.Timeout = 0; /* Don't time out */
3980         cmd->Request.CDB[0] = opcode;
3981         cmd->Request.CDB[1] = type;
3982         memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */
3983
3984         cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command);
3985         cmd->ErrorDescriptor.Addr.upper = 0;
3986         cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct);
3987
3988         writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET);
3989
3990         for (i = 0; i < 10; i++) {
3991                 tag = readl(vaddr + SA5_REPLY_PORT_OFFSET);
3992                 if ((tag & ~3) == paddr32)
3993                         break;
3994                 schedule_timeout_uninterruptible(HZ);
3995         }
3996
3997         iounmap(vaddr);
3998
3999         /* we leak the DMA buffer here ... no choice since the controller could
4000            still complete the command. */
4001         if (i == 10) {
4002                 printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n",
4003                         opcode, type);
4004                 return -ETIMEDOUT;
4005         }
4006
4007         pci_free_consistent(pdev, cmd_sz, cmd, paddr64);
4008
4009         if (tag & 2) {
4010                 printk(KERN_ERR "cciss: controller message %02x:%02x failed\n",
4011                         opcode, type);
4012                 return -EIO;
4013         }
4014
4015         printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n",
4016                 opcode, type);
4017         return 0;
4018 }
4019
4020 #define cciss_soft_reset_controller(p) cciss_message(p, 1, 0)
4021 #define cciss_noop(p) cciss_message(p, 3, 0)
4022
4023 static __devinit int cciss_reset_msi(struct pci_dev *pdev)
4024 {
4025 /* the #defines are stolen from drivers/pci/msi.h. */
4026 #define msi_control_reg(base)           (base + PCI_MSI_FLAGS)
4027 #define PCI_MSIX_FLAGS_ENABLE           (1 << 15)
4028
4029         int pos;
4030         u16 control = 0;
4031
4032         pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
4033         if (pos) {
4034                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4035                 if (control & PCI_MSI_FLAGS_ENABLE) {
4036                         printk(KERN_INFO "cciss: resetting MSI\n");
4037                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE);
4038                 }
4039         }
4040
4041         pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
4042         if (pos) {
4043                 pci_read_config_word(pdev, msi_control_reg(pos), &control);
4044                 if (control & PCI_MSIX_FLAGS_ENABLE) {
4045                         printk(KERN_INFO "cciss: resetting MSI-X\n");
4046                         pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE);
4047                 }
4048         }
4049
4050         return 0;
4051 }
4052
4053 /* This does a hard reset of the controller using PCI power management
4054  * states. */
4055 static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev)
4056 {
4057         u16 pmcsr, saved_config_space[32];
4058         int i, pos;
4059
4060         printk(KERN_INFO "cciss: using PCI PM to reset controller\n");
4061
4062         /* This is very nearly the same thing as
4063
4064            pci_save_state(pci_dev);
4065            pci_set_power_state(pci_dev, PCI_D3hot);
4066            pci_set_power_state(pci_dev, PCI_D0);
4067            pci_restore_state(pci_dev);
4068
4069            but we can't use these nice canned kernel routines on
4070            kexec, because they also check the MSI/MSI-X state in PCI
4071            configuration space and do the wrong thing when it is
4072            set/cleared.  Also, the pci_save/restore_state functions
4073            violate the ordering requirements for restoring the
4074            configuration space from the CCISS document (see the
4075            comment below).  So we roll our own .... */
4076
4077         for (i = 0; i < 32; i++)
4078                 pci_read_config_word(pdev, 2*i, &saved_config_space[i]);
4079
4080         pos = pci_find_capability(pdev, PCI_CAP_ID_PM);
4081         if (pos == 0) {
4082                 printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n");
4083                 return -ENODEV;
4084         }
4085
4086         /* Quoting from the Open CISS Specification: "The Power
4087          * Management Control/Status Register (CSR) controls the power
4088          * state of the device.  The normal operating state is D0,
4089          * CSR=00h.  The software off state is D3, CSR=03h.  To reset
4090          * the controller, place the interface device in D3 then to
4091          * D0, this causes a secondary PCI reset which will reset the
4092          * controller." */
4093
4094         /* enter the D3hot power management state */
4095         pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr);
4096         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4097         pmcsr |= PCI_D3hot;
4098         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4099
4100         schedule_timeout_uninterruptible(HZ >> 1);
4101
4102         /* enter the D0 power management state */
4103         pmcsr &= ~PCI_PM_CTRL_STATE_MASK;
4104         pmcsr |= PCI_D0;
4105         pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr);
4106
4107         schedule_timeout_uninterruptible(HZ >> 1);
4108
4109         /* Restore the PCI configuration space.  The Open CISS
4110          * Specification says, "Restore the PCI Configuration
4111          * Registers, offsets 00h through 60h. It is important to
4112          * restore the command register, 16-bits at offset 04h,
4113          * last. Do not restore the configuration status register,
4114          * 16-bits at offset 06h."  Note that the offset is 2*i. */
4115         for (i = 0; i < 32; i++) {
4116                 if (i == 2 || i == 3)
4117                         continue;
4118                 pci_write_config_word(pdev, 2*i, saved_config_space[i]);
4119         }
4120         wmb();
4121         pci_write_config_word(pdev, 4, saved_config_space[2]);
4122
4123         return 0;
4124 }
4125
4126 /*
4127  *  This is it.  Find all the controllers and register them.  I really hate
4128  *  stealing all these major device numbers.
4129  *  returns the number of block devices registered.
4130  */
4131 static int __devinit cciss_init_one(struct pci_dev *pdev,
4132                                     const struct pci_device_id *ent)
4133 {
4134         int i;
4135         int j = 0;
4136         int rc;
4137         int dac, return_code;
4138         InquiryData_struct *inq_buff;
4139
4140         if (reset_devices) {
4141                 /* Reset the controller with a PCI power-cycle */
4142                 if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev))
4143                         return -ENODEV;
4144
4145                 /* Now try to get the controller to respond to a no-op. Some
4146                    devices (notably the HP Smart Array 5i Controller) need
4147                    up to 30 seconds to respond. */
4148                 for (i=0; i<30; i++) {
4149                         if (cciss_noop(pdev) == 0)
4150                                 break;
4151
4152                         schedule_timeout_uninterruptible(HZ);
4153                 }
4154                 if (i == 30) {
4155                         printk(KERN_ERR "cciss: controller seems dead\n");
4156                         return -EBUSY;
4157                 }
4158         }
4159
4160         i = alloc_cciss_hba();
4161         if (i < 0)
4162                 return -1;
4163
4164         hba[i]->busy_initializing = 1;
4165         INIT_HLIST_HEAD(&hba[i]->cmpQ);
4166         INIT_HLIST_HEAD(&hba[i]->reqQ);
4167         mutex_init(&hba[i]->busy_shutting_down);
4168
4169         if (cciss_pci_init(hba[i], pdev) != 0)
4170                 goto clean0;
4171
4172         sprintf(hba[i]->devname, "cciss%d", i);
4173         hba[i]->ctlr = i;
4174         hba[i]->pdev = pdev;
4175
4176         init_completion(&hba[i]->scan_wait);
4177
4178         if (cciss_create_hba_sysfs_entry(hba[i]))
4179                 goto clean0;
4180
4181         /* configure PCI DMA stuff */
4182         if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64)))
4183                 dac = 1;
4184         else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))
4185                 dac = 0;
4186         else {
4187                 printk(KERN_ERR "cciss: no suitable DMA available\n");
4188                 goto clean1;
4189         }
4190
4191         /*
4192          * register with the major number, or get a dynamic major number
4193          * by passing 0 as argument.  This is done for greater than
4194          * 8 controller support.
4195          */
4196         if (i < MAX_CTLR_ORIG)
4197                 hba[i]->major = COMPAQ_CISS_MAJOR + i;
4198         rc = register_blkdev(hba[i]->major, hba[i]->devname);
4199         if (rc == -EBUSY || rc == -EINVAL) {
4200                 printk(KERN_ERR
4201                        "cciss:  Unable to get major number %d for %s "
4202                        "on hba %d\n", hba[i]->major, hba[i]->devname, i);
4203                 goto clean1;
4204         } else {
4205                 if (i >= MAX_CTLR_ORIG)
4206                         hba[i]->major = rc;
4207         }
4208
4209         /* make sure the board interrupts are off */
4210         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_OFF);
4211         if (request_irq(hba[i]->intr[SIMPLE_MODE_INT], do_cciss_intr,
4212                         IRQF_DISABLED | IRQF_SHARED, hba[i]->devname, hba[i])) {
4213                 printk(KERN_ERR "cciss: Unable to get irq %d for %s\n",
4214                        hba[i]->intr[SIMPLE_MODE_INT], hba[i]->devname);
4215                 goto clean2;
4216         }
4217
4218         printk(KERN_INFO "%s: <0x%x> at PCI %s IRQ %d%s using DAC\n",
4219                hba[i]->devname, pdev->device, pci_name(pdev),
4220                hba[i]->intr[SIMPLE_MODE_INT], dac ? "" : " not");
4221
4222         hba[i]->cmd_pool_bits =
4223             kmalloc(DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4224                         * sizeof(unsigned long), GFP_KERNEL);
4225         hba[i]->cmd_pool = (CommandList_struct *)
4226             pci_alloc_consistent(hba[i]->pdev,
4227                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4228                     &(hba[i]->cmd_pool_dhandle));
4229         hba[i]->errinfo_pool = (ErrorInfo_struct *)
4230             pci_alloc_consistent(hba[i]->pdev,
4231                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4232                     &(hba[i]->errinfo_pool_dhandle));
4233         if ((hba[i]->cmd_pool_bits == NULL)
4234             || (hba[i]->cmd_pool == NULL)
4235             || (hba[i]->errinfo_pool == NULL)) {
4236                 printk(KERN_ERR "cciss: out of memory");
4237                 goto clean4;
4238         }
4239         spin_lock_init(&hba[i]->lock);
4240
4241         /* Initialize the pdev driver private data.
4242            have it point to hba[i].  */
4243         pci_set_drvdata(pdev, hba[i]);
4244         /* command and error info recs zeroed out before
4245            they are used */
4246         memset(hba[i]->cmd_pool_bits, 0,
4247                DIV_ROUND_UP(hba[i]->nr_cmds, BITS_PER_LONG)
4248                         * sizeof(unsigned long));
4249
4250         hba[i]->num_luns = 0;
4251         hba[i]->highest_lun = -1;
4252         for (j = 0; j < CISS_MAX_LUN; j++) {
4253                 hba[i]->drv[j].raid_level = -1;
4254                 hba[i]->drv[j].queue = NULL;
4255                 hba[i]->gendisk[j] = NULL;
4256         }
4257
4258         cciss_scsi_setup(i);
4259
4260         /* Turn the interrupts on so we can service requests */
4261         hba[i]->access.set_intr_mask(hba[i], CCISS_INTR_ON);
4262
4263         /* Get the firmware version */
4264         inq_buff = kzalloc(sizeof(InquiryData_struct), GFP_KERNEL);
4265         if (inq_buff == NULL) {
4266                 printk(KERN_ERR "cciss: out of memory\n");
4267                 goto clean4;
4268         }
4269
4270         return_code = sendcmd_withirq(CISS_INQUIRY, i, inq_buff,
4271                 sizeof(InquiryData_struct), 0, CTLR_LUNID, TYPE_CMD);
4272         if (return_code == IO_OK) {
4273                 hba[i]->firm_ver[0] = inq_buff->data_byte[32];
4274                 hba[i]->firm_ver[1] = inq_buff->data_byte[33];
4275                 hba[i]->firm_ver[2] = inq_buff->data_byte[34];
4276                 hba[i]->firm_ver[3] = inq_buff->data_byte[35];
4277         } else {         /* send command failed */
4278                 printk(KERN_WARNING "cciss: unable to determine firmware"
4279                         " version of controller\n");
4280         }
4281         kfree(inq_buff);
4282
4283         cciss_procinit(i);
4284
4285         hba[i]->cciss_max_sectors = 2048;
4286
4287         rebuild_lun_table(hba[i], 1, 0);
4288         hba[i]->busy_initializing = 0;
4289         return 1;
4290
4291 clean4:
4292         kfree(hba[i]->cmd_pool_bits);
4293         if (hba[i]->cmd_pool)
4294                 pci_free_consistent(hba[i]->pdev,
4295                                     hba[i]->nr_cmds * sizeof(CommandList_struct),
4296                                     hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4297         if (hba[i]->errinfo_pool)
4298                 pci_free_consistent(hba[i]->pdev,
4299                                     hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4300                                     hba[i]->errinfo_pool,
4301                                     hba[i]->errinfo_pool_dhandle);
4302         free_irq(hba[i]->intr[SIMPLE_MODE_INT], hba[i]);
4303 clean2:
4304         unregister_blkdev(hba[i]->major, hba[i]->devname);
4305 clean1:
4306         cciss_destroy_hba_sysfs_entry(hba[i]);
4307 clean0:
4308         hba[i]->busy_initializing = 0;
4309         /* cleanup any queues that may have been initialized */
4310         for (j=0; j <= hba[i]->highest_lun; j++){
4311                 drive_info_struct *drv = &(hba[i]->drv[j]);
4312                 if (drv->queue)
4313                         blk_cleanup_queue(drv->queue);
4314         }
4315         /*
4316          * Deliberately omit pci_disable_device(): it does something nasty to
4317          * Smart Array controllers that pci_enable_device does not undo
4318          */
4319         pci_release_regions(pdev);
4320         pci_set_drvdata(pdev, NULL);
4321         free_hba(i);
4322         return -1;
4323 }
4324
4325 static void cciss_shutdown(struct pci_dev *pdev)
4326 {
4327         ctlr_info_t *tmp_ptr;
4328         int i;
4329         char flush_buf[4];
4330         int return_code;
4331
4332         tmp_ptr = pci_get_drvdata(pdev);
4333         if (tmp_ptr == NULL)
4334                 return;
4335         i = tmp_ptr->ctlr;
4336         if (hba[i] == NULL)
4337                 return;
4338
4339         /* Turn board interrupts off  and send the flush cache command */
4340         /* sendcmd will turn off interrupt, and send the flush...
4341          * To write all data in the battery backed cache to disks */
4342         memset(flush_buf, 0, 4);
4343         return_code = sendcmd(CCISS_CACHE_FLUSH, i, flush_buf, 4, 0,
4344                 CTLR_LUNID, TYPE_CMD);
4345         if (return_code == IO_OK) {
4346                 printk(KERN_INFO "Completed flushing cache on controller %d\n", i);
4347         } else {
4348                 printk(KERN_WARNING "Error flushing cache on controller %d\n", i);
4349         }
4350         free_irq(hba[i]->intr[2], hba[i]);
4351 }
4352
4353 static void __devexit cciss_remove_one(struct pci_dev *pdev)
4354 {
4355         ctlr_info_t *tmp_ptr;
4356         int i, j;
4357
4358         if (pci_get_drvdata(pdev) == NULL) {
4359                 printk(KERN_ERR "cciss: Unable to remove device \n");
4360                 return;
4361         }
4362
4363         tmp_ptr = pci_get_drvdata(pdev);
4364         i = tmp_ptr->ctlr;
4365         if (hba[i] == NULL) {
4366                 printk(KERN_ERR "cciss: device appears to "
4367                        "already be removed \n");
4368                 return;
4369         }
4370
4371         mutex_lock(&hba[i]->busy_shutting_down);
4372
4373         remove_from_scan_list(hba[i]);
4374         remove_proc_entry(hba[i]->devname, proc_cciss);
4375         unregister_blkdev(hba[i]->major, hba[i]->devname);
4376
4377         /* remove it from the disk list */
4378         for (j = 0; j < CISS_MAX_LUN; j++) {
4379                 struct gendisk *disk = hba[i]->gendisk[j];
4380                 if (disk) {
4381                         struct request_queue *q = disk->queue;
4382
4383                         if (disk->flags & GENHD_FL_UP) {
4384                                 cciss_destroy_ld_sysfs_entry(hba[i], j, 1);
4385                                 del_gendisk(disk);
4386                         }
4387                         if (q)
4388                                 blk_cleanup_queue(q);
4389                 }
4390         }
4391
4392 #ifdef CONFIG_CISS_SCSI_TAPE
4393         cciss_unregister_scsi(i);       /* unhook from SCSI subsystem */
4394 #endif
4395
4396         cciss_shutdown(pdev);
4397
4398 #ifdef CONFIG_PCI_MSI
4399         if (hba[i]->msix_vector)
4400                 pci_disable_msix(hba[i]->pdev);
4401         else if (hba[i]->msi_vector)
4402                 pci_disable_msi(hba[i]->pdev);
4403 #endif                          /* CONFIG_PCI_MSI */
4404
4405         iounmap(hba[i]->vaddr);
4406
4407         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(CommandList_struct),
4408                             hba[i]->cmd_pool, hba[i]->cmd_pool_dhandle);
4409         pci_free_consistent(hba[i]->pdev, hba[i]->nr_cmds * sizeof(ErrorInfo_struct),
4410                             hba[i]->errinfo_pool, hba[i]->errinfo_pool_dhandle);
4411         kfree(hba[i]->cmd_pool_bits);
4412         /*
4413          * Deliberately omit pci_disable_device(): it does something nasty to
4414          * Smart Array controllers that pci_enable_device does not undo
4415          */
4416         pci_release_regions(pdev);
4417         pci_set_drvdata(pdev, NULL);
4418         cciss_destroy_hba_sysfs_entry(hba[i]);
4419         mutex_unlock(&hba[i]->busy_shutting_down);
4420         free_hba(i);
4421 }
4422
4423 static struct pci_driver cciss_pci_driver = {
4424         .name = "cciss",
4425         .probe = cciss_init_one,
4426         .remove = __devexit_p(cciss_remove_one),
4427         .id_table = cciss_pci_device_id,        /* id_table */
4428         .shutdown = cciss_shutdown,
4429 };
4430
4431 /*
4432  *  This is it.  Register the PCI driver information for the cards we control
4433  *  the OS will call our registered routines when it finds one of our cards.
4434  */
4435 static int __init cciss_init(void)
4436 {
4437         int err;
4438
4439         /*
4440          * The hardware requires that commands are aligned on a 64-bit
4441          * boundary. Given that we use pci_alloc_consistent() to allocate an
4442          * array of them, the size must be a multiple of 8 bytes.
4443          */
4444         BUILD_BUG_ON(sizeof(CommandList_struct) % 8);
4445
4446         printk(KERN_INFO DRIVER_NAME "\n");
4447
4448         err = bus_register(&cciss_bus_type);
4449         if (err)
4450                 return err;
4451
4452         /* Start the scan thread */
4453         cciss_scan_thread = kthread_run(scan_thread, NULL, "cciss_scan");
4454         if (IS_ERR(cciss_scan_thread)) {
4455                 err = PTR_ERR(cciss_scan_thread);
4456                 goto err_bus_unregister;
4457         }
4458
4459         /* Register for our PCI devices */
4460         err = pci_register_driver(&cciss_pci_driver);
4461         if (err)
4462                 goto err_thread_stop;
4463
4464         return err;
4465
4466 err_thread_stop:
4467         kthread_stop(cciss_scan_thread);
4468 err_bus_unregister:
4469         bus_unregister(&cciss_bus_type);
4470
4471         return err;
4472 }
4473
4474 static void __exit cciss_cleanup(void)
4475 {
4476         int i;
4477
4478         pci_unregister_driver(&cciss_pci_driver);
4479         /* double check that all controller entrys have been removed */
4480         for (i = 0; i < MAX_CTLR; i++) {
4481                 if (hba[i] != NULL) {
4482                         printk(KERN_WARNING "cciss: had to remove"
4483                                " controller %d\n", i);
4484                         cciss_remove_one(hba[i]->pdev);
4485                 }
4486         }
4487         kthread_stop(cciss_scan_thread);
4488         remove_proc_entry("driver/cciss", NULL);
4489         bus_unregister(&cciss_bus_type);
4490 }
4491
4492 static void fail_all_cmds(unsigned long ctlr)
4493 {
4494         /* If we get here, the board is apparently dead. */
4495         ctlr_info_t *h = hba[ctlr];
4496         CommandList_struct *c;
4497         unsigned long flags;
4498
4499         printk(KERN_WARNING "cciss%d: controller not responding.\n", h->ctlr);
4500         h->alive = 0;           /* the controller apparently died... */
4501
4502         spin_lock_irqsave(CCISS_LOCK(ctlr), flags);
4503
4504         pci_disable_device(h->pdev);    /* Make sure it is really dead. */
4505
4506         /* move everything off the request queue onto the completed queue */
4507         while (!hlist_empty(&h->reqQ)) {
4508                 c = hlist_entry(h->reqQ.first, CommandList_struct, list);
4509                 removeQ(c);
4510                 h->Qdepth--;
4511                 addQ(&h->cmpQ, c);
4512         }
4513
4514         /* Now, fail everything on the completed queue with a HW error */
4515         while (!hlist_empty(&h->cmpQ)) {
4516                 c = hlist_entry(h->cmpQ.first, CommandList_struct, list);
4517                 removeQ(c);
4518                 if (c->cmd_type != CMD_MSG_STALE)
4519                         c->err_info->CommandStatus = CMD_HARDWARE_ERR;
4520                 if (c->cmd_type == CMD_RWREQ) {
4521                         complete_command(h, c, 0);
4522                 } else if (c->cmd_type == CMD_IOCTL_PEND)
4523                         complete(c->waiting);
4524 #ifdef CONFIG_CISS_SCSI_TAPE
4525                 else if (c->cmd_type == CMD_SCSI)
4526                         complete_scsi_command(c, 0, 0);
4527 #endif
4528         }
4529         spin_unlock_irqrestore(CCISS_LOCK(ctlr), flags);
4530         return;
4531 }
4532
4533 module_init(cciss_init);
4534 module_exit(cciss_cleanup);